Remote Sensing

for Sustainability
 Taylor & Francis Series in
Remote Sensing Applications
Series Editor
Qihao Weng
Indiana State University
Terre Haute, Indiana, U.S.A.

Remote Sensing Applications for the Urban Environment, George Z. Xian

Remote Sensing of Impervious Surfaces in Tropical and Subtropical Areas,
Hui Lin, Yuanzhi Zhang, and Qihao Weng
Global Urban Monitoring and Assessment through Earth Observation,
edited by Qihao Weng
Remote Sensing of Natural Resources, edited by Guangxing Wang and Qihao Weng
Remote Sensing of Land Use and Land Cover: Principles and Applications,
Chandra P. Giri
Remote Sensing of Protected Lands, edited by Yeqiao Wang
Advances in Environmental Remote Sensing: Sensors, Algorithms, and
Applications, edited by Qihao Weng
Remote Sensing of Coastal Environments, edited by Yeqiao Wang
Remote Sensing of Global Croplands for Food Security,
edited by Prasad S. Thenkabail, John G. Lyon, Hugh Turral, and Chandashekhar M. Biradar
Global Mapping of Human Settlement: Experiences, Data Sets, and Prospects,
edited by Paolo Gamba and Martin Herold
Hyperspectral Remote Sensing: Principles and Applications, Marcus Borengasser,
William S. Hungate, and Russell Watkins
Remote Sensing of Impervious Surfaces, edited by Qihao Weng
Multispectral Image Analysis Using the Object-Oriented Paradigm, Kumar Navulur
Urban Remote Sensing, edited by Qihao Weng and Dale A. Quattrochi
Remote Sensing
for Sustainability

Edited by Qihao Weng

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2017 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

Version Date: 20160916

International Standard Book Number-13: 978-1-4987-0071-9 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmit-
ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.

Library of Congress Cataloging‑in‑Publication Data

Names: Weng, Qihao, editor.

Title: Remote sensing for sustainability / [edited by] Qihao Weng.
Description: Boca Raton : Taylor & Francis, a CRC title, part of the Taylor &
Francis imprint, a member of the Taylor & Francis Group, the academic
division of T&F Informa, plc, [2017] | Includes bibliographical references
and index.
Identifiers: LCCN 2016023093 | ISBN 9781498700719 (hardback : acid-free paper)
Subjects: LCSH: Environmental monitoring--Remote sensing. | Sustainable
development--Remote sensing. | Sustainable urban development--Remote
sensing. | Sustainable agriculture--Remote sensing. | Natural
resources--Remote sensing. | Renewable energy sources--Remote sensing.
Classification: LCC GE45.R44 R47 2017 | DDC 363.7/063028--dc23
LC record available at https://lccn.loc.gov/2016023093

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
Contents
Foreword ...................................................................................................................ix
Preface.......................................................................................................................xi
Acknowledgments ...................................................................................................xix
Editor ......................................................................................................................xxi
Contributors ......................................................................................................... xxiii

Section i Remote Sensing for Sustainable cities

Chapter 1 Extraction of Parameters from Remote Sensing Data

for Environmental Indices for Urban Sustainability ............................ 3
John C. Trinder

Chapter 2 Earth Observation for Urban and Spatial Planning ........................... 29

Mattia Marconcini, Annekatrin Metz, Thomas Esch,
and Julian Zeidler

Chapter 3 Assessment of Urban Growth in the Pearl River Delta, China,

Using Time Series Landsat Imagery .................................................. 45
Lei Zhang and Qihao Weng

Chapter 4 InSAR Monitoring of Land Subsidence for Sustainable Urban

Planning ............................................................................................. 61
Abduwasit Ghulam, Mark Grzovic, Maitiniyazi Maimaitijiang,
and Mamat Sawut

Chapter 5 A Tale of Two Cities: Urbanization in Greensboro,

North Carolina, and Guiyang, Guizhou, China ................................. 81
Honglin Xiao and Qihao Weng

v
vi Contents

Section ii Remote Sensing for Sustainable

natural Resources

Chapter 6 Role of Remote Sensing in Sustainable Grassland Management:

A Review and Case Studies for a Mixed-Grass Prairie Ecosystem ... 103
Alexander Tong, Bing Lu, Yuhong He, and Xulin Guo

Chapter 7 Classifying Tree Species Using Fine Spatial Resolution Imagery

to Support the Conservation of an Endangered Bird Species in
Hawaii .............................................................................................. 121
Qi Chen

Chapter 8 Remote Sensing of Forest Damage by Diseases and Insects ........... 145
Gang Chen and Ross K. Meentemeyer

Chapter 9 Monitoring Water Quality with Remote Sensing Image Data ......... 163
Bunkei Matsushita, Wei Yang, Lalu Muhamad Jaelani,
Fajar Setiawan, and Takehiko Fukushima

Section iii Remote Sensing for Sustainable

environmental Systems

Chapter 10 Urban Air Quality Studies Using EO Data ...................................... 193

Xuefei Hu

Chapter 11 Heat Hazard Monitoring with Satellite-Derived Land Surface

Temperature...................................................................................... 211
Yitong Jiang and Qihao Weng

Chapter 12 Remote Sensing Identification of Threshold Zones along

a Mediterranean to Arid Climatic Gradient ..................................... 229
Maxim Shoshany

Chapter 13 Soil Moisture Using Optical Remote Sensing and Ground

Measurements: A Case Study from Pakistan ................................... 243
Mudassar Umar, Siraj Munir, Iftikhar Ali, Salman Qureshi,
Claudia Notarnicola, Said Rahman, and Qihao Weng
Contents vii

Section iV Remote Sensing for Sustainable energy

Chapter 14 Earth Observation and Its Potential to Implement a Sustainable

Energy Supply—A German Perspective .......................................... 269
Thomas Esch, Markus Tum, and Annekatrin Metz

Chapter 15 Use of Nighttime Imaging Data to Assess Decadal Trends

in Energy Use in China .................................................................... 277
Yanhua Xie and Qihao Weng

Chapter 16 Support of Wind Resource Modeling Using Earth

Observation—A European Perspective on the Status
and Future Options ........................................................................... 297
Thomas Esch, Charlotte Bay Hasager, Paul Elsner,
Janik Deutscher, Manuela Hirschmugl, Annekatrin Metz,
and Achim Roth

Chapter 17 Assessing Solar Energy Potential and Building Energy Use

in Indianapolis Using Geospatial Techniques .................................. 317
Yuanfan Zheng and Qihao Weng

Index ...................................................................................................................... 351

Foreword
Dr. Qihao Weng has authored another outstanding remote sensing book entitled
Remote Sensing for Sustainability (CRC Press, 2016). Well known for his expertise in
remote sensing and GIS analysis of urban and environmental systems, this professor
from Indiana State University, director of the Center for Urban and Environmental
Change, series editor for the Taylor & Francis Series in Remote Sensing Applications,
and co-editor-in-chief of the ISPRS Journal of Photogrammetry and Remote Sensing
also is involved in the Group on Earth Observations or GEO. Established in 2005,
GEO is a partnership of governments and organizations with a vision toward a future
when global decisions benefiting humankind are informed by Earth observations
and information derived from coordinated and sustained remote sensing (http://
www.earthobservations.org). Professor Weng’s work as the task lead and coordina-
tor for GEO SB-04, Global Urban Observation and Information Task, led him to the
idea to write a book on the use of remote sensing for multiple and sustainable soci-
etal benefits, including biodiversity, disaster resilience, energy and mineral resources
management, food security, sustainable agriculture, infrastructure and transporta-
tion management, public health surveillance, sustainable urban development, and
water resources management. In other words, his book covers remote sensing of
all the requirements to sustain life on Earth and ensure environments that promote
human health, productivity, and security.
In this book, Remote Sensing for Sustainability, Weng recognizes that grand
and global challenges require international collaboration and cooperation in order
to optimize remote sensing data acquisition, storage, access, and dissemination. He
understands the value of sharing methods and techniques of image processing and
data analysis for producing accurate and actionable knowledge from images, point
clouds, sensor networks, ground measurements, and modeled predictions. Experts
from around the world were gathered to report on specialties of remote sensing for
sustainable cities, natural resources, environmental systems, and energy. Chapters
within these major topic areas cover critical issues of urban planning and growth,
endangered species conservation, water and air quality, forest damage by diseases
and insects, energy supplies, and sustainable wind and solar energy potential—all
using geospatial data, techniques, and analyses. As Barbara Ryan, GEO Secretariat
Executive Director, notes in the Preface of this book, there is a critical need to under-
stand the character of urban environments where approximately half of the human
global population live in only about 3% of the Earth surface. Rapidly expanding,
dense, sprawling, and vulnerable to human and natural impacts, urban systems must
be understood in order to make timely and informed decisions aimed toward min-
imizing deforestation, air and water pollution, energy consumption, food deserts,
marginal housing, and limited green spaces. This book makes a giant step forward

ix
x Foreword

in advancing global cooperation in remote sensing efforts and sharing knowledge

desperately needed for Earth sustainability in a changing climate.

Marguerite Madden
Center for Geospatial Research Department of Geography
The University of Georgia
Athens, GA
Preface
TOWARD A SUSTAINABLE EARTH THROUGH REMOTE SENSING
SuStainability and Earth ObSErvatiOn
In recent years, we have witnessed an emerging trend worldwide to pursue green
technologies and low-carbon economies and lifestyles. However, the sustainabil-
ity issue is not new. At the 1992 United Nations Conference on Environment and
Development (UNCED) in Rio, UNCED Principle Three characterized sustainable
development as “the right to development must be fulfilled so as to equitably meet
developmental and environmental needs of present and future generations,” while
Principle Four stated that “in order to achieve sustainable development, environmen-
tal protection shall constitute an integral part of the development process and cannot
be isolated from it.” These principles were reaffirmed at the 2002 Johannesburg
World Summit on Sustainable Development and have since produced a profound
implication for use and stewardship of natural resources, ecology, and environment
(Weng and Yang 2003). In practice, sustainable development is a multifaceted con-
cept and has been viewed from many perspectives, depending on one’s personal
experience, viewpoint, and discipline (Weng and Yang 2003). As we consider sus-
tainable development from an ecological perspective, we must have an appreciation
of ecosystem integrity, which implies the existence of the system structure and func-
tion, maintenance of system components, interactions among them, and the resul-
tant dynamic of the ecosystem (Campbell and Heck 1997). Shaller (1990) suggested
that “sustainable agriculture over the long-term enhances environmental quality and
the resource base upon which agriculture depends, provides for basic human food
and fiber needs, is economically viable, and enhances the quality of life for farmers
and society as a whole.” This definition illustrates well three pillars of sustainable
development: ecological, economic, and social objectives (Weng and Yang 2003).
The ecological objective seeks to preserve the integrity of the ecosystem, while the
economic objective attempts to maximize human welfare within the existing capital
stock and technologies and use economic units (i.e., money or perceived value) as a
measurement standard (Campbell and Heck 1997). The social objective stresses the
needs and desires of people and uses standards of well-being and social empower-
ment. The ecological perspective for sustainable development balances the ecologi-
cal, economic, and societal values, and falls at the intersection of the spheres that
represent the three components (Weng and Yang 2003). Incoordination among the
three components will likely result in failure to achieve sustainability (Zonneveld
and Forman 1990).
The spatial and temporal scales are key elements in assessing ecological and envi-
ronmental sustainability, because they module the objectives of what we wish to
sustain, over what time scale and in which geographical scope (Weng 2014). With
more than one-half of the world population living in cities, the 21st century has
become the first urban century. Cities are human-central ecosystems and are the

xi
xii Preface

most complex of all human settlements. While the economic and societal values
are stressed, the ecological value is often ignored, and therefore, the ecological
perspective of sustainable development asserts the importance of the coordination
and balance among the three objectives for sustainability of cities (Weng and Yang
2003). Because of the nature of cities as dynamic, complex settlements and human-
central ecosystems, the development of evaluation methods for sustainable cities
should reflect these characteristics (Weng and Yang 2003). The US Climate Change
Science Program (CCSP 2008) defines one of its five goals to be “understanding the
sensitivity and adaptability of different natural and managed ecosystems and human
systems to climate and related global changes.” Although vulnerabilities of settle-
ments to impacts of climate change vary regionally, they generally include some or
many of the following impact concerns: health, water and infrastructures, severe
weather events, energy requirements, urban metabolism, sea level rise, economic
competitiveness, opportunities and risks, and social and political structures. CCSP
(2008) further recommends that research on climate change effects on human settle-
ments in the United States be given a much higher priority in order to provide better
metropolitan-area scale decision-making.
Earth observation technology, in conjunction with in situ data collection, has been
used to observe, monitor, measure, and model many of the components that com-
prise natural and human ecosystems cycles (Weng 2012a). Driven by the societal
needs and improvement in sensor technology and image processing techniques, we
have witnessed a great increase in research and development, technology transfer,
and engineering activities worldwide since the turn of the 21st century. Commercial
satellites acquire imagery at spatial resolutions previously only possible to aerial
platforms, but these satellites have advantages over aerial imageries including their
capacity for synoptic coverage, inherently digital format, short revisit time, and capa-
bility to produce stereo image pairs conveniently for high-accuracy 3D mapping
thanks to their flexible pointing mechanism (Weng 2012b). Hyperspectral imag-
ing affords the potential for detailed identification of materials and better estimates
of their abundance in the Earth’s surface, enabling the use of remote sensing data
collection to replace data collection that was formerly limited to laboratory testing
or expensive field surveys (Weng 2012b). While LiDAR technology provides high-
accuracy height and other geometric information for urban structures and vegeta-
tion, radar technology has been re-invented since the 1990s mainly because of the
increase of spaceborne radar programs (Weng 2012b). These technologies are not
isolated at all. In fact, their integrated uses with more established aerial photography
and multispectral remote sensing techniques have been the main stream of current
remote sensing research and applications (Weng 2012b).
With these recent advances, techniques of and data sets from remote sensing and
Earth observation have become an essential tool for understanding the Earth, moni-
toring the world’s natural resources and environments, managing exposures to natu-
ral and man-made risks and disasters, and helping the sustainability and productivity
of natural and human ecosystems (Weng 2012b). This book aims at introducing to
the current state of remote sensing knowledge needed for sustainable development
and management with selected studies. These studies either explore the methods
and techniques of remote sensing for application to various aspects of sustainable
Preface xiii

development or provide important insights into sustainability science from the per-
spective of remote sensing technology. Therefore, this book would be of great value
for students, professors, and researchers in both remote sensing and sustainability
science. It can help narrow the gap between the two disciplines. This book may
also serve as a textbook for undergraduate and graduate students or as an impor-
tant supplement for those majoring in sustainability, remote sensing, geography,
geosciences, planning, environmental science and engineering, civic engineering,
resources science, land use, energy, and geographic information system (GIS). On
college campuses, we have recently witnessed an ever-increasing number of classes
and programs on sustainability or related subjects. Remote sensing is emerging as
an essential geospatial tool in sustainability; this book would meet the need of those
classes and programs. In addition, this book may be used as a reference book for
sustainability officers, practitioners, and professionals alike in the government, com-
mercial, and industrial sectors. Since its contents cover numerous applications of
remote sensing to sustainability, this book, indeed, provides a useful toolbox.

intErnatiOnal COOpEratiOn and COllabOratiOn

On Earth ObSErvatiOn fOr a SuStainablE Earth

The 2002 World Summit on Sustainable Development in Johannesburg highlighted

the urgent need for coordinated observations relating to the state of the Earth. The
First Earth Observation Summit in Washington, DC, in 2003 adopted a declara-
tion to establish the ad hoc intergovernmental Group on Earth Observations (ad hoc
GEO) to draft a 10-Year Implementation Plan. Since 2003, GEO has been work-
ing to strengthen the cooperation and coordination among global observing systems
and research programs for integrated global observations. The GEO process has out-
lined a framework document calling for Global Earth Observation System of Systems
(GEOSS) and has defined nine areas of societal benefits (Group on Earth Observation
2008), including the following:

• Reducing loss of life and property from natural and human-induced

disasters,
• Understanding environmental factors affecting human health and well-being,
• Improving the management of energy resources,
• Understanding, assessing, predicting, mitigating, and adapting to climate
variability and change,
• Improving water resource management through better understanding of the
water cycle,
• Improving weather information, forecasting and warning,
• Improving the management and protection of terrestrial, coastal and marine
ecosystems,
• Supporting sustainable agriculture and combating desertification, and
• Understanding, monitoring and conserving biodiversity.

On September 25, 2015, the United Nations adopted a set of sustainable develop-
ment goals (SDGs), each of which has specific targets to be achieved over the next
xiv Preface

15 years (United Nations Development Programme 2015). These goals represent

the United Nation’s responses to numerous societal challenges and the efforts to
build a sustainable Earth. Through large-scale, repetitive acquisition of the Earth
surface image data, remote sensing can provide essential information and knowl-
edge to supplement statistical analyses in the assessment of indicators toward the
attainment of the SDGs. Because Earth observation offers an indispensable tool to
measure and monitor the progress toward SDGs, in the recently developed “GEO
Strategic Plan 2016–2025: Implementing GEOSS,” GEO is determined to develop a
concerted direction with the SDGs (Group on Earth Observations 2015). The GEO
Global Urban Observation and Information Initiative (GI-17) has set the following
goals for the period 2012–2015: (1) improving the coordination of urban observa-
tions, monitoring, forecasting, and assessment initiatives worldwide; (2) supporting
the development of a global urban observation and analysis system; (3) producing
up-to-date information on the status and development of the urban system—from
a local to a global scale; (4) filling existing gaps in the integration of global urban
observation with data that characterize urban ecosystems, environment, air qual-
ity and carbon emission, indicators of population density, environmental quality,
quality of life, and the patterns of human environmental and infectious diseases;
and (5) developing innovative techniques in support of effective and sustainable
urban development. These goals will be implemented by developing and expanding
selected activities and programs that the GEO Global Urban Observation (SB-04)
task team has been working (Weng et al. 2014). This book intends to contribute
to the GEO’s Strategic Plan by addressing and exemplifying a number of societal
benefit areas of remote sensing data sets, methods, and techniques for sustainable
development.

SynOpSiS Of thE bOOk

This book consists of four sections. Section I deals with remote sensing for sustain-
able cities; Section II discusses remote sensing techniques and methods for forest
resources; Section III presents remote sensing studies for sustainable environmental
systems, with topics ranging from air, water, to land; and Section IV includes various
contributions in remote sensing of sustainable energy systems.
Section I includes five chapters dealing with theories and methods as well as
practical applications of sustainable development for cities using remote sensing.
Chapter 1 provides key concepts and principles for assessing sustainability of cit-
ies. It then discusses the typical parameters derivable from remotely sensing imag-
ery that can be used to define indicators for sustainable cities. Chapter 2 introduces
a selection of applications and data products that provide support for day-to-day
decision-making activities in urban and regional planning. Chapters 3 through 5
present various studies in which remotely sensed data, methods, and techniques
are used for studying cities or urban clusters. In Chapter 3, Zhang and Weng moni-
tor urban growth process in the Pearl River Delta, Guangdong Province, China, by
using time series Landsat imagery from 1987 to 2014. This chapter demonstrates
the effectiveness of time series data mining for assessing urban growth pattern
over a long period and the usefulness of generated data set and information for
Preface xv

exploring the relationship between urban growth and environmental sustainabil-

ity. In Chapter 4, the damage of subsidence on urban structures in the St. Louis
Metropolitan area, Missouri, is explored by using Synthetic Aperture Radar images
in the period from 1992 to 2011. Their results show hot spots of ongoing and poten-
tial land collapses in the region, which are valuable not only to individual home-
owners but also to city planners, insurance companies, and regional policymakers
charged with assessing risks of abandoned coal mines. In the last chapter in the
section, Chapter 5, Xiao and Weng examine urban growth in the North Carolina
urban crescent from the Research Triangle Park (Raleigh–Durham–Chapel Hill)
to Greensboro, and Guiyang City to Anshun City in Guizhou Province, China,
from the 1980s to the 2010s, when accelerated industrialization and urbanization
occurred. They compare and contrast the spatial patterns, paths, and driving forces
of urbanization in the two regions and countries of different socioeconomic devel-
opment stages.
Section II focuses on remote sensing methods and techniques for sustainable nat-
ural resources. In Chapter 6, Tong et al. present a few case studies on the application
of remote sensing in grassland management for a mixed-grass prairie ecosystem in
North America; on the basis of the case studies, they further discuss challenges and
opportunities for remote sensing in grassland management. Chapter 7 assesses the
relationship between species extinction and biodiversity loss using the example of
palila (Loxioides bailleui), an endangered bird species on the island of Hawaii. To
understand its population trend, tree species in its habitat were identified analyzed
with high spatial resolution satellite imagery at both pixel and object levels. Chapter
8 takes this direction further along biodiversity and conservation by examining how
evolved remote sensing techniques can be employed to investigate forest damages
by diseases and insects. The last chapter in Section II, Chapter 9, focuses on water
resources. Matsushita and his colleagues first survey major satellite sensors for water
quality studies, followed by a discussion of representative algorithms for water area
delineation, atmospheric correction, and water quality parameter estimation. This
chapter ends with a proposed framework for water quality assessment using remote
sensing.
What remote sensing methods and techniques can do for the sustainability of
environmental systems and how they do it are addressed in Section III. In Chapter
10, Hu explores the use of satellite remote sensing data to expand ground network
of PM2.5 observation by using aerosol optical depth (AOD). Various AOD prod-
ucts and methods that are widely used in PM2.5 concentration estimation were
assessed. This discussion was followed by a case study in the Atlanta metropolitan
area, Georgia, to estimate ground-level PM2.5 concentration from MODIS AOD.
Chapter 11 continues the discussion on public health but from the perspective of
heat hazards. Jiang and Weng develop methods to analyze daily and hourly varia-
tions of land surface temperature (LST), which were derived from remotely sensed
thermal infrared image data, and discuss the impact of evapotranspiration on LST
over a variety of urban surfaces. Both Chapters 12 and 13 study ecosystems in
semi-arid and arid regions but with distinct approaches of remote sensing. The for-
mer investigates the tendency of desertification along the Mediterranean to arid
transition zones in central Israel, and the latter assesses soil moisture condition in
xvi Preface

the Umer Kot region of Pakistan. In Chapter 12, Shoshany develops three indica-
tors using remote sensing data, that is, Green Vegetation Cover, life-forms’ sub-
pixel compositions, and spatial recovery versus erosion potential, and by assessing
these indicators, he suggests that no significant shift occurred in the transition zone
between 1996 and 2011, but that there was high vulnerability to future degrada-
tion. Soil moisture holds the key to drought and agricultural monitoring in semi-
arid and arid regions. Chapter 13 investigates the potential of near-infrared and red
reflectance space (i.e., the RSSMM method) and temperature vegetation dryness
index (TVDI) for the assessment of soil moisture. Results of the remote sensing
methods were compared with in situ soil moisture measurements at different land
surface depths of 0–15, 15–30, and 30–45 cm. RSSMM was found satisfactory in
determining soil moisture, but TVDI provided a more reliable estimate of moisture
condition.
The last section of this book, Section IV, examines the issues of energy use and
sustainable energy sources using remote sensing technology. The development of
renewable energies is one of the key challenges in the 21st century. In this context, it
is of central importance to focus on the review of surface potentials, the determina-
tion of suitable sites, the consideration of user interests, and the detection of trends
and impacts on the landscape. To meet these requirements and tasks, timely, spatial,
and thematic high-quality geospatial data are indispensable. In Chapter 14, Esch
et al. exemplify applications of remote sensing and related technologies and geo-
information products for land management in Germany, which is geared toward the
development of potentials of renewable energies. Chapter 15 assesses the capability
of DMSP-OLS nighttime light imagery for analyzing the decadal trends of energy
consumption (EC) in China, from 2000 to 2012. Here, Xie and Weng demonstrate
a moderate to rapid growth of EC for coastal and capital cities, but a slow growth
for the majority of central, northeastern, and western cities. They further find the
total and urban EC at the prefectural level to be regionally clustered, which may
have an implication in future Chinese energy policy and the spatial distribution of
EC. The last two chapters examine renewable energy sources. While Chapter 16
focuses on wind energy, Chapter 17 evaluates solar energy potential. Wind speed
and wind flow are strongly influenced by land surface properties. Three different
remote sensing–based parameters can help characterize wind resources: (1) land
cover and land use, (2) digital elevation models, and (3) phenological information.
In Chapter 16, Esch et al. discuss how Earth Observation (EO) data can be used
to support wind resource modeling, especially the possibilities brought about by
the Copernicus Sentinel satellites. They conclude that by using EO-based infor-
mation on the surface (e.g., roughness) and in situ wind measurements, realistic
wind fields for sufficiently large areas can be derived by considering shadowing
effects and wind shear as well. Chapter 17 provides two case studies to demonstrate
the applicability of remote sensing techniques on sustainable energy development
in Indianapolis, Indiana. The first case study demonstrates a method to estimate
the solar energy potential of building roofs, and the second case study examines
the correlation between energy use and building morphological attributes such as
ground area, total floor area, height, surface area, compactness, aspect ratio, and
orientation.
Preface xvii

REFERENCES
Group on Earth Observation, 2008. GEO Information Kit. http://www.earthobservations.org
/documents/geo_information_kit.pdf.
Group on Earth Observations, 2015. The GEO 2016–2025 Strategic Plan: Implementing
GEOSS, https://www.earthobservations.org/geoss_wp.php, last accessed March 5,
2016.
United Nations Development Programme, 2015. Sustainable Development Goals (SDGs),
http://www.undp.org/content/undp/en/home/sdgoverview/post-2015-development
-agenda.html, last accessed March 5, 2016.
United Nations Environment Programme, 1992. Rio Declaration on Environment and
Development. United Nations publication, Sales No. E.73.II.A.14 and corrigendum,
Chap. I. http://www.unep.org/documents.multilingual/default.asp?documentid=78&art
icleid=1163.
U.S. Climate Change Science Program (CCSP), 2008. Analyses of the Effects of Global
Change on Human Health, Welfare, and Human Systems. A Report by the U.S. Climate
Change Science Program and the Subcommittee on Global Change Research. [Gamble,
J.L. (ed.), K.L. Ebi, F.G. Sussman, T.J. Wilbanks]. U.S. Environmental Protection
Agency, Washington, DC.
Weng, Q. 2012a. Remote sensing of impervious surfaces in the urban areas: Requirements,
methods, and trends. Remote Sensing of Environment, 117(2), 34–49.
Weng, Q. 2012b. An Introduction to Contemporary Remote Sensing. New York: McGraw-Hill
Professional, 320 pp.
Weng, Q. 2014. Scale Issues in Remote Sensing. Hoboken, NJ: John Wiley & Sons, Inc.,
352 pp.
Weng, Q., Esch, T., Gamba, P., Quattrochi, D.A. and Xian, G. 2014. Global urban observation
and information: GEO’s effort to address the impacts of human settlements. In Weng,
Q. (ed.). Global Urban Monitoring and Assessment through Earth Observation. Boca
Raton, FL: CRC Press/Taylor & Francis, Chapter 2, pp. 15–34.
Acknowledgments
I thank all the contributors for making this endeavor possible. Furthermore, I offer
my deepest appreciation to all the reviewers, who have taken precious time from their
busy schedules to review the chapters submitted to this book. Finally, I am indebted
to my family for their enduring love and support. It is my hope that the publication of
this book will provide inspiration to students, researchers, and practitioners to con-
duct more in-depth studies on remote sensing for sustainability. It becomes apparent
that the realization of many societal benefits of Earth Observation technology and
sustaining our Earth requires international collaboration and cooperation.
The reviewers of the chapters for this book are listed below in alphabetical order:
Wanda De Keersmaecker, Chengbin Deng, Pinliang Dong, Paolo Gamba, Feng Gao,
Anatoly Gitelson, Peter Hofmann, Husi Letu, Yangfan Li, Lu Liang, Dengsheng
Lu, Ryo Michishita, Zina Mitraka, Hiroyuki Miyazaki, Daniel Odermatt, Dale A.
Quattrochi, Conghe Song, Hannes Taubenboeck, Cuizhen Wang, Guangxing Wang,
Charles Wong, Changshan Wu, George Xian, Klemen Zaksek, Hui Zeng, Caiyun
Zhang, Qingling Zhang, Xiaoyang Zhang, Wei Zhuang, and Bin Zou.

Qihao Weng

xix
Editor
Dr. Qihao Weng is the director of the Center for Urban and Environmental Change
and a professor of geography at Indiana State University. In addition, he serves as
editor-in-chief of ISPRS Journal of Photogrammetry and Remote Sensing and is
the series editor for Taylor & Francis Series in Remote Sensing Applications. He
received his PhD degree in geography from the University of Georgia in 1999. In the
same year, he joined the University of Alabama as an assistant professor. Since 2001,
he has been a member of the faculty in the Department of Earth and Environmental
Systems at Indiana State University, where he has taught courses on remote sensing,
digital image processing, remote sensing–GIS integration, GIS, and environmental
modeling. He has mentored 15 doctoral and 13 master students.
Dr. Weng’s research focuses on remote sensing and GIS analysis of urban eco-
logical and environmental systems, land-use and land-cover change, environmental
modeling, urbanization impacts and sustainability, and human–environment inter-
actions. He is the author of more than 180 peer-reviewed journal articles and other
publications and 9 books. Dr. Weng has worked extensively with optical and thermal
infrared remote sensing data, and more recently with LiDAR data, primarily for
urban heat island study, land-cover and impervious surface mapping, urban growth
detection, image analysis algorithms, and the integration with socioeconomic charac-
teristics, with financial support from US funding agencies that include NSF, NASA,
USGS, USAID, NOAA, the National Geographic Society, and Indiana Department
of Natural Resources. From the American Association of Geographers (AAG),
Dr. Weng has received the Outstanding Contributions Award in Remote Sensing in
2011 and the Willard and Ruby S. Miller Award in 2015 for his outstanding contribu-
tions to geography. In May 2008, he received a prestigious NASA senior fellowship.
Dr. Weng is also the recipient of the Robert E. Altenhofen Memorial Scholarship
Award by the American Society for Photogrammetry and Remote Sensing (1999),
the Best Student-Authored Paper Award by the International Geographic Information
Foundation (1998), and the 2010 Erdas Award for Best Scientific Paper in Remote
Sensing by ASPRS (first place). At Indiana State, he received the Theodore Dreiser
Distinguished Research Award in 2006 and was selected as a Lilly Foundation
Faculty Fellow in 2005. Dr. Weng has given more than 90 invited talks (includ-
ing keynote addresses, public speeches, colloquia, and seminars) and has presented
more than 100 papers at professional conferences (including co-presenting).
Dr. Weng is currently the coordinator for GEO’s GI-17, Global Urban Observation
and Information Task (2016–2019). He has been the organizer and program com-
mittee chair of the biennial IEEE/ISPRS/GEO-sponsored International Workshop
on Earth Observation and Remote Sensing Applications conference series since
2008. He previously held the following positions: national director of the American
Society for Photogrammetry and Remote Sensing (2007–2010), chair of AAG China
Geography Specialty Group (2010–2011), secretary of the ISPRS Working Group
VIII/1 (2004–2008), and panel member of US DOE’s Cool Roofs Roadmap and
Strategy in 2010.

xxi
Contributors
Iftikhar Ali Takehiko Fukushima
Vienna University of Technology Faculty of Life and Environmental
(Tu Wien) Sciences
Department of Geodesy and University of Tsukuba
Geoinformation (GEO) Tsukuba, Ibaraki, Japan
Vienna, Austria
Abduwasit Ghulam
Gang Chen Center for Sustainability
Department of Geography and Earth Saint Louis University
Sciences St. Louis, Missouri
and
Laboratory for Remote Sensing and Mark Grzovic
Environmental Change (LRSEC) Center for Sustainability
University of North Carolina at and
Charlotte Department of Earth and Atmospheric
Charlotte, North Carolina Sciences
Saint Louis University
Qi Chen St. Louis, Missouri
Department of Geography
University of Hawaii at Mānoa Xulin Guo
Honolulu, Hawaii University of Saskatchewan
Saskatoon, Saskatchewan, Canada
Janik Deutscher
Joanneum Research Charlotte Bay Hasager
Institute for Information and Technical University of Denmark
Communication Technologies Department of Wind Energy
Graz, Austria Risø Campus
Roskilde, Denmark
Paul Elsner
Birkbeck Yuhong He
University of London University of Toronto Mississauga
Department of Geography, Environment Mississauga, Ontario, Canada
and Development Studies
London, United Kingdom Manuela Hirschmugl
Joanneum Research
Thomas Esch Institute for Information and
German Aerospace Center (DLR) Communication Technologies
German Remote Sensing Data Center Graz, Austria
(DFD)
Weßling, Germany

xxiii
xxiv Contributors

Xuefei Hu Ross K. Meentemeyer

Department of Environmental Health Center for Geospatial Analytics
Rollins School of Public Health and
Emory University, Atlanta, Georgia Department of Forestry and
Environmental Resources
Lalu Muhamad Jaelani North Carolina State University
Geomatics Engineering Department Raleigh, North Carolina
Institut Teknologi Sepuluh Nopember
Surabaya, Indonesia Annekatrin Metz
German Aerospace Center (DLR)
Yitong Jiang German Remote Sensing Data Center
Center for Urban and Environmental (DFD)
Change Weßling, Germany
Department of Earth and
Environmental Systems Siraj Munir
Indiana State University Pakistan Space and Upper Atmosphere
Terre Haute, Indiana Research Commission
Islamabad, Pakistan
Bing Lu
University of Toronto Mississauga and
Mississauga, Ontario, Canada
Department of Remote Sensing and GIS
Maitiniyazi Maimaitijiang Institute of Space Technology
Center for Sustainability Karachi, Pakistan
Saint Louis University
St. Louis, Missouri Claudia Notarnicola
EURAC Research
and Institute for Applied Remote Sensing
Bolzano, Italy
College of Management
Xinjiang Agricultural University Salman Qureshi
Urumqi, Xinjiang, China School of Architecture
Birmingham City University
Mattia Marconcini Birmingham, United Kingdom
German Aerospace Center (DLR)
German Remote Sensing Data Center and
(DFD)
Weßling, Germany Department of Geography
Humboldt University of Berlin
Bunkei Matsushita Berlin, Germany
Faculty of Life and Environmental
Sciences
University of Tsukuba
Tsukuba, Ibaraki, Japan
Contributors xxv

Said Rahman Alexander Tong

Pakistan Space and Upper Atmosphere University of Toronto Mississauga
Research Commission Mississauga, Ontario, Canada
Regional Office
Peshawar, Pakistan John C. Trinder
School of Civil and Environmental
and Engineering
The University of New South Wales
Department of Remote Sensing and GIS Sydney, New South Wales, Australia
Institute of Space Technology
Karachi, Pakistan Markus Tum
German Aerospace Center (DLR)
Achim Roth German Remote Sensing Data Center
German Aerospace Center (DLR) (DFD)
German Remote Sensing Data Center Weßling, Germany
(DFD)
Weßling, Germany Mudassar Umar
Pakistan Space and Upper Atmosphere
Mamat Sawut Research Commission
Center for Sustainability and
Saint Louis University Department of Remote
St. Louis, Missouri Sensing Commission and GSIS
Institute of Space Technology
and Karachi, Pakistan

College of Resources and Qihao Weng

Environmental Sciences Center for Urban and Environmental
Xinjiang University Change
Urumqi, Xinjiang, China Department of Earth and
Environmental Systems
Fajar Setiawan Indiana State University
Graduate School of Life and Terre Haute, Indiana
Environmental Sciences
University of Tsukuba Honglin Xiao
Tsukuba, Ibaraki, Japan History and Geography
Elon University
Maxim Shoshany Elon, North Carolina
Mapping and GeoInformation
Engineering
Faculty of Civil and Environmental
Engineering
Technion, Israel Institute of Technology
Haifa, Israel
xxvi Contributors

Yanhua Xie Lei Zhang

Center for Urban and Environmental Center for Urban and Environmental
Change Change
Department of Earth and Department of Earth and
Environmental Systems Environmental Systems
Indiana State University Indiana State University
Terre Haute, Indiana Terre Haute, Indiana

Wei Yang and

Center for Environmental
Remote Sensing The State Key Laboratory of
Chiba University Information Engineering in
Chiba, Japan Surveying, Mapping and Remote
Sensing
Julian Zeidler Wuhan University
German Aerospace Center (DLR) Wuhan, People’s Republic of China
German Remote Sensing Data Center
(DFD) Yuanfan Zheng
Weßling, Germany Center for Urban and
Environmental Change
Department of Earth and
Environmental Systems
Indiana State University
Terre Haute, Indiana
Section I
Remote Sensing
for Sustainable Cities
 1 Extraction of Parameters
from Remote Sensing
Data for Environmental
Indices for Urban
Sustainability
John C. Trinder

CONTENTS
1.1 Introduction ......................................................................................................3
1.2 Toward a Sustainable Community .................................................................... 5
1.3 Sustainability Indicators ................................................................................... 6
1.3.1 Defining Indicators ............................................................................... 6
1.3.2 Application of SIs ................................................................................. 8
1.4 Indicators for Urban Sustainability ................................................................ 11
1.5 Application of Remote Sensing for Urban Sustainability .............................. 13
1.5.1 Available Remotely Sensed Data ........................................................ 13
1.5.2 Urban Sustainability Measures Derived by Remote Sensing ............. 16
1.5.2.1 Measurement of Impervious Surfaces in Relation
to Open Space ...................................................................... 18
1.5.2.2 Growth in Fragmentation of Open Space versus
Impervious Surfaces ............................................................ 19
1.5.2.3 Mapping and Analysis of Sustainable Transport Systems ....20
1.5.2.4 Compactness of Cities and Mapping of Township Layout .... 21
1.5.2.5 Deforestation and Changes in Land Cover .......................... 22
1.5.2.6 Biodiversity .......................................................................... 23
1.6 Conclusions ..................................................................................................... 23
References ................................................................................................................24

1.1 INTRODUCTION
It has been recognized over the past few decades that actions by humans have modi-
fied and altered the energy and mass exchanges that occur between the atmosphere,
oceans, and biota, and researchers now understand that the changes being wrought
on the planet could be beyond the resilience of natural systems to absorb. The con-
sequence of these changes can also be a loss or a severe decline in the ecosystem

3
4 Remote Sensing for Sustainability

services on which we rely, thus affecting our quality of life. McGlade (2007) referred
to even more serious consequences of a green backlash, where dramatic shifts in the
structure and behavior of ecosystems can occur without warning.
Sustainable development has been proposed as a means of ensuring that human
impacts are within the capacity of the Earth’s environment* to cope with changes.
While there have been many definitions presented, sustainability refers to the adop-
tion of practices in relation to environmental use and management that provide a
satisfactory standard of living for today’s population and that do not impair the
capacity of the environment to provide for and support the needs of future genera-
tions. Alternatively, sustainable development is that which meets the needs of society
today without foreclosing the needs or options of the future (Blanco et al. 2001;
Mahi 2001). The concept of sustainability in respect of the use of the environment’s
resources includes the notion that the outputs derived, whether they are from land,
water, or air, can be produced continuously over time, and that a balance can be
achieved between the rate of economic growth, the use of resources, and environ-
mental quality, thus minimizing the risk of long-term environmental degradation.
Sustainable practice is one which is sensitive to ecological constraints and seeks to
minimize the undesirable effects of exploitation and use, and which might negatively
affect the longer-term viability of a resource. It is also one in which the full economic
and environmental replacement costs associated with the use of a resource should
be met. Turner (1993) described strong versus weak sustainability, which is based on
“the economic concept of capital, defined as a stock of resources with the capacity
to give rise to the flow of goods and services.” Strong sustainability requires “the
stock of natural capital to be maintained above critical levels” (Turner 1993). Weak
sustainability presumes that the “total capital stock does not decline” and all types of
capital are substitutable. Karlsson et al. (2007) indicated that where current policies
and actions are heading in regard to human welfare, on the scale of weak to strong
sustainability, can be mapped out provided they are formulated in monetary terms.
Brandon and Lombardi (2011) referred to community capital that contributes to
humans’ well-being, which includes built and financial, human and social, and natu-
ral capital, each being measured in different ways. When they are out of balance,
chaos or disaster can occur.
Kates (2000) reviewed the relationship between population and consumption in
terms of the formula I = P × C, where I is environmental degradation or resource
depletion, P is the number of people or households, and C is the transformation of
energy, materials, and information. This simple formula shows that as population
increases, resource depletion also increases. In addition, the value of C must be con-
trolled according to an optimal level of transformation of natural capital. Therefore,
as populations increase, or a certain section of the population wishes to increase its
use of resources to the same level as other groups in a society, since natural capital
available for transformation effectively remains constant, there will be fewer natu-
ral resources available per capita and a redistribution of resources will be required.

* In the context of this chapter, environment refers to the surroundings or conditions in which a person,
animal, or plant lives or operates, while ecology is the branch of biology that deals with the relations
of organisms to one another and to their physical surroundings.
Extraction of Parameters from Remote Sensing Data 5

The maintenance of sustainability therefore becomes even more difficult and yet
more critical.
Sustainable development cannot be divorced from issues of equity, welfare,
lifestyle, and the expectation of improved standards of living in most countries.
The Principles of the 1992 Rio Declaration, which were reaffirmed at the 2002
Johannesburg World Summit on Sustainable Development, define the roles of the
stakeholders in sustainable development, and rights and responsibilities in devel-
opment processes. The Johannesburg Declaration went on to refer to “…the three
components of sustainable development, economic development, social development
and environmental protection as interdependent and mutually reinforcing pillars.”

1.2 TOWARD A SUSTAINABLE COMMUNITY

There have been many papers proposing approaches to achieving sustainability in
nations, in regions, and by individuals. Gallopin and Raskin (2002) compared a
number of scenarios that have been used or may be used in the future for predict-
ing the characteristics of a sustainable human society. They include market forces,
policy reform, eco-communalism, through to muddling through. The global com-
munity is currently dominated by market forces in which there is often an absence
of controls over development and therefore there are tensions between development
and sustainability goals. The policy reform approach of Gallopin and Raskin (2002)
was based on the assumption of consensus and strong political will to achieve a
sustainable future. Radermaker (2004) compared the impact of several approaches
to economic and political developments and concluded that a balanced philosophy
must be based on the concept of a global ecosocial market, consensus, and respect
for civil rights and human equity, where human behavior is agreed globally by social
contract. It means that there needs to be a consensus on protection of resources and
respecting the need for all humans to have an adequate quality of life with access
to essential resources. It is a long-term view of how the global population should
cooperate to secure environmentally sound developments, but one which he believed
is essential to achieve a sustainable and equitable use of resources. Azapagic and
Perdan (2005a,b) have presented a procedure for decision-making that includes all
stakeholders in a development process defining their preferences and choosing the
most suitable alternative for the development, implementation of the chosen alterna-
tive, and assessment of the outcomes. The decision-making process was based on the
concept of Multiple Criteria Decision Analysis, which was described in some detail.
In order to assess the sustainability of a society and its consequent well-being,
organizations in many countries, regional organizations, nongovernmental organiza-
tions, and private organizations have attempted to develop sustainability indicators
(SIs). Hundreds of indicators have been produced under the three pillars of eco-
nomic, social, and environmental. A fourth pillar, institutional, has also recently
been added. In judging the importance of the three original pillars, Jesinghaus
(2007) showed that economy is the most important pillar, with a weight of 45%,
while social has a weighting of 35% and environment has a weighting of 20%. If
these weights of importance lead to a degradation of the environment, then it seems
a greater weight may be required for the environment.
6 Remote Sensing for Sustainability

While many of the concepts referred to above may be far removed from the
practical implementation of remote sensing technologies, it can be argued that remote
sensing and geographic information system (GIS) technologies have important roles
to play in assisting in the understanding of the physical impacts of development
and, therefore, by inference, their impacts on human well-being. Also, by virtue of
information extracted from the analysis of geospatial data, unsustainable practices
being undertaken may be identified and their likely consequences may be predicted
if they are continued. Such analyses must involve experts in remote sensing and GIS,
as well as those in ecology, biology, sociology, human resources, and politics.
A description of sustainability indicators (SIs) will be given Section 1.3, followed
by those for urban environments and a demonstration of how remote sensing
technologies can be incorporated for the practical assessment of sustainability of urban
areas. This chapter will concentrate on applications of remote sensing technologies
for assessing urban sustainability as well as land-use practices outside urban areas,
since urban dwellers depend on land in the vicinity of urban environments for much
of their well-being.

1.3 SUSTAINABILITY INDICATORS

1.3.1 Defining inDicators
Becker (1997) defined the approaches that can be taken for assessing sustainability
using “an exact measurement of single factors and their combination into meaningful
parameters” and indicators “as an expression of complex situations by a variable that
compresses information into a more readily understandable form” (Harrington et al.
1993). Tanguay et al. (2010) stated that an observed datum or variable becomes an
indicator, only when its role in the evaluation of a phenomenon has been established.
An index (or composite indicator) is a synthesis of indicators. Moldan and Dahl
(2007) stated that “Indicators of sustainability should measure characteristics of the
human-environmental system that ensure its continuity and functionality far into
the future.”
Hák et al. (2007) provided a detailed scientific assessment of SIs, in the treatise
implemented by SCOPE and UNEP, together with the IHDP and EEA, and sponsored
by the ICSU.* This is a comprehensive coverage of SI from the perspective of the
three pillars and includes methodological aspects, system and sectorial approaches,
and case studies. A great deal can be learned from this volume about SIs. Rao (1998)
stated that SIs are designed to monitor progress and assess the effectiveness and
impact of policies on natural resource development. Becker (1998) reported on the
proliferation of papers and recommendations on developing SIs since the UN Rio
Earth Summit in 1992. The UN, Organisation for Economic Co-operation and
Development (OECD), the World Bank, and many other organizations have devel-
oped sets of indicators. OECD countries use 23 so-called indices based on natural

* SCOPE, Scientific Committee on Problems of the Environment; UNEP, UN Environmental

Programme; IHDP, International Human Dimensions Programme on Global Environmental Change;
EEA, European Environmental Agency; ICSU, International Council for Science.
Extraction of Parameters from Remote Sensing Data 7

sciences, policy performance, accounting framework, and synoptic indices. There

are also aggregate indices that cover countries or regions.
Petrosyan (2014) developed a new composite indicator, composite appraising
supportive progress (CASP), derived from 12 other indicators, with the weighting
based on the number of papers published on each indicator. While there are advan-
tages to having a single measure of sustainability, there are considerable questions
about the weights used to produce composite indices, which show little scientific
justification.
Moldan and Dahl (2007) referred to the different time scales of indicators for the
three pillars. Economic indicators will normally have a short-term effect, while the
effects of environment indicators will be longer term. In addition, they suggested
that SIs might be more easily understood if they are formed into frameworks, based
on a hierarchy of subdomains, with the three pillars being the basis for one such
framework. Brandon and Lombardi (2011) stated that frameworks have been devel-
oped in order to link indicators to policy processes and also for developing messages
to decision-makers.
Olalla-Tárraga (2006) claimed that a reductionist approach in which the three
areas, economic, social, and environmental, are separated has failed to provide a sat-
isfactory set of SIs that can be practically implemented. He presented eight concep-
tual frameworks that have been published by various authors, namely, domain-based,
issue-based, goal-based, sectoral, causal, comparative, ecosystemic, and combina-
tions. His solution was the hierarchical concept shown in Figure 1.1, in which the
three areas (economic, social, and environmental) are each subdivided into area,
objective, attribute, and indicators. An attempt will be made to relate environmental

Dimension Area Objective Attribute

Economy
Indicators

Sustainable
development Environment

Social

FIGURE 1.1 Hierarchical framework of sustainable indicator system. (From Olalla-

Tárraga M. A. 2006. A conceptual framework for assessing urban ecological systems, Int. J.
Sustainable Dev. World Ecol. 13: 1–15. With permission.)
8 Remote Sensing for Sustainability

indicators for urban areas to such a framework later in this chapter. Links may exist
between the different criteria. This is a complex approach, but one which may be
developed further in the future.
Olalla-Tárraga (2006) has also commented that a problem with SIs is the limited
amount of data available to compile them. Becker (2007) provided a general dis-
course on frameworks for SIs that should lead to a more holistic approach to their
implementation. She stated that limited tools are currently available for assessing
the nonlinear multidimensional problem of sustainable development, but frame-
works can assist in understanding the issues. Gutiérrez-Espeleta (2007) has also
raised the issue that a new approach is necessary in the development of SIs. He
has proposed a multilevel typology in which the indicators for Environment and
Society can be classified into five generations. The first and second generations
would measure a single characteristic of an environmental (or societal) issue, while
the third, fourth, and fifth generations would be integrated measures. He has cited
two indices—the Environmental Vulnerability Index (EVI) and Environmental
Sustainability Index (ESI)—that have been developed as high-order indices and
show promise. There are perhaps synergies with this approach and frameworks, but
they have been developed along different lines. A major issue in demonstrating the
usefulness of indices is the availability of data to test the use of indices (Brandon
and Lombardi 2011). In summary, while frameworks have not yet resulted in a use-
ful set of SI, the concept is believed by a number of researchers to lead to a better
approach in the future.

1.3.2 application of sis

Becker (1997) listed the complex criteria as shown in Table 1.1 for developing SIs
that have been compiled with reference to the work of a number of scientists. These
are very relevant to developing SIs for the physical aspects of the environment that
can be determined by remote sensing and that should be suitable for contributions
to a framework approach to SIs in the future. Becker (1997) has shown examples
of environmental indicators for agriculture that could include such items as “yield
trends, coefficients for limited resources, material and energy flows and balances,
soil health, modeling and bioindicators.” Indicators used in practice are usually
application specific but are expected to be unbiased, sensitive to changes, and
convenient to communicate and collect.
Stevenson and Li (2001) referred to the concepts of pressure, state, and response
developed by other authors in discussing the development of SIs applicable for agri-
culture. Pressure refers to human activity, such as in farming; state refers to “the
state of the environment and resources,” for example, water quality and soil erosion;
and response refers to economic and environment agents, such as policy makers. As
an example, they stated that indicators are a simplification of the impact of agricul-
ture on the environment and that simple measures such as indicators are inadequate,
without setting targets that represent levels of sustainability of agricultural activities.
They argued that SIs must provide information on which to base decisions at the
local or regional scale and must be a guiding framework for practitioners. They then
derived a methodology for a meaningful set of indicators for agriculture that includes
TABLE 1.1
Criteria for the Selection of SIs Adapted for Their Relevance to Remote Sensing
1. Scientific Quality 2. Ecosystem Relevance 3. Data Management 4. Sustainability Paradigm
1.1 Indicator really measures what it is 2.1 Changes as the system moves away 3.1 Cost effective 4.1 What is to be sustainable?
supposed to detect from equilibrium 3.2 Data available 4.2 Participatory definition
1.2 Indicator measures significant 2.2 Distinguishes agro-systems moving 3.3 Quantifiable 4.3 Adequate rating of single
aspect away from sustainability 3.4 Representative aspects
1.3 Problem specific 2.3 Identifies key factors leading to 3.5 Transparent 4.4 Resource efficient
1.4 Distinguishes between causes and unsustainability 3.6 Geographically relevant 4.5 Carry capacity
effects 2.4 Warning of irreversible processes 3.7 Relevant to users 4.6 Health protection
1.5 Can be reproduced and repeated 2.5 Proactive in forecasting future trends 3.8 User friendly 4.7 Target values
over time 2.6 Covers full cycles through time 3.9 Widely accepted 4.8 Time horizon
1.6 Uncorrelated, independent 2.7 Corresponds to aggregation level 3.10 Easy to measure 4.9 Social welfare
1.7 Unambiguous 2.8 Highlights links to other system 3.11 Easy to document 4.10 Equity
Extraction of Parameters from Remote Sensing Data

levels 3.12 Easy to interpret

2.9 Permits trade-off detection and 3.13 Comparable across borders over
assessment between system time
components and levels
2.10 Can be related to other indicators

Source: Becker B. 1997. Sustainability assessment: A review of values, concepts, and methodological approaches, Issues in Agriculture 10, CGIAR, Washington, DC,
The World Bank. With permission.
9
10 Remote Sensing for Sustainability

all stakeholders and goal setting. No actual values of indicators were given, which
demonstrates that the concept of SI is only progressing slowly, and in many cases,
no assessment of the sustainability of human activity is yet possible. In the future,
values will be required for the indicators; otherwise, assessment of sustainability
will not be achievable.
Hueting and Reijnders (2004) viewed the issue of SIs from an economic perspec-
tive, referring to Sustainable National Income (SNI) as the goal of countries. They
also argued that physical aspects of the environment must be included in measures
of sustainability and gave the example of the cod stocks in Canada in which physi-
cal aspects were inadequately considered before the industry collapsed. Hart (1999)
demonstrated a new approach in SIs that included the interconnections between the
three pillars of sustainability: social, environmental, and economic. She has stated
that the indicators should be relevant, understandable, reliable, and timely, and has
described a large number of indicators that cover many aspects of human activity,
including production, energy, transport, education, health, recreation, ecosystem,
land use, resource use, and many more. Most of these indicators are based on such
measures as percentage of land covered by impervious materials, volumes of water
used, harvest rates compared with growth rates, and so on. Indicators for sustainabil-
ity of land practices include sustainable land use, sustainable resource use, amount
of tree cover, sustained returns, impact on soil or water, diversity of products, and
conservation of native habitats.
Böhringer and Jochem (2007) reviewed 11 indices and commented that they did
not cover all factors related to sustainability. They remarked on the arbitrariness
of the weighting systems in compiling the indicators. A comprehensive set of
Biodiversity Indicators has been presented by Biggs et al. (2007) covering such
topics as gene-level, species-based, population abundance, ecosystem-level, and
composite indicators. They also provided a list of single-variable indicators in use
that are related to a reference value. Many of these indicators should be measurable
by earth observation technologies.
Van Woerden et al. (2007) described a set of indicators related to the Global
Environmental Outlook covering indicators related to the atmosphere, land, and
water. Barrios and Kimoto (2006) used an analytical approach, based on principal
components and sparse principal component analyses of a number of core indicators
of sustainable development for the Philippines for deriving relevant indicators.
In Phillis and Andriantiatsaholiniaina (2001) and Andriantiatsaholiniaina
et al. (2004), fuzzy logic was used to determine a combined measure of sustain-
ability and sensitivity to various indicators. They argued that the fragmentation of
information about sustainability measures makes it difficult to assess the effects
of developments on the environment. Hence, they attempted to use a systematic
rule-based system that combined many measures of sustainability. Since these
measures are never well defined, they have used fuzzy definitions of the measures,
expressed verbally in such terms as good, bad, medium, and so on. A sensitivity
value was determined for the factors that contribute to a lack of sustainability.
The research concluded that there is no unique sustainable path, and hence poli-
cies need to be chosen to determine the most sustainable strategies. Cornelissen
et al. (2001) also used fuzzy set theory to link human expectations expressed
Extraction of Parameters from Remote Sensing Data 11

linguistically to SIs presented numerically. The conclusion from this review is

that the determination of sustainability is complex and not clearly defined and
that there is often a time lag between development and its negative impact.
It is well known that remote sensing technologies, because of the regional and
multitemporal coverage of the images, can enable measurement of many variables
more economically than other field-based or manual methods. It is necessary to iden-
tify those indicators that can be measured reliably by remote sensing on a regular
basis and that are reproducible, without bias, and truly reflect the characteristics of
the environment when it is changing. Appropriate indicators must be determined
in association with experts in the particular fields in which the indicators are being
developed. While there are redundancies in Table 1.1, many of the criteria for
determining SIs are relevant to the capabilities of remote sensing, particularly those
in columns 1, 2, and 3.

1.4 INDICATORS FOR URBAN SUSTAINABILITY

Sustainability of urban areas should define sustainable urban form, which can also
be seen from both planning and landscape perspectives. From the planning view-
point, a sustainable urban form is defined by its compactness, mixed use, density,
sustainable transport, diversity, and greening (Jabareen 2006). Therefore, compact-
ness is one of the design concepts for a sustainable city.
The growth in cities is causing increasing stress on many aspects of the urban
environment. According to the World Health Organization (WHO), 54% of the
global population in 2014 lived in cities and that percentage is growing annually at
a rate of more than 1.5%. Rees and Wackernagel (1996) believed that the end of the
20th century was a turning point in the history of human civilization and provided
many examples of how humans are increasingly consuming more of the Earth’s
resources, to the extent that a full-world assumption of economics should be adopted
rather than that of an empty world. They then went on to discuss the ecological foot-
print and human load of cities, which is increasing with time for many developed
countries. They concluded that “no city or urban region can achieve sustainability
on its own” because cities depend on resources from the hinterland, stretching from
nearby areas to globally through exports and imports. Therefore, “a prerequisite for
sustainable cities is sustainable use of the global hinterland.” They listed the advan-
tages of city living, based on economies of scale, possibilities of reduced energy con-
sumption, and recycling, among others. However, they were pessimistic about the
ability of humans to make positive steps to achieve more sustainable living. Alberti
(1996) stated that cities alter the local and global environment, because they alter the
land, import vast amounts of food water and energy, and export waste. She stated
that assessing sustainability of an urban environment requires assessment of the fol-
lowing aspects:

• Direct transformation of the physical structure and habitat

• Use of natural resources (renewable and nonrenewable)
• Release of emissions and waste
• Human health and well-being
12 Remote Sensing for Sustainability

Alberti (1996) also provided many examples of measures of urban sustainability,

some of which are measurable by remote sensing technologies, for example, urban
land cover, open areas, transport networks, and the effects of land use on accessibil-
ity to transport, urban renewal areas, derelict areas, air quality, and accessible green
space.
Freedberg (2010) stated that there are six specific principles for urban sustainability:

• Greater viability of public transport

• Affordable housing that meets a range of demographic groups
• Greater access to quality education and jobs
• Supporting neighborhoods that are engaging in sustainable practices
• Engagement of governments in providing financial support for sustainable
activities
• Investment in healthy, safe, and walkable neighborhoods

Several of these principles can be monitored by remote sensing technologies.

Olalla-Tárraga (2006) stressed the need to take an integrated approach to develop-
ing indicators for urban environments that are a complex mixture of buildings,
backyard gardens, open space, transport links, impervious surfaces, drainage systems,
deteriorating atmosphere including heat sinks, and many more.
Shen et al. (2011) explored SIs used in a number of cities around the world, listing
37 indicators under the following headings: Environmental, Economic, Social, and
Governance. The Environmental Indicators were as follows:

• En1—Geographically balanced settlement

• En2—Freshwater
• En3—Wastewater
• En4—Quality of ambient air and atmosphere (which would include the
effect of heat sinks)
• En5—Noise pollution
• En6—Sustainable land use
• En7—Waste generation and management
• En8—Effective and environmentally sound transportation systems
• En9—Mechanisms to prepare and implement environmental plans

Tanguay et al. (2010) compiled 188 urban SIs derived from 17 studies in published
literature. They noted that of the 188 indicators, only 72% were used in one or two
studies, while very few were found in more than 5 of the 17 studies. The distribution of
indicators in the social, environmental, and economic categories revealed the overlap
between sustainability dimensions and their descriptions. By further aggregating the
indicators and reviewing their significance, they concluded on a list of 29 indicators,
only 3 of which were related to the environment: ecological footprint, space allotted
to nature conservation relative to area of territory, and percentage of waterways with
excellent water quality. While the authors rated the important economic and social
aspects of sustainable city living, they did not consider the sustainability of city
dwellers in terms of the input of environmental capital.
Extraction of Parameters from Remote Sensing Data 13

Brandon and Lombardi (2011) have adopted the framework for assessment of the
sustainability of developments of the built environment, based on the concept of
a Cosmonomic Idea of Reality incorporating 15 modalities in a specific order in
which the latter modalities depend on the earlier ones. They comprise the following:
numerical, spatial, kinematics, physical, biological, sensitive, analytic, historical,
communicative, social, economic, aesthetic, juridical, ethical, and creedal. These
modalities have been aggregated into two levels, where the highest level comprises
physical environment, human cultural capital, and financial institutional capital. The
second level of five main groups of urban policy activities comprises three each of
the modalities listed above.
Goddard et al. (2010) reviewed the importance of neighborhood gardens and
their effect on biodiversity in cities. They suggested that the scale of the gardens
is important for different taxa, some being able to survive in small regions,
whereas others require much larger areas than neighborhood gardens to thrive.
They demonstrated that gardens are socioecological constructs at different scales
depending on the location, neighborhood, and individuals responsible.
There is little guidance from the above discussion on which SIs should be adopted
to assess the sustainability of urban environments based on remote sensing technolo-
gies, but some examples will be considered in the next section.

1.5 APPLICATION OF REMOTE SENSING

FOR URBAN SUSTAINABILITY
1.5.1 available remotely senseD Data
Foody (2003) stated that remote sensing can provide a wealth of environmental data
over a range of spatial and temporal scales and therefore can play a major role in the
provision of indicators of environmental conditions for sustainable development and
associated decision-making. The reason why remote sensing technologies have not
been presented as important for assessing the sustainability of urban environments
may be partially attributed to the lack of data with suitable resolutions. However,
with recent developments in high-resolution sensors, remote sensing technologies
should now be able to make significant contributions to the assessment of sustainable
or unsustainable practices. Also, the repeat coverage that is available with satellite
remote sensing at the current high spatial resolutions should mean that they will
be even more valuable for assessing trends in sustainability of urban areas. The
characteristics of remote sensing systems that acquire data suitable for determining
the sustainability of urban areas are presented in Table 1.2.
Optical systems that acquire high-resolution images from the air, either by RPAS
(Remotely Piloted Aerial Systems, UAS, or drones) or by piloted aircraft, and
imaging systems on Earth-orbiting satellites record reflected solar radiation from
the Earth’s surface. Because of the relatively low flying heights of airborne sen-
sors, there is usually greater flexibility in decisions of selection of airborne sensors.
While they are listed in Table 1.2, the following discussion on resolutions will refer
to satellite sensors. Most modern satellites are referred to as agile and hence can
point forward/backward to obtain stereo images and sideways to acquire images of
14 Remote Sensing for Sustainability

TABLE 1.2
Examples of Data Acquisition Systems for Urban Remote Sensing
Data Type Characteristics Examples
RPAS image Small cameras with spatial Multipropeller copter or fixed wing RPAS
data resolutions ranging from 1–2 cm under control of a remote operator
to >10 cm
Spectral resolution—RGB or CIR
Temporal resolution—as required in
daylight
Manned Digital aerial cameras A range of commercial digital aerial cameras
aircraft image Spatial resolution—3–5 cm to are available with resolutions ranging from
data >50 cm 40 Mpixels to more than 250 Mpixels
Spectral resolution—RGB, CIR HyMap hyperspectral sensor with 128 bands
Hyperspectral sensor acquiring with wavelengths from 0.45 to 2.48 μm and
hundreds of bands 3–10 m spatial resolution
Temporal resolution—as required in AVIRIS is an optical sensor with 4- to 20-m
daylight spatial resolution, 224 spectral bands, with
wavelengths from 0.4 to 2.5 μm
Satellite High-resolution images with spatial IKONOS II with 0.80-m panchromatic
images resolutions from approximately images and 3.2 MSS
30 cm to several meters WorldView-1 with 0.50-m panchromatic only
Spectral resolution—panchromatic CartoSat2B with 0.8-m panchromatic images
(pan) with a single band to GeoEye 1 with 0.41-m panchromatic images
multiple bands (MSS) and 1.65-m MSS
Temporal resolution—from many WorldView-2 with 0.4-m pan images and
days to 1 or 2 days depending on 1.85-m MSS with 8 bands
location and capabilities of the Pléiades with 50-cm panchromatic and 2-m
satellite MSS
WorldView-3 with 31-cm panchromatic and
8 bands MSS
Multiple microsatellites
Synthetic Microwave sensors installed in SAR wavelengths used for satellite remote
aperture radar satellites or aircraft and operate in sensing
(SAR) cloudy conditions and day or night X-band—3 cm
Spatial resolution <1 m for airborne C-band—8 cm
and ranging up to >10 m for space L-band—25 cm
borne with various levels of Most satellite-borne SAR sensors acquire
polarization either X-band or C-band
Spectral resolution—variable RADARSAT-2—C-band
Temporal resolution—variable TerraSAR-X—X-band
according to number of satellites TanDEM-X—X-band
available and orbits COSMO-SkyMed—X-band
ALOS-2—L-band
Commercial airborne systems
Lidar Currently airborne only. Elevation Many commercial systems are available with
posts acquired at 2 to >10 posts/m2 similar characteristics
Multispectral sensors available
Extraction of Parameters from Remote Sensing Data 15

the terrain surface from neighboring orbits, thus reducing the time between multiple
imaging of the same area on the terrain.
The resolution of remote sensing images can be expressed in terms of spatial,
spectral, and temporal resolution. A brief description of the three levels of resolution
will be presented for completeness.

• Spatial resolution of data is generally described as the footprint of the

sensor or the area that the sensor views instantaneously on the terrain
surface, often referred to as ground sampling distance. This is a function of
sensor altitude, detector size, sensor optics, and the system configuration.
The dimensions of the smallest feature visible on the terrain surface are
a function of the spatial resolution, the contrast and color of the object in
relation to its background, and its shape (e.g., very narrow linear features
are often detectable on satellite images when they are not expected to be).
While sensors on satellite platforms are available with a very broad range of
spatial resolutions, because of the requirements of urban remote sensing to
detect buildings, pavements, roads, parking lots, the fragmentation of open
space, and similar issues, only the so-called high-resolution satellite images
with spatial resolutions on the order of 2 m and smaller will be discussed
in this chapter. This is supported by Sliuzas et al. (2010) who stated that the
identification of small urban objects or objects in complex environments
requires a minimum spatial resolution of 5 m. They also argued that the
required spatial resolution of images should be determined by the smallest
object to be identified in the images. The availability of images with spatial
resolutions smaller than 2 m over the past decade provides new opportunities
for satellite remote sensing applications in sustainability studies in urban
areas.
• Spectral resolution refers to the range and width of wavelengths that can be
resolved by a sensor.
• Temporal resolution refers to the frequency of coverage of a certain area on
the terrain surface by a sensor. The repeat cycle of orbiting satellites with
optical systems viewing vertically only is usually many days, but this can
be reduced by a constellation of satellites or by including tilt capability on
the satellites so that they can view the same areas on the ground, weather
permitting, from successive orbits.

In the design of sensors on satellites, there is necessarily a compromise between

the demands for spatial, spectral, and temporal resolutions. For example, currently,
there is the WorldView-3 satellite with a single panchromatic very high spatial reso-
lution of 0.31 m and eight multispectral bands with spatial resolutions of 1.24 m.
The spectral resolution of this satellite may be relatively coarse for analysis of some
materials in urban areas. On the other hand, currently available airborne hyperspec-
tral sensors have very high spectral resolutions in hundreds of bands but a relatively
coarse spatial resolution (e.g., HyMap system referred to in Table 1.2).
Optical image satellite systems that are commonly used for urban applications
have been described by Rashed and Jürgens (2010) and in the comprehensive review
16 Remote Sensing for Sustainability

paper by Weng (2012). Gamba et al. (2011) referred to several problems with the
geometry of high spatial resolution optical images, since a digital elevation model
(DEM) with an accuracy on the order of 1 m is required to produce orthophotos from
such data over urban areas. If high buildings exist in the image, they will be subject
to relief distortions that can only be corrected if the DEM includes the heights of
buildings. In addition, large distortions will occur in images that are acquired with
the satellite tilted sideways to image areas at a specific time. Image interpretation
may also be compromised for the tilted images.
Synthetic Aperture Radar (SAR) systems are active remote sensing systems that
emit microwave radiation from an antenna and record the time of travel and the
intensity of the radiation returned to the antenna. SAR systems have the advantage
that they can be used day or night and in cloudy conditions, and the polarization of
emitted and received radiation can be varied as well, providing additional potential
for extracting information from the data. Soergel (2010) has described some applica-
tions of SAR systems for the analysis of urban areas.
Airborne lidar (Light Detection and Ranging—also written as LiDAR) data are
based on a scanning laser that emits a pulse (with a wavelength in the infrared region
of the electromagnetic spectrum) toward the terrain surface, and the distance to the
terrain surface can be determined from the time of travel of the pulse to the terrain
and back to the sensor.
Together with knowledge of the position and altitude of the aircraft, a dense point
cloud on the terrain surface can be determined to represent the position and eleva-
tion of discrete posts at a density of two or more posts per square meter. In addition,
the intensities of the returned laser pulses from the terrain or objects on the terrain
are recorded, which can represent an infrared image. Airborne lidar can assist in
the extraction of three-dimensional (3D) information about man-made features and
therefore details of the built environment.

1.5.2 Urban sUstainability measUres DeriveD by remote sensing

Taking into consideration the above review of the literature on SIs for urban environ-
ments, the aspects of an urban environment that can be assessed for SIs by remote
sensing technologies are presented in Table 1.3, where an attempt has been made to
develop three levels of a framework for these environmental indicators. With refer-
ence to Figure 1.1, in Table 1.3, Area is the Urban Sustainability Measure, Objective
is suggested as corresponding to the measurements made by the remote sensing
technologies, while Attributes are the parameters derived by remote sensing tech-
nologies. While some Indicators, as shown in Figure 1.1, may be determined for
the attributes, there is no adequate information available yet to present these details.
The measures suggested in Table 1.3 would need to be considered in conjunction
with other environmental, social, and economic SIs, which cannot be measured by
remote sensing technologies, for the overall framework for assessing sustainability
of urban areas.
In terms of the framework developed by Brandon and Lombardi (2011), while they
are less specific to the measurements derived by remote sensing technologies, the first
five modalities—numerical, spatial, kinematics, physical, and biological—would
Extraction of Parameters from Remote Sensing Data 17

TABLE 1.3
Assessable Sustainability Aspects in the Urban Environment by Remote
Sensing Technologies
Objective—Measurement Attribute—Parameters
Area—Urban by Remote Sensing Derived by Remote Sensing
No. Sustainability Measure Technologies Technologies
1 Balanced development— Measurement of impervious Ratio of area of impervious
fraction of built versus open surfaces in relation to open surfaces to open spaces
space space
2 Transformation of the Growth in fragmentation of Diversity, dominance,
physical structure and open space versus fragmentation
habitat from green space to impervious surfaces
impervious surfaces
3 Effective and Mapping and analysis of Transport Mode Index
environmentally sound transport systems to
transportation systems demonstrate effectiveness
4 Consideration of healthy, Determine compactness of Size
safe, and walkable/cycle cities, mapping of walking Density
neighborhoods and cycle paths, and Degree of distribution
township layout Clustering
5 Consumption of natural Measurement of Land use/land cover changes
resources (renewable and deforestation and changes
nonrenewable), from in land cover over time
hinterland and its impact
6 Effects on biodiversity Changes in local vegetation Land-use conversion and loss
and native flora and of habitats
potential habitats for fauna
7 Release of emissions and Determine surface water Not covered
waste, especially into quality and chemical
waterways and the content of atmosphere
atmosphere

apply to remote sensing technologies. Numerical refers to the accounting process

that can be developed from remote sensing, spatial refers to the spatial details
extracted, kinematics refers to transport and mobility, physical refers to the physi-
cal environment, and biological refers to health and ecological protection as well
as biodiversity. While both frameworks should be relevant to urban sustainability,
that shown in Figure 1.1 will be the basis of a demonstration of the contributions of
remote sensing to urban sustainability in the following.
There is a need for greater attention to the future development of SIs for urban
areas because more than half of the world’s population lives in cities. Such indi-
cators should also be used for monitoring whether an environment is becoming
less sustainable as developments occur or populations grow, leading to a need for
modification of decisions affecting an urban area. Each of the Indicators listed in
Table 1.3 will be discussed below, except for Item 7, “Release of emissions and
18 Remote Sensing for Sustainability

waste, especially into waterways and the atmosphere,” which is discussed else-
where in this book.

1.5.2.1 Measurement of Impervious Surfaces in Relation to Open Space

A great deal has been written on the measurement of impervious surfaces, which
cover the full range of methods available for image classification of such surfaces, for
example, in the review paper by Weng (2012). There are a number of issues raised in
Weng’s paper that are relevant to this chapter. Buildings and roads can be detected
with optical sensors with spatial resolutions of 0.25 to 0.5 m (which is a much smaller
resolution than referred to in Section 1.5.1), although shadows may cause problems
with tall buildings. There is also the possibility of confusion of roads and buildings
with surrounding features, and therefore, fusion of several forms of data of the area
may prove beneficial. Weng (2012) has suggested that the investigation of the spectral
diversity of impervious surfaces together with 3D characteristics and temporal
changes should prove beneficial in urban studies. Salah et al. (2010) investigated
various machine learning and ensemble learning approaches for the extraction of
impervious surfaces created by buildings and roads and also extracted the ground
and vegetation from high-resolution aerial photography and airborne lidar data over
urban areas and found in excess of 90% accuracy for most methods.
The investigation by Wu and Yuan (2011) using high-resolution satellite images for
extraction of impervious surfaces was based on pixel and object-oriented methods.
They have estimated that approximately 40%–50% of pixels are mixed pixels, for
high-resolution satellite data such as IKONOS and Quickbird multispectral data.
They used a normalized spectral mixture analysis, regression trees, artificial neural
networks, and object-oriented approaches for the extraction of impervious surfaces
with overall accuracies of approximately 90%. Shadows from buildings and trees
need to be differentiated since they hide different types of surfaces. Canters et al.
(2011) also discussed the extraction of impervious surfaces from both high-resolution
and medium-resolution images and suggested that the use of lower-cost medium-
resolution images together with subpixel classification methods could be a better
approach. However, this claim is not in agreement with other statements made above.
Satellite SAR images have been used for extraction of buildings and roads in urban
areas, sometimes with mixed success (Soergel 2010), but high-resolution airborne
SAR images should provide better extraction capabilities than satellite SAR images.
While it was expressed earlier that spatial resolutions of 2 m or better are desirable
for urban studies, the application of lower-resolution hyperspectral images has been
extensively explored by Roessner et al. (2011) in studying impervious materials in
urban areas. Images were derived from the HyMap sensor with pixel sizes ranging
from 3 to 6 m, details of which are given in Table 1.2. The study extracted end
members describing distinct spectral signatures of different surface materials in
several German cities. The study revealed the increased information derived from
the hyperspectral images compared with lower spectral resolution images used in
previous studies, which assumed that the surfaces are completely impervious. The
higher spectral resolution hyperspectral images, though lower in spatial resolution,
allowed more detailed mapping of the surface materials in urban areas and the
separation of semipermeable surfaces, which might include cobblestones or gravel,
Extraction of Parameters from Remote Sensing Data 19

from completely impervious surfaces. This would affect the level of runoff and, thus,
the hydrological cycle of urban areas. Also, there may be significant influences on
the assessments of microclimates in urban areas. The availability of hyperspectral
images should allow better definition of ground cover in urban areas and associated
ecological factors and also allow for studies on the effects of developments on
biodiversity. Time series would also allow for studies on changes in these factors.
The definition of an attribute for impervious surfaces is complex, but the ratio of
areas of fully impervious surfaces to open spaces is suggested as the attribute for
this SI in Table 1.3. Further work would be required to determine the influence of
partially impervious surfaces on urban sustainability.

1.5.2.2 Growth in Fragmentation of Open Space

versus Impervious Surfaces
The growth in fragmentation in urban and suburban areas is suggested as a further SI
for urban areas, since the transformation of fragments of open space into impervious
surfaces is an important indicator of urban development. Greenhill et al. (2003)
derived two parameters to act as ecological indicators for suburban areas:

• The weighted mean patch size (WMPS), which provides information about
the size distribution of vegetation patches. The mean patch size is the aver-
age area of vegetative patches within a window. The weighted mean patch
size includes information on both patch size and number.
• Lacunarity, which provides an indicator of the spatial clustering of such
patches and is dependent on the number of interpatch nonvegetated pixels
within a square box that are summarized in a histogram. The mean and
variance of the counts are used to calculate the lacunarity.

These parameters could be used to assess the impact of fragmentation of open

space caused by urban developments that change the distribution of vegetated versus
nonvegetated areas. Certain values of WMPS and lacunarity could correspond to
maintaining a relatively low density of housing and a good clustering of local green
areas, which would be suitable for diversification of flora and fauna. On the other
hand, high values of these parameters would demonstrate the primarily impervious
environment, which provides little opportunity for diversification of flora and fauna,
and the need for the provision of such environments elsewhere in the urban area for
an adequate lifestyle for the inhabitants.
Lein (2014) argued that limiting sustainability studies to regional scales enables
the linking of the economic, social, and environmental factors of sustainability
to the landscape unit. In addition, the revelation of unsustainable trends in aspects of
the environment should lead to a change in behavior or development programs. He
then presented advantages of remote sensing methods for assessing sustainability:
the capacity of obtaining unique measurements of reflectances from the Earth’s
surface, the repeatability of the observations, and the archival capacity of the
data. He espoused the possibility of developing indices for sustainability, which
should have the following attributes: relevance, concept integrity, reliability, scale
appropriateness, scale sensitivity, and robustness, many of which are similar to those
20 Remote Sensing for Sustainability

recommended by Becker (1997), as shown in Table 1.1. Lein thus provided indices that
can be based on metrics derived from remote sensing, namely, impervious surfaces,
fragmentation, diversity, and dominance. The estimates of impervious surfaces are
based on NDVI. Fragmentation of the landscape is dependent on the following
formula: F = (n − 1)/(c − 1), where n is the number of patches in a kernel and c is the
number of cells considered. Diversity = (Σp × ln(p)), where Σ is the summation of all
land types in the study area, p is the proportion of each land type in the spatial unit
of measure (pixel), and ln is the natural logarithm. Dominance is defined by the most
abundant land type by lnS + Σpk × lnpk, where S is the number of habitat types and
pk is the proportion of area in habitat k. As these metrics were assessed using Earth
observation data, monitoring could lead to a dynamic management process. Lein
demonstrated an ecological integrity component that is dependent of the diversity,
dominance, fragmentation, and impervious surfaces. These components could be
monitored to assess trends in sustainability. The use of high-resolution satellite data
could provide greater granularity to the information extracted in this research, thus
enabling more detailed analysis of fragmentation, diversity, and dominance.
Sapena and Ruiz (2015) have defined several metrics to determine the rate of growth
patterns on the European Urban Atlas databases in 2006 and 2012. The measures
used to describe fragmentation are as follows: urban density, which estimates the
proportion of developed areas, comprising housing, commercial, industrial and
landmark buildings, roads, barren land, and leisure areas, as a ratio of the classified
urban areas in the region; weighted standard distance, which measures concentration
or scattering around the centroid of the objects of a given class, weighted by the
size of each object; Euclidean nearest neighbor (ENN) mean distance, which is
also a measure of scattering, being the mean of the distances between the edges of
the objects of the same class; Shannon diversity, which represents the abundance
and evenness of the classes; and edge contrast ratio, which quantifies the degree of
contrast between objects from different land uses, excluding the road network.
On the basis of the above discussion, the primary attribute for this indicator is
fragmentation together with diversity and dominance, which have been added to
Table 1.3. The measures proposed by Lein (2014) and Sapena and Ruiz (2015) are
demonstrations of how such attributes could be measured using remote sensing data.

1.5.2.3 Mapping and Analysis of Sustainable Transport Systems

The task of determining the overall effectiveness of transport routes is beyond the
capacity of remote sensing. However, an essential task is to provide data for the
spatial analysis, together with GIS tools that enable transport experts to deter-
mine the effectiveness of the transport infrastructure. Zhang and Guindon (2006)
and Guindon and Zhang (2007) have described the process for developing a set of
Sustainable Transportation Performance Indicators, in which the basis for land-
use mapping were combined with socioeconomic data derived from the national
census, for the determination of such parameters as urban population density and
compactness of cities. The Transport Mode Index, which was adopted as the attri-
bute in Table 1.3, measures the impact of the land-use mix and urban form on the
feasibility of various modes of transportation. The authors stated that while indi-
cators are important for demonstrating the sustainability of transport systems in
Extraction of Parameters from Remote Sensing Data 21

Canadian cities, simple measures such as density and compactness are poor indica-
tors of energy consumption, whereas land-use mix is a better indicator. This would
be the cue for the application of remote sensing technologies since land-use analysis
can be undertaken economically over urban areas using medium-resolution images.
However, they argued that more research is required to develop better indicators that
incorporate median travel distance.

1.5.2.4 Compactness of Cities and Mapping of Township Layout

Compactness is one of the “design concepts of sustainable urban form” (Jabareen
2006). The logic behind it is that a more compact city results in less travel and there-
fore lower energy consumption, leading to environmentally sustainable transport
systems, less fragmentation of neighboring lands, walkable environments, and elim-
ination or significant reduction in urban sprawl. The scale of analysis of compact-
ness of urban areas may determine the methods and resolutions of images used for
measurement by remote sensing technologies. Large metropolitan areas may require
regional-scale measurements while walkable cities may require larger-scale images
and manual mapping operations.
A number of parameters have been used by Tsai (2005) to define compactness
versus urban sprawl. These include size of the metropolitan area, which varies
according to the extent of sprawl; density, expressed in terms of land occupation per
capita; degree of distribution of development in a metropolitan area; and clustering
or centralization of the metropolis.
Tsai (2005) then discussed the Moran, Geary, and Gini coefficients, including
simulation of various forms of urban areas based on population or employment that
can be assessed as to their suitability for describing compactness. He found that the
Moran coefficient was useful as a metric for distinguishing between compact urban
areas and urban sprawl, but it was not able to differentiate between a circular and a
linear shape of cities. There was no index that completely described compactness
versus sprawl, but the Moran and Gini coefficients were suitable for some examples
of urban form. Population or employment cannot be determined by remote sensing
technologies. However, the 3D form determined by airborne lidar may be a surrogate
for population required by the Moran coefficient, although it is unlikely to be able to
distinguish between actual population and employment locations. The 2D geometry
of urban areas required for the Gini coefficient may be determined from medium- to
high-resolution images, although it would require manual interpretation of the images.
Herold et al. (2003) analyzed four urban spatial metrics using high-resolution
satellite data to determine their characteristics for several types of land covers rang-
ing from forests to high-density single-unit residential areas. They supported the use
of spatial metrics as descriptors of built-up structures and open areas. Huang et al.
(2007) suggested additional measures of compactness that included an area weighted
average shape index, a patch fractal dimension, centrality, several compactness indi-
ces, ratio of open space, density, and several measures related to social indices for
comparisons of the form of many cities around the world using medium-resolution
remotely sensed data. They found significant differences between developed world
cities in America, Australia, and parts of Europe, which tend to be subject to sprawl,
and those generally more compact cities in Asia.
22 Remote Sensing for Sustainability

Sim and Mesev (2011) extracted similar parameters as Tsai (2005) (size, density,
continuity of development, and scattering of developments) but also added shape
and loss of green space as indicators of urban sprawl. They used different statisti-
cal methods for indicators, including entropy and ENN. Therefore, from the above
discussion as shown in Table 1.3, the attributes for urban compactness are assumed
to be those defined by Tsai above, namely, size of the metropolitan area, density,
degree of distribution, and clustering.
Locating and mapping the lengths of walking paths and mapping township lay-
outs are primarily a manual mapping task that can be undertaken with medium- to
high-resolution optical images for the production of digital orthophotos.

1.5.2.5 Deforestation and Changes in Land Cover

Since urban areas depend on imports of products derived from outside the urban
areas, and further in some cases, a considerable amount of land clearing and defor-
estation occurs to service urban areas. In order to define the effects of these land-use
changes, a detailed analysis of particular cities and the resulting changes in surround-
ing areas would have to be undertaken. Medium-resolution remote sensing technolo-
gies should be applicable for these purposes. For example, NASA (2010) revealed,
based on the application of the QuickScat satellite radar system, that Beijing has
quadrupled its size in 9 years. Foody (2003) discussed aspects of maps of forest
and forest change, estimation of forest biomass, biodiversity, and drought, showing
how remote sensing can satisfy some of the broadly based indicators of changes in
open space areas, and that satellite remote sensing would be the most economical
approach. As an example, Brazil’s National Institute for Space Research (INPE) has
used satellite data for monitoring deforestation for more than 15 years based on a
range of medium-resolution satellite images, demonstrating the conversion of forests
to farming in the Amazon region. INPE has developed an almost real-time moni-
toring system to detect illegal land clearing. High-resolution images are required
to determine the exact area of clearing and for extracting small areas of clearing.
Therefore, the study of land-use changes over the period due to urbanization could
be undertaken economically by processing time series remote sensing images.
The information derived from remote sensing can also be directly related to mea-
suring important socioeconomic impacts. Rates of land cover change and drought
will strongly influence vegetation yield, which could substantially affect human
health and well-being of neighboring cities. These factors will, for example, influ-
ence the demand for and rate of fertilizer application, which may be associated with
downstream pollution. Issues such as soil erosion are a major concern for land users
and are also strongly associated with consequential impacts, including the silting of
lakes and damage to hydroelectric power stations.
Bacchus et al. (2000, 2003) investigated the detrimental effects on the health of
the vegetation, in the case of pond cypress, caused by the withdrawal of groundwater
from aquifers in Florida. Their investigations were based on laboratory spectrometry
studies (in the visible, near-infared [NIR], and mid-infrared regions of the spectrum)
of dried milled branch tips collected from natural stands of pond cypress stands,
both in summer and in winter before bud-break. They found that the NIR spec-
tral response was more affected by stress than by site-related factors. Visual effects
Extraction of Parameters from Remote Sensing Data 23

of stress that were found to be well correlated with the spectrometry studies were
also evident. They believed that the chemical changes in the vegetation that were
revealed in the spectrometry studies could be used as an indicator of unsustainable
withdrawal of water from the aquifers. This research is an important development
in the use of indicators for detecting unsustainable practices leading to stress on
vegetation caused by inadequate water. In a similar manner, Chisholm et al. (2003)
studied moisture stress on Eucalyptus camaldulensis (River Red Gum) in Australia,
using high-resolution spectral data at the leaf level. Their results indicate that even
low levels of stress can be detected from such data before they become visible, and
therefore spectral reflectance regions that would be indicators of moisture stress in
vegetation and hence act as appropriate SIs may be developed. Withdrawal of water
from aquifers is often used as a source of water for urban areas and could be a further
manifestation of the impacts of urbanization on surrounding areas. These impacts
need to be assessed by appropriate remote sensing technologies. A general term for
the attributes that could be adopted is land use/land cover changes but there are
many impacts related to the effects of urbanization on the hinterland that should be
included in a detailed set of indicators.

1.5.2.6 Biodiversity
Land cover change threatening biodiversity, and a major variable in the loss of nutri-
ents from productive lands, can be mapped and monitored by a range of remote sens-
ing data sources. This may require high-resolution images or spectrally unmixing
approaches to determine the class composition of mixed pixels to capture land cover
modifications systematically and on a repetitive basis. Vegetation indices and change
detection techniques derived from images from high-resolution optical sensors per-
mit the mapping, monitoring, and measurement of the areal extent of the change.
Hepinstall-Cymerman (2011) demonstrated that analyzing biodiversity requires
knowledge of the existing fauna and vegetation, which also involves field surveys and
documentation. He described a land cover change avian biodiversity model based on
remote sensing, which, together with adequate knowledge of the avian fauna, pro-
vided details of the impacts on the species. Land-use conversion and loss of habitats
are shown as attributes for the effects on biodiversity.

1.6 CONCLUSIONS
This chapter aims to present some principles for assessing sustainability of develop-
ment in urban areas and to describe the ways in which remote sensing can be used
in this process. Definitions of sustainable development have been given, and the
approach to its assessment based on SIs is described. While a number of different
indicators are currently available for the three pillars (social, economic, and envi-
ronmental), there appears to be no consensus on the most appropriate indicators for
sustainability of the environment, and especially for urban areas. The list suggested
in Table 1.3 could represent some components of a framework for SIs in urban areas,
but considerable work would still be required to further develop and test the set of SIs.
Since more than 50% of the global population now lives in urban areas and this
percentage is increasing, there is urgency in determining and assessing SIs for urban
24 Remote Sensing for Sustainability

areas. Unless the sustainability of urban areas is addressed, there may be little chance
of real sustainability of the environment being achieved. Decision-making, which
affects urban environments, in many cases, is dominated by market forces often with
inadequate consideration of effects on the environment. Interdisciplinary collabora-
tions between the remote sensing community and experts in a range of scientific
fields such as ecology, biology, sociology, human resources, and politics, who can
take responsibility for assessing the sustainability of urban communities, should be
developed so that their combined expertise can determine the impacts of unsustainable
practices in urban areas and the hinterland, and decision-makers can be notified of
the need to change current practices. While remote sensing technologies will not be
the only tools for assessing sustainability, they should make an important contribu-
tion to this multidisciplinary process, provided they satisfy scientific criteria, such as
being subject to strict calibration and validation. A great deal has yet to be learned
about these processes and how the full potential of remote sensing can be achieved in
this very important issue of environmental sustainability of urban areas.

REFERENCES
Alberti M. 1996. Measuring urban sustainability, Environ. Impact Assess. Rev. 16: 381–424.
Andriantiatsaholiniaina L. A., Kouikoglou V. S., and Phillis Y. A. 2004. Evaluating strategies
for sustainable development: Fuzzy logic reasoning and sensitivity analysis. Ecol. Econ.
48: 149–172.
Azapagic A. and Perdan S. 2005a. An integrated sustainability decision-support framework.
Part I: Problem structuring. Int. J. Sustainable Dev. World Ecol. 12: 98–111.
Azapagic A. and Perdan S. 2005b. An integrated sustainability decision-support framework.
Part II: Problem analysis. Int. J. Sustainable Dev. World Ecol. 12: 112–131.
Bacchus S. T., Archibald D. D., Brook G. A. et al. 2003. Near-infrared spectroscopy of a
hydroecological indicator: New tool for determining sustainable yield for Floridan aqui-
fer system. Hydrol. Processes 17: 1785–1809.
Bacchus S. T., Hamazaki T., Britton K. O., and Haines B. L. 2000. Soluble sugar composi-
tion of pond-cypress: A potential indicator of ground-water perturbations. J. Am. Water
Resour. Assoc. 36(1): 55–65.
Barrios E. and Kimoto K. 2006. Some approaches to the construction of sustainable develop-
ment index for the Philippines, Int. J. Sustainable Dev. World Ecol. 13: 277–288.
Becker B. 1997. Sustainability assessment: A review of values, concepts, and methodological
approaches, Issues in Agriculture 10, CGIAR, Washington, DC, The World Bank.
Becker J. 1998. Sustainable development assessment for local land uses. Int. J. Sustainable
Dev. World Ecol. 5: 59–69.
Becker J. 2007. How frameworks can help operationalize sustainable development indicators,
World Futures. 63: 137–150.
Biggs R., Scholes R. J., ten Brink J. E., and Vačkář D. 2007. Biodiversity indicators. In Hák
T., Moldan B., and Dahl A. J. (Eds.), Sustainability Indicators: A Scientific Assessment,
pp. 249–270, Island Press, Washington, DC.
Blanco H., Wautiez F., Llavero A., and Riveros C. 2001. Sustainable development indicators
in Chile: To what extent are they useful and necessary? Eure-Revista Latinoamericana
de Estudios Urbano Regionales. 27: 85–95.
Böhringer C. and Jochem P. E. P. 2007. Measuring the immeasurable—A survey of sustain-
ability indices. Ecol. Econ. 63: 1–8.
Extraction of Parameters from Remote Sensing Data 25

Brandon P. and Lombardi P. 2011. Evaluating Sustainable Development in the Built

Environment, Second Edition, Wiley-Blackwell, Hoboken, New Jersey.
Canters F., Batelaan O., van de Voorde T. et al. 2011. Use of impervious surface data obtained
from remote sensing in distributed hydrological modeling of urban areas, In Yang X.
(Ed.), Urban Remote Sensing, pp. 255–273, Wiley-Blackwell, Hoboken, New Jersey.
Chisholm L. A., Cooke J., Erdmann B. et al. 2003. Preliminary Investigations into Observed
River Red Gum Decline along the River Murray below Euston: Technical Report 03/03.
Australia: Vic: Murray-Darling Basin Commission.
Cornelissen A. M. G., van den Berg J., Koops W. J. et al. 2001. Assessment of the contribution
of sustainability indicators to sustainable development: A novel approach using fuzzy
set theory. Agric. Ecosyst. Environ. 86: 173–185.
Foody G. M. 2003. Remote sensing of tropical forest environments: Towards the monitoring
of environmental resources for sustainable development, Int. J. Remote Sens. 24(20):
4035–4046.
Freedberg M. 2010. In Schaffer D., Vollmer D. (Eds.), Pathways to Urban Sustainability:
Research and Development on Urban Systems: Summary of a Workshop by Committee
on the Challenge of Developing Sustainable Urban Systems, pp. 9–10, National
Research Council, Washington, DC.
Gallopin G. C. and Raskin P. D. 2002. Global Sustainability Bending the Curve. Routledge,
London.
Gamba P., Erll’Acqua F., Stasolla M. et al. 2011. Limits and challenges of optical very-high-
resolution satellite remote sensing for urban applications, In Yang X. (Ed.), Urban
Remote Sensing, pp. 36–47, Wiley-Blackwell, Hoboken, New Jersey.
Greenhill D. R., Ripke L. T., Hitchman A. P. et al. 2003. Characterization of suburban areas
for land use planning using landscape ecological indicators derived from Ikonos-2
multispectral imagery, IEEE Trans. GRSS 41(9): 2015–2023.
Goddard M. A, Dougill A. J., and Benton T. G. 2010. Scaling up from gardens: Biodiversity
conservation in urban environments, Trends Ecol. Evol. 25(2): 90–98.
Guindon B. and Zhang Y. 2007. Using satellite remote sensing to survey transport-related
urban sustainability. Part II. Results of a Canadian urban assessment, Int. J. Appl. Earth
Observ. Geoinf. 9: 276–293.
Gutiérrez-Espeleta E. E. 2007. Further work needed to develop sustainable development
indicators. In Hák T., Moldan B., and Dahl A. J. (Eds.), Sustainability Indicators: A
Scientific Assessment, pp. 351–360, Island Press, Washington, DC.
Hák T., Moldan B., and Dahl A. J. (Eds.) 2007. Sustainability Indicators: A Scientific
Assessment, Island Press, Washington, DC.
Harrington L., Jones P. G., and Winograd M. 1993. Measurements and indicators of sustain-
ability. Report of a Consultancy Team, Centro Internacional de Agricultura Tropical
(CIAT), Cali, Colombia.
Hart M. 1999. Guide to Sustainable Indicators. 2nd Ed., Hart Environmental Data, North
Andover, MA.
Hepinstall-Cymerman J. 2011. Ecological modeling in urban environments: Predicting
changes in biodiversity in response to future urban development. In Yang X. (Ed.),
Urban Remote Sensing, pp. 359–370, Wiley-Blackwell, Hoboken, New Jersey.
Herold M., Couclelis H., and Clarke C. C. 2003. The role of spatial metrics in the analysis and
modeling of urban land use change, Comput. Environ. Urban Syst. 29: 369–399.
Huang J., Lu X. X., and Sellers J. M. 2007. A global comparative analysis of urban form:
Applying spatial metrics and remote sensing, Landscape Urban Plann. 82: 184–197.
Hueting R. and Reijnders L. 2004. Broad sustainability contra sustainability: The proper con-
struction of sustainability indicators, Ecol. Econ. 50: 249–260.
Jabareen Y. R. 2006. Sustainable urban forms, J. Plann. Educ. Res. 26: 38–52.
26 Remote Sensing for Sustainability

Jesinghaus J. 2007. Indicators: Boring statistics or the key to sustainable development? In Hák
T., Moldan B., and Dahl A. J. (Eds.), Sustainability Indicators: A Scientific Assessment,
pp. 351–360, 83–96, Island Press, Washington, DC.
Karlsson S., Dahl A. L., Biggs R. et al. 2007. Meeting conceptual challenges, In Hák T.,
Moldan B., and Dahl A. J. (Eds.), Sustainability Indicators: A Scientific Assessment,
pp. 27–48. Island Press, Washington, DC.
Kates R. W. 2000. Population and consumption: From more to enough. In Schmandt J. and
Ward C.H (Eds.), Sustainable Development: The Challenge of Transition, pp. 79–99,
Cambridge University Press: Cambridge.
Lein J. K. 2014. Toward a remote sensing solution for regional sustainability assessment and
monitoring, Sustainability 6: 2067–2086.
Mahi P. 2001. Developing environmentally acceptable desalination projects. Desalination
138: 167–172.
McGlade J. 2007. Foreword: Finding the right indicators for policymaking, In Hák T., Moldan
B., and Dahl A. J. (Eds.), Sustainability Indicators: A Scientific Assessment, Island
Press, Washington, DC.
Moldan B. and Dahl A. J. 2007. Challenges to sustainability indicators. In Hák T., Moldan
B., and Dahl A. J. (Eds.), Sustainability Indicators: A Scientific Assessment, pp. 1–24,
Island Press, Washington, DC.
NASA. 2010. Beijing Quadrupled in Size in a Decade, NASA Finds. http://www.jpl.nasa.gov
/news/news.php?feature=4641, accessed July 31, 2015.
Olalla-Tárraga M. A. 2006. A conceptual framework for assessing urban ecological systems,
Int. J. Sustainable Dev. World Ecol. 13: 1–15.
Petrosyan A. 2014. Proposal of composite appraising supportive progress beyond twelve (12)
economic sustainability indices, J. Econ. Dev. Stud. 2(2): 547–569.
Phillis Y. A. and Andriantiatsaholiniaina L. A. 2001. Sustainability: An ill-defined concept and
its assessment using fuzzy logic. Ecol. Econ. 37: 435–456.
Radermaker F. J. 2004. Balance or Destruction, Oekosoziales Forum Europa, Vienna, Austria.
Rao D. P. 1998. Remote sensing and GIS for sustainable development: An overview, Int. Arch.
Photogramm. Remote Sens. XXXII(7): 156–163.
Rashed T. and Jürgens C. 2010. (Eds) Remote Sensing of Urban and Suburban Areas,
Heidelberg, Springer.
Rees W. and Wackernagel M. 1996. Urban ecology footprints: Why cities cannot be
sustainable—And why they are the key to sustainability, Environ. Impact Assess. Rev.
16: 223–248.
Roessner S., Segl K., Bochow M. et al. 2011. Potential of hyperspectral remote sensing for
analysing the urban environment. In Yang X. (Ed.), Urban Remote Sensing, pp. 50–61,
John Wiley & Sons, Hoboken, New Jersey.
Salah M., Trinder J., Shaker A. et al. 2010. Integrating multiple classifiers with fuzzy major-
ity voting for improved land cover classification, Int. Arch. Photogramm. Remote Sens.
Spat. Inf. Sci. 39(3A): 7–12.
Sapena M. and Ruiz L. A. 2015. Analysis of urban development by means of multi-temporal
fragmentation metrics from LULC data, Int. Arch. Photogramm. Remote Sens. Spat. Inf.
Sci. XL-7/W3: 1411–1418.
Shen L.-Y., Ochoa J. J., Shah M. N., and Zhang X. 2011. The application of urban sustainabil-
ity indicators—A comparison between various practices, Habitat Int. 35:17–29.
Sim S. and Mesev V. 2011. Measuring urban sprawl and compactness: Case study Orlando,
International Cartographic Conference, CO-437, pp. 1–10, Paris, France.
Sliuzas R., Kuffer M., and Masser I. 2010. In Rashed T. and Jürgens C. (Eds.), Remote Sensing
of Urban and Suburban Areas, pp. 67–84, Heidelberg, Springer.
Soergel U. (Ed.) 2010. Radar Remote Sensing of Urban Areas, Heidelberg, Springer.
Extraction of Parameters from Remote Sensing Data 27

Stevenson M. and Li H. 2001. Indicators of sustainability as a tool min agricultural develop-

ment: Partitioning scientific and participatory process, Int. J. Sustainable Dev. World
Ecol. 8: 57–65.
Tanguay G. A., Rajaonson J., Lefebvre J.-F. et al. 2010. Measuring the sustainability of cities:
An analysis of the use of local indicators, Ecol. Indic. 10: 407–418.
Tsai Y.-H. 2005. Quantifying urban form: Compactness versus ‘sprawl’, Urban Stud. 42(1):
141–161.
Turner R. K. 1993. Sustainability: Principles and practices. In Turner R. K. (Ed.), Sustainable
Environmental Economics and Management: Principles and Practices, pp. 3–36,
London, Belhaven Press.
Van Woerden J., Singh A., and Demkine V. 2007. Core set of UNEP GEO indicators among
global environmental indices, indicators and data, In Hák T., Moldan B., and Dahl
A. J. (Eds.), Sustainability Indicators: A Scientific Assessment, pp. 343–350, Island
Press, Washington, DC.
Weng Q. 2012. Remote sensing of impervious surfaces in the urban areas: Requirements,
methods, and trends, Remote Sens. Environ. 117: 34–49.
Wu C. and Yuan F. 2011. Remote sensing of high resolution urban impervious surfaces,
In Yang X. (Ed.), Urban Remote Sensing, pp. 241–254, Wiley-Blackwell Hoboken,
New Jersey.
Zhang Y. and Guindon B. 2006. Using satellite remote sensing to survey transport-related
urban sustainability. Part 1: Methodologies for indicator quantification, Int. J. Appl.
Earth Observ. Geoinf. 8: 149–164.
 2 Earth Observation
for Urban and
Spatial Planning
Mattia Marconcini, Annekatrin Metz,
Thomas Esch, and Julian Zeidler

CONTENTS
2.1 Earth Observation Applications in Urban and Spatial Planning .................... 29
2.2 Selected Applications ..................................................................................... 30
2.2.1 Aseessment of Impervious Urban Surface ......................................... 30
2.2.2 Mapping of Urban Growth ................................................................. 33
2.2.3 Characterization of Settlement Pattern ............................................... 38
2.3 Conclusions and Outlook ................................................................................ 39
Acknowledgments.................................................................................................... 42
References ................................................................................................................ 43

2.1 EARTH OBSERVATION APPLICATIONS

IN URBAN AND SPATIAL PLANNING
At present, the two most critical phenomena affecting cities worldwide are urbaniza-
tion and climate change. Indeed, on the one hand, the United Nations (UN 2014) esti-
mate that nowadays 54% of the human population is living in urban areas (up from
34% in the 1960s); on the other hand, global climate changes are directly affecting
the economy of cities as well as the quality of urban environments.
In such context, cities play a dual role: they are part of the problem and a key
part of the solution (Kamal-Chaoui and Alexis 2009). In particular, despite covering
~2%–3% of the emerged land, cities are responsible for 30%–40% of greenhouse
gas emissions (or 70% if all the human activities are taken into account) (UN 2011),
which directly affect population health and result in flooding, storms, heat, drought,
sea-level rise, and damages to infrastructures and buildings. Nevertheless, by means
of proper managing strategies, they can also lessen the adverse impacts of climate
change through a wide range of system-specific actions and become sustainable and
livable environments. However, this is a challenging task for urban planners as they
must take into consideration all the complex and interconnected features of urban
systems for properly balancing their socioeconomic and environmental dynamics.
In this framework, it is of paramount importance to provide the decision mak-
ers with proper and reliable information and to directly engage the citizenry in

29
30 Remote Sensing for Sustainability

implementing suitable adaptation and mitigation actions. To this aim, in addition to

in situ measurements as well as socioeconomic and environmental variables, Earth
Observation (EO)–based products are also increasingly used at present as part of the
planning intelligence; indeed, they proved to be of great support to environmental
and climate managers in a variety of applications. Nevertheless, there still exist sev-
eral methodological and technological gaps between the pressing requirement from
the users and the current state-of-the-art methodologies. As an example, there is a
growing need for assimilating EO-based products into urban models or for facilitat-
ing the use of geo-spatial products by nonexperts as well.
This contribution introduces selected geo-information products and their related
application in support of local and regional urban planning, such as the assessment
of impervious urban surface, the mapping of urban growth, and the characteriza-
tion of settlement patterns. The corresponding studies were conducted for five regions
of interest, including Antwerp (Belgium), Helsinki (Finland), London (United
Kingdom), Madrid (Spain), and Milan (Italy).

2.2 SELECTED APPLICATIONS

2.2.1 Aseessment of ImpervIous urbAn surfAce
Urban growth is associated not only with the construction of new buildings but
also (and in more general terms) with a consistent increase of all the impervious
surfaces (hence including roads, parking lots, squares, pavements, and railroads as
well), which do not allow water to penetrate, forcing it to run off. Effectively map-
ping the impervious surface area is then of high importance because it is related
to the risk of urban floods, the urban heat island phenomenon, and the reduction
of ecological productivity. However, to date, this task has been mostly carried out
by photointerpretation of very high resolution (VHR) airborne optical imagery or
in situ surveys, which are generally costly and time-consuming and hence forbid
a systematic regional-scale mapping as well as regular updates. To this purpose,
EO data have started being used since they proved capable of improved detection
capabilities, larger-scale analysis, and lower costs. In this context, one of the cur-
rent state-of-the-art methodologies has been presented by Esch et al. (2009), which
allows one to automatically estimate the percentage of impervious surface (PIS)
by analyzing the Normalized Difference Vegetation Index (NDVI) calculated
from single-date Landsat Thematic Mapper scenes. Nonetheless, the performances
might strongly vary depending on (i) the quality of the Landsat image used for the
analysis and (ii) the availability of local railway and road network vector layers, as
well as VHR optical imagery needed to train the employed empirical regression
model based on support vector regression (SVR). To overcome such drawbacks,
we improved the above-described technique by (i) considering the mean temporal
NDVI calculated from a series of Landsat-8 (LS8) scenes acquired over the area
of interest in a 1-year time frame (which allows a drastic reduction in the effect of
vegetation phenology and prevents problems related to cloud coverage and shadow
in specific scenes) and (ii) deriving the training points for the SVR model starting
from OpenStreetMap data (OSM 2015). In particular, samples are extracted in
Earth Observation for Urban and Spatial Planning 31

selected areas from OpenStreetMap layers corresponding to impervious surfaces.

Next, they are first rasterized and then aggregated at the 30-m spatial resolution
of Landsat imagery. An SVR model is finally used to correlate the resulting train-
ing information with the mean temporal NDVI and hence estimate the PIS for the
whole study area. Specifically, only pixels denoted as urban in the Global Urban
Footprint (GUF) mask described by Esch et al. (2012) are preserved. An example
of the estimated PIS for Helsinki (Finland) based on a series of 25 LS8 scenes
collected for two path-row combinations covering the city in the entire 2014 is
illustrated in Figure 2.1.
A quantitative assessment of the effectiveness of the obtained results has been
carried out by means of WorldView-2 (WV2) multispectral scenes acquired at 2-m
spatial resolution and available for the five considered study cases (see Table 2.1). In
particular, given the spatial detail offered by the WV2 images, we could mark with a
very high degree of confidence all the impervious structures included in the different
study regions. To this aim, we first computed for each scene the NDVI and manually
identified the most suitable threshold that allows the exclusion of all the green areas
(i.e., nonimpervious); then, we refined the resulting mask by extensive photointerpre-
tation and aggregated it to 30-m spatial resolution. Finally, we compared the result-
ing WV2-based reference PIS to the corresponding portion of the 2014 PIS products
obtained with the proposed method.
To this aim, three different measures have been considered:

• The Pearson’s correlation coefficien, which measures the strength of the

linear relationship between two variables, and it is defined as the covariance
of the two variables divided by the product of their standard deviations; in
particular, it is largely used in the literature for validating the output of
regression models (as in our case).
• The Mean Error (ME), which is calculated as the difference between the
estimated value (i.e., the 2014 LS8-based PIS) and the reference value (i.e.,
the WV2-based reference PIS) averaged over all the pixels of the image.
• The Mean Absolute Error (MAE), which is calculated as the absolute dif-
ference between the estimated value (i.e., the 2014 LS8-based PIS) and the
reference value (i.e., the WV2-based reference reference) averaged over all
the pixels of the image.

The results of this comparison are reported in Table 2.2 and are extremely prom-
ising. Indeed, we obtained a mean correlation of 0.8271 and average ME and MAE
equal to −0.09 and 13.33, respectively, which confirms the great effectiveness of the
LS8-based PIS products. However, it is also worth pointing out that because of the
different acquisition geometries, the WV2 and LS8 images generally exhibit a very
small shift. Nevertheless, despite being limited, such displacement often results in a
one-pixel shift between the LS8-based PIS and the WV2-based reference PIS aggre-
gated at a 30-m resolution. This somehow affects the computation of the MAE and
of the correlation coefficient (which, however, resulted in highly satisfactory values).
Instead, the bias does not alter the ME, which always exhibited values close to 0, thus
confirming the capabilities of the implemented technique.
 32

Helsinki
% Impervious surface
100
0
0 2.5 5 10 km
Remote Sensing for Sustainability

FIGURE 2.1 PIS for Helsinki (FI), modeled on the basis of multitemporal indices derived from Landsat-8 TM data.
Earth Observation for Urban and Spatial Planning 33

TABLE 2.1
Acquisition Dates and Size of the WV2 Images Available for the Five Study
Sites
Acquisition Date (DD.MM.YYYY) Size (Pixel)
Antwerp 31.07.2014 5404 × 7844
Helsinki 21.04.2014 12,468 × 9323
London 28.08.2013 7992 × 8832
Madrid 20.12.2013 10,094 × 13,105
Milan 14.05.2014 8418 × 7957

TABLE 2.2
Pearson’s Correlation Coefficient, Mean Error (ME), and Mean Absolute
Error (MAE) Obtained from the Comparison of the 2014 PIS Obtained
with the Proposed LS8-Based Method and the WV2-Based Reference PIS
Pearson’s Correlation Coefficient ME MAE
Antwerp 0.8713 −2.69 12.07
Helsinki 0.7847 0.44 15.60
London 0.8094 −1.32 13.79
Madrid 0.8079 2.45 12.43
Milan 0.8623 0.67 12.76
Mean 0.8271 −0.09 13.33

2.2.2 mAppIng of urbAn growth

Reliably delineating the urban growth that occurred in the last decades is of great
importance to properly model the temporal evolution of urbanization and, hence,
to better estimate future trends and implement suitable planning strategies. In
Europe, all major cities have access nowadays to digital databases with highly
detailed information about the urban extent often down to the single-building level
(mostly derived from VHR optical airborne imagery taken every 2 to 5 years).
Nonetheless, the information available for the past 20 years is often of poor quality
because urban extent maps were mostly generated manually by in situ surveys at
that time.
For this purpose, we developed a novel approach for delineating the urban area
extent from ESA radar imagery, namely, ERS-1/2 SAR Precision Image and Envisat
ASAR Image Mode Precision products acquired between 1992 and 2012 at 30 ×
30 m spatial resolution with 12.5 m pixel distance (Marconcini et al. 2014). The cor-
responding block scheme is reported in Figure 2.2.
There are two main assumptions of the implemented method. On the one hand,
we suppose that for the investigated region of interest, a binary mask M is avail-
able, which delineates the current extent of urban areas, for example, from local
cadastral data or, alternatively, from GUF (Esch et al. 2012) or OpenStreetMap
 34
ERS PRI/ASAR IMP
multitemporal series
GUF

2012 M
A Masking Classification Urban
extent
t1 tN
S
tN
Pre-
Sigma0 Feature Speckle
processing
extraction divergence M
A Masking Classification
Urban
t1 tN tN tN extent
S
tN–1
Sigma0 Feature
extraction d
Urban
tN–1 tN–1 extent

t2

M
A Masking Classification
Urban
extent
S
t1
Sigma0 Feature Speckle
extraction divergence

t1 t1
Remote Sensing for Sustainability

FIGURE 2.2 Schematic view of the work flow for the spatiotemporal mapping of urbanization based on ERS/ASAR time series data.
Earth Observation for Urban and Spatial Planning 35

data (OSM 2015). Specifically, we solely consider built-up structures without lin-
ear elements as streets, roads, or railways because it is not feasible to properly
identify them by means of radar data. On the other hand, we assume that human
settlements in the study area experienced urban expansion rather than shrinkage
in the last decades, which is generally reasonable given the current global trend of
urbanization.
Given a multitemporal sequence of backscattering ERS/ASAR images available
for the investigated region in the periods of interests, we sort them by acquisition
date from the newest to the oldest; then, we apply calibration and terrain correction
to each scene and finally co-register the entire series. Next, we consider the first
item of the series A and, in line with the strategy adopted for deriving the GUF, we
extract its speckle divergence texture feature S, which represents an estimate of the
local true image texture and exhibits high values solely in correspondence of hetero-
geneous and highly structured built-up areas.
Under the considered working hypothesis, one pixel cannot be categorized as
belonging to the built-up class earlier in the time series, if this does not occur at a
later time. Accordingly, we mask A and S by using M and only keep those pixels
labeled as urban in M, which are then provided as input to the unsupervised clas-
sification scheme presented in Esch et al. (2013), which allows one to automatically
derive the corresponding built-up extent. This process is then iteratively applied to
the next images of the multitemporal sequence and, for each scene, the correspond-
ing A and S are masked by means of the built-up mask obtained for the previous item
of the series.
It might occur that, for specific periods of interest (e.g., 1 or 2 years), several
ERS/ASAR scenes have been acquired over the study area. In this case, to prop-
erly account for the generally stable behavior of the urban areas (mostly associated
with high backscattering values) compared to the other information classes (which
might show high values only under specific conditions), it might be beneficial to first
calculate the mean temporal backscattering and then derive its speckle divergence.
This allows an improvement in the performances and a reduction of the effect of the
speckle noise.
The outcome of the spatiotemporal urbanization mapping for Madrid (Spain)
is presented in Figure 2.3, showing the urban area extent derived from ERS-1,
ERS-2, and ASAR imagery acquired in 1992 (light brown), 2000 (red), 2006
(blue), and 2010 (dark blue). Moreover, the change polygons provided by the
European CORINE Land Cover data set (Figure 2.3a) are compared to the
changes identified based on the SAR time series (Figure 2.3b). Considering
the building block–related change indication, a block was only assigned as
change area if a sufficient proportion of change pixels were identified within the
corresponding area.
To assess the effectiveness of the proposed methodology, we considered the city
of Madrid as the representative validation case. Indeed, the municipality has a sur-
face of ~600 km2 and includes both high- and low-density residential areas, as well
as large portions of rural and forest areas. Accordingly, we evaluated the accuracies
of the corresponding built-up extent products generated for the years 2000, 2006,
 36

N 0 0.5 1 2 km 0 0.5 1 2 km
(e) (a) (b)

(a) (b)

(c) 0 0.5 1 2 km (d) 0 0.5 1 2 km

(c)

(d)

1900
2000 0 2.25 4.5 9 km
2006

FIGURE 2.3 Urban growth estimated for specific hot spots (a–d) and the entire city (e) of Madrid (Spain) as defined from CORINE data.
(Continued)
Remote Sensing for Sustainability
 N 0 0.5 1 2 km 0 0.5 1 2 km
(j) (f ) (g)

(f ) (g)

(h) 0 0.5 1 2 km (i) 0 0.5 1 2 km

(h)
Earth Observation for Urban and Spatial Planning

(i)

1992
2000
2006 0 2.25 4.5 9 km
2010

FIGURE 2.3 (CONTINUED) Urban growth estimated for specific hot spots (f–i) and the entire city (j) of Madrid (Spain) as defined from the pre-
sented methodology based on ERS/ASAR SAR time series data (12.5 m spatial resolution).
37
38 Remote Sensing for Sustainability

TABLE 2.3
OA%, PA%, UA%, and Kappa Coefficient Computed for the 2000, 2006,
and 2010 Built-Up Extent Products Generated for the City of Madrid (Spain)
PA% UA%
Non– Non–
Year OA% Built-Up Built-Up Built-Up Built-Up Kappa
2010 89.40 89.30 89.51 89.57 89.24 0.7880
2006 85.98 85.70 86.28 86.30 85.67 0.7197
2000 87.29 86.60 87.99 87.92 86.68 0.7459

and 2010, by means of VHR optical imagery from Google Earth. In particular, VHR
data are available for the following dates:

• September 15, 2000—suitable for the assessment of the year 2000 built-up
extent product
• July 5, August 5, and September 10, 2006—suitable for the assessment of
the year 2006 built-up extent product
• October 2, 2009—suitable for the assessment of the year 2010 built-up
extent product

In our analysis, we used a blind interpretation approach, where randomly selected

points have been labeled as urban/nonurban via photointerpretation before generat-
ing the built-up extent maps with the presented technique. For each of the three
considered years, we labeled 1000 built-up and 1000 non–built-up points randomly
distributed within the boundaries of the Madrid municipality. Specifically, these
validation points were chosen not to be falling within a buffer of 10 m between built-
up and non–built-up areas. Then, we compared the labels manually associated to
the resulting set of 2000 points against the corresponding ones of the built-up extent
products. Finally, different statistics were computed, namely, percentage overall
accuracy (OA%), percentage producer’s accuracy (PA%), percentage user’s accuracy
(UA%), and kappa coefficient (which also takes into consideration errors and their
type). Results reported in Table 2.3 assess the capabilities of the proposed technique,
which always exhibited very high accuracies.

2.2.3 chArActerIzAtIon of settlement pAttern

In order to define and implement effective adaptation and mitigation strategies for
a certain study area (from the local to the regional scale), it is important to properly
characterize its settlement pattern. This involves not only gathering a comprehensive
knowledge about form, size, and distribution of the corresponding types of settle-
ments but also quantitatively assessing their significance with respect to each other.
As an example, the local impact of a big but isolated city might be lower than that
of a smaller city connected to many others in its surrounding. In this framework, we
Earth Observation for Urban and Spatial Planning 39

implemented a novel technique that, by means of graph theory and spatial network
analysis, allows one to properly model the relationships between different settle-
ments and to quantitatively describe their mutual relevance (Esch et al. 2014). In par-
ticular, given a binary mask delineating urban and nonurban areas within the region
of interest, we first identify all settlement objects (defined as clusters composed of
pixels labeled as urban and connected via at least one edge or corner). Then, a spatial
network is created where the nodes are associated with the centroids of the differ-
ent objects, while the edges connect neighboring settlements within a predefined
Euclidean distance from each other. Next, several attributes are computed for each
node describing the geometrical properties of the corresponding settlement, for
example, area, perimeter, solidity, equivalent diameter, shape index, and eccentric-
ity. Likewise, weights are calculated for each edge characterizing the link between
the two connected objects, for example, minimum Euclidean distance, number of
crossed edges, and cohesion index (see Esch et al. 2014).
To characterize the impact of each node (and hence of the corresponding settle-
ment object), we finally compute different relevance indices that jointly take into
account its attributes as well as the weights associated with all its edges. For instance,
we considered the degree centrality (defined as the total number of edges connected
to a given node), the betweenness centrality (defined as the number of times that one
node is included in the shortest path between any two other nodes in the network),
or the local dominance (defined as the ratio between the degree centrality and the
number of edges for which the given node has a size greater than that of the neighbor
to which it is connected).
Many experimental trials have been carried out at different scales for assess-
ing the performances of the presented method, which proved to be an effective and
promising tool for supporting both quantitative and qualitative settlement pattern
analyses.
In Figure 2.4, we report the spatial network obtained for the greater London area
starting from the corresponding portion of the GUF data set (Esch et al. 2012) fur-
ther split up and subdivided by the administrative district boundaries (for this reason,
the city of London consists of different polygons associated with its 33 boroughs).
A minimum Euclidean distance of 1 km has been considered when computing the
edges. The size of different nodes is proportional to the area of the corresponding
settlement, whereas their color varies based on the resulting degree centrality. Here,
it is worth noting that nodes appearing in darker tones are those involved in the high-
est number of interactions and that most of them are located at the western side of
London (i.e., where then the city receives more pressure from the outside).

2.3 CONCLUSIONS AND OUTLOOK

The constantly increasing availability and accessibility of modern remote sensing
technologies provides new opportunities for a wide range of urban applications such
as mapping and monitoring of the urban environment (land cover, land use, mor-
phology, urban structural types), socioeconomic estimations (population density),
characterization of urban climate (microclimate, human health conditions), analy-
sis of regional and global impacts (groundwater and climate modeling, urban heat
 40

London
Degree centrality
(1 km)
0–7
7–12
12–18
18–24
24–33
33–46
46–63
63–78
78–94
94–122
122–165
0 5 10 20 km
165–221

FIGURE 2.4 Spatial network obtained for the Greater London area. The size of the nodes is proportional to the area of the corresponding settlement, whereas
Remote Sensing for Sustainability

their color varies depending on the computed degree centrality. Edges connect objects within a minimum Euclidean distance of 1 km from each other.
Earth Observation for Urban and Spatial Planning 41

islands), or urban security and emergency preparedness (sustainability, vulnerabil-

ity). In this contribution, we have introduced a selection of applications and example
products that have been developed for providing additional and innovative data that
can support day-to-day decision-making in urban and regional planning.
For most of the applications dealing with urban environments, the basic challenge
is related to the spectral heterogeneity and morphological complexity of built-up
areas. The spectral heterogeneity originates from the enormous diversity of differ-
ent materials forming the urban landscape. Thereby, some land cover types such as
vegetation, bare soil, or water are also found in nonurban environments. Moreover,
certain surfaces (e.g., bare soil and specific construction materials of buildings or
pavements) can hardly be differentiated from each other through their spectral sig-
nature. Regarding the morphological complexity, urban areas are characterized by
structural elements featuring diverse scales and shapes. In order to accurately cap-
ture the morphological properties of urban objects, a very high spatial resolution
of the sensor system is required. However, although an increased spatial resolution
certainly expands the spectrum of urban applications, this development comes along
with new challenges in terms of an automated image analysis. On the one hand,
the observable heterogeneity within the specific object types increases significantly
since many local but often nonrelevant characteristics appear (e.g., roof lights and
chimneys on top of buildings or cars, street furniture, and sign postings on streets).
On the other hand, urban features are hence formed by groups of pixels with similar
spectral signatures.
To address these challenges arising from an improved spatial resolution, recent
studies have increasingly used object-oriented analysis approaches. Compared to the
established pixel-based approaches, these techniques facilitate an improved consid-
eration of spectral, geometric, textural, contextual, and hierarchical characteristics.
The previous remarks regarding urban remote sensing stress that the suit-
able approach, technology, and data are highly dependent on the thematic focus
and the spatial scale of the analysis. Medium-resolution multispectral data (e.g.,
Landsat, Spot, and IRS-P6) are best suited for regional analyses since they cover
areas of up to 32,000 km² with one image ensuring cost-effective analyses. At the
same time, the spatial resolution is still sufficient to discriminate built-up areas
from nonurban regions based on spectral and textural characteristics. Because
of their direct link to morphological properties, high- and medium-resolution
SAR images provide particularly robust features for the detection of settlements.
However, the applicability of SAR data for local analysis of the urban structures
is still limited since the complex geometrical and physical characteristics of met-
ropolitan areas and the varying appearance and visibility of objects subject to
the line of sight lead to significant distortions of and ambiguities in the resulting
radar images.
To cope with the heterogeneity and complexity of urban areas, VHR multi-
spectral systems such as IKONOS or QuickBird are required. Indeed, their sen-
sors provide images in four spectral bands featuring a ground resolution of 4 m
(IKONOS) and 2.44 m (QuickBird) supplemented by a panchromatic channel with
a geometric resolution of 1 m (IKONOS) and 61 cm (QuickBird). Some drawbacks
of these data are the limitation of the spectral resolution to four bands—facilitating
42 Remote Sensing for Sustainability

only a very rough reconstruction of the spectral signature—and the limited spa-
tial coverage of a few hundred square kilometers by one image. Hence, analyses
of complete metropolitan areas and major or mega cities demand a data volume
that significantly increases the complexity and expense for image processing and
classification.
The immense spectral resolution of hyperspectral sensor systems enables the-
matically comprehensive and spatially detailed characterizations of the urban envi-
ronment. However, current hyperspectral sensor systems showing a spatial resolution
that is useful for urban applications are limited to airborne platforms. The first high-
resolution hyperspectral satellite sensor EnMAP is supposed to be launched by 2017.
This system will feature a spatial resolution of 30 m and cover the spectral range
of 420–2450 nm with approximately 200 bands. Heldens et al. (2011) reviewed 146
publications to give an outlook on the capabilities of the EnMAP mission for urban
applications.
The synchronism and coexistence of economic activities, environmental threats,
infrastructural deficits, poverty, and population growth mark a significant challenge
to urban planning. Therefore, future research has to focus on integrated interdisci-
plinary studies to understand the multidimensional and complex interactions of urban
systems and to analyze and assess the effects of plans, actions, and concepts. An
important step toward the improvement of the generated information products and
their acceptance by decision makers consists in the adaptation to holistic approaches
on complex urban systems. Hence, the appropriate concepts have to integrate and
correlate multiple analysis tools (image analysis software, geographic information
system), data types (satellite images, vector data and statistics), and data sources
(EO, in situ survey, census). The synergetic use of various data sources and their
combined analysis increases the quality and information content of the resulting
products, opens new levels of information, and enhances the possibilities of integrat-
ing the resulting data and information into existing systems and concepts. However,
in view of regional, national, or even global monitoring tasks, there is still some
effort needed with respect to the availability and accessibility of remote sensing data
and the operationalization of image processing in order to allow for cost- and time-
efficient analyses and a rapid provision of the required information. Thereby, new
sensor systems such as RapidEye and GeoEye will improve the capabilities of urban
remote sensing application, particularly in terms of providing detailed time series of
multispectral imagery.

ACKNOWLEDGMENTS
The results of the studies introduced in this chapter were generated in the context of
the FP7 DECUMANUS project. The SAR data used in this study was provided by
ESA based on the Category-1 proposal no. 16969.
Earth Observation for Urban and Spatial Planning 43

REFERENCES
Esch, T., V. Himmler, G. Schorcht, M. Thiel, T. Wehrmann, F. Bachofer, C. Conrad, M.
Schmidt, and S. Dech, Large-area assessment of impervious surface based on integrated
analysis of single-date Landsat-7 images and geospatial vector data. Remote Sensing of
Environment, 113, 1678–1690, 2009.
Esch, T., H. Taubenböck, A. Roth, W. Heldens, A. Felbier, M. Thiel, M. Schmidt, A. Müller,
and S. Dech. TanDEM-X mission-new perspectives for the inventory and monitoring
of global settlement patterns. Journal of Applied Remote Sensing, 6(1), 061702, 2012;
21 pp., doi: 10.1117/1.JRS.6.061702.
Esch, T., M. Marconcini, A. Felbier, A. Roth, W. Heldens, M. Huber, M. Schwinger, H.
Taubenböck, A. Müller, and S. Dech, Urban footprint processor—Fully automated pro-
cessing chain generating settlement masks from global data of the TanDEM-X Mission.
IEEE Geoscience and Remote Sensing Letters, 10(6), 1617–1621, 2013.
Esch, T., M. Marconcini, D. Marmanis, J. Zeidler, S. Elsayed, A. Metz, and S. Dech.
Dimensioning urbanization—An advanced procedure for characterizing human settle-
ment properties and patterns using spatial network analysis. Applied Geography, 55,
212–228, 2014.
Heldens, W., U. Heiden, T. Esch, E. Stein, and A. Müller. Can the future EnMAP mission
contribute to urban applications? A literature survey. Remote Sensing 3(9), 1817–1846,
2011.
Kamal-Chaoui, L. and R. Alexis. Competitive Cities and Climate Change, OECD Regional
Development Working Papers no. 2, OECD Publishing, 2009.
Marconcini, M., A. Metz, T. Esch, and J. Zeidler, Global urban growth monitoring by means
of SAR data. Proceedings of the IEEE International Geoscience and Remote Sensing
Symposium (IGARSS 2014), Quebec, Canada, July 13–18, 2014.
OpenStreetMap: http://www.openstreetmap.org/ December 21, 2015.
United Nations, Human Settlements Programme, Cities and Climate Change: Global Report
on Human Settlements 2011, Earthscan, 2011.
United Nations, Department of Economic and Social Affairs, Population Division. World
Urbanization Prospects: The 2014 Revision, Highlights (ST/ESA/SER.A/352), 2014.
 3 Assessment of Urban
Growth in the Pearl River
Delta, China, Using Time
Series Landsat Imagery
Lei Zhang and Qihao Weng

CONTENTS
3.1 Introduction .................................................................................................... 45
3.2 Case Study ......................................................................................................46
3.2.1 Study Area ..........................................................................................46
3.2.2 Methodology ....................................................................................... 48
3.2.2.1 Data Preprocessing .............................................................. 48
3.2.2.2 Calculation of BCI and LST ................................................ 49
3.2.2.3 Gap Filling and Smoothing.................................................. 49
3.2.2.4 Selection of Stable Time Series ........................................... 50
3.2.2.5 Semi-Supervised KFCM Algorithm.................................... 50
3.3 Results............................................................................................................. 52
3.3.1 Quantitative Characteristics of Urban Growth ................................... 52
3.3.2 Assessment of Urban Growth ............................................................. 55
3.3.3 Clustering Accuracy Assessment ....................................................... 55
3.4 Conclusions ..................................................................................................... 57
References ................................................................................................................ 58

3.1 INTRODUCTION
Urban growth has a significant influence on urban environments, including climate
change (Liao et al. 2015; Pathirana et al. 2014), biochemical cycles (Hutyra et al. 2014),
and environment quality (Panagopoulos et al. 2015; Zhao et al. 2015). Assessment
of urban growth is needed for sustainable development and studies on ecological con-
sequences. Since remote sensing technology provides spatially consistent data with
high spatial resolution and high temporal frequency, remote sensing imagery makes
it possible to analyze and model urban growth over long periods at various scales in a
timely and cost-effective manner. Multitemporal coarse or medium spatial resolution
imagery has been commonly applied for analysis of urban growth, such as DMSP/
OLS (Defense Meteorological Satellite Program/Operational Linescan System)
nighttime light data (Liu et al. 2012; Ma et al. 2012; Zhang and Seto 2011), Landsat

45
46 Remote Sensing for Sustainability

archive (Bagan and Yamagata 2012; Michishita et al. 2012; Sun et al. 2013; Xian and
Crane 2005), China–Brazil Earth Resources Satellites images, and HJ-1 images (Du
et al. 2015). The main drawback to using coarse or moderate spatial resolution imag-
ery is the mixed pixel problem, which leads to a salt-and-pepper effect caused by
spectral heterogeneity. Multitemporal high spatial resolution imagery has also been
used to extract urban areas, such as Spot 5 imagery (Durieux et al. 2008; Jacquin et
al. 2008) and RapidEye and IRS data (Dupuy et al. 2012). Although these methods
have been applied to analyze urban growth successfully, the successive imagery with
high temporal resolution was difficult to obtain. Additionally, the intraclass spectral
variability problem is inevitable in high spatial resolution imagery.
Multitemporal analysis for urban growth usually requires single classification or
segmentation of all the stacked images and is limited to provide detailed change
information because of low temporal resolution. Therefore, time series imagery
applied in differentiating land cover has attracted increased attention from research-
ers in recent years, because temporal domain has showed its advantages in resolving
class confusion between classes with similar spectral characteristics (Bhandari et
al. 2012; Schneider 2012). Specifically, Landsat time series have been successfully
applied to map dynamics of urban areas because of their long record of continuous
measurement at effective spatial resolution and temporal frequency (Gao et al. 2012;
Li et al. 2015; Sexton et al. 2013a,b). However, these methods focused on spectral
differences or temporal consistency after classification. Little attention was paid to
temporal data mining method to differentiate urban areas from other land cover
using dense time series Landsat images.
Since time series clustering has been shown to be effective in time series data min-
ing (Fu 2011; Liao 2005), in this study, we aimed at extracting urban areas using a semi-
supervised fuzzy time series clustering method through the Biophysical Composition
Index (BCI) (Deng and Wu 2012) and Land Surface Temperature (LST) time series
and applied the method to the Pearl River Delta, China, from 1987 to 2014. Kernel
fuzzy C-means (KFCM), proposed by Zhang and Chen (2003), was introduced in this
study, because it could provide a more robust signal-to-noise ratio and is less sensi-
tive to cluster shapes in comparison to other clustering algorithms (Du et al. 2005).
BCI and LST time series images were derived because of their strong correlation with
urban areas. BCI aimed to identify different urban biophysical compositions and has
been demonstrated to be effective in identifying the characteristics of impervious sur-
faces and vegetation and in distinguishing bare soil from impervious surfaces. LST,
as a significant parameter in urban environmental analysis, tends to be positively cor-
related with urban expansion (Weng and Hu 2008; Yuan and Bauer 2007).

3.2 CASE STUDY

3.2.1 Study AreA
The Pearl River Delta, as the third most important economic district of China, is
located in the developmental core region of Guangdong Province, between 21°N–23°N
and 111°E–115°E. It has experienced rapid urbanization since the reform process
started in the late 1970s in China. The Pearl River Delta has a subtropical climate
Assessment of Urban Growth in the Pearl River Delta, China 47

with an average annual temperature of 21°C–23°C, including a dry season from

October to April and a wet season from May to September.
Quantifying and analyzing the urban growth in the Pearl River Delta are important
to characterize the effects of anthropogenic activities on urban environments during
the past years. In our study, 239 Landsat images, covering the period from 1987 to
2014, were used to monitor urban area dynamics. Landsat imagery, including TM,
ETM+ (including SLC-off data), and OLI data with cloud cover less than 50%, was
ordered and downloaded from the USGS Earth Explorer (Reference system: WRS-2,
Path: 122, Row: 44). A clipped region from Landsat imagery with an area of 16,824
km2, covering five cities, Guangzhou, Foshan, Zhongshan, Dongguan, and Shenzhen,
was used to monitor urban growth. These five cities were the most developed cities in
the Pearl River Delta. The geographic location of the study area is shown in Figure 3.1.

113°E 113°30´E 114°E

660000 680000 700000 720000 740000 760000 780000 800000 820000 840000

640000
2640000

620000
2620000
2600000

600000
23°30´N

23°30´N
2580000

530000
2560000

560000
2540000

23°N
23°N

540000
2520000

520000
2500000

500000
22°30´N

22°30´N
2480000

480000
2460000

460000

660000 680000 700000 720000 740000 760000 780000 800000 820000 840000
113°E 113°30´E 114°E
N
Projection: UTM, Zone 49N
W E 0 5 10 20 30 40 Pixel size: 30 m
km Datum: WGS-84
S
Ellipsoid: WGS_1984

FIGURE 3.1 Geographical location of the case study area, Pearl River Delta.
48 Remote Sensing for Sustainability

Time series Landsat images

Time series BCI images Time series LST images

Gap filling and smoothing

Stable time series

Labeled time series Semi-supervised kernel-based fuzzy C-means

Annual urban growth patterns

Accuracy assessment

FIGURE 3.2 Procedures for mapping annual urban areas in the Pearl River Delta using
time series Landsat images.

3.2.2 Methodology
This study intended to quantify the spatiotemporal patterns of urban areas in the
Pearl River Delta using dense time series Landsat images from 1987 to 2014. In
this study, the proposed method included five steps (Figure 3.2): First, time series
Landsat subsets were registered in the same projection and converted to sur-
face reflectance using the Landsat Ecosystem Disturbance Adaptive Processing
System (LEDAPS) (Masek et al. 2006). Then, cloud, cloud shadow, and snow
were masked and their values were set to null. Next, the BCI and LST time
series were computed from preprocessed data. Gap filling (Garcia 2010; Wang
et al. 2012) and smoothing were adopted to fill missing values in the time series.
After that, stable time series were selected based on time series decomposition
(Verbesselt et al. 2010). Finally, the semi-supervised KFCM algorithm was per-
formed to clustering time series to map annual urban areas during the study
period.

3.2.2.1 Data Preprocessing

All the Landsat data were registered to the 1984 World Geodetic System Universal
Transverse Mercator (WGS-84 UTM) Zone 49 North projection system and resam-
pled at 30-m spatial resolution. Each Landsat subset was standardized to surface
reflectance using the LEDAPS method (Masek et al. 2006), which applied MODIS
(moderate-resolution imaging spectroradiometer) atmospheric correction routines to
Landsat L1 data products. Cloud, cloud shadow, and snow mask were calculated
using the Fmask algorithm for all scenes (Zhu and Woodcock 2012). The locations
of SLC-off data were identified using band-specific gap mask files in each SLC-off
data product.
Assessment of Urban Growth in the Pearl River Delta, China 49

3.2.2.2 Calculation of BCI and LST

BCI (Deng and Wu 2012), involving Tasseled Cap (TC) transformation and the
V–I–S triangle model, was given as

TC1 + TC3
− TC2
BCI = 2 , (3.1)
TC1 + TC3
+ TC2
2

where TCi (i = 1, 2, and 3) were three normalized TC components: TC1 was high
albedo, TC3 was low albedo, and TC2 was vegetation. Each derived TC component
was then linearly normalized to the range from 0 to 1.
The LST was calculated using the radiative transfer equation method (Sobrino et
al. 2004):

[ Lλ − L ↑ − τ(1 − ε) L ↓]
BT = , (3.2)
τε

where BT was the atmospherically corrected pixel value as brightness temperature.

L↑, L↓, and τ, derived from an atmospheric correction tool (Barsi et al. 2005), were
upwelling radiance, downwelling radiance, and transmittance, respectively. ε was
emissivity, derived from NDVI. Lλ was the pixel values as radiance derived from
digital numbers DN:

Lλ = gain × DN + bias, (3.3)

where gain and bias were gain value and bias value for the specific band, respectively.
Then, the radiance was converted to surface temperature (Chander and Markham
2003):

K2
T= , (3.4)
 K1 
ln  +1
 BT 

where T was the temperature in Kelvin (K) and K1 and K2 were prelaunch calibra-
tion constants. For Landsat 5 TM, K1 = 607.76 W/(m2 sr μm) and K2 = 1260.56 K; for
Landsat 7 ETM+, K1 = 666.09 W/(m2 sr μm) and K2 = 1282.71 K; for Landsat TIRS
10, K1 = 774.89 W/(m2 sr μm) and K2 = 1321.08 K.

3.2.2.3 Gap Filling and Smoothing

Discontinuities existed in the time series Landsat data due to missing values caused
by cloud cover and SLC-off data, which made uncertainties under incomplete time
series in a subsequent analysis. Gap filling and smoothing were needed to improve the
continuity and consistency in the time series. The gap filling method (Garcia 2010;
50 Remote Sensing for Sustainability

Wang et al. 2012), as a penalized least square regression based on three-dimensional

discrete cosine transform (DCT-PLS), was adopted to predict missing values. It used
information from both spatial and temporal variability to provide robust gap filling
and it required no ancillary data sets such as alternative geospatial data sets or digital
elevation models to model missing values. A smoothing gap filled time series was
essential to improve the accuracy of the phenology derived from the reconstructed
time series (Atkinson et al. 2012). Fourier fitting, as a most common and useful
method, was adopted for smoothing gap filled time series BCI and LST data. Fourier
analysis has showed promise in monitoring interannual vegetation changes (Geerken
2009). For the BCI and LST time series, Fourier fitting reduced the amount of noise,
which mitigated the effects of outliers, anomalies, and spurious values in the time
series.

3.2.2.4 Selection of Stable Time Series

Stable time series were derived from BCI time series. Stable time series means
time series of labeled land covers, which have not experienced land cover transition
during the study period. Since there was no exponential growth in the time series,
an additive decomposition model was applied to separate BCI time series into three
distinct components:

T = Tt + St + It, (3.5)

where T was the observed data at time t, Tt was a nonseasonal secular trend com-
ponent, St was a seasonal component, and It was an irregular component. Tt was
estimated using the regression model as

Tt = β 0 + β1t + β2t. (3.6)

β 0, β1, and β2 were regression coefficients. The seasonal component St was derived
from the detrend time series using a parametric regression model. Detrend time series
were computed by subtracting the trend component Tt from the original time series.
Given the trend component Tt and the seasonal component St, the irregular compo-
nent was estimated as

It = T − Tt − St. (3.7)

The time series with the trend component as constant and without obvious phe-
nology circle in seasonal variables were selected as stable time series.

3.2.2.5 Semi-Supervised KFCM Algorithm

Time series BCI and LST images were processed using a semi-supervised KFCM
to obtain clustering results. Given time series data X = {x1, x 2,…, xn}, xk ∈ Rd(k =
1, 2,…,n), where d was temporal dimension and n was the number of samples.
Assessment of Urban Growth in the Pearl River Delta, China 51

KFCM partitions X into c fuzzy subsets by minimizing the following objective

function:

c n

Jm (U , V ) = 2 ∑ ∑ u (1 − K (x , v )),
i =1 k =1
m
ik k i (3.8)

where c was the number of clusters, vi was the ith cluster centroid, uik was the mem-
bership of xk in class i, and Σ i uik = 1; m ∈[1, +∞] was the weighting exponent deter-
mining the fuzziness of the clusters. K(xk, vi) was the kernel function, aiming to map
xk from the input space X to a new space with higher dimensions. In this study, radial
basis function kernel was adopted:

(
K ( x k , vi ) = exp − x k − vi /σ 2 ,
2
) (3.9)

where the parameter σ was computed by

 2 
∑
n

1 xi − m 
σ=  i =1
 . (3.10)
c n

In order to search for new clusters, the objective function was minimized:

c n

Min Jm (U , V ) = 2 ∑ ∑ u (1 − K (x , v ))
i =1 k =1
m
ik k i (3.11)

s.t. ∑u
i =1
ik = 1, k = 1, 2,, n. (3.12)

The Lagrange function converted the constrained objective as an unconstrained

optimization model. By optimizing the objective function, the membership uik and
centroid vi could be updated:

1
 1  m−1
 (1 − K ( x , v )) 
k i
uik = 1
(3.13)
  m−1
∑ 1
c

j =1  (1 − K ( x k , vi )) 
 
52 Remote Sensing for Sustainability

∑ u K (x , v )x
n
m
ik k i k
v = k =1
. (3.14)
∑ u K (x , v )
i n
m
ik k i
k =1

Labeled time series samples were derived from stable time series, and unlabeled
samples were derived from the remaining time series. Given time series data X con-
sisted of Xl and Xu, Xl was labeled samples and Xu was unlabeled samples. “l” and “u”
indicate labeled or unlabeled data, respectively.
The whole process of semi-supervised KFCM algorithm was shown as follows:

1. Initialize the values of σ and uik using Xl and Xu. For Xl, the value of compo-
nent uik was set to 1 if the data xk were labeled with class i, and 0 otherwise.
For Xu, positive random values within [0,1] were set to unlabeled data. The
( )
m
Σ nk ′=1 uikl x kl
initial set of centroid vi was calculated as vi =
0
, where n′ was
( )
m
the number of labeled data. Σ nk ′=1 uikl

2. Update the membership uik in Xu and centroid vi until the objective function
was minimized.

Finally, inconsistent labeled pixels were mapped comparing the LST L and BCI
B clustering results. For those pixels, if the maximum membership max(uik)L of the
pixel k in L was higher than max(uik)B in B, the pixel was labeled as the class with
max(uik)L in L, and vice versa. However, if the values were equal, the pixel was
labeled as the class with max(uik)L .

3.3 RESULTS
3.3.1 QuAntitAtive ChArACteriStiCS of urbAn growth
The urban area distributions for the Pearl River Delta from 1987 to 2014 are shown
in Figure 3.3. The dark gray represents urban area, the medium gray shows water
bodies, and the light gray shows nonurban area. Because image numbers in 1989,
1992, 1997, and 1998 were fewer than 3 and cloud cover was more than 50% for all
images, urban areas in these 4 years were not analyzed.
The annual urban area maps in the Pearl River Delta in Figure 3.3 show a dra-
matic urban expansion from 1987 to 2014. Urban areas increased from 598 km2 in
1987 to 5768 km2 in 2014. To evaluate urban growth in the study area, urban areas by
year were calculated from annual clustering maps in Figure 3.4. The spatial distribu-
tions of annual urban areas could be divided into four periods.

1. For the period 1987–1991, the Pearl River Delta experienced no significant
change in urban areas. The Pearl River Delta was in the early phases of
development during this period. Urban areas increased from 3.56% of the
study area in 1987 to 5.97% in 1993, with an annual average rate of 13.57%.
The average increased urban area per year was 101.53 km2/year.
Assessment of Urban Growth in the Pearl River Delta, China 53

1987 1988 1990 1991

1993 1994 1995 1996

1999 2000 2001 2002

2003 2004 2005 2006

2007 2008 2009 2010

2011 2012 2013 2014

FIGURE 3.3 Urban areas from 1987 to 2014 in the Pearl River Delta.

2. For the period 1993–2000, large urban areas formed within and around
existing urban areas. The Pearl River Delta experienced its first rapid
growth period as a result of the deep development of China’s reform and
opening up. Urban areas increased from 6.81% in 1993 to 19.07% in 2000
with an annual average rate of 22.52%. The average increased urban area
per year was 412.62 km2/year.
 54

7000
5768.07
6000 5416.39 5521.08 5624.82
4909.33 4962.11 5144.95
5000 4520.44 4768.12 4791.91
4168.19
4000 3672.34 3849.43
3208.43 3396.20
3000 2526.27
2065.25
2000 1395.65 1701.85
1004.72 1145.35
598.59 848.59 948.23

Urban areas (km2)

1000
0
1987 1988 1990 1991 1993 1994 1995 1996 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Year

FIGURE 3.4 Annual urban areas in the Pearl River Delta, from 1987 to 2014.
Remote Sensing for Sustainability
Assessment of Urban Growth in the Pearl River Delta, China 55

3. For the period 2001–2006, the Pearl River Delta was still experiencing a
rapid growth, but the urban expansion was slower than that in the previous
period. In this period, the Southeast Asian economic crisis affected urban
development to some degree. The driving forces of continuous urban expan-
sion were mainly because China had a fast-growing economy since it joined
the World Trade Organization in 2001. Urban areas increased from 20.19%
in 2000 to 28.34% in 2006, yielding an annual average rate of 6.73%. The
average increased urban area per year was 228.65 km2/year.
4. For the period 2007–2014, Pearl River Delta entered a period of stable devel-
opment and urban development was slowed down. Urban areas increased
from 28.48% in 2007 to 34.28% in 2014 with an annual average rate of
2.55%. The average increased urban area per year was 122.02 km2/year.

3.3.2 ASSeSSMent of urbAn growth

Since the 1980s, rapid urbanization in the Pearl River Delta was strongly linked to
economic growth and population attributed to reform policies. From 1980 to 2000,
the average annual gross domestic product (GDP) growth of the Pearl River Delta
grew up to 16.9%. In this period, the driving force of urban growth in the study area
was the establishment of Special Economic Zones of Shenzhen, followed by foreign
investment from overseas investors. Since the new century, the Pearl River Delta
has been continuing to experience strong economic growth with an average annual
GDP growth of up to 15% between 2000 and 2007. However, the GDP growth has
slowed down since 2008 because of the global financial crisis and industrial struc-
ture adjustment. The GDP growth particularly slowed down to below 10% since
2011. According to the Hong Kong Trade Development Council, the Pearl River
Delta enjoyed a per-capita GDP of RMB 93,114 in 2013 (approximately $15,000),
with real GDP growing by an average of 9.4%.
The urban growth types differed during four periods: 1987–1991, 1993–2000,
2001–2006, and 2007–2014. In the 1987–1991 period, urban growth was domi-
nated by scattered development. In the 1993–2000 period, scattered development
was decreased and strip development was increased. Urban areas mainly expanded
along transportation networks. According to the fifth national population census,
the population of the Pearl River Delta was 4 million in 2000, rising by 1.94 million
when compared to the population in 1990. From 2001 to 2014, strip development and
compact development became dominant. City clusters and metropolitan stretches
came into being in the Pearl River Delta. Accompanying the urbanization process
was population growing to 47.72 million in 2008. Since 2011, the population growth
affected by economic growth and industrial structure adjustment began to stabilize
(Pearl River Delta Region Planning and Guangdong Statistical Yearbook 2014).

3.3.3 CluStering ACCurACy ASSeSSMent

In this study, historical imagery acquired from Google Earth was used as reference
data for each year. Since available historical imagery cannot cover the whole study
 56

100
91.32 90.86 91.18 89.63 89.51 89.76 90.54 91.26
90 88.23 88.45 88.32 89.22 88.17 89.74 90.05
85.33 84.1
81.65 80.12 81.34
79.16 78.23 80.64 78.9
80

70

Overall accuracy (%)

60

50
1987 1988 1990 1991 1993 1994 1995 1996 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Year

FIGURE 3.5 Annual clustering accuracy of the proposed method.

Remote Sensing for Sustainability
Assessment of Urban Growth in the Pearl River Delta, China 57

area and historical imagery from Google Earth for some period was few, reference
samples were also selected from time series BCI images using visual judgments.
Stable time series derived from stacked cloud free BCI images were used as refer-
ence data. Then, the stratified random sampling method was employed for selecting
reference data for each year. Five hundred samples were randomly selected to each
class and divided into two subsets. One subset was used for classifier training and the
second was used for accuracy assessment. It would help eliminate the bias resulting
from using the same samples for both training and testing. The annual clustering
accuracy is shown in Figure 3.5.
The annual clustering accuracy yielded 78.23% to 91.32%, which shows the
effectiveness and feasibility of the time series clustering method. However, the
average clustering accuracy in the 1987–1999 period was 81.05%, and the average
clustering accuracy in the 2000–2014 period was 89.75%. The reason was that dif-
ferent temporal dimensions in each year led to different clustering accuracy. The
available annual average image number before 2000 was smaller than 3, but the
available annual average image number after 2000 was larger than 14. The main
reason for this phenomenon was that the low temporal resolution in each year could
obscure land cover changes and reduce the separability of temporal characteristics
for urban areas and nonurban areas. The vegetation phenology characteristic would
particularly be weakened. Time series clustering also showed values of imagery
with cloud contamination or SLC-off data in identifying urban areas. Although
cloud contamination and SLC-off data caused significant temporal noise and
resulted in incomplete time series, gap filling and smoothing were helpful for solv-
ing missing data problems through enhancing temporal resolution of time series
Landsat imagery.

3.4 CONCLUSIONS
This chapter explores urban growth in the Pearl River Delta using time series Landsat
imagery from 1987 to 2014 based on the time series clustering method. The annual
spatiotemporal patterns of urban areas in the Delta were quantified. The results indi-
cated that urban areas in study areas increased rapidly from 3.56% to 34.28% during
the 1987–2014 period. The method proposed in this chapter verifies the feasibility
and effectiveness of the time series clustering method in assessing urban growth
patterns using time series Landsat imagery. Results from this study can be used by
policymakers for urban planning and management and for hydrological modeling to
determine the effect of increasing urban areas on urban environments of the study
area. This study can also be valuable for exploring the mechanisms of urban areas
and environmental relationship for sustainable urban planning and management.
However, how to evaluate the differences between two time series and how time
series components affect time series clustering remain unanswered in this study.
Future studies may include similarity metrics before using the clustering method to
solve the above issues.
58 Remote Sensing for Sustainability

REFERENCES
Atkinson, P. M., Jeganathan, C., Dash, J., & Atzberger, C. (2012). Inter-comparison of four
models for smoothing satellite sensor time-series data to estimate vegetation phenology.
Remote Sensing of Environment, 123, 400–417.
Bagan, H., & Yamagata, Y. (2012). Landsat analysis of urban growth: How Tokyo became the
world’s largest megacity during the last 40years. Remote Sensing of Environment, 127,
210–222.
Barsi, J. A., Schott, J. R., Palluconi, F. D., & Hook, S. J. (2005, August). Validation of a
web-based atmospheric correction tool for single thermal band instruments. In Optics &
Photonics 2005 (pp. 58820E–58820E). International Society for Optics and Photonics.
Bhandari, S., Phinn, S., & Gill, T. (2012). Preparing Landsat Image Time Series (LITS) for
monitoring changes in vegetation phenology in Queensland, Australia. Remote Sensing,
4(6), 1856–1886.
Chander, G., & Markham, B. (2003). Revised Landsat-5 TM radiometric calibration pro-
cedures and postcalibration dynamic ranges. IEEE Transactions on Geoscience and
Remote Sensing, 41(11), 2674–2677.
Deng, C., & Wu, C. (2012). BCI: A biophysical composition index for remote sensing of urban
environments. Remote Sensing of Environment, 127, 247–259.
Du, W., Inoue, K., & Urahama, K. (2005). Robust kernel fuzzy clustering. In Fuzzy systems
and Knowledge Discovery (pp. 454–461). Springer Berlin Heidelberg.
Du, P., Xia, J., & Feng, L. (2015). Monitoring urban impervious surface area change using
China–Brazil Earth Resources Satellites and HJ-1 remote sensing images. Journal of
Applied Remote Sensing, 9(1), 096094–096094.
Dupuy, S., Barbe, E., & Balestrat, M. (2012). An object-based image analysis method for
monitoring land conversion by artificial sprawl use of RapidEye and IRS data. Remote
Sensing, 4(2), 404–423.
Durieux, L., Lagabrielle, E., & Nelson, A. (2008). A method for monitoring building
construction in urban sprawl areas using object-based analysis of Spot 5 images and
existing GIS data. ISPRS Journal of Photogrammetry and Remote Sensing, 63(4),
399–408.
Fu, T. C. (2011). A review on time series data mining. Engineering Applications of Artificial
Intelligence, 24(1), 164–181.
Gao, F., de Colstoun, E. B., Ma, R., Weng, Q. et al. (2012). Mapping impervious sur-
face expansion using medium-resolution satellite image time series: A case study in
the Yangtze River Delta, China. International Journal of Remote Sensing, 33(24),
7609–7628.
Garcia, D. (2010). Robust smoothing of gridded data in one and higher dimensions with miss-
ing values. Computational Statistics & Data Analysis, 54(4), 1167–1178.
Geerken, R. A. (2009). An algorithm to classify and monitor seasonal variations in vegeta-
tion phenologies and their inter-annual change. ISPRS Journal of Photogrammetry and
Remote Sensing, 64(4), 422–431.
Hutyra, L. R., Duren, R., Gurney, K. R., Grimm, N., Kort, E. A., Larson, E., & Shrestha, G.
(2014). Urbanization and the carbon cycle: Current capabilities and research outlook
from the natural sciences perspective. Earth’s Future, 2(10), 473–495.
Jacquin, A., Misakova, L., & Gay, M. (2008). A hybrid object-based classification approach
for mapping urban sprawl in periurban environment. Landscape and Urban Planning,
84(2), 152–165.
Li, X., Gong, P., & Liang, L. (2015). A 30-year (1984–2013) record of annual urban dynamics
of Beijing City derived from Landsat data. Remote Sensing of Environment, 166, 78–90.
Liao, T. W. (2005). Clustering of time series data—A survey. Pattern Recognition, 38(11),
1857–1874.
Assessment of Urban Growth in the Pearl River Delta, China 59

Liao, J., Wang, T., Jiang, Z., Zhuang, B. et al. (2015). WRF/Chem modeling of the impacts of
urban expansion on regional climate and air pollutants in Yangtze River Delta, China.
Atmospheric Environment, 106, 204–214.
Liu, Z., He, C., Zhang, Q., Huang, Q., & Yang, Y. (2012). Extracting the dynamics of
urban expansion in China using DMSP-OLS nighttime light data from 1992 to 2008.
Landscape and Urban Planning, 106(1), 62–72.
Ma, T., Zhou, C., Pei, T., Haynie, S., & Fan, J. (2012). Quantitative estimation of urbanization
dynamics using time series of DMSP/OLS nighttime light data: A comparative case
study from China’s cities. Remote Sensing of Environment, 124, 99–107.
Masek, J. G., Vermote, E. F., Saleous, N. E., Wolfe, R., Hall, F. G., Huemmrich, K. F., Gao, F.,
Kutler, J., & Lim, T. K. (2006). A Landsat surface reflectance dataset for North America,
1990–2000. IEEE Geoscience and Remote Sensing Letters, 3(1), 68–72.
Michishita, R., Jiang, Z., & Xu, B. (2012). Monitoring two decades of urbanization in the
Poyang Lake area, China through spectral unmixing. Remote Sensing of Environment,
117, 3–18.
Panagopoulos, T., Duque, J. A. G., & Dan, M. B. (2015). Urban planning with respect to
environmental quality and human well-being. Environmental Pollution. DOI:10.1016/j
.envpol.2015.07.038.
Pathirana, A., Denekew, H. B., Veerbeek, W., Zevenbergen, C., & Banda, A. T. (2014). Impact
of urban growth-driven landuse change on microclimate and extreme precipitation—
A sensitivity study. Atmospheric Research, 138, 59–72.
Schneider, A. (2012). Monitoring land cover change in urban and peri-urban areas using dense
time stacks of Landsat satellite data and a data mining approach. Remote Sensing of
Environment, 124, 689–704.
Sexton, J. O., Song, X. P., Huang, C., Channan, S., Baker, M. E. et al. (2013a). Urban growth
of the Washington, DC–Baltimore, MD metropolitan region from 1984 to 2010 by
annual, Landsat-based estimates of impervious cover. Remote Sensing of Environment,
129, 42–53.
Sexton, J. O., Urban, D. L., Donohue, M. J., & Song, C. (2013b). Long-term land cover
dynamics by multi-temporal classification across the Landsat-5 record. Remote Sensing
of Environment, 128, 246–258.
Sobrino, J. A., Jiménez-Muñoz, J. C., & Paolini, L. (2004). Land surface temperature retrieval
from LANDSAT TM 5. Remote Sensing of Environment, 90(4), 434–440.
Sun, C., Wu, Z. F., Lv, Z. Q., Yao, N., & Wei, J. B. (2013). Quantifying different types of urban
growth and the change dynamic in Guangzhou using multi-temporal remote sensing data.
International Journal of Applied Earth Observation and Geoinformation, 21, 409–417.
Verbesselt, J., Hyndman, R., Zeileis, A., & Culvenor, D. (2010). Phenological change detec-
tion while accounting for abrupt and gradual trends in satellite image time series. Remote
Sensing of Environment, 114(12), 2970–2980.
Wang, G., Garcia, D. et al. (2012). A three-dimensional gap filling method for large geophysi-
cal datasets: Application to global satellite soil moisture observations. Environmental
Modelling & Software, 30, 139–142.
Weng, Q., & Hu, X. (2008). Medium spatial resolution satellite imagery for estimating and
mapping urban impervious surfaces using LSMA and ANN. IEEE Transactions on
Geoscience and Remote Sensing, 46(8), 2397–2406.
Xian, G., & Crane, M. (2005). Assessments of urban growth in the Tampa Bay watershed
using remote sensing data. Remote Sensing of Environment, 97(2), 203–215.
Yuan, F., & Bauer, M. E. (2007). Comparison of impervious surface area and normalized dif-
ference vegetation index as indicators of surface urban heat island effects in Landsat
imagery. Remote Sensing of Environment, 106(3), 375–386.
Zhang, D. Q., & Chen, S. C. (2003, June). Kernel-based fuzzy and possibilistic c-means clustering.
In Proceedings of the International Conference Artificial Neural Network (pp. 122–125).
60 Remote Sensing for Sustainability

Zhang, Q., & Seto, K. C. (2011). Mapping urbanization dynamics at regional and global scales
using multi-temporal DMSP/OLS nighttime light data. Remote Sensing of Environment,
115(9), 2320–2329.
Zhao, W., Zhu, X., Sun, X., Shu, Y., & Li, Y. (2015). Water quality changes in response to
urban expansion: Spatially varying relations and determinants. Environmental Science
and Pollution Research, 1–15.
Zhu, Z., & Woodcock, C. E. (2012). Object-based cloud and cloud shadow detection in
Landsat imagery. Remote Sensing of Environment, 118, 83–94.
 4 InSAR Monitoring
of Land Subsidence
for Sustainable
Urban Planning
Abduwasit Ghulam, Mark Grzovic,
Maitiniyazi Maimaitijiang, and Mamat Sawut

CONTENTS
4.1 Introduction .................................................................................................... 61
4.2 Study Area and Data ....................................................................................... 63
4.2.1 Study Area .......................................................................................... 63
4.2.2 Data ..................................................................................................... 65
4.3 Methods .......................................................................................................... 65
4.3.1 SBAS Processing ................................................................................66
4.3.2 PSI Processing .................................................................................... 67
4.4 Results and Discussion ................................................................................... 67
4.4.1 Spatial and Temporal Pattern of Land Subsidence from 1992 to 2000 ... 67
4.4.2 Spatial and Temporal Pattern of Land Subsidence from 2007 to 2011 ....71
4.4.3 Validation of InSAR Deformation Using Leveling Data.................... 73
4.4.4 Implications for Sustainable Urban Planning ..................................... 74
4.5 Conclusion ...................................................................................................... 75
Acknowledgment ..................................................................................................... 76
References ................................................................................................................ 76

4.1 INTRODUCTION
Underground mining can cause a number of environmental problems such as build-
ing collapse and road damage (Al-Rawahy 1995; Donnelly et al. 2001; Guéguen et
al. 2009; Yerro et al. 2014), as well as disruptions in groundwater aquifers, under-
ground gas, water, electricity, and sewage systems (Dong et al. 2013). With the larg-
est reported bituminous coal reserves in the United States, Illinois has been one of
the nation’s major coal producers for nearly 150 years, and coal has been mined in
73 of the state’s 102 counties (http://www.eia.doe.gov). Many of the modern and
historic coal mining areas are located under towns or important infrastructure such
as major roadways and railways. The Illinois State Geological Survey estimates that

61
62 Remote Sensing for Sustainability

approximately 201,000 acres of urban and built-up lands may be close to under-
ground mines with 333,000 housing units being exposed to possible mine subsidence.
Coal production–related ground deformation generally occurs in three stages: the
initial phase, which usually corresponds to less than 15% of the total subsidence
for an individual coal seam; the fast main phase, corresponding to approximately
75% of the maximum subsidence; and the residual phase, which presents a decreas-
ing subsidence rate (Guéguen et al. 2009) that can occur over decades. Generally,
subsidence occurs during active mining, and its occurrence can be directly related
to the location of active mining. The amount of residual subsidence after active min-
ing is often small (less than 10% relative to the total subsidence) and can easily
be overlooked after the mine is exhausted. However, residual subsidence associated
with abandoned mines can continue for years—or decades. Monitoring the long-
term residual land subsidence is of great interest not only to individual homeowners,
insurance companies, and regional policymakers charged with assessing risk and
the development of hazard mitigation plans, but also to city planners and developers.
Conventional ground-based deformation monitoring techniques (e.g., Global
Positioning Systems [GPSs] and leveling) are limited to discrete and sparse sites
and are not able to provide detailed and comprehensive ground deformation over
large areas. The Differential Interferometric Synthetic Aperture Radar (DInSAR)
technique developed in the late 1980s has demonstrated its potential for high-density
spatial mapping of ground displacement and has become an important tool for moni-
toring temporal and spatial ground movements (Chaussard et al. 2013; Gabriel et al.
1989; Massonnet and Feigl 1998). DInSAR can achieve an accuracy comparable to
field measurements, but at a much higher spatial density, larger coverage, and with
low cost (Raucoules et al. 2009; Zhang et al. 2011). However, the standard DInSAR
technique is subject to uncertainties caused by temporal and spatial decorrelations as
well as atmosphere artifacts (Sousa et al. 2011; Zebker and Villasenor 1992; Zebker
et al. 1997). To overcome these limitations, advanced time-series InSAR techniques
(timeSAR) have been proposed, utilizing multiple interferograms derived from a
large set of SAR images. There are two schools of timeSAR techniques: Persistent
Scatterer Interferometry (PSI) (Ferretti et al. 2000, 2001; Hooper et al. 2004;
Kampes 2006) and Small Baseline Subset (SBAS) (Berardino et al. 2002; Lanari
et al. 2004, 2007; Usai 2003). Millimetric precision of velocity estimation is possible
with timeSAR approach.
Each of these methods has its advantages and limitations. For example, PSI is
based on analysis of persistent point targets and works better in urban areas than
in suburban or vegetated areas (Bell et al. 2008; Colesanti et al. 2003; Raucoules
et al. 2013), providing high-resolution measurement of surface motions (Prati et al.
2010). Limitations associated with PSI processing include the following: (1) a large
number of SAR images (at least 20 images) is required to obtain reliable results
(Guéguen et al. 2009); (2) it assumes the temporal deformation to be linear; and
(3) it is known to underestimate high deformation rates because of temporal unwrap-
ping issues (Raucoules et al. 2009). In contrast, SBAS exploits distributed scatterers
using small baseline interferogram subsets and performs better for both urban and
nonurban vegetated areas, and also in areas with high deformation rates (Chaussard
et al. 2014; Gourmelen et al. 2010; Hooper 2006). SBAS is able to estimate nonlinear
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 63

deformation rates but may not be optimal for localized displacements that may
affect, for example, small areas or single buildings.
In multitemporal InSAR processing, both PSI and SBAS approaches are opti-
mized to obtain ground displacement rates with a nominal accuracy of milli-
meters per year (Yan et al. 2012). These methods have been successfully applied to
detect ground movements caused by not only natural phenomena such as tectonic
or volcanic activities (Chaussard et al. 2014; Yan et al. 2012), salt movement (Abir
et al. 2015), glacial rebound and landslides (Lauknes et al. 2010; Liu et al. 2013),
ground dissolution (Gutiérrez et al. 2011; Paine et al. 2012), and freeze–thaw of per-
mafrost (Chen et al. 2013; Short et al. 2014), but also anthropogenic activities includ-
ing underground mining (coal, oil and gas, etc.) (Abdikan et al. 2014; Grzovic and
Ghulam 2015; Guéguen et al. 2009; Herrera et al. 2007; Raucoules et al. 2003; Yerro
et al. 2014), groundwater extraction (Chaussard et al. 2013; Zhu et al. 2013), tunnel
construction (Knospe and Busch 2009; Strozzi et al. 2011), and load increase of con-
structions (Abidin et al. 2011; Mazzotti et al. 2009).
The objectives of this chapter are to map risk areas of structural damage and
infrastructure failure attributed to mining extraction using both PSI and SBAS tech-
niques and to demonstrate the potential of InSAR for sustainable urban planning that
avoids developing over subsiding areas.

4.2 STUDY AREA AND DATA

4.2.1 Study AreA
The St. Louis Metropolitan area is located just south of the confluence of the
Missouri and Mississippi rivers, straddling the Missouri–Illinois border (Figure
4.1). The study area encompasses all of St. Louis City and St. Louis County as
well as portions of Jefferson and St. Charles Counties in Missouri, and Madison,
Monroe, and St. Clair Counties in Illinois. The region is susceptible to many geo-
logic hazards (e.g., surface subsidence, sinkholes) because of the presence of numer-
ous sinkholes and abandoned underground mines (Louchios et al. 2013; MoDNR
2014a,b). The area presents significant problems because of the density of urban
infrastructure and population. It is situated between the relatively higher topogra-
phy of the Ozark Plateau in the southwest and the lower topography of the Illinois
Basin. The elevation ranges from 68 to 530 m (Jordan et al. 2014b). It has diverse
land cover classes in the area such as forest, agriculture, pasture, and urban (Jordan
et al. 2012). The climate of this region is continental type with distinct seasons
characterized by wide ranges in temperature and irregular annual and seasonal
precipitation (Jordan et al. 2014b).
The surface cover in the St. Louis metropolitan area varies from glacial deposits
to stream alluvium (Schultz 1993). Much of the surface of the study area is covered
by loess consisting of sandy silts, silts, and clays that are 3–25 m thick (Schultz
1993). The floodplains of Mississippi and Missouri and their tributaries are covered
with alluvium consisting of fluvial clays, silts, sands, and gravels and are up to 25 m
thick (Schultz 1993). Underneath this surface cover are gently dipping (<1° northeast)
Paleozoic limestones, dolostones, shales, and sandstones of marine origin (Harrison
 91°0´W 90°30´W 90°0´W 89°30´W
64

ERS1/2
Illinois 2 (asc) Track 1
Track 16 26 (d sc)
PALSAR
Missouri

39°0´N

39°0´N
MADISON
ST CHARLES

ST LOUIS ST LOUIS CITY

ST. CLAIR
38°30´N

38°30´N
FRANKLIN

JEFFERSON MONROE
Study area
GPS base stations
Survey benchmarks
PALSAR footprint
38°0´N

38°0´N
ERS 1/2 footprint
0 20 40
City/county boundary km

91°0´W 90°30´W 90°0´W 89°30´W

FIGURE 4.1 Location of the study area. The background is a shaded relief map from ArcGIS online map services. Solid and dashed rectangles refer
to PALSAR and ERS1/2 image footprint. Black circles are the locations of GPS base stations that are referenced during InSAR processing. Long-term
Remote Sensing for Sustainability

leveling measurements are available, and survey benchmarks are located at the solid triangle.
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 65

1997). The oldest bedrock units are in the Ozark Plateau exposed in the southwest
third of the study area and consist of Cambrian, Ordovician, and Devonian dolos-
tones, sandstones, and shales. To the northeast are younger Pennsylvania rocks in
the Illinois Basin that consist of limestones, shales, and sandstones. In between the
Ozark Plateau and the Illinois Basin are Mississippian limestones and sandstones
topped by Pennsylvanian shales. The St. Louis Fault Zone is the only structure that
significantly influences surface features. The fault zone trends north-northeast and
appears to control the course of the Mississippi River, causing it to flow across the
northeastern flank of the Ozark Plateau. Schultz (1993) presented evidence for right-
lateral strike-slip motion on the fault, but it is inconclusive as most of the fault zone
lies beneath the Mississippi River.

4.2.2 dAtA
Thirty-seven C-band SAR images were obtained from the European Space Agency’s
ERS-1 and ERS-2 SAR satellites. The ERS-1/2 data were acquired on descending
orbits for the period between 1992 and 2000. We also selected 16 of the Phased
Array type L-band Synthetic Aperture Radar (PALSAR) images acquired from
2007 to 2011. The PALSAR sensor onboard the Advanced Land Observing Satellite
(ALOS) from the Japan Aerospace Exploration Agency (JAXA) collects data with a
46-day repeat orbit cycle. The PALSAR images were acquired from ascending orbit
direction in Fine Beam Single Polarization (FBS) and Fine Beam Dual Polarization
(FBD) observation mode with the off-nadir angle of 34.3°. Note that both satellite
radar systems only detect displacement in the line-of-sight (LOS) direction.
For validation, we used leveling data collected from 1988 to 2008. The level-
ing data were measured monthly as relative movement to permanent benchmarks
deployed in the study area. National Elevation Dataset (NED) with a spatial resolu-
tion of 10 m was used to remove topographic effect, flattening, coregistration, and
geocoding of InSAR data.

4.3 METHODS
An integrated PSI and SBAS approach can maximize the spatial density of coherent
pixels, allowing the identification of persistent scatterers that dominate the scattering
from the resolution cell and slowly varying filtered phase pixels representing distrib-
uted targets whose phases decorrelate little over short time intervals (Hooper 2008).
We utilized both SBAS and PSI processing of ERS-1/2 data. However, only SBAS
was performed on PALSAR images owing to the small number of samples. Another
challenge for InSAR processing was the fact that a large portion of the area was cov-
ered by vegetation and water, which significantly decreases the coherence of InSAR
pairs. To overcome this limitation, we processed the ERS-1/2 data by two smaller
sections divided by the Mississippi River, St. Louis part on the west of the river and
the Belleville part on the east of the river. Compared to C-band SAR images, better
coherence and coregistration can be achieved with L-band data especially over veg-
etated terrains because of the better alignment of the reference and repeat images in
the range direction (Rosen et al. 1996). We were able to process entire PALSAR data
66 Remote Sensing for Sustainability

without splitting the image coverage into smaller portions. While the details of PSI
and SBAS approaches can be found in other literature, for completeness, we provide
a brief description of the methods with pertinent parameter settings.

4.3.1 SBAS ProceSSing

Interferograms were generated using SARscape modules of ENVI software from
Exelis VIS Information Solutions. Normal baseline was set to 50% of the critical
baseline and 900 days threshold was used for the temporal baseline. After coreg-
istration and interferogram generation, NED 10 m Digital Elevation Model (DEM)
resampled to 25 m was used to remove the flat earth phase component. Goldstein
filtering (Goldstein and Werner 1998) with optimized parameters (Ghulam et al.
2010) were used to remove interferogram noise. A complex multilook operation was
performed to produce ground resolution of approximately 25 m. Baseline-dependent
phase residuals attributed to DEM inaccuracies (Berardino et al. 2002) were removed
using a predefined linear displacement model that jointly estimates the DEM error
and the low-pass displacement parameters. The efficacy of using a priori known dis-
placement models in removing the DEM error has also been reported by a number of
studies (Lauknes et al. 2010; Berardino et al. 2002).
Both minimum cost flow (MCF) network (Costantini 1998) and Delaunay 3D
(D3D) methods (Hooper and Zebker 2007) was used to unwrap the phase values
only known within [−π, π]. D3D unwrapping is more reliable than the standard two-
dimensional approaches because it exploits the temporal information and unwraps
the phase difference between neighboring pixels in the time domain to reduce atmo-
spheric and orbital effects. However, it requires higher redundancy of SAR observa-
tions. For a given SBAS epoch, the MCF or D3D method was used depending on the
redundancy of DInSAR pairs determined by a plot of time versus relative position
with respect to the super master image. A spatial coherence thresholding of 0.35
was used to exclude decorrelated areas from the phase unwrapping so that a reliable
phase unwrapping was feasible only with highly coherent pixels. By screening all of
the wrapped and unwrapped interferograms, problems related to inaccurate orbits
and uncoherent pairs were identified and further corrected or excluded.
Temporal variations in the refractive index attributed to water vapor can lead
to a significant phase delay that is associated with elevation (Cavalie et al. 2007;
Delacourt et al. 1998). This error component along with the phase error caused by
the atmospheric signal is correlated in space and needs to be removed. To that end,
we employed a two-step inversion process. First, we estimated a phase delay eleva-
tion profile for each interferogram and displacement velocity using a linear inver-
sion model. A wavelet number of levels, which refers to the power of base 2, was
used to determine the residual topography. This value was set as a function of of
the reference DEM, which was used for the interferogram flattening. We processed
SAR data with 25 m resolution with NED 10 m DEM resampled to 25 m; therefore,
a wavelet number of level 1 was used. This means that information coarser than
50 m was removed. The higher the levels, the lower in terms of spatial frequency
will be removed. Then, the first estimated residual topography and velocity phase
components were subtracted from input wrapped interferograms. The difference,
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 67

the remaining error component attributed to the phase delays owing to atmospheric
signals, which is spatially correlated, was unwrapped and added to the first estimated
displacement velocity phase. Second, the sum of the first estimated velocity and the
remaining error phase was unwrapped again and reflattened, and an SBAS second
inversion was conducted to jointly estimate the error source and refined velocity.
Atmospheric phase screening was conducted to estimate atmospheric effects, which
were removed using a low-pass spatial filter with a 1.2 km × 1.2 km window on each
single acquisition followed by a high-pass filter on the time series images.
Sixty-three interferogram pairs were generated from 16 PALSAR images over the
entire study area. The number of pairs generated from 37 ERS-1/2 images for the
west and east of the Mississippi River was 86 and 94, respectively.

4.3.2 PSi ProceSSing

PSI is intended for the analysis of point targets characterized by high coherence
behavior. A total of 37 ERS-1/2 images were processed with the PSI algorithm
(Ferretti et al. 2001). The topographic reference was determined from the same
DEM as for the SBAS, which was used to remove the topographic and ellipsoidal
height components of the phase. Persistent scatterers were selected based on the
coherence value over 0.65. Then, atmospheric phases were removed from each dif-
ferential interferogram by filtering the spatial and temporal domains according to
their spatial and temporal correlations using the same method utilized for SBAS. To
unwrap the complex phase, the MCF method was integrated with a region growing
procedure to improve the performances in areas with low signal-to-noise ratio due
to predominant vegetation cover (Costantini 1998). Then, mean annual LOS velocity
and displacement time series were calculated.

4.4 RESULTS AND DISCUSSION

4.4.1 SPAtiAl And temPorAl PAttern of lAnd SuBSidence
from 1992 to 2000

Figure 4.2a through d show estimated mean LOS velocity in millimeters per year
from the 37 ERS-1/2 images using both SBAS (Figure 4.2a and b) and PSI (Figure
4.2c and d) methods. The mean LOS velocity image was draped on the mean inten-
sity image. Positive LOS velocities (blue colors) represent movement toward the sat-
ellite, and negative LOS velocities (red colors) represent movement away from the
satellite. Spatial patterns of ground deformation observed by both SBAS and PSI
methods over St. Louis, Missouri (Figure 4.2a and c) and Belleville, Illinois (Figure
4.2b and d) are almost identical. In other words, deformation signals detected by
InSAR were confirmed by both methods, demonstrating distinct subsidence or uplift
patterns on several locations in the study site.
From 1992 to 2000, relative uplift, as demonstrated by the blue cool colors in
Figure 4.2a and c, was observed over parks and cemeteries in south-east central
areas of St. Louis city. The uplift might be attributed to an increase in precipita-
tion in the 1990s (Jordan et al. 2014a,b) and tree growth. In Belleville, Illinois, the
68 Remote Sensing for Sustainability

(a) (c)

A1

A2 A3

A4
A5
A6

(b) (d)

B4 B6
B1

B5
B7
B2
B3 B8

0 2.5 5 10 km

LOS Velocity (mm/year)

< –9.0 –1.4 to –0.5 0.6–8.0
–8.9 to –1.5 –0.4 to 0.5 >8.0

FIGURE 4.2 Mean LOS velocity (in millimeters per year) estimated by SBAS (a and b)
and PSI (c and d) methods using 37 ERS-1/2 images. Background image is mean intensity.
Positive LOS velocities (blue colors) represent movement toward the satellite; negative LOS
velocities (red colors) represent movement away from the satellite. Dashed rectangles are
typical areas with distinct subsidence or uplift patterns that will be analyzed further in suc-
ceeding sections.
 InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 69

deformation patterns from 1992 to 2000 were mostly subsidence, which correspond
to the locations of underground mining and abandoned mine workings (Figure 4.2b
and d).
Both the SBAS and PSI data sets show average annual rates of movement rang-
ing from −9 to +8 mm/year. It is worth noting that the SBAS estimated deformation
represents a pixel dimension of approximately 20 × 20 m, and the PSI estimates were
retrieved at a full pixel resolution of 4 × 20 m in the azimuth and range directions,
respectively. In order to compare time series, all pixels within a selected region of
interest (ROI) were averaged. Figure 4.3 shows the estimated total LOS displacement

30 30
A1 A2
20 20
Deformation (mm)

SBAS Deformation (mm) SBAS

10 PSI 10 PSI

0 0
–10 –10
–20 –20
–30 –30
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year
30 30
A3 A4
20 20
Deformation (mm)

Deformation (mm)

SBAS
10 10 PSI

0 0
–10 SBAS –10
PSI
–20 –20
–30 –30
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year
30 30
A5 A6
20 20
Deformation (mm)

Deformation (mm)

10 10
0 0
–10 SBAS –10 SBAS
PSI PSI
–20 –20
–30 –30
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year

FIGURE 4.3 Time series deformation history of selected regions of interest shown in
Figure 4.2. Labels A and B correspond to the St. Louis, Missouri, site and the Belleville,
Illinois, site. (Continued)
 70 Remote Sensing for Sustainability

20 20
B1 B2
10 10
SBAS SBAS
Deformation (mm)

Deformation (mm)
0 PSI 0 PSI
–10 –10
–20 –20
–30 –30
–40 –40
–50 –50
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year
20 20
B3 B4
10 10 SBAS
SBAS
Deformation (mm)

Deformation (mm)
0 PSI 0 PSI
–10 –10
–20 –20
–30 –30
–40 –40
–50 –50
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year
20 20
B5 B6
10 SBAS
10
SBAS
Deformation (mm)

Deformation (mm)

0 PSI 0 PSI
–10 –10
–20 –20
–30 –30
–40 –40
–50 –50
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year
20 20
B7
10 10
Deformation (mm)

Deformation (mm)

0 0
B8
–10 –10
–20 –20
–30 SBAS –30 SBAS
PSI PSI
–40 –40
–50 –50
1991 1993 1995 1997 1999 2001 1991 1993 1995 1997 1999 2001
Year Year

FIGURE 4.3 (CONTINUED) Time series deformation history of selected regions of inter-
est shown in Figure 4.2. Labels A and B correspond to the St. Louis, Missouri, site and the
Belleville, Illinois, site.
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 71

time series averaged for selected ROIs using both SBAS and PSI methods. Labels
A1–A6 and B1–B8 correspond to the St. Louis, Missouri, and Belleville, Illinois,
sites. Main observations include the following: (1) ground deformation history from
SBAS and PSI was highly consistent, demonstrating a match between highs and
lows of the deformation trend line; (2) the amplitude of total deformation varied
from +30 to −30 mm for most of the ROIs except for location B4, which experienced
significant (up to −50 mm) subsidence from 1992 to 2000; (3) overall, SBAS-derived
deformation trends were smoother than PSI results, which was possibly due to the
fact that a better estimation of nonlinear deformation (note: PSI assumes a linear
deformation trend) as well as residual topography was achieved with SBAS using a
wavelet number approach; and (4) SBAS results seem to have a greater spatial cover-
age compared to PSI, especially over vegetated areas.

4.4.2 SPAtiAl And temPorAl PAttern of lAnd SuBSidence

from 2007 to 2011

Figure 4.4 shows the effects of subsidence in the St. Louis metropolitan region. The
mean velocity deformation map results (Figure 4.4a) for the period from January 2,
2007 to January 13, 2011 are from the ALOS PALSAR data set using the SBAS
method. Note that not all subsidence takes place over underground mines. In Figure
4.4b, an area affected by karst processes (purple outline) is shown. The karst plain
covers most of Monroe county and a small portion of St. Claire County, but much of
this area is agriculture or undeveloped.
Mining on the Missouri side of the Mississippi River was active up into the
1940s, but today mining activity is confined to the open pit mining of limestone
(Seeger 2014). On the Illinois side, however, mining continues and consists primar-
ily of underground coal mining. Active underground mines are required by law to
have a surface subsidence mitigation plan as well as the requirement to monitor and
repair mine subsidence damages that occur (Bauer 2008). Mine subsidence from
these active mines causes little damage to surface infrastructure after initial collapse
of the surface because the method used for mining leaves little to no open space
underground (Bauer 2008). There are, however, many abandoned underground coal
mines in the region (Chenoweth et al. 2004a,b, 2005). For these mines, the potential
for collapse still exists after many years because the mining techniques used leave
open spaces underground after mining has stopped (Bauer 2006, 2008).
Figure 4.4c shows the time series of ground subsidence, indicated by the red line,
for the large southern subsiding area in Figure 4.4d, which is an area in Illinois
affected by mine subsidence during the period. Groundwater depth (blue line) and the
total yearly precipitation (light gray shading) for the area are also plotted. The loca-
tion of the well where groundwater measurements were taken is indicated by the star
on Figure 4.4a. We can see that a drastic subsidence occurred when the groundwater
level drops during the drought year (2007), possibly caused by the excessive ground-
water withdrawals. It suggests that the deformation pattern observed in the area is
likely to be related to the drop in the groundwater level. Then, subsidence slows
or reverses to uplift when groundwater levels rise. Note, however, that the surface
72 Remote Sensing for Sustainability

90º20΄W 90º0΄W 89º40΄W

(a) (c)
10 14
39º0΄N

groundwater depth (dm)

Displacement (mm)
N 12
–10

Precipitation (cm)
10
–30 8
ES
Precipitation

ARL
38º50΄N

CH MADISON –50
GW depth
6
ST Displacement
4
–70
2
–90 0
38º40΄N

2006 2007 2008 2009 2010

STLOUIS Year
STLOUIS CITY
(d)
38º30΄N

ST. CLAIR
JEFF
38º20΄N

Groundwater well

Counties
MONROE
Underground coal mines
Sinkhole areas
Mean LOS velocity (mm/yr) 25
38º10΄N

50.0 –50.0 km

(b)
(e) 2003

(g) 2010 (f ) 2005

FIGURE 4.4 (a) Mean LOS velocity (in millimeters per year) using the small baseline
subset (SBAS) method for the ALOS PALSAR data set, a period from January 2, 2007 to
January 13, 2011. Bright colors indicate movement away from the satellite and dark colors
indicate movement toward the satellite. Light gray outlines indicate underground coal mine
locations. Purple outlines indicate sinkhole-prone areas. The star indicates the location of the
well where groundwater depths were taken. (b) A Google Earth image from 2014 showing a
typical sinkhole-prone area. (c) Time series plot of ground deformation for the southern subsid-
ing area shown in (d). The red line indicates ground deformation, the blue line denotes ground-
water depth from the well indicated in (a), and the light gray shaded area indicates yearly total
precipitation. (d) Close-up of indicated area showing mean LOS velocity of an area undergoing
development during the period ALOS PALSAR images were acquired. (e through g) Google
Earth images showing recent urban development of the area shown in the figure.
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 73

subsidence, unlike groundwater, never returns to the level it started at. One may
expect that subsidence closely follows the groundwater recharge cycle, for example,
ground uplifts or subsidence corresponding to increase or decrease of groundwater
tables. This is particularly true for areas where there are no underground mine pil-
lar failures. The fact that ground subsidence did not reverse, corresponding to water
level changes after a drought year, hints a high likelihood of pillar failure.
Several significant subsidence zones can be found in Figure 4.4d (see the red
spots). Aided by aerial photos and Google images, we determined that these red
spots have experienced recent urban developments for both residential and school
building (Figure 4.4e, f, and g). It is apparent that development on the areas located
on the upper-left corner of Figure 4.4d continued despite reported land subsidence
incidents in a nearby school (red spot on the lower-right portion of Figure 4.4d).
InSAR data measured subsidence of up to −25.6 mm/year at the two subsiding areas.

4.4.3 VAlidAtion of inSAr deformAtion uSing leVeling dAtA

In order to validate InSAR-derived measurements, we selected two bench mark (BM)
points (BM3 and BM29) where two decades of leveling measurements were avail-
able. Figure 4.5 shows the deformation history of BM3 and BM29 obtained from
SBAS and leveling measurements, which confirms the accuracy of InSAR measure-
ments. It is evident from Figure 4.5 that BM3 experienced a linear deformation from
1990 to 2011 while BM29 had an abrupt change in deformation trend in 1997. The
overall trend observed in BM29 indicates that there was a possible land collapse
or failure in underground mine pillars occurring in 1997, which was followed by a
no deformation period.
The room-and-pillar mining technique, which leaves vacant space underground
after abandonment of the mine, is a common type of mining technique used in the
majority of the abandoned mines in our study area (Grzovic and Ghulam 2015).
In this type of mining extraction, a series of pillars are left to support the roof of

30 30
BM3 InSAR BM29 InAR
20 BM3 20 BM29
Deformation (mm)

Deformation (mm)

Linear fit of InSAR Linear fit of InAR

10 Linear fit of BM3 10 Linear fit of BM29

0 0
–10 –10
–20 –20
–30 –30
1991 1995 1999 2003 2007 2011 1991 1995 1999 2003 2007 2011
(a) Year (b) Year

FIGURE 4.5 Comparison of InSAR-derived deformation against field measured data at

BMs located in Belleville, Illinois. (a) Deformation trend at BM3. (b) Deformation trend at
BM29. It indicates an abrupt change in deformation trend at BM29 in 1998. This type of land
collapse might be attributed to an increase of crustal overburden due to a new development
over abandoned mining areas or changes in soil properties from lawn watering or rainfall.
74 Remote Sensing for Sustainability

the mine while rooms are formed by extracting coal (Bauer 2008). Arguably, the
deformation trend observed from BM29 may indicate a possible pillar failure, which
might have been caused by residual land subsidence accelerated by extreme changes
in precipitation, groundwater level, or a new development followed by excessive lawn
watering or water use.
As shown in Figure 4.5, the SBAS time series exhibit a larger spread around
the fitted linear curve than the BM survey measurements. This is due to the fact
that ground-truth data from leveling represent measurements of a single point while
InSAR-derived deformation represents a 20-m pixel average subsidence. One may
expect discrepancies between absolute deformation values of leveling and SBAS, but
consistency of overall deformation trends between the two measurements should be
regarded as compelling.

4.4.4 imPlicAtionS for SuStAinABle urBAn PlAnning

Commercial coal mining in Illinois started around 1810 in Jackson County, and
mining for coal, fluorspar, lead, limestone, and zinc has occurred in 72 counties
(Bauer 2008). The Illinois Department of Natural Resources estimates that there
are approximately 5000 closed or abandoned coal mines in 53 counties across the
state, about half of which have not been mapped or documented. Approximately
201,000 acres of urban and built-up lands may be close to underground mines with
333,000 housing units being exposed to possible mine subsidence. In recent decades,
subsidence damage to structures has become increasingly common in Illinois as
new developments expanded over abandoned, long-forgotten mines. Developing
low-cost, effective tools to monitor potential risk from long-term residual land sub-
sidence and predicting impending collapse before it actually causes economic loss
is of great importance not only to individual property owners but also to insurance
companies and regional policymakers charged with assessing risk related to aban-
doned coal mines.
Our preliminary studies indicated that sudden collapses of buildings, roads, and
other infrastructure have occurred in the region. Although the thickness of the min-
ing roof (depth to mine working) is far greater than the increasing overburden from
urban developments, ground subsidence can be accelerated after a new development
built on abandoned mining areas. Arguably, this might be explained by a combined
effect of increasing overburden, lawn watering, abrupt changes in precipitation pat-
terns, and groundwater table.
For example, the largest subsidence was observed at the B4 location, reaching
approximately −50 mm/year over the 1992–2000 observation period. The promi-
nent red spot on the upper-right corner of the B4 rectangle (Figure 4.2b) was right
over Wolf Branch Middle School, Swansea, Illinois. The school was built around
2003 and quickly started to develop cracks and severe building damage by 2006.
Significant mining subsidence at Bridle Lane, Swansea, Illinois, and surrounding
areas (including the Wolf Branch Middle School) was also detected from InSAR
maps of 2007–2011 (Figure 4.4d through g). These findings serve as compelling
evidences that InSAR can reveal risk areas to avoid for new construction and guiding
suggestions for prioritizing resources for targeted surveying, which, when combined
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 75

with soil properties, groundwater withdrawal, and modern fault activities and dis-
tribution of abandoned mining areas, provides valuable information regarding
impending land collapses. Doing so will (1) benefit policymakers and city planners
in making informed decisions about future developments and (2) save potential loss
of investments in such hazardous areas.

4.5 CONCLUSION
Coal mining can cause a number of environmental and social problems, such as
building collapse, road damage, and disruptions in groundwater aquifers that are
the result of mining-induced persistent land subsidence. In recent decades, subsidence
damage to structures has become increasingly common as towns and cities expanded
over abandoned, long-forgotten mines. Therefore, developing cost-effective methods
to monitor long-term residual land subsidence is of great interest not only to individ-
ual homeowners but also to city planners, insurance companies, and regional policy-
makers charged with assessing risk related to abandoned coal mines. The objective
of this chapter was to develop new techniques for using space-based data to detect
irregular patterns of land subsidence and to design efficient strategies for identify-
ing and mapping areas with elevated risk of structural and infrastructural damage.
Thirty-seven C-band (ERS1/2) and 16 L-band (PALSAR) Synthetic Aperture Radar
(SAR) images acquired from 1992 to 2000 and from 2007 to 2011 were processed
with SBAS and PSI techniques to investigate land deformation. The correlation
between land deformation and mining activities, regional geological background,
groundwater, and precipitation were discussed. The results showed hot spots of
ongoing and potential land collapses in the region with an average subsidence of
−9 mm/year. Validated by conventional surveying, GPS measurements, and related
geological survey, this study presented a comparative study between SBAS and PSI
methods for displacement mapping, in an area made challenging by a humid climate,
extensive surface waters, and dense vegetation cover.
Our results confirmed that subsidence took place mostly above the mining areas
where deformation was well known and had experienced land collapses and sig-
nificant infrastructure damage. Observed ground deformation, precipitation, and the
variations of groundwater table show unquestionable similarities. Mechanisms that
cause uplift over parks and cemeteries, however, are not clear and may be of various
sources: unloading attributed to water level rise, pore pressure variations, swelling
of soil or tree growth, and so on.
In most of the validation and comparison, we used vertical subsidence rates con-
verted from the observed LOS rates. The limitation of this work is that only descend-
ing ERS-1/2 or ascending ALOS acquisitions were utilized for deformation mapping.
Therefore, InSAR-derived LOS deformation represents the movement away from or
toward the satellite. However, both GPS and leveling measurements conducted in
the study area showed that ground displacement was occurring mainly vertically.
We have projected the LOS displacement onto the vertical direction using satellite
look angle and LOS displacement vector assuming that the horizontal ground dis-
placement is negligible. However, this assumption may need comprehensive valida-
tion with additional information, for example, pairing ascending and descending
76 Remote Sensing for Sustainability

InSAR measurements or extensive GPS network. It should be noted that our work
mainly focused on detecting ground deformation and identifying potential collapse
hot spots. Detailed investigation on the mechanisms and structural, anthropogenic,
and geological causes of mining related land collapses will be our future research.

ACKNOWLEDGMENT
This work was supported by Saint Louis University’s President’s Research Fund.

REFERENCES
Abdikan, S., M. Arikan, F. B. Sanli, and Z. Cakir. 2014. Monitoring of coal mining subsidence
in peri-urban area of Zonguldak city (NW Turkey) with persistent scatterer interferom-
etry using ALOS-PALSAR. Environmental Earth Sciences 71 (9):4081–4089.
Abidin, H. Z., H. Andreas, I. Gumilar, Y. Fukuda, Y. E. Pohan, and T. Deguchi. 2011. Land
subsidence of Jakarta (Indonesia) and its relation with urban development. Natural
Hazards 59 (3):1753–1771.
Abir, I., S. Khan, A. Ghulam, S. Tariq, and M. Shah. 2015. Active tectonics of western Potwar
Plateau-Salt Range, Northern Pakistan from InSAR observations and seismic imaging.
Remote Sensing of Environment 168:265–275.
Al-Rawahy, S. 1995. Effects of mining subsidence observed by time-lapse seismic reflection
profiling, Durham University.
Bauer, R. A. 2006. Mine Subsidence in Illinois: Facts for Homeowners. Champaign, Illinois:
Illinois State Geological Survey.
Bauer, R. A. 2008. Planned Coal Mine Subsidence in Illinois: A Public Information Booklet:
Illinois State Geological Survey Circular, Champaign, Illinois.
Bell, J. W., F. Amelung, A. Ferretti, M. Bianchi, and F. Novali. 2008. Permanent scatterer
InSAR reveals seasonal and long-term aquifer-system response to groundwater pump-
ing and artificial recharge. Water Resources Research 44 (2).
Berardino, P., G. Fornaro, R. Lanari, and E. Sansosti. 2002. A new algorithm for surface
deformation monitoring based on small baseline differential SAR interferograms. IEEE
Transactions on Geoscience and Remote Sensing 40 (11):2375–2383.
Cavalie, O., M. P. Doin, C. Lasserre, and P. Briole. 2007. Ground motion measurement in the
Lake Mead area, Nevada, by differential synthetic aperture radar interferometry time
series analysis: Probing the lithosphere rheological structure. Journal of Geophysical
Research—Solid Earth 112 (B3).
Chaussard, E., F. Amelung, H. Abidin, and S.-H. Hong. 2013. Sinking cities in Indonesia:
ALOS PALSAR detects rapid subsidence due to groundwater and gas extraction.
Remote Sensing of Environment 128:150–161.
Chaussard, E., S. Wdowinski, E. Cabral-Cano, and F. Amelung. 2014. Land subsidence in
central Mexico detected by ALOS InSAR time-series. Remote Sensing of Environment
140:94–106.
Chen, F., H. Lin, W. Zhou, T. Hong, and G. Wang. 2013. Surface deformation detected by
ALOS PALSAR small baseline SAR interferometry over permafrost environment of
Beiluhe section, Tibet Plateau, China. Remote Sensing of Environment 138:10–18.
Chenoweth, C. A., M. E. Barrett, and S. D. Elrick. 2004a. Directory of Coal Mines in Illinois
7.5-minute Quadrangle Series: O’Fallon Quadrangle St. Clair County. Champaign,
Illinois: Illinois State Geological Survey.
Chenoweth, C. A., S. D. Elrick, and M. E. Barrett. 2004b. Directory of Coal Mines in Illinois
7.5-minute Quadrangle Series: French Village Quadrangle St. Clair County. Champaign,
Illinois: Illinois State Geological Survey.
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 77

Chenoweth, C. A., S. D. Elrick, and M. E. Barrett. 2005. Directory of Coal Mines in Illinois
7.5-minute Quadrangle Series: Collinsville Quadrangle Madison and St. Clair Counties.
Champaign, Illinois: Illinois State Geological Survey.
Colesanti, C., A. Ferretti, F. Novali, C. Prati, and F. Rocca. 2003. SAR monitoring of progres-
sive and seasonal ground deformation using the permanent scatterers technique. IEEE
Transactions on Geoscience and Remote Sensing 41 (7):1685–1701.
Costantini, M. 1998. A novel phase unwrapping method based on network programming.
IEEE Transactions on Geoscience and Remote Sensing 36 (3):813–821.
Delacourt, C., P. Briole, and J. Achache. 1998. Tropospheric corrections of SAR interfero-
grams with strong topography. Application to Etna. Geophysical Research Letters 25
(15):2849–2852.
Dong, S., S. Samsonov, H. Yin, S. Ye, and Y. Cao. 2013. Time-series analysis of subsidence
associated with rapid urbanization in Shanghai, China measured with SBAS InSAR
method. Environmental Earth Sciences:1–15.
Donnelly, L. J., H. De La Cruz, I. Asmar, O. Zapata, and J. D. Perez. 2001. The monitoring and
prediction of mining subsidence in the Amaga, Angelopolis, Venecia and Bolombolo
Regions, Antioquia, Colombia. Engineering Geology 59 (1–2):103–114.
Ferretti, A., C. Prati, and F. Rocca. 2000. Nonlinear subsidence rate estimation using perma-
nent scatterers in differential SAR interferometry. IEEE Transactions on Geoscience
and Remote Sensing 38 (5):2202–2212.
Ferretti, A., C. Prati, and F. Rocca. 2001. Permanent scatterers in SAR interferometry. IEEE
Transactions on Geoscience and Remote Sensing 39 (1):8–20.
Gabriel, A. K., R. M. Goldstein, and H. A. Zebker. 1989. Mapping small elevation changes
over large areas: Differential radar interferometry. Journal of Geophysical Research:
Solid Earth (1978–2012) 94 (B7):9183–9191.
Ghulam, A., R. Amer, and R. Ripperdan. 2010. A filtering approach to improve deformation
accuracy using large baseline, low coherence DInSAR phase images. In 2010 IEEE
International Geoscience and Remote Sensing Symposium. Honolulu, Hawaii: IEEE.
Goldstein, R. M., and C. L. Werner. 1998. Radar interferogram filtering for geophysical appli-
cations. Geophysical Research Letters 25 (21):4035–4038.
Gourmelen, N., F. Amelung, and R. Lanari. 2010. Interferometric synthetic aperture radar–
GPS integration: Interseismic strain accumulation across the Hunter Mountain fault in
the eastern California shear zone. Journal of Geophysical Research: Solid Earth (1978–
2012) 115 (B9).
Grzovic, M., and A. Ghulam. 2015. Evaluation of land subsidence from underground coal
mining using TimeSAR (SBAS and PSI) in Springfield, Illinois. Natural Hazards
79:1739–1751.
Guéguen, Y., B. Deffontaines, B. Fruneau et al. 2009. Monitoring residual mining subsidence
of Nord/Pas-de-Calais coal basin from differential and Persistent Scatterer Interferometry
(Northern France). Journal of Applied Geophysics 69 (1):24–34.
Gutiérrez, F., J. P. Galve, P. Lucha, C. Castañeda, J. Bonachea, and J. Guerrero. 2011.
Integrating geomorphological mapping, trenching, InSAR and GPR for the identifica-
tion and characterization of sinkholes: A review and application in the mantled evaporite
karst of the Ebro Valley (NE Spain). Geomorphology 134 (1):144–156.
Harrison, R. W. 1997. Bedrock geologic map of the St. Louis 30′ × 60′ quadrangle, Missouri
and Illinois. U.S. Department of the Interior, U.S. Geological Survey.
Herrera, G., R. Tomas, J. M. Lopez-Sanchez et al. 2007. Advanced DInSAR analysis on mining
areas: La Union case study (Murcia, SE Spain). Engineering Geology 90 (3–4):148–159.
Hooper, A. 2008. A multi-temporal InSAR method incorporating both persistent scatterer and
small baseline approaches. Geophysical Research Letters 35:L16302.
Hooper, A. 2006. Persistent scatter radar interferometry for crustal deformation studies and
modeling of volcanic deformation. Ph.D. thesis, Stanford University.
78 Remote Sensing for Sustainability

Hooper, A., and H. A. Zebker. 2007. Phase unwrapping in three dimensions with application to
InSAR time series. Journal of the Optical Society of America A—Optics Image Science
and Vision 24 (9):2737–2747.
Hooper, A., H. Zebker, P. Segall, and B. Kampes. 2004. A new method for measuring defor-
mation on volcanoes and other natural terrains using InSAR persistent scatterers.
Geophysical Research Letters 31 (23).
Jordan, Y. C., A. Ghulam, and M. L. Chu. 2014a. Assessing the impacts of future urban devel-
opment patterns and climate changes on total suspended sediment loading in surface
waters using geoinformatics. Journal of Environmental Informatics 24 (2):65–79.
Jordan, Y. C., A. Ghulam, and S. Hartling. 2014b. Traits of surface water pollution under
climate and land use changes: A remote sensing and hydrological modeling approach.
Earth-Science Reviews 128:181–195.
Jordan, Y. C., A. Ghulam, and R. B. Herrmann. 2012. Floodplain ecosystem response to cli-
mate variability and land-cover and land-use change in Lower Missouri River basin.
Landscape Ecology 27 (6):843–857.
Kampes, B. M. 2006. Radar Interferometry: Persistent Scatterer Technique. Vol. 12: Springer.
Knospe, S., and W. Busch. 2009. Monitoring a tunneling in an urbanized area with Terrasar-X
interferometry—Surface deformation measurements and atmospheric error treatment.
Paper read at Geoscience and Remote Sensing Symposium, 2009 IEEE International,
IGARSS 2009.
Lanari, R., F. Casu, M. Manzo et al. 2007. An overview of the small baseline subset algorithm:
A DInSAR technique for surface deformation analysis. Pure and Applied Geophysics
164 (4):637–661.
Lanari, R., O. Mora, M. Manunta, J. J. Mallorquí, P. Berardino, and E. Sansosti. 2004. A
small-baseline approach for investigating deformations on full-resolution differen-
tial SAR interferograms. IEEE Transactions on Geoscience and Remote Sensing 42
(7):1377–1386.
Lauknes, T. R., A. Piyush Shanker, J. F. Dehls, H. A. Zebker, I. H. C. Henderson, and Y. Larsen.
2010. Detailed rockslide mapping in northern Norway with small baseline and persistent
scatterer interferometric SAR time series methods. Remote Sensing of Environment 114
(9):2097–2109.
Liu, P., Z. Li, T. Hoey et al. 2013. Using advanced InSAR time series techniques to moni-
tor landslide movements in Badong of the Three Gorges region, China. International
Journal of Applied Earth Observation and Geoinformation 21:253–264.
Louchios, A. G., S. D. Elrick, C. P. Korose, and P. Wise. 2013. Coal Industry in Illinois.
Champaign, Illinois: Illinois State Geological Survey.
Massonnet, D., and K. L. Feigl. 1998. Radar interferometry and its application to changes in
the Earth’s surface. Reviews of Geophysics 36 (4):441–500.
Mazzotti, S., A. Lambert, M. Van der Kooij, and A. Mainville. 2009. Impact of anthropogenic
subsidence on relative sea-level rise in the Fraser River delta. Geology 37 (9):771–774.
MoDNR. 2014a. MO 2014 Inventory of Mines, Occurences, and Prospects (IMOP). Missouri
Department of Natural Resources (MoDNR), Missouri Geological Survey (MGS),
Geological Survey Program (GSP).
MoDNR. 2014b. MO 2014 Sinkholes. Missouri Department of Natural Resources (MoDNR),
Missouri Geological Survey (MGS), Geological Survey Program (GSP), Environmental
Geology Section (EGS).
Paine, J. G., S. M. Buckley, E. W. Collins, and C. R. Wilson. 2012. Assessing collapse risk in
evaporite sinkhole-prone areas using microgravimetry and radar interferometry. Journal
of Environmental and Engineering Geophysics 17 (2):75–87.
Prati, C., A. Ferretti, and D. Perissin. 2010. Recent advances on surface ground deforma-
tion measurement by means of repeated space-borne SAR observations. Journal of
Geodynamics 49 (3–4):161–170.
InSAR Monitoring of Land Subsidence for Sustainable Urban Planning 79

Raucoules, D., B. Bourgine, M. De Michele et al. 2009. Validation and intercomparison of per-
sistent scatterers interferometry: PSIC4 project results. Journal of Applied Geophysics
68 (3):335–347.
Raucoules, D., C. Cartannaz, F. Mathieu, and D. Midot. 2013. Combined use of space-borne
SAR interferometric techniques and ground-based measurements on a 0.3 km2 subsidence
phenomenon. Remote Sensing of Environment 139:331–339.
Raucoules, D., C. Maisons, C. Camec, S. Le Mouelic, C. King, and S. Hosford. 2003.
Monitoring of slow ground deformation by ERS radar interferometry on the Vauvert
salt mine (France)—Comparison with ground-based measurement. Remote Sensing of
Environment 88 (4):468–478.
Rosen, P. A., S. Hensley, H. A. Zebker, F. H. Webb, and E. J. Fielding. 1996. Surface deforma-
tion and coherence measurements of Kilauea volcano, Hawaii, from SIR-C radar inter-
ferometry. Journal of Geophysical Research—Planets 101 (E10):23109–23125.
Schultz, A. P. 1993. Map showing surficial geology of the St. Louis 30 × 60 minute quadrangle.
Seeger, C. 2014. MO 2014 Inventory of Mines, Occurences, and Prospects (SHP). Rolla,
Missouri: Missouri Geological Survey.
Short, N., A.-M. LeBlanc, W. Sladen, G. Oldenborger, V. Mathon-Dufour, and B. Brisco. 2014.
RADARSAT-2 D-InSAR for ground displacement in permafrost terrain, validation from
Iqaluit Airport, Baffin Island, Canada. Remote Sensing of Environment 141:40–51.
Sousa, J. J., A. J. Hooper, R. F. Hanssen, L. C. Bastos, and A. M. Ruiz. 2011. Persistent
Scatterer InSAR: A comparison of methodologies based on a model of temporal defor-
mation vs. spatial correlation selection criteria. Remote Sensing of Environment 115
(10):2652–2663.
Strozzi, T., R. Delaloye, D. Poffet, J. Hansmann, and S. Loew. 2011. Surface subsidence
and uplift above a headrace tunnel in metamorphic basement rocks of the Swiss Alps
as detected by satellite SAR interferometry. Remote Sensing of Environment 115
(6):1353–1360.
Usai, S. 2003. A least squares database approach for SAR interferometric data. IEEE
Transactions on Geoscience and Remote Sensing 41 (4):753–760.
Yan, Y., M.-P. Doin, P. Lopez-Quiroz et al. 2012. Mexico City subsidence measured by InSAR
time series: Joint analysis using PS and SBAS approaches. IEEE Journal of Selected
Topics in Applied Earth Observations and Remote Sensing 5 (4):1312–1326.
Yerro, A., J. Corominas, D. Monells, and J. J. Mallorqui. 2014. Analysis of the evolution
of ground movements in a low densely urban area by means of DInSAR technique.
Engineering Geology 170:52–65.
Zebker, H. A., P. A. Rosen, and S. Hensley. 1997. Atmospheric effects in interferometric syn-
thetic aperture radar surface deformation and topographic maps. Journal of Geophysical
Research—Solid Earth 102 (B4):7547–7563.
Zebker, H. A., and J. Villasenor. 1992. Decorrelation in interferometric radar echoes. IEEE
Transactions on Geoscience and Remote Sensing 30 (5):950–959.
Zhang, Y. H., J. X. Zhang, H. A. Wu, Z. Lu, and G. T. Sun. 2011. Monitoring of urban sub-
sidence with SAR interferometric point target analysis: A case study in Suzhou,
China. International Journal of Applied Earth Observation and Geoinformation 13
(5):812–818.
Zhu, W., Q. Zhang, X. L. Ding, C. Y. Zhao, C. S. Yang, and W. Qu. 2013. Recent ground defor-
mation of Taiyuan basin (China) investigated with C-, L-, and X-bands SAR images.
Journal of Geodynamics 70:28–35.
 5 A Tale of Two Cities
Urbanization in Greensboro,
North Carolina, and
Guiyang, Guizhou, China
Honglin Xiao and Qihao Weng

CONTENTS
5.1 Introduction .................................................................................................... 81
5.2 Study Areas..................................................................................................... 83
5.3 Data and Methodology ...................................................................................84
5.3.1 Data Collection and Preprocessing .....................................................84
5.3.2 Image Classification............................................................................ 85
5.4 Urban Land Changes ...................................................................................... 89
5.5 Urbanization Process Analysis ....................................................................... 91
5.5.1 Urbanization in Guizhou, China......................................................... 91
5.5.2 Urbanization in North Carolina ..........................................................92
5.6 Comparison of the Urbanization Process .......................................................94
5.6.1 Vertical versus Horizontal Development ............................................94
5.6.2 Centralization versus Decentralization............................................... 95
5.6.3 Centrally Planned versus Market Driven ...........................................96
5.6.4 Urbanized versus Urbanizing .............................................................97
5.7 Conclusion ......................................................................................................97
References ................................................................................................................ 98

5.1 INTRODUCTION
Industrialization and urbanization are the primary revolutionary forces that remake
and reshape the world (Carpenter 1966). Urbanization is the process when other types
of land cover, such as agriculture and vegetation, are largely replaced by paved con-
crete surfaces. In the past several decades, significant population increases, migra-
tion, and accelerated socioeconomic development have intensified the urbanization
process in both China and the United States. Not surprisingly, this rapid urbanization
process, along with the ongoing social and economic transitions, has been creating
many environmental problems. However, there are important differences in devel-
opment stages, cultures and societies, urban structures, construction materials, and
physical settings that result in differences of magnitudes, patterns, spatial extents,
and natures of urban growth between the two countries. These differences have then

81
82 Remote Sensing for Sustainability

further resulted in variations in the urban heat budget, surface runoff, water quality,
local climate, vegetation change, and other environmental processes between the
two (Boggs and Sun 2011; Cervero 2000; Gibbard et al. 2005; Jones et al. 1990). In
China, land use and land cover patterns have undergone a fundamental change since
the inception of the economic reform and open-door policies in 1978. Urban growth
has been sped up, and extreme stress on the environment has occurred. Massive
amounts of agricultural and forest land are disappearing each year, converted to
urban or related uses. In recent years, urbanization in China has taken place at an
unprecedented pace and will continue over the next decades. The level of urban-
ization in China has risen from 18% in 1978 to 30% in 1995, and to 42% in 2004.
According to the latest sixth national census data, as of November 1, 2010, China’s
urban population stood at 665.57 million, with an urbanization level of 49.68%. At
the rate of 35 million people per year moving to the cities, China will have nearly
1 billion in the cities and towns by 2030 or a 70% urban population (The Economist
2014). Although the United States is a far more developed country than China, it
also loses a large amount of agricultural and forest land each year to urbanization as
a result of the increasing population, desires for larger and better living spaces, and
economic development. North Carolina in particular has changed significantly since
1990. In 1990, the state was home to 6.6 million people; by 2000, its population sur-
passed the 8 million mark, an increase of 21.4% from 1990. From 2000 to 2010, the
state grew 18.5% compared with the 9.7% overall growth in the United States. This
has resulted in a population of 9.5 million people and led North Carolina to become
the 10th most populous state in the United States (Bauerlein 2011). By 2030, North
Carolina is predicted to have a population of more than 13 million—an increase of
more than 30% from 2010—making it the seventh most populated state in the coun-
try (US Bureau of the Census).
Understanding urbanization characteristics, patterns, and processes is an effec-
tive way to reduce the risk level that the rapid urbanization has brought about to
the environment. Measuring the degrees of the impact from urbanizations in coun-
tries with different cultures, societies, technology backgrounds, and historical tradi-
tions has long been an interest of scientific inquiry (Carpenter 1966; Wirth 1938).
Although no two places experienced identical urban transformation, it is obvious
that there would be similarities in the processes. It is particularly important to com-
pare the urbanization processes in the United States and China. These two countries
account for approximately a quarter of the world population and have the two largest
economies globally. Urbanization from the two countries has had and will have an
even bigger impact on the surface of the Earth.
In satisfying the need, this study compared the urbanization processes in the
North Carolina urban crescent from Raleigh–Durham–Chapel Hill (Research
Triangle Park, RTP) to Greensboro (TRIAD) and Guiyang–Anshun in southwestern
Guizhou Province, China, from the 1980s to the 2010s. Specifically, this study will
first use Landsat TM images from 1991, 2001, and 2007 for Guizhou and TM images
from 1989, 2001, and 2010 for North Carolina to extract land use/land cover (LULC)
information. It will then examine how the urban areas had developed and what the
similarities and differences in the urbanization processes were.
A Tale of Two Cities 83

5.2 STUDY AREAS

In order to make the study results comparable, two study sites were carefully selected.
Guizhou Province in the southwestern China (Figure 5.1a) and North Carolina in the
southeastern United States (Figure 5.1b) were ideal for this study for the following
reasons: First, both provinces/states are relatively rural and rely mostly on agricul-
tural products despite having experienced tremendous urban growth in the past
several decades. Second, North Carolina is located in the Piedmont area of the
Appalachian Mountain leaning toward the Ocean, while Guizhou Province is on
the slope of the Himalaya Mountains. Third, their statuses and degrees of develop-
ment in comparison to their respective countries are similar. Combined together,
Raleigh–Durham–Chapel Hill is ranked the 25th largest urban cluster in the United
States. Guiyang, the capital of Guizhou Province, is ranked the 27th largest city in
China. Additionally, Greensboro, the third largest city in North Carolina and part of
the Piedmont Triad that consists of the area within and surrounding the three major
cities of Greensboro, Winston-Salem, and High Point, has approximately 225,000
residents, while Anshun (the third largest city in Guizhou Province) has a popula-
tion size of 220,000 in its metro area. Furthermore, the TRIAD–RTP area is one of
the most developed areas in North Carolina, while the area that includes Guiyang,
Anshun, Qingzheng, and Pingba County is among the most developed and flattest
areas in Guizhou.
The physical settings are very comparable as well. Guiyang–Anshun in Guizhou
Province is located in the east side of the Yunnan–Guizhou Plateau. It covers
6497 km2 and has a population of 10 million. It has a subtropical wet monsoon
climate. The mean annual temperature is 20°C, with the hottest month being July
and the coldest being January. Annual average precipitation is 1140 mm, with a
distinct summer wet season and a winter dry season. Average monthly humidity
ranges from 74% to 78%. LULC varies from agricultural fields (with crops such as
rice, corn, soybeans, wheat, rapeseed, oats, barley, and sweet potatoes) to natural

(a)
(b)
ince
Prov
hou
Guiz
Qingzheng Guiyang
Dingba
Anshun
N

FIGURE 5.1 The study areas. (a) Guiyang–Anshun, Guizhou, China. (b) RTP–Greensboro,
North Carolina.
84 Remote Sensing for Sustainability

forests in remote areas and local Feng-Shui (literally wind and water) preserves.
In addition, there are barren land, water, and urban/built-up areas. Vegetation is
subtropical with largely evergreen trees, although deciduous trees can also be found
in some areas.
RTP-Greensboro in North Carolina is characterized by hilly, rolling land. It cov-
ers an area of 10,756 km2. According to the recent census data, this area has become
one of the most populous areas in the United States with a population size of approx-
imately 1.4 million (US Census Bureau 2010). It has a humid, subtropical climate.
Winters are short and mild, while summers are usually very hot and humid. Its
average January temperatures range from −1°C to 10°C and July temperatures range
from 20°C to 27°C. The average annual precipitation in the Piedmont is approxi-
mately 1200 mm. This region, located at the slope area of the Smoky Mountains,
contains many of the state’s largest cities. Elevations in the area vary from 100 to
400 m above the mean sea level. The major agricultural products include tobacco,
corn, cotton, hay, peanuts, and vegetable crops. Forests are mainly mixed deciduous
and evergreen trees. The area houses the nation’s largest furniture, tobacco, brick,
and textile producers. Metalworking, chemicals, mining, and paper are also impor-
tant industries.
Since the beginning of China’s economic reforms in 1978, China has experienced
dramatic changes in its environment. Because it is dominated by Karst topography,
Guiyang–Anshun has experienced more significant environmental problems than
most other parts of the nation (Xiao et al. 2003). In recent years, especially since
1990, many of the high-quality agricultural lands are no longer used to produce
crops. Accelerated economic growth has led to the encroachment of urban areas
into nearby agricultural lands. New industrial districts have occupied much of the
available flat land near major cities, while in the countryside, farmers use flat land to
build new houses. In comparison, North Carolina has experienced dramatic change
in the past three decades, with impressive growth and remarkable economic and edu-
cational development. This once rural state has been rapidly urbanized, and in 1992,
North Carolina officially became an urban state when the percentage of the urban
population passed 51%. North Carolina has undergone a 10% population increase
since 2000. Most of the increased population occurred in the Research Triangle,
an area composed of the cities of Raleigh, Durham, and Chapel Hill. With RTP at
its core, the region is one of the country’s fastest-growing metropolitan areas. In
recent years, although the RTP area has been hit hard by the economic recession in
the textile and furniture industries, the information technology and biotechnology
industries are booming.

5.3 DATA AND METHODOLOGY

5.3.1 Data ColleCtion anD PreProCessing
In order to make the comparison more meaningful and, at the same time, cover a
sufficiently long period, Landsat TM scenes were carefully chosen for each study
area. Although obtaining images of exactly matching times in the same years over
the same period proved impossible because of the limitation in data availability, we
A Tale of Two Cities 85

had managed to obtain three cloud-free images for each area that were acquired
in approximately the same seasons and covered a similar time frame. The images
for Guizhou were acquired in November or December 1991, 2000, and 2007 and
the images for North Carolina were obtained in July or August 1989, 2001, and
2010. Because these images in each area were taken at a similar time of year, the
phonological effect was minimized. Other ancillary data, such as county bound-
ary files, topographic maps, vegetation maps of Guizhou, DEMs, historical aerial
photographs, and LULC records from the two study areas, were also collected for
the study.
The satellite images were corrected to remove atmospheric effects by using Erdas
Imagine’s Atmospheric Correct Extension (ATCOR 3), which used an algorithm
based on the methods of Richter (1996) and Zhang et al. (2002). The images were
then georectified using ground control points collected by GPS. The images were
resampled to 30 m pixel size for all bands using the nearest-neighbor method. All
the data were projected to a common UTM coordinate system for the purpose of
analysis.

5.3.2 image ClassifiCation

On the basis of the spatial resolution of the satellite images used and the study objec-
tive, LULC types for Guizhou were classified in accordance with the classification
system developed by the Resource and Environment Information Center of Chinese
Academy of Sciences (CAS) (Peng et al. 2011) (Table 5.1). For North Carolina, LULC
was classified with a modified Anderson’s system (Anderson et al. 1976) in refer-
ence to the US Geological Survey NLCD system (Homer et al. 2007) (Table 5.2). A
hybrid image classification of unsupervised and supervised classifiers was applied.
Studies have found that hybrid classifiers are particularly valuable in the analysis
of LULC change, particularly in the area where there is complex variability in the
spectral patterns or individual cover types (Lillesand et al. 2008; Mas 1999). Hybrid
unsupervised/supervised classifications have been conducted in many LULC stud-
ies, and in most cases, the results are better than those from supervised or unsuper-
vised classifications alone (García and Álvarez 1994; Rutchey and Vilcheck 1994).
The unsupervised classification was carried out using the Iterative Self-Organizing
Data Analysis algorithm to identify spectral clusters in the images. On the basis
of the results of the unsupervised classifications, training sites were chosen from
the images. For each image, spectral signatures for the training sites were carefully
chosen and examined. A maximum likelihood classifier was then employed for the
image classification. Land cover types for Guizhou were first classified into 11 types,
including water, urban construction land, rural settlement, transportation and min-
ing land, dense forest land, shrub land, fruit land, tea gardens, grass land, rice paddy
land, arid agricultural land, vegetable land, and rocky desertification land. Land
covers for North Carolina were first classified into 12 types of second-order land
covers, including water, open space, low intensity, medium intensity, high intensity,
deciduous forest, evergreen forest, mixed forest, shrub/scrub, grassland/herbaceous,
pasture/hay, and cultivated crops. They were then combined to form into the first-
order lands (Table 5.2).
86 Remote Sensing for Sustainability

TABLE 5.1
Land Cover Classification Scheme for Guizhou, China
Level I Types Level II Types
Order Type Order Type Description
1 Water area 40 Water Mainly rivers and lakes and looking nearly
black and homogeneous with clear boundary.
2 Urban or 21 Urban Built-up areas in cities and towns. Looking
built-up construction bright green on RS image.
land
22 Rural Mainly big villages, and usually scattered into
settlements arable lands. Look like urban built-up areas
but much smaller on RS image.
23 Transportation Mainly mining areas and transportation
and mining facilities, such as motorways. Looking bright
land green and discernable by its line shape.
3 Wood 31 Dense forest Patches of secondary coniferous forest, mainly
land land dense redpines. Looking dark red and being
discernable on RS image with clear boundary.
32 Shrub land Mainly growing dense broad-leaf shrubbery
usually less than 3 m tall, usually distributing
on mountainous areas with bright red and
veined patches on RS image.
33 Fruit tree land Mainly different types of fruit trees.
34 Tea garden land Mainly tea gardens. Looking shiny red but
homogeneous with clear boundary on RS
image.
4 Grass land 40 Grass land Covered by grasses, usually containing lots of
thorns and scattered shrubbery with crown
density<10%. Looking light pink with veins.
5 Arable 51 Paddy land Mainly used for growing rice. Usually
land distributing in bottom of valleys and basins
and lying near rivers and lakes. Being
homogeneous patches with clear boundary.
52 Arid land Mainly used for growing corn and wheat with
poor irrigation facility. Usually lying in hilly
areas and looking bright gray on false-colored
RS image.
53 Vegetable land Mainly used for growing different kinds of
vegetables. Usually distributed around cities
and towns as patches with clear boundaries.
Normally looking pinkish on RS image.
6 Barren 60 Rocky Rocky karst hills with lot of rocks exposed and
land desertification bare, and usually covered by limited grasses
land and unused. Looking dark gray on RS image.
A Tale of Two Cities 87

TABLE 5.2
Land Cover Classification Scheme for North Carolina
Level I Types Level II Types
Order Type Order Type Description
1 Water area 40 Water All areas of open water, generally with less than
25% cover of vegetation or soil, looking
nearly black and homogeneous with clear
boundary.
2 Urban or 21 Open space Includes areas with a mixture of some
built-up constructed materials, but mostly vegetation in
the form of lawn grasses. Impervious surfaces
account for less than 20% of total cover. These
areas most commonly include large-lot
single-family housing units, parks, golf
courses, and vegetation planted in developed
settings for recreation, erosion control, or
aesthetic purposes. Looking bright green on
RS image.
22 Low Includes areas with a mixture of constructed
intensity materials and vegetation. Impervious surfaces
account for 20%–49% of total cover. These
areas most commonly include single-family
housing units. Looking like urban built-up
areas but much smaller on RS image.
23 Medium Includes areas with a mixture of constructed
intensity materials and vegetation. Impervious surfaces
account for 50%–79% of the total cover.
These areas most commonly include
single-family housing units. Looking bright
green and discernable by its line shape.
24 High Includes highly developed areas where people
intensity reside or work in high numbers. Examples
include apartment complexes, row houses, and
commercial/industrial. Impervious surfaces
account for 80%–100% of the total cover.
3 Wood land 31 Deciduous Areas dominated by trees generally greater than
forest 5 m tall, and greater than 20% of total
vegetation cover. More than 75% of the tree
species shed foliage simultaneously in
response to seasonal change. Looking dark red
and being discernable on RS image with clear
boundary.
(Continued)
88 Remote Sensing for Sustainability

TABLE 5.2 (CONTINUED)

Land Cover Classification Scheme for North Carolina
Level I Types Level II Types
Order Type Order Type Description
32 Evergreen Areas dominated by trees generally greater than
forest 5 m tall, and greater than 20% of total
vegetation cover. More than 75% of the tree
species maintain their leaves all year. Canopy
is never without green foliage.
33 Mixed forest Areas dominated by trees generally greater than
5 m tall, and greater than 20% of total
vegetation cover. Neither deciduous nor
evergreen species are greater than 75% of total
tree cover.
34 Shrub/scrub Areas dominated by shrubs; less than 5 m tall
with shrub canopy typically greater than 20% of
total vegetation. This class includes true shrubs,
young trees in an early successional stage, or
trees stunted from environmental conditions
with bright red and veined patches on RS image.
4 Grassland 40 Grassland Areas dominated by grammanoid or herbaceous
herbaceous herbaceous vegetation, generally greater than 80% of total
vegetation. These areas are not subject to
intensive management such as tilling, but can
be utilized for grazing. Looking light pink
with veins.
5 Agricultural 51 Pasture/hay Areas of grasses, legumes, or grass-legume
land mixtures planted for livestock grazing or the
production of seed or hay crops, typically on a
perennial cycle. Pasture/hay vegetation
accounts for greater than 20% of total
vegetation. Being homogeneous patches with
clear boundary.
52 Cultivated Areas used for the production of annual crops,
crops such as corn, soybeans, vegetables, tobacco,
and cotton, and also perennial woody crops
such as orchards and vineyards. Crop
vegetation accounts for greater than 20% of
total vegetation. This class also includes all
land being actively tilled and looking bright
gray on false-colored RS image.
53 Vegetable Mainly used for growing different kinds of
land vegetables. Usually distributed around cities
and towns as patches with clear boundaries.
Normally looking pinkish on RS image.
(Continued)
A Tale of Two Cities 89

TABLE 5.2 (CONTINUED)

Land Cover Classification Scheme for North Carolina
Level I Types Level II Types
Order Type Order Type Description
6 Woody 69 Woody Areas where forest or shrub land vegetation
wetlands wetlands accounts for greater than 20% of vegetative
cover and the soil or substrate is periodically
saturated with or covered with water.
7 Barren land 70 Rock/sand/ Barren areas of bedrock, desert pavement,
clay scarps, talus, slides, volcanic material, glacial
debris, sand dunes, strip mines, gravel pits,
and other accumulations of earthen material.
Generally, vegetation accounts for less than
15% of total cover. Looking dark gray on
RS image.

It is difficult to classify land cover types by remote sensing interpretation alone,

especially in the mountainous karst areas in Guizhou, because of the influence of
rugged terrain and fragmented land covers. The authors’ knowledge of and famil-
iarity to this study areas played an important role in the LULC classification pro-
cess. Postclassification sorting was performed to enhance the classification results.
Occasionally, Google Earth was used to help correct the misclassifications. The over-
all accuracy for the three images in Guizhou was 89.3%, 90.5%, and 91.5%, respec-
tively, while the overall accuracy for the three maps in North Carolina was 91.3%,
93.4%, and 90.6%, respectively. To identify the spatial–temporal patterns of LULC
changes, and to examine the conversion from one class to another, a thematic change
detection analysis was conducted using the Image Analyst Extension from ArcGIS
software. LULC maps were superimposed to create change maps between different
periods. To emphasize changes in urban and built-up, other types of changes were
dropped from the analysis of change detection.

5.4 URBAN LAND CHANGES

Both study areas have experienced significant changes on urban land during the
past two decades (Figures 5.2 and 5.3). In Guizhou, urban land use only accounted
for 2.5% of the study area in 1989. It increased to 4.75% in 2000, and then further
jumped to 16.79% in 2007. In less than 20 years, the urban land increased sixfold,
from 40,322 to 269,530 acres. Of the lands changed to urban, rice paddy land contrib-
uted the most at 88,310 acres (Figure 5.4a). Substantial amounts of dry agricultural
and forest lands had also been converted into urban use. The loss of dry agricultural
land was slightly less than rice paddy at 77,330 acres, which was followed by forest at
53,160 acres and then grassland at 17,460 acres. The amounts of shrub land, non-rice
crop land, and barren land that were converted into urban were 13,010, 12,552, and
14,790 acres, respectively.
90 Remote Sensing for Sustainability

Legend
County line ng
ya
Urban 1991 Qingzheng ui
G
Urban 2000

Urban 2007
ba
ng
Pi

An
sh
un N

0 5 10 20 30 40
km

FIGURE 5.2 Urban expansion map in Guiyang–Anshun, Guizhou, China, from 1991 to 2007.

Alamance Orange
D
ur
ha
m
Greensboro
Du
Guilford

rh
am
ill

Burlington
H
el
ap
Ch

Raleigh
Randolph

e
ak

Cary
W

Chatham

Legend
N
County line
Urban 1989
Urban 2001
km
Urban 2010 0 5 10 20 30 40

FIGURE 5.3 Urban expansion map for RTP–Greensboro, North Carolina, from 1989 to
2010.
A Tale of Two Cities 91

100,000

80,000
Area (acres)
60,000

40,000

20,000

le
ss

re

en
sh

dy

ab
(a)
ra

tu
Bu

rr
ad
G

et
ul

Ba
p

g
ic

ce

Ve
gr

Ri
A

400,000

300,000
Area (acres)

200,000

100,000

0
t

nd

d
ay
s

te
re

-h
la

va
Fo

ss

re

lti

(b)
ra

tu

Cu
G

s
Pa

FIGURE 5.4 The amount of land converted into urban area. (a) Guiyang–Anshun from
1989 to 2007. (b) RTP–TRIAD in North Carolina from 1989 to 2010.

In the RTP–TRIAD area, urban and built-up land accounted for 5.18% of the
total study area in 1989 (Figure 5.3). It more than tripled to 18.01% in 2001, and then
almost doubled to 31.37% in 2010. In the past two decades, 397,790 acres of forest
lands had been converted into urban or built-up land (Figure 5.4b). Almost equal
amounts of pasture/hay land and cultivated lands were transformed into houses,
offices, impervious surfaces, golf courses, and so on. Their loss stood at 159,109 and
162,385 acres, respectively. In addition, approximately 30,000 acres of grassland
was converted into urban and built-up from 1989 to 2010.

5.5 URBANIZATION PROCESS ANALYSIS

5.5.1 Urbanization in gUizhoU, China
Guizhou has experienced tremendous change in the urban area since the induction of
China’s open-door policy in the late 1970s. It has gone through two stages of urban
development during the study period. During the first phase from 1989 to 2001,
urban development was focused on the renewal of the urban space as a result of
gentrification. Much of the effort was put on improving the quality of life and human
92 Remote Sensing for Sustainability

welfare in the urban areas. As a result, services and infrastructure such as water and
sewerage systems, sanitation, transport, gas pipeline, and solid waste collection and
disposal in the cities such as Guiyang and Anshun were greatly improved. During
that time, urban land development was mainly accomplished through government-
affiliated companies. However, organizations and individuals could acquire non-
transferable land-use rights from the governments to build houses for their workers
through a special land allocation policy (Du et al. 2011).
Close to the end of the first study period from 1989 to 2000, China had made
tremendous progress in its economic growth that paved the way for China’s future
modernization. On the basis of the achievement since the open-door policy, the cen-
tral government decided to speed up the urbanization process in order to stimulate
the economy growth by promoting urbanization at a rate of 1.5% to 2% annually
(Song and Ding 2007). This urbanization policy was facilitated in several combined
factors including the government, the marketplace, and foreign and domestic invest-
ment in real estate and manufacturing. In order to stimulate the demand for urban
land and urban labor, the government issued policies to make it easy for companies
and state-owned enterprises to obtain land for urban construction. For example, the
Constitution Amendment Act to Article 10 in 1988 made transferring of the land use
right to others possible. On the basis of this Amendment, companies, organizations,
and individuals could negotiate and purchase land-use rights with governments or in
the secondary land market from existing rights-holders (Du et al. 2011). During this
period, the state-owned or collective-owned land developers could often work with
the government to purchase land at prices that were far below the market prices for
the sake of urban development.
During the second study period from 2001 to 2007, the main strategies were to
expand the established cities by creating new urban districts to absorb the hundreds
of thousands of migrants from the countryside and reduce the amount of peasants
by merging rural townships and villages (Che xiang bing cun). A large percentage of
the urban sprawl was a result of the establishment of several suburban Economic and
Technological Development Zones (EDZs), such as “Jingyang New Development
District” in Guiyang City, where the local government attracted companies to estab-
lish business or to build factories and real estate companies to build thousands of
residential building complexes by providing low-priced land, incentives, and free
infrastructure. To boost this new urban development, the city moved all of its govern-
ment developments from the old core to this newly established city. Approximately
half of the urban land expansion in Guiyang was caused by government-led develop-
ment of EDZs (Ding 2003; Song and Ding 2007).

5.5.2 Urbanization in north Carolina

During the two study periods, the urban growth in the RTP–TRIAD area had
unique expansion patterns as well. In the first phase, urban expansion permeated
almost everywhere throughout the study area (Figure 5.3). However, the expansions
were not distributed evenly among the various types of areas because location was
a critical determinant on the types of growth each area might have. For example,
the growth in the RTP and Greensboro area were mostly in highly developed areas
A Tale of Two Cities 93

for commercial/industrial buildings and apartment complexes; in the rest of the

area, the growth was mainly sporadic single-family housing units. In the early
years of the first study period, because of easy access, relatively cheap land, and
convenience of doing business, the new development expanded around the previ-
ously established urban/business core. The urban growth patterns depended heav-
ily on the prior density: growth went where there was room for it, filling in the
urban tracts and lower-density edge of the city tracts. However, as the open lands
were filled up toward their practical maximum density, their growth slowed down
(Hartgen 2003).
As the land parcels became harder to assemble, prices rose, and the remaining
lots were too expensive. Many companies or real estate realtors started to build their
new businesses or homes in the areas that were not too far away from the urban core,
much cheaper with low density and still close enough for daily commuters. This
urban development phenomenon has been called counter-urbanization, a process
of population de-concentration characterized by urban areas or communities with
smaller sizes, lower densities, and more local homogeneity (Brian 1977, 1980). Most
of the growth has become housing and entertainment communities for the burgeon-
ing RTP and Greensboro areas. It is estimated that approximately half of the work-
ing population were living in places at least 15 min away from their workplaces.
As more and more land went toward low-density housing subdivisions, streets, golf
courses, strip malls, and other related developments, this came at the cost of mas-
sive tracts of forest land and farmland, sometimes wasted because of the low land
use efficiency. The sprawling growth of suburban areas was largely responsible for
the loss of almost half a million acres of farmland between 1997 and 2002 in North
Carolina (Stuart 2010). During the time, land consumption was increasing four times
faster than the rate of population growth. When government policies were issued to
make it more difficult for new development to get approval, it pushed the additional
growth to the region’s remote areas.
The urbanization process accelerated in the beginning of the second study period
at around 2001 when more land was converted into residential houses to fulfill hous-
ing demand. The financial institutions artificially created an easier environment for
people to access home mortgages, which boosted a housing bubble in this area. On
top of a primary residence home, many households also purchased additional houses
for investment/speculative purposes. This development pattern, however, took a
sharp turn in the second study phase. Since approximately 2006, the area was hit
hard by the economic meltdown and when the housing bubble burst. The decreased
house values and the sluggish economy had caused a serious budget crisis for local
governments in maintaining normal operations. In order to find ways to fill the bud-
get deficit, some cities in the study area had launched projects targeted at urban
renewal or development, although much different from the earlier versions in the
1950s to the 1960s. For example, the town of Burlington in Alamance County was
once a hub for the textile industry. In recent years, because of the challenges from
foreign countries, most of the factories were closed down. This had led to mass
unemployment and residential houses foreclosed or abandoned. To deal with the
budget crisis and find a new economic engine, the city initiated a massive project
by relocating its downtown shopping center to a location right off the nearby major
94 Remote Sensing for Sustainability

highway, interstate highway I-85/40, aimed at drawing business from the travelers or
nearby towns in order to revitalize its troubled economy. This new trend of renewal
has resulted in urban sprawl into the less congested areas while creating easier access
to the freeways and expressways.

5.6 COMPARISON OF THE URBANIZATION PROCESS

Although both of the two study areas experienced tremendous increases in urban
area and shared many similarities in the urbanization process, the two study areas
had profoundly different cultures and economic and political systems. Consequently,
they followed differing ways of development.

5.6.1 VertiCal VersUs horizontal DeVeloPment

The urbanization in China can be characterized as vertically expanded, in con-
trast to horizontally expanded in the United States. Guiyang–Anshun had 269,530
acres of urban land with a population of 8 million, while RTP–TRIAD only had
a population of approximately 2 million but with 833,897 acres of urban area. In
order to use the limited land efficiently and effectively, the Chinese cities gravi-
tated toward building concentrated housing in tall complexes. To maximize land
use efficiency, city governments issued regulations ensuring that newly built com-
mercial and residential buildings reached a minimum number of floors. In 1991, the
average number of floors for the residential buildings was approximately six, which
increased to approximately 15 in 2008. The tallest building in Guiyang was 28 sto-
ries in 1990. In 2008, buildings of 30 to 40 stories became a norm in Guiyang and
Anshun. As a result, the cities were congested with clusters of so-called building
forests (Figure 5.5a) with not much open space between the buildings. On the other
hand, the urban area in the United States was more spread out with low density
(Figure 5.5b). Residential buildings in the United States were mostly one to two
stories with front yards and backyards. Houses had ample space between with lawns
and fences/shrubs separating them. This difference in the dimension of the urban
development has resulted in the observed difference in transformations of the urban
layout between the two study areas. The Chinese cities were much more congested,
with more homogeneous construction materials and roof and surface types. The
US cities, on the other hand, were diverse with large portions of land allocated to
nonurban use.
The vertical development pattern faced a change over the past few years in
Guiyang. As the Chinese economy continued to boom, the relatively affluent citi-
zens wanted to have a US standard of living. Consequently, many luxurious single-
house villas were built to meet this demand. The scale of horizontal development in
Guiyang was small, but had a great impact on its environment. The majority of the
development was at the expense of the limited arable land in this karst region. The
US cities, in contrast, had changed to add more vertical development. As the land
became scarcer and more expensive, large backyards and the unused lands between
communities were consequently more expensive owing to opportunity cost. For
example, in the city of Chapel Hill, the lot size of the houses built in the years of
A Tale of Two Cities 95

(a) (b)

FIGURE 5.5 Vertical versus horizontal development in Guizhou and North Carolina. Both
photos were captured from Google Earth. (a) A small part of Jingyang New Development
Zone in Guiyang City. (b) A typical residential area in the RTP–Greensboro area.

the study period had been shrinking and there were more high-rise complexes being
built than before. In addition, these new construction projects were inserted into the
forest or open space between the established communities as in-fill developments
instead of leap-frog clusters into new open areas.

5.6.2 Centralization VersUs DeCentralization

During the study periods, most urban growth in Guizhou Province had been cen-
tered on the old established cores, before gradually radiating to the surrounding
areas. Cities expanded outward in the form of rings through different development
stages. The first phase started in the early 1990s from the renewal or redevelop-
ment of the old city cores, where aging buildings were demolished and replaced
by modern skyscrapers and residential high-rise complexes. After 1995, the city
started its second phase of development by encroaching from different directions
into the closest available lands. Once the new expansion could not keep up with
the ever-increasing demand, it would then continue its new phase of expansion.
In China, there was a big difference between the quality of life in the city and
the surrounding rural area in terms of public facilities, services, job opportuni-
ties, and so on, and thus it had become a universal goal to own an apartment in
the urban area for people both living inside the cities and in the rural villages.
When possible, they would spend their life savings to purchase a house in the
city. Because of the massive influx of the rural people into the cities, the city
boundaries had been pushed further away into the neighboring agricultural or
other nonurban lands.
96 Remote Sensing for Sustainability

The urbanization process in North Carolina was more marked in contrast with
that of Guiyang, especially during the first study period. Instead of agglomerat-
ing around the old city cores, urban sprawl was evident throughout the study area.
Most urban areas experienced a population shift from the central city to suburban
regions. Directly opposite to the situation in China, the quality of life in the suburban
areas was much better than the condensed urban area. This decentralized urban-
ization phenomenon was commonly visible around the US cities, where the urban
expansions were scattered and far away from the central urban cores (Schneider
and Woodcock 2008). Many new shopping malls, public facilities, parking lots, and
highway extended into those once rural areas to accompany the newly built residen-
tial communities. Furthermore, the materials used for urban construction were very
different between the urban core and the local communities. In order to sustain the
weight of the high-rises in the downtown area, brick, cement, and steel were the
main materials for the construction. For the suburban cities, most buildings (espe-
cially residential ones) had wooden frames. This construction design in the suburbs
had a more natural environment than the city core.

5.6.3 Centrally PlanneD VersUs market DriVen

The urbanization process in China was closely associated with the government’s
policies and plans. In China, all the land belonged either to the state or collectives.
Individuals by law are unable to own land, but could obtain land use rights for
specific purposes. The rights could not be transferred or sold in the market. When
necessary, the government could take away the land from individuals for use in
other government planned purposes. As a result, the degree, extent, and pattern of
urban development were largely controlled by the government. In the early 1990s,
the urban development was limited within the established old urban cores because
of the government’s policy to protect the farmland (Xiao and Weng 2007). In the
1990s, because of worries about food shortages as a result of farmland loss and
population increases, the central government issued a series of strict regulations
preventing land conversion. Starting around 2001, the urbanization process accel-
erated because of the government’s policy shift to stimulate economy growth by
promoting urbanization (Song and Ding 2007). For the first time, private real estate
developers were allowed to share the market with the government, and the urbaniza-
tion process accelerated at an unprecedented speed as the result. During this period,
most of the urban expansion was a result of the establishment of New Development
Zones (NDZs) by the government. For example, the site of Jingyang New District in
Guiyang was a prisoner training camp surrounded by farmland before 2000. After
the government decided to turn this area into another new city center by providing
low-priced land and free infrastructure to factories and real estate developers, this
once remote area had grown into a metro area as large as the old urban area. New
urban districts expanded from the existing cities into the adjacent rural areas by tak-
ing over large areas of rice paddies or other lands (Ding 2003).
The urbanization process in the United States was mainly driven by market
forces, since it was the developers and individuals and not government policies that
drove most of the urban development. The housing markets as a whole functioned
A Tale of Two Cities 97

systematically and orderly from many seemingly random decisions between the con-
sumer preferences and the products and services from the entrepreneurs and pro-
ducers (Blair 1995). Decisions about what land to develop, and whether to build
single houses or condominiums, were based on the developers’ judgments about the
conditions of the market. The land developers would completely take on the risks
and rewards associated with their own decisions. When developers saw opportuni-
ties to make profits from developing certain areas, they would first buy the lands
from the owners, and then improve the properties by turning them into homes, office
buildings, or shopping malls, and later sell or lease them to businesses and fami-
lies willing to pay their prices (McDonald 1997; Staley 1999). In the land market,
the developers and consumers made decisions on the land and building prices from
different sources including the home supply, office availability, and factory invento-
ries (Henderson 1988). In summation, the urban area expanded in accordance with
the developers and the consumers’ interests. Over the past two decades, because
there was a strong demand for homes and office buildings, large areas of farmland
and forest land had been turned into urban and built-up areas in the RTP area. In
Burlington, while one side of the city was being abandoned, the area closest to inter-
state I-40/85 boomed because of the new modern shopping malls and cheap but
high-quality residential homes.

5.6.4 UrbanizeD VersUs Urbanizing

North Carolina was much more urbanized than Guizhou before the study period. It
had undergone several decades of speedy urbanization after World War II. Its infra-
structure, transportation system, and public facilities were very well developed at
the beginning of the study period. The creation of the RTP in the 1950s had helped
greatly in elevating this area into one of the most urbanized areas in the United
States. Guizhou Province, however, was at the early stage of its development with
only 2.5% urban land use. Back then, China’s urbanization rate was generally among
some of the lowest in the world, and Guizhou was one of the least developed areas
in the nation. Because of the difference in development stages, the characteristics of
urbanization were quite different between the two places. The type of development
in North Carolina was much more patched, in-filling, and sporadic. Most of the con-
struction was to improve the already established ones and was fairly spread through-
out the study area. A large portion of the urban expansion was building golf courses,
baseball fields, or parks to improve the quality of life. In Guizhou, on the other hand,
the development was geared more toward fundamental infrastructure. Most of the
urban increase was to build basic infrastructures such as airports, New Development
Districts, new residential centers and neighborhoods, or modern highways to loop
around the cities and connect them to other cities.

5.7 CONCLUSION
Our study result indicates that the United States and China share a lot of similari-
ties and differences in the processes of urbanization. In the past two decades, urban
areas have grown significantly at an alarming speed of approximately 500% in the
98 Remote Sensing for Sustainability

two study sites. In Guiyang–Anshun, Guizhou Province, China, the urban area grew
from 2.75% in 1991 to 16.78% in 2007, while in RTP–TRIAD, North Carolina, the
increase was from 5.18% in 1989 to 31.37% in 2010. As a result of the urban growth, a
great deal of nonurban land was lost. In Guizhou, the urban land was converted from
vegetable land, shrub land, grass land, dry agriculture, and rice paddies. In North
Carolina, the losses were mostly from areas of forest, grassland/shrub, pasture/hay,
and cultivated lands.
Because of the differences in political systems, economic development stages,
cultures, and physical settings, the two study areas followed different paths of urban-
ization. In Guizhou, because of the shortage of land resources, the cities were more
congested and built with skyscrapers. Cities in North Carolina, on the other hand,
were developed with low density. Because of the common pursuit of better living
environments and amenities, the urban areas were more spread out horizontally.
In Guizhou, urban areas grew out from the old established cores and radiated into
the surrounding areas. These areas were developed according to the central govern-
ment’s plans. In North Carolina, the developers and the landowners chose where and
what type of growth urban expansion should be. In addition, because the two study
sites were in different development stages, the nature of the urban expansion was
quite different. Guizhou’s urban growth focused more on infrastructure and basic
facilities, while the growth in North Carolina was to build parks, golf courses, and
athletic fields and other amenities to improve the quality of life.

REFERENCES
Anderson, J. R., Hardy, E. E., Roach, J. T., and Witmer, R. E., 1976. A Land Use and Land
Cover Classification System for Use with Remote Sensor Data: U.S. Geological Survey
Professional Paper 964, 28 pp.
Bauerlein, V., 2011. Population Jumps in North Carolina. The Wall Street Journal, March 3,
2011.
Blair, J. P., 1995. Local Economic Development. Beverly Hills, California: Sage Publications.
Boggs, J. L., and Sun, G., 2011. Urbanization alters watershed hydrology in the Piedmont of
North Carolina. Ecohydrology, 4: 256–264.
Brian J. L. Berry (ed.), 1977. Urbanization and Counter Urbanization. London: Sage
Publications.
Brian, B. J. L., 1980. Urbanization and counterurbanization in the United States. Annals of the
American Academy of Political and Social Science, Changing Cities: A Challenge to
Planning, 451: 13–20.
Carpenter, D., 1966. Urbanization in the United States and Japan. Studies in Comparative
International Development, 2(3): 38–49.
Cervero, R., 2000. Efficient Urbanization: Economic Performance and the Shape of the
Metropolis. Policy Working Paper. Lincoln Institute of Land Policy. Cambridge, MA.
Ding, C., 2003. Land policy reform in China: Assessment and prospects. Land Use Policy,
20: 109–120.
Du, H., Ma, Y., and An, Y., 2011. The impact of land policy on the relation between hous-
ing and land prices: Evidence from China. The Quarterly Review of Economics and
Finance, 51(1): 19–27.
García, M. C., and Álvarez, R., 1994. TM digital processing of a tropical forest region in
southeastern Mexico. International Journal of Remote Sensing, 15, 1611–1632.
A Tale of Two Cities 99

Gibbard, S. G., Caldeira, K., Bala, G., Phillips, T.J., Wickett, M. 2005. Climate effects of
global land cover change. Geophysical Research Letters, 32(L23705): 4.
Hartgen, D. T., 2003. Highways and Sprawl in North Carolina. Policy Summary.
Henderson, J. V., 1988. Urban Development: Theory, Fact, and Illusion. Oxford University
Press, New York.
Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Herold, N., McKerrow, A.,
VanDriel, J. N., and Wickham, J., 2007. Completion of the 2001 National Land Cover
Database for the Conterminous United States. Photogrammetric Engineering and
Remote Sensing, 73(4): 337–341.
Jones, P. D., Groisman, P. Ya, Coughlan, M., Plummer, N., Wang, W-C., and Karl, T. R., 1990.
Assessment of urbanization effects in time series of surface air temperature over land.
Nature 347, 169–172.
Lillesand, T., Kiefer, R. W., and Chipman, J., 2008. Remote Sensing and Image Interpretation,
6th Edition, Wiley. 968 pp.
Mas, J., 1999. Monitoring land-cover changes: A comparison of change detection techniques.
International Journal of Remote Sensing, 20: 139–152.
McDonald, J. F., 1997. Fundamentals of Urban Economics. Prentice Hall, Inc., Englewood
Cliffs, New Jersey.
Peng, J., Xu, Y. Q., Cai, Y. L., and Xiao, H. L., 2011. The role of policies in land use/cover
change since the 1970s in ecologically fragile karst areas of Southwest China: A case
study on the Maotiaohe watershed. Environmental Science & Policy. 14(4): 408–418.
Richter, R., 1996. A spatially adaptive fast atmospheric correction algorithm. International
Journal of Remote Sensing, 17: 1201–1214.
Rutchey, K., and Vilcheck, L., 1994. Development of an everglades vegetation map using
a SPOT image and the global positioning system. Photogrammetric Engineering and
Remote Sensing, 60(6): 767–775.
Schneider, A., and Woodcock, C. E., 2008. Compact, dispersed, fragmented, extensive? A
comparison of urban growth in twenty-five global cities using remotely sensed data,
pattern metrics and census information. Urban Studies, 45: 659.
Song, Y., and Ding, C., 2007. Urbanization in china: Critical issues in an era of rapid growth.
Lincoln Institute of Land Policy, Cambridge, MA. 302 pp.
Staley, S. R., 1999. “Urban Sprawl” and the Michigan Landscape: A Market-Oriented
Approach. A joint project for Michigan from the Mackinac Center for Public Policy and
the Reason Public Policy Institute.
Stuart, A. W., 2010. Recent Population Change in North Carolina. North Carolina Museum of
History Office of Archives and History, N.C. Department of Cultural Resources.
The Economist, 2014. Building the dream, Special Report, The Economist Newspaper Limited.
Apr 19, 2014.
U.S. Census Bureau. 2010. Population and housing unit counts for the United States, regions,
divisions, and American Indian, Alaska Native, and Native Hawaiian Areas, from http://
www.census.gov/2010census/.
Wirth, L., 1938. Urbanism as a way of life. The American Journal of Sociology, 44(1) 1–24.
Xiao, H., Brook, G., and Lo, C., 2003. Evidence from satellite images of human-induced
environment change in Guizhou Province, China 1991–1998. Asian Geographer, 22
(1 and 2).
Xiao, H., and Weng, Q., 2007. The impact of land use and land cover changes on land sur-
face temperature in a karst area of China. Journal of Environmental Management, 85:
245–257.
Zhang, Y., Guindon, B., and Cihlar, J., 2002. An image transform to characterize and com-
pensate for spatial variations in thin cloud contamination of Landsat images. Remote
Sensing of Environment, 82: 173–187.
Section II
Remote Sensing for Sustainable
Natural Resources
 6 Role of Remote Sensing
in Sustainable Grassland
Management
A Review and Case
Studies for a Mixed-Grass
Prairie Ecosystem
Alexander Tong, Bing Lu,
Yuhong He, and Xulin Guo

CONTENTS
6.1 Introduction .................................................................................................. 104
6.2 Case Studies .................................................................................................. 105
6.2.1 Assessing Grassland Health Using Remote Sensing–Derived
Biophysical Properties ...................................................................... 107
6.2.2 Mapping Chlorophyll Content as an Indicator of Grassland Health....107
6.2.3 Investigating Grassland Disturbance Using Remote Sensing .......... 110
6.2.4 Determining the Effects of Climatic Factors on Grassland
Health Using Remote Sensing .......................................................... 111
6.2.5 Mapping Habitat for Endangered Species Using Remote Sensing ... 112
6.3 Conclusions, Challenges, and Opportunities ................................................ 114
6.3.1 Conclusion ........................................................................................ 114
6.3.2 Challenges and Opportunities .......................................................... 115
6.3.2.1 Image Acquisition and Processing ..................................... 115
6.3.2.2 Species-Level Monitoring .................................................. 115
6.3.2.3 Vegetation Structure Mapping ........................................... 116
6.3.2.4 Snow, Topography, and Soil Moisture Mapping ................ 116
6.3.2.5 Logistics ............................................................................. 116
Acknowledgment ................................................................................................... 117
References .............................................................................................................. 117

103
104 Remote Sensing for Sustainability

6.1 INTRODUCTION
Grasslands are critically important ecosystems that serve as habitats for a variety of
species of perennial grasses and other herbaceous vegetation, birds, animals, and
insects. However, a majority of grassland biomes have been transformed, converted,
or altered by human intervention on a global scale, with very few natural intact areas
remaining. In North America, the grassland biome was once the most extensive, but
has become one of the most threatened ecosystems (Samson and Knopf 1994). Since
the remaining grassland ecosystems are inherently fragile, effective management strat-
egies are important to address impacts driven by anthropogenic activities such as over-
grazing, urban expansion, agricultural intensification, invasive species, and wildfire
suppression under a changing climate. In fact, grassland management strategies have
been moving toward integrated ecosystem and landscape-based approaches to address
the impacts by focusing on areas such as biodiversity conservation, habitat restoration,
and sustainable resource management (Bizikova 2009; Estrada-Carmona et al. 2014;
Jun 2006). Such broad-scale plans bring partners and stakeholders together to realize
common and shared objectives that consider both local and landscape-wide needs.
Monitoring grassland changes across space and time is the first step leading to
effective management plans. Since grasslands are central to the livelihoods of more
than a billion low-income people, managing grasslands has been a balance between
competing demands, especially between economic returns and ecosystem services.
Traditional ways of grassland monitoring that rely on field surveys are typically
expensive and labor-intensive. In addition, field-based methods only provide localized
information that presents a challenge when extrapolating it over large areas, and the
information is thus often insufficient for land managers. Alternatively, remote sensing
has become increasingly important for grassland ecosystem monitoring and guid-
ing sustainable management practices (Boval and Dixon 2012). Once remote sensing
technologies and techniques are validated at a local level, they can be easily general-
ized and used for long-term monitoring at a range of spatial and temporal scales.
Remote sensing imagery with a variety of spatial and temporal resolutions can be
utilized for different management purposes with modest budgets. For practical and eco-
nomic reasons, multispectral image data including Système Probatoire d’Observation
de la Terre (SPOT), Landsat, Moderate Resolution Imaging Spectroradiometer
(MODIS), and Advanced Very High Resolution Radiometer (AVHRR) images
acquired from spaceborne sensors are commonly used for studying large geographi-
cal areas. Whereas SPOT and Landsat images offer high and medium spatial reso-
lution data, respectively, they also have correspondingly lower temporal resolutions
in comparison to the coarse spatial, but high temporal resolutions of MODIS and
AVHRR data. Still, the most extensively used satellite imagery for research and appli-
cation has been Landsat, as it offers a long historic archive that can be used to map
long-term spatiotemporal vegetation changes. More capable sensors that could resolve
ground features more accurately are spaceborne sensors offering very high spatial
resolutions such as GeoEye-1, WorldView-2/3, or Pleiades-1A with spatial resolutions
of 0.41, 0.46, and 0.5 m, respectively. Less commonly used for study because of the
high cost of acquisition is airborne hyperspectral imagery such as AVIRIS or CASI,
which provides very high spectral and spatial resolution data, but is consequently not
Role of Remote Sensing in Sustainable Grassland Management 105

practical for long-term studies/monitoring application. Recently, lightweight digital

cameras and hyperspectral sensors that can be mounted on unmanned aerial vehicles
(UAVs) have been utilized for grassland surveys (Laliberte and Rango 2011; Laliberte
et al. 2011; Rango et al. 2006; Von Bueren et al. 2015). UAVs as an emerging remote
sensing platform can provide imagery with spatial resolution higher than 0.1 m, which
is an unprecedented data source for species-level research. Such platforms are capable
of providing very high spatial resolution images at required temporal resolutions that
can be used for finer-scale grassland investigation. However, the current acquisition
cost and processing time for such high spatial resolution imagery may be only appro-
priate for hot spot monitoring, but untenable for long-term large-area monitoring.
Over the past decade, studies using remote sensing for grassland management have
been conducted all over the world, including North America (Listopad et al. 2015;
Mirik and Ansley 2012), South America (Bradley and Millington 2006; Di Bella
et al. 2011), Europe (Psomas et al. 2011; Redhead et al. 2012), Asia (Cui et al. 2012;
Leisher et al. 2012; Zhang et al. 2008), Africa (Olsen et al. 2015), and Australia
(Guerschman et al. 2009; Lawes and Wallace 2008). Remote sensing of spatiotem-
poral changes in vegetation attributes such as biochemical (e.g., chlorophyll pigment)
and biophysical (e.g., leaf area index [LAI]) properties, plant biomass, canopy cover,
and vegetation height can inform ecosystem status. Additionally, these attributes can
be applied to monitor the effects of wildfire disturbances, habitat loss, or climate
change on grasslands. In these studies, the most commonly used remote sensing
technique for ecosystem monitoring applications has been the use of empirical–
statistical models. These models involve establishing a relationship between in
situ biochemical or biophysical measurements with spectral vegetation indices
calculated using ground-level spectral reflectance measurements or optical remote
sensing imagery.
Intensive research using remote sensing within the past decade has been con-
ducted in an endangered mixed-grass prairie ecosystem to evaluate vegetation con-
ditions across space and time in relation to local environmental factors, climate
conditions, and disturbance events. The implications of the research for validating
remote sensing technologies and techniques for the mixed-grass prairies are provid-
ing valuable information and insight to land managers for guiding future manage-
ment practices; several bodies of this research are presented in the following section.

6.2 CASE STUDIES

Of the three North American prairie types—tallgrass, mixed-grass, and shortgrass,
the mixed-grass prairies have seen some of the worst decline, with remnants still
found in parts of southern Alberta and Saskatchewan, Canada. The Government of
Canada recognized the ecological importance of preserving an intact area of the
endangered mixed-grass prairie ecosystem in 1981 and the Grasslands National Park
(GNP, N 49°12′, W 107°24′) was soon established in 1988 (Figure 6.1). However,
portions of the GNP are fragmented into small parcels as a consequence of land
within and neighboring the park boundaries being privately held and used for agricul-
tural or grazing purposes. Nevertheless, Parks Canada, an agency of the Government of
Canada, has been tasked with operating and protecting the GNP in efforts to conserve
 106

00

2900
W E

28

270
0
00
S

27
260
0
2600

0
2800

290
27
00
2600
25
00
Fescue grassland
Moist mixed grassland
Mixed grassland 280
0
3000

Contour (100-ft intervals)

Frenchman river
West block 0 2.5 5 10 15
km

Badlands Valley area Upland area

FIGURE 6.1 Map delineating the extent of the mixed grass prairies in Canada, the GNP, and three typical landscape units including badlands, valley
Remote Sensing for Sustainability

area, and upland area in the GNP.

Role of Remote Sensing in Sustainable Grassland Management 107

and restore the rich diversity of species and highly specialized communities of plants
and animals in their native state that have evolved in response to a variety of stresses,
such as drought, grazing, and fire (Anderson 2006; Shorthouse and Larson 2010).
Here, we introduce a select number of studies demonstrating the application of
remote sensing technologies and techniques for the management of the mixed-grass
prairies at the GNP. From a remote sensing perspective, the GNP has offered a unique
challenge of assessing heterogeneous canopies composed of several dominant spe-
cies, often with dead material or exposed soil. We focus on the role of remote sensing
for addressing several key challenges currently affecting the ecological integrity of
the GNP and indicate the utility of these studies for sustainable management needs.
The importance of deriving vegetation attributes (e.g., LAI and chlorophyll content)
for evaluating grassland health is first discussed, followed by the introduction of
how remote sensing can be used for assessing the effect of wildfire disturbances and
climate conditions, and finally we outline how remote sensing–derived information
can aid in habitat mapping of an endangered grassland species.

6.2.1 Assessing grAsslAnd HeAltH Using remote

sensing–derived BiopHysicAl properties
Biophysical properties derived from remote sensing are direct indicators of grass-
land ecological status and can help land managers to assess the health of the eco-
system across space and aid in sustainable management. Many studies for the GNP
(e.g., Banerjee et al. 2011; He et al. 2006a,b, 2007a,b, 2009; Tong and He 2013) have
focused on using remote sensing–derived LAI to study vegetation spatial patterns of
the ecosystem. He et al. (2009) linked LAI to other vegetation biophysical properties
such as dead biomass, which solved the difficulty of mapping biophysical information
owing to insufficient sampling coverage for the GNP, and many of these biophysical
measures have been further incorporated in aid of spatial fire fuel modeling, habitat
modeling for species at risk, and biomass monitoring for a reintroduced herd of plains
bison (Figure 6.2). Fire and habitat modeling and biomass monitoring are all impor-
tant remote sensing applications to address some of the current ecological challenges
facing Parks Canada. For instance, fire modeling can assist in the implementation of
appropriate areas for prescribed burning, habitat modeling can aid in species reintro-
duction and protection of critical habitats for species at risk, while biomass monitor-
ing is essential for determining the effects of grazing regimes in the GNP.

6.2.2 mApping cHloropHyll content

As An indicAtor of grAsslAnd HeAltH

Leaf chlorophylls are inherently related to the photosynthetic capacity of plants and
thus provide a measure of productivity to guide sustainable management practices. A
study conducted by Tong (2014) used empirical–statistical models to estimate chlo-
rophylls at a range of spatial scales. In specific, the leaf-level relationships between
chlorophylls and corresponding spectral reflectance were scaled up to generate can-
opy and landscape-level chlorophyll maps through a scaling-up procedure proposed
by Wong and He (2013) using SPOT-5 and CASI-550 images (Figure 6.3). Such
 108

Dead biomass (ton/ha)

Leaf area index
0–0.5 N
0–0.5 N
0.5–1.5 W E
0.5–1.5 W E
1.5–2.0 S
1.5–2.5 S
2.0–3.5 10 km 2.5–3.5
3.5–5.6 10 km
3.5–6

FIGURE 6.2 LAI map with a map accuracy of 66.7% obtained from SPOT-4 image using the regression model between Adjusted Transformed Soil
Adjusted Vegetation Index (ATSAVI) and LAI. Dead biomass map with an accuracy of 51.2% obtained from LAI map using the regression model
between dead biomass and LAI. (Adapted from He, Y., X. Guo, and J. F. Wilmshurst. 2009. Reflectance measures of grassland biophysical structure.
International Journal of Remote Sensing 30 (10): 2509–2521. doi:10.1080/01431160802552751.)
Remote Sensing for Sustainability
Role of Remote Sensing in Sustainable Grassland Management 109

(a)

0 2.5 5 10
km

(b)
N

Chlorophyll content (mg/m2)

0.1–20 100–150

20–40 150–300

40–60 300–500

60–100 500–900

0 0.25 0.5 1 1.5

km

FIGURE 6.3 Chlorophyll maps derived from (a) SPOT-5 image acquired June 8, 2012, with
a map accuracy of 63.52%; (b) CASI-550 image acquired June 23, 2012, with a map accuracy
of 72.88%. (Adapted from Tong, A. 2014. Estimating Grassland Chlorophyll Content for a
Mixed Grassland: Exploring the Performance of the Empirical–Statistical and the Physical
Modelling Approach. MSc dissertation, University of Toronto.)
110 Remote Sensing for Sustainability

chlorophyll maps can provide land managers and owners the ability to accurately
assess the vigor of vegetation in areas affected by agricultural or grazing practices
and facilitate the implementation of strategies for sustainable resource management.
These maps may also be useful to detect areas that are water stressed and help assess
areas vulnerable to fire.

6.2.3 investigAting grAsslAnd distUrBAnce Using remote sensing

Disturbances on the grasslands (e.g., wildfire, grazing) are important factors driving
the evolution of grasslands. Parks Canada (2010) understood this early on and, as part
of the current GNP management framework, has reintroduced disturbance regimes
(e.g., prescribed burning and light to moderate grazing by plains bison) that were
considered natural processes before human interference and essential to ecological
restoration efforts. Remote sensing can be applied to investigate the occurrence of
these disturbances, evaluate their influence, and provide feedback and support for
present and future sustainable grassland management practices. Given that the GNP
is a protected area, the potential for uncontrolled disturbances is still a possibility as
demonstrated by an unexpected wildfire that occurred in April 2013. Lu et al. (2015)
investigated this wildfire disturbance by identifying different levels of burn severity
(Figure 6.4) and investigated the postfire grassland recovery process (Figure 6.5)
using Landsat images. The degree of burn severity was determined to be a function
of the amount of prefire dead biomass and the elevation of the landscape. The grass-
lands showed a high resilience and recovered quickly after the fire disturbance. This
study highlighted the importance of dead biomass estimation for fire risk manage-
ment and invasive species control for local ecosystem balance.

High severity
Moderate severity
Low severity
0 1 2 4 6
Very low severity km

FIGURE 6.4 Burned area in GNP and burn severity estimated using Landsat imagery.
(Adapted from Lu, B., Y. He, and A. Tong. 2015. Evaluation of spectral indices for estimat-
ing burn severity in semi-arid grasslands. International Journal of Wildland Fire 25 (2):
147–157.)
Role of Remote Sensing in Sustainable Grassland Management 111

(a) (b)

NDVI
1
0.8 N
0.6
0.4 0 4.5 9
0.2 km
(c) 0

FIGURE 6.5 Normalized Difference Vegetation Index (NDVI) map derived from Landsat
images showing the postfire vegetation recovery. (a) Six weeks after fire; (b) 10 weeks after
fire; (c) 3 months after fire. (Adapted from Lu, B., Y. He, and A. Tong. 2015. Evaluation of
spectral indices for estimating burn severity in semi-arid grasslands. International Journal
of Wildland Fire 25 (2): 147–157.)

6.2.4 determining tHe effects of climAtic fActors

on grAsslAnd HeAltH Using remote sensing

Environmental changes driven by climatic factors, primarily temperature and pre-

cipitation regimes, are important for determining ecosystem health. At the same
time, understanding the spatial and temporal variability of evapotranspiration (ET)
is also important for identifying soil and plant water stress that affects the health
of vegetation (Girolimetto and Venturini 2013; Ritchie 1998; Yang et al. 2012).
Remote sensing can provide important information on water availability and guide
human activities such as irrigation or other management practices to help alleviate
plant drought stress. For semiarid grasslands such as the GNP, vegetation greenness
varies dramatically across space over years, as indicated by AVHRR Normalized
Difference Vegetation Index (NDVI) values shown in Figure 6.6. For example, in
a typical dry year, 2001, AVHRR NDVI values were generally low with an NDVI
value less than 0.2 across the GNP, while in a wet year, 2002, AVHRR NDVI values
are higher than 0.35. He (2014) thus investigated the effect of precipitation on veg-
etation cover for the GNP using climate data and SPOT-4 and AVHRR images. The
amount and timing of precipitation were found to effectively control the ecosystem
dynamics (Figure 6.7). Years with decreased vegetation coverage were associated
with increased ET as a result of higher temperature, which stressed vegetation and
was exacerbated by water deficiency in the soil. Although He (2014) did not directly
estimate ET using remote sensing, studies have established its utility for a grass-
lands environment (e.g., Nosetto et al. 2005; Yang et al. 2012). Precipitation affected
upland grass communities the highest, followed by valley grass communities, while
the effect on riverside shrubs was not significant. The highest relationships occurred
between percent vegetation cover and precipitation from the previous 80-day period,
112 Remote Sensing for Sustainability

1988 1989 1990 1991 1992

1993 1994 1995 1996 1997

1998 1999 2000 2001 2002

2003 2004 2005 2006 2007

NDVI 0 15 30
0.00 0.10 0.15 0.20 0.25 0.30 0.35 1.00 River Holding boundary km

FIGURE 6.6 Vegetation greenness (i.e., NDVI derived from July AVHRR imagery) varies
from 1988 to 2007. (Adapted from He, Y. 2014. The effect of precipitation on vegetation cover
over three landscape units in a protected semi-arid grassland: Temporal dynamics and suitable
climatic index. Journal of Arid Environments 109: 74–82. doi:10.1016/j.jaridenv.2014.05.022.)

suggesting that the lag effect of precipitation and the precipitation accumulated from
the previous 80 days contributed the most to vegetation growth. This result could be
used to aid land managers in predicting plant stress and implement water manage-
ment strategies.

6.2.5 mApping HABitAt for endAngered species Using remote sensing

In recent decades, wildlife populations have been on a drastic decline owing to
human activities involving land conversion, pesticide use, or overhunting. In order
to effectively guide land management practices to conserve, protect, and prevent
further declines in wildlife populations, it has been necessary to investigate met-
rics that can quantify the biological activity of threatened and endangered species
(e.g., foraging or reproduction behaviour), the biophysical and biochemical fea-
tures of habitats, and the spatiotemporal changes of habitats as a result of human
intervention. Such data need to be acquired at a broad spatial scale, which cannot
Role of Remote Sensing in Sustainable Grassland Management 113

NDVI vs. accumulated precipitation

1.0
Upland grasslands
0.9
Valley grassland
0.8
Shrub communities
0.7
0.6
r

0.5
0.4
0.3
0.2
0.1
0.0
May 20–Jul. 10

May 10–Jul. 10

May 01–Jul. 10

Apr. 20–Jul. 10

Apr. 10–Jul. 10

Apr. 01–Jul. 10
Jun. 20–Jul. 10

Jun. 10–Jul. 10

Jun. 01–Jul. 10
Jul. 01–Jul. 07

Time duration

FIGURE 6.7 The correlation coefficients calculated between NDVI during the peak grow-
ing season (July 10) and accumulated precipitation from previous 10–100 days with a 10-day
interval over three vegetation communities—upland/sloped grass, valley grass, and riparian
shrub from 1988 to 2007. The dashed line is the significant value at the 0.05 level. (Adapted
from He, Y. 2014. The effect of precipitation on vegetation cover over three landscape units
in a protected semi-arid grassland: Temporal dynamics and suitable climatic index. Journal
of Arid Environments 109: 74–82. doi:10.1016/j.jaridenv.2014.05.022.)

be easily achieved using traditional field survey methods. Remote sensing provides
the capability to investigate the aforementioned metrics by quantifying the spa-
tiotemporal changes in land use, vegetation type and cover, ecological processes
and conditions, and wildlife corridors (Flaherty et al. 2014; Neumann et al. 2015;
Olsen et al. 2007).
The habitats for various species in the North American prairie have experienced
some of the worst decline in the past few decades (Klimek et al. 2007), and protected
areas such as the GNP have established themselves as a haven for endemic species
that have seen their habitats destroyed elsewhere, including the loggerhead shrike,
an open country bird that was once widely distributed throughout North America.
Remote sensing has also been applied in this area to support endangered species
protection. Using remote sensing imagery, Shen et al. (2013) applied SPOT-4 imag-
ery to map loggerhead shrike nesting sites in the GNP, which offers the opportunity
to monitor the spatial distribution of the species for future conservation strategies
(Figure 6.8). Findings indicated that nests in highly elevated open areas away from
roads with scattered shrubs, particularly thorny species, were important for active
shrike nesting sites. The study indicated that future park management goals should
focus on preserving native grass and thorny shrub species within the shrike’s breed-
ing range.
114 Remote Sensing for Sustainability

N
W E
S

0 2.5 5 10
km

FIGURE 6.8 The suitable shrike habitat mapping based on a logistic model for loggerhead
shrike in the West Block of GNP in 2006. (Adapted from Shen, L., Y. He, and X. Guo. 2013.
Exploration of loggerhead shrike habitats in Grassland National Park of Canada based on
in situ measurements and satellite-derived Adjusted Transformed Soil-Adjusted Vegetation
Index (ATSAVI). Remote Sensing 5 (1): 432–453. doi:10.3390/rs5010432.)

6.3 CONCLUSIONS, CHALLENGES, AND OPPORTUNITIES

6.3.1 conclUsion
In the research capacity, remote sensing techniques supporting grassland manage-
ment have been validated for many grassland biomes, such as the mixed-grass prai-
ries at the GNP. By quantifying vegetation biophysical and biochemical properties
and establishing empirical relationships at the local scale, upscaling procedures can
extrapolate these relationships to produce vegetation maps for the entire study area.
Traditional field-based methods have only been able to measure conditions at the
local scale, whereas remote sensing has provided the ability to generate cost-effective
vegetation maps that are a time-saving solution to accurately assess the entire extent
of an area; subsequently, the application of remote sensing has not only provided
vegetation health information but also detected and assessed grassland disturbances
such as impacts of grazing or wildfires, grassland response to climatic changes, and
mapped endangered animal habitats, to name a few. Results and information from
the presented case studies have helped understand the impacts of current manage-
ment decisions and can further facilitate future sustainable management policies and
plans.
Role of Remote Sensing in Sustainable Grassland Management 115

6.3.2 cHAllenges And opportUnities

6.3.2.1 Image Acquisition and Processing
Remote sensing provides an economical and capable means for detecting, moni-
toring, and assessing ground features, but challenges still remain. For spaceborne
sensors that operate on a predefined orbit and trajectory, the revisit time is fixed. If
inclement weather (e.g., heavy cloud cover) is present over the area when the satel-
lite sensor acquires an image, the image data may be unusable. Additionally, it is
challenging to obtain spaceborne imagery with both high spatial and temporal reso-
lution. For airborne mounted sensors, the time of image acquisition has to be con-
sidered, as well as the high costs of long-term data acquisition. UAV systems have
increasingly become an alternative platform for image acquisition owing to their
low cost of operation, high flexibility, ability to be deployed quickly and repeatedly,
and capability to acquire imagery at very high spatial resolutions (sub-decimeter)
(Laliberte and Rango 2011). Since UAV systems are normally operated at a range of
low altitudes (50–300 m), it is possible to acquire images of different spatial resolu-
tions, and weather conditions have minimal interference on image acquisition. With
the quick response time of grasslands, particularly the mixed-grass prairie ecosys-
tems at the GNP, UAVs offer a high temporal capacity to monitor and quantify the
change in ecosystem response to disturbances that otherwise could not be detected
with traditional spaceborne and airborne platforms.
Techniques used to analyze image data require specialized training and human
error may be introduced. Imagery requires calibration and validated by field data,
which are often limited to a few sites around accessible areas, and is time-consuming
and therefore expensive to collect. This is not a weakness per se, as local level mea-
surements can be extrapolated and scaled across the image. We point this out here to
emphasize the need for collaboration and sharing of resources between agencies, land
managers, and other stakeholders to reduce costs (Marsett et al. 2006). In all, careful
planning and training in applying remote sensing can help alleviate these limitations.

6.3.2.2 Species-Level Monitoring

Understanding changes in species composition is crucial for conserving biodiversity,
managing invasive species, and conducting sustainable grazing practices. For most
applications, at least for homogeneous canopies, medium spatial resolution Landsat
imagery has the capability to map vegetation at a range of scales (Xie et al. 2008).
Conversely, heterogeneous grassland canopies, such as those in the GNP, feature sev-
eral dominant species in addition to fractional dead material or soil components that
compound the ability to accurately map and assess vegetation from the species level.
Yet, even with the advent of very high spatial resolution imagery, the mixed pixel
problem (i.e., a pixel that contains more than one spectral signature belonging to
several ground features) remains a challenge for retrieving ground information from
heterogeneous canopies. Radiative transfer modeling may be an avenue to accurately
estimate vegetation parameters, but its inherent complexity makes them impractical
for land managers.
116 Remote Sensing for Sustainability

6.3.2.3 Vegetation Structure Mapping

Optical remote sensing has been applied extensively for investigating grassland bio-
physical and biochemical properties. However, extracting canopy structure infor-
mation (vegetation density profiles) from optical remote sensing images remains a
challenge since incoming radiation is primarily intercepted and reflected at the top
of the canopy surface. On the other hand, airborne Light Detection and Ranging
(LiDAR) systems are active sensors that emit and receive light pulses that can pen-
etrate and reach beneath a canopy and therefore estimate canopy height, LAI, or
aboveground biomass effectively. LiDAR systems provide an all-day capability
and may be operated in slightly inclement weather conditions for vegetation stud-
ies. LiDAR has been widely applied for investigating vegetation structure in forest
and agricultural areas (Drake et al. 2002; Hawbaker et al. 2009; Houldcroft et al.
2005) yet has rarely been deployed to study grasslands, especially for investigating
vegetation structure. Given the relatively low to sparse vegetation density and can-
opy heights in grassland environments, the challenge of capturing the subtle struc-
ture of grassland canopies remains to be addressed (Bork and Su 2007; Hellesen
and Matikainen 2013; Su 2004). To this end, with decreasing acquisition costs, the
opportunity to explore the potential of LiDAR for grassland management is fast
becoming an attractive option to researchers and land managers. Nevertheless, UAV
multispectral images of the same area but from different directions may also be
used to produce vegetation structure information in grasslands, using Structure from
Motion techniques (Smith and Vericat 2015). Specifically, the optical point clouds
of vegetation can be produced from multiple fly-overs of the same sites from many
different directions.

6.3.2.4 Snow, Topography, and Soil Moisture Mapping

Spaceborne Interferometric Synthetic Aperture Radar (InSAR) sensors are increas-
ingly being explored to resolve ground features. They are active sensors that emit
and receive long wavelengths in the microwave portion of the spectrum that are not
susceptible to atmospheric scattering, thereby allowing an all-day and all-weather
imaging capability. Canada has been a pioneer in the field of InSAR, with the deploy-
ment of RADARSAT-1 and RADARSAT-2. For grassland research, RADARSAT-2
shows potential for snow mapping, ground surface topography mapping, and soil
moisture detection. To date, RADARSAT-2 has been validated in grassland studies
for soil moisture retrieval (e.g., Bertoldi et al. 2014; Xing et al. 2014) and image clas-
sification (e.g., Buckley and Smith 2011) with success. However, its application and
implementation at an operational level for land managers have yet to be explored.

6.3.2.5 Logistics
Remote sensing may not be well understood by stakeholders involved with manage-
ment policies and plans, and is thus imperative that researchers are able to engage
and demonstrate the utility and application of remote sensing. The extent and focus
of research for grassland management using remote sensing has been meager rela-
tive to studies that have been conducted for areas with greater perceived economic
value (e.g., precision agriculture and forests). To fully appreciate the current trend
Role of Remote Sensing in Sustainable Grassland Management 117

of remote sensing research, we used Scopus, a comprehensive bibliographic data-

base, and our survey revealed only 488 results inclusive from 1979 to 2015 that
were relevant to grassland remote sensing and management. In comparison, forest
remote sensing and management returned 3399 results and, similarly, agriculture
remote sensing and management returned 3126 results inclusive from 1976 to 2015.
Only by actively promoting awareness of the importance of grasslands for research
can we better understand the ecological status of the grassland environment. Given
the lesser focus of grasslands research within the remote sensing community, it is
likely that less emphasis has been placed on educational initiatives and opportuni-
ties for stakeholders involved with grassland management policies and plans and
is thus imperative that researchers are able to actively engage and demonstrate the
utility and application of remote sensing. Through collaborative efforts between both
parties, sharing of knowledge, needs, and resources along the way of the research
process should be able to lead to the production of spatial products for use at an
operational level. Future work may consider integrating remote sensing data with
information from other disciplines such as ecology, geography, sociology, or eco-
nomics to better understand the complexities of human–environmental interactions.

ACKNOWLEDGMENT
The support of the Natural Sciences and Engineering Research Council of Canada
Discovery Grants to Dr. Yuhong He and Dr. Xulin Guo is gratefully acknowledged.

REFERENCES
Anderson, R. C. 2006. Evolution and origin of the central grassland of North America:
Climate, fire, and mammalian grazers. Journal of the Torrey Botanical Society 133(4):
626–647. doi:10.3159/1095-5674(2006)133[626:EAOOTC]2.0.CO;2.
Banerjee, S., Y. He, X. Guo, and B. C. Si. 2011. Spatial relationships between leaf area index
and topographic factors in a semiarid grassland: Joint multifractal analysis. Australian
Journal of Crop Science 5(6): 756–763.
Bertoldi, G., S. Della Chiesa, C. Notarnicola, L. Pasolli, G. Niedrist, and U. Tappeiner.
2014. Estimation of soil moisture patterns in mountain grasslands by means of SAR
RADARSAT2 images and hydrological modeling. Journal of Hydrology 516: 245–257.
doi:10.1016/j.jhydrol.2014.02.018.
Bizikova, L. 2009. Challenges and lessons learned from integrated landscape management
(ILM) projects. Winnipeg: International Institute for Sustainable Development.
Bork, E. W. and J. G. Su. 2007. Integrating LIDAR data and multispectral imagery for
enhanced classification of rangeland vegetation: A meta analysis. Remote Sensing of
Environment 111(1): 11–24. doi:10.1016/j.rse.2007.03.011.
Boval, M. and R. M. Dixon. 2012. The importance of grasslands for animal production and
other functions: A review on management and methodological progress in the tropics.
Animal 6(5): 748–762. doi:10.1017/S1751731112000304.
Bradley, A. V. and A. C. Millington. 2006. Spatial and temporal scale issues in determining
biomass burning regimes in Bolivia and Peru. International Journal of Remote Sensing
27(11): 2221–2253. doi:10.1080/01431160500396550.
Buckley, J. R. and A. M. Smith. 2011. Comparing RADARSAT 2 and TerraSAR-X Quad-Pol
SAR imagery of grasslands. doi:10.1109/IGARSS.2011.6050022.
118 Remote Sensing for Sustainability

Cui, X., Z. G. Guo, T. G. Liang, Y. Y. Shen, X. Y. Liu, and Y. Liu. 2012. Classification man-
agement for grassland using MODIS data: A case study in the Gannan Region, China.
International Journal of Remote Sensing 33(10): 3156–3175. doi:10.1080/01431161.2
011.634861.
Di Bella, C. M., M. A. Fischer, and E. G. Jobbágy. 2011. Fire patterns in north-eastern
Argentina: Influences of climate and land use/cover. International Journal of Remote
Sensing 32(17): 4961–4971. doi:10.1080/01431161.2010.494167.
Drake, J. B., R. O. Dubayah, D. B. Clark, R. G. Knox, J. B. Blair, M. A. Hofton, R. L. Chazdon,
J. F. Weishampel, and S. Prince. 2002. Estimation of tropical forest structural charac-
teristics, using large-footprint Lidar. Remote Sensing of Environment 79(2–3): 305–319.
doi:10.1016/S0034-4257(01)00281-4.
Estrada-Carmona, N., A. K. Hart, F. A. J. DeClerck, C. A. Harvey, and J. C. Milder. 2014.
Integrated landscape management for agriculture, rural livelihoods, and ecosystem
conservation: An assessment of experience from Latin America and the Caribbean.
Landscape and Urban Planning 129: 1–11. doi:10.1016/j.landurbplan.2014.05.001.
Flaherty, S., P. W. W. Lurz, and G. Patenaude. 2014. Use of LiDAR in the conservation
management of the endangered red squirrel (Sciurusi vulgaris L.). Journal of Applied
Remote Sensing 8(1). doi:10.1117/1.JRS.8.083592.
Girolimetto, D. and V. Venturini. 2013. Water stress estimation from NDVI-Ts plot and the wet
environment evapotranspiration. Advances in Remote Sensing 2: 283–291. http://dx.doi
.org/10.4236/ars.2013.24031.
Guerschman, J. P., M. J. Hill, L. J. Renzullo, D. J. Barrett, A. S. Marks, and E. J. Botha. 2009.
Estimating fractional cover of photosynthetic vegetation, non-photosynthetic vegetation
and bare soil in the Australian tropical savanna region upscaling the EO-1 Hyperion
and MODIS sensors. Remote Sensing of Environment 113(5): 928–945. doi:10.1016/j
.rse.2009.01.006.
Hawbaker, T. J., N. S. Keuler, A. A. Lesak, T. Gobakken, K. Contrucci, and V. C. Radeloff.
2009. Improved estimates of forest vegetation structure and biomass with a LiDAR-
optimized sampling design. Journal of Geophysical Research: Biogeosciences 114(3).
doi:10.1029/2008JG000870.
He, Y. 2014. The effect of precipitation on vegetation cover over three landscape units in a
protected semi-arid grassland: Temporal dynamics and suitable climatic index. Journal
of Arid Environments 109: 74–82. doi:10.1016/j.jaridenv.2014.05.022.
He, Y., X. Guo, and B. C. Si. 2007a. Detecting grassland spatial variation by a wavelet
approach. International Journal of Remote Sensing 28(7): 1527–1545. doi:10.1080
/01431160600794621.
He, Y., X. Guo, and J. Wilmshurst. 2006a. Studying mixed grassland ecosystems I: Suitable
hyperspectral vegetation indices. Canadian Journal of Remote Sensing 32(2): 98–107.
He, Y., X. Guo, J. Wilmshurst, and B. C. Si. 2006b. Studying mixed grassland ecosystems II:
Optimum pixel size. Canadian Journal of Remote Sensing 32(2): 108–115.
He, Y., X. Guo, and J. F. Wilmshurst. 2007b. Comparison of different methods for measuring
leaf area index in a mixed grassland. Canadian Journal of Plant Science 87(4): 803–813.
He, Y., X. Guo, and J. F. Wilmshurst. 2009. Reflectance measures of grassland bio-
physical structure. International Journal of Remote Sensing 30(10): 2509–2521.
doi:10.1080/01431160802552751.
Hellesen, T. and L. Matikainen. 2013. An object-based approach for mapping shrub and tree
cover on grassland habitats by use of liDAR and CIR orthoimages. Remote Sensing 5(2):
558–583. doi:10.3390/rs5020558.
Houldcroft, C. J., C. L. Campbell, I. J. Davenport, R. J. Gurney, and N. Holden. 2005.
Measurement of canopy geometry characteristics using LiDAR laser altimetry: A feasi-
bility study. IEEE Transactions on Geoscience and Remote Sensing 43(10): 2270–2282.
doi:10.1109/TGRS.2005.856639.
Role of Remote Sensing in Sustainable Grassland Management 119

Jun, H. 2006. Effects of integrated ecosystem management on land degradation control and
poverty reduction. In Environment, Water Resources and Agricultural Policies: Chapter
3, 63–72. Danvers: Organization for Economic Co-operation and Development.
Klimek, S., A. Richter gen. Kemmermann, M. Hofmann, and J. Isselstein. 2007. Plant spe-
cies richness and composition in managed grasslands: The relative importance of field
management and environmental factors. Biological Conservation 134(4): 559–570.
doi:10.1016/j.biocon.2006.09.007.
Laliberte, A. and A. Rango. 2011. Image processing and classification procedures for analy-
sis of sub-decimeter imagery acquired with an unmanned aircraft over arid rangelands.
GIScience and Remote Sensing 48(1): 4–23. doi:10.2747/1548-1603.48.1.4.
Laliberte, A. S., M. A. Goforth, C. M. Steele, and A. Rango. 2011. Multispectral remote sens-
ing from unmanned aircraft: Image processing workflows and applications for range-
land environments. Remote Sensing 3(11): 2529–2551. doi:10.3390/rs3112529.
Lawes, R. A. and J. F. Wallace. 2008. Monitoring an invasive perennial at the landscape
scale with remote sensing. Ecological Management and Restoration 9(1): 53–59.
doi:10.1111/j.1442-8903.2008.00387.x.
Leisher, C., S. Hess, T. M. Boucher, P. van Beukering, and M. Sanjayan. 2012. Measuring the
impacts of community-based grasslands management in Mongolia’s Gobi.” PLoS ONE
7(2). doi:10.1371/journal.pone.0030991.
Listopad, C. M. C. S., R. E. Masters, J. Drake, J. Weishampel, and C. Branquinho. 2015.
Structural diversity indices based on airborne LiDAR as ecological indicators for man-
aging highly dynamic landscapes. Ecological Indicators 57: 268–279. doi:10.1016/j
.ecolind.2015.04.017.
Lu, B., Y. He, and A. Tong. 2015. Evaluation of spectral indices for estimating burn severity in
semi-arid grasslands. International Journal of Wildland Fire 25(2): 147–157.
Marsett, R. C., J. Qi, P. Heilman, S. H. Biedenbender, M. C. Watson, S. Amer, M. Weltz,
D. Goodrich, and R. Marsett. 2006. Remote sensing for grassland management
in the arid southwest. Rangeland Ecology and Management 59(5): 530–540.
doi:10.2111/05-201R.1.
Mirik, M. and R. J. Ansley. 2012. Comparison of ground-measured and image-classified mes-
quite (Prosopis glandulosa) canopy cover. Rangeland Ecology and Management 65(1):
85–95. doi:10.2111/REM-D-11-00073.1.
Neumann, W., S. Martinuzzi, A. B. Estes, A. M. Pidgeon, D. Dettki, G. Ericsson, and V. C.
Radeloff. 2015. Opportunities for the application of advanced remotely-sensed data in
ecological studies of terrestrial animal movement. Movement Ecology 3(8). doi:10.1186
/s40462-015-0036-7.
Nosetto, M. D., E. G. Jobbágy, and J. M. Paruelo. 2005. Land-use change and water losses:
The case of grassland afforestation across a soil textural gradient in Central Argentina.
Global Change Biology 11(7): 1101–1117. doi:10.1111/j.1365-2486.2005.00975.x.
Olsen, J. L., S. Miehe, P. Ceccato, and R. Fensholt. 2015. Does EO NDVI seasonal metrics
capture variations in species composition and biomass due to grazing in semi-arid grass-
land savannas? Biogeosciences 12(14): 4407–4419. doi:10.5194/bg-12-4407-2015.
Olsen, L. M., V. H. Dale, and T. Foster. 2007. Landscape patterns as indicators of ecologi-
cal change at Fort Benning, Georgia. Landscape and Urban Planning 79(2): 137–149.
doi:10.1016/j.landurbplan.2006.02.007.
Parks Canada. 2010. Grasslands National Park of Canada Management Plan. http://www
.pc.gc.ca/eng/pn-np/sk/grasslands/plan/~/media/pn-np/sk/grasslands/pdf/prairies-grass
lands-plan-062010_e.ashx.
Psomas, A., M. Kneubühler, S. Huber, K. Itten, and N. E. Zimmermann. 2011. Hyperspectral
remote sensing for estimating aboveground biomass and for exploring species richness
patterns of grassland habitats. International Journal of Remote Sensing 32(24): 9007–
9031. doi:10.1080/01431161.2010.532172.
120 Remote Sensing for Sustainability

Rango, A., A. Laliberte, C. Steele, J. E. Herrick, B. Bestelmeyer, T. Schmugge, A. Roanhorse,

and V. Jenkins. 2006. Using unmanned aerial vehicles for rangelands: Current
applications and future potentials. Environmental Practice 8(3): 159–168. doi:10.1017
/S1466046606060224.
Redhead, J., M. Cuevas-Gonzales, G. Smith, F. Gerard, and R. Pywell. 2012. Assessing the
effectiveness of scrub management at the landscape scale using rapid field assess-
ment and remote sensing. Journal of Environmental Management 97(1): 102–108.
doi:10.1016/j.jenvman.2011.12.005.
Ritchie, J. T. 1998. Soil water balance and plant water stress. In Systems Approaches for
Sustainable Agricultural Development: Understanding Options for Agricultural
Production, ed. G. Y. Tsuji, G. Hoogenboom, and P. K. Thornton, 41–54. Dordrecht:
Kluwer Academic Publishers.
Samson, F. and E. L. Knopf. 1994. Prairie conservation in North America. Bioscience 44(6):
418–421. doi: 10.2307/1312365.
Shen, L., Y. He, and X. Guo. 2013. Exploration of loggerhead shrike habitats in Grassland
National Park of Canada based on in situ measurements and satellite-derived Adjusted
Transformed Soil-Adjusted Vegetation Index (ATSAVI). Remote Sensing 5(1): 432–453.
doi:10.3390/rs5010432.
Shorthouse, J. D. and D. J. Larson. 2010. Grasslands and grasslands arthropods of Canada. In
Arthropods of Canadian Grasslands (Volume 1): Ecology and Interactions in Grassland
Habitats, ed. J. D. Shorthouse and K. D. Float, 1–24. Ottawa: Biological Survey of Canada.
Smith, M. W. and D. Vericat. 2015. From experimental plots to experimental landscapes:
Topography, erosion and deposition in sub-humid badlands from structure-from-
motion photogrammetry. Earth Surface Processes and Landforms 40(12): 1656–1671.
doi:10.1002/esp.3747.
Su, Q. 2004. DEM modelling, vegetation characterization and mapping of Aspen Parkland
Rangeland using LiDAR data. PhD dissertation, University of Alberta.
Tong, A. 2014. Estimating grassland chlorophyll content for a mixed grassland: Exploring the
performance of the empirical–statistical and the physical modelling approach. MSc dis-
sertation, University of Toronto.
Tong, A. and Y. He. 2013. Comparative analysis of SPOT, Landsat, MODIS, and AVHRR
Normalized Difference Vegetation Index data on the estimation of Leaf Area Index in
a mixed grassland ecosystem. Journal of Applied Remote Sensing 7(1). doi:10.1117/1
.JRS.7.073599.
Wong, K. and Y. He. 2013. Estimating grassland chlorophyll content using remote sensing
data at leaf, canopy, and landscape scales. Canadian Journal of Remote Sensing 39(2):
155–166. doi:10.5589/m13-021.
Von Bueren, S. K., A. Burkart, A. Hueni, U. Rascher, M. P. Tuohy, and I. J. Yule. 2015.
Deploying four optical UAV-based sensors over grassland: Challenges and limitations.
Biogeosciences 12(1): 163–175. doi:10.5194/bg-12-163-2015.
Xie, Y., Z. Sha, and M. Yu. 2008. Remote sensing imagery in vegetation mapping: A review.
Journal of Plant Ecology 1(1): 9–23. doi:10.1093/jpe/rtm005.
Xing, M., B. He, X. Li, and X. Quan. 2014. Soil moisture retrieval using RADARSAT-2 and
HJ-1 CCD data in grassland. doi:10.1109/IGARSS.2014.6947164.
Yang, Y., S. Shang, and L. Jiang. 2012. Remote sensing temporal and spatial patterns of
evapotranspiration and the responses to water management in a large irrigation district
of North China. Agricultural and Forest Meteorology 164: 112–122. doi:10.1016/j
.agrformet.2012.05.011.
Zhang, Y., Z. Chen, B. Zhu, X. Luo, Y. Guan, S. Guo, and Y. Nie. 2008. Land desertifica-
tion monitoring and assessment in Yulin of Northwest China using remote sensing and
geographic information systems (GIS). Environmental Monitoring and Assessment
147(1–3): 327–337. doi:10.1007/s10661-007-0124-2.
 7 Classifying Tree
Species Using Fine
Spatial Resolution
Imagery to Support
the Conservation of
an Endangered Bird
Species in Hawaii
Qi Chen

CONTENTS
7.1 Introduction .................................................................................................. 122
7.2 Study Area and Data ..................................................................................... 124
7.2.1 Study Area ........................................................................................ 124
7.2.2 Remotely Sensed Images and Field Data ......................................... 125
7.3 Methods ........................................................................................................ 125
7.3.1 Discriminant Analysis ...................................................................... 125
7.3.2 SVM.................................................................................................. 127
7.3.3 Random Forest .................................................................................. 127
7.3.4 Classification and Accuracy Assessment Strategy ........................... 127
7.4 Results and Discussion ................................................................................. 129
7.4.1 Classification with the Four Conventional Worldview-2 Bands ......... 129
7.4.2 Classification with the Inclusion of Four New Worldview-2 Bands ... 129
7.4.3 Comparison of Different Classifiers ................................................. 137
7.4.4 Pixel-Based versus Object-Based Classification............................... 139
7.5 Conclusions ................................................................................................... 139
Acknowledgments.................................................................................................. 140
References .............................................................................................................. 140

121
122 Remote Sensing for Sustainability

7.1 INTRODUCTION
Habitat loss and species extinction are two main challenges of conserving biodiver-
sity in the 21st century (Gottschalk et al. 2005) and serious threats to sustainabil-
ity (Vucetich et al. 2015). Hawaii is the “Endangered Species Capital of the United
States,” with just 0.2% of the US land area but approximately 25% of the US endan-
gered species (Eldredge and Evenhuis 2003). Because of the introduction of thousands
of new species and increased human activities, approximately 90% of the dry land
ecosystem in the state is now completely gone and at least 120 Hawaiian plant species
under conservation have less than 50 individual plants left. As a result, the avifauna has
been decimated and approximately 100 bird species has become extinct (USGS 2006).
Palila (Loxioides bailleui) is probably the most-known endangered bird species in
Hawaii—a status derived primarily from its ties to a 1979 landmark case redefining the
scope and intent of the federal Endangered Species Act (Riddle 2010). Palila feeds on
immature seeds of mamane trees (Sophora chrysophylla), an endemic dry-forest tree
species that occurs widely throughout the main Hawaiian Islands and ranges from near
shoreline to tree line (>3000 m elevation) (Banko 2002). Nevertheless, palila and its hab-
itat have experienced a multitude of threats including feral sheep and goats that browse
mamane trees (Scowcroft and Giffin 1983), feral cats that depredate nests and adult
birds (Hess et al. 2004), fungus such as Armillaria that kills mamane trees (Gardner and
Trujillo 2001), and alien parasitoid wasps competing with palila for caterpillars that are
fed to nestlings (Leonard et al. 2008; Oboyski et al. 2004). Although palila used to be
widely distributed in lowlands of the island of Hawaii, these threats reduce its current
habitat to an extent that is less than 5% of its historical range. Currently, palila’s habitat
is limited to a relatively small area (less than 30 km2) on the western slope of Mauna
Kea, a high and dormant volcano (Hess et al. 1999; Johnson et al. 2006) (see Figure 7.1).
In recent years, significant efforts have been made to restore palila’s habitat and
population, for example, by reducing the number of ungulates and trapping rats
(Banko 2002; Banko and Farmer 2006). Despite these efforts, a steady decline in
palila population has happened, especially during the last 5 years (Leonard et al.
2008). The latest survey showed that there are only approximately 1200 birds left,
down from approximately 4400 in 2003 (DLNR 2010). What is particularly perplex-
ing is that no clear causes can be identified for the population decline, making it diffi-
cult to take effective measures for palila conservation. Mamane and naio (Myoporum
sandwicense) are the two tree species in this habitat. However, mamane is more criti-
cal for palila’s survival since palila almost exclusively eats immature mamane seeds.
Therefore, it is important to separate mamane and naio trees so that the abundance
of mamane trees and their spatial distribution can be derived to understand the palila
population dynamics. Since the crowns of mamane and naio are relatively small
(with an average crown diameter of 4–5 m) and they can grow next to each other in
the habitat, it is very difficult, if not impossible, to classify these two species using
conventional medium spatial resolution satellite imagery such as Landsat.
During the last decade, commercial fine spatial resolution satellite imagery has
emerged as a powerful and cost-effective tool for detailed vegetation mapping (e.g.,
Adelabu and Dube 2015; Boggs 2010; Ji and Wang 2015; Lin et al. 2015; Morales
et al. 2012; Murray et al. 2010; Pouteau et al. 2011; Puissant et al. 2014; van Lier et
Classifying Tree Species Using Fine Spatial Resolution Imagery 123

Focal study area boundary

Naio trees N
0 0.5 1 2
km Mamane trees

(b)

(a) (c)

Kauai
Maui
Oahu
Study area

Hawaii

FIGURE 7.1 The study site and sampled trees. (a and b) From Worldview-2 and (c) from
IKONOS. (b and c) Correspond to an area outlined by the yellow box in (a).

al. 2009). In October 2009, a new satellite called Worldview-2 from DigitalGlobe
Inc. was launched, which has up to 1.84 m spatial resolution in multispectral (MS)
bands and 0.46 m for the panchromatic band. One of the main differences between
Worldview-2 and other fine spatial resolution satellites such as IKONOS, QuickBird,
and Geoeye-1 is that it carries four new MS bands (called coastal, yellow, red-edge, and
near-infrared [NIR]2) in addition to the conventional four bands (blue, green, red,
and NIR) (Table 7.1). Worldview imagery has recently been used for mapping
124 Remote Sensing for Sustainability

TABLE 7.1
Characteristics of Worldview-2 and IKONOS MS Bands
IKONOS Worldview-2
Spectral ranges at half-maximum
Blue 445–516 nm 450–510 nm
Green 506–595 nm 510–580 nm
Red 632–698 nm 630–690 nm
NIR 757–853 nm 770–895 nm
Coastal N/A 400–450 nm
Yellow N/A 585–625 nm
Red-edge N/A 705–745 nm
NIR-2 N/A 860–1040 nm
Spatial resolutiona
At-nadir 3.2 m 1.8 m

Note: N/A, not available.

a Images were provided with a spatial resolution of 4 m for IKONOS and 2 m for Worldview-2.

detailed vegetation information such as species (e.g., Cho et al. 2015; Heumann 2011;
Li et al. 2015; Robinson et al. 2016). However, relatively few studies have compared
it with the first-generation commercial satellites such as IKONOS and investigate
the potential of its fine spatial resolution and additional spectral bands for mapping
vegetation at the species level (Pu and Landry 2012).
The main objective of this study is to compare the capability of Worldview-2
and IKONOS for classifying mamane and naio tree species in the palila habitat
of Hawaii. To make the comparison less limited to a particular algorithm, three
types of representative classification algorithms are tested: discriminant analysis
(DA), support vector machine (SVM), and random forest (RF). The article is
organized as follows: the study area and data will be introduced first, followed
by an explanation of the classification algorithms and analytical strategy, and
then results will be reported and discussed. The major research findings will be
summarized in the end.

7.2 STUDY AREA AND DATA

7.2.1 Study AreA
The study site is on the western slope of Mauna Kea (elevation, approximately 1900
to 2700 m), which lies above the trade-wind inversion with relatively light cloud cover
and rainfall (Juvik and Juvik 1998) (Figure 7.1). Rainfall averaged 35–75 cm and
temperature averaged 11.1°C ± 1.5°C annually (Giambelluca et al. 1986). Mamane
and naio are the two dominant tree species in the habitat. The majority of mamane
occurs at the upper elevation (above 2300 m). At lower elevations, mixed stands of
mamane and naio can be found, dominated by the latter (Hess et al. 1999; Scott et al.
1986). The trees are relatively short with a mean height of 4.7 ± 2.0 m. The average
Classifying Tree Species Using Fine Spatial Resolution Imagery 125

canopy cover is around 30%, resembling open-canopy woodland with an understory

grass layer. The focal study area is a strip of 2-km swath where airborne lidar data
were also acquired to characterize vegetation structure (Figure 7.1).

7.2.2 remotely SenSed ImAgeS And FIeld dAtA

Worldview-2 and IKONOS images were acquired at a local time of 10:54 a.m. on
December 26, 2009, and 11:03 a.m. on January 8, 2008, respectively. The MS images
of Worldview-2 and IKONOS were resampled to 2 and 4 m, respectively, and both
types of images were orthorectified and georeferenced with the UTM Zone 5N
(WGS84) coordinate system by the data providers.
To collect ground truth for classification, individual mamane and naio trees
were located in the field between March 14 and March 21, 2010. A Garmin
Venture HC GPS and a 4 ft–by–3 ft hard copy IKONOS false-color map over-
lain with grid lines of 500-m intervals were used to locate trees (Figure 7.2). A
Trimble GeoXT GPS with higher positioning accuracy was carried in the field
as well, but it was found that using the Garmin Venture HC GPS gave sufficient
information to accurately locate trees in the map owing to the open canopy of the
landscape (Riddle 2010). GIS polygon files of individual tree crowns were manu-
ally created later in the computer laboratory with the Worldview-2 image as the
base layer. As a result, a total of 44 mamane polygons and 40 naio polygons were
created within the focal study area, which correspond to 438 mamane pixels and
743 naio pixels in the Worldview-2 image and 112 mamane pixels and 183 naio
pixels in the IKONOS image.

7.3 METHODS
A wide range of classification algorithms exist in the literature, which are either
parametric or nonparametric. The parametric algorithms typically assume that each
class follows a Gaussian distribution while the nonparametric algorithms do not
necessarily make such assumption and thus are more flexible to handle classes of
different distributions. Among the algorithms tested in this study, DA is parametric
and SVM and RF are nonparametric. For completeness, the main ideas of these algo-
rithms are briefly summarized next while RF is introduced in slightly more details
since it is relatively new.

7.3.1 dIScrImInAnt AnAlySIS

DA is a classical classification algorithm that derives its discriminant function from
Bayes’ theorem and assumes that each class’s conditional probability is Gaussian
and their prior probabilities are equal. The discriminant function has many different
variations depending on how to estimate the variance–covariance matrix ∑ of each
class. The most generic discriminant function is to estimate ∑ for individual classes,
which is called quadratic discriminant analysis (QDA). When the number of features
is large and the sample size is small, estimating ∑ could be unstable. One way to
handle this problem is to assume that all classes have the same Gaussian distribution
126 Remote Sensing for Sustainability

(a)

(b)

(c)

FIGURE 7.2 Examples of mamane and naio trees in the field. (a) An isolated mamane tree,
(b) an isolated naio tree, and (c) naio (left) and mamane (right) trees are growing next to each
other. (Courtesy of Ryan Riddle.)

so that sample data can be pooled to estimate just one Gaussian distribution, which
is linear discriminant analysis (LDA). For DA, the decision boundary of classifica-
tion is constructed implicitly by comparing the discriminant functions of different
classes. For QDA, the decision boundary is usually curvilinear, while for LDA, the
decision boundary is linear.
Classifying Tree Species Using Fine Spatial Resolution Imagery 127

7.3.2 SVm
Different from DA, SVM explicitly constructs a linear decision boundary (called
hyperplane when the number of features is greater than three) to separate two classes
(Cortes and Vapnik 1995). It is possible that there exist multiple hyperplanes to sepa-
rate two classes for the training sample (Duda et al. 2000). SVM finds one unique
optimal hyperplane by maximizing the distance from a hyperplane to the nearest
point of either class (Hastie et al. 2009). Although SVM is a linear classifier, features
can be expanded to a higher feature space to fit a hyperplane, which could corre-
spond to a nonlinear decision boundary in the original lower feature space through
a technique called kernel trick (Abe 2005). The kernels tested in this study are qua-
dratic (or second-order polynomial) and Gaussian radial basis function.

7.3.3 rAndom ForeSt

RF is a recent addition to the nonparametric classifiers (Breiman 2001). It was devel-
oped from several techniques including classification and regression tree (CART)
and bootstrap aggregation (or bagging). The main idea of RF is to build a large
number of de-correlated trees and let them vote for the most popular class (Hastie et
al. 2009). When an individual tree is built, RF differs from regular CART in that
(1) the tree is based on the bootstrap (random sampling with replacement) instead of the
original sample (Breiman 1996) and (2) the splitting of a node is based on a random
subset instead of all features (Ho 1998). Such randomization leads to de-correlated
trees, which are particularly useful for reducing the variance of prediction. If large
individual trees are grown, they can also have low bias. The low bias and variance of
estimates make RF popular in many fields (Hastie et al. 2009).
Many benefits are associated with the bootstrapping used in RF. A bootstrap sample
usually has repetitive values for some observations and leaves approximately one-third
of the observations not included. These observations, called out-of-bag, are thus not used
for building the tree from the bootstrap sample. Approximately, an observation is out-
of-bag for one-third of the trees. By comparing each observation’s true class label with
the majority vote of class predictions from these out-of-bag trees, an overall out-of-bag
prediction error can be calculated, without further using skills such as cross-validation.
The out-of-bag prediction error is useful for tuning parameters such as the number of
trees, the number of features randomly selected for node splitting, and the leaf node size.
Out-of-bag observations can also be used to rank the importance of individual
features. After an individual tree is grown, it can be used to predict the classes of
out-of-bag observations and calculate the prediction accuracy. If the values for a
particular feature are randomly permuted among the out-of-bag observations, the
prediction accuracy will decrease. The amount of accuracy decrease averaged over
all trees can be used as a measure of the importance of that feature.

7.3.4 clASSIFIcAtIon And AccurAcy ASSeSSment StrAtegy

A total of five specific algorithms, including two DA algorithms (QDA and LDA),
SVMs with two different kernels (quadratic and Gaussian radial basis function),
128 Remote Sensing for Sustainability

and RF, were tested. It is hypothesized that compared to IKONOS, Worldview-2

will produce higher classification accuracy with its (1) finer spatial solution and
(2) more spectral bands. To test this hypothesis, three types of classification
were conducted: the first was to use all four IKONOS bands (blue, green, red, and
NIR), the second was to use the four Worldview-2 conventional bands (blue, green,
red, and NIR), and the third was to use all eight Worldview-2 bands including the
four new bands (coastal, yellow, red-edge, and NIR2). The comparison between
IKONOS and Worldview-2 using the same four conventional MS bands focuses
the evaluation on the benefits of Wordlview-2’s finer spatial resolution. The com-
parison of using Worldview-2 four versus eight bands evaluates the usefulness of
the new MS bands.
Although both images were orthorectified by the data providers, their geoloca-
tion accuracies were further verified by overlaying the images with a digital sur-
face model derived from airborne lidar data and visually checking distinct features
such as roads and isolated trees. It was found that the Worldview-2 image matches
well with airborne lidar data while IKONOS has some shifts (up to ~5 m in some
places). To handle this issue, image-to-image registration was conducted by col-
lecting a total of 158 ground control points well distributed over the study area
and warping the IKONOS image to match Worldview-2 using a second-order poly-
nomial transformation in ENVI 4.7 (ITT VIS Corp.). The root mean square error
of image registration is 0.4 m. The classification was conducted using the digital
numbers without being further calibrated to reflectance because each image was
classified separately and radiometric correction usually has minimal effects on
classification of a single-date image (Song et al. 2001). Mamane and naio trees are
the only two tree species in the study area, with other vegetation types being short
scrubs and grass. A canopy height model derived from lidar was used to mask out
all pixels that are less than 1 m so that the classification was focused on mamane
and naio trees only.
Tenfold cross-validation was used to evaluate the overall classification accuracy.
This means that the field data were broken into 10 folds, based on which model devel-
opment and test were performed for 10 iterations. Each time, one fold was used for
testing and the rest were used for model development. For RF, a stepwise approach
instead of an exhaustive search of all possible combinations of parameters was used
to determine the model parameters based on the out-of-bag prediction errors. The
stepwise approach was to calibrate one parameter each time while keeping others
fixed. To reduce the computation demand, the minimal leaf node was first deter-
mined, followed by the number of subset features for node splitting, and finally the
number of trees.
The focus of this study is to perform classification at the pixel level. Nevertheless,
the airborne discrete-return lidar data acquired in this site provide the possibility
of mapping individual tree crown objects (Chen et al. 2006). To evaluate such
potential, I first considered individual pixels within each manually delineated ref-
erence tree crown polygon as individual groups (i.e., objects) and then produced
the average spectral values of each tree pixel group, which was further used for
classification. After the object-level classification was done, all pixels within each
reference polygon were labeled as the same tree species and the classification
Classifying Tree Species Using Fine Spatial Resolution Imagery 129

accuracy was summarized at the pixel level to compare with results from the pixel-
level classification.

7.4 RESULTS AND DISCUSSION

7.4.1 clASSIFIcAtIon wIth the Four conVentIonAl worldVIew-2 BAndS
Table 7.2 lists the accuracy of individual classifiers and their means when the
four conventional MS bands (blue, green, red, and NIR) of either IKONOS or
Worldivew-2 are used. It was found that Worldview-2 has consistently higher accu-
racy than IKONOS. For example, at the pixel level, when the IKONOS image was
classified, the mean classification accuracy of different classifiers was 68.6%; when
the Worldview-2 image was classified, the mean accuracy increased up to 71.9%
(Figure 7.3). For individual classifiers, the accuracy increase varied from 1.3% for
QDA to 6.1% for RF.
At the object level, the accuracy increase using Worldview-2 over IKONOS was
even larger. When the IKONOS image was classified, the average classification
accuracy of different classifiers was 77.1%; however, when the Worldview-2 image
was classified, the average accuracy increased by 7.9% (reaching 85.0%) (Figure
7.3). For individual classifiers, the accuracy increase varied from 5.2% for LDA to
12.8% for SVM-GRB. The higher overall classification accuracy of Wordview-2
compared to IKONOS is supported by the spectral scatterplots shown in Figure 7.4b
and c: it is evident that the two tree species are more separable in the scatterplots of
Worldview-2 than IKONOS.
There are a couple of factors that might contribute to the accuracy difference
between Worldview-2 and IKONOS images. For example, there are slight differ-
ences between the spectral sensitivity curves of corresponding individual bands
of the two sensors. The sun–target–sensor geometry is a little different because
these two images were taken from two different dates. The atmospheric conditions
could have negligible effects since both images were taken in clear days. Fire is the
major natural hazard risk at this study site, but no fires have occurred over this area
for years. Therefore, little changes of vegetation are expected between the differ-
ent image acquisition dates to make a significant contribution to the difference in
the classification accuracy. Although it is difficult to quantify the contribution of
each factor to the accuracy difference, it is expected that the higher classification
accuracy of Worldview-2 is mainly attributed to its finer spatial resolution, which
can capture more details of vegetation especially along the edges of the individual
tree crowns.

7.4.2 clASSIFIcAtIon wIth the IncluSIon oF Four new worldVIew-2 BAndS

Table 7.2 also summarizes the classification accuracy when all eight bands of
Worldview-2 were used. It was found that the inclusion of the four new bands
increased the classification accuracy for most classifiers. At the pixel level, the
additional four new bands increased the mean classification accuracy from 71.9%
to 75.0%. For individual classifiers, the accuracy increase varied from 1.3% of
 130

TABLE 7.2
Tenfold Cross-Validation Overall Accuracy of Species Classification Using Different Algorithms
Pixel Based Object Based
IKONOS Worldview-2 Worldview-2 IKONOS Worldview-2 Worldview-2
Classifiera (4 Bands) (4 Bands) (8 Bands) (4 Bands) (4 Bands) (8 Bands) Mean
QDA 69.1% 70.4% 75.5% 77.6% 82.9% 84.4% 76.7%
LDA 68.9% 70.3% 74.5% 80.2% 85.4% 86.5% 77.6%
SVM-Quad 68.6% 71.5% 73.4% 78.0% 84.0% 84.5% 76.7%
SVM-GRB 68.2% 72.8% 74.1% 74.2% 87.0% 86.5% 77.1%
RF 68.2% 74.3% 77.4% 75.3% 85.9% 84.7% 77.6%
Mean 68.6% 71.9% 75.0% 77.1% 85.0% 85.3%

Note: The highest accuracy of each column is underlined. The higher accuracy between two specific algorithms (QDA vs. LDA; SVM-Quad vs. SVM-GRB) of the same
type of classifier is in boldface. The accuracy of RF is also in boldface to facilitate comparison.
a LDA, linear discriminant analysis; QDA, quadratic discriminant analysis; RF, random forest; SVM-GRB, support vector machine with Gaussian radial basis kernel; SVM-

Quad, support vector machine with quadratic kernel.

Remote Sensing for Sustainability
Classifying Tree Species Using Fine Spatial Resolution Imagery 131

90
IKONOS-2 (4 bands)
Worldview-2 (4 bands)
Worldview-2 (8 bands)
80

70

60
Overall accuracy (%)

50

40

30

20

10

0
Pixel-based Object-based

FIGURE 7.3 Comparison of mean classification accuracy (averaged over classifiers) when
(1) different sensors and bands are used (four IKONOS bands, four Worldview-2 conventional
MS bands, or all eight Worldview-2 bands) and (2) the classification is performed at the pixel
level or at the object level.

SVM-GRB to 5.1% of LDA. At the object level, the mean classification accuracy
increases by a small margin from 85.0% to 85.3%. However, when the individual
classifiers were examined, only three of them (QDA, LDA, and SVM-Quad) had
an increase in classification accuracy, and the other two (SVM-GRB and RF) had a
minor decrease in classification accuracy. It is not exactly clear why SVM-GRB and
RF had slightly worse accuracy. However, these two methods are more complex than
the others and thus they may run into the problem of overfitting.
Usually, a classifier with more parameters can model more flexible decision
boundaries, but they also need more training data to constrain the classifier. The
needed training data also increase with the number of input features (Richard and
Jia 2006). SVM-GRB and RF can model more flexible decision boundaries than
the other three classifiers (see Figure 7.5c and d), which implies that they essen-
tially need more training data to constrain the construction of decision boundary.
However, when the classification is conducted at the object level, the sample size is
relatively small and the inclusion of the four additional bands makes the situation
even worse, which can explain, to some extent, why SVM-GRB and RF had slightly
lower accuracy when the four new bands were added. In other words, the lower
132 Remote Sensing for Sustainability

IKONOS-2 (pixel level) IKONOS-2 (pixel level) IKONOS-2 (pixel level)

260
250 500
240
220 200 400
Green

200

Red

NIR
300
180 150
160 200
100
140
100
120
160 180 200 160 180 200 100 150 200 250
(a) Blue Blue Red

IKONOS-2 (object level) IKONOS-2 (object level) IKONOS-2 (object level)

180 450
220
210 160 400
200 350
Green

190 140

NIR
Red

180 300
170 120
250
160
100
150 200
160 170 180 190 200 160 170 180 190 200 100 120 140 160 180
(b) Blue Blue Red
Worldview-2 (object level) Worldview-2 (object level) Worldview-2 (object level)
400 160 400
380 150
140 350
360
130
Green

340 300
NIR
Red

120
320
110 250
300
100
280 200
90
260 280 300 260 280 300 100 120 140 160
(c) Blue Blue Red

FIGURE 7.4 Scatterplots of (a) object-based mean pixel values for IKONOS bands,
(b) object-based mean pixel values for Worldview-2 bands, and (c) individual pixel values for
Worldview-2 bands. + represents mamane and * represents naio.

accuracy of these two classifiers is more related to the algorithm itself instead of the
new bands.
The out-of-bag observations in RF were also used to rank the importance of indi-
vidual bands of Worldview-2 for classification. It is interesting to see that the top four
bands for tree species classification at the pixel level were coastal, yellow, red-edge,
and NIR-2, which are exactly the four new bands (Figure 7.6a). At the object level,
the importance of individual bands for RF classification was different. However, the
coastal band was still among the top four bands of the highest importance (Figure
7.6b). The analysis above indicates that the four new bands are valuable in improving
the classification of tree species at the study site.
 1: LDA 2: QDA 5: Random forest
450 450 450

400 400 400

350 350 350

NIR

NIR
NIR
300 300 300

250 250 250

200 200 200

100 120 140 160 180 100 120 140 160 180 100 120 140 160 180
Red Red Red

3: SVM (Quadratic kernel) 4: SVM (GRB kernel)

450 450 0.8

400 400
0.6
350 350
0.4

NIR
NIR
300 300
Overall accuracy

250 250 0.2

Classifying Tree Species Using Fine Spatial Resolution Imagery

200 200
0
100 120 140 160 180 100 120 140 160 180 1 2 3 4 5
Red Red Classifier
(a)

FIGURE 7.5 Decision boundaries and overall accuracy of different classifiers for IKONOS and Worldview-2 images: (a) object-level classification with
IKONOS bands. (Continued)
133
 134

1: LDA 2: QDA 5: Random forest

400 400 400

350 350 350

NIR
NIR
NIR
300 300 300

250 250 250

200 200 200

100 120 140 160 100 120 140 160 100 120 140 160
Red Red Red

3: SVM (Quadratic kernel) 4: SVM (GRB kernel)

400 400 1

0.8
350 350

0.6
300 300

NIR
NIR
0.4
250 250
Overall accuracy

0.2
200 200
0
100 120 140 160 100 120 140 160 1 2 3 4 5
Red Red Classifier
(b)

FIGURE 7.5 (CONTINUED) Decision boundaries and overall accuracy of different classifiers for IKONOS and Worldview-2 images: (b) object-level
classification with Worldview-2 bands. (Continued)
Remote Sensing for Sustainability
 1: LDA 2: QDA 5: Random forest
500 500 500

400 400 400

300 300 300

NIR
NIR
NIR
200 200 200

100 100 100

100 150 200 250 100 150 200 250 100 150 200 250
Red Red Red

3: SVM (Quadratic kernel) 4: SVM (GRB kernel)

0.8
500 500

400 400 0.6

300 300 0.4

NIR
NIR
Overall accuracy

200 200 0.2

Classifying Tree Species Using Fine Spatial Resolution Imagery

100 100
0
100 150 200 250 100 150 200 250 1 2 3 4 5
Red Red Classifier
(c)

FIGURE 7.5 (CONTINUED) Decision boundaries and overall accuracy of different classifiers for IKONOS and Worldview-2 images: (c) pixel-level
classification with IKONOS bands. (Continued)
135
 1: LDA 2: QDA 5: Random forest
136

500 500 500

400 400 400

NIR
NIR
300 300 300

NIR
200 200 200

100 100 100

100 150 200 100 150 200 100 150 200
Red Red Red

3: SVM (Quadratic kernel) 4: SVM (GRB kernel)

0.8
500 500

0.6
400 400

0.4

NIR
NIR
300 300
Overall accuracy

200 200 0.2

100 100
0
100 150 200 100 150 200 1 2 3 4 5
Red Red Classifier
(d)

FIGURE 7.5 (CONTINUED) Decision boundaries and overall accuracy of different classifiers for IKONOS and Worldview-2 images: (d) pixel-level
Remote Sensing for Sustainability

classification with Worldview-2 bands. + represents mamane and * represents naio.

Classifying Tree Species Using Fine Spatial Resolution Imagery 137

2.5
Out-of-bag feature importance

1.5

0.5

0
Coastal Blue Green Yellow Red Red-edge NIR NIR2
(a) Worldview-2 band

0.5
0.45
0.4
Out-of-bag feature importance

0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
Coastal Blue Green Yellow Red Red-edge NIR NIR2
(b) Worldview-2 band

FIGURE 7.6 Feature importance ranked by random forest for (a) pixel-level classification
and (b) object-level classification.

7.4.3 compArISon oF dIFFerent clASSIFIerS

When the two DA methods were compared, it was found that QDA has slightly higher
accuracy than LDA at the pixel level (Table 7.2). The accuracy increase varied from
0.1% for four Worldview-2 conventional bands to 1.0% for all eight Worldview-2
bands. However, LDA was clearly better than QDA at the object level. The accuracy
increase varied from 2.1% for eight Worldview-2 bands to 2.6% for four IKONOS
bands. This again is likely related to the fewer observations for training at the object-
level classification. LDA pools the two classes to estimate the variance–covariance
138 Remote Sensing for Sustainability

matrix while QDA estimates the variance–covariance matrix for each class sepa-
rately. Therefore, LDA has a higher ratio of training sample size to the number of
model parameters, which is advantageous in model estimation when the number of
observations for model training is small.
Another interesting result in this study is that, on average, LDA (the simplest clas-
sifier) has the same highest (77.6%) classification accuracy as RF (the most flexible
classifier) (Table 7.2). The good performance of LDA is not surprising if we inspect
the scatterplots of the two tree species (Figure 7.4). Although there is considerable
amount of overlay between two classes for some bands, each class seems to have one
unique mean and Gaussian-like distribution. Despite the fact that the more complex
algorithms (such as SVM and RF) can more easily model classes of multiple-mode
or non-Gaussian distributions, LDA and QDA could achieve high classification accu-
racy as long as the assumption of Gaussian distribution for each class is satisfied
(Hastie et al. 2009). Although it seems that mamane trees are more spread out than
naio trees for some bands, the degree of spread is not dramatically different from
each other, which justifies the relatively high classification accuracy of LDA com-
pared to QDA, especially at the object level.
It is worthwhile to note that even though the mean classification accuracy is
very similar among different classifiers (Table 7.2), each classifier could generate
dramatically different decision boundaries (Figure 7.5). The decision boundary
for LDA is linear owing to the common variance–covariance matrix between two
classes. QDA usually has a curvilinear decision boundary because each class has
different variance–covariance matrices. Both SVM-Quad and SVM-GRB gener-
ate smooth nonlinear boundaries, but it seems that SVM-GRB has more flexibility
in generating different shapes of decision boundaries. Among these classifiers,
RF is most versatile in terms of generating decision boundaries varying from a
straight line (e.g., Figure 7.5b) to very complex and localized curves (see Figure
7.5c and d). If we assume that each class follows a Gaussian-like distribution, it
seems that, at the pixel level, RF has overfitting problems caused by the significant
overlay between two classes in the spectral space (Figure 7.5c and d). According
to Occam’s razor (which states that the simplest model explaining the data is pre-
ferred) (Duda et al. 2000), LDA is a good choice for the tree species classification
at this study site.
There are divergent results in the literature regarding the relative performance of
parametric and nonparametric classifiers. For example, Camps-Valls and Bruzzone
(2005) compared LDA with several kernel-based nonparametric classifiers including
SVM, regularized radial basis function neural networks, kernel Fisher discriminant
analysis, and regularized AdaBoost for classifying the nine classes in the AVIRIS
NW Indiana’s Indian Pines 1992 data set. It was found that LDA had significantly
lower accuracy than the nonparametric classifiers. However, a recent study by
Szuster et al. (2011) showed that Maximum Likelihood Classifier (equivalent to QDA
in this study) had similar performance with neural network and SVM when ASTER
images were used for classifying seven land cover and land use types in the coastal
zone of Koh Tao island, Thailand. These contrasting results agree with the No Free
Lunch Theorem: there is no context- or problem-independent reason to favor one
classifier over another (Duda et al. 2000). The apparent superiority of one algorithm
Classifying Tree Species Using Fine Spatial Resolution Imagery 139

over another in a particular study usually depends on the distribution of data and the
match between an algorithm and the problem it addresses.

7.4.4 pIxel-BASed VerSuS oBject-BASed clASSIFIcAtIon

It was found that the object-based classification can significantly increase the classi-
fication accuracy compared to pixel-based classification. On average, the increase in
the overall classification accuracy was 8.5%, 13.1%, and 10.3% when the four IKONOS
bands, four conventional Worldview-2 bands, and all eight Worldview-2 bands were
used, respectively. Note that the accuracy increase is larger for Worldview-2 than for
IKONOS, probably because the finer spatial resolution of Worldview-2 has a better
characterization of pixels at the crown edge.
Although object-based image analysis (OBIA) has become popular over the past
decade owing to the increasing use of fine spatial resolution remotely sensed data
(Blaschke 2010), only a few studies have compared the classification accuracy of
object-based classification with pixel-based classification (e.g., Cleve et al. 2008;
Gao 2008). For example, Cleve et al. (2008) found that the object-based classification
approach provided a 17.97% higher overall accuracy than the pixel-based approach
for mapping the wildland–urban interface with aerial photos. It is encouraging that
the object-level classification at this study also produces positive results. One of the
main reasons why the object-based approach outperforms the pixel-based approach
is the reduction of spectral heterogeneity (e.g., caused by shade) of tree canopy using
the object-based approach. More research is needed in the future to conduct object-
level classification with individual tree objects automatically delineated using fine
spatial resolution imagery (e.g., Chopping 2011) or airborne lidar data (e.g., Chen et
al. 2006).

7.5 CONCLUSIONS
This study compared Worldview-2 and IKONOS for identifying mamane and naio
trees in the palila habitat of the island of Hawaii using both parametric and nonpara-
metric classifiers. Classification was conducted at the pixel and object levels. The
major findings are as follows:

• On average, the four conventional Worldview-2 bands (blue, green, red, and
NIR) can achieve 3.3% and 7.9% higher overall accuracy than the same four
IKONOS bands at the pixel level and object level, respectively. If individual
classifiers are considered, the highest accuracy improvement is 6.1% at the
pixel level and 12.8% at the object level.
• The inclusion of four new Worldview-2 bands (coastal, yellow, red-edge,
and NIR2) enhanced the average overall accuracy by 3.1% and 0.3% at the
pixel level and object level, respectively. The analysis of feature importance
using RF indicated that the four new bands were among the top four bands
(out of the eight Worldview-2 bands) for classifying tree species at the pixel
level. Coastal band was among the top four bands for classification at the
object level.
140 Remote Sensing for Sustainability

• The five classifiers (QDA, LDA, SVM-Quad, SVM-GRB, and RF) had
similar average overall accuracies varying from 76.7% (QDA and SVM-
Quad) to 77.6% (LDA and RF). However, each classifier could generate
dramatically different decision boundaries. Among these classifiers, LDA
has the simplest linear decision boundary while achieving the highest clas-
sification accuracy. This indicates that LDA is a good choice for tree species
at this study site.
• Classification at the object level, compared to that at the pixel level, can
significantly increase the overall accuracy by 8.5%, 13.1%, and 10.3% when
the four IKONOS bands, four conventional Worldview-2 bands, and all
eight Worldview-2 bands were used, respectively. In a nutshell, it was found
that both the finer spatial resolution and the additional spectral bands of
Worldview-2 can improve the classification of tree species at such a critical
habitat.

Conserving endangered species and maintaining nature’s intrinsic values for

future generations are an important obligation of human beings and a key aspect
of sustainability. However, in recent years, species conservation has experienced
a rapid decline in the dialogue of global sustainability, which has shifted its
focus to issues such as climate change. One driver of this concerning trend is
the insufficient funding and high cost of monitoring species and their habitat
in the field. The technology innovation of high-resolution yet low-cost satellite
imagery can greatly increase the efficiency in habitat characterization and thus
can help avert this alarming trend, especially in remote areas and developing
countries where the accessibility of high-quality remotely sensed data is very
limited. The encouraging findings of this research will further motivate the use
of state-of-the-art remote sensing technology to address pressing sustainability
issues in our society.

ACKNOWLEDGMENTS
The field data of this study were collected by Ryan Riddle. The Worldview-2 and
IKONOS images were provided by DigitalGlobe Inc. and the GeoEye Foundation,
respectively. Thanks are also extended to Greg Asner and Paul Banko for their dis-
cussion of the palila project.

REFERENCES
Abe, S., 2005. Support Vector Machines for Pattern Classification. London: Springer.
Adelabu, S., Dube, T., 2015. Employing ground and satellite-based QuickBird data and ran-
dom forest to discriminate five tree species in a Southern African Woodland. Geocarto
International 30(4), 457–471.
Banko, P., 2002. Restoring palila on the big island. Environmental Review 9(6), 1–8.
Banko, P., Farmer, C., 2006. Palila restoration: Lessons from long-term research. USGS Fact
Sheet FS-2006-3104.
Blaschke, T., 2010. Object based image analysis for remote sensing. ISPRS Journal of
Photogrammetry and Remote Sensing 65(1), 2–16.
Classifying Tree Species Using Fine Spatial Resolution Imagery 141

Boggs, G. S., 2010. Assessment of SPOT 5 and QuickBird remotely sensed imagery for map-
ping tree cover in savannas. International Journal of Applied Earth Observation and
Geoinformation 12(4), 217–224.
Breiman, L., 1996. Bagging predictors. Machine Learning 26(2), 123–140.
Breiman, L., 2001. Random forests. Machine Learning 45(1), 5–32.
Camps-Valls, G., Bruzzone, L., 2005. Kernel-based methods for hyperspectral image clas-
sification. IEEE Transactions on Geoscience and Remote Sensing 43(6), 1352–1362.
Chen, Q., Baldocchi, D. D., Gong, P., Kelly, M., 2006. Isolating individual trees in a savanna
woodland using small footprint LIDAR data. Photogrammetric Engineering and Remote
Sensing 72(8), 923–932.
Cho, M. A., Malahlela, O., Ramoelo, A., 2015. Assessing the utility WorldView-2 imagery for
tree species mapping in South African subtropical humid forest and the conservation
implications: Dukuduku forest patch as case study. International Journal of Applied
Earth Observation and Geoinformation 38, 349–357.
Chopping, M., 2011. CANAPI: Canopy analysis with panchromatic imagery. Remote Sensing
Letters 2(1), 21–29.
Cleve, C., Kelly, M., Kearns, F. R., Moritz, M., 2008. Classification of the wildland–urban
interface: A comparison of pixel- and object-based classifications using high-resolution
aerial photography. Computers Environment and Urban Systems 32(4), 317–326.
Cortes, C., Vapnik, V., 1995. Support-vector networks. Machine Learning 20(3), 273–297.
DLNR, 2010. Hawaii’s endangered palila bird numbers dropping. http://www.staradvertiser
.com/news/breaking/102911719.html (accessed on October 1, 2010).
Duda, R., Hart, P., Stork, D., 2000. Pattern Classification (2nd ed.). New York: John Wiley
& Sons.
Eldredge, L. G., Evenhuis, N. L., 2003. Hawaii’s biodiversity: A detailed assessment of the
numbers of species in the Hawaiian Islands. Bishop Museum Occasional Papers 76,
1–28.
Gao, J., 2008. Mapping of land degradation from ASTER data: A comparison of object-based
and pixel-based methods. GIScience & Remote Sensing 45(2), 149–166.
Gardner, D. E., Trujillo, E. E., 2001. Association of Armillaria mellea with mämane decline at
Pu’u La’au. Newsletter of the Hawaiian Botanical Society 40, 33–34.
Giambelluca, T. W., Nullet, M. A., Schroeder, T. A., 1986. Hawaii Rainfall Atlas, Report R76,
Hawai’i Division of Water and Land Development, Department of Land and Natural
Resources, Honolulu.
Gottschalk, T., Huettmann, F., Ehlers, M., 2005. Thirty years of analysing and modelling avian
habitat relationships using satellite imagery data: A review. International Journal of
Remote Sensing 26(12), 2631–2656.
Hastie, T., Tibshirani, R., Friedman, J., 2009. The Elements of Statistical Learning: Data
Mining, Inference, and Prediction (2nd ed.). New York: Springer.
Hess, S. C., Banko, P. C., Brenner, G. J., Jacobi, J. D., 1999. Factors related to the recovery of
subalpine woodland on Mauna Kea, Hawai’i. Biotropica 31(2), 212–219.
Hess, S. C., Banko, P. C., Goltz, D. M., Danner, R. M., Brinck, K. W., 2004. Strategies for
reducing feral cat threats to endangered Hawaiian birds. In Timms, R. M., Gorensel,
W. P. (Eds.), Proceedings of the 21st Vertebrate Pest Conference, University of California,
Davis, CA.
Heumann, B. W., 2011. An object-based classification of mangroves using a hybrid decision
tree—Support vector machine approach. Remote Sensing 3(11), 2440–2460.
Ho, T. K., 1998. The random subspace method for constructing decision forests. IEEE
Transactions on Pattern Analysis and Machine Intelligence 20(8), 832–844.
Ji, W., Wang, L., 2015. Discriminating Saltcedar (Tamarix ramosissima) from sparsely dis-
tributed Cottonwood (Populus euphratica) using a summer season satellite image.
Photogrammetric Engineering & Remote Sensing 81(10), 795–806.
142 Remote Sensing for Sustainability

Johnson, L., Camp, R. J., Brinck, K. W., Banko, P. C., 2006. Long-term population monitor-
ing: Lessons learned from an endangered passerine in Hawai’i. Wildlife Society Bulletin
34(4), 1055–1063.
Juvik, J., Juvik, S., 1998. Atlas of Hawaii (3rd ed.). Honolulu: University of Hawai’i Press.
Leonard Jr., D. L., Banko, P. C., Brinck, K. W., Farmer, C., Camp, R. J., 2008. Recent surveys
indicate rapid decline of palila population. Elepaio 68(4), 27–30.
Li, D., Ke, Y., Gong, H., Li, X., 2015. Object-based urban tree species classification using bi-
temporal WorldView-2 and WorldView-3 images. Remote Sensing 7(12), 16917–16937.
Lin, C., Popescu, S.C., Thomson, G., Tsogt, K., Chang, C.I., 2015. Classification of tree spe-
cies in overstorey canopy of subtropical forest using QuickBird images. PLoS One
10(5), e0125554.
Morales, R. M., Idol, T., Chen, Q., 2012. Differentiation of Acacia koa forest stands across
an elevation gradient in Hawai’i using fine-resolution remotely sensed imagery.
International Journal of Remote Sensing 33(11), 3492–3511.
Murray, H., Lucieer, A., William, R., 2010. Texture-based classification of sub-Antarctic vege-
tation communities on Heard Island. International Journal of Applied Earth Observation
and Geoinformation 12(3), 138–149.
Oboyski, P. T., Slotterback, J. W., Banko, P. C., 2004. Differential parasitism of seed-eating
Cydia (Lepidoptera: Tortricidae) by native and alien wasp species relative to eleva-
tion in subalpine Sophora (Fabaceae) forests on Mauna Kea, Hawaii. Journal of Insect
Conservation 8, 229–240.
Pouteau, R., Meyer, J. Y., Stoll, B., 2011. A SVM-based model for predicting distribution
of the invasive tree Miconia calvescens in tropical rainforests. Ecological Modelling
222(15), 2631–2641.
Pu, R., Landry, S., 2012. A comparative analysis of high spatial resolution IKONOS and
WorldView-2 imagery for mapping urban tree species. Remote Sensing of Environment
124, 516–533.
Puissant, A., Rougier, S., Stumpf, A., 2014. Object-oriented mapping of urban trees using
Random Forest classifiers. International Journal of Applied Earth Observation and
Geoinformation 26, 235–245.
Richards, J., Jia, X., 2006. Remote Sensing Digital Image Analysis: An Introduction (4th ed.).
Berlin, Germany: Springer-Verlag.
Riddle, R., 2010. Remote sensing techniques for classifying the habitat of an endangered bird
species on Mauna Kea. MA thesis, Department of Geography, University of Hawaii at
Manoa.
Robinson, T. P., Wardell-Johnson, G. W., Pracilio, G., Brown, C., Corner, R., van Klinken,
R. D., 2016. Testing the discrimination and detection limits of WorldView-2 imagery on a
challenging invasive plant target. International Journal of Applied Earth Observation
and Geoinformation 44, 23–30.
Scott, J. M., Mountainspring, S., Ramsey, F. L., Kepler, C. B., 1986. Forest bird communities
of the Hawaiian Islands: Their dynamics, ecology, and conservation. Studies in Avian
Biology 9, 1–431.
Scowcroft, P. G., Giffin, J. G., 1983. Feral herbivores suppress māmane and other browse spe-
cies on Mauna Kea, Hawai’i. Journal of Range Management 36(5), 638–645.
Song, C., Woodcock, C. E., Seto, K. C., Pax-Lenney, M., Macomber, S. A., 2001. Classification
and change detection using Landsat TM data: When and how to correct atmospheric
effects? Remote Sensing of Environment 75, 230–244.
Szuster, B., Chen, Q., Borger, M., 2011. A comparison of classification techniques to sup-
port land cover and land use analysis in tropical coastal zones. Applied Geography 31,
525–532.
Classifying Tree Species Using Fine Spatial Resolution Imagery 143

USGS, 2006. Hawaii forest bird interagency database project: Collecting, understanding, and
sharing population data on Hawaiian forest birds. USGS Fact Sheet. http://biology.usgs
.gov/pierc/Native_Birds.html (accessed on October 1, 2008).
van Lier, O. R., Fournier, R. A., Bradley, R. L., Thiffault, N., 2009. A multi-resolution satel-
lite imagery approach for large area mapping of ericaceous shrubs in Northern Quebec,
Canada. International Journal of Applied Earth Observation and Geoinformation 11(5),
334–343.
Vucetich, J. A., Bruskotter, J. T., Nelson, M. P., 2015. Evaluating whether nature’s intrinsic
value is an axiom of or anathema to conservation. Conservation Biology 29(2), 321–332.
 8 Remote Sensing of
Forest Damage by
Diseases and Insects
Gang Chen and Ross K. Meentemeyer

CONTENTS
8.1 Introduction ..................................................................................................145
8.2 Trends of Remotely Detecting Forest Disease and Insect Outbreaks ..........147
8.3 Remote Sensing Characteristics and Qualifications .....................................149
8.3.1 Spectral Characteristics ....................................................................150
8.3.2 Spatial Characteristics ......................................................................151
8.3.3 Temporal Characteristics ..................................................................151
8.4 A Review of Remote Sensing Algorithms ....................................................152
8.4.1 Thresholding .....................................................................................152
8.4.2 Classification.....................................................................................153
8.4.3 Statistical Regression ........................................................................154
8.4.4 Change Detection .............................................................................155
8.4.5 Additional Approaches .....................................................................156
8.5 Challenges and Opportunities ......................................................................156
8.5.1 Early Warning of Forest Damage .....................................................156
8.5.2 Consistent Monitoring of Long-Term, Historical Forest Damage ....157
8.5.3 Differentiating among Compound Disturbances..............................157
8.6 Conclusion ....................................................................................................157
Acknowledgment ...................................................................................................158
References ..............................................................................................................158

8.1 INTRODUCTION
Forests are an integral part of natural ecosystems, providing numerous ecological,
economic, social, and cultural services (Boyd et al. 2013; Chen et al. 2015a). For
example, they store approximately 45% of terrestrial carbon (C) and remain as a
large net C sink by capturing one-quarter of the anthropogenic carbon dioxide (CO2)
each year (Bonan 2008; Pan et al. 2011). However, environmental change (e.g., severe
drought) and global trade have increased forest vulnerability to a range of natural
disturbances, including diseases and insects (Asner 2013; Boyd et al. 2013; Wang
et al. 2008; Wingfield et al. 2015). Forest diseases are caused by pathogens that are
infectious and transmissible, such as bacteria, fungi, viruses, and helminths. Insects
attack different parts of the tree, with defoliators feeding on leaves or needles, and

145
146 Remote Sensing for Sustainability

bark/wood borers boring into the bark/wood. While some pathogen and insect spe-
cies are native to local ecosystems, many of the recent disturbances arise from the
nonindigenous species that may pose more pernicious and unpredictable threats to
forest health (Boyd et al. 2013). Over the past few decades, the frequency and inten-
sity of disease- and insect-caused forest disturbances have dramatically increased,
leading to extensive tree mortality in key forest biomes worldwide. Examples include
the sudden oak death epidemic in western United States, outbreaks of mountain
pine beetle in Canada’s boreal forest, bronze bug damage in plantation forests in
South Africa, and the spread of bark beetles in central Europe and Scandinavia
(Fassnacht et al. 2014; Meentemeyer et al. 2015; Oumar and Mutanga 2014; Wulder
et al. 2009). Figure 8.1 illustrates two typical symptoms of forest damage attributed
to the outbreaks of mountain pine beetles and the infectious disease sudden oak
death, respectively.
Sustainable forest management is essential to mitigating the destructive impacts
of diseases or insects on forest ecosystems. This is especially true when major dis-
turbance events have the potential to reduce the dominant native species, causing a
permanent change in forest structure. One prerequisite for effective management is to
understand the spatial distribution and severity of forest damage. Consequently, miti-
gation efforts can be performed to limit the population and the spread of pathogens or
insects on infected or susceptible host trees. Although conventional field mensuration
remains the most accurate way to quantify stages of infestation, it becomes time-con-
suming and costly when pathogen or insect populations reach epidemic levels. Remote
sensing provides a timely and accurate approach to scale up field measurements and
characterize spatially explicit information about the Earth’s surface at landscape to
regional scales. Recent developments in spaceborne and airborne sensors have further

Mountain pine beetle Plant pathogen

Dendroctonus ponderosae P. ramorum

(a) (b)

FIGURE 8.1 Landscape-scale forest mortality caused by (a) mountain pine beetle and
(b) sudden oak death. The infected trees show distinct symptoms of (a) red needles and (b) brown-
to-gray leaf lesions, respectively. (Courtesy of (a) Ministry of Forests, Lands and Natural Resource
Operations, http://www2.gov.bc.ca/gov/content/governments/organizational-structure/ministries
-organizations/ministries/forest-and-natural-resource-operations, and (b) California Oak
Mortality Task Force.)
Remote Sensing of Forest Damage by Diseases and Insects 147

advanced our ability to collect Earth observation data across multiple spatial, tempo-
ral, and spectral scales, making remote sensing feasible to monitor forest disturbances
(e.g., variations in forest biophysical and biochemical parameters) in response to the
disease and insect outbreaks of varying stages of invasion. Such rapid and accurate
delineation of large-area forest damage allows decision makers to take prompt and
informed actions, supporting the sustainable management of forests.
The main objective of this chapter is to provide a brief survey of remote sens-
ing assessment of forest damage by diseases and insects. Emphasis is directly laid
on mapping forest disturbances with satellite and airborne Earth observation data.
The following sections are organized to (i) summarize the recent trends of applying
remote sensing to detect forest disease and insect outbreaks, (ii) investigate remote
sensing characteristics and its qualifications for studying the topic, (iii) provide a
brief review of remote sensing algorithms, and (iv) discuss several remaining chal-
lenges that face researchers and decision makers in sustainable forest management.

8.2 TRENDS OF REMOTELY DETECTING FOREST

DISEASE AND INSECT OUTBREAKS
While the idea of applying remote sensing to detect disease- and insect-induced for-
est damage was considered as early as the 1970s and 1980s (e.g., Heller and Bega
1973; Nelson 1983; Rock et al. 1986), only recently (since the late 1990s) did the
topic receive considerable attention for managing emerging outbreak (Table 8.1).
Two reasons possibly explain slow adoption. First, a growing number of studies
showed that the frequency and intensity of forest disease and insect attacks signifi-
cantly increased over the past two decades as a result of climate change and glo-
balization (see a brief review by Boyd et al. [2013]). There was a growing need to
understand the mechanisms (e.g., spatial patterns) of the landscape-scale disease and
insect progression informing effective mitigation strategies. Second, the collected
Earth observation data have increased immensely during the same time period. The
large volumes of data sets with relatively cheap acquisition costs, for example, the
opening of more than four decades of Landsat archive (Woodcock et al. 2008), made
it easier to systematically analyze the impact of certain diseases or insects in specific
areas of interest. Ironically, one of the recent challenges facing many researchers is
how to better handle such big data.
Geographically, research hotspots were primarily located in North America (e.g.,
Canada and the United States) and Europe (e.g., Germany, Norway, Spain, Sweden,
and the United Kingdom), with several other studies conducted in Australia, China,
and South Africa. Please note that the case studies cited in Table 8.1 were collected
by searching Elsevier’s ScienceDirect database with the following formula: remote
sensing AND forest AND (disease OR insect). We also respectively substituted
pathogen for disease, and pest for insect in the search. The results were further
refined by removing the studies that did not contain a significant remote sensing
component or did not target specific disease or insect types.
Compared to forest diseases, insects appeared to be more intensively studied
using remote sensing (Table 8.1). This reflects the fact of high tree mortality induced
by insects as well as their globally widespread occurrence. For example, among the
148 Remote Sensing for Sustainability

TABLE 8.1
Types of Diseases and Insects, and the Corresponding Regions, Countries,
and Case Studies
Region Type of Disease or Insect Country Case Study
Europe Autumnal moth (Epirrita Sweden Babst et al. 2010
autumnata)
Bark beetle Ips grandicollis Germany Fassnacht et al. 2014
Bark beetle Ips typographus L. Germany Kautz 2014
Beech leaf-miner weevil Spain Rullán-Silva et al. 2015
(Rhynchaenus fagi)
Fungal spore Ganoderma sp. United Sadyś et al. 2014
Kingdom
Insect Physokermes inopinatus Sweden Olsson et al. 2012
Pine processionary moth Spain Sangüesa-Barreda et al. 2014
(Thaumetopoea pityocampa)
Pine sawfly (Neodiprion sertifer Norway Solberg et al. 2006
(Geoffrey))
North Black-headed budworm (Acleris Canada Luther et al. 1997
America gloverana (Walsingham))
Blister rust fungus (Cronartium United States Hatala et al. 2010
ribicola)
Eastern hemlock looper Canada Fraser and Latifovic 2005
(Lambdina fiscellaria)
Eastern spruce budworm United States Wolter et al. 2009
(Choristoneura fumiferana)
Emerald ash borer (Agrilus United States Pontius et al. 2008
planipennis Fairmaire)
Gypsy moth (Lymantria dispar L.) United States de Beurs and Townsend 2008;
Townsend et al. 2012;
Thayn 2013
Hemlock woolly adelgid (Adelges United States Siderhurst et al. 2010
tsugae Annand)
Jack pine budworm Canada, Leckie et al. 2005; Radeloff et al.
(Choristoneura pinus pinus United States 1999
(Free.))
Mountain pine beetle Canada, Assal et al. 2014; Bright et al.
(Dendroctonus ponderosae United States 2012; Cheng et al. 2010; Coops
Hopkins) et al. 2009; Goodwin et al. 2008;
Hatala et al. 2010; Meddens
et al. 2011; Meigs et al. 2011,
2015; Raffa et al. 2013; Skakun
et al. 2002; Walter and Platt
2013; Wulder et al. 2008, 2009
Spruce budworm (Choristoneura Canada, Wolter et al. 2008
fumiferana) United States
(Continued)
Remote Sensing of Forest Damage by Diseases and Insects 149

TABLE 8.1 (CONTINUED)

Types of Diseases and Insects, and the Corresponding Regions, Countries,
and Case Studies
Region Type of Disease or Insect Country Case Study
Sudden oak death (Phytophthora United States Kelly and Meentemeyer 2002;
ramorum) Lamsal et al. 2011; Liu et al.
2006, 2007; Meentemeyer et al.
2008; Pu et al. 2008
Western spruce budworm United States Meigs et al. 2011, 2015
(Choristoneura freemani)
Others Aphid (Essigella californica) Australia Goodwin et al. 2005
Bark beetle (Ips grandicollis) Australia Verbesselt et al. 2009
Fungal pathogen (Sphaeropsis Australia Goodwin et al. 2005
sapinea)
Insect (Thaumastocoris South Africa Oumar and Mutanga 2014;
peregrinus) Oumar et al. 2013
Mopane worm (Gonimbrasia South Africa Adelabu et al. 2014
belina)
Pine caterpillar (Dendrolimus China Huang et al. 2010
superans Butler, Dendrolimus:
Lasiocampidae, Lepidoptera)

United States’ 20 major diseases and insects that caused 6.4 million acres of
tree mortality in 2011, 60% were insects; mountain pine beetle (Dendroctonus
ponderosae Hopkins) alone killed 3.8 million acres of trees (USDA Forest Service
2012). Several other insects, such as bark beetle Ips grandicollis, gypsy moth
(Lymantria dispar L.), and jack pine budworm (Choristoneura pinus pinus [Free.]),
have also been well studied across forest biomes (Table 8.1). In contrast, remote
detection of the disease impacts on forest ecosystems was less studied. One excep-
tion is sudden oak death caused by the invasive plant pathogen Phytophthora ramo-
rum (Rizzo et al. 2005), which received considerable attention as a result of rapid
transmission and widespread mortality of oak and tanoak trees in coastal forests of
California and Oregon (Table 8.1).

8.3 REMOTE SENSING CHARACTERISTICS AND QUALIFICATIONS

The premise of utilizing remote sensing to detect disease- or insect-infested forests is
that the damaged trees show distinct symptoms capable of being observed remotely.
Depending on the type or stage of damage, the symptoms may indicate the decline in
chlorophyll/water quantity in foliage, leaf discoloration, defoliation, or treefall gaps.
For effective monitoring, Earth observation data acquired from satellite or airborne
sensors are expected to capture the differences in the reflected radiation from dam-
aged versus healthy trees. In this section, we base our discussion on the previous
150 Remote Sensing for Sustainability

research efforts to demonstrate the qualifications of remote sensing for monitoring

forest disturbances attributed to diseases and insects.

8.3.1 Spectral characteriSticS

The spectral values in a forest image scene are often biased to representing the upper
layer traits of tree canopies. While the top-down manner of photographing vegeta-
tion lacks the ability to characterize the entire tree, it is possible to link the status of
canopy to forest health because diseases or insects substantially affect a tree’s ability
to photosynthesize and store moisture in foliage. One consequence is the noticeable
change in foliage color (i.e., discoloration). For example, needles on pine trees turn
red in the red-attack stage by mountain pine beetle (Wulder et al. 2006). Oak trees
visually appear brown and freeze-dried as a result of sudden oak death (Kelly and
Meentemeyer 2002). Remote sensors with the capacity to record the visible portion
of the electromagnetic spectrum (wavelengths from approximately 400 to 700 nm)
are able to detect these symptoms, which appear similarly in the human visual sys-
tem. However, disease- and insect-mediated forest mortality is a gradual process.
Some early-stage symptoms cannot be easily observed; for instance, unhealthy trees
with reduced chlorophylls may only appear to be slightly brighter than the healthy
trees in the visible spectral range owing to reduced absorbance of the blue and red
wavelengths by foliage (Knipling 1970). Sensors with the capacity to further detect
the near-infrared spectrum (wavelengths from approximately 700 to 1300 nm) are
probably more sensitive to such physiological stress. Similarly, the amount of energy
reflected in the short-wave infrared range (wavelengths from approximately 1300 to
2500 nm) is correlated with vegetation moisture (Laurent et al. 2005). Today’s remote
sensing technologies are already capable of recording the radiation reflected in those
spectral ranges. To further advance the performance of remote detection, research-
ers utilized a variety of spectral indices (i.e., combinations of spectral bands) and
have repetitively confirmed their effectiveness in monitoring forest damage subject
to disease and insect attacks (see case studies in Table 8.1). Examples of the indi-
ces include normalized difference vegetation index (NDVI; Tucker 1979), enhanced
vegetation index (Liu and Huete 1995), disturbance index (DI; Healey et al. 2005),
normalized difference moisture index (NDMI; Jin and Sader 2005), normalized dif-
ference infrared index (Jackson et al. 2004), and enhanced wetness difference index
(EWDI; Skakun et al. 2003).
While multispectral imagery has proven its potential to assess the status of
damaging diseases and insects, previous studies discovered that the subtle spectral
discrepancies between healthy and damaged trees (e.g., during the previsual green
mortality stage) can be better detected by fine-spectral resolution data, that is, dozens
to hundreds of narrow and contiguous spectral bands acquired through hyperspec-
tral imaging (Coops et al. 2003; Hatala et al. 2010). On the basis of this technology,
researchers have further developed narrowband vegetation indices, some of which
were freshly designed (e.g., transformed chlorophyll absorption reflectance index;
Haboudane et al. 2002), while the others were simple modifications of the traditional
vegetation indices by means of substituting narrowband for broadband reflectance
(e.g., red edge NDVI; Gitelson and Merzlyak 1994). Although not as common as the
Remote Sensing of Forest Damage by Diseases and Insects 151

broadband indices yet, narrowband indices have shown the potential to explain the
physiological changes in the forests suffering damage from insects (Fassnacht et al.
2014; Oumar et al. 2013).

8.3.2 Spatial characteriSticS

Recent development in remote sensing allows us to perceive spatial details on the
Earth’s surface at varying scales, for example, 1 km/500 m/250 m MODIS, 30 m/15 m
Landsat, 10 m/5 m SPOT-5, 4 m/1 m IKONOS, 1.2 m/0.3 m Worldview-3, and
centimeter-level aerial photos. This offers forest practitioners a range of choices for
balancing the accuracy of detecting disease or insect occurrence and data acquisition
cost. Typically, coarse to moderate-resolution imagery has been traditionally applied
to measure forest structural change at the landscape scale. For example, de Beurs
and Townsend (2008) applied MODIS data with a 250-m spatial resolution to moni-
tor more than 16,000 km2 of insect defoliation of hardwood forests by gypsy moth.
Fraser and Latifovic (2005) showed that 1-km-resolution SPOT VEGETATION data
were sufficient for mapping a 350,000-km2 area of coniferous forest mortality in
Quebec, Canada, caused by the eastern hemlock looper. A higher-severity distur-
bance event may lead to a more satisfactory detection result, because the infected
tree patches tend to be larger on average.
However, challenges arise if the majority of the damaged trees are within small,
discrete patches. High–spatial resolution satellite and airborne imagery are more
suitable for fine-scale detection and have proven to be feasible in previous studies
(e.g., Adelabu et al. 2014; Cheng et al. 2010; Kautz 2014; Meddens et al. 2011; Wulder
et al. 2008). It should be noted that a unique consideration of processing such type of
data sets is the recent paradigm shift from pixel-based to object-based image analy-
sis, that is, geographic object-based image analysis (GEOBIA; Blaschke et al. 2014).
Because a high-resolution pixel often covers a portion of a tree or a small tree cluster,
the corresponding pixel value may contain a high spectral variation as a result of
the complex forest 3D structure and sun–tree–sensor geometry (Chen et al. 2011).
Compared to the traditional pixel-based modeling, GEOBIA extracts image objects
(groups of pixels) to represent meaningful geographic objects, for the purpose of
reducing spectral noises and increasing mapping accuracies.

8.3.3 temporal characteriSticS

The size of Earth observation data archives is growing at an unprecedented pace.
With rich time series data, it becomes feasible to extract the trajectories of dis-
ease and insect progression over a long term (e.g., Meigs et al. 2011; Vogelmann
et al. 2009; Walter and Platt 2013). Because most of the infected trees do not die
instantly, many forest disease or insect studies tend to apply annual or biannual
imagery to characterize the spatiotemporal patterns of forest change. To mitigate
the impact of seasonal variation, multidate images are preferably collected in the
same months or the same seasons. Of the variety of date archives, Landsat time
series have been the most widely used (see case studies in Table 8.1). This is pos-
sibly attributed to the features of four decades of data storage with minimized
152 Remote Sensing for Sustainability

temporal gaps, free data access, and global coverage (Woodcock et al. 2008).
However, as we are entering the remote sensing big data era, we expect to see an
increasing application of diverse data archives for long-term forest health monitor-
ing in the near future.

8.4 A REVIEW OF REMOTE SENSING ALGORITHMS

To date, a variety of remote sensing algorithms have been developed to measure for-
est damage caused by diseases and insects. The main principle is to extract the dif-
ferences in spectral reflectance between healthy and infected trees, as well as among
the infected trees during varying stages of decline. Here, we provide a brief review
of those algorithms and categorize them into five groups: thresholding, classification,
change detection, statistical regression, and the others, with details described below.

8.4.1 threSholding
Compared to healthy trees, damaged trees have distinct symptoms, such as reduced
moisture, discolored foliage, and defoliated canopy. A thresholding method defines
one or multiple thresholds to extract the pixels representing damaged trees from the
entire forest image scene. While the operation appears simple, the success of apply-
ing thresholding largely depends on the effective description of forest symptoms and
the accurate definition of threshold(s).
Describing the symptoms of forest damage has been primarily relying on image
spectral indices. Some of those indices were specifically designed to assess forest
disturbances. For example, Coops et al. (2006) created a red–green index, the ratio
of QuickBird red to green wavelengths, to extract the red-attack damage (i.e., foliage
color turning red from green) in the mountain pine beetle–infested coniferous forests.
Their results confirmed the potential of using a simple threshold to red–green index
values for separating the infected from the healthy trees. For many other studies,
however, thresholding methods often directly employed or modified the existing indi-
ces that had not been intentionally developed for monitoring infestation. For example,
multiple thresholds were applied to Landsat NDMI for extracting beetle-infested trees
and forest regrowth after disturbance events (Coops et al. 2010; Goodwin et al. 2008).
Similarly, Coops et al. (2009) calculated DI using 1-km-resolution MODIS images
covering a part of the terrestrial land base of Canada. They found that those DI pixel
values larger than ±1 standard deviation of the long-term mean were consistent with
the areas flagged as infested using aerial survey. To further improve the thresholding
performance, Skakun et al. (2003) created an EWDI through combing three differ-
ent dates of wetness bands (derived from the Landsat TM tasseled cap transforma-
tion). Likewise, Olsson et al. (2012) modified the classic NDVI index by substituting
the green band for the red band in the equation. The new index GNDVI was found
to outperform NDVI, and negative GNDVI values indicated damage. Overall, the
thresholding methods are simple to implement, with thresholds typically defined with
assistance of field survey and manual photo interpretation. One major limitation for
Remote Sensing of Forest Damage by Diseases and Insects 153

thresholding is that it is only suitable to identify major stages of forest disturbances,

for example, extracting heavily damaged trees from healthy ones.

8.4.2 claSSification
Land-cover classification using imagery to differentiate between land-cover types
was developed almost immediately after the advent of remote sensing. The suitabil-
ity of using image classification to measure forest damage is based on the fact that
the distinct symptoms of infected forests make them appear as new land-cover types.
It also seems to be consistent and convenient to apply one classification framework
to extract not only the damaged/healthy forests but also the other land-cover types
coexisting with forests, for example, grasses, shrubs, built-ups, and water.
Of the variety of classification algorithms, the classic supervised maximum like-
lihood classifier (MLC) demonstrated continued success in forest disease and insect
monitoring. For example, MLC was effectively applied to Landsat imagery for dif-
ferentiating mountain pine beetle–induced red attacks from non-red attacks (Walter
and Platt 2013). MLC and Landsat imagery were also used to extract gypsy moth–
caused defoliation from the nondefoliated trees (Thayn 2013). In addition, previous
studies suggested that the application of MLC to classify high–spatial resolution
imagery has the potential to detect forest damage of multiple stages. For example,
Leckie et al. (2005) was able to estimate jack pine budworm–induced four classes of
discoloration (nil–trace, light, moderate, and severe) through the application of MLC
and 2.5-m-resolution aerial imagery acquired from the multispectral electro-optical
imaging sensor. Meddens et al. (2011) and Bright et al. (2012) independently used
aerial photography and MLC to classify beetle-caused tree mortality into green, red
(dead trees with red needles), and gray (dead trees without needles) tree classes with
the same overall accuracy of 87%. When integrated with hyperspectral imagery,
MLC was found to be a viable solution to estimate forest stress during the early pre-
visual stage of a sudden oak death outbreak (Pu et al. 2008).
Novel machine learning methods, as a complement to classic classifiers, have been
introduced to the domain of remote sensing classification since the 1990s. Support
vector machines (SVMs) are a successful example, which have proven to be feasible
to detect three levels of insect defoliation ranging from nonimpacted undefoliated
plants to partly defoliated plants and finally refoliating plants after severe defoliation
in an African savanna (Adelabu et al. 2014). When applied to classify hyperspectral
imagery acquired from HyMap, SVMs were found to have notable high overall accu-
racies mapping bark beetle–caused tree mortality, with the best result reaching as
high as 97% accuracy (Fassnacht et al. 2014). Random forests (RFs) act as another
popular machine learning method in classification. In a case study of mapping insect
defoliation levels with RapidEye 5-m-resolution imagery, Adelabu et al. (2014) com-
pared RFs and SVMs, and found comparable results. It should be noted that one
outstanding feature of RF is that it can rank all the input variables based on their
importance (Breiman 2001), which facilitates result analysis by identifying the most
crucial spectral bands or indices in disease and insect mapping.
154 Remote Sensing for Sustainability

A subpixel classification scheme is needed if the spatial resolution of image pixels

is too coarse to detect small, fragmented disturbances in a patchy distribution. To
do so, spectral mixture analysis (SMA) provides a viable means, which is typically
based on the assumption that the spectral value of each pixel is a linear combination
of the reflectance from surface materials (endmembers) weighted by their factions.
For example, Radeloff et al. (1999) performed SMA on Landsat TM imagery to clas-
sify jack pine budworm defoliation levels in a mixed forest stand and found a strong
negative correlation between SMA-derived green needle fraction and field-measured
budworm population (r = −0.94). With SMA and 0.5-m-resolution multispectral imag-
ery, Goodwin et al. (2005) quantified the fractional abundance of three endmembers:
sunlit canopy, shadow, and soil. Their results suggested a possibility of using the sunlit
canopy image fraction to describe crown/leader color in the forests affected by dam-
aging agents. When it comes to classifying hyperspectral imagery, the high spectral
noises in data often challenge the performance of classifiers. To address the issue when
using HyMap imagery, Hatala et al. (2010) employed the mixture-tuned matched-filter
algorithm, an improved SMA through maximizing the target response and minimiz-
ing background spectral signatures, to classify whitebark pine stress and mortality.

8.4.3 StatiStical regreSSion

Statistical regression analysis allows practitioners to estimate not only the discrete
stages of forest disturbances (e.g., damaged vs. healthy) but also continuous defolia-
tion or tree mortality levels from none to 100%. Compared to most classification
methods, regression has the capacity to demonstrate the significance of the selected
explanatory variables derived from remote sensing imagery. Such information can
inform sustainable forest management, for example, predicting forest vulnerability
in response to disease or insect attacks.
Logistic regression has been shown as a simple solution for identifying forest sta-
tus of being damaged or not. For example, this model was applied to estimate an out-
break of black-headed budworm in Western Newfoundland, Canada, with a proven
success to distinguish susceptible trees from those that were not (Luther et al. 1997).
However, such analysis may not be sufficient for developing effective mitigation
strategies. Researchers have expressed higher interests in understanding the detailed
(i.e., continuous) tree damage levels. To do so, classic multiple linear regression was
widely used to link remote sensing–derived metrics (e.g., spectral bands, spectral
indices, and topographic variables) with field-measured damage indicators, such as
defoliation intensity (de Beurs and Townsend 2008; Pontius et al. 2008), basal area
(Siderhurst et al. 2010), leaf area index (Solberg et al. 2006), foliar nitrogen and plant
growth vigor (McNeil et al. 2007), concentration of total chlorophyll (Cheng et al.
2010), and leaf water content (Cheng et al. 2010). Their studies also indicated the
suitability of applying multiple regression to analyze a wide range of remote sensing
data types (e.g., MODIS, Landsat, lidar, and the hyperspectral).
Recent sensor development has increased image spectral resolution and extended
the coverage of data spectral range. However, this poses a challenge to regression
modeling, that is, high dimensionality and collinearity of remotely sensed explan-
atory variables. To address this issue, Verbesselt et al. (2009) applied the least
Remote Sensing of Forest Damage by Diseases and Insects 155

absolute shrinkage and selection operator (LASSO) to model bark beetle–induced

tree mortality in Pinus radiata plantations. Compared to the standard data fitting
method of least squares, LASSO is an alternative regularized version to minimizing
the residual sum of squares “under a constraint on the sum of the absolute values of
regression coefficient estimates” (Verbesselt et al. 2009). Another solution is partial
least squares regression (also known as projection to latent structures), which finds
new hyperplanes for minimizing the variance between impendent and dependent
variables (Geladi and Kowalski 1986). Researchers have confirmed its effectiveness
of mitigating the variable multicollinearity effects in studying insect-caused forest
damage (Oumar and Mutanga 2014; Oumar et al. 2013; Wolter et al. 2008).
The aforementioned regression models are considered as fixed effects, that is,
treating all the variables as nonrandom. However, Rullán-Silva et al. (2015) argued
that a mixed-effects model, containing both fixed and random effects, is more appro-
priate for estimating the percentage of defoliation caused by beech leaf-miner weevil.
The addition of random effects to a fixed-effects model was found to better account
for the variability possibly introduced by environmental uncertainties (Rullán-Silva
et al. 2015). While the mixed-effects models are relatively new to the field of remote
sensing, we note that their merits have been increasingly recognized in forest ecol-
ogy (Bolker et al. 2009).

8.4.4 change detection

Change detection employs multitemporal imagery (i.e., time series data) to measure
the spatial patterns of forest disturbances through time. In contrast with using single-
date imagery to identify damaged trees, this approach analyzes shifts in spectral
reflectance across multiple dates. Accordingly, extra considerations are required to
deal with spectral variation through time that arises from both forest disturbances
and differences in atmospheric conditions and the sun–view–tree geometries (Chen
et al. 2011; Song et al. 2001).
Previous efforts showed two ways of conducting change detection. First, the spectral
discrepancies between multidate images are calculated through differencing the same
spectral bands or indices from the base year (before disturbance) and the disturbance
year(s). This is followed by applying thresholding, statistical regression, or classifica-
tion to extract the pixels containing higher spectral variation (indicating damaged trees)
than the others (e.g., de Beurs and Townsend 2008; Townsend et al. 2012; Wulder et al.
2008). Second, change detection focuses on measuring forest damage directly through
all the spectral bands or indices. For example, Babst et al. (2010) applied principal com-
ponent analysis to transfer multidate NDVI images (derived from Landsat time series)
into new principal components. They discovered that the second principal component
contained crucial information representing the change of NDVI, which was corre-
lated with the level of defoliation caused by autumnal moth. Additionally, because the
spectral discrepancies among the Landsat time series include both real and noisy false
changes, Kennedy et al. (2010) developed a LandTrendr temporal segmentation algo-
rithm to capture only the salient features of the trajectory (representing real changes)
using a multilevel model fitting strategy. This algorithm was employed by Meigs et al.
(2011) to successfully characterize the impacts of bark beetle on tree mortality.
156 Remote Sensing for Sustainability

8.4.5 additional approacheS

In addition to the aforementioned mainstream methods, several other algorithms
have been developed for unique considerations in disease and insect monitoring.
For example, the occurrence of tree dieback is associated with specific forest envi-
ronmental factors (e.g., distance from hosts to target trees; Kelly and Meentemeyer
2002). Liu et al. (2006) modeled such ecological compatibility with Markov random
field, which was used to refine the results from a noncontextual SVM classification.
To deal with nontraditional data types, such as lidar for characterizing forest 3D
structure (Chen and Hay 2011; Lim et al. 2003), Zhang (2008) applied mathematical
morphology to process lidar point clouds for identifying small gaps in mangrove
forests owing to natural disturbances, including the outbreaks of insects. Bright et al.
(2012) employed lidar to estimate forest aboveground carbon. When integrated with
the beetle-caused tree mortality map from an MLC classification, the carbon storage
map was able to clearly reveal the impact of insect severity on forest carbon loss.

8.5 CHALLENGES AND OPPORTUNITIES

Advancements in remote sensing data acquisition and analysis have remarkably
improved the feasibility of assessing landscape-scale forest disturbances induced by
diseases or insects. However, challenges remain. In this section, we identify some of
those challenges and suggest potential solutions.

8.5.1 early Warning of foreSt damage

Forests that are infected by diseases or insects do not die instantly. The detection
of early stage forest damage offers forest managers an opportunity to perform effi-
cient disease and insect control. During this stage, the infected trees may only show
a slight decline in chlorophyll levels and leaf water content. Previous efforts have
confirmed the potential of applying hyperspectral remote sensing to assist with early
detection of tree stress (e.g., Fassnacht et al. 2014; Pu et al. 2008). However, most of
the sensors were mounted on airborne platforms (e.g., CASI, HyMap, and AVIRIS),
making data acquisition an expensive process. To date, only a few satellite sensors
(e.g., EO-1 Hyperion) are operational, although their application has been restrained
because of limited spatial coverage and high spectral noises. To address the chal-
lenge, developing Landsat-like hyperspectral sensors is a promising solution. For
example, NASA’s hyperspectral infrared imager (HyspIRI) mission will mount two
instruments on a satellite in low Earth orbit. Once launched, HyspIRI will deliver
global coverage hyperspectral imagery at the 10-nm spectral resolution from the vis-
ible, short-wave infrared range to the thermal infrared range (NASA 2015). Another
potential solution is to assemble a small, inexpensive hyperspectral unmanned air-
craft system (UAS; see a recent review by Pajares [2015]). While such a system still
has small spatial coverage, its highly operational flexibility combined with a proper
sampling strategy makes early warning feasible. One limitation, however, is the obli-
gation to meet UAS regulations and policies that may vary considerably from region
to region.
Remote Sensing of Forest Damage by Diseases and Insects 157

8.5.2 conSiStent monitoring of long-term, hiStorical foreSt damage

While several remote sensing programs (e.g., AVHRR, Landsat, or SPOT) have been
operational for three to four decades, many new types of sensors appeared only
recently, such as those featuring high spatial resolution, hyperspectral resolution,
and the ability to characterize forest 3D structure. These new sensors do have a
higher capacity to detect forest stress and mortality; however, their data archives
often have limited temporal and spatial coverage. This poses a challenge for consis-
tently monitoring the long-term, historical impacts of diseases and insects on for-
ests. One dilemma facing many researchers is that the study area was only partially
covered by the data acquired from high-performance sensors for limited periods.
Choosing the data that have full coverage (e.g., Landsat) can be one solution, while
combining data from multiple sensors can be another solution (e.g., using Landsat
data to fill in the gaps that lack hyperspectral imagery). In the latter case, the devel-
oped algorithms should have the capacity to accommodate varying types of remote
sensing data across spatial, spectral, and temporal scales, so that all the results can
be compared using consistent criteria.

8.5.3 differentiating among compound diSturbanceS

Forests are a natural ecosystem. The disturbances affecting the same forested regions
may come from a range of sources. Besides insect and disease, other natural disasters
(e.g., wildfire and wind) or anthropogenic activities (e.g., logging) can lead to com-
pound disturbances. It is also possible that one disturbance regime (e.g., wildfire) may
influence forest responses to another disturbance (e.g., disease), resulting in interact-
ing disturbances (Turner 2010). Recent remote sensing studies have been limited on
the topic of differentiating between disease-/insect-caused forest damage and other
types of damage. One major challenge is that single sensors are typically not suitable
to complete this task. For example, in a study of estimating burn severity in a for-
est that had experienced pre-fire disease outbreaks, Chen et al. (2015b) found similar
spectral reflectance in burned and diseased trees using Landsat imagery. Therefore, a
likely solution is the development of a multisensor approach, taking advantage of the
strengths from individual sensors, for example, Landsat time series for temporal analy-
sis of disease and insect progression, hyperspectral imaging for tracking the early signs
of forest damage, and lidar for assessing the change in forest vertical profiles. Data
integration maximizes practitioners’ ability to estimate changes in forest biophysical
and biochemical parameters, augmenting accurate assessments of forest damage.

8.6 CONCLUSION
Global forest ecosystems face high frequencies of landscape-level disturbances
resulting from disease and insect epidemics. Over the past decades, remote sensing
tools have improved detection of forest disturbances in a timely and cost-effective
manner. As sensor technologies advance, richer Earth observation data with higher
spatial, spectral, and temporal resolutions are expected to offer better choices
to assess varying stages of disease/insect invasion in a range of forest biomes.
158 Remote Sensing for Sustainability

Accordingly, algorithms for modeling spectra–disturbance relationships will need

to be continually refined or redeveloped to take advantage of new data and novel
landscape changes caused by nonnative, invasive pathogens and insects.

ACKNOWLEDGMENT
This research was supported by the University of North Carolina at Charlotte through
a Junior Faculty Development Award.

REFERENCES
Adelabu, S., Mutanga, O., & Adam, E. (2014). Evaluating the impact of red-edge band
from Rapideye image for classifying insect defoliation levels. ISPRS Journal of
Photogrammetry and Remote Sensing, 95, 34–41.
Asner, G. P. (2013). Geography of forest disturbance. Proceedings of the National Academy of
Sciences of the United States of America, 110, 3711–2.
Assal, T. J., Sibold, J., & Reich, R. (2014). Remote sensing of environment modeling a histori-
cal mountain pine beetle outbreak using Landsat MSS and multiple lines of evidence.
Remote Sensing of Environment, 155, 275–288.
Babst, F., Esper, J., & Parlow, E. (2010). Landsat TM/ETM+ and tree-ring based assessment
of spatiotemporal patterns of the autumnal moth (Epirrita autumnata) in northernmost
Fennoscandia. Remote Sensing of Environment, 114, 637–646.
Blaschke, T., Hay, G. J., Kelly, M. et al. (2014). Geographic object-based image analysis—Towards
a new paradigm. ISPRS Journal of Photogrammetry and Remote Sensing, 87, 180–191.
Bolker, B. M., Brooks, M. E., Clark, C. J., Geange, S. W., Poulsen, J. R., Stevens, M. H. H., &
White, J. S. S. (2009). Generalized linear mixed models: A practical guide for ecology
and evolution. Trends in Ecology and Evolution, 24, 127–135.
Bonan, G. B. (2008). Forests and climate change: Forcings, feedbacks, and the climate ben-
efits of forests. Science, 320, 1444–1449.
Boyd, I. L., Freer-Smith, P. H., Gilligan, C. A, & Godfray, H. C. J. (2013). The consequence
of tree pests and diseases for ecosystem services. Science, 342(6160), 1235773.
doi:10.1126/science.1235773.
Breiman, L. (2001). Random forests. Machine Learning, 45, 5–32.
Bright, B. C., Hicke, J. A., & Hudak, A. T. (2012). Estimating aboveground carbon stocks of a
forest affected by mountain pine beetle in Idaho using lidar and multispectral imagery.
Remote Sensing of Environment, 124, 270–281.
Chen, G., & Hay, G. J. (2011). A support vector regression approach to estimate forest bio-
physical parameters at the object level using airborne lidar transects and QuickBird data.
Photogrammetric Engineering and Remote Sensing, 77, 733–741.
Chen, G., Hay, G. J., Castilla, G., & St-Onge, B. (2011). A multiscale geographic object-based
image analysis to estimate lidar-measured forest canopy height using Quickbird imag-
ery. International Journal of Geographical Information Science, 25, 877–893.
Chen, G., Metz, M. R., Rizzo, D. M., Dillon, W. W., & Meentemeyer, R. K. (2015a). Object-
based assessment of burn severity in diseased forests using high-spatial and high-
spectral resolution MASTER airborne imagery. ISPRS Journal of Photogrammetry and
Remote Sensing, 102, 38–47.
Chen, G., Metz, M. R., Rizzo, D. M., & Meentemeyer, R. K. (2015b). Mapping burn sever-
ity in a disease-impacted forest landscape using Landsat and MASTER imagery.
International Journal of Applied Earth Observation and Geoinformation, 40, 91–99.
Remote Sensing of Forest Damage by Diseases and Insects 159

Cheng, T., Rivard, B., Sánchez-Azofeifa, G. A., Feng, J., & Calvo-Polanco, M. (2010).
Continuous wavelet analysis for the detection of green attack damage due to mountain
pine beetle infestation. Remote Sensing of Environment, 114, 899–910.
Coops, N., Stanford, M., Old, K., Dudzinski, M., Culvenor, D., & Stone, C. (2003). Assessment
of dothistroma needle blight of Pinus radiata using airborne hyperspectral imagery.
Ecology and Epidemiology, 93, 1524–1532.
Coops, N. C., Johnson, M., Wulder, M. A., & White, J. C. (2006). Assessment of QuickBird
high spatial resolution imagery to detect red attack damage due to mountain pine beetle
infestation. Remote Sensing of Environment, 103, 67–80.
Coops, N. C., Wulder, M. A., & Iwanicka, D. (2009). Large area monitoring with a MODIS-
based Disturbance Index (DI) sensitive to annual and seasonal variations. Remote
Sensing of Environment, 113(6), 1250–1261.
Coops, N. C., Gillanders, S. N., Wulder, M. A., Gergel, S. E., Nelson, T., & Goodwin, N. R.
(2010). Assessing changes in forest fragmentation following infestation using time
series Landsat imagery. Forest Ecology and Management, 259, 2355–2365.
de Beurs, K. M., & Townsend, P. A. (2008). Estimating the effect of gypsy moth defoliation
using MODIS. Remote Sensing of Environment, 112(10), 3983–3990.
Fassnacht, F. E., Latifi, H., Ghosh, A., Joshi, P. K., & Koch, B. (2014). Assessing the potential
of hyperspectral imagery to map bark beetle-induced tree mortality. Remote Sensing of
Environment, 140, 533–548.
Fraser, R. H., & Latifovic, R. (2005). Mapping insect-induced tree defoliation and mortal-
ity using coarse spatial resolution satellite imagery. International Journal of Remote
Sensing, 26, 193–200.
Geladi, P., & Kowalski, B. R. (1986). Partial least-squares regression: A tutorial. Analytica
Chimica Acta, 185, 1−17.
Gitelson, A., & Merzlyak, M. (1994). Spectral reflectance changes associated with autumn
senescence of Aesculus hippocastanum L. and Acer platanoides L. leaves. Journal of
Plant Physiology, 143, 286–292.
Goodwin, N., Coops, N. C., & Stone, C. (2005). Assessing plantation canopy condition
from airborne imagery using spectral mixture analysis and fractional abundances.
International Journal of Applied Earth Observation and Geoinformation, 7, 11–28.
Goodwin, N. R., Coops, N. C., Wulder, M. A., Gillanders, S., Schroeder, T. A., & Nelson, T.
(2008). Estimation of insect infestation dynamics using a temporal sequence of Landsat
data. Remote Sensing of Environment, 112, 3680–3689.
Haboudane, D., Miller, J. R., Tremblay, N., Zarco-Tejada, P. J., & Dextraze, L. (2002).
Integrated narrow-band vegetation indices for prediction of crop chlorophyll content
for application to precision agriculture. Remote Sensing of Environment, 81, 416–426.
Hatala, J. A., Crabtree, R. L., Halligan, K. Q., & Moorcroft, P. R. (2010). Landscape-scale pat-
terns of forest pest and pathogen damage in the Greater Yellowstone Ecosystem. Remote
Sensing of Environment, 114, 375–384.
Healey, S. P., Cohen, W. B., Yang, Z., & Krankina, O. N. (2005). Comparison of Tasseled Cap-
based Landsat data structures for use in forest disturbance detection. Remote Sensing of
Environment, 97, 301–310.
Heller, R.C., & Bega, R. V. (1973). Detection of forest diseases by remote sensing. Journal of
Forestry, 71, 18–21.
Huang, L., Ning, Z., & Zhang, X. (2010). Impacts of caterpillar disturbance on forest net pri-
mary production estimation in China. Ecological Indicators, 10, 1144–1151.
Jackson, T. L., Chen, D., Cosh, M., Li, F., Anderson, M., Walthall, C., Doriaswamy, P., &
Hunt, E. R. (2004). Vegetation water content mapping using Landsat data derived nor-
malized difference water index for corn and soybeans. Remote Sensing of Environment,
92, 475–482.
160 Remote Sensing for Sustainability

Jin, S., & Sader, S. A. (2005). MODIS time-series imagery for forest disturbance detection
and quantification of patch size effects. Remote Sensing of Environment, 99, 462–470.
Kautz, M. (2014). On correcting the time-lag bias in aerial-surveyed bark beetle infestation
data. Forest Ecology and Management, 326, 157–162.
Kelly, M., & Meentemeyer, R. K. (2002). Landscape dynamics of the spread of sudden oak
death. Photogrammetric Engineering & Remote Sensing, 68, 1001–1009.
Kennedy, R. E., Yang, Z., & Cohen, W. B. (2010). Detecting trends in forest disturbance and
recovery using yearly Landsat time series: 1. LandTrendr—Temporal segmentation
algorithms. Remote Sensing of Environment, 114, 2897–2910.
Knipling, E. B. (1970). Physical and physiological basis for the reflectance of visible and near-
infrared radiation from vegetation. Remote Sensing of Environment, 1, 155–159.
Lamsal, S., Cobb, R. C., Hall Cushman, J., Meng, Q., Rizzo, D. M., & Meentemeyer, R. K. (2011).
Spatial estimation of the density and carbon content of host populations for Phytophthora
ramorum in California and Oregon. Forest Ecology and Management, 262, 989–998.
Laurent, E. J., Shi, H., Gatziolis, D., LeBouton, J. P., Walters, M. B., & Liu, J. (2005). Using
the spatial and spectral precision of satellite imagery to predict wildlife occurrence pat-
terns. Remote Sensing of Environment, 97, 249–262.
Leckie, D. G., Cloney, E., & Joyce, S. P. (2005). Automated detection and mapping of crown
discolouration caused by jack pine budworm with 2.5 m resolution multispectral imag-
ery. International Journal of Applied Earth Observation and Geoinformation, 7, 61–77.
Lim, K., Treitz, P., Wulder, M., & Flood, M. (2003). LiDAR remote sensing of forest structure.
Progress in Physical Geography, 1, 88–106.
Liu, H., & Huete, A. (1995). A feedback based modification of the NDVI to minimize can-
opy background and atmospheric noise. IEEE Transactions on Geoscience and Remote
Sensing, 33, 457–465.
Liu, D., Kelly, M., & Gong, P. (2006). A spatial–temporal approach to monitoring forest dis-
ease spread using multi-temporal high spatial resolution imagery. Remote Sensing of
Environment, 101(2), 167–180.
Liu, D., Kelly, M., Gong, P., & Guo, Q. (2007). Characterizing spatial–temporal tree mortality pat-
terns associated with a new forest disease. Forest Ecology and Management, 253, 220–231.
Luther, J. E., Franklin, S. E., Hudak, J., & Meades, J. P. (1997). Forecasting the susceptibility
and vulnerability of balsam fir stands to insect defoliation with Landsat thematic mapper
data. Remote Sensing of Environment, 59, 77–91.
McNeil, B. E., de Beurs, K. M., Eshleman, K. N., Foster, J. R., & Townsend, P. A. (2007).
Maintenance of ecosystem nitrogen limitation by ephemeral forest disturbance: An
assessment using MODIS, Hyperion, and Landsat ETM+. Geophysical Research
Letters, 34, L19406.
Meddens, A. J. H., Hicke, J. A., & Vierling, L. A. (2011). Evaluating the potential of multispec-
tral imagery to map multiple stages of tree mortality. Remote Sensing of Environment,
115, 1632–1642.
Meentemeyer, R. K., Dorning, M. A., Vogler, J. B., Schmidt, D., & Garbelotto, M. (2015).
Citizen science helps predict risk of emerging infectious disease. Frontiers in Ecology
and the Environment, 13, 189–194.
Meentemeyer, R. K., Rank, N. E., Shoemaker, D. A., Oneal, C. B., Wickland, A. C., Frangioso,
K. M., & Rizzo, D. M. (2008). Impact of sudden oak death on tree mortality in the Big
Sur ecoregion of California. Biological Invasions, 10, 1243–1255.
Meigs, G. W., Kennedy, R. E., & Cohen, W. B. (2011). A Landsat time series approach to char-
acterize bark beetle and defoliator impacts on tree mortality and surface fuels in conifer
forests. Remote Sensing of Environment, 115, 3707–3718.
Meigs, G. W., Kennedy, R. E., Gray, A. N., & Gregory, M. J. (2015). Spatiotemporal dynamics
of recent mountain pine beetle and western spruce budworm outbreaks across the Pacific
Northwest Region, USA. Forest Ecology and Management, 339, 71–86.
Remote Sensing of Forest Damage by Diseases and Insects 161

NASA (2015). HyspIRI Mission Study. Available on: https://hyspiri.jpl.nasa.gov.

Nelson, R. F. (1983). Detecting forest canopy change due to insect activity using Landsat
MSS. Photogrammetric Engineering and Remote Sensing, 49, 1303–1314.
Olsson, P.-O., Jönsson, A. M., & Eklundh, L. (2012). A new invasive insect in Sweden—
Physokermes inopinatus: Tracing forest damage with satellite based remote sensing.
Forest Ecology and Management, 285, 29–37.
Oumar, Z., & Mutanga, O. (2014). Integrating environmental variables and WorldView-2
image data to improve the prediction and mapping of Thaumastocoris peregrinus
(bronze bug) damage in plantation forests. ISPRS Journal of Photogrammetry and
Remote Sensing, 87, 39–46.
Oumar, Z., Mutanga, O., & Ismail, R. (2013). Predicting Thaumastocoris peregrinus dam-
age using narrow band normalized indices and hyperspectral indices using field spectra
resampled to the hyperion sensor. International Journal of Applied Earth Observation
and Geoinformation, 21, 113–121.
Pajares, G. (2015). Overview and current status of remote sensing applications based on unmanned
aerial vehicles (UAVs). Photogrammetric Engineering & Remote Sensing, 81, 281–329.
Pan, Y., Birdsey, R. A, Fang, J. et al. (2011). A large and persistent carbon sink in the world’s
forests. Science, 333, 988–93.
Pontius, J., Martin, M., Plourde, L., & Hallett, R. (2008). Ash decline assessment in emer-
ald ash borer-infested regions: A test of tree-level, hyperspectral technologies. Remote
Sensing of Environment, 112, 2665–2676.
Pu, R., Kelly, M., Anderson, G. L., & Gong, P. (2008). Using CASI hyperspectral imagery to
detect mortality and vegetation stress associated with a new hardwood forest disease.
Photogrammetric Engineering & Remote Sensing, 74, 65–75.
Radeloff, V. C., Mladenoff, D. J., & Boyce, M. S. (1999). Detecting Jack Pine budworm defo-
liation using spectral mixture analysis: Separating effects from determinants. Remote
Sensing of Environment, 169, 156–169.
Raffa, K. F., Powell, E. N., & Townsend, P. A. (2013). Temperature-driven range expansion of
an irruptive insect heightened by weakly coevolved plant defenses. Proceedings of the
National Academy of Sciences of the United States of America, 110, 2193–2198.
Rizzo, D. M., Garbelotto, M. & Hansen, E. M. (2005). Phytophthora ramorum: Integrative
research and management of an emerging pathogen in California and Oregon forests.
Annual Review of Phytopathology, 43, 309–335.
Rock, B. N., Vogelmann, J. E., Williams, D. L., Vogelmann, A. F., & Hoshizaki, T. (1986).
Remote detection of forest damage. BioScience, 36, 439–445.
Rullán-Silva, C., Olthoff, A. E., Pando, V., Pajares, J. A., & Delgado, J. A. (2015). Remote
monitoring of defoliation by the beech leaf-mining weevil Rhynchaenus fagi in northern
Spain. Forest Ecology and Management, 347, 200–208.
Sadyś, M., Skjøth, C. A., & Kennedy, R. (2014). Back-trajectories show export of airborne
fungal spores (Ganoderma sp.) from forests to agricultural and urban areas in England.
Atmospheric Environment, 84, 88–99.
Sangüesa-Barreda, G., Camarero, J. J., García-Martín, A., Hernández, R., & de la Riva, J.
(2014). Remote-sensing and tree-ring based characterization of forest defoliation and
growth loss due to the Mediterranean pine processionary moth. Forest Ecology and
Management, 320, 171–181.
Siderhurst, L. A., Griscom, H. P., Hudy, M., & Bortolot, Z. J. (2010). Changes in light levels
and stream temperatures with loss of eastern hemlock (Tsuga canadensis) at a southern
Appalachian stream: Implications for brook trout. Forest Ecology and Management,
260, 1677–1688.
Skakun, R. S., Wulder, M. A., & Franklin, S. E. (2003). Sensitivity of the thematic mapper
enhanced wetness difference index to detect mountain pine beetle red-attack damage.
Remote Sensing of Environment, 86, 433–443.
162 Remote Sensing for Sustainability

Solberg, S., Næsset, E., Holt, K., & Christiansen, E. (2006). Mapping defoliation during a
severe insect attack on Scots pine using airborne laser scanning, 102, 364–376.
Song, C., Woodcock, C. E., Seto, K. C., Lenney, M. P., & Macomber, S. A. (2001). Classification
and change detection using Landsat TM data: When and how to correct atmospheric
effects? Remote Sensing of Environment, 75, 230–244.
Thayn, J. B. (2013). Using a remotely sensed optimized Disturbance Index to detect insect
defoliation in the Apostle Islands, Wisconsin, USA. Remote Sensing of Environment,
136, 210–217.
Townsend, P. A., Singh, A., Foster, J. R., Rehberg, N. J., Kingdon, C. C., Eshleman, K. N., &
Seagle, S. W. (2012). A general Landsat model to predict canopy defoliation in broad-
leaf deciduous forests. Remote Sensing of Environment, 119, 255–265.
Tucker, C. (1979). Red and photographic infrared linear combinations for monitoring vegeta-
tion. Remote Sensing of Environment, 8, 127–150.
Turner, M. G. (2010). Disturbance and landscape dynamics in a changing world. Ecology, 91,
2833–2849.
USDA Forest Service (2012). Major Forest Insect and Disease Conditions in the United
States: 2011. Available online: http://www.fs.fed.us/foresthealth/publications/Conditions
Report_2011.pdf.
Verbesselt, J., Robinson, A., Stone, C., & Culvenor, D. (2009). Forecasting tree mortality using
change metrics derived from MODIS satellite data. Forest Ecology and Management,
258, 1166–1173.
Vogelmann, J. E., Tolk, B., & Zhu, Z. (2009). Monitoring forest changes in the southwestern
United States using multitemporal Landsat data. Remote Sensing of Environment, 113,
1739–1748.
Walter, J. A., & Platt, R. V. (2013). Multi-temporal analysis reveals that predictors of mountain
pine beetle infestation change during outbreak cycles. Forest Ecology and Management,
302, 308–318.
Wang, C., He, H. S., & Kabrick, J. M. (2008). A risk rating study to predict oak decline and
recovery in the Missouri Ozark Highlands, USA. GIScience and Remote Sensing, 45,
406–425.
Wingfield, M. J., Brockerhoff, E. G., Wingfield, B. D., & Slippers, B. (2015). Planted forest
health: The need for a global strategy. Science, 349, 832–836.
Wolter, P. T., Townsend, P. A., & Sturtevant, B. R. (2009). Estimation of forest structural
parameters using 5 and 10 meter SPOT-5 satellite data. Remote Sensing of Environment,
113, 2019–2036.
Wolter, P. T., Townsend, P. A., Sturtevant, B. R., & Kingdon, C. C. (2008). Remote sens-
ing of the distribution and abundance of host species for spruce budworm in Northern
Minnesota and Ontario. Remote Sensing of Environment, 112(10), 3971–3982.
Woodcock, C. E., Allen, A. A., Belward, A. S. et al. (2008). Free access to Landsat imagery.
Science, 320, 1011.
Wulder, M. A., Dymond, C. C., White, J. C., Leckie, D. G., & Carroll, A. L. (2006). Surveying
mountain pine beetle damage of forests: A review of remote sensing opportunities.
Forest Ecology and Management, 221, 27–41.
Wulder, M. A., Ortlepp, S. M., White, J. C., Coops, N. C., & Coggins, S. B. (2009). Monitoring
the impacts of mountain pine beetle mitigation. Forest Ecology and Management, 258,
1181–1187.
Wulder, M. A., White, J. C., Coops, N. C., & Butson, C. R. (2008). Multi-temporal analy-
sis of high spatial resolution imagery for disturbance monitoring. Remote Sensing of
Environment, 112, 2729–2740.
Zhang, K. (2008). Identification of gaps in mangrove forests with airborne LIDAR. Remote
Sensing of Environment, 112, 2309–2325.
 9 Monitoring Water
Quality with Remote
Sensing Image Data
Bunkei Matsushita, Wei Yang,
Lalu Muhamad Jaelani, Fajar Setiawan,
and Takehiko Fukushima

CONTENTS
9.1 Introduction ..................................................................................................163
9.2 Major Satellite Sensors for Monitoring Water Quality ................................164
9.3 Water Delineation from Satellite Imagery....................................................169
9.4 Atmospheric Correction for Using Satellite Data .........................................174
9.5 Estimation of Water Quality Parameters from Remote Sensing Data .........177
9.5.1 Bio-Optical Model ............................................................................177
9.5.2 Estimations of the Biogeochemical Parameters of Water Bodies ....177
9.5.3 Estimations of the Inherent Optical Properties of Water Bodies .....180
9.5.4 Estimations of the Apparent Optical Properties of Water Bodies ....181
9.6 A Hybrid Approach for Developing a Universal Algorithm ........................182
9.7 Conclusions ...................................................................................................184
Acknowledgments..................................................................................................184
References ..............................................................................................................184

9.1 INTRODUCTION
Water covers approximately 74% of the Earth’s surface, and it plays many important
roles in the lives of all human being. Inland and coastal waters in particular have a
direct interface with human society by providing value for food supplies, industrial
uses, transportation, commerce, and human health (UNEP 2006). However, many
water bodies have encountered severe environmental problems (e.g., eutrophication)
in recent decades as a result of human interventions and climate change (Ayres et al.
1996; Haddeland et al. 2014). It is thus crucial to monitor and understand the amount
and quality of these water bodies as well as their biogeochemical processes in order
to achieve the effective management and sustainable use of the water resources
(United Nations Open Working Group 2014).
Since conventional water-monitoring methods (e.g., water sampling from a boat)
are very time-, labor-, and cost-consuming, the maintenance of steady monitor-
ing is difficult for local and national governments with meager financial resources,

163
164 Remote Sensing for Sustainability

especially in developing countries. For example, even when a lake is monitored by

an in situ sampling method, the number of sampling locations and the frequency
of water sampling may not accurately represent the actual status of the entire lake,
because the water quality can dramatically change both spatially and temporally
(Kiefer et al. 2015). However, with the progress of satellite sensors and water qual-
ity estimation algorithms, remote sensing is regarded as a quite useful technique for
studying bodies of water, as it can provide synoptic observations at very frequent
intervals for water bodies, at relatively low cost (Dekker and Hestir 2012).
In this chapter, we focus on how to obtain water quality data from remote sensing
techniques that can be used toward achieving the Sustainable Development Goal 6.3
(Monitoring Ambient Water Quality; United Nations Open Working Group 2014).
As noted above, we will describe major satellite sensors and the advances needed
to enable further progress in optical water-quality retrieval, several representative
algorithms for water area delineation, atmospheric correction, and water quality esti-
mations, plus a hybrid approach for the universal application of these algorithms.

9.2 MAJOR SATELLITE SENSORS

FOR MONITORING WATER QUALITY
Since the late 1970s, many attempts have been made to observe water quality from
space. The first ocean-color sensor, Coastal Zone Color Scanner (CZCS), was
launched in October 1978 aboard the Nimbus-7 satellite (Hovis et al. 1980). With four
spectral bands in visible range (443, 520, 550, and 670 nm), and one spectral band in
the thermal infrared spectrum (10.5–12.5 μm), the CZCS was the only spaceborne
sensor that observed global ocean colors during the period 1978–1986 (Table 9.1 and
Figure 9.1). However, since the CZCS was just a proof-of-concept sensor for the first
generation of ocean-color observation at that time, there were many limitations in
its instrument design such as a lack of a sufficient number of spectral bands in the
visible and near-infrared (NIR) range as well as relatively low radiometric sensitivity
(Gordon and Wang 1994; Hu et al. 2012a). In light of these limitations, the data pro-
vided by the CZCS should be used with caution. For example, Tassan (1988) reported
that when CZCS data were used, high concentrations of gelbstoff and suspended
minerals reduced the accuracy of both chlorophyll-a estimations and atmospheric
corrections.
The second generation of ocean-color sensor, Sea-viewing Wide Field-of-View
Sensor (SeaWiFS), was launched in 1997 and ended its mission in 2010 (Table 9.1
and Figure 9.1). The SeaWiFS has four additional bands for ocean observations: two
in the visible range (412 and 490 nm) and two in the NIR region (765 and 865 nm).
The band around 412 nm is useful for separating the detrital and viable phytoplank-
ton signals (Ciotti and Bricaud 2006), and the band centered on 490 nm can assist in
the detection of accessory pigments (Pan et al. 2010). The two NIR bands centered
on 765 and 865 nm contribute to atmospheric correction (Gordon and Wang 1994).
Compared to the CZCS, the radiometric sensitivity of the SeaWiFS was also largely
improved (the mean signal-to-noise ratio [SNR] is between 424 and 790 for six vis-
ible bands; Hu et al. 2012a). However, the spatial and temporal resolutions of the
SeaWiFS are still similar to those of the CZCS.
TABLE 9.1
Summary of Indicative Satellite Sensors Used for Studying Waters
No. of Bands Used
Satellite for Water Quality
Sensors Orbit Active Period Monitoring Resolution (m) Revisit Time Mean SNR
CZCS Sun-synchronous 1978–1986 VIS 4 800 1–2 days 193–220
SeaWiFS Sun-synchronous 1997–2010 VIS 6, NIR 2 1100 1–2 days 424–790 for VIS and 183–219 for
NIR
MODISA Sun-synchronous 2002 to present VIS 7, NIR 2 1000 1–2 days 1366–2401 for VIS and 806–995
for NIR
MERIS Sun-synchronous 2002–2012 VIS 8, NIR 7 300 3 days 513–1088 for VIS and 218–667 for
NIR (excluding band 11 of 54)
GOCI Geostationary 2010 to present VIS 6, NIR 2 500 1 hour 579–833 for VIS and 596–587 for
NIR
TM Sun-synchronous 1984–2011 VIS 3, NIR 1, SWIR 2 30 16 days 29–72 for VIS, 17 for NIR, and
6–10 for SWIR
ETM+ Sun-synchronous 1999 to present VIS 3, NIR 1, SWIR 2 30 16 days 41–78 for VIS, 13 for NIR, and
Monitoring Water Quality with Remote Sensing Image Data

6–10 for SWIR

OLI Sun-synchronous 2013 to present VIS 4, NIR 1, SWIR 2 30 16 days 144–478 for VIS, 67 for NIR, and
14–30 for SWIR

Note: Mean SNR values are obtained from Hu et al. (2012a) except for SNR of OLI (obtained from Franz et al. 2015).
165
166 Remote Sensing for Sustainability

Visible
10
8
6
4
2
0
400 450 500 550 600 650 700
Wavelength (nm)

Near-infrared
10
8
6
4
2
0
700 750 800 850 900 950 1000
Wavelength (nm)

SWIR
10
8
6
4
2
0
1000 1200 1400 1600 1800 2000 2200 2400
Wavelength (nm)

FIGURE 9.1 Location of the spectral bands of several satellite sensors that have been used for
studying waters. 1: GOCI, 3: OLI, 4: ETM+, 5: TM, 6: MERIS, 7: MODIS, 8: SeaWiFS, 9: CZCS.

The Moderate Resolution Imaging Spectroradiometer (MODIS), launched in

December 1999 (aboard the Terra satellite MODIST) and May 2002 (aboard the
Aqua satellite MODISA), is a sensor for both land and water observations. However,
the water-leaving radiance obtained from MODIST is almost unusable for ocean-
color applications owing to the uncertainties and instabilities of the sensor observed
since the start of the mission (Franz et al. 2008). The MODISA has 36 spectral
bands in total, of which 9 bands are used for ocean-color remote sensing (Table 9.1
and Figure 9.1). Generally, the MODISA ocean-color bands are similar to those of
the SeaWiFS, with three bands in blue and two bands in green, but two bands in
red wavelengths are different from those of the SeaWiFS (which has only one red
band). The additional red band of the MODISA (around 678 nm) contributes to the
Monitoring Water Quality with Remote Sensing Image Data 167

detection of chlorophyll fluorescence. The radiometric sensitivity of the MODISA

ocean-color bands were further improved with the mean SNR values of 1366–2401
in the visible range and 806–995 in the NIR region (Hu et al. 2012a). The seven
MODISA land bands are often applied to high-turbidity inland and coastal waters
because of their higher spatial resolutions (250–500 m) and longer-wavelength bands
in the short-wave infrared region (SWIR; Hu 2009; Hu et al. 2010; Wang and Shi
2007).
During a similar active period of the MODIS, the European Space Agency launched
another ocean-color sensor (the Medium Resolution Imaging Spectroradiometer
[MERIS]) in March 2002. This mission was ended in 2012. Compared to the three
ocean-color sensors described above, the MERIS provided higher spectral and spa-
tial resolutions, but slightly lower temporal resolution and radiometric sensitivity
(Table 9.1 and Figure 9.1). The new visible band around 620 nm made the MERIS
a unique sensor that is able to distinguish cyanobacterial blooms from other algal
blooms (Kutser et al. 2006). Another unique band of the MERIS is around 709 nm.
This band has been demonstrated to be very useful for estimating chlorophyll-
a concentrations and detecting algal blooms in turbid inland and coastal waters
(Dall’Olmo et al. 2003; Gilerson et al. 2010; Gitelson et al. 2008; Gower et al. 2005).
Although the spectral characteristics, temporal resolutions, and radiometric sen-
sitivities of the current ocean-color sensors are adequate to monitor water quality,
the relatively coarse spatial resolutions of these data sets (300–1000 m) often limit
many applications in inland waters with small size. Therefore, several land sensors
with relatively fine spatial resolution (30 m) such as the TM (Thematic Mapper), the
ETM+ (Enhanced Thematic Mapper Plus), and the OLI (Operational Land Imager)
aboard Landsat satellites (Table 9.1 and Figure 9.1) have also been used for assess-
ing water quality parameters (e.g., Dekker et al. 2002; Oyama et al. 2009; Tyler et
al. 2006; Wu et al. 2015). In addition to the relatively fine spatial resolution of the
sensors aboard Landsat, another advantage of the use of Landsat data is the longev-
ity of its successive missions (1972 to present; Lobo et al. 2015). However, the wide
bandwidth, infrequent revisit time, and lower radiometric sensitivity of Landsat are
substantial limitations for applying Landsat data to water quality monitoring.
Unlike all of the abovementioned sensors, the Geostationary Ocean Color Imager
(GOCI), launched in June 2010 by the Korea Aerospace Research Institute (KARI),
is the first and so far the only sensor in geostationary orbit for monitoring ocean
color from space (Ryu et al. 2012). As a benefit owing to its geostationary orbit,
GOCI can provide eight images during daylight with 500 m spatial resolution and
higher radiometric sensitivity (>579, Table 9.1). The bands of GOCI are similar to
those of MODIS ocean bands, but it uses less one green band than MODIS, centered
at 531 nm (Figure 9.1). However, its lack of bands around SWIR wavelengths is a
significant limitation for atmospheric correction in turbid inland and coastal waters
(Wang et al. 2012). In addition, the lack of bands around 620 nm makes the GOCI
unable to distinguish cyanobacterial blooms from other algal blooms (Kutser et al.
2006), and its lack of bands around 709 nm makes it difficult to accurately retrieve
the chlorophyll-a concentrations in waters with higher concentrations of colored dis-
solved organic matter (CDOM) and tripton (Gitelson et al. 2008). Another limita-
tion of GOCI is its spatial coverage. GOCI covers only the regions covering Korea,
168 Remote Sensing for Sustainability

China, and Japan, and it cannot provide a global data set because it is designed to
monitor the ocean color around the Korean Peninsula (130°E, 36°N at the center) on
a geostationary orbit (Ryu et al. 2012).
Thus, all past and present ocean-color or land sensors have limitations for
monitoring water quality. The following four aspects of future satellite sensors for
remotely retrieving water quality parameters should be improved: temporal resolu-
tion, spatial resolution, spectral resolution, and radiometric sensitivity (Mouw et
al. 2015). First, higher temporal resolution such as acquiring one or more satel-
lite image per hour during daylight would allow not only observations of fast pro-
cesses of water bodies (e.g., the tidal and diurnal variability of water quality) but
also the accumulation of more available satellite data without cloud contaminations
(Ruddick et al. 2014). This high temporal resolution could be achieved by a geosta-
tionary platform.
Second, finer spatial resolution is required for monitoring water quality in inland
and coastal waters and for detecting algal blooms in open oceans. Bissett et al.
(2004) showed that the optimal spatial resolution for monitoring near-shore waters
(within 200 m of shore) is 100 m. This spatial resolution may also meet the require-
ment of some inland waters larger than 0.09 km 2—corresponding to 9 pixels (i.e.,
100 m × 100 m × 9 pixels)—as Verpoorter et al. (2012) reported that it is diffi-
cult to verify water bodies smaller than 9 pixels relative to image noise. However,
there are numerous water bodies smaller than 0.09 km 2 (Verpoorter et al. 2014)
and water bodies with spatially heterogeneous water quality parameters (Kutser
2004). Further improvements in spatial resolution are thus necessary for these water
bodies.
Third, good spectral resolution is highly desirable for many water-quality remote
sensing studies such as those concerning the retrieval of water quality parame-
ters, the detection of cyanobacterial blooms or Trichodesmium distribution, and
the removal of bottom-reflectance effects (Mouw et al. 2015). However, satellite-
based measurements with too small spectral intervals often face the problems of
how to accurately remove gas absorption effects, how to select appropriate band
for efficient data processing, and how to store a large volume of data (Lee et al.
2014). Lee et al. (2014) found that the remote sensing reflectance of the neighbor-
ing bands is highly correlated when the spectral distance between two neighbor-
ing spectral bands is less than 10 nm, and even when this distance is increased to
20 nm, the spectral information loss is still very limited. Accordingly, they sug-
gested that future ocean-color satellite sensors should have 15 bands space almost
evenly between 400 and 700 nm at 10-nm bandwidth intervals. For the wavelength
beyond 700 nm, at least two NIR bands and two SWIR bands are necessary for
atmospheric correction in clear and turbid waters, respectively (Gordon and Wang
1994; Wang and Shi 2007). In addition, a band around 709 nm is also necessary for
estimating chlorophyll-a concentrations and detecting algal blooms in turbid inland
and coastal waters (Dall’Olmo et al. 2003; Gilerson et al. 2010; Gitelson et al. 2008;
Gower et al. 2005).
Finally, previous studies have stated that the radiometric sensitivity of future
ocean-color sensors should have SNRs larger than 1000 for wavelengths between
350 and 720 nm, larger than 600 for wavelengths between 720 and 900 nm, and
Monitoring Water Quality with Remote Sensing Image Data 169

between 100 and 200 nm for SWIR bands (Hu et al. 2012a; Wang 2007). Since the
SNR is in proportion to the pixel size, the bandwidth for a specific band, and the inte-
gration time, a longer integration time is necessary if we want to ensure high SNRs
as well as fine spatial and spectral resolutions. A sensor on a geostationary satellite
would probably achieve this goal because “the sensor can stare at the desired pixels”
(Lee et al. 2014). A detailed review of the opportunities and challenges presented
by geostationary sensors was provided by Ruddick et al. (2014). Summaries of past,
present, and planned ocean-color sensors such as the Pre-Aerosol, Clouds, and ocean
Ecosystem (PACE) and the Ocean and Land Color Imager (OLCI) can be found in
an appendix of a report published by the International Ocean Colour Coordinating
Group (IOCCG 2012).

9.3 WATER DELINEATION FROM SATELLITE IMAGERY

To monitor water quality from satellite data, the first step is the extraction of the
water body from the data. This step can also serve to detect cloud-contaminated
pixels in water bodies. Many algorithms and techniques have been developed for
delineating water bodies from optical remote sensing data. These methods can be
summarized as belonging to three categories: (1) digitizing water bodies via a visual
assessment of satellite images; (2) using image classification techniques such as
supervised or unsupervised classification procedures, spectral transformation, and
texture analysis; and (3) using a single or multiple algebraically operated bands such
as the band ratio and spectral water index, combined with appropriate threshold
values (Verpoorter et al. 2012). Methods based on the spectral water index have
been widely used to delineate water bodies from satellite imagery, because not
only can this approach efficiently enhance water features while suppressing or even
eliminating non-water features, it is also easy to use and less computationally time-
consuming than other methods.
The first spectral water index devised for delineating water from land is the nor-
malized difference water index (NDWI), which was proposed by McFeeters (1996)
and can be expressed as follows:

NDWI = (dgreen − dNIR)/(dgreen + dNIR), (9.1)

where dgreen and dNIR are the digital number (DN) at green and NIR bands,
respectively. The NDWI values can range from −1 to 1. The selection of the green
band is done to maximize the typical reflectance of water features. The use of NIR
band in the NDWI can not only minimize the water reflectance (strong absorption by
water), but also take advantage of the high reflectance at the NIR band by terrestrial
vegetation and soil features (even higher than the reflectance of the green band).
Accordingly, water features will have positive NDWI values, and land features will
have zero or negative values (McFeeters 1996). Zero was thus set as the threshold to
distinguish water bodies from land by McFeeters (1996) (i.e., if NDWI > 0, then =
water; otherwise, non-water).
Xu (2006) found that the NDWI cannot efficiently suppress the signal from
built-up land, and thus extracted water bodies were still mixed with built-up land
170 Remote Sensing for Sustainability

noise. For example, Figure 9.2a through c show a Landsat5/TM sub-scene for a
part of Wuhan City (China), an NDWI image of the same place, and extracted
water bodies from the NDWI image, respectively, showing that many built-up
lands near Lake Dong (surrounded by a circle in Figure 9.2a) show positive NDWI
values in Figure 9.2b and thus were misclassified as water bodies (white color in
Figure 9.2c).

The Yangtze Lake Sha

Lake Dong

(a) (b)
–1.0

0.0

1.0

(c) (d)
–1.0

0.0

1.0

(e)

FIGURE 9.2 NDWI and MNDWI images for part of Wuhan City, China. (a) Landsat5/TM
image of part of Wuhan City (RGB: 541), acquired on October 24, 2003. (b) NDWI image of
the same area. (c) Extracted water bodies from the NDWI image. (d) MNDWI image of the
same area. (e) Extracted water bodies from the MNDWI image.
Monitoring Water Quality with Remote Sensing Image Data 171

To further suppress these built-up land noises, Xu (2006) proposed a modified

NDWI (MNDWI), which uses a SWIR band instead of the NIR band in the NDWI.
The MNDWI can be expressed as follows:

MNDWI = (dgreen − dSWIR)/(dgreen + dSWIR), (9.2)

where dSWIR is the DN value at a SWIR band such as Landsat5/TM band 5. Figure 9.2d
and e show the MNDWI image for the same Landsat5/TM sub-scene and extracted
water bodies from the MNDWI image, respectively. Compared to the NDWI image
(Figure 9.2b), most of the MNDWI values of the built-up lands became zero or
negative values and thus were removed from the water bodies (Figure 9.2e). This is
because both built-up land and water reflect green light more than they reflect NIR
light, but built-up land also reflects much greater light at the SWIR band than that at
the green band, which is different from water features (Xu 2006). In addition, since
the soil and vegetation still reflect more SWIR light than green light, the MNDWI
can keep negative values for them like the NDWI does.
Although the MNDWI has taken the normalization form, which can reduce shadow
noise (Xu 2006), the MNDWI still suffers from mountain and cloud shadows when a
satellite image includes these types of surfaces. Figure 9.3a through c show a Landsat5/
TM sub-scene of Lake Maninjau (Indonesia), an MNDWI image of the lake and its sur-
rounding area, and the extracted water bodies from the MNDWI image, respectively,
showing that many cloud and mountain shadows are misclassified as water bodies. This
is because spectral features of cloud and mountain shadows at green and SWIR bands
are similar to those of water bodies and thus show positive MNDWI values (Figure 9.4).
Verpoorter et al. (2012) suggested the use of digital elevation model data with the
hillshade algorithm and ArcGIS hydrology tools in ArcMap program to reduce the
noise attributed to cloud and mountain shadows. It was also found that these cloud
and mountain shadowed pixels still show negative NDWI values, indicating that the
NDWI also has the potential to suppress these shadow noises (Figure 9.3d and e).
This is because the background of these shadowed pixels is vegetation (Figure 9.3a),
which reflects more NIR light than SWIR light (Figure 9.4). Table 9.2 shows value
ranges of NDWI and MNDWI for several different land cover types; only water fea-
tures have positive values for both NDWI and MNDWI.
These findings suggest that the combined use of the MNDWI and NDWI can effi-
ciently suppress not only built-up noise but also vegetation shadow noise attributed to
clouds and mountains. Figure 9.3f shows the extracted water bodies by combining the
MNDWI image with the NDWI image (i.e., if the MNDWI > 0 and the NDWI > 0,
then = waters; otherwise, non-waters). The results demonstrate that most of the cloud
and mountain noises were markedly suppressed. However, a misclassification owing to
a different type of cloud remained (white pixels surrounded by a circle in Figure 9.3f).
An example spectrum of this type of cloud is also shown in Figure 9.4 (i.e., cloud
2 in the figure). Similar to the water features, this cloud type showed NDWI and
MNDWI values that were both larger than zero. However, the spectral brightness of
both cloud types is greater than those of other land cover types, and thus it is easy to
differentiate the cloud from other land cover types using a brightness index such as
the Modified Brightness Index proposed by Verpoorter et al. (2012).
172 Remote Sensing for Sustainability

(a) (b) (c)

–1.0

0.0

1.0

(d) (e) (f)

–1.0

1.0
0.0

FIGURE 9.3 MNDWI and NDWI images for Lake Maninjau, Indonesia. (a) Landsat5/TM
image of Lake Maninjau (RGB 541), acquired on September 11, 1989. (b) MNDWI image of
the area. (c) Extracted water bodies from the MNDWI image. (d) NDWI image of the same
area. (e) Extracted water bodies from the NDWI image. (f) Extracted water bodies from the
combination of MNDWI and NDWI images.

Setting an appropriate threshold for the spectral water index is a challenge for
accurately delineating water bodies from other land cover features. McFeeters (1996)
and Xu (2006) set zero as the default threshold value for water delineation. However,
Xu (2006) also found that a more accurate water body extraction could be achieved
if the zero threshold value was adjusted slightly. The variation of threshold value is
mainly attributed to the mixed pixel problem, in which a pixel is considered to con-
tain both water and land features.
Monitoring Water Quality with Remote Sensing Image Data 173

100 300
Cloud shadow
90 Mountain shadow 270
Lake water
80 240
Cloud 1
70 Cloud 2 210

Digital number (cloud)

60 180
Digital number

50 150

40 120

30 90

20 60

10 30

0 0
1 2 3 4 5 7

FIGURE 9.4 Spectral reflectance patterns of cloud shadow, mountain shadow, lake water,
and clouds (the right y axis) obtained from Landsat5/TM image of Lake Maninjau, Indonesia,
acquired on September 11, 1989.

TABLE 9.2
NDWI and MNDWI Value Ranges for Different Land Cover Types
Cloud Mountain
Water Vegetation Soil Built-Up Shadow Shadow
NDWI + 0 or − 0 or − + 0 or − 0 or −
MNDWI + 0 or − 0 or − 0 or − + +

Using simulation data, Ji et al. (2009) extensively investigated how the threshold val-
ues of several spectral water indices (including McFeeters’ NDWI and Xu’s MNDWI)
were influenced by the fractions of soil and vegetation at a given water fraction. They
found that the MNDWI shows the most stable threshold at a given water fraction, and
a zero threshold value was suitable only for extracting pixels with a water fraction
close to 100% for both the NDWI and MNDWI. It can thus be considered that the zero
threshold value of the MNDWI or NDWI probably provides an underestimated water
area compared to the true water body, because the shoreline of a water body is probably
a mix of water and land features owing to the image’s spatial resolution, especially for
the coarse spatial resolution satellite images (e.g., those provided by SeaWiFS, MODIS,
and MERIS). It should also be noted that optical shallow water areas, water areas cov-
ered by aquatic vegetation, and water areas with high turbidity caused by strong algal
blooms and mineral particles can also provide negative values of spectral water indices
and thus influence accurate water body delineations (Verpoorter et al. 2012).
174 Remote Sensing for Sustainability

9.4 ATMOSPHERIC CORRECTION FOR USING SATELLITE DATA

For a water pixel, the optical signal received by a satellite sensor (i.e., the top of
atmosphere [TOA] reflectance, ρtoa) at wavelength λ can generally be obtained by the
following equation (Gordon and Wang 1994):

ρtoa(λ) = ρr(λ) + [ρa(λ) + ρra(λ)] + t(λ)ρw(λ), (9.3)

where ρr(λ) is the reflectance from Rayleigh scattering, [ρa(λ) + ρra(λ)] is the reflec-
tance from the sum of aerosol scattering and the interaction between Rayleigh and
aerosol scattering (i.e., the aerosol multiple-scattering reflectance), t(λ) is the diffuse
transmittances of the atmospheric column, and ρw(λ) is the water-leaving reflectance.
Figure 9.5 shows two examples of in situ–measured water-leaving reflectance and
the corresponding TOA reflectance measured by MERIS in Deadman Bay, Florida
(clear water) and Lake Kasumigaura, Japan (turbid water), respectively. It is clear

0.06

0.05

0.04
In situ measured
Rrs (sr–1)

0.03 MERIS measured

0.02

0.01

0
413 443 490 510 560 620 665 681 709 754
(a) Wavelength (nm)

0.06

0.05

0.04 In situ measured

Rrs (sr–1)

0.03 MERIS measured

0.02

0.01

0
413 443 490 510 560 620 665 681 709 754
(b) Wavelength (nm)

FIGURE 9.5 Examples of in situ–measured water-leaving reflectance and corresponding

TOA reflectance measured by MERIS in Deadman Bay, Florida (a) and Lake Kasumigaura,
Japan (b).
Monitoring Water Quality with Remote Sensing Image Data 175

in these examples that the atmosphere contributes much more signal than the water
body itself. According to the IOCCG (2000, 2010), in most water areas, only 10% of
the signals recorded by the satellite sensors originate from water bodies in the visible
wavelengths (e.g., Figure 9.5a). It is therefore essential to carry out atmospheric cor-
rection before applying the satellite data to quantitative estimations of water quality
parameters.
The key point for atmospheric correction is to estimate the aerosol type and
amount for a given pixel because the Rayleigh-scattering component (ρr) can be pre-
dicted a priori. The most general approach for estimating the aerosol contribution is
to use two specific bands (or reference bands) for which the water-leaving reflectance
(ρw) can be assumed to be zero. The aerosol contribution at these two specific bands
can be estimated as follows:

[ρa(λ1) + ρra(λ1)] = ρtoa(λ1) − ρr(λ1) (9.4)

[ρa(λ2) + ρra(λ2)] = ρtoa(λ2) − ρr(λ2). (9.5)

Two specific bands are required because we need to estimate not only the magni-
tude of the aerosol’s contribution but also its dependence on wavelength for extrapo-
lating (or interpolating) the aerosol contribution to the other shorter bands. For this,
the estimated aerosol multiple scatterings must first be converted to aerosol single
scatterings (ρas) at the same bands through a look-up table (LUT) as follows (Gordon
and Wang 1994):

[ρa(λ1) + ρra(λ1)] → ρas(λ1) (9.6)

[ρa(λ2) + ρra(λ2)] → ρas(λ2). (9.7)

The atmospheric correction parameter epsilon (i.e., ε(λ1, λ2)) is then calculated
as follows:

ε(λ1, λ2) = ρas(λ1)/ρas(λ2). (9.8)

The values of ε(λ1, λ2) correspond to different aerosol models. For a given aero-
sol model, epsilon values for other bands can be extrapolated using the following
equation:

ε(λ, λ2) = exp[ln(ε(λ1, λ2))/(λ2 − λ1)*(λ2 − λ)]. (9.9)

Aerosol single scatterings for all wavelengths (ρas(λ)) can then be estimated using
the following equation:

ρas(λ) = ρas(λ2)*ε(λ, λ2). (9.10)

176 Remote Sensing for Sustainability

The obtained aerosol single scatterings can be reconverted to the correspond-

ing aerosol multiple scatterings for all wavelengths through the LUT prepared in
advance:

ρas(λ) → [ρa(λ) + ρra(λ)]. (9.11)

Finally, the water-leaving reflectance (ρw(λ)) at all wavelengths can be derived as

follows:

ρw(λ) = (ρtoa(λ) − {ρr(λ) + [ρa(λ) + ρra(λ)]})/t(λ). (9.12)

For clear waters (e.g., open oceans, or Figure 9.5a), the water-leaving reflec-
tance at NIR bands can be assumed to be zero owing to the very high absorp-
tion of pure water and little or no contribution from suspended particles in these
bands. Generally, SeaWiFS bands 7 and 8, MODIS bands 15 and 16, MERIS
bands 12 and 13, and GOCI bands 7 and 8 can be used as two specific bands
(i.e., λ1 and λ 2) for the purpose of atmospheric correction. However, for turbid
waters (e.g., some inland and coastal water, or Figure 9.5b), these two specific
NIR bands should be shifted to longer wavelengths such as SWIR bands because
high concentrations of particulate matters make the water-leaving reflectance in
the NIR bands nonnegligible (Wang and Shi 2005). At the SWIR bands, even for
turbid water, water-leaving reflectance can still be assumed to be zero because
the absorption of pure water at these wavelengths is much higher than that at NIR
bands (Shi and Wang 2009).
Unfortunately, there is no existing satellite sensor with SWIR bands for ocean-
color remote sensing. Although the MODIS land bands at SWIR wavelengths (e.g.,
bands 5 and 7) have shown potential for atmospheric correction over highly tur-
bid water (Wang and Shi 2007), SNR values of the two SWIR bands that are too
low make atmospherically corrected water-leaving reflectance at visible bands quite
noisy (Hu et al. 2012a; Wang and Shi 2012).
Another approach for solving the atmospheric correction problem over turbid
waters is to estimate the water-leaving reflectance at two NIR reference bands by a
bio-optical model instead of assuming them to be zero (e.g., Bailey et al. 2010; Hu
et al. 2000; Jaelani et al. 2015; Stumpf et al. 2003; Wang et al. 2012). The estimated
water-leaving reflectance values at the two NIR reference bands are then removed
from the corresponding TOA reflectance to meet Equations 9.4 and 9.5. This
approach is available for the satellite sensors without SWIR bands (e.g., SeaWiFS,
MERIS, and GOCI).
A summary of existing atmospheric correction algorithms for turbid waters can
be found in papers by the IOCCG (2010) and Jaelani et al. (2013). Most of these algo-
rithms have been integrated into widely used software packages such as SeaDAS
(SeaWiFS Data Analysis System; http://seadas.gsfc.nasa.gov/) and BEAM (Basic
ERS & ENVISAT AATSR and MERIS; http://www.brockmann-consult.de/cms
/web/beam), which are easily available and well-supported platforms.
Monitoring Water Quality with Remote Sensing Image Data 177

9.5 ESTIMATION OF WATER QUALITY PARAMETERS

FROM REMOTE SENSING DATA
9.5.1 Bio-optical Model
The fundamental relationship for retrieving water quality parameters is the bio-optical
model, which can be expressed by the following equations (Gordon et al. 1988):

rrs(λ) = g0μ(λ) + g1μ(λ)2 (9.13)

μ(λ) = bb(λ)/[a(λ) + bb(λ)], (9.14)

where g0 = 0.089, g1 = 0.125, and a(λ) and bb(λ) are the total absorption and
backscattering coefficients, respectively. rrs(λ) is the subsurface remote sensing
reflectance, which can be calculated from the above-surface remote sensing
reflectance (Rrs(λ)) as follows (Lee et al. 2002):

rrs(λ) = Rrs(λ)/[0.52 + 1.7 Rrs(λ)]. (9.15)

a(λ) and bb(λ) can be further expressed as the sum of the water constituents’ absorp-
tion and the backscattering coefficients, respectively, as follows (Mobley 1994):

a(λ) = aw(λ) + aph(λ) + adg(λ) (9.16)

bb(λ) = bbw(λ) + bbp(λ), (9.17)

where aw(λ), aph(λ), and adg(λ) are the absorption coefficients of pure water,
phytoplankton, and the combination of detritus (often referred to as aNAP) and
gelbstoff (often referred to as a CDOM), respectively, and bbw(λ) and bbp(λ) are the
backscattering coefficients of pure water and particulate matter, respectively. The
parameter bbp(λ) is often further expressed as the sum of backscattering coefficients
of phytoplankton (bbph(λ)) and nonalgal particles (bbNAP(λ)).

9.5.2 estiMations of the BiogeocheMical paraMeters of Water Bodies

Mouw et al. (2015) recommended a minimum list for desired biogeochemical
parameters from remote sensing: the concentrations of chlorophyll-a (Chl-a), total
suspended matter, particulate organic matter, particulate inorganic matter, dissolved
organic matter, and dissolved inorganic matter. Algorithms for estimating Chl-a
have been well developed (Matthews 2011; Odermatt et al. 2012).
Two types of algorithms have been widely used for estimating Chl-a from remote
sensing data: band ratio–based and baseline-based algorithms (Table 9.3). The band
ratio–based algorithm can be generally expressed as

Chl-a = f(band ratio), (9.18)

 178

TABLE 9.3
Summary for Several Representative Algorithms for Estimating Chlorophyll-a (Chl-a) Concentration
Band Combination
Category Algorithm (Band Center, nm) Advantage Disadvantage References
Band ratio Generally Can reduce effects from non- Intolerance to spectrally related errors of
based phytoplankton matters remote sensing reflectance (e.g.,
instrument noise, imperfect
atmospheric correction)
OCx 443, 490, 510, 555 Suitable for low Chl-a between 0.3 Intolerance to optically complex waters O’Reilly et al. 1998
and 10 mg m−3
Two-band model 665, 709 Suitable for moderately turbid waters Other turbid waters Gilerson et al. 2010
(e.g., remote sensing reflectance at
865 nm between 0.0001 and 0.001)
Three-band model 665, 709, 754 Suitable for very turbid waters (e.g., Other turbid waters Gitelson et al. 2008
remote sensing reflectance at
865 nm between 0.001 and 0.01)
SAMO-LUT 550, 665, 709, 754 Suitable for very turbid and even for Other turbid waters Yang et al. 2011
extremely turbid waters
Baseline Generally Tolerance to spectrally related errors Intolerance to optically complex waters
based of remote sensing reflectance
CIA 443, 555, 670 Suitable for very low Chl-a (≤0.25 mg Other Chl Hu et al. 2012b
m−3)
FLH 665, 681, 709 Suitable for moderate Chl-a (10–30 Other Chl Gower 1980
mg m−3)
MCI 681, 709, 754 Suitable for high Chl-a (>25 mg m−3) Other Chl Gower et al. 2005
FAI 667, 859, 1240 Suitable for extremely high Chl-a Other Chl Hu 2009
(e.g., surface scums)
Remote Sensing for Sustainability
Monitoring Water Quality with Remote Sensing Image Data 179

where f represents a function that can be linear or nonlinear. For clear or optically
simple waters (e.g., open oceans), the band ratio of Rrs at blue and green bands is
widely used for estimating Chl-a (e.g., OCx form-based algorithms; O’Reilly et al.
1998). This is because the Rrs at the blue band is strongly related to chlorophyll
absorption, whereas at the green band, it is relatively insensitive to changes in Chl-a
values owing to the weak absorption of chlorophyll at this band, and it can thus be
used as a reference.
However, for optically complex waters with moderate turbidities, the Rrs values in
the red and NIR bands are preferred for the retrieval of Chl-a (Dekker 1993; Gitelson
and Kondratyev 1991; Han and Rundquist 1997). This is because the Rrs values at
the blue and green bands are affected not only by phytoplankton but also by other
constituents (i.e., NAP and CDOM) (Gitelson et al. 2009). The use of the band ratio
of Rrs(NIR)/Rrs(red) can minimize the effects from NAP and CDOM based on four
assumptions: (i) at the red band, aph + aw >> aNAP + aCDOM; (ii) at the NIR band, aw >>
aph + aNAP + aCDOM; (iii) at both the red and NIR bands, a >> bb; and (iv) for backscat-
tering coefficients, bb(red) ≈ bb(NIR) (Gitelson et al. 2008). According to the above
assumptions, the Rrs(NIR)/Rrs(red) can be approximated as

Rrs(NIR)/Rrs(red) ≈ [aph(red) + aw(red)]/aw(NIR). (9.19)

Since aw(λ) can be considered constant, the Rrs(NIR)/Rrs(red) value varies only
with aph(red), which is strongly related to Chl-a. For example, Gilerson et al. (2010)
proposed a two-band model, which uses the band ratio of Rrs at 709 and 665 nm (i.e.,
Rrs(709)/Rrs(665)) to estimate Chl-a in moderately turbid waters.
For optically complex waters with higher turbidities, the above assumptions
from (i) to (iii) will become invalid (Dall’Olmo et al. 2005; Gitelson et al. 2008).
Therefore, a three-band index that uses one Rrs at the red band (around 665 nm)
and two Rrs at NIR bands (around 709 and 754 nm) was designed to address the
effects of NAP and CDOM in these waters (Dall’Olmo et al. 2003; Gitelson et al.
2008). First, the Rrs around the wavelengths 665 and 709 nm were inverted, and their
difference was calculated to minimize the absorption effects of NAP and CDOM
(i.e., Rrs−1 (665) − Rrs−1 (709)). This step is based on two assumptions: (i) aNAP(665) ≈
aNAP(709) and aCDOM(665) ≈ aCDOM(709), and (ii) bb(665) ≈ bb(709). Therefore,
Rrs−1 (665) − Rrs−1 (709) can be approximated as [aph(665) + aw(665) − aph(709) − aw(709)]/bb.
Then, the R rs at the third wavelength around 754 nm is used to further minimize
the effects by bb. This step is based on another two assumptions: (i) bb(665) ≈ bb(709) ≈
bb(754), and (ii) aw(754) >> aph(754) + aNAP(754) + aCDOM(754) + bb(754). The three-band
index is thus designed as  Rrs−1 (665) − Rrs−1 (709)  × Rrs (754), and can be approximated
according to the above assumptions as

 Rrs−1 (665) − Rrs−1 (709)  × Rrs (754) ≈ [aph (665) + aw (665)

− aph (709) − aw (709)]/aw (754).

(9.20)
180 Remote Sensing for Sustainability

Similar to the two-band index (Equation 9.19), the three-band index varies only
with aph and thus can provide accurate Chl-a estimations in more turbid waters.
However, the assumption of aw(754) >> aph(754) + aNAP(754) + aCDOM(754) +
bb(754) will be invalid in extremely turbid waters (Le et al. 2009; Yang et al. 2010).
A four-band index (Le et al. 2009), an enhanced three-band index (Yang et al. 2010),
and the Semi-Analytical Model Optimizing and Look-Up Tables (SAMO-LUT)
algorithm (Yang et al. 2011) were then proposed for this type of waters. Since the
independent variable in the SAMO-LUT algorithm is still a three-band index, the
SAMO-LUT algorithm is considered a type of band ratio–based algorithm.
The second type of Chl-a estimation algorithm, that is, the baseline-based algo-
rithms, can be generally expressed as

Chl-a = f(RH), (9.21)

where RH is the relative height of a reflectance peak (or trough, Rrs(λ2)) from a
baseline formed linearly between Rrs values at two adjacent bands (i.e., Rrs(λ1) and
Rrs(λ3) with λ1 < λ2 < λ3):

RH = Rrs(λ2) − [Rrs(λ1) + (λ2 − λ1)/(λ3 − λ1)*(Rrs(λ3) − Rrs(λ1))]. (9.22)

A number of baseline-based algorithms have been proposed for estimating Chl-a

in waters with various trophic states (Table 9.3). For example, Hu et al. (2012b) pro-
posed a color index–based algorithm (CIA) for ultra-oligotrophic oceans; Gower
(1980) developed a fluorescence line height algorithm for mesotrophic and eutrophic
waters; Gower et al. (2005) and Hu (2009) suggested a maximum chlorophyll index
(MCI) and a floating algae index, respectively, for hypertrophic waters.
Generally, the tolerance to spectrally related errors of Rrs (e.g., instrument noise
and imperfect atmospheric correction) is a great advantage of these baseline-based
algorithms (Hu 2009; Hu et al. 2012b). However, the baseline-based algorithms often
suffer from effects of non-phytoplankton matter (e.g., NAP and CDOM). In contrast,
some band ratio–based algorithms can reduce the effects from NAP and CDOM,
but they are intolerant to spectrally related errors of Rrs. This is probably because of
their form designs (Equations 9.19, 9.20, and 9.22; Hu et al. 2012b). Table 9.3 gives a
summary of several representative algorithms for Chl-a estimation in various waters.
Other algorithms for estimating water quality parameters are described in two
comprehensive reviews (Matthews 2011; Odermatt et al. 2012).

9.5.3 estiMations of the inherent optical properties of Water Bodies

The recommended standard products of inherent optical properties (IOPs) from
remote sensing are the total absorption coefficient (a(λ)), the absorption coefficient
of each constituent (i.e., aph(λ), aNAP(λ), and aCDOM(λ)), and the backscattering
coefficient of particulate matters (bbp(λ)) (Mouw et al. 2015). The bbp(λ) is usually
expressed as follows (Lee et al. 2002):

bbp(λ) = bbp(λ0)(λ0/λ)Y, (9.23)

Monitoring Water Quality with Remote Sensing Image Data 181

where λ 0 is the reference wavelength and Y is the spectral slope. For clear waters,
λ 0 = 555 nm was used (Lee et al. 2002). However, for turbid waters, λ 0 = 750 nm
was suggested (Yang et al. 2013). From Equations 9.14 and 9.17, b bp(λ 0) can be
expressed as

bbp(λ0) = μ(λ0)a(λ0)/(1 − μ(λ0)) − bbw(λ0). (9.24)

For clear waters, a(λ0) can be obtained from an empirical relationship, which is
the function of rrs at 443, 490, 555, and 667 nm (Lee et al. 2002). For turbid waters,
since λ0 = 750 nm, a(λ0) ≈ aw(λ0) can be assumed because the water absorption at
this wavelength is much larger than that from other constituents (Yang et al. 2013).
The spectral slope Y can be estimated by an empirical equation for clear waters
(Equation 9.25; Lee et al. 2002) and a semi-analytical equation for turbid waters
(Equation 9.26; Yang et al. 2013):

Y = 2.2(1 − 1.2exp(−0.9rrs(443)/rrs(555))) (9.25)

Y = −372.99(log[μ(750)/μ(780)])2 + 37.286(log[μ(750)/μ(780)]) + 0.84. (9.26)

Then, the total absorption coefficient can be obtained as

a(λ) = (1 − μ(λ))*(bbw(λ) + bbp(λ))/μ(λ). (9.27)

It is still much more challenging to separate the absorption coefficient of each

constituent from the total absorption coefficient because more assumptions are
required. A detailed review of IOP retrievals was provided by the IOCCG (2006).

9.5.4 estiMations of the apparent optical properties of Water Bodies

Except for Rrs(λ), the recommended standard products of apparent optical properties
from remote sensing are the diffuse attenuation coefficient (Kd(λ)) and the euphotic
zone depth (Z eu) (Mouw et al. 2015). Both of these parameters are of great importance
in studying and modeling the physical, chemical, and biological processes in water
bodies (Kirk 1994; Mobley 1994). Semi-analytical algorithms have been proposed
to remotely estimate Kd(λ) and Z eu based on the retrievals of the total absorption and
backscattering coefficients (i.e., a and bb; Lee et al. 2005a, 2007).
Kd(λ) is commonly defined as the vertically averaged value of the spectral diffuse
attenuation coefficient at depth z (i.e., Kd(λ, z)) over the water layer (Lee et al. 2005a).
Kd(λ, z) is defined as follows (Gordon et al. 1980):

Kd(λ, z) = −(1/Ed(λ, z))*(dEd(λ, z)/dz), (9.28)

where Ed(λ, z) is the downwelling irradiance at depth z. The maximum depth for
vertically averaging Kd(λ, z) values is generally defined as the depth where the
downwelling irradiance is reduced to 10% of the surface irradiance (Lee et al. 2005a).
182 Remote Sensing for Sustainability

A general model of Kd(λ) was suggested by Lee et al. (2005a) through numerical
simulations of radiative transfer in water bodies:

Kd(λ) = (1 + 0.005θa) a(λ) + 4.18(1 − 0.52e−10.8 a(λ)) bb(λ), (9.29)

where θa is the solar zenith angle in air, which can be derived from the location and
measuring time.
Z eu is practically defined as the depth at which photosynthetic available radiation
(PAR) is 1% of its surface value (Kirk 1994). In other words, if the vertical transmit-
tance of the PAR at a depth z (i.e., TPAR(z)) reaches 1%, this depth is the Z eu. TPAR(z)
can be derived as follows (Lee et al. 2007):

TPAR(z) = exp[−Kd(PAR, z)*z], (9.30)

where Kd(PAR, z) is the attenuation coefficient of PAR between the water surface
and the depth z, which can be estimated from a semi-analytical model (Lee et al.
2005b):

Kd(PAR, z) = K1 + K2/(1 + z)0.5, (9.31)

where K1 and K2 are functions of the total absorption and backscattering coefficients
at 490 nm (i.e., a(490) and bb(490)) as well as the solar zenith angle (θa):

K1 = [−0.057 + 0.482 a(490) + 4.221b b (490)][1 + 0.09 sin(θa)] (9.32)

K 2 = [0.183 + 0.702a(490) − 2.567b b (490)][1.465 − 0.667 cos(θa)] . (9.33)

Since a(λ) and bb(λ) can be derived from Rrs (Lee et al. 2002; Yang et al. 2013),
the Kd(λ) and Z eu values can also be remotely derived. The above models (Equations
9.29 through 9.33) have been successfully applied to both clear oceanic waters (Lee
et al. 2005b, 2007) and turbid inland waters (Yang et al. 2014, 2015).

9.6 A HYBRID APPROACH FOR DEVELOPING

A UNIVERSAL ALGORITHM
A universal algorithm should have the ability to address various water types ranging
from clear ultra-oligotrophic to turbid hypertrophic systems. However, previous
studies have shown that each algorithm has its strengths and limitations for different
water types. For example, Wang and Shi (2007) reported that the standard NIR
atmospheric correction algorithm can provide good-quality ocean-color products for
the global open oceans and offshore waters, but the SWIR atmospheric correction
algorithm should be used for turbid coastal waters to produce similar quality. Hu
Monitoring Water Quality with Remote Sensing Image Data 183

et al. (2012b) reported that the CIA can only be applied to estimate Chl-a for ultra-
oligotrophic oceans with Chl-a concentrations lower than 0.25 mg m−3, whereas for
oceanic waters with Chl-a concentrations higher than 0.3 mg m−3, the OC4 algorithm
should be used. Matsushita et al. (2015) also reported that a blue-green algorithm
such as OC4 can be used in optically simple lakes, but a red-NIR algorithm such as
the two-band model or a three-band model must be used for optically complex turbid
lakes to minimize effects from non-phytoplankton matter. These findings indicate
that a hybrid approach is desirable to address waters with various optical properties.
Figure 9.6 illustrates a conceptual diagram for a hybrid approach. In addition
to candidate estimation algorithms for atmospheric correction and water quality
parameters (e.g., Chl-a), another type of algorithm is required for selecting the most
appropriate algorithm for different water types. Several studies have attempted to
address this issue. For example, Wang and Shi (2007) proposed the use of a turbid
water index to identify productive or turbid waters for which the SWIR atmospheric
correction should be used instead of the standard NIR atmospheric correction algo-
rithm. Gomez et al. (2011) developed two normalized difference indices, that is,
[Rrs(560) − Rrs(443)]/[Rrs(560) + Rrs(443)] < 0 and [Rrs(709) − Rrs(665)]/[Rrs(709) +
Rrs(665)] > 0 for first separating Mediterranean lakes of the European Union into
two water types, and then different Chl-a retrieval algorithms are selected for each

toa( )

Switching AC
algorithm

WT 1 WT 2 WT n

AC algorithm 1 AC algorithm 2 AC algorithm n

w( )

Switching Chl-a
algorithm

WT 1 WT 2 WT n
Chl-a estimation Chl-a estimation Chl-a estimation
algorithm 1 algorithm 2 algorithm n

Chl-a

FIGURE 9.6 Conceptual diagram for a hybrid approach. AC, atmospheric correction; Chl-a,
chlorophyll-a concentration; WT, water type.
184 Remote Sensing for Sustainability

water type to obtain the most accurate Chl-a concentrations for the whole study area.
Moore et al. (2014) proposed an optical water type (OWT) framework for selecting
and blending two Chl-a retrieval algorithms. Matsushita et al. (2015) suggested a
hybrid algorithm that uses an MCI to switch three different types of Chl-a estimation
algorithms for clear, moderate-turbid, and high-turbid waters, respectively. These
attempts indicate that the hybrid approach has significant potential toward the devel-
opment of a universal algorithm for monitoring water quality with remote sensing
image data.

9.7 CONCLUSIONS
The remote sensing technique has been widely used for detecting, mapping, and
quantifying water bodies. In this chapter, we introduced algorithms that can be used
to obtain water quality data from remote sensing, and we described major suitable
satellite sensors, water body delineation, atmospheric correction, and water quality
estimation. These remotely obtained water quality data may be useful for meeting
Sustainable Development Goal 6.3 proposed by the United Nations Open Working
Group. Although the usefulness of remote sensing data is widely recognized, a few
cases remain in which these data have been operationally used in water resource
management. Difficulties still exist regarding both the hardware and software of
remote sensing systems used to obtain additional scientific data for studies of the
sustainable use of water resources. For the hardware, a more powerful satellite
sensor with sufficient spatial resolution, temporal resolution, spectral resolution,
and radiometric sensitivity is necessary for collecting more useful information. For
the software, a hybrid algorithm that combines the advantages of each individual
algorithm is needed for a universal application. Nevertheless, the remote sensing
technique will remain a powerful tool in sustainability studies by providing images
of Earth on a daily basis.

ACKNOWLEDGMENTS
This research was supported in part by JSPS and LIPI under the JSPS-LIPI Joint
Research Program (BDB28034), the Grants-in-Aid for Scientific Research of MEXT
from Japan (No. 25420555), and also by the Environment Research and Technology
Development Fund (S-9-4-(1)) of the Ministry of the Environment, Japan. The
authors also would like to thank the anonymous reviewers for their valuable com-
ments on the manuscript.

REFERENCES
Ayres, W., Busia, A., Hirji, R., Lintner, S., McCalla, A., & Robelus, R., 1996. Integrated
lake and reservoir management: World Bank Approach and Experience, World Bank
Technical Paper, 358, World Bank, Washington.
Bailey, S. W., Franz, B. A., & Werdell, P. J., 2010. Estimation of near-infrared water-leaving
reflectance for satellite ocean color data processing. Optics Express, 18, 7521–7527.
Bissett, W. P., Amone, R. A., Davis, C. O., Dickey, T. D., Dye, D., Kohler, D. D. et al., 2004.
From meters to kilometers. Oceanography, 17, 32–42.
Monitoring Water Quality with Remote Sensing Image Data 185

Ciotti, A. M., & Bricaud, A., 2006. Retrievals of a size parameter for phytoplankton and spec-
tral light absorption by colored detrital matter from water-leaving radiances at SeaWiFS
channels in a continental shelf region off Brazil. Limnology and Oceanography:
Methods, 4, 237–253.
Dall’Olmo, G., Gitelson, A., & Rundquist, D. C., 2003. Towards a unified approach for remote
estimation of chlorophyll-a in both terrestrial vegetation and turbid productive waters.
Geophys. Res. Lett., 30.
Dall’Olmo, G., Gitelson, A., Rundquist, D. C., Leavitt, B., Barrow, T., & Holz, J. C.,
2005. Assessing the potential of SeaWiFS and MODIS for estimating chlorophyll
concentration in turbid productive waters using red and near-infrared bands. Remote
Sensing of Environment, 96, 176–187.
Dekker, A. G., 1993. Detection of water quality parameters for eutrophic waters by high
resolution remote sensing. Ph.D. dissertation, Vrije Universiteit, Amsterdam, The
Netherlands.
Dekker, A. G. & Hestir, E. L., 2012. Evaluating the feasibility of systematic inland water
quality monitoring with satellite remote sensing, CSIRO: Water for a Healthy Country
National Research Flagship.
Dekker, A. G., Vos, R. J., & Peters, S. W. M., 2002. Analytical algorithms for lake water
TSM estimation for retrospective analyses of TM and SPOT sensor data. International
Journal of Remote Sensing, 23, 15–35.
Franz, B., Bailey, S. W., Kuring, N., & Werdell, P. J., 2015. Ocean color measurements with
the Operational Land Imager on Landsat-8: Implementation and evaluation in SeaDAS.
Journal of Applied Remote Sensing, 9, 096070, 1–16.
Franz, B., Kwiatowska, E. J., Meister, G., & McClain, C. R., 2008. Moderate resolution imag-
ing spectroradiometer on Terra: Limitations for ocean color applications. Journal of
Applied Remote Sensing, 2, 023525, 1–17.
Gilerson, A., Gitelson, A., Zhou, J., Gulrin, D., Moses, W., Ioannou, I., & Ahmed, S., 2010.
Algorithms for remote sensing of chlorophyll-a in coastal and inland waters using red
and near infrared bands. Optics Express, 18, 24109–24125.
Gitelson, A., Dall’Olmo, G., Moses, W., Rundquist, D. C., Barrow, T., Fisher, T. R. et al.,
2008. A simple semi-analytical model for remote estimation of chlorophyll-a in turbid
waters: Validation. Remote Sensing of Environment, 112, 3582–3593.
Gitelson, A., Gurlin, D., Moses, W., & Barrow, T., 2009. A bio-optical algorithm for the remote
estimation of the chlorophyll-a concentration in case-2 waters. Environment Research
Letters, 4 (045003), 5.
Gitelson, A., & Kondratyev, K. Y., 1991. On the mechanism of formation of maximum in
the reflectance spectra near 700 nm and its application for remote monitoring of water
quality. Trans. Doklady USSR Acad. Sci.: Earth Sci. Sections 306, 1–4.
Gomez, J. A. D., Alonso, C. A., & Garcia, A. A., 2011. Remote sensing as a tool for monitoring
water quality parameters for Mediterranean Lakes of European Union water framework
directive (WFD) and as a system of surveillance of cyanobacterial harmful algae blooms
(SCyanoHABs). Environ. Monit. Assess., 181, 317–334.
Gordon, H., Brown, O., Evans, R., Brown, R., Smith, R., Baker, K. et al., 1988. A semi-
analytic radiance model of ocean color. Journal of Geophysical Research, 93,
10909–10924.
Gordon, H. R., Smith, R. C., & Zaneveld, J. R. V., 1980. Introduction to ocean optics. Proc.
SPIE Soc. Pot. Eng., 6, 1–43.
Gordon, H. R., & Wang, M., 1994. Retrieval of water-leaving radiance and aerosol optical
thickness over the oceans with SeaWiFS: A preliminary algorithm. Applied Optics, 33,
443–452.
Gower, J., 1980. Observations of in-situ fluorescence of chlorophyll a in Saanich Inlet.
Boundary-Layer Meteorology, 18, 235–245.
186 Remote Sensing for Sustainability

Gower, J., King, S., Borstad, G., & Brown, L., 2005. Detection of intense plankton blooms
using the 709 nm band of the MERIS imaging spectrometer. International Journal of
Remote Sensing, 26, 2005–2012.
Haddeland, I., Heinke, J., Biemans, H., Eisner, S., Flörke, M., Hanasaki, N., Konzmann, M.,
Ludwig, F., Masaki, Y., Schewe, J., Stacke, T., Tessler, Z. D., Wada, Y., & Wisser, D.,
2014. Global water resources affected by human interventions and climate change.
PNAS, 111, 3251–3256.
Han, L. H., & Rundquist, D. C., 1997. Comparison of NIR/RED ratio and first derivative
of reflectance in estimating algal-chlorophyll concentration: A case study in a turbid
reservoir. Remote Sensing of Environment, 62, 253–261.
Hovis, W. A., Clark, D. K., Anderson, F., Austin, R. W., Wilson, W. H., Baker, E. T., Ball, D.,
Gordon, H. R., Mueller, J. L., Al-Sayed, S. Z., Sturm, B., Wrigley, R. C., & Yentsch,
C. S., 1980. Nimbus-7 Coastal Zone Color Scanner: System description and initial
imagery. Science, 210, 60–63.
Hu, C., 2009. A novel ocean color index to detect floating algae in the global oceans. Remote
Sensing of Environment, 113, 2118–2129.
Hu, C., Carder, K. L., & Muller-Karger, F. E., 2000. Atmospheric correction of SeaWiFS
imagery over turbid coastal waters: A practical method. Remote Sensing of Environment,
74, 195–206.
Hu, C., Feng, L., Lee, Z., Davis, C. O., Mannino, A., McClain, C. R. et al., 2012a. Dynamic
range and sensitivity requirements of satellite ocean color sensors: Learning from the
past. Applied Optics, 51, 6045–6062.
Hu, C., Lee, Z., & Franz, B., 2012b. Chlorophyll a algorithms for oligotrophic oceans: A novel
approach based on three-band reflectance difference. Journal of Geophysical Research,
117, C01011 (doi:01010.01029/02011JC007395).
Hu, C. et al., 2010. Moderate Resolution Imaging Spectroradiometer (MODIS) observations
of cyanobacteria blooms in Taihu Lake, China. Journal of Geophysical Research—
Oceans, 115, C04002.
IOCCG, 2000. Remote sensing of ocean colour in coastal and other optically-complex waters.
In S. Sathyendranath (Ed.), Report of the International Ocean-Colour Coordinating
Group, No. 3. Dartmouth, Canada: IOCCG.
IOCCG, 2006. Remote sensing of inherent optical properties: Fundamentals, tests of algo-
rithms, and applications. In Z. P. Lee (Ed.), No. 5. Dartmouth, Canada: IOCCG.
IOCCG, 2010. Atmospheric correction for remotely-sensed ocean-colour products. In M.
Wang (Ed.), Reports of the International Ocean-Colour Coordinating Group, No. 10.
Dartmouth, Canada: IOCCG.
IOCCG, 2012. Mission Requirements for Future Ocean-Colour Sensors. In McClain, C. R.
and Meister, G. (Eds.), Reports of the International Ocean-Colour Coordinating Group,
No. 13. Dartmouth, Canada: IOCCG.
Jaelani, L. M., Matsushita, B., Yang, W., & Fukushima, T., 2013. Evaluation of four MERIS
atmospheric correction algorithms in Lake Kasumigaura, Japan. International Journal
of Remote Sensing, 34, 8967–8985.
Jaelani, L. M., Matsushita, B., Yang, W., & Fukushima, T., 2015. An improved atmospheric
correction algorithm for applying MERIS data to very turbid inland waters. International
Journal of Applied Earth Observation and Geoinformation, 39, 128–141.
Ji, L., Zhang, L., & Wylie, B., 2009. Analysis of dynamic thresholds for the normalized
difference water index. Photogrammetric Engineering & Remote Sensing, 75,
1307–1317.
Kiefer, I., Odermatt, D., Anneville, O., Wuest, A., & Bouffard, D., 2015. Application of
remote sensing for the optimization of in-situ sampling for monitoring of phytoplankton
abundance in a large lake. Science of the Total Environment, 527–528, 493–506.
Monitoring Water Quality with Remote Sensing Image Data 187

Kirk, J. T. O., 1994. Light and Photosynthesis in Aquatic Ecosystems. Cambridge: University
Press.
Kutser, T., 2004. Quantitative detection of chlorophyll in cyanobacterial blooms by satellite
remote sensing. Limnology and Oceanography, 49, 2179–2189.
Kutser, T., Metsamaa, L., Strombeck, N., & Vatmae, E., 2006. Monitoring cyanobacterial
blooms by satellite remote sensing. Estuarine, Coastal and Shelf Science, 67,
303–312.
Le, C., Li, Y., Zha, Y., Sun, D., Huang, C., & Lu, H., 2009. A four-band semi-analytical model
for estimating chlorophyll a in highly turbid lakes: The case of Taihu Lake, China.
Remote Sensing of Environment, 113, 1175–1182.
Lee, Z. P., Carder, K. L., & Arnone, R., 2002. Deriving inherent optical properties from
water color: A multi-band quasi-analytical algorithm for optically deep waters. Applied
Optics, 41, 5755–5772.
Lee, Z. P., Darecki, M., Carder, K. L., Davis, C. O., Stramski, D., & Rhea, W. J., 2005b.
Diffuse attenuation coefficient of downwelling irradiance: An evaluation of remote sens-
ing methods. Journal of Geophysical Research, 110, C2, C02017.
Lee, Z. P., Du, K., & Arnone, R., 2005a. A model for the diffuse attenuation coefficient of
downwelling irradiance. Journal of Geophysical Research, 110, C2, C02016.
Lee, Z. P., Shang, S., Hu, C., & Zibordi, G., 2014. Spectral interdependence of remote-sensing
reflectance and its implications on the design of ocean color satellite sensors. Applied
Optics, 53, 3301–3310.
Lee, Z. P., Weidemann, A., Kindle, J., Arnone, R., Carder, K. L., & Davis, C., 2007. Euphotic
zone depth: Its derivation and implication to ocean-color remote sensing. Journal of
Geophysical Research, 112, C03009.
Lobo, F. L., Costa, M. P. F., & Novo, E. M. L. M., 2015. Time-series analysis of Landsat-
MSS/TM/OLI images over Amazonian waters impacted by gold mining activities.
Remote Sensing of Environment, 157, 170–184.
Matsushita, B., Yang, W., Yu, G., Oyama, Y., Yoshimura, K., & Fukushima, T., 2015. A
hybrid algorithm for estimating the chlorophyll-a concentration across different trophic
states in Asian inland waters. ISPRS Journal of Photogrammetry and Remote Sensing,
102, 28–37.
Matthews, M., 2011. A current review of empirical procedures of remote sensing in inland
and near-coastal transitional waters. International Journal of Remote Sensing, 32,
6855–6899.
McFeeters, S. K., 1996. The use of normalized difference water index (NDWI) in the
delineation of open water features. International Journal of Remote Sensing, 17,
1425–1432.
Mobley, C., 1994. Light and water: Radiative Transfer in Natural Waters. New York: Academic
Press.
Moore, T. S., Dowell, M. D., Bradt, S., & Verdu, A. R., 2014. An optical water type framework
for selecting and blending retrievals from bio-optical algorithms in lakes and coastal
waters. Remote Sensing of Environment, 143, 97–111.
Mouw, C. B. et al., 2015. Aquatic color radiometry remote sensing of coastal and inland
waters: Challenges and recommendations for future satellite missions. Remote Sensing
of Environment, 160, 15–30.
Odermatt, D., Gitelson, A., Brando, V. E., & Schaepman, M., 2012. Review of constituent
retrieval in optically deep and complex waters from satellite imagery. Remote Sensing
of Environment, 118, 116–126.
O’Reilly, J., Maritorena, S., Mitchell, B., Siegel, D., Carder, K., Garver, S. et al., 1998. Ocean
color chlorophyll algorithms for SeaWiFS. Journal of Geophysical Research, 103,
24937–24953.
188 Remote Sensing for Sustainability

Oyama, Y., Matsushita, B., Fukushima, T., Matsushige, K., & Imai, A., 2009. Application
of spectral decomposition algorithm for mapping water quality in a turbid lake (Lake
Kasumigaura, Japan) from Landsat TM data. ISPRS Journal of Photogrammetry and
Remote Sensing, 64, 73–85.
Pan, X., Mannino, A., Russ, M. E., Hooker, S. B., & Harding, L. W., 2010. Remote sensing of
phytoplankton pigment distribution in the United States northeast coast. Remote Sensing
of Environment, 114, 2403–2416.
Ruddick, K., Neukermans, G., Vanhellemont, Q., & Jolivet, D., 2014. Challenges and
opportunities for geostationary ocean colour remote sensing of regional seas: A review
of recent results. Remote Sensing of Environment, 146, 63–76.
Ryu, J. H., Han, H. J., Cho, S., Park, Y. J., & Ahn, Y. H., 2012. Overview of Geostationary
Ocean Color Imager (GOCI) and GOCI Data Processing System (GDPS). Ocean
Science Journal, 47, 223–233.
Shi, W., & Wang, M., 2009. An assessment of the black ocean pixel assumption for MODIS
SWIR bands. Remote Sensing of Environment, 113, 1587–1597.
Stumpf, R. P., Arnone, R. A., Gould, R. W., Martinolich, P. M., & Ransibrahmanakul, V.,
2003. A partly coupled ocean-atmosphere model for retrieval of water-leaving radiance
from SeaWiFS in coastal waters. In SeaWiFS Postlaunch Technical Report Series,
NASA Technical Memorandum 2003-206892, Vol. 22, S. B. Hooker and E. R. Firestone
(Eds.), 51–59. Greenbelt, MD: NASA Goddard Space Flight Center.
Tassan, S., 1988. The effect of dissolved yellow substance on the quantitative retrieval of
chlorophyll and total suspended sediment concentrations from remote measurements of
water colour. International Journal of Remote Sensing, 9, 787–797.
Tyler, A. N., Svab, E., Preston, T., Presing, M., & Kovacs, W. A., 2006. Remote sensing
of the water quality of shallow lakes: A mixture modeling approach to quantifying
phytoplankton in water characterized by high-suspended sediment. International
Journal of Remote Sensing, 27, 1521–1537.
UNEP, 2006. Marine and coastal ecosystems and human wellbeing: A synthesis report based
on the findings of the Millennium Ecosystem Assessment. UNEP, 76 pp.
United Nations Open Working Group, 2014. Outcome document—Open Working Group
on Sustainable Development Goals (19th July 2014). United Nations, New York, New
York. URL: http://sustainabledevelopment.un.org/content/documents/1579SDGs%20
Proposal.pdf.
Verpoorter, C., Kutser, T., Seekell, D. A., & Tranvik, L. J., 2014. A global inventory of
lakes based on high-resolution satellite imagery. Geophysical Research Letters, 41,
6396–6402.
Verpoorter, C., Kutser, T., & Tranvik, L. J., 2012. Automated mapping of water bodies using
Landsat multispectral data. Limnology and Oceanography: Methods, 10, 1037–1050.
Wang, M., 2007. Remote sensing of the ocean contributions from ultraviolet to near-infrared
using the shortwave infrared bands: Simulations. Applied Optics, 46, 1535–1547.
Wang, M., & Shi, W., 2005. Estimation of ocean contribution at the MODIS near-infrared
wavelengths along the east coast of the US: Two case studies. Geophysical Research
Letters, 32, L13606.
Wang, M., & Shi, W., 2007. The NIR–SWIR combined atmospheric correction approach for
MODIS ocean color data processing. Optics Express, 15, 15722–15733.
Wang, M., & Shi, W., 2012. Sensor noise effects of the SWIR bands on MODIS-derived ocean
color products. IEEE Transactions on Geoscience and Remote Sensing, 50, 3280–3292.
Wang, M., Shi, W., & Jiang, L., 2012. Atmospheric correction using near-infrared bands
for satellite ocean color data processing in the turbid Western Pacific Region. Optics
Express, 20, 741–753.
Monitoring Water Quality with Remote Sensing Image Data 189

Wu, G., Cui, L., Liu L., Chen, F., Fei, T., & Liu, Y., 2015. Statistical model development
and estimation of suspended particulate concentrations with Landsat 8 OLI images of
Dongting Lake, China. International Journal of Remote Sensing, 36, 343–360.
Xu, H., 2006. Modification of normalized difference water index (NDWI) to enhance open
water features in remotely sensed imagery. International Journal of Remote Sensing,
27, 3025–3033.
Yang, W., Matsushita, B., Chen, J., & Fukushima, T., 2011. Estimating constituent
concentrations in case II waters from MERIS satellite data by semi-analytical model
optimizing and look-up tables. Remote Sensing of Environment, 115, 1247–1259.
Yang, W., Matsushita, B., Chen, J., Fukushima, T., & Ma, R., 2010. An enhanced three-
band index for estimating chlorophyll-a in turbid case-II waters: Case studies of Lake
Kasumigaura, Japan and Lake Dianchi, China. IEEE Geoscience and Remote Sensing
Letters, 7, 655–659.
Yang, W., Matsushita, B., Chen, J., Yoshimura, K., & Fukushima, T., 2013. Retrieval of
inherent optical properties for turbid inland waters from remote-sensing reflectance.
IEEE Transactions on Geoscience and Remote Sensing, 51, 3761–3773.
Yang, W., Matsushita, B., Chen, J., Yoshimura, K., & Fukushima, T., 2014. Application of a
semianalytical algorithm to remotely estimate diffuse attenuation coefficient in turbid
inland waters. IEEE Geoscience and Remote Sensing Letters, 11, 1046–1050.
Yang, W., Matsushita, B., Yoshimura, K., Chen, J., & Fukushima, T., 2015. A modified
semianalytical algorithm for remotely estimating euphotic zone depth in turbid inland
waters. IEEE Journal of Selected topics in Applied Earth Observations and Remote
Sensing, 8, 1545–1554.
Section III
Remote Sensing for Sustainable
Environmental Systems
 10 Urban Air Quality
Studies Using EO Data
Xuefei Hu

CONTENTS
10.1 PM2.5 and Its Health Effects ......................................................................... 193
10.2 Remote Sensing Applications for PM2.5 Concentration Prediction .............. 195
10.2.1 Data ................................................................................................... 195
10.2.1.1 AOD Data........................................................................... 195
10.2.1.2 Fire Data ............................................................................ 199
10.2.1.3 LIDAR Data .......................................................................200
10.2.2 Methods ............................................................................................200
10.2.2.1 Physically Based Methods .................................................200
10.2.2.2 Statistical Methods............................................................. 201
10.3 Case Study ....................................................................................................203
10.3.1 Data ...................................................................................................203
10.3.2 Methodology .....................................................................................203
10.3.3 Results...............................................................................................204
References .............................................................................................................. 205

10.1 PM2.5 AND ITS HEALTH EFFECTS

PM2.5 refers to particles with an aerodynamic diameter of less than 2.5 μm and is
also known as fine particulate matter, including dust, dirt, soot, smoke, and liquid
droplets. PM2.5 originates from both natural and anthropogenic sources (Alves et al.
2000). Some particulates are emitted naturally from volcano eruptions, dust storms,
and forest and grassland fires. However, natural sources only make a small contribu-
tion to the total concentration, and anthropogenic sources are more important. For
instance, human activities, including the burning of fossil fuels in road vehicles,
power plants, and certain industrial processes, contribute a significant amount of
fine particles. In addition, secondary aerosols derived from precursors emitted from
various sources such as cars, trucks, power plants, and industrial facilities are also
a major contributor. Secondary aerosol formation occurs because of chemical reac-
tion in the atmosphere generally downwind a distance from the original emission
sources (Hodan and Barnard 2004). Besides outdoor sources, PM2.5 is also produced
by indoor activities such as cooking (e.g., frying). The composition of particles varies
and depends on their sources. For example, mineral dust is generally made of min-
eral oxides. Primary aerosols that are directly emitted from emission sources into
the atmosphere may include sulfur dioxide (SO2), nitrogen oxides (NOx), volatile

193
194 Remote Sensing for Sustainability

organic compounds (VOCs), ammonia (NH3), elemental carbon (EC), and crustal
materials such as soil and ash. Secondary aerosols may include ammonium sulfate
and ammonium nitrate, which are typically formed through chemical reaction of
precursors such as ammonia, SO2, and NOx (Squizzato et al. 2013). Organic carbon
(OC) is one of a large group of compounds (VOCs) and can be either primary or
secondary. It is formed by a variety of processes, including combustion and second-
ary organic aerosol formation. In general, coarse particles are mainly made up of
primary aerosols, and fine particles are heavily contributed by secondary aerosols
(Deng et al. 2013).
Numerous epidemiological and experimental studies have shown that PM2.5, one
of the criterion air pollutants regulated by the US National Ambient Air Quality
Standard, is associated with various adverse health outcomes including cardiovascu-
lar and respiratory mobility and mortality (Dominici et al. 2006). Peters et al. (2001)
suggested that the elevated concentrations of fine particles may increase the risk of
myocardial infarction after a few hours or 1 day of exposure to those air pollutants.
Riediker et al. (2004) found that in-vehicle exposure to PM2.5 is associated with
cardiovascular effects and may cause pathophysiologic changes including inflam-
mation, coagulation, and cardiac rhythm. Miller et al. (2007) revealed that long-
term exposure to PM2.5 is associated with increased risk of cardiovascular disease
and death for postmenopausal women. Puett et al. (2009) further demonstrated that
chronic exposure to PM2.5 tends to increase the risk of all-cause and cardiovascu-
lar mortality. Madrigano et al. (2013) found that long-term exposure to area PM2.5
was associated with the occurrence of acute myocardial infarction. Neophytou et
al. (2014) revealed that occupational PM2.5 exposure increases the risk of incident
ischemic heart disease in both aluminum smelting and fabrication facilities. Sunyer
and Basagaña (2001) reported that fine particles are associated with the risk of death
in patients with chronic obstructive pulmonary disease (COPD), and their findings
pointed out the adverse impact of fine particles on the trigger of death in COPD
patients. Bose et al. (2015) demonstrated that even low indoor PM2.5 levels may cause
a systematic inflammatory response in COPD. Norris et al. (1999) found a significant
association between emergency department visit for asthma in children and fine par-
ticulate matter. Lin et al. (2002) reported that exposure to fine particles emitted from
heavy traffic contributes to childhood asthma hospitalization. Brauer et al. (2002)
also revealed a positive association between respiratory diseases (e.g., respiratory
infection and asthma) and fine particles in children. Spira-Cohen et al. (2011) sug-
gested that pollution-related asthma exacerbations in children living near roadways
are mainly attributed to the diesel soot fraction of PM2.5. Habre et al. (2014) reported
that exposure to indoor and outdoor PM2.5 may exacerbate cough and wheeze symp-
toms in asthmatic children. Laden et al. (2006) found an increase in overall mortal-
ity associated with each 10 μg/m3 increase in PM2.5, indicating that cardiovascular
mortality and lung cancer mortality were positively associated with ambient PM2.5
concentrations. Kioumourtzoglou et al. (2016) found strong evidence showing poten-
tially harmful effects of long-term exposure to PM2.5 on neurodegeneration. Chung
et al. (2015) pointed out that long-term exposure to PM2.5 was associated with mor-
tality in the elderly population of the eastern United States.
Urban Air Quality Studies Using EO Data 195

10.2 REMOTE SENSING APPLICATIONS FOR PM2.5

CONCENTRATION PREDICTION
Most studies of the associations between PM2.5 and health effects have relied on
ground measurements typically from regulatory monitoring systems, such as United
States Environmental Protection Agency (USEPA) Air Quality System to estimate
PM2.5 exposure. Although the EPA measurements are considered ground truth and
the most accurate, the spatial distribution of those monitoring stations is sparse
and uneven. For example, the United States has an extensive monitoring network and
has approximately 1200 stations. However, those stations only cover ~30% of ~3000
counties in the conterminous United States. In addition, many of those stations are
located in or near urban areas, which leaves ~30% of the US population living in
suburban and rural areas without any PM2.5 monitors. Thus, the limited coverage of
ground monitoring stations limited the epidemiological studies to areas near moni-
toring sites. In addition, collecting PM2.5 measurements from ground monitoring
station is also expensive and time-consuming.
Satellite remote sensing data, given its comprehensive spatiotemporal coverage,
have the potential to expand ground network by estimating PM2.5 concentrations
using aerosol optical depth (AOD) in areas where monitoring stations are not avail-
able or too sparse. AOD measures light extinction by aerosol scattering and absorp-
tion in an atmospheric column, and AOD retrieved from visible channels is most
sensitive to particles with size from 0.1 to 2 μm (Kahn et al. 1998). Thus, it can be
considered a representation of the loadings of fine particles in an atmospheric col-
umn. In addition, Wang and Christopher (2003) reported a good correlation between
the satellite-derived AOD and PM2.5. Using satellite-derived AOD to estimate PM2.5
exposure is not only time and cost efficient, but could substantially improve esti-
mates of population exposure to PM2.5 (van Donkelaar et al. 2010). In contrast, using
ground measurements inevitably introduces measurement errors to estimation of
population exposure (Zeger et al. 2000).

10.2.1 Data
10.2.1.1 AOD Data
To date, AOD derived from a number of satellite sensors, including MODerate
resolution Imaging Spectroradiometer (MODIS) (Hu et al. 2013; Liu et al.
2007a; Zhang et al. 2009), Multiangle Imaging SpectroRadiometer (MISR)
(Liu et al. 2007a), Geostationary Operational Environmental Satellite Aerosol/
Smoke Product (GASP) (Liu et al. 2009; Paciorek et al. 2008), and Multi-Angle
Implementation of Atmospheric Correction (MAIAC) (Hu et al. 2014a,b,c), has
been used in previous studies to estimate PM 2.5 concentrations. In addition, the
new Visible Infrared Imaging Radiometer Suite (VIIRS) AOD product will poten-
tially have wide applications for PM 2.5 concentration prediction when the data
quality becomes acceptable. Table 10.1 lists all available AOD products, their
characteristics, and the studies in which these AOD products were used for PM2.5
concentration estimation.
196 Remote Sensing for Sustainability

TABLE 10.1
Available AOD Products, Their Characteristics, and Related Studies
AOD Products Spatial Resolutions Studies
GASP 4 km Chudnovsky et al. 2012b; Green et al. 2009; Liu et al.
2009; Paciorek et al. 2008 etc.
MODIS 10 km Hu et al. 2013; Lee et al. 2011; Lin et al. 2015; Ma et al.
2015; Wang and Christopher 2003 etc.
MISR 17.6 km Liu et al. 2004, 2007b; Ma et al. 2014; van Donkelaar et al.
2006 etc.
MAIAC 1 km Chudnovsky et al. 2012a, 2013; Hu et al. 2014a,b;
Just et al. 2015; Lee et al. 2015 etc.
VIIRS 6 km Schliep et al. 2015 etc.

10.2.1.1.1 Geostationary Operational Environmental

Satellite Aerosol/Smoke Product
GASP retrieves AOD over the contiguous United States from the GOES-East and
GOES-West visible imagery with a temporal resolution of 30 min (up to 15 min) and
a spatial resolution of 4 km. Unlike instruments onboard polar-orbiting satellites
that provide only one daily AOD retrieval, the GOES satellite’s geostationary orbit
allows AOD retrievals with higher frequency. As a result, the GASP AOD product
with high temporal resolution has the ability to capture rapidly changing aerosol
conditions attributed to dust storms and fire events.
The GASP AOD has been used as an important proxy for PM2.5. Paciorek et al.
(2008) examined the association between GASP AOD and ground-level PM2.5 con-
centrations and found that GASP AOD has the potential to improve exposure esti-
mates for epidemiological studies, considering its higher temporal coverage than
MODIS and MISR. Green et al. (2009) evaluated the quality of GASP AOD and its
potential to predict surface PM2.5. The results point out that using satellite AOD to
predict surface PM2.5 needs to take several factors into account, including seasonal
and diurnal variations in particle size distribution, relative humidity, and seasonal
change in boundary layer height. Liu et al. (2009) developed a two-stage general-
ized addictive model (GAM) with GASP AOD, meteorological fields, and land use
variables as predictors to estimate ground-level PM2.5 concentrations in a domain
centered in Massachusetts. The results show that the model with GASP AOD out-
performs the non-AOD model and has a greater predicting power. Chudnovsky et al.
(2012b) used a mixed-effects model with the control for the day-to-day variability
in the PM2.5–AOD relationship to predict PM2.5 concentrations in the New England
region from GASP AOD. The predicted PM2.5 concentrations have a good agreement
with observations, which indicates that accounting for daily variability in the PM2.5–
AOD relationship is essential to obtain spatiotemporally resolved PM2.5 exposure.

10.2.1.1.2 Moderate Resolution Imaging Spectroradiometer

The MODIS aerosol product monitors AOD over both ocean and land and provides
daily near-global observations at a spatial resolution of 10 km. Two types of MODIS
Urban Air Quality Studies Using EO Data 197

AOD products are available, including MOD04 and MYD04. MOD04 contains
data collected from the TERRA satellite since 2000, while MYD04 includes data
acquired from the AQUA platform from 2002 onward. Compared to previous efforts
such as GOES and AVHRR, MODIS expands spectral channels for aerosol measure-
ment, and as a result, MODIS not only can retrieve AOD over dark vegetated/soiled
land by the Dark-Target aerosol retrieval approach but also derive AOD over some
brighter surfaces such as deserts using the Deep-Blue algorithm through the blue
band. In addition, MODIS is also the first to provide a real solution for AOD retriev-
als over land on a global scale.
MODIS AOD has been widely used in PM2.5 concentration predictions. For
instance, Wang and Christopher (2003) examined the relationship between MODIS
AOD and hourly surface PM2.5 at seven locations in Jefferson county, Alabama.
Zhang et al. (2009) examined the relation between MODIS AOD and PM2.5 over
the 10 USEPA-defined geographic regions in the United States, and the results show
good correlations over the eastern United States in summer and fall. Hu (2009) used
geographically weighted regression (GWR) to derive a spatially complete PM2.5
surface covering the conterminous United States from MODIS AOD. Lee et al.
(2011) developed a mixed-effects model that allows day-to-day variability in daily
PM2.5–AOD relationships to predict PM2.5 concentrations in the New England region
from MODIS AOD. Kloog et al. (2012) developed an advanced three-stage model
to predict spatiotemporally resolved PM2.5 concentrations using MODIS AOD in
the Mid-Atlantic States. The model has the capability to predict PM2.5 exposure in
areas where AOD is missing. Hu et al. (2013) developed a GWR model with MODIS
AOD, meteorological fields, and land use variables as predictors to predict PM2.5
concentrations in the southeastern United States by incorporating spatially varying
relationships between PM2.5 and AOD. Lin et al. (2015) retrieved ground-level PM2.5
concentrations from MODIS AOD by conducting both vertical and humidity correc-
tions. Xie et al. (2015) developed a mixed-effects model to derive daily estimations
of surface PM2.5 in Beijing from 3-km-resolution MODIS AOD. Ma et al. (2015)
developed a two-stage statistical model to estimate ambient PM2.5 concentrations
from 2004 to 2013 in China using MODIS collection 6 AOD data.

10.2.1.1.3 Multiangle Imaging Spectroradiometer

MISR is an instrument onboard the Terra satellite in a sun-synchronous orbit, and
its five viewing angles (0°, 26.1°, 45.6°, 60.0°, and 70.5°) allow changes in reflec-
tion that provides the means to distinguish different types of aerosols. The swath of
the MISR instrument is ~400 km, which allows it to view the entire Earth surface
every 9 days. The MISR aerosol product is at a spatial resolution of 17.6 km derived
by the MISR Standard Aerosol Retrieval Algorithm from MISR top-of-atmosphere
radiances. Kahn et al. (2005) compared MISR AOD and Aerosol Robotic Network
(AERONET) data on a global scale and obtained high correlation coefficients
between them for various sites.
Many studies have examined the relationship between MISR AOD and surface
PM2.5 concentrations. Liu et al. (2004) used the global chemical transport models
output and obtained the local PM2.5–AOD conversion factors over the contigu-
ous United States to predict surface PM2.5 concentrations from MISR AOD. van
198 Remote Sensing for Sustainability

Donkelaar et al. (2006) used the factors affecting the relationship between PM2.5
and AOD simulated from GEOS-CHEM to estimate ground-level PM2.5 concentra-
tions from both MISR and MODIS AOD, and they found significant spatial variation
of the relationship between the annual mean ground-level measurements and PM2.5
estimated from MODIS and MISR. Liu et al. (2007b) used the MISR fractional AOD
and aerosol transport model constraints to predict ground-level PM2.5 concentrations
and its major constituents in the continental United States. The results show that
using fractional AOD can significantly improve the estimating power of the model,
compared with similar models with total-column AOD as the single predictor. Liu
et al. (2007a) compared the ability of MISR and MODIS AOD to predict ground-
level PM2.5 concentrations in St. Louis, MO, and found that MISR achieves higher
prediction accuracy, while MODIS provides better spatial coverage. Ma et al. (2014)
developed a national-scale GWR model to estimate daily PM2.5 concentrations in
China with AOD data fused from MISR and MODIS data.

10.2.1.1.4 Multi-Angle Implementation of Atmospheric Correction

The MAIAC AOD product is derived from MODIS radiances using a newly devel-
oped MAIAC algorithm. The algorithm has a global scope and can retrieve AOD
over both dark and bright surfaces at a spatial resolution of 1 km. Validation over the
continental United States shows that MAIAC and operational Collection 5 MODIS
Dark Target AOD have a similar accuracy over dark and vegetated surfaces, while
MAIAC achieved higher accuracy over brighter surfaces (Lyapustin et al. 2011). In
addition, MAIAC provides greater spatial coverage and more AOD retrievals than
MODIS (Hu et al. 2014a).
Chudnovsky et al. (2012a) investigated the relationship between MAIAC AOD
and PM2.5 concentrations measured by EPA ground monitors at a variety of spatial
scales and found that the correlation between PM2.5 and AOD decreased significantly
with the decrease of AOD resolution. Chudnovsky et al. (2013) further compared
the relationship between MAIAC AOD and PM2.5 measured from 84 EPA ground
monitors with the PM2.5–AOD relationship using MODIS AOD, and the results show
that the correlation coefficient for MAIAC is slightly higher, and MAIAC AOD is
more capable of capturing spatial patterns of PM2.5. In addition, MAIAC AOD can
also help increase the number of days for PM2.5 prediction. Hu et al. (2014a) devel-
oped an advanced two-stage model with MAIAC AOD as the primary predictor
and meteorological fields and land use variables as secondary predictors to estimate
spatiotemporally resolved PM2.5 concentrations in the southeastern United States at
1 km resolution. The model includes a linear mixed-effects model to account for the
day-to-day variability in the PM2.5–AOD relationship and a geographical weighted
regression model to explain the spatial variability. Hu et al. (2014b) further inves-
tigated the 10-year spatiotemporal trend of PM2.5 concentrations in the southeast-
ern United States using PM2.5 at 1 km resolution estimated from MAIAC AOD.
Kloog et al. (2014) predicted daily PM2.5 at 1 km resolution across the northeastern
United States, including New England, New York, and New Jersey, from 2003 to
2011 using MAIAC AOD. Lee et al. (2015) applied a separate mixed-effects model
to predict daily PM2.5 concentrations at 1 km resolution in the southeastern United
States for the years 2003–2011 using MAIAC AOD. Just et al. (2015) calibrated the
Urban Air Quality Studies Using EO Data 199

relationship between PM2.5 and MAIAC AOD using ground monitors, land use, and
meteorological features and predicted daily PM2.5 concentrations at 1 km resolution
across the great Mexico City area for 2004–2014.

10.2.1.1.5 Visible Infrared Imaging Radiometer Suite

The VIIRS instrument is onboard the Suomi National Polar Partnership, which was
launched in late 2011 and is a polar-orbiting satellite with afternoon overpass (similar
to Aqua, ~1:30 p.m. local time). VIIRS provides two high-resolution AOD products,
including Intermediate Product at a spatial resolution of 750 m and Environmental
Data Record with a resolution of 6 km. Two types of AOD products make it possible
for air quality applications to examine spatial detail on an urban scale and also moni-
tor large-scale events like wildfires.
Because of the recent launch, applications of VIIRS AOD products for PM2.5
prediction have been limited but are promising. Schliep et al. (2015) developed a
hierarchical autoregressive model with daily spatially varying coefficients to esti-
mate daily average PM2.5 across the conterminous United States from VIIRS AOD.
Because of the quality assurance protocol of VIIRS AOD, many missing AOD data
contribute to the ineffectiveness of AOD in their model.

10.2.1.2 Fire Data

Another remote sensing product that can be used for PM2.5 concentration esti-
mation is fire data. Fires, including prescribed burning and wildfires, are impor-
tant sources of fine particulate matter. Zeng et al. (2008) estimated a maximum
increase of 25 μg/m3 in PM2.5 concentrations within a day owing to prescribed
burning emissions. Zhang et al. (2008) conducted a research in Beijing with the
time span from July 2002 to July 2003 and found that biomass burning contributes
18%–38% of PM2.5 organic carbon. Jaffe et al. (2008) reported that the increase in
PM2.5 attributed to fires reaches a high fraction of the annual National Ambient Air
Quality Standards in summer and has a significant contribution to regional haze
in the western United States. Zhang et al. (2010) revealed that 13% of PM 2.5 con-
centrations are attributed to biomass burning annually in the southeastern United
States. Tian et al. (2009) reported that prescribed burning contributes 55% and
80% of PM2.5 mass concentrations in January and March of 2002, respectively.
Christopher et al. (2009) found a threefold increase of PM2.5 concentrations col-
lected from ground monitors in Birmingham, Illinois, during fire events, com-
pared to background values, although those monitors are hundreds of miles away
from the fire sources.

10.2.1.2.1 MODIS Fire Product

The MODIS fire products are produced for the MODIS sensors onboard two Earth
Observing System (EOS) satellites: Terra and Aqua. The two satellites are in sun-
synchronous orbits with different local overpass times (~1:30 p.m. for Aqua and
~10:30 a.m. for Terra). As a result, MODIS can provide two overpasses over a location
in a single day, which is important because the differences in fires between morning
and afternoon can be detected. Two products exist. One is the MODIS Active Fire
Product, which provides actively burning fire locations at satellite overpass times,
200 Remote Sensing for Sustainability

and the other is the MODIS Burned Area Product, which provides the burned area.
The active fire products are generated daily at 1 km spatial resolution.
The use of MODIS fire in PM2.5 concentration prediction has been limited. Hu
et al. (2014c) incorporated MODIS fire counts in an advanced two-stage model to
examine if remotely sensed fire data can help improve the prediction accuracy of
PM2.5 concentrations in the southeastern United States and conducted a sensitivity
analysis to determine the optimum buffer radius centered around each PM2.5 moni-
toring site, which is crucial to count the fire incidents that may have impact on the
corresponding monitors. The results show that when the radius reaches 75 km, fire
count data achieve the greatest predictive power of PM2.5 concentrations. A compari-
son between the fire model and the nonfire model shows that the prediction accuracy
increases more substantially at the sites with higher fire occurrence when MODIS
fire count data are incorporated in the model, indicating that remotely sensed fire
count data can provide a measurable improvement in PM2.5 concentration prediction,
particularly in areas where fires occur frequently.

10.2.1.3 LIDAR Data

Light Detection and Ranging (LIDAR) data have also been used for estimating sur-
face PM2.5 concentrations. A LIDAR system can use a laser to measure aerosol scat-
tering as a function of height in the atmosphere, and the advantage of the LIDAR
method over satellite instruments is that it can calculate optical depth for selected
ranges in the atmosphere. Engel-Cox et al. (2006) pointed out that LIDAR appor-
tionment of the fraction of AOD within the planetary boundary layer (PBL) can
achieve better agreement with surface PM2.5 than does the total column amount.
He et al. (2008) used LIDAR measurements to obtain significant improvements in
correlation between AOD with surface extinction and PM2.5 by considering aerosol
vertical distribution. Chu et al. (2013) demonstrated that surface PM2.5 can be better
estimated using AOD normalized by haze layer height derived from LIDAR aerosol
extinction profiles than that using AOD only.

10.2.2 MethoDs
10.2.2.1 Physically Based Methods
Liu et al. (2004) developed a simple approach using outputs from a global chemis-
try and transport model (CTM) to examine the physical basis for the relationship
between PM2.5 and AOD. The local PM2.5–AOD conversion factors were obtained
using simulated PM2.5 concentrations and AOD. Simulated PM2.5 concentrations
include sea salt and dust mass concentrations from the Global Ozone Chemistry
Aerosol Radiation and Transport (GOCART) model and mass concentrations for
− +
SO 2−
4 , NO 3 , NH 4 , EC, and OC derived from GEOS-CHEM, while simulated AOD
is composed of AOD values for sea salt and dust from GOCART and those for other
particulate species from GEOS-CHEM. The results show that the predicted PM2.5
concentrations from MISR AOD are strongly correlated with EPA PM2.5 measure-
ments, and the proposed method has the capability to reduce the uncertainty in esti-
mated PM2.5 concentrations owing to the discrepancy of correlations between lower
Urban Air Quality Studies Using EO Data 201

and upper tropospheric aerosols. van Donkelaar et al. (2006) extended this method
to estimate PM2.5 concentrations from both MISR and MODIS and examined the
factors affecting the PM2.5–AOD relationship using the GEOS-CHEM simulation.
The GEOS-CHEM aerosol simulation was conducted for each day from January
2001 to October 2002 at the MODIS and MISR overpass times to obtain local values
for each parameter required in the PM2.5 prediction model. The results show that the
relative vertical profile is the most influential factor affecting the spatial relationship
between predicted and measured PM2.5 concentrations, while the temporal variation
in AOD is the most important factor that affects the temporal relationship between
predicted and measured concentrations. Lin et al. (2015) also used a physically based
method to retrieve ground-level PM2.5 concentrations from MODIS AOD by taking
account of the effect of the main aerosol characteristics. The effects on hygroscopic
growth, particle mass extinction efficiency, and size distribution are estimated and
incorporated in the PM2.5–AOD relationship.

10.2.2.2 Statistical Methods

Statistical models have also been used to predict PM2.5 concentrations from remotely
sensed AOD, and the prediction models have been evolving from using AOD as
the only predictor to the incorporation of multiple predictors including meteorologi-
cal and land use variables and from one-stage models to advanced multiple-stage
models. Wang and Christopher (2003) explored the relationship between MODIS
AOD and hourly PM2.5 mass concentrations at seven locations in Jefferson County,
Alabama, and used an empirical linear regression model to derive PM2.5 concentra-
tions from MODIS AOD. The results indicate that MODIS AOD has great poten-
tial for air quality applications. However, this model assumes that the relationships
between PM2.5 and AOD are constant over space and time, and a single linear PM2.5–
AOD relationship was established for all sampling sites and days, which may lead
to bias for PM2.5 prediction. Hu (2009) found that the PM2.5–AOD relationships are
spatially inconsistent across the conterminous United States and fitted a GWR model
with MODIS AOD as the independent variable to investigate the PM2.5–AOD rela-
tionship using the 2-year average PM2.5 and AOD data. GWR is a linear regression
model that can model spatial varying relationship and reveal how the PM2.5–AOD
relationship changes over space by generating a continuous surface of estimates for
each predictor at each local location instead of a universal value. The results show
that it is appropriate to estimate PM2.5 surface from AOD using GWR, and the esti-
mated PM2.5 reaches an accuracy of 84%. Ma et al. (2014) developed a national-scale
GWR model to estimate daily PM2.5 concentrations in China from fused satellite
AOD, and their results confirm the satisfactory performance of the GWR model. Lee
et al. (2011) developed a mixed-effects model with MODIS AOD as the predictor to
predict daily PM2.5 concentrations in the New England region. The mixed-effects
model with random intercept and slopes allows day-to-day variability in the PM2.5–
AOD relationship and, as a result, can calculate a PM2.5–AOD slope separately for
each day. The results suggest that the proposed method makes it possible to deter-
mine spatial and temporal patterns of PM2.5 concentrations in a relatively large study
domain from remotely sensed AOD. Xie et al. (2015) also developed a mixed-effects
model to derive daily estimations of surface PM2.5 in Beijing from 3-km-resolution
202 Remote Sensing for Sustainability

MODIS AOD. The model can account for daily variations of the PM2.5–AOD rela-
tionship and shows good performance in model predictions. In addition to AOD
values, many other parameters have been demonstrated as influential factors that
can affect the relationship between PM2.5 and AOD and, as a result, have the poten-
tial to be used as predictors for PM2.5 concentration estimation. Liu et al. (2005)
incorporated meteorological fields, including PBL height and relative humidity, as
covariates in linear regression models to predict PM2.5 concentrations and found that
they are highly significant predictors of PM2.5. Furthermore, Liu et al. (2007a) intro-
duced other meteorological parameters such as air temperature and wind speed to be
included in a general linear regression model for PM2.5 concentration prediction, and
the results show that both wind speed and air temperature are significant predictors
of PM2.5. Moving one step further, Liu et al. (2009) used generalized additive mod-
els incorporated with land use information, including road length and population
density, to predict daily PM2.5 concentrations on a regional scale and demonstrate
that land use variables are effective predictors of PM2.5 concentrations. Kloog et al.
(2011) extended previous work by introducing numerous land use variables that can
be potentially used in statistical models for PM2.5 concentration prediction, includ-
ing percent of open spaces, elevation, major roads, PM2.5 point emissions, and area-
source point emissions. The results indicate that by including land use variables, the
proposed model outperforms previous AOD-PM2.5 models. Hu et al. (2013) incor-
porated forest cover in the GWR model for PM2.5 concentration estimation in the
southeastern United States and found that forest cover is a statistically significant
predictor of PM2.5. Hu et al. (2014c) further tested the inclusion of MODIS fire counts
in the linear mixed-effects model to examine if fire count data are an effective pre-
dictor of PM2.5. The results show that the prediction accuracy improved more from
the nonfire model to the fire model at sites with higher fire occurrence, and the inclu-
sion of fire count data can provide a measurable improvement in PM2.5 concentration
estimation, particularly in areas and seasons prone to fire events. In addition, to
account for both spatial and temporal variability in the PM2.5–AOD relationship and
predict PM2.5 concentrations in areas and days where and when AOD is not available,
multistage statistical models are becoming more and more popular. Liu et al. (2009)
proposed a two-stage GAM model to examine the spatial and temporal variability
in PM2.5 concentrations separately. The first stage aims to account for the temporal
variability. The dependent variable is the daily PM2.5 concentrations, and all the
covariates are averaged spatially and therefore only vary temporally. The second
stage is to explain the spatial variability. All the covariates in this stage are averaged
over the entire period and therefore only vary spatially. The results show that the first
stage model contributes more to the overall model performance because temporal
variability dominates the overall PM2.5 variability. Hu et al. (2014a) developed a two-
stage model incorporated with meteorological fields and land use variables to predict
1-km PM2.5 surface from MAIAC AOD in the southeastern United States. The model
includes a first-stage linear mixed-effects model to account for the day-to-day vari-
ability in the relationship between PM2.5 and AOD and a second-stage GWR model
to explain the spatial variability in the PM2.5–AOD relationship. Hence, the model
has the capability to predict spatiotemporally resolved PM2.5 concentrations with
high accuracy. Ma et al. (2015) also developed a two-stage spatial statistical model
Urban Air Quality Studies Using EO Data 203

to estimate ambient PM2.5 concentrations from 2004 to 2013 in China. The model
includes a linear mixed-effects model as the first stage and a GAM model as the
second stage to calibrate the spatiotemporal relationship between PM2.5 and AOD.
Kloog et al. (2011) and Kloog et al. (2012) introduced a third-stage GAM model to
estimate daily PM2.5 concentrations in the study domain for days when AOD data
were unavailable. The model includes a smooth function for the coordinates of each
grid cell centroid and a random intercept for each grid cell, uses predicted PM2.5
concentrations from previous stages as the dependent variable and the mean of PM2.5
measurements collected from monitors on that day as the predictor, and is similar to
universal kriging. In addition, to account for temporal variation, the model was fit
for each 2-month period of each year to generate separate spatial surfaces for those
periods.

10.3 CASE STUDY

A case study was conducted to estimate ground-level PM2.5 concentrations from
MODIS AOD in Atlanta, Georgia, in 2011using statistical methods. The study area
was approximately 130 km × 130 km, covering the Atlanta Metro area. The domain
included a large urban center and surrounding suburban areas.

10.3.1 Data
The 24-h averaged PM2.5 concentrations in 2011 were collected from 13 EPA Federal
Reference Monitors and downloaded from the EPA’s Air Quality System Technology
Transfer Network (http://www.epa.gov/ttn/airs/airsaqs/). The 2011 MODIS AOD
data at a spatial resolution of 10 km were downloaded from the Earth Observing
System Data Gateway at the Goddard Space Flight Center (https://ladsweb.nascom
.nasa.gov/data/search.html). Hourly meteorological parameters were obtained from
the North American Land Data Assimilation System Phase 2 (http://ldas.gsfc.nasa
.gov/nldas/) at a spatial resolution of ~13 km. Elevation data were obtained from
the National Elevation Dataset (http://ned.usgs.gov) at a spatial resolution of 1 arc-
sec (~30 m). Road data were extracted from ESRI StreetMap USA (Environmental
Systems Research Institute, Inc., Redland, California). Forest cover data were gener-
ated using 2011 land cover maps downloaded from the National Land Cover Database
(http://www.mrlc.gov). Primary PM2.5 emissions were obtained from the 2011 EPA
National Emission Inventory facility emissions reports. All data were integrated for
model fitting and PM2.5 prediction, a 1 km × 1 km square buffer was generated for
each PM2.5 monitoring site and MODIS pixel centroid, and meteorological fields and
AOD values were assigned using the nearest-neighbor approach. Forest cover and
elevation were averaged, while road length and point emissions were summed over
the 1 km × 1 km square buffer. A fitting data set with 1320 records was obtained.

10.3.2 MethoDology
A linear mixed-effects model (Kloog et al. 2011) with 24-h averaged PM2.5 mea-
surements as the dependent variable, MODIS AOD as the primary predictor, and
204 Remote Sensing for Sustainability

meteorological and land use variables as secondary predictors is developed to esti-

mate ground-level PM2.5 concentrations. The model explains the temporal variabil-
ity in the PM2.5–AOD relationship by including day-specific random intercepts and
slopes for AOD and meteorological variables. The model structure can be expressed
as

PM2.5,st = (b0 + b0 ,t ) + (b1 + b1,t ) AOD st + (b2 + b2,t )Relative Humidity st

+ (b3 + b3,t ) Wind Speed st + b4 Elevation s

+ b5 Major Roadss + b6Forest Cover 2011s + b7Point Emissions 2011s

+ ε st (b0 ,t b1,t b2,t b3,t ) ~ N [(0, 0, 0, 0), Ψ ],

where PM2.5,st is the measured ground-level PM2.5 concentration (μg/m3) at site s

on day t; b 0 and b 0,t (day-specific) are the fixed and random intercept, respectively;
AODst is the MODIS AOD value (unitless) at site s on day t; b1 and b1,t (day-specific)
are the fixed and random slopes for AOD, respectively; Relative Humidityst is the
relative humidity at site s on day t; b2 and b2,t (day-specific) are the fixed and random
slopes for relative humidity, respectively; Wind Speedst is the 2-m wind speed (m/s)
at site s on day t; b3 and b3,t (day-specific) are the fixed and random slopes for wind
speed, respectively; Elevations is elevation values (m) at site s; Major Roadss is road
length values (m) at site s; Forest Cover 2011s is forest cover values (unitless) at site s;
Point Emissions 2011s is point emissions (tons per year) at site s; and Ψ is an unstruc-
tured variance–covariance matrix for the random effects.
A 10-fold cross-validation (CV) is conducted to assess the model performance,
and statistical indicators, including the coefficient of determination (R2) and the
square root of the mean squared prediction errors (RMSPE), are calculated.

10.3.3 Results
The 2011 annual mean PM2.5 surface for the Atlanta region is illustrated in Figure 10.1.
The mean PM2.5 concentration for the entire study domain in 2011 is 10.06 μg/m3.

(µg/m3)

N
>11

10.5

10

9.5

<9
0 50 km

FIGURE 10.1 The 2011 annual mean PM2.5 surface in the Atlanta region.
Urban Air Quality Studies Using EO Data 205

The results show that high PM2.5 concentrations appear in urban areas, while low
concentrations occur in rural and suburban areas. In addition, the results also show
that model fitting generates an R2 of 0.7 and an RMSPE of 3.15 μg/m3, while CV gen-
erates an R2 of 0.6 and an RMSPE of 3.45 μg/m3, indicating a good fit of the model.

REFERENCES
Alves, C.A., Pio, C.A., & Duarte, A.C. (2000). Particulate size distributed organic: Compounds
in a forest atmosphere. Environmental Science & Technology, 34, 4287–4293.
Bose, S., Hansel, N., Tonorezos, E., Williams, D., Bilderback, A., Breysse, P., Diette, G., &
McCormack, M.C. (2015). Indoor particulate matter associated with systemic inflam-
mation in COPD. Journal of Environmental Protection, 6, 566–572.
Brauer, M., Hoek, G., Van Vliet, P., Meliefste, K., Fischer, P.H., Wijga, A., Koopman, L.P.,
Neijens, H.J., Gerritsen, J., Kerkhof, M., Heinrich, J., Bellander, T., & Brunekreef, B.
(2002). Air pollution from traffic and the development of respiratory infections and asth-
matic and allergic symptoms in children. American Journal of Respiratory and Critical
Care Medicine, 166, 1092–1098.
Christopher, S.A., Gupta, P., Nair, U., Jones, T.A., Kondragunta, S., Yu-Ling, W., Hand, J.,
& Xiaoyang, Z. (2009). Satellite remote sensing and mesoscale modeling of the 2007
Georgia/Florida Fires. IEEE Journal of Selected Topics in Applied Earth Observations
and Remote Sensing, 2, 163–175.
Chu, D.A., Tsai, T.-C., Chen, J.-P., Chang, S.-C., Jeng, Y.-J., Chiang, W.-L., & Lin, N.-H.
(2013). Interpreting aerosol lidar profiles to better estimate surface PM2.5 for columnar
AOD measurements. Atmospheric Environment, 79, 172–187.
Chudnovsky, A.A., Kostinski, A., Lyapustin, A., & Koutrakis, P. (2012a). Spatial scales of
pollution from variable resolution satellite imaging. Environmental Pollution, 172,
131–138.
Chudnovsky, A.A., Lee, H.J., Kostinski, A., Kotlov, T., & Koutrakis, P. (2012b). Prediction of
daily fine particulate matter concentrations using aerosol optical depth retrievals from
the Geostationary Operational Environmental Satellite (GOES). Journal of the Air &
Waste Management Association, 62, 1022–1031.
Chudnovsky, A.A., Tang, C., Lyapustin, A., Wang, Y., Schwartz, J., & Koutrakis, P. (2013).
A critical assessment of high resolution aerosol optical depth (AOD) retrievals for fine
particulate matter (PM) predictions. Atmospheric Chemistry and Physics Discussions,
13, 14581–14611.
Chung, Y., Dominici, F., Wang, Y., Coull, B.A., & Bell, M.L. (2015). Associations between
long-term exposure to chemical constituents of fine particulate matter (PM(2.5)) and
mortality in Medicare enrollees in the eastern United States. Environmental Health
Perspectives, 123, 467–474.
Deng, X.J., Wu, D., Yu, J.Z., Lau, A.K.H., Li, F., Tan, H.B., Yuan, Z.B., Ng, W.M., Deng, T.,
Wu, C., & Zhou, X.J. (2013). Characterization of secondary aerosol and its extinction
effects on visibility over the Pearl River Delta Region, China. Journal of the Air & Waste
Management Association, 63, 1012–1021.
Dominici, F., Peng, R.D., Bell, M.L., Pham, L., McDermott, A., Zeger, S.L., & Samet,
J.M. (2006). Fine particulate air pollution and hospital admission for cardiovascular
and respiratory diseases. Jama—Journal of the American Medical Association, 295,
1127–1134.
Engel-Cox, J.A., Hoff, R.M., Rogers, R., Dimmick, F., Rush, A.C., Szykman, J.J., Al-Saadi, J.,
Chu, D.A., & Zell, E.R. (2006). Integrating lidar and satellite optical depth with ambient
monitoring for 3-dimensional particulate characterization. Atmospheric Environment,
40, 8056–8067.
206 Remote Sensing for Sustainability

Green, M., Kondragunta, S., Ciren, P., & Xu, C. (2009). Comparison of GOES and MODIS
Aerosol Optical Depth (AOD) to Aerosol Robotic Network (AERONET) AOD and
IMPROVE PM2.5 Mass at Bondville, Illinois. Journal of the Air & Waste Management
Association, 59, 1082–1091.
Habre, R., Moshier, E., Castro, W., Nath, A., Grunin, A., Rohr, A., Godbold, J., Schachter, N.,
Kattan, M., & Coull, B. (2014). The effects of PM2.5 and its components from indoor
and outdoor sources on cough and wheeze symptoms in asthmatic children. Journal of
Exposure Science and Environmental Epidemiology, 24, 380–387.
He, Q.S., Li, C.C., Mao, J.T., Lau, A.K.H., & Chu, D.A. (2008). Analysis of aerosol ver-
tical distribution and variability in Hong Kong. Journal of Geophysical Research-
Atmospheres, 113.
Hodan, W.M., & Barnard, W.R. (2004). Evaluating the Contribution of PM2.5 Precursor Gases
and Re-entrained Road Emissions to Mobile Source PM2.5 Particulate Matter Emissions.
MACTEC Under Contract to the Federal Highway Administration. MACTEC Federal
Programs, Research Triangle Park, NC.
Hu, X., Waller, L.A., Al-Hamdan, M.Z., Crosson, W.L., Estes Jr, M.G., Estes, S.M., Quattrochi,
D.A., Sarnat, J.A., & Liu, Y. (2013). Estimating ground-level PM2.5 concentrations
in the southeastern U.S. using geographically weighted regression. Environmental
Research, 121, 1–10.
Hu, X., Waller, L.A., Lyapustin, A., Wang, Y., Al-Hamdan, M.Z., Crosson, W.L., Estes Jr,
M.G., Estes, S.M., Quattrochi, D.A., Puttaswamy, S.J., & Liu, Y. (2014a). Estimating
ground-level PM2.5 concentrations in the Southeastern United States using MAIAC
AOD retrievals and a two-stage model. Remote Sensing of Environment, 140, 220–232.
Hu, X., Waller, L.A., Lyapustin, A., Wang, Y., & Liu, Y. (2014b). 10-year spatial and temporal
trends of PM2.5 concentrations in the southeastern US estimated using high-resolution
satellite data. Atmospheric Chemistry and Physics, 14, 6301–6314.
Hu, X., Waller, L.A., Lyapustin, A., Wang, Y., & Liu, Y. (2014c). Improving satellite-driven
PM2.5 models with Moderate Resolution Imaging Spectroradiometer fire counts in the
southeastern U.S. Journal of Geophysical Research: Atmospheres, 119, 2014JD021920.
Hu, Z.Y. (2009). Spatial analysis of MODIS aerosol optical depth, PM2.5, and chronic coro-
nary heart disease. International Journal of Health Geographics, 8.
Jaffe, D., Hafner, W., Chand, D., Westerling, A., & Spracklen, D. (2008). Interannual varia-
tions in PM2.5 due to wildfires in the western United States. Environmental Science &
Technology, 42, 2812–2818.
Just, A.C., Wright, R.O., Schwartz, J., Coull, B.A., Baccarelli, A.A., Tellez-Rojo, M.M.,
Moody, E., Wang, Y., Lyapustin, A., & Kloog, I. (2015). Using high-resolution satellite
aerosol optical depth to estimate daily PM2.5 geographical distribution in Mexico City.
Environmental Science & Technology, 49, 8576–8584.
Kahn, R., Banerjee, P., McDonald, D., & Diner, D.J. (1998). Sensitivity of multiangle imag-
ing to aerosol optical depth and to pure-particle size distribution and composition over
ocean. Journal of Geophysical Research—Atmospheres, 103, 32195–32213.
Kahn, R.A., Gaitley, B.J., Martonchik, J.V., Diner, D.J., Crean, K.A., & Holben, B. (2005).
Multiangle Imaging Spectroradiometer (MISR) global aerosol optical depth validation
based on 2 years of coincident Aerosol Robotic Network (AERONET) observations.
Journal of Geophysical Research—Atmospheres, 110.
Kioumourtzoglou, M.-A., Schwartz, J.D., Weisskopf, M.G., Melly, S.J., Wang, Y., Dominici, F.,
& Zanobetti, A. (2016). Long-term PM(2.5) exposure and neurological hospital admis-
sions in the northeastern United States. Environmental Health Perspectives, 124, 23–29.
Kloog, I., Chudnovsky, A.A., Just, A.C., Nordio, F., Koutrakis, P., Coull, B.A., Lyapustin, A.,
Wang, Y., & Schwartz, J. (2014). A new hybrid spatio-temporal model for estimating
daily multi-year PM2.5 concentrations across northeastern USA using high resolution
aerosol optical depth data. Atmospheric Environment, 95, 581–590.
Urban Air Quality Studies Using EO Data 207

Kloog, I., Koutrakis, P., Coull, B.A., Lee, H.J., & Schwartz, J. (2011). Assessing temporally
and spatially resolved PM2.5 exposures for epidemiological studies using satellite aero-
sol optical depth measurements. Atmospheric Environment, 45, 6267–6275.
Kloog, I., Nordio, F., Coull, B.A., & Schwartz, J. (2012). Incorporating local land use regres-
sion and satellite aerosol optical depth in a hybrid model of spatiotemporal PM2.5 expo-
sures in the mid-Atlantic States. Environmental Science & Technology, 46, 11913–11921.
Laden, F., Schwartz, J., Speizer, F.E., & Dockery, D.W. (2006). Reduction in fine particu-
late air pollution and mortality—Extended follow-up of the Harvard six cities study.
American Journal of Respiratory and Critical Care Medicine, 173, 667–672.
Lee, H.J., Liu, Y., Coull, B.A., Schwartz, J., & Koutrakis, P. (2011). A novel calibration
approach of MODIS AOD data to predict PM2.5 concentrations. Atmospheric Chemistry
and Physics, 11, 7991–8002.
Lee, M., Kloog, I., Chudnovsky, A., Lyapustin, A., Wang, Y., Melly, S., Coull, B., Koutrakis,
P., & Schwartz, J. (2015). Spatiotemporal prediction of fine particulate matter using
high-resolution satellite images in the Southeastern US 2003–2011. Journal of Exposure
Science and Environmental Epidemiology, 26, 377–384.
Lin, C., Li, Y., Yuan, Z., Lau, A.K.H., Li, C., & Fung, J.C.H. (2015). Using satellite remote
sensing data to estimate the high-resolution distribution of ground-level PM2.5. Remote
Sensing of Environment, 156, 117–128.
Lin, S., Munsie, J.P., Hwang, S.A., Fitzgerald, E., & Cayo, M.R. (2002). Childhood asthma hospi-
talization and residential exposure to state route traffic. Environmental Research, 88, 73–81.
Liu, Y., Franklin, M., Kahn, R., & Koutrakis, P. (2007a). Using aerosol optical thickness to
predict ground-level PM2.5 concentrations in the St. Louis area: A comparison between
MISR and MODIS. Remote Sensing of Environment, 107, 33–44.
Liu, Y., Koutrakis, P., Kahn, R., Turquety, S., & Yantosca, R.M. (2007b). Estimating fine par-
ticulate matter component concentrations and size distributions using satellite-retrieved
fractional aerosol optical depth: Part 2—A case study. Journal of the Air & Waste
Management Association, 57, 1360–1369.
Liu, Y., Paciorek, C.J., & Koutrakis, P. (2009). Estimating regional spatial and temporal vari-
ability of PM2.5 concentrations using satellite data, meteorology, and land use informa-
tion. Environmental Health Perspectives, 117, 886–892.
Liu, Y., Park, R.J., Jacob, D.J., Li, Q.B., Kilaru, V., & Sarnat, J.A. (2004). Mapping annual
mean ground-level PM2.5 concentrations using Multiangle Imaging Spectroradiometer
aerosol optical thickness over the contiguous United States. Journal of Geophysical
Research—Atmospheres, 109.
Liu, Y., Sarnat, J.A., Kilaru, A., Jacob, D.J., & Koutrakis, P. (2005). Estimating ground-
level PM2.5 in the eastern United States using satellite remote sensing. Environmental
Science & Technology, 39, 3269–3278.
Lyapustin, A., Wang, Y., Laszlo, I., Kahn, R., Korkin, S., Remer, L., Levy, R., & Reid, J.S.
(2011). Multiangle implementation of atmospheric correction (MAIAC): 2. Aerosol
algorithm. Journal of Geophysical Research—Atmospheres, 116.
Ma, Z., Hu, X., Huang, L., Bi, J., & Liu, Y. (2014). Estimating ground-level PM2. 5 in China
using satellite remote sensing. Environmental Science & Technology, 48, 7436–7444.
Ma, Z., Hu, X., Sayer, A.M., Levy, R., Zhang, Q., Xue, Y., Tong, S., Bi, J., Huang, L., &
Liu, Y. (2015). Satellite-based spatiotemporal trends in PM2.5 concentrations: China,
2004–2013. Environmental Health Perspectives, 124, 184–192.
Madrigano, J., Kloog, I., Goldberg, R., Coull, B.A., Mittleman, M.A., & Schwartz, J.
(2013). Long-term exposure to PM2.5 and incidence of acute myocardial infarction.
Environmental Health Perspectives, 121, 192–196.
Miller, K.A., Siscovick, D.S., Sheppard, L., Shepherd, K., Sullivan, J.H., Anderson, G.L., &
Kaufman, J.D. (2007). Long-term exposure to air pollution and incidence of cardiovas-
cular events in women. New England Journal of Medicine, 356, 447–458.
208 Remote Sensing for Sustainability

Neophytou, A.M., Costello, S., Brown, D.M., Picciotto, S., Noth, E.M., Hammond, S.K.,
Cullen, M.R., & Eisen, E.A. (2014). Marginal structural models in occupational epide-
miology: Application in a study of ischemic heart disease incidence and PM2.5 in the
US aluminum industry. American Journal of Epidemiology.
Norris, G., YoungPong, S.N., Koenig, J.Q., Larson, T.V., Sheppard, L., & Stout, J.W. (1999).
An association between fine particles and asthma emergency department visits for chil-
dren in Seattle. Environmental Health Perspectives, 107, 489–493.
Paciorek, C.J., Liu, Y., Moreno-Macias, H., & Kondragunta, S. (2008). Spatiotemporal
associations between GOES aerosol optical depth retrievals and ground-level PM2.5.
Environmental Science & Technology, 42, 5800–5806.
Peters, A., Dockery, D.W., Muller, J.E., & Mittleman, M.A. (2001). Increased particulate air
pollution and the triggering of myocardial infarction. Circulation, 103, 2810–2815.
Puett, R.C., Hart, J.E., Yanosky, J.D., Paciorek, C., Schwartz, J., Suh, H., Speizer, F.E., &
Laden, F. (2009). Chronic fine and coarse particulate exposure, mortality, and coro-
nary heart disease in the Nurses’ Health Study. Environmental Health Perspectives, 117,
1697–1701.
Riediker, M., Cascio, W.E., Griggs, T.R., Herbst, M.C., Bromberg, P.A., Neas, L., Williams,
R.W., & Devlin, R.B. (2004). Particulate matter exposure in cars is associated with car-
diovascular effects in healthy young men. American Journal of Respiratory and Critical
Care Medicine, 169, 934–940.
Schliep, E., Gelfand, A., & Holland, D. (2015). Autoregressive spatially varying coefficients
model for predicting daily PM 2.5 using VIIRS satellite AOT. Advances in Statistical
Climatology, Meteorology and Oceanography, 1, 59–74.
Spira-Cohen, A., Chen, L.C., Kendall, M., Lall, R., & Thurston, G.D. (2011). Personal expo-
sures to traffic-related air pollution and acute respiratory health among Bronx school-
children with asthma. Environmental Health Perspectives, 119, 559–565.
Squizzato, S., Masiol, M., Brunelli, A., Pistollato, S., Tarabotti, E., Rampazzo, G., & Pavoni,
B. (2013). Factors determining the formation of secondary inorganic aerosol: A case
study in the Po Valley (Italy). Atmospheric Chemistry and Physics, 13, 1927–1939.
Sunyer, J., & Basagaña, X. (2001). Particles, and not gases, are associated with the risk of
death in patients with chronic obstructive pulmonary disease. International Journal of
Epidemiology, 30, 1138–1140.
Tian, D., Hu, Y., Wang, Y., Boylan, J.W., Zheng, M., & Russell, A.G. (2009). Assessment of
biomass burning emissions and their impacts on urban and regional PM2.5: A Georgia
case study. Environmental Science & Technology, 43, 299–305.
van Donkelaar, A., Martin, R.V., Brauer, M., Kahn, R., Levy, R., Verduzco, C., & Villeneuve,
P.J. (2010). Global estimates of ambient fine particulate matter concentrations from
satellite-based aerosol optical depth: Development and application. Environmental
Health Perspectives, 118, 847–855.
van Donkelaar, A., Martin, R.V., & Park, R.J. (2006). Estimating ground-level PM2.5 using
aerosol optical depth determined from satellite remote sensing. Journal of Geophysical
Research-Atmospheres, 111.
Wang, J., & Christopher, S.A. (2003). Intercomparison between satellite-derived aerosol
optical thickness and PM2.5 mass: Implications for air quality studies. Geophysical
Research Letters, 30, 2095.
Xie, Y., Wang, Y., Zhang, K., Dong, W., Lv, B., & Bai, Y. (2015). Daily estimation of
ground-level PM2.5 concentrations over Beijing using 3 km resolution MODIS AOD.
Environmental Science & Technology, 49, 12280–12288.
Zeger, S.L., Thomas, D., Dominici, F., Samet, J.M., Schwartz, J., Dockery, D., & Cohen, A.
(2000). Exposure measurement error in time-series studies of air pollution: Concepts
and consequences. Environmental Health Perspectives, 108, 419–426.
Urban Air Quality Studies Using EO Data 209

Zeng, T., Wang, Y., Yoshida, Y., Tian, D., Russell, A.G., & Barnard, W.R. (2008). Impacts of
prescribed fires on air quality over the southeastern United States in spring based on
modeling and ground/satellite measurements. Environmental Science & Technology, 42,
8401–8406.
Zhang, H., Hoff, R.M., & Engel-Cox, J.A. (2009). The relation between Moderate Resolution
Imaging Spectroradiometer (MODIS) Aerosol Optical Depth and PM2.5 over the United
States: A geographical comparison by US Environmental Protection Agency regions.
Journal of the Air & Waste Management Association, 59, 1358–1369.
Zhang, T., Claeys, M., Cachier, H., Dong, S., Wang, W., Maenhaut, W., & Liu, X. (2008).
Identification and estimation of the biomass burning contribution to Beijing aerosol
using levoglucosan as a molecular marker. Atmospheric Environment, 42, 7013–7021.
Zhang, X., Hecobian, A., Zheng, M., Frank, N.H., & Weber, R.J. (2010). Biomass burning
impact on PM2.5 over the southeastern US during 2007: Integrating chemically spe-
ciated FRM filter measurements, MODIS fire counts and PMF analysis. Atmospheric
Chemistry and Physics, 10, 6839–6853.
 11 Heat Hazard Monitoring
with Satellite‑Derived
Land Surface Temperature
Yitong Jiang and Qihao Weng

CONTENTS
11.1 Introduction .................................................................................................. 211
11.2 Methodology ................................................................................................. 213
11.2.1 Study Area ........................................................................................ 213
11.2.2 Data Sets ........................................................................................... 215
11.2.2.1 Data for Linear Spectral Mixture Analysis ....................... 215
11.2.2.2 Data for Downscaling TIR Images .................................... 215
11.2.2.3 Data for ET Estimation ...................................................... 216
11.2.3 Data Processing ................................................................................ 216
11.2.3.1 Data Preparation and Preprocessing .................................. 217
11.2.3.2 Vegetation–Impervious Surface–Soil Model .................... 217
11.2.3.3 Downscaling LST .............................................................. 217
11.2.3.4 The Land Surface Moisture Model .................................... 218
11.3 Results........................................................................................................... 220
11.3.1 Pure Impervious Surfaces, Soil, and Vegetation Cover Pixels ......... 220
11.3.2 The Changes of Spatial Distribution of Daily and Hourly LST ....... 222
11.3.3 Instantaneous LST and Latent Flux over Pure Impervious
Surfaces, Soil, and Vegetation Cover with an Hour or Day Interval.... 222
11.4 Conclusions ................................................................................................... 226
References .............................................................................................................. 227

11.1 INTRODUCTION
Land surface temperature (LST) is a key parameter in heat hazard monitoring. The
trade-off between spatial and temporal resolutions of currently available thermal
infrared (TIR) images and the need for obtaining TIR have been discussed (Agam et
al. 2007; Kustas et al. 2003; Schmugge et al. 1998; Weng 2009). The sensors on polar
orbiting satellites, such as Landsat sensors and the Advanced Spaceborne Thermal
Emission and Reflectance Radiometer (ASTER), can provide TIR data with rela-
tively high spatial resolution. However, their low temporal resolutions are not suf-
ficient for monitoring the diurnal change of LST. Although the Moderate-Resolution
Imaging Spectroradiometer (MODIS) and the Advanced Very High Resolution
Radiometer (AVHRR) produce one to two images per day for the same area, cloud

211
212 Remote Sensing for Sustainability

coverage reduces the usage of the image data and thus increases the time between
two image acquisitions. The Geostationary Operational Environmental Satellite
(GOES) imager on the geostationary satellite has a much higher frequency of obser-
vation, which is every 15 min, but with a much coarser spatial resolution of 4 km.
Therefore, a common solution for characterizing heat waves is to downscale GOES
images from 4 to 1 km while keeping its temporal resolution.
Thermal sharpening, also called downscaling, is the way to increase the spatial
and temporal resolution of TIR images (Weng et al. 2014). The spatial sharpening
techniques focus on downscaling the surface temperature of a sensor to higher reso-
lution with its visible and near-infrared bands (Weng and Fu 2014). The downscaling
of Landsat LST based on its normalized difference vegetation index (NDVI) is an
example of spatial sharpening. Downscaling temporal sharpening techniques are to
downscale surface temperature from a coarser spatial resolution but higher temporal
resolution sensor to generate high temporal resolution images (Weng and Fu 2014).
To downscale GOES or SEVIRI LST to generate hourly or every 15 min LST are
examples of temporal downscaling. There are two major ways to increase the spatial
resolution of TIR images: the emissivity-based method (Nichol 2009) and the veg-
etation cover–based method (Agam et al. 2007; Jiang and Weng 2013; Kustas et al.
2003). The vegetation cover–based method was developed upon the assumption that
vegetation cover is the primary driver of LST variation. Besides vegetation cover,
researchers also discovered other factors that affect LST. Such auxiliary data include
solar irradiation, albedo, topography, thermal inertia, and surface moisture (Zakšek
and Oštir 2012).
The knowledge of surface moisture in urban areas is significant because it links
LST to air temperature. A combination of air temperature, wind speed, humidity,
and short- and long-wave radiation exposure determines a person’s thermal comfort
during urban heat island effects and heat hazard. An accurate estimation of sur-
face moisture would aid the public health reaction during extreme weather events.
Evapotranspiration (ET), the sum of evaporation and plant transpiration, is one of
the important measures of the surface moisture, and it has profound impact on a
person’s thermal comfort. Therefore, the interactions between ET and LST are worth
studying. Transpiration is the process in which water is absorbed by plants’ roots
and is moved to pores on the underside of leaves, where it changes to vapor and is
released to the atmosphere. It is controlled by several atmospheric factors: tempera-
ture, relative humidity, topographic condition, wind and air movement, and soil and
moisture availability (Christensen et al. 2008; Strelcova et al. 2013). In agricultural
and forest areas, the studies on transpiration were usually conducted on individual
or several species (Strelcova et al. 2013). However, in urban areas, urban forests and
urban street trees contain a diverse mix of species from many regions worldwide.
Therefore, the studies of transpiration in urban areas specifically considered the
diversity of the species and hypothesized that species played a more important role
than meteorological variables (Pataki et al. 2011). Urban built environments increase
urban runoff and slow down the infiltration rate, which reduces the transpiration rate
and limits the rooting depth of urban trees (Barthens et al. 2009). Plant density var-
ies in urban areas as well. Hagishima et al. (2007) suggested that small-size plants
and low spatial density vegetation may result in an increase in ET and latent heat
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature 213

flux in urban areas. The land cover of urban surfaces affects urban tree transpira-
tion because energy fluxes and ambient humidity may influence the leaf-to-air vapor
pressure, consequently prolonging or shortening the stomatal closure (Kjelgren and
Montague 1998).
Evaporation over impervious surfaces is one of the least studied topics in the
fields of urban hydrology and microclimatology (Ramamurthy and Bou-Zeid 2014).
The urban area plays an important role in global climate change because of the urban
heat island effects, heat wave hazards, and urban runoff. To measure the impact
of urban climate in regional- or global-scale models, one of the areas that has not
been studied thoroughly is urban surface moisture. Impervious surfaces contribute
to urban evaporation (Ramamurthy and Bou-Zeid 2014). With the LST derived from
satellite images in various combinations of temporal and spatial resolutions, and the
maturity of the techniques of surface temperature derivation, the LST is becom-
ing more and more popular for heat-related hazard monitoring. However, no single
currently available TIR satellite system is capable of capturing all features of ET
dynamics of agricultural fields, such as weekly or monthly trends, day-to-day fluc-
tuations, and peaks (Cammalleri et al. 2014). Therefore, thermal downscaling and
fusion seem promising to combine the advantages of multiple data sets, although the
wide range of spectral, spatial, and temporal resolutions may affect the consistency
of ET estimation (Cammalleri et al. 2014).
In urban areas, the land surface can be depicted by three major components:
vegetation cover, impervious surfaces, and soil (Ridd 1995; Weng and Lu 2009).
In remote sensing, the land cover surface in each pixel can be described by the per-
centage of vegetation cover, impervious surfaces, and soil. The surface moisture on
pure vegetation cover, impervious surfaces, and soil can be measured by vegetation
transpiration and evaporation over impervious surfaces and soil. The objectives of
this study are (1) to produce the instantaneous LST with hourly and daily temporal
resolution by downscaling the GOES LST from 4 to 1 km and MODIS LST from
960 to 30 m; and (2) to estimate the instantaneous latent flux by the downscaled
LST, and compare the variation of LST and latent flux over vegetation, impervious
surfaces, and soil before and after precipitation.

11.2 METHODOLOGY
11.2.1 Study AreA
The study area, Marion County, Indiana, is located in the Midwestern United States
(Figure 11.1). It is the county seat of Indianapolis, the capital and largest city in
the State of Indiana. Marion County is part of the Indianapolis–Carmel–Anderson
Metropolitan Statistical Area. It covers both the urban and suburban areas of the
city of Indianapolis. The county is located on a flat plain, which makes it possible
for the city to expand in all directions. Indianapolis was the 14th largest city in the
United States by population in 2014. The area lies between latitudes 39°55′37.75″ to
39°37′54.28″ and longitudes −86°19′33.65″ to −85°56′10.11″. According to the 2010
Census, the total area of Marion County is 403.01 square miles. 98.34% is land, and
1.66% is water. According to Census 2000, Indianapolis is the most populous city in
 86°20´0˝W 86°15´0˝W 86°10´0˝W 86°5´0˝W 86°0´0˝W
214

39°55´0˝N
39°55´0˝N
100°0´0˝W 90°0´0˝W 80°0´0˝W 70°0´0˝W

50°0´0˝N

50°0´0˝N
Montreal

39°50´0˝N
39°50´0˝N

UNITED Boston
STATES Chicago Detroit
New York
Philadelphia
Washington

40°0´0˝N
Marion Country, Indiana

40°0´0˝N
Atlanta
Dallas
39°45´0˝N

39°45´0˝N
Houston ,

30°0´0˝N
30°0´0˝N
Esri, HERE, DeLorme, Mapmy India
© Open Street Map contributors, and
Gulf of the GIS user community
MEXICO Mexico
Havana
100°0´0˝W 90°0´0˝W 80°0´0˝W 70°0´0˝W

0 375 750 1500 2250 3000

Miles
39°40´0˝N

39°40´0˝N
N

86°20´0˝W 86°15´0˝W 86°10´0˝W 86°5´0˝W 86°0´0˝W

0 2.5 5 10 15 20
Miles
Remote Sensing for Sustainability

FIGURE 11.1 The location of the study area, Marion County, Indiana.
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature 215

the State of Indiana and the 12th largest city in the United States. From 2000 to 2010,
the population continued to increase from 781,926 to 820,445. The rate of increase is
4.9%. The estimated population in 2015 is 853,173.
Indianapolis has undergone land use/land cover (LULC) change because of the
urbanization process. According to the National Land Cover Database (NLCD)
2006, the conversion to developed land from 2001 to 2006 mainly took place in
the suburban areas, specifically between the circle of Interstate Highway 465 and
the county boundary, with the largest changes in the Southern and Eastern fringes.
Sparse land cover change in terms of density took place in the urban areas within
the circle of Highway 465. According to the NLCD 2001–2006 land cover change
data, 3.65% of the total land cover was changed from 2001 to 2006 in Indianapolis,
Indiana, among which, 51% were changed from cultivated land to developed land,
31% were changed from a lower level to a higher level of developed land (e.g., open
space to low density developed, medium density developed to high density devel-
oped), and 3.6% were changed from forest to developed.
From 2006 to 2011, more changes from lower intensity to higher intensity of devel-
oped land than from cultivated crop to developed land were found in Indianapolis.
Among the total changed land covers, the first three biggest amounts of land cover
change were from a lower level of developed land cover to a higher level of devel-
oped land cover: 21.7% were changed from Developed Open Space, to Developed,
Medium Intensity. 12.1% were changed from Developed, Lower Intensity to Medium
Intensity, and 11.2% were changed from Developed Open Space to High Intensity.
Changes from cultivated crops to open space and low-intensity developed land were
the next two categories; they were 9.6% and 8.1%, respectively. Changes from decid-
uous forest to developed open space, low intensity, and medium intensity were rela-
tively less; they were 2.9% in total. Changes from pasture/hay to developed open
space and low intensity were 3.0%.

11.2.2 dAtA SetS

11.2.2.1 Data for Linear Spectral Mixture Analysis
Landsat TM images that overpass path 21 row 32 on June 17, 2001, and July 1, 2006,
were used for applying Linear Spectral Mixture Analysis (LSMA) and producing
images of the percentage of vegetation in each pixel. The percentage of impervious
surfaces is from NLCD 2001 and 2006.

11.2.2.2 Data for Downscaling TIR Images

GOES-8 band 4 and band 5 images from June 15 to 17, 2001, were selected for down-
scaling. These images were chosen because they were acquired within 48 h after a
precipitation, because according to Ramamurthy and Bou-Zeid (2014), the highest
evaporation from impervious surface were found 48 h after precipitations. The aux-
iliary data included MODIS NDVI and enhanced vegetation index products; albedo
in visible, NIR, and short-wave bands; emissivity in MODIS band 31 and band 32;
Shuttle Radar Topography Mission (SRTM) digital elevation data; and percentage of
imperviousness from NLCD 2001.
216 Remote Sensing for Sustainability

MODIS LST products acquired from DOY 179 to 200 in 2006 were used to down-
scale to Landsat resolution. The auxiliary data included NDVI, fractional vegetation
cover, emissivity, broadband albedo from the Landsat image collected on July 1,
2006, as well as the SRTM elevation data.

11.2.2.3 Data for ET Estimation

Hourly solar radiation data were acquired from the National Solar Radiation
Database (n.d.). Daily solar radiation was configured by averaging hourly data.
Hourly wind speed, atmospheric temperature, air pressure, and relative humidity
were acquired from the Indiana State Climate office (iClimate.org). The data were
collected in the International Airport of Indianapolis. Percent of Imperviousness
was downloaded from NLCD 2011. Downscaled LSTs were produced by GOES
and MODIS from Section 11.2.2.2, and fractional vegetation cover data were
configured by LSMA in Section 11.2.2.1.

11.2.3 dAtA ProceSSing

The flowchart of methodology is shown in Figure 11.2. It is a combination of LST
downscaling, LSMA, and the Two-Source Energy Balance (TSEB) model. In the
LST downscaling, the hourly LST was produced by downscaling GOES LST from
4 to 1 km; the daily LST was produced by downscaling MODIS LST from 960 to
30 m. In the LSMA, the percentage of soil, vegetation, and impervious surfaces in
each pixel was calculated. The TSEB model generated the instantaneous latent flux,
and latent flux over pure soil, vegetation, and impervious surfaces was compared on
an hourly and daily bases.

GOES, MODIS, Landsat

Downscaling

Finer scale instantaneous LST each hour

and each day at satellite passing time
TSEB
Landsat, NLCD

LSMA Net, ground, sensible fluxes

%Soil %Veg %Impervious Latent flux

Latent flux over three types of surfaces

FIGURE 11.2 Flowchart of methodology.

Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature 217

11.2.3.1 Data Preparation and Preprocessing

The atmospheric effect of Landsat TM images was corrected using the simplified
dark object subtraction method (Song et al. 2001), and the reference values rescaling
gain factor and bias factor values were from Chander et al. (2009). Landsat LSTs
were computed according to Coll et al. (2010). GOES LSTs were computed by split-
window algorithm according to Jiménez-Muñoz and Sobrino (2008).

11.2.3.2 Vegetation–Impervious Surface–Soil Model

LSMAs were applied to derive fractional vegetation cover. First, principal compo-
nent analysis (PCA) was applied to all the spectral bands of Landsat images, and
the highest ranked components were selected and plotted. Second, the endmembers
were selected. According to Johnson et al. (1992), the potential endmembers lay at
the vertices of these PCA bands’ scatterplots. The three selected endmembers were
vegetation, soil, and impervious surfaces. Details about the selection of endmembers
and the estimation of the fraction were discussed in Weng et al. (2008). To estimate
the fraction of impervious surface, different combinations of three or four endmem-
bers were compared. The endmembers included high albedo, low albedo, vegetation,
and soil. Since this study was conducted in an urban area, the criteria for selecting
the best-suited fraction images included high-quality fraction images for urban land-
scape, low error, and the distinction among typical LULC types. The fractional soil
cover was produced based on the percentage of impervious surface and vegetation
cover—besides the vegetation and impervious surfaces, the rest of each pixel was
assumed to be soil. Water bodies were masked out.

11.2.3.3 Downscaling LST

The GOES-8 Imager possesses one visible band and four infrared bands. The data
were downloaded from the National Oceanic and Atmospheric Administration
(NOAA) comprehensive large array-data stewardship system (http://www.nsof
.class.noaa.gov/saa/products/search?datatype_family=GVAR_IMG). The NOAA
Weather and Climate Toolkit was used to export the AREA files to TIFF for-
mat. AREA files are count data with calibration coefficients. The toolkit uses the
calibration information contained in the calibration block of the AREA file and
converts the raw counts (10 bit precision) to brightness temperatures for the IR
channels. Then, GOES LSTs were retrieved using the split-window algorithm and
coefficients from Jiménez-Muñoz and Sobrino (2008). Since LST is strongly influ-
enced by parameters such as solar irradiation, albedo, topography, thermal inertia,
and vegetation cover (Weng et al. 2004), the corresponding auxiliary data were
obtained for downscaling. The downscaling method followed Zakšek and Oštir
(2012). First, PCA was applied on auxiliary data in finer spatial resolution; second,
the highest ranked principal components were upscaled to coarser resolution; third,
regressions between LST and upscaled components were formed; last, the LSTs
at finer spatial resolution were estimated by finer resolution components and the
regressions (Figure 11.3). The GOES LSTs were downscaled from 4 to 1 km every
hour; the MODIS LSTs are downscaled from 960 to 240 m, and then from 240 to
30 m each day at the satellite passing time.
218 Remote Sensing for Sustainability

Low-
High- resolution
resolution thermal
auxiliary bands
data

Split window
PCA

Low-
High- Low- resolution
resolution Upscaling resolution LST
components components

Low- Regression
resolution
auxiliary
data High-
resolution High-
LST Validation resolution
LST

FIGURE 11.3 The flowchart for LST downscaling.

11.2.3.4 The Land Surface Moisture Model

The method of estimating LE followed Weng et al. (2013). Net radiation (Rn) was
calculated as

Rn = (1 − α) Rshort + εεa σTa4 − εσTs4 , (11.1)

where R n is the net radiation, σ is the Stefan–Boltzmann constant (5.67 × 10 −8W/

m 2K4), εa is atmospheric emissivity, α is broadband albedo, R short is short-wave
radiation, ε is surface broadband emissivity, Ta is atmospheric temperature, and
Ts is surface temperature. Broadband albedo was calculated according to Liang
(2000), and broadband emissivity was calculated according to Ogawa et al.
(2003).
Atmospheric emissivity was calculated as

1/ 7
e 
εa = 1.24  a  , (11.2)
 Ta 

where ea is atmospheric water vapor pressure, which was estimated based on satura-
tion water vapor pressure and relative humidity (Weng et al. 2013).
Ground heat flux was calculated based on net radiation and a coefficient (cg) that
describes the influence of surface cover material, seasonality, and diurnal change.
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature 219

The land cover was classified into water, bare soils, grass, forest, urban, and agricul-
ture, and cg values vary among different land cover types:

G = cg × Rn. (11.3)

Since the study area is the same, the cg values were adopted from Weng et al.
(2013) in Table 11.1.
Sensible heat flux (H) was calculated separately for non-vegetated areas and veg-
etated areas. The non-vegetated areas and vegetated areas are differentiated by the
amount of vegetation cover, and the value of 0.5 was used as the threshold:

H = fnon-veg Hnon-veg + fveg Hveg (11.4)

Ts − Ta
H non-veg = ρa cp (11.5)
RAH + Rs

Tc − Ta
H veg = ρa cp , (11.6)
RAH

where ρa is the air density in kg/m³, cp is the specific heat of air at constant pressure
in J/(kg·K), Ts and Tc are surface temperature for non-vegetated and vegetated areas,
and Ta is air temperature. RAH is the aerodynamic resistance in s/m, and Rs is the
resistance to heat flow in the boundary layer immediately above soil surface. RAH
was calculated as

  zu − d 0     z − d0  
 ln   − Ψ M   ln  t  − ΨH 
 z0 M     z0 H 
RAH = 
. (11.7)
k 2u

TABLE 11.1
cg Values According to LULC Types
for Ground Heat Flux
LULC Types cg
Water 0.35
Bare soils 0.30
Grass 0.30
Forest 0.15
Urban 0.40
Agriculture 0.30
220 Remote Sensing for Sustainability

According to Weng et al. (2013), zu and zt are the heights at which the wind speed
u and atmospheric temperature are measured, which are 10 and 2 m, respectively. d 0
is the displacement height, and z0M and z0H are the roughness lengths for momentum
and heat transport, respectively. ΨM and ΨH are stability correction functions for
momentum and heat, respectively, and k is von Karman’s constant, which is equal
to 0.4 (Weng et al. 2013). This equation may be simplified by removing ΨM and ΨH.
Rs can be calculated as

1
Rs = , (11.8)
a + bus

where a is the free convective velocity, which is equal to 0.04 m/s, b is a coefficient
to represent the typical soil surface roughness, and us is the wind speed over soil
surface at the height of 0.05–0.2 m.
LE was also calculated for vegetated and non-vegetated areas separately:

LE = fnon-vegLEnon-veg + fvegLEveg (11.9)

LEnon-veg = Rn,non-veg − G − Hnon-veg (11.10)

∆
LE veg = α PT fG Rn,veg, (11.11)
∆+γ

where αPT is the Priestley–Taylor parameter, which is equal to 1.26, γ is the psychro-
metric constant, Δ is the slope of saturation vapor pressure–temperature curve. fG is
the fraction of the leaf area index that is green, and it is equal to unity when it is not
available. If LEnon-veg is negative, it was set to zero, Hnon-veg was recomputed as the
residual of Equation 11.10.

11.3 RESULTS
11.3.1 Pure imPerviouS SurfAceS, Soil, And vegetAtion cover PixelS
The spatial distribution of vegetation, soil, and impervious surface fractions, as well
as the pure pixels of the three land cover types (red dots), are shown in Figure 11.4.
Figure 11.4a through c are in 30 m resolution, and Figure 11.4d through f are in
1 km resolution. A resolution of 30 m shows the distribution of the three land types
much clearer than the 1-km resolution. The high-percentage impervious surfaces
were distributed in the central urban areas, along the major roads and highways, and
in the commercial sites along the highway. The vegetation cover is mainly located
in the urban forest along the White River, beside Eagle Creek Reservoir and Geist
Reservoir, in low-density residential land in the north and in agricultural fields in the
south. The distribution of pure soil pixels was mainly in the agriculture fields, open
space, and airport in the south.
 0 2.5 5 10 15 20 0 2.5 5 10 15 20 0 2.5 5 10 15 20
Miles Miles Miles
Pure impervious pixels Pure vegetation pixels Pure soil pixels N
% Imperviousness at 30 m % Vegetation cover at 30 m % Soil at 30 m
0–0.09 0.39–0.52 –0.32–0.10 0.37–0.48 –0.67––0.07 0.21–0.33
0.09–0.25 0.52–0.67 0.10–0.25 0.48–0.59 –0.07––0.08 0.33–0.48
0.25–0.39 0.67–0.83 0.25–0.37 0.59–0.69 0.08–0.21 0.48–0.66
0.83–1 0.69–1.24 0.66–1.32
(a) (b) (c)

0 2.5 5 10 15 20 0 2.5 5 10 15 20
0 2.5 5 10 15 20 Miles
Miles Miles % Soil at 1 km
% Imperviousness at 1 km % Vegetation cover at 1 km
0–0.09 0.41–0.53 –0.04–0.10 0.35–0.45 –0.76––0.27 0.18–0.36
0.09–0.26 0.53–0.66 0.10–0.24 0.45–0.56 –0.27––0.02 0.36–0.54
0.26–0.41 0.66–0.99 0.24–0.35 0.56–0.89 0.02–0.18 0.54–0.99
0.99 and above 0.89–0.90 1.00–1.03
(d) (e) (f )
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature

FIGURE 11.4 The percentage of imperviousness, vegetation cover, and soil in each pixel. The red points are the locations of pure impervious surface, vegeta-
tion cover, and soil pixels. (a) Percentage of imperviousness at 30 m resolution; (b) percentage of vegetation cover at 30 m resolution; (c) percentage of soil at
30 m resolution; (d) percentage of imperviousness at 1 km resolution; (e) percentage of vegetation at 1 km resolution; and (f) percentage of soil at 1 km resolution.
221
222 Remote Sensing for Sustainability

In the 1-km-resolution images, there are no 100% pure impervious surface and
vegetation cover pixels. Therefore, the highest concentration pixels with 99% of
impervious surface and 89% of vegetation cover were used as the pure pixels. The
pure pixels of the three land cover types were selected to monitor the characteristics
of the daily and hourly LST and Latent flux changes over the tree land cover types.

11.3.2 the chAngeS of SPAtiAl diStribution of dAily And hourly lSt

The spatial and temporal variability of LST at 1 km resolution is shown in Figure
11.5. It was produced using GOES 4 km LST and the downscaling technique. Taking
June 16, 2001, for example, from 9:00 to 15:00, the downscaled LST was able to
record the changes in spatial distribution over time. At 9:00, the built-up areas were
300–305 K, and the vegetated areas were 295–300 K; at 10:00, the LST for the
whole area increased 5–10 K; at 11:00, the LST of central urban area, airport, and
commercial land along the Interstate Highway reached 315 K; at noon, the built-
up area remained above 310 K; the high LST shrank at 14:00, and the LST of the
whole area decreased below 310 K at 15:00. The highest LST was found at the center
of Indianapolis, Indiana, and it lasted longer than surrounding areas. Compared to
4-km-resolution images, the heat hazard can be located more accurately and effi-
ciently with the increased spatial and temporal resolution. With exposure and vul-
nerability information, the heat hazard risk areas can be decided (Jiang et al. 2015).

11.3.3 inStAntAneouS lSt And lAtent flux over Pure imPerviouS SurfAceS,
Soil, And vegetAtion cover with An hour or dAy intervAl
The instantaneous LST and latent flux over vegetation, soil, and impervious sur-
face show various responses to rainfall during multiple precipitations in July 2006
(Figure 11.6). The three land cover types had a similar pattern of LST change. In
particular, the LSTs of soil and impervious surfaces were parallel to each other.
This is because the impervious surface usually has the highest sensible heat. Soil
moisture and evaporation from soil may lower down the surface temperature. The
pattern of LST of vegetation cover is a little different: it tended to have smoother
change of LST during rainfalls. It may attributed to evaporation and transpiration,
which adjusted the direct impact from solar radiation and reflectance from nearby
objects. The impervious surfaces had the highest LST before and after the precipi-
tations, it was 1–2 K higher than soil, and 8–10 K higher than vegetation cover.
Precipitation tended to make the LST difference between the three land surfaces
smaller. After the 6.32-cm precipitation on 193 days of the year, the LST of imper-
vious surfaces dropped 2 K, and the LST of soil surfaces dropped less than 2 K in
2 days. Precipitation has much smaller impact on LST over vegetation cover. There
is no apparent temperature drop after precipitation over vegetation cover.
Figure 11.6b explains the LE changes over three land cover types over time.
Vegetation cover and soil surface share the similar pattern of latent flux change dur-
ing multiple precipitations. The latent flux of vegetation cover and soil was high
not long after the 0.64-cm precipitation on DOY 179, and it gradually dropped
until it received the next rainfall. After the 6.32-cm precipitation, the latent flux of
 0 2.5 5 10 15 20 0 2.5 5 10 15 20 0 2.5 5 10 15 20
Miles Miles Miles

09:00 10:00 11:00

0 2.5 5 10 15 20Miles0 2.5 5 10 15 20 0 2.5 5 10 15 20 Miles

Miles
12:00 14:00 15:00
N Land surface temperature (K)

0
0

05
0 5
32

–
–

–3
31 31
5– 95
in
0
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature

9
15

2 2
M
3

30 30
0
3 3
5– 10–
223

FIGURE 11.5 Selected hourly LSTs at 1 km resolution on June 16, 2001.

224 Remote Sensing for Sustainability

310 7
308
6
306
5

Precipitation (cm)
304
302 4
LST (K)

300
298 3
296 2
294
1
292
290 0
175 180 185 190 195 200 205
DOY 2006
Precipitation Impervious Soil Vegetation
(a)

400.00 7
350.00 6
300.00
5

Precipitation (cm)
250.00
LE (W/m2)

4
200.00
3
150.00
2
100.00
50.00 1

0.00 0
175 180 185 190 195 200 205
DOY 2006
Precipitation Impervious Soil Vegetation
(b)

FIGURE 11.6 The instantaneous LST (a) and latent flux (b) in July 2006.

vegetation cover and soil increased and then dropped within a week. The pattern of
latent flux over impervious surfaces is markedly different from the natural land cov-
ers such as vegetation and soils. The LE over impervious surfaces generated sharp
increase and decrease within a week after the rainfall, and it was higher than soil
surfaces and close to the amount from vegetation. Lacking the ability to store water,
the impervious surfaces contribute considerable evaporation in the urban area and
thus alter the natural energy and water cycle.
Figure 11.7a shows the changes of LST before and after a 2.54-cm precipitation.
During and shortly after the precipitation, there is a drop of LST from three types of
land cover types, among which the LST over soil decreased more rapidly, and LST
of impervious surfaces remained higher compared to vegetation and soil. There was
a sudden drop of LST around midnight. The LST of the three land cover types in the
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature 225

second day was relatively smooth. The highest LST from impervious surface and the
LST of soil and impervious surface were almost the same. The figure only shows
48 h after the precipitation.
Figure 11.7b shows the change of latent flux over the three land cover types after
the 2.54-cm precipitation. During the rainfall, the latent flux from soil is much higher
than that from vegetation and impervious surfaces. However, after the precipitation, the
latent flux over impervious surface is more than 100 W/m2 higher than soil and more
than 250 W/m2 higher than vegetation cover at noon. This result indicates that the evap-
oration over impervious surfaces may remain high for more than 48 h in the Midwest.

320 1.40

310 1.20

1.00

Precipitation (cm)
300
0.80
LST (K)

290
0.60
280
0.40

270 0.20

260 0.00
10:30
12:30
14:30
16:30
18:30
20:30
22:30
0:30
2:30
4:30
6:30
8:30
10:30
12:30
14:30
16:30
18:30
20:30
22:30
0:30
2:30
4:30
6:30
8:30
10:30
6/15 6/16 6/17
Precipitation Impervious Soil Vegetation
(a)

700.00 1.40
600.00 1.20
500.00
Precipitation (cm)

1.00
400.00
LE (W/m2)

300.00 0.80
200.00 0.60
100.00
0.40
0.00
–100.00 0.20
–200.00 0.00
10:30
12:30
14:30
16:30
18:30
20:30
22:30

10:30
12:30
14:30
16:30
18:30
20:30
22:30

10:30
0:30
2:30
4:30
6:30
8:30

0:30
2:30
4:30
6:30
8:30

6/15 6/16 6/17

Precipitation Impervious Soil Vegetation
(b)

FIGURE 11.7 The hourly LST (a) and latent flux (b) in June 2001 after precipitation.
226 Remote Sensing for Sustainability

11.4 CONCLUSIONS
This chapter introduced a case study of using the satellite-derived LST to moni-
tor the heat hazard. It focuses on two aspects: first, using the thermal downscal-
ing technique to increase the spatial and temporal resolution; second, applying the
TSEB model in urban areas, categorizing the urban land cover into impervious sur-
faces, soil, and vegetation, and discussing the characteristics of LST, evaporation,
and transpiration over three land cover types. This research is meaningful because
the surface moisture in urban areas is one of the least studied field. To monitor the
heat hazard accurately, the impacts from surface moisture cannot be ignored, and
an efficient way to estimate air temperature based on the satellite-derived LST is
desirable.
The overall methodology was a combination of LST downscaling, LSMA, and
the TSEB model. In the LST downscaling, the hourly LST was produced by down-
scaling GOES LST from 4 to 1 km; the daily LST was produced by downscaling
MODIS LST from 960 to 30 m. In the LSMA, the percentage of soil, vegetation,
and impervious surfaces in each pixel was calculated. The TSEB model generated
the instantaneous latent flux, and latent flux over pure soil, vegetation, and imper-
vious surfaces was compared on an hourly and daily basis. The approach of using
the TSEB model to estimate the urban surface energy balance maximized the use
of multispectral remote sensing data and minimized the amount of meteorologi-
cal data from weather stations. In addition, to increase the spatial and temporal
resolution of LST, the downscaling process also made use of the satellite data,
including the thermal bands from GOES and MODIS and the auxiliary data at
finer resolutions.
The findings for the study site are the temporal variability of surface mois-
ture, namely, the latent flux from vegetation, soil, and impervious surfaces in
Indianapolis, Indiana. Using soil and vegetation as control, the impervious sur-
faces contribute to urban evaporation after precipitation. This finding agrees with
Ramamurthy and Bou-Zeid (2014) in a small scale with the Princeton urban can-
opy model. This indicates that the approached method is applicable in urban areas
with large scales.
The study succeeds in accounting for urban moisture in the following aspects: the
LSMA picked the pure pixels, LST downscaling successfully generated finer spatial
and temporal scale LST, and TSEB successfully estimated the similar pattern of ET
over vegetation and soil surfaces and the prompt response from impervious surfaces
after precipitation. Therefore, the proposed method is promising in the estimation of
ET over vegetation, soil, and impervious surfaces in urban areas.
In the original model, the parameters describing the roughness condition of land
surface, namely, d 0, z0M, and z0H, were fixed according to the LULC types. Although
each land cover type may yield similar morphological parameters, each city has its
own unique morphological characteristics, which relate to history, culture, tradition,
planning, and topography; it varies over regions, countries, and continents. Contexts
that affect the urban setting include hydraulic factors, transportation hubs, trade cen-
ters, defensive sites, and religious factors. Therefore, in future research, the estima-
tion of morphological parameters from LiDAR data is highly desirable.
Heat Hazard Monitoring with Satellite‑Derived Land Surface Temperature 227

REFERENCES
Agam, N., Kustas, W. P., Anderson, M. C., Li, F., and Neale, C. M. U. (2007). A vegetation
index based technique for spatial sharpening of thermal imagery. Remote Sensing of
Environment, 107(2007), 545–558.
Barthens, J., Day, S. D., Harris, J. R., Wynn, T. M., and Dove, J. E. (2009). Transpiration and
root development of urban trees in structural soil stormwater reservoirs. Environmental
Management, 44(4), 646–657.
Cammalleri, C., Anderson, M. C., Gao, F., Hain, C. R., and Kustas, W. P. (2014). Mapping
daily evapotranspiration at field scales over rainfed and irrigated agricultural areas using
remote sensing data fusion. Agricultural and Forest Meteorology, 186(2014), 1–11.
Chander, G., Markham, B. L., & Helder, D. L. (2009). Summary of current radiometric
calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote
Sensing of Environment, 113, 893–903.
Christensen. L., Tague, C. L., and Baron, J. S. (2008). Spatial patterns of simulated transpiration
response to climate variability in a snow dominated mountain ecosystem. Hydrological
Processes, 22(18), 3576–3588.
Coll, C., Galve, J. M., Sanchez, J. M., and Caselles, V. (2010). Validation of Landsat-7/ETM+
thermal-band calibration and atmospheric correction with ground-based measurements.
IEEE Transactions on Geoscience and Remote Sensing, 48, 547–555.
Hagishima, A., Narita, K. I., and Tanimoto, J. (2007). Field experiment on transpiration from
isolated urban plants. Hydrological Processes, 21(9), 1217–1222.
Jiang, Y., Fu, P., and Weng, Q. (2015). Downscaling GOES land surface temperature for
assessing heat wave health risks. IEEE Geoscience and Remote Sensing Letters, 12(8),
1605–1609.
Jiang, Y. and Weng. Q. (2013). Estimating LST using a vegetation-cover-based thermal sharpen-
ing technique. IEEE Geoscience and Remote Sensing Letters, 10(5), 1249–1252.
Jiménez-Muñoz, J. and Sobrino, J. A. (2008). Split-window coefficients for land surface tem-
perature retrieval from low-resolution thermal infrared sensors. IEEE Geoscience and
Remote Sensing, 5(4), 806–809.
Johnson, P. E., Smith, M. O., and Adams, J. B. (1992). Simple algorithms for remote determi-
nation for mineral abundances and particles sizes from reflectance spectra. Journal of
Geophysical Research, 97, 2649–2657.
Kjelgren, R., and Montague, T. (1998). Urban tree transpiration over turf and asphalt surfaces.
Atmospheric Environment, 32(1), 35–41.
Kustas, W. P., Norman, J. M., Anderson, M. C., and French, A. N. (2003). Estimating subpixel
surface temperatures and energy fluxes from the vegetation index–radiometric tempera-
ture relationship. Remote Sensing of Environment, 85(2003), 429–440.
Liang, S. (2000). Narrowband to broadband conversions of land surface albedo I: Algorithms.
Remote Sensing of Environment, 76, 213–238.
National Solar Radiation Database (n.d.). Retrieved from http://rredc.nrel.gov/solar/old_data
/ nsrdb/1991-2010/targzs/targzs_by_state.html.
Nichol, J. E. (2009). An emissivity modulation method for spatial enhancement of thermal
satellite images in urban heat island analysis. Photogrammetric Engineering & Remote
Sensing, 60(5), 547–556.
Ogawa, K., Schmugge, T., Jacob, F., and French, A. (2003). Estimation of land surface window
(8–12 μm) emissivity from multispectral thermal infrared remote sensing—A case study
in a part of Sahara Desert. Geophysical Research Letters, 30(2), 1067.
Pataki, D. E., McCarthy, H. R., Litvak, E., and Pincetl, S. (2011). Transpiration of urban forests
in the Los Angeles metropolitan area. Ecological Applications, 21(3), 661–677.
Ramamurthy, P. and Bou-Zeid, E. (2014). Contribution of impervious surfaces to urban evapora-
tion. Water Resources Research, 50, 2889–2902.
228 Remote Sensing for Sustainability

Ridd, M. K. (1995). Exploring a V–I–S (Vegetation–Impervious Surface–Soil) model for

urban ecosystem analysis through remote sensing: Comparative anatomy for cities.
International Journal of Remote Sensing, 16, 2165–2185.
Schmugge, T. J., Kustas, W. P., and Humes, K. S. (1998). Monitoring land surface fluxes using
ASTER observations. IEEE Transactions on Geoscience and Remote Sensing, 36(5),
1421–1430.
Song, C., Woodcock, C. E., Seto, K. C., Lenney, M. P. and Macomber, S. A. (2001). Classification
and change detection using Landsat TM data: When and how to correct atmospheric
effects? Remote Sensing of Environment, 75, 230–244.
Strelcova, K., Kurjak, D., Lestianska, A., Kovalcikova, D., Ditmarova, L., Skvarenina, J., and
Ahmed, Y. A. R. (2013). Differences in transpiration of Norway spruce drought stressed
trees and trees well supplied with water. Biologia, 68(6), 1118–1122.
Weng, Q. (2009). Thermal infrared remote sensing for urban climate and environmental studies:
Methods, applications, and trends. ISPRS Journal of Photogrammetry and Remote Sensing,
64(4), 335–344.
Weng, Q. and Fu, P. (2014). Modeling diurnal land temperature cycles over Los Angeles using
downscaled GOES imagery. ISPRS Journal of Photogrammetry and Remote Sensing,
97, 78–88.
Weng, Q., Fu, P., and Gao, F. (2014). Generating daily land surface temperature at Landsat
resolution by fusing Landsat and MODIS data. Remote Sensing of Environment, 145,
55–67.
Weng, Q., Hu, X., and Lu, D. (2008). Extracting impervious surfaces from medium spatial res-
olution multispectral and hyperspectral imagery: A comparison. International Journal
of Remote Sensing, 29, 3209–3232.
Weng, Q., Hu, X., Quattrochi, D. A., and Liu, H. (2013). Assessing intra-urban surface energy
fluxes using remotely sensed ASTER imagery and routine meteorological data: A
case study in Indianapolis, U.S.A. IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing, doi: 10.1109/JSTARS.2013.2281776.
Weng, Q. and Lu D. (2009). Landscape as a continuum: An examination of the urban landscape
structures and dynamics of Indianapolis city, 1991–2000, by using satellite images.
International Journal of Remote Sensing, 30(10), 2547–2577.
Weng, Q., Lu, D., and Schubring J. (2004). Estimation of land surface temperature-vegetation
abundance relationship for urban heat island studies. Remote Sensing of Environment,
89(2004), 467–483.
Zakšek, K. and Oštir, K. (2012). Downscaling land surface temperature for urban heat island
diurnal cycle analysis, Remote Sensing of Environment, 117, 114–124.
 12 Remote Sensing
Identification of
Threshold Zones along
a Mediterranean to
Arid Climatic Gradient
Maxim Shoshany

CONTENTS
12.1 Introduction .................................................................................................. 229
12.2 Ecosystem Transitions along a Climatic Gradient........................................ 230
12.3 Remote Sensing Techniques and Their Implementation .............................. 231
12.3.1 Green Vegetation Cover .................................................................... 231
12.3.2 Life-Form Compositions................................................................... 235
12.3.3 Inverse Erosion and Recovery Potentials ......................................... 237
12.4 Summary and Conclusions ...........................................................................240
References .............................................................................................................. 241

12.1 INTRODUCTION
Debates regarding desertification have been heated especially since Thomas and
Middleton (1994) challenged the marching desert myth in response to earlier sug-
gestions by Lamprey (1975) that the Sahara desert boundary in Sudan is advancing
by 5.5 km every year. Controversies in interpreting spatial and temporal informa-
tion regarding shifts in desert threshold zones are not limited to the Sahel region
(e.g., Nicholson 2011) but also characterize other semiarid regions (Veron et al.
2006) and the semiarid margins of the Mediterranean in particular (Safriel 2009;
Thornes 2000). Although there is a consensus regarding the important role of remote
sensing in monitoring desertification, there are disagreements regarding the inter-
pretation of satellite imagery for this purpose. This is well exemplified by the
disagreement of Hein and De Ridder (2006) with claims regarding greening of the
Sahel made by Herrmann et al. (2005). According to Veron et al. (2006), contro-
versies regarding desertification emerge mainly from methodological and termino-
logical differences. They suggest that improvements in exploring desertification over
large regions may be obtained by strengthening the synergy between ecology and

229
230 Remote Sensing for Sustainability

Beit Guvrin Dense shrublands

of Avisur Highland

Judean

Desert
Mountains
Open shrublands
of Amazia
West to east transect
North to south transect
0

Judean
30
0
20
Desert fringe batha Negev 0
north of Dvira Forest Desert 10

Arid ecosystems
south of Lehavim

FIGURE 12.1 Map of the climatic gradient of the Judean Mountains with photographs of
the four main ecosystems along the gradient and location of transects.

remote sensing. Within such an approach, this remote sensing study implements veg-
etation phenomenologies in investigating ecosystems’ transition between semiarid
Mediterranean to arid region. Three modes of vegetation change are analyzed: green
vegetation cover (GVC) change between years of extreme high and low rainfall lev-
els, life-forms’ composition change, and variations in spatial erosion versus recovery
potentials. Landsat images of the climatic gradient between the Judean Mountains
in central Israel and the Judean Desert in the east and the Negev Desert in the south
(Figure 12.1) facilitate this eco-exploration.

12.2 ECOSYSTEM TRANSITIONS ALONG A CLIMATIC GRADIENT

Mediterranean to arid climatic gradients characterize wide regions at the western
sides of South and North America, South Africa, and Australia, and in the southern
and eastern margins of the Mediterranean Basin. Ecosystems along these gradients
vary between woodlands, dense shrublands, open shrublands, dwarf-shrublands
(batha), and arid vegetation (herbaceous species with scattered patches of dwarf-
shrubs). These life-form transitions are governed by rainfall, lithology, and soil,
together with strong anthropogenic influences attributed to fire, grazing, and wood-
cutting. Desertification representing long-term (possibly irreversible) soil erosion
and loss of vegetation productivity was reported in several well-known studies
Remote Sensing Identification of Threshold Zones 231

(e.g., Cerda and Lavee 1999; Geeson et al. 2002; Hill et al. 2008; Lavee et al. 1998)
for the margins of the Mediterranean regions. The interplay between land use change
and climate parameters was hypothesized by Milan et al. (2005) and Puigdefabregas
and Mendizabal (1998) and is thought to be responsible for the desert-like conditions
found in the vast Mediterranean regions of eastern Tunisia, Libya, Egypt, and in
Almeria in Spain. Spatiotemporal changes in shrubs, dwarf-shrubs, and grass cover
characterize this land degradation (e.g., Maestre et al. 2009; Papanastasis et al. 2003;
Peinado et al. 1995). Identification of threshold zones between these vegetation cover
typologies is instrumental for studying desertification in semiarid Mediterranean
landscapes. According to Fagre and Charles (2009), thresholds can be identified
“where small changes in an environmental driver produce large, persistent responses
in an ecosystem.” In the Mediterranean to arid transition context, this refers to dis-
continuity among vegetation and soil conditions along gradual climate change as
a result of ecosystems’ hysteresis (Meron et al. 2004), or from the loss of recovery
potential.
The climatic gradients between the Judean Mountains in central Israel and their
eastern and southern desert margins represent a wide range of differences in natural
settings (e.g., rainfall and lithology) and anthropogenic influences. Implementation
of remote sensing on west to east versus north to south gradients is expected to allow
one to explore the formation of sharp transition zones between ecosystems in general
and of desert thresholds in particular.

12.3 REMOTE SENSING TECHNIQUES

AND THEIR IMPLEMENTATION
In this chapter, we will integrate three techniques, developed and implemented in
earlier studies for assessing modes of vegetation change along the climatic gradi-
ents of the Judean Mountains. The first method is based on a multitemporal tech-
nique developed by Shoshany et al. (1994, 1995, 1996), the second method presents
an algorithm for life-form decomposition as implemented by Shoshany and Svoray
(2002), and the third method concerning estimation of spatial potentials for erosion
versus recovery adopts the work reported by Shoshany (2012).

12.3.1 Green VeGetation CoVer

Vegetation cover is a primary indicator for the location of desert threshold zones.
A study of vegetation cover changes between years of extreme rainfall levels was
found to be most informative regarding the type of response of different ecosys-
tems along a climatic gradient to rainfall changes. This was conducted for the years
1991 and 1992, which had a unique combination of rainfall regimes in the Eastern
Mediterranean; specifically, 1991 was the 60-year record for low precipitation and
1992 was the year with the highest rainfall during this time frame (Kutiel et al.
1995; Shoshany et al. 1996). Here, we conducted a similar study while extending
the investigation from 2 to 16 years, between 1996 and 2011. For this purpose, we
used the same empirical multiple regression vegetation reflectance (VR) model,
232 Remote Sensing for Sustainability

which was developed by Shoshany et al. (1994) specifically for estimating GVC in
our study area:

GVC (%) = 1.48 + 2.5R[1] − 25.5R[2] − 266.97R[3] + 275.5R[4]. (12.1)

For constructing this model, the reflectance data for small plot areas at Landsat
TM channels (R[1]…R[4]) were acquired using the NASA Radiometer, while veg-
etation cover was calculated by classifying close-range photographs of these plot
areas into green vegetation and non-vegetation areas. The VR model was tested for
26 plot areas at four sites along the climatic gradient between the Judean Mountains
and the Judean Desert; it was found to be highly significant with an R2 value of 0.88
and a P value of 0.001.
This model was reapplied here to five radiometrically corrected and geometri-
cally co-registered images of the end of the summer, for 3 years of extreme high and
low rainfall (Table 12.1) in between 2 years with close to average rainfall (1996 and
2011).
Two transects were selected along the climatic gradients representing the less-
disturbed landscape sequences from the Judean Mountains to the Judean Desert in
the east and the Negev Desert in the south. The high spatial and temporal variability
of GVC values represents both the high GVC characteristic of protected shrublands
and forest plantations versus the low green values in areas undergoing high grazing
pressures and frequent fires.
A step-like form of change can be used for generalizing GVC changes in both
directions (Table 12.1 and Figure 12.2):

• North to south transect (Figure 12.2b): Two steps of a sharp GVC decrease
characterize the transition from dense shrublands to the desert fringe batha
(dwarf-shrubs). The first step occurs where dense shrublands, which recov-
ered (Shoshany 2002) in areas with rainfall above 400 mm/year (section
I with GVC values between 22 and 37%), are rapidly transformed into
composites of highly disturbed shrublands dissected by stripes of fields
and other land uses. Further south, the landscape is characterized by open
shrublands of moderate density (section II with GVC values between 17%
and 23%) in areas of approximately 350 mm/year average rainfall. The
transition of this area into desert fringe Batha (section III with GVC values
between 8% and 12%) of Sarcopoterium spinosum spread over large extents
occurs within less than 2 km. Such a sharp transition is the result of historic
high grazing pressures followed by frequent fires ignited by army training
in this area. The two sharp transition zones of this transect are thus primar-
ily anthropogenic, where heavy soil erosion in the northern parts of section
III would severely slow down any future recovery trends. The transition
between the desert fringe batha and the arid unit (section IV with GVC
between 6% and 13%) does not form a sharp threshold, as can be inferred
from their overlap in GVC values.
• West to east transect (Figure 12.2a): The GVC change along this transect can
be well described by both a step-like form and a convex transition from the
Remote Sensing Identification of Threshold Zones 233

TABLE 12.1
GVC and Life-Form Composition in Sections along the Climatic Gradient
between the Judean Mountains and the Judean Desert in the East
and the Negev Desert in the South (Figure 12.1)
Year 1996 1999 2003 2008 2011 Life-Form Composition
Percentage of the 102% 66% 143% 70% 94%
MAP
West–east section I 21.08 17.46 25.07 23.74 22.72
(5.38) (4.12) (5.63) (4.35) (5.18)
West–east section II 15.06 11.05 16.71 16.04 14.15
(3.65) (3.12) (5.51) (5.18) (4.70)
West–east 5.29 5.37 10.19 6.04 6.04
section III (2.55) (2.67) (2.71) (2.06) (2.62)
Shrubs Dwarf- Soil and
shrubs rocks
North–south 34.41 22.61 36.70 30.50 25.80 0.73 0.04 0.23
section I (2.06) (2.49) (2.75) (2.67) (2.79)
North–south 21.78 17.39 24.94 22.76 17.50 0.32 0.23 0.45
section II (3.09) (1.86) (2.89) (2.72) (2.46)
North–south 10.95 8.94 11.95 11.47 8.45 0.065 0.585 0.35
section III (3.10) (1.85) (2.83) (2.68) (2.40)
North–south 7.67 6.16 12.56 8.64 6.71 0.05 0.31 0.64
section IV (2.08) (2.50) (2.90) (2.64) (2.78)

Source: Summary of rainy season for 2011/2012 and the main hydrological characteristics. Reported by
the Israel Water Authority.
Note: Rainfall data represent the percentage of the mean annual precipitation (MAP).

crests of the Judean Mountains in the west to the desert margins in the east.
Two sharp transition zones can be detected. The western one has GVC values
between 17% and 25% of the open shrubland declining toward the highly
disturbed zone with GVC levels between 11% and 17%, representing built-up,
rural, and agricultural land uses with scattered islands of natural vegetation
(Sharakas et al. 2003; Shoshany et al. 1995, 1996). The eastern one represents
a drop-down within a distance of 5 km from an average GVC of 15% as
obtained for the seminatural desert fringe batha mixed with tree plantations
to an average GVC of 6% characterizing the arid ecosystem. The sharp tran-
sition between the desert fringe batha and the arid ecosystem corresponds to
the boundary between the desert and brown lithosols and the Rendzina soil
(Dan and Raz 1970). The rate of GVC that increases west from this boundary
is jointly governed by rainfall and both positive and negative human impacts:
tree planting, on the one hand, and grazing and fires, on the other.

Generally, profiles from all years fit well and describe similar spatial forms of
change beyond significant temporal differences owing to changes in precipitation
 234 Remote Sensing for Sustainability

35

30 1996
1999
25 2003
Green vegetation cover

2008
20 Section I 2011

15

Section II
10

5
Desert threshold at a Section III
boundary between soil units
0
690,000 695,000 700,000 705,000 710,000 715,000 720,000 725,000 730,000 735,000
West to east coordinates
(a) –5

50
1996 Beit Guvrin
45 Avisur
1999 Dvira Forest Highland
Amazia
40 open shrublands Dense
2003 Shrublands
35
Green vegetation cover

2008 Desert fringe

30 batha
2011 Section I
25
Arid ecosystem
20
Section II
15

10
Section III
Section IV
5

0
3,460,000 3,465,000 3,470,000 3,475,000 3,480,000 3,485,000 3,490,000 3,495,000 3,500,000 3,505,000 3,510,000
(b) South to north coordinates

FIGURE 12.2 GVC along transects. (a) GVC change along the west to east transect.
(b) GVC change along the north to south transect.

between years. This result is most meaningful, considering the fact that the rainfall
fluctuated between 66% and 143% of the mean annual precipitation (MAP) between
1996 and 2011. Within the framework of temporal changes in threshold zones, it is
most important to examine the GVC variations in locations of highest transitions in
both transects.
Remote Sensing Identification of Threshold Zones 235

The boundary between the desert fringe batha section and the arid section in the
west to east transect, which is attributed to soil transition, seems to be maintained
even during years of extremely high rainfall (such as 2003). However, when the pro-
files are examined in detail, there seems to be short extents in the desert fringe batha
section in which their GVC values are only slightly higher (~2%) than those of the
arid section. During the years with extremely low rainfall, the GVC of these desert
fringe batha section drops to its level in the arid section. Permanent desertification
may take place in this section with repeated fires, wood-cutting, and gazing, fol-
lowed by soil erosion.
The boundary between the open shrublands (section II) and the desert fringe
batha (section III) in the north to south transect shows a catastrophic shift owing
to overgrazing and fire. The GVC changed by only 3% in section III between 1999
and 2003, suggesting that it lost some of its recovery potential. Section II shows an
increase in GVC that is twice as high between these years (from 17% to 25%). The
open shrublands thus maintain some level of resilience to dry periods. However,
since the lithology and rainfall are similar on both sides of this partition line, fre-
quent fires may transform it relatively quickly into a degraded ecosystem similar to
Unit III.

12.3.2 Life-form Compositions

The spatial heterogeneity of life-form compositions at the subpixel level results in the
non-uniqueness of spectral and phenological information at moderate and low spatial
resolutions. Although the mapping of trees, shrubs, and herbaceous growth in large
homogenous areas was reported in a number of studies, an estimation of life-form
compositions in general, and of dwarf-shrubs in particular, in Mediterranean and
semiarid environments having high spatial heterogeneity have received very limited
attention (e.g., Hamada et al. 2013; Shoshany and Svoray 2002). An adaptive phe-
nological (seasonal) and zonal unmixing technique was developed by Shoshany and
Svoray (2002) in order to estimate the area fraction of these life-forms for each pixel.
The fundamental principles underlying this technique are as follows (Figure 12.3):

a. Selecting vegetation, soil, and rock endmembers and performing the

unmixing separately for the semiarid and arid regions for each season (end
of winter, beginning of summer, and end of summer).
b. Estimating the fraction of each life-form based on subtracting the corre-
sponding seasonal vegetation fractions. For example, fractional cover of
herbaceous growth is estimated according to the difference between the
end of the winter (end of March) vegetation fraction and that of the begin-
ning of the summer (beginning of June).

A north to south transect (following the same line as in Section 12.3.1) represent-
ing changes in life-form compositions allowed us to assess the types of life-form
transitions along the climatic gradient (Figure 12.4) for 1995, a year having 113% of
the MAP. The spatial structure of life-form compositions along the transects fits well
the structure inferred based on the GVC. However, life-form compositions provide
 236 Remote Sensing for Sustainability

Rλ = Rvegλ × Fveg winter + R soil X fraction of soil

End of the winter: all life-forms are
photosyntetically active

FShrubs = Fveg summerEnd FHerbaceous = Fveg winter – Fveg summerBegin

Adaptive
spectral unmixing;
Selecting different endmembers
for each season for shrubs (semi-arid)
and dwarf-shrubs (arid) dominated environments
End of summer: Fdwarf-Shrubs = Beginning of summer:
only shrubs are Fveg summerBegin – Fveg summerEnd only shrubs and dwarf-shrubs
photosyntetically active are photosyntetically active

Rλ = Rvegλ × Fveg summerEnd + R soil X fraction of soil Rλ = Rvegλ × Fveg summerBegin + R soil X fraction of soil

FIGURE 12.3 Schematic description of the algorithm for life-forms’ unmixing.

1.2
Shrubs Amazia
Dvira Forest open shrublands Beit Guvrin
Dwarf
1 shrubs
Fractional cover of life–forms

Soil and
0.8 rocks

0.6

0.4

0.2

0
3,465,000 3,470,000 3,475,000 3,480,000 3,485,000 3,490,000 3,495,000 3,500,000 3,505,000 3,510,000
–0.2 South to North coordinate

FIGURE 12.4 Life-forms’ composition along the north to south transect.

much more explicit information regarding the nature of the transitions within and
between threshold zones:

• The sharp southern boundary of the Dense Shrublands of the Avisur

Highland (shrub fraction = 0.76) at Beit Guvrin was caused by the dissec-
tion of the natural vegetation by fields, grazing areas, archeological sites,
and army training areas.
• The Open Shrublands’ recovery boundary: approximately 3 km south of
Beit Guvrin; a sharp boundary marks the northern edge of the Amazia
open shrublands (shrub fraction = 0.32; dwarf-shrub fraction = 0.23; soil
and rock fraction = 0.45) where the landscape is recovering under con-
trolled grazing intensities, with fields extending only in the lower parts of
the Wadis.
Remote Sensing Identification of Threshold Zones 237

• The southern boundary of the Open Shrublands of Amazia that emerged

through frequent fires and historically high grazing intensities. The area
south of this boundary consists the desert fringe batha that is dominated by
large extents of dwarf-shrubs and bare soil patches (shrub fraction = 0.065;
dwarf-shrub fraction = 0.585; soil and rock fraction = 0.35).
• Transition from dwarf-shrub–dominated areas to an arid ecosystem (shrub
fraction = 0.05; dwarf-shrub fraction = 0.31; soil and rock fraction = 0.64)
of scattered patches of dwarf-shrubs with large extents of bare ground and
rocks, typical of the Northern Negev (Desert).

In these two latter desert fringe zones, grazing and fires are balanced by the rela-
tively quick colonization (and recolonization) of the S. spinosum, forming a hetero-
geneous patchy pattern of bare soil and dwarf-shrubs. In the arid zone, dwarf-shrub
patches are smaller than those found in the desert fringe batha. The separation line
between these two units corresponds to a rainfall level of approximately 270 mm/
year. In our study area, it passes at the vicinity of the Dvira Forest.

12.3.3 inVerse erosion and reCoVery potentiaLs

Thresholds in the approach presented by Shoshany (2012) are boundaries of geodi-
versity rather than lines separating homogenous units. Geodiversity in this context
concerns the spatial heterogeneity of vegetation and soil patch patterns. Degradation
and recovery cycles in desert fringe ecosystems are characterized by dynamic
changes in the spatial heterogeneity with mutual processes taking place between the
vegetation and soil patterns (Forman 1995; Shoshany and Kelman 2006). During
degradation, vegetation continuum is perforated by soil patches, dissected, frag-
mented, isolated, contracted, and dissolved. In parallel, soil patches grow, aggre-
gate, connect, and expand. Inverse processes are taking place during recovery. As
described in Section 12.3.2, along the north to south transect, the shift between the
desert fringe batha and the arid ecosystem is characterized by a change from 2/3
dwarf-shrubs’ cover and 1/3 bare soil to 2/3 bare soil and 1/3 dwarf-shrubs. Along
the west to east transect, the shift is from approximately 15% to 6% GVC (note that
S. spinosum has low greenness at the end of the summer). Spatial variations between
these cover distributions are of significant impact on the degradation and recovery
of these ecosystems. Large dwarf-shrub patches may shrink and still sustain during
long and hot summers and during drought periods, while maintaining their recovery
potential when habitat conditions improve. Isolated scattered plants under such con-
ditions may dry out and die. Soil protection by vegetation in this latter case is low
and thus increases potential erosion by runoff and wind (e.g., Ludwig et al. 2005;
Okin et al. 2009; Puigdefabregas 2008). At the Landsat TM resolution, spatial het-
erogeneity of plant patterns is expressed by variations in vegetation versus soil/rock
fractional cover. For analyzing potential erosion and potential recovery, Shoshany
(2012) suggested to employ the following mathematical morphology technique of
dilation on F as fractional cover of vegetation or soil:

F ⊕ K = Max {f(xi + q, j + p) + Kq , p | −n < q < n; −n < p < n}, (12.2)

238 Remote Sensing for Sustainability

where Kq,p is the structuring element (kernel) of the dimensions 2n + 1 × 2n + 1

(n = 2 in this study).
Dilation of fractional cover maps would extend areas of maximal cover. The over-
all areal change between the original and dilated maps highly depends on the spatial
distribution of the maximal values. In a scattered pattern, the change will occur
around any core of maximal vegetation cover or inversely of high soil cover, thus
producing wide areal changes. In clumped areas of high cover, the change will take
place only in the margins of these clumps and thus will cover limited area. Clumped
patches of high vegetation fractional cover associated with scattered pattern of low
soil fractional cover would represent areas of potential recovery. Clumped patterns
of high soil fractional cover associated with scattered pattern of low vegetation cover
fraction would represent erosion potential. Within this description, the determina-
tion of threshold zone is expected to be configured according to two phenomenolo-
gies: first, by sharp increase in soil fractional cover reaching almost complete cover
(Threshold Type I), and second, by the shift from higher fractional vegetation cover
to higher fractional soil cover (Threshold Type II).
Figure 12.5 presents the methodological outline. In the first stage, spectral unmix-
ing facilitates estimating the subpixel fractions of vegetation, soil, and rocks. In the
second stage, the cover proportion map of vegetation and soil surface types was
dilated separately using Equation 12.2, indicating the vegetation recovery and soil
erosion potentials. In the third stage, recovery versus erosion potentials are deter-
mined by subtracting their dilated levels.
The Landsat image of October 1987 was used for detecting erosion and recov-
ery potentials as a reference point before the high rainfall fluctuations that took
place in 1991, 1992, 1999, 2003, and 2008. The position of the two threshold zones,
as inferred by the new technique (Figure 12.6a and b), corresponds well with the

Landsat TM
spectral unmixing

Fraction of
Vegetation fraction
soil and bare rocks

Erosion potential: Recovery potential:

dilated soil fraction dilated veg. fraction

Threshold Type I:
where one of the dilated cover
is reaching domination

Threshold Type II:

where
Erosion potential = Recovery potential

FIGURE 12.5 Schematic description of the erosion and recovery potentials’ algorithm.
Remote Sensing Identification of Threshold Zones 239

1
Shrub cover change potential
0.9
Bare rock fraction change potential
0.8
Erosion and recovery potentials

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5000 10,000 15,000 20,000 25,000
(a) Distance from west to east (m)

0.6

0.4
Threshold Type II
Recovery potential–erosion potential

0.2

0
0 5000 10,000 15,000 20,000 25,000
–0.2

–0.4

–0.6

Threshold Type I
–0.8

–1
(b) Distance from west to east (m)

FIGURE 12.6 Potential erosion and potential recovery (a) and their subtraction (b).
240 Remote Sensing for Sustainability

boundaries detected in the west to east profile of GVC (Figure 12.2a). However, the
new technique presents both a more explicit and distinctive expression of the desert
transitions. In the GVC map, the transition from desert to the desert fringe ecosystem
is characterized by a gradual increase from a level of 6% to 10%, whereas the erosion
potential at the same area decreases from 0.85 to 0.15. Within the arid area, the dif-
ference between the recovery and erosion potentials is negative, indicating an excess
of erosion potential. The upper boundary of the transition zone is clearly detected
where the difference approaches zero (Threshold Type II).

12.4 SUMMARY AND CONCLUSIONS

GVC, life-form compositions, erosion, and recovery potentials highly vary along the
climatic gradients between the Judean Mountains and their arid margins. In addition
to high local fluctuations attributed to abrupt changes in the natural and anthropo-
genic habitat conditions, step-like transitions were detected. While most of these
transitions result from differences in the type and intensity of human disturbance,
the boundary between the desert fringe ecosystem and the arid ecosystem of the
Judean Desert (west to east transect) follows a soil boundary between lithosols and
Rendzina. The boundary between the desert fringe and the arid ecosystems in the
north to south gradient is weak, showing only a slight shift in GCV and with some
overlap in values. The study of multitemporal changes showed that GVC levels fol-
low rainfall fluctuations to a certain extent. Only the highly disturbed unit (section
III) in the north to south section showed a minor recovery in GVC values after the
1999 to 2003 rainfall fluctuations.
Changes in life-form compositions fit well with the structure of change inferred
through the study of GVC variations. In addition, it adds a significant insight into
the type of change occurring. Thus, for example, whereas the shift from the arid to
the desert fringe ecosystems in the north to south sections is low in terms of GVC
values representative of the end of the summer, the life-form composition inferred
through seasonal changes in vegetation fraction indicates that there is a switch from
the dominance of bare soil and rock to the dominance of dwarf-shrubs. However,
potential erosion versus potential recovery seems to provide the most sensitive tool
for detecting the formation of boundaries and threshold zones in areas of low and
moderate shrub (including dwarf-shrub) cover. It implies the role of spatial distribu-
tion of surface conditions in determining ecosystems responses to changes in habitat
conditions.
By combining all the information extracted in these studies, it is reasonable to
suggest that no significant shift in the threshold zones occurred between the Judean
Mountains and their arid margins between 1996 and 2011. However, information
from the GVC indicates a high sensitivity and the possibility of a shift in the bound-
ary between the desert fringe Batha and the arid zone in both transects. Recent evi-
dence regarding the accumulation of water deficits and the drying of planted forests
(Dorman et al. 2012) and of shrublands (Shoshany and Karnibad 2015) in this region
may enhance the effect of human disturbance, reduce recovery potential, and result
in a significant desertification of the Mediterranean to arid transition zones.
Remote Sensing Identification of Threshold Zones 241

REFERENCES
Cerda, A., and Lavee, H. (1999) The effect of grazing on soil and water losses under arid and
Mediterranean climates. Implications for desertification. Pirineos, 153–154, 159–174.
Dan, Y., and Raz, Z. (1970) The soil formations map of Israel. Beit Dagan, Israel: The Volcani
Centre.
Dorman, M., Svoray, T., and Prevolozki, A. (2012) Drying trees in the Jerusalem Pine forests
of Israel—From a high altitude point of view. Ecol. Environ., 3, 230–237. (In Hebrew.)
Fagre, D.B., and Charles, C.W. (2009) Thresholds of climate change in ecosystems: Final
Report, Synthesis and Assessment Product 4.2 U.S. Climate Change Science Program
and the Subcommittee on Global Change Research.
Forman, R.T.T. (1995) Land Mosaics: The Ecology of Landscapes and Regions. Cambridge
University Press. 632 pp.
Geeson, N., Brandt, J., and Thornes, J.B. (2002) Mediterranean Desertification: A Mosaic of
Processes and Responses. John Wiley & Sons.
Hamada, Y., Stow, D.A., Roberts, D.A., Franklin, J., and Kyriakidis, P.C. (2013) Assessing
and monitoring semi-arid shrublands using object-based image analysis and multiple
endmember spectral mixture analysis. Environ. Monit. Assess., 185, 3173–3190.
Hein, L., and De Ridder, N. (2006) Desertification in the Sahel: A reinterpretation. Glob.
Chang. Biol., 12, 751–758.
Herrmann, S.M., Anyamba, A., and Tucker, C.J. (2005) Recent trends in vegetation dynam-
ics in the African Sahel and their relationship to climate. Glob. Environ. Chang., 15,
394–404.
Hill, J., Stellmes, M., Udelhoven, Th., Röder, A., and Sommer, S. (2008) Mediterranean
desertification and land degradation Mapping related land use change syndromes based
on satellite observations. Glob. Planet. Change, 64, 146.
Kutiel, P., Lavee, H., and Shoshany, M. (1995) The influence of a climatic gradient upon
vegetation dynamics along a Mediterranean arid transect. J. Biogeogr., 22, 1065–1071.
Lamprey, H.F. (1975) Report on the desert encroachment reconnaissance in northern Sudan.
UNESCO/UNEP. Desertification Control Bull., 17, 1–7.
Lavee, H., Imeson, A.C., and Sarah, P. (1998) The impact of climate change on geomorphology
and desertification along a Mediterranean–arid transect. Land Degrad. Dev., 9, 407–422.
Ludwig, J.A., Wilcox, B.P., Breshears, D.D., Tongway, D.J., and Imeson, A.C. (2005)
Vegetation patches and runoff-erosion as interacting ecohydrological processes in semi-
arid landscapes. Ecology, 86(2), 288–297.
Maestre, F.T., Bowker, M.A., Puche, M.D., Belen Hinnojosa, M., Martinez, I., Garcia-Palacios,
P., Castillo, A.P., Soliveres, S., Luzuriaga, A.L., Sanchez, A.M., Carreira, J.A., Gullardo,
A., and Escudero, A. (2009) Shrub encroachment can reverse desertification in semi-arid
Mediterranean grasslands. Ecol. Lett., 12, 930–941.
Meron, E., Gilad, E., von Hardenberg, J., Shachak, M., and Zarmi, Y. (2004) Vegetation pat-
terns along a rainfall gradient. Chaos, Solitons Fractals, 19, 367–376.
Milan, M.M., Estrela, M.J., Sanz, M.J., Mantilla, E., Martin, M., Pastor, F., Salvador, R.,
Valledo, R., Alonso, L., Gangoitti, G., Ilardia, J.L., Navazo, M., Albizuri, A., Artinano,
B., Ciccioli, P., Kallos, G., Carvalho, R.A., Andres, D., Hoff, A., Werhahn, J., Seufert,
G., and Versino, B. (2005) Climatic feedbacks and desertification: The Mediterranean
model. J. Clim., 18, 684–701.
Nicholson, S.E. (2011) Dryland Climatology, Desertification Chapter, Cambridge University
Press, 431–447.
Okin, G.S., Parsons, A.J., Wainwright, J., Herrick, J.E., Bestelmeyer, B.T., Peters, D.C., and
Fredrickson, E.L. (2009) Do changes in connectivity explain desertification? Bioscience,
59, 237–244.
242 Remote Sensing for Sustainability

Papanastasis, V.P., Kyriakakis, S., Kazakis, G., Abid, M., and Doulis, A. (2003) Plant cover
as a tool for monitoring desertification in mountain Mediterranean rangelands. Manage.
Environ. Qual., 14(1), 69–81.
Peinado, M., Alcaraz, F., Aguirre, J.L., Delgadillo, J., and Aguado, I. (1995) Shrubland forma-
tions and association in Mediterranean–desert transitional zones of northwestern Baja
California. Vegetatio, 117, 165–179.
Puigdefabregas, J. (2008) The role of vegetation patterns in structuring runoff and sediment
fluxes in drylands. Earth Surf. Processes Landforms. 30, 133–147.
Puigdefabregas, J., and Mendizabal, T. (1998) Perspectives on desertification: Western
Mediterranean. J. Arid Environ., 39, 209–224.
Safriel U. (2009) Status of desertification in the Mediterranean region. In: Water Security,
Land Degradation and Desertification in the Mediterranean Region. Edited by Rubio,
J.L., Rubio, J.L., Daussa, R., Blum, W., and Pedrazzini, F. Springer: Berlin, Germany;
pp. 33–43.
Sharakas, O., Hammad, A.A., Nubani, A., and Abdullah, A. (2003) Land Degradation Risk
Assessment in the Palestinian Central Mountains Utilizing Remote Sensing and GIS
Technique; Geography Department, Birzeit University: Birzeit, Palestine, 97 pp.
Shoshany, M. (2002) Landscape fragmentation and soil erodibility in south and north fac-
ing slopes during ecosystems recovery: An analysis from multi-date air photographs.
Geomorphology, 45, 3–20.
Shoshany, M. (2012) Desert threshold identification by maximizing potential instability in soil
and shrubs’ patterns: A remote sensing study. Land Degrad. Dev., 23, 331–338.
Shoshany, M., and Karnibad, L. (2015) Relative water use efficiency and shrublands biomass
along a semi-arid climatic gradient: A remote sensing study. Remote Sens. (online), 7,
2283–2301.
Shoshany, M., and Kelman, E. (2006) Mutuality in soil and vegetation pattern characteristics:
Modeling with cellular automata. Geomorphology, 77(1–2), 35–46.
Shoshany, M., Kutiel, P., and Lavee, H. (1995) Seasonal vegetation cover changes as indica-
tors of soil types along a climatological gradient: A mutual study of environmental pat-
terns and controls using remote sensing. Int. J. Remote Sens., 16, 2137–2151.
Shoshany, M., Kutiel, P., and Lavee, H. (1996) Monitoring temporal vegetation cover changes
in Mediterranean and arid ecosystem using a remote sensing technique: Case study of
the Judean Mountain and the Judean Desert. J. Arid Environ., 32, 1–13.
Shoshany, M., Kutiel, P., Lavee, H., and Eichler, M. (1994) Remote sensing of vegetation
cover along a climatological gradient. ISPRS J. Photogramm. Remote Sens. 49, 1–8.
Shoshany, M., and Svoray, T. (2002) Multi-date adaptive spectral unmixing and its applica-
tion for the analysis of ecosystems’ transition along a climatic gradient. Remote Sens.
Environ., 81, 1–16.
Thomas, D.S.G., and Middleton, N.J. (1994) Desertification: Exploding the Myth. John Wiley
& Sons Ltd., Chichester, UK. xiv + 194 pages. ISBN 0-471-94815-2.
Thornes, J.B. (2000) Mediterranean desertification: The issues. In: Mediterranean
Desertification: Research Results and Policy Implications. Edited by Balabanis, P.,
Peter, D., Ghazi, A., and Tsogas, M. European Commission, EUR 19303, Brussels;
pp. 9–17.
Veron, S.R., Paruelo, J.M., and Oesterheld, M. (2006) Assessing desertification. J. Arid
Environ., 66, 751–763.
 13 Soil Moisture Using
Optical Remote
Sensing and Ground
Measurements
A Case Study from Pakistan
Mudassar Umar, Siraj Munir, Iftikhar Ali,
Salman Qureshi, Claudia Notarnicola,
Said Rahman, and Qihao Weng

CONTENTS
13.1 Introduction .................................................................................................. 243
13.2 Data and Methodology .................................................................................246
13.2.1 Study Area ........................................................................................246
13.2.2 Remote Sensing Data ........................................................................246
13.2.3 Gravimetric Measurements ..............................................................248
13.2.4 The NIR–Red Spectral Reflectance Space ....................................... 249
13.2.5 Development of a RSSM Model ....................................................... 249
13.2.6 Development of a TVDI ................................................................... 252
13.3 Results and Discussion ................................................................................. 255
13.3.1 Ts–NDVI Space (Combination of LST and NDVI) .......................... 255
13.3.2 Spatial Variation of TVDI ................................................................ 257
13.3.3 Comparison of TVDI with Gravimetric Measurements................... 259
13.4 Conclusion .................................................................................................... 262
Acknowledgments.................................................................................................. 262
References .............................................................................................................. 263

13.1 INTRODUCTION
Soil moisture is an important component of the terrestrial environment that sig-
nificantly regulates water circulation and surface energy exchanges between land
surface and the atmosphere (Jackson 1993; Vereecken et al. 2014). It is an important
factor affecting the budget of the hydrological cycle, specifically by separating rain-
fall into runoff, surface infiltration, and evapotranspiration (Jawson and Niemann
2007). Many scientific applications require soil moisture data to represent the initial

243
244 Remote Sensing for Sustainability

state of soil moisture such as numerical weather prediction and climate projections
because they play a key role in hydro-meteorological processes (Gao et al. 2014). A
precise estimation of soil moisture can help in improving the forecasting of precipi-
tation, temperature, droughts, and floods (Albergel et al. 2013). There are generally
four methods to retrieve soil moisture: (i) ground-based soil sampling or sensors,
(ii) land surface model using meteorological data, (iii) remote sensing from airborne
or satellite data, and (iv) data assimilation techniques that integrate remote sensing
signals into land surface models (Qin et al. 2013).
The spatial variability of soil moisture is high owing to the significant het-
erogeneity of soil, topography, vegetation, and precipitation (Crow et al. 2012),
making field campaigns challenging and time-intensive. Furthermore, the gravi-
metric measurements (i.e., ground measurements) are simply representative over
a small spatial scale. In data assimilation and land surface models, the model
grid-box size is greater than the gravimetric measurements (Qin et al. 2013).
Remote sensing provides effective methods for collecting information from a
range of samples over a large area in relatively shorter and repeated intervals of
time (Nichols et al. 2011).
Microwave remote sensing offers a great potential to measure soil moisture across
varying spatial and temporal scales (Singh et al. 2005). The poor spatial resolution
of passive microwave and the strong sensitivity of active microwave over vegetation
cover and in surface roughness limit the capability of microwave remote sensing
to infer soil moisture (Srivastava et al. 2003). The weakness of microwave sens-
ing soil moisture is the impact of ground vegetation cover. Only longer-wavelength
microwave could penetrate thick vegetation cover and can appropriately sense soil
moisture. To overcome this problem, a number of approaches using visible, near-
infrared (NIR), and thermal infrared wavelengths were evaluated for soil moisture
assessment.
Remote sensing data have been effectively used for the estimation of soil mois-
ture near surface, but this information is limited to a few centimeters below the
surface (Santos et al. 2014). Remote sensing does not allow access to the whole
root zone where water can be absorbed by the roots (Liu et al. 2012). Soil mois-
ture has varying spectral patterns across different wavelengths as remote sensing
data from visible to microwave lengths can effectively be used for monitoring the
variations in soil moisture (Wang and Qu 2009). Vegetation indices can be used to
relate soil moisture in the root zone and soil moisture influences vegetation growth
and thus changes the spectral characteristics of vegetation (Wang et al. 2010). The
spectral reflectance decreases with increasing soil moisture in the visible and NIR
range. Some studies have shown soil moisture modeling with visible and NIR data
such as the vegetation anomaly index combining NIR–red data with land surface
temperature (LST) for studying land surface parameters (vegetation anomaly, soil
moisture), and vegetation indices serve as indicators of drought as well (Huete et
al. 1992). The soil moisture reflectance method is dependent on the relationship
between soil moisture and vegetation (Zhan et al. 2007). Zhan et al. (2007) vali-
dated the NIR–red observed soil moisture with gravimetric measurements, and the
effective soil depth for moisture estimation was 5 cm and was found to be signifi-
cant. Nevertheless, the major limitations in case of optical remote sensing of soil
Soil Moisture Using Optical Remote Sensing and Ground Measurements 245

moisture relate to surface roughness, soil structure, and the presence of organic
matters (Wang and Qu 2009).
The use of remote sensing for the estimation of surface energy fluxes and sur-
face soil water dates back to 1970. For instance, it was used by meteorologists to
estimate the surface energy fluxes and soil water (Carlson et al. 1984; Price 1980,
1982). Several other studies have approached them with similar methods (Kustas
and Norman 1996; Sandholt et al. 2002). Information about surface energy and
water status has been determined by many researchers by developing the relation-
ship between remotely sensed LST and normalized difference vegetation index
(NDVI) (Amiri et al. 2009; Carlson et al. 1994; Rajasekar and Weng 2009). The
correlation between thermal and visible/NIR wavelengths has proven to be useful
for appropriate monitoring of vegetation and water stress. A method was devel-
oped for mapping the surface moisture and land surface energy fluxes in the 1990s
(Price 1990), which was referred to as the triangle method. This method lies on the
interpretation of the pixel distribution in Ts–NDVI space. If an image contains a
wide range of soil water content and vegetation cover, the space presents a triangle.
This triangle is formed because surface temperature decreases as vegetation cover
increases. The concept of the triangle was first given by Price (1990) and later fur-
ther elaborated by several researchers (Gillies et al. 1997) while Sandholt et al.
(2002) and others adopted and applied the triangle method (Chauhan et al. 2003;
Stisen et al. 2008; Wang et al. 2006). Wang et al. (2007) stated that MODIS land
parameters (LST, NDVI) are significantly correlated with ground soil moisture.
Chauhan et al. (2003) applied this approach to obtain nearly accurate, high soil
moisture by linking microwave-derived soil moisture with optical parameters.
Li and Dong (1996) stated that a relationship can be developed using satellite-
derived NDVI, surface temperature, and soil moisture to estimate moisture con-
tents in deep soil. Guo et al. (1997) estimated soil moisture at 20 cm depth using
vegetation and surface temperature information and reported that satellite data
have a significant relationship to the mentioned depth. Li et al. (2008) analyzed
the spatiotemporal variability of land surface moisture based on vegetation and
temperature using the triangle method. Wang et al. (2010) examined a relation-
ship between TVDI and soil moisture to estimate soil moisture from 10- to 20-cm
depths. They stated that satellite data have a significant relationship only to the
upper layers (10 and 20 cm). The relationship was not significant for deep soil
layers (20–50 cm). The developed estimation model provides reliable estimates of
soil moisture at depths of 20 cm but cannot estimate moisture levels in deeper soil
layers (Wang et al. 2010). It was suggested that the combination of LST and NDVI
could provide better estimates of vegetation and soil moisture on the land surface.
Ts–NDVI space is related to surface evapotranspiration rate and has been used for
the assessment of temperature and soil moisture condition (Goetz 1997; Prihodko
and Goward 1997).
Sandholt et al. (2002) proposed a moisture index called temperature vegeta-
tion dryness index (TVDI) based on the simplified interpretation of Ts–NDVI
space. Xin et al. (2006) evaluated the potential of TVDI by using Pathfinder data
(8 km) to assess the soil moisture status by using routine level measurements of
soil moisture at the station level. The potential of the moderate resolution imaging
246 Remote Sensing for Sustainability

spectroradiometer (MODIS) is investigated to capture spatial variability in soil

moisture with intensive measurements (Patel et al. 2009). Less focus has been
given in investigating the potential of TVDI from Landsat TM data for soil mois-
ture assessment.
The main aim of this study was to evaluate soil moisture through remotely sensed
data and its comparison using field measurements in the Umer Kot, Sindh province
of Pakistan. To achieve this goal, our objectives were to develop a remote sens-
ing soil moisture model (RSSMM) using NIR–red spectral reflectance space and to
develop TVDI by combining LST and NDVI for estimating soil moisture. However,
despite numerous work on soil moisture estimation through TVDI, a comparison
between RSSMM and TVDI has not been carried out.

13.2 DATA AND METHODOLOGY

13.2.1 Study AreA
The Umer Kot district located in the Sindh province of Pakistan was selected as
the study area. This region is geographically situated between 24°10′ to 25°45′ north
latitude and 69°04′ to 71°06′ east longitude. The total area of the district is 5608 km2.
The study area comprises two distinct land covers, the irrigated portion in the north-
west and barony/desert in the north-southeast. The agricultural land of the study
area is very fertile. The main crops grown in the irrigated belt of the study area are
wheat and chili. The other crops such as sugarcane, mustard, sunflower, ispaghol,
and saunf are also successfully grown in the area. Bajra, guwar, and till are grown in
the rainfed belt (desert) of the area. The general classification of the land cover of this
study area is shown in Figure 13.1. The study area bears distinct climatic conditions;
the irrigated land within the study area has temperate conditions, that is, neither
extremely hot nor very cold in winter as compared to the eastern portion of the study
area. The summer heat is reduced by constant blowing of the southwest breeze from
the sea. The climatic pattern of the eastern (desert) portion is tropical, hot, and dry.
In this area, air temperature can rise above 40°C during May to August and drops
down to 6°C in December and January. The average annual rainfall is approximately
222 mm, which mostly occurs during the monsoon.
The Sindh province was severely affected by drought in past years (FAO 2001),
and the study area is part of the drought-affected region. It is a well-established fact
that soil moisture is an essential parameter for early drought prediction. The main
reason behind the selection of this region was that it is a mixture of semiarid and
desert land. The total population of the district is 2,065,590 in 2010 and is heav-
ily dependent on agricultural production. Hence, the assessment of moisture condi-
tions using advanced remote sensing techniques will help in the effective handling
of drought and other hydrological conditions.

13.2.2 remote SenSing dAtA

Optical and thermal infrared band data of the Landsat 5 Thematic Mapper (TM)
were used in this study. The scene was acquired on May 3, 2011. The data acquisition
 69°15´0˝E 69°30´0˝E 69°45´0˝E 70°0´0˝E 70°15´0˝E

Pakistan
Study area Umarkot

0 255 510 1020

km

Sindh

Legend
Roads Umarkot
Vegetation Water
0 5 10 20 30 40
km Desert Settlements
0 4080 160 240 320 Bare soil
km
Soil Moisture Using Optical Remote Sensing and Ground Measurements

69°15´0˝E 69°30´0˝E 69°45´0˝E 70°0´0˝E 70°15´0˝E

FIGURE 13.1 Study area and its land cover.

247
248 Remote Sensing for Sustainability

date has clear atmospheric condition and images were acquired through the USGS
Earth Resource Observation Systems Data Center, which has corrected radiometric
and geometrical distortions of the images to a quality level of 1G before delivery.
The image covering the Umer Kot district was a subset from the geometrically cor-
rected image (Path 152, Row 41). The image was atmospherically corrected by using
the approach of Markham and Barker (1986). They gave Equation 13.1 without con-
sidering the impact of topography (Ekstrand 1996) and the atmosphere (Liang et al.
1997). Hence, combined surface and atmospheric reflectance of earth is given by

πLd 2
ρ= , (13.1)
ESUNcos(SZ)

where
ρ = unitless planetary reflectance at the satellite
π = 3.1415
L = spectral radiance at sensor aperture in mW cm−2 ster−1 μm−1
d2 = earth–sun distance in astronomical units
ESUN = mean solar exoatmospheric irradiance in mW cm−2 μm−1
SZ = sun zenith angle when scene is recorded

13.2.3 grAvimetric meASurementS

The field measurements were carried out by using the gravimetric method (Evett et
al. 2008). The soil moisture content can be determined by weight as the ratio of the
mass of water present to the dry weight of the soil sample. The water mass is deter-
mined by drying the soil to constant weight and measuring the soil sample mass after
and before drying. The gravimetric measurements of soil moisture were carried out
at 24 selected sites in the study area. The representative random sampling was used
to collect all the soil samples. Soil moisture was measured at three different depths:
0–15, 15–30, and 30–45 cm. All measurements were carried out between 8:00 and
16:00 h local mean time (GMT+05), while the equatorial crossing time of Landsat
was between 9:30 and 10:30 h (local). A local apparatus soil auger (Figure 13.2a

(a) (b) (c)

FIGURE 13.2 (a and b) Field sample collection using soil auger. (c) Samples in the labora-
tory oven.
Soil Moisture Using Optical Remote Sensing and Ground Measurements 249

and b) was used for the measurement. The collected samples were analyzed in the
laboratory (Figure 13.2c); the weight of each sample was 50 g. The samples were
heated in an oven for 18 h at 105°C. The dry weight of each sample was noted and
moisture was calculated by the following formula:

Wt.Soil (g) − Dry soil (g)

Moisture (%) = * 1000. (13.2)
Dry Soil (g)

13.2.4 the nir–red SpectrAl reflectAnce SpAce

Vegetation strongly absorbs blue and red light and reflects green in the NIR spec-
trum. The reflectance of bare soil is high in the red to NIR spectral region. Densely
vegetated area shows smaller reflectance in red and higher reflectance in NIR. This
concept lays the theory of NDVI, which helps assess the changes in vegetation frac-
tion, leaf area index, and chlorophyll content. Richardson and Weigand (1977) devel-
oped a NIR–red spectral space with MSS and a perpendicular vegetation index for
distribution of vegetation in space. The NIR–red spectral reflectance space was con-
structed by using atmospherically corrected TM band 3 and band 4. The NIR–red
spectral scatterplot constructs a triangular shape (Figure 13.3). The NIR–red space
developed by TM band 3 and band 4 validates the distribution of vegetation in space
(Richardson and Weigand 1977).
In Figure 13.3, the change in surface vegetation cover is represented by line AD
from the densely vegetated area (A) and sparse vegetation (E) to bare soil (D). Line
BC refers to soil moisture status, extending from the wet area (B) to semiarid to
dry soil (C). Line BC, which shows the direction of dryness, helps in inferring the
drought severity. The scatterplot of the NIR–red reflectance space demonstrates the
triangular shape. Surface coverage and surface condition can be described signifi-
cantly in this space. Relationships exist among surface spectrum, land cover types,
and dryness condition. This prompts us to build a NIR–red spectral space model for
soil moisture monitoring.

13.2.5 development of A rSSm model

Figure 13.3 shows that dryness gradually rises from B to C and climaxes at
C. Line BC is the soil line and is mathematically expressed by the following
equation:

RNIR = MRred + I, (13.3)

where RNIR is the reflectance of the NIR band, Rred is the reflectance of the red band,
M is the slope of the soil line, and I is the intercept.
According to the perpendicular vegetation index (PVI) developed by Richardson
and Weigand (1977), any point in the NIR–red reflectance space to the line vertical
to the soil line represents surface soil moisture. Line L dissects the coordinate origin
250 Remote Sensing for Sustainability

0.5
Densely vegetated
A
0.4
Dry soil

C
NIR reflectance

E
0.3 Sparsely vegetated
Soil line

Bare surface
D
0.2

Wet soil
B
0.10
0.10 0.2 0.3 0.4 0.5
Red reflectance

FIGURE 13.3 Conceptual diagram of NIR–red reflectance space.

and is vertical to the soil line (Figure 13.4). Hence, the normal function of line L can
be formulated from the soil line expression

 1 
RNIR =  − R . (13.4)
 M red 

The distance from any points in the NIR–red reflectance space to line L represents
surface soil moisture conditions. The farther the distance, the stronger the drought
and the less soil moisture and vice versa. Let there be a random point K (Rred, RNIR);
then, the vertical distance from K (Rred, RNIR) to line L can calculated as follows:

1
KM = ( RNIR + MRred ). (13.5)
M2 +1

The objects with some reflectance near the wet area of Figure 13.3 have a moisture
content close to 1, while the objects with high reflectance are in the direction of C
showing dryness and have a moisture content close to 0. The following model has
been established by subtracting the normalized values of Equation 13.5 from 1:

1
RSSMM = 1 − ( RNIR + MRred ). (13.6)
M2 +1

The soil line extracted from NIR and red band reflectance of the observed area in
Figure 13.5 is expressed as follows:

RNIR = 0.8745Rred + 0.0745. (13.7)

 Soil Moisture Using Optical Remote Sensing and Ground Measurements 251

0.5

A
0.4 C
NIR reflectance Dry soil

0.3 K
ine
il l

PV
So

I
0.2 D

M
Wet soil
B
0.10
O 0.2 0.3 0.4
0.10
Red reflectance
L

FIGURE 13.4 Sketch of soil moisture modeling.

0.40
Red reflectance vs. NIR reflectance
R2 = 0.8808, y = 0.8745x + 0.0745
0.35

0.30
NIR reflectance

0.25

0.20

0.15

0.10

0.05

0.00
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Red reflectance

FIGURE 13.5 Soil line extraction.

It can be seen from Equation 13.7 that the slope (M) = 0.8745 and the intercept
(I) = 0.0745; introducing into Equation 13.5, soil moisture can be obtained over
the study area:

1
RSSMM = 1 − ( RNIR + 0.8745 Rred ). (13.8)
0.87452 + 1
252 Remote Sensing for Sustainability

13.2.6 development of A tvdi

Soil moisture is dependent on LST and vegetation. Chauhan et al. (2003) showed that
there can be a strong correlation between soil moisture, NDVI, and LST for a region.
NDVI is calculated with adjusted apparent reflectance of NIR and red:

ρ( band 4) − ρ( band 3)
NDVI = . (13.9)
ρ( band 4) + ρ( band 3)

Emissivity of natural land surfaces can be measured by using the method pro-
posed by Sobrino et al. (2004). On the basis of brightness temperature, NDVI, and
emissivity, LST is calculated as follows:

ρT
LST = , (13.10)
ρ + λ*T*lnε

where
λ = wavelength of emitted radiance for which the average of limiting wave-
lengths (λ = 11.5 μm) is used (Markham et al. 1985)
h*c
ρ= , where ρ = Boltzmann constant (1.38* 10 −23 J/K), h = Planck’s constant
σ
(6.626 × 10 −34 Js), and c = velocity of light (3 × 108 m/s)
T = effective at-satellite temperature in Kelvin (K)
ε = emissivity of land surface (obtained from thermal band of Landsat TM)

The relationship between LST and NDVI was negative. This proves that as
vegetation increases, the surface temperature tends to decrease, and vice versa.
This result allows the development of a triangle method. The method of mapping
soil moisture using LST and NDVI is known as the triangle method (Carlson 2007;
Price 1990).
In the LST–NDVI triangle (Figure 13.6), the highest LST (Tsmax) along the dry
edge establishes dry surface when the soil wetness is approaching 0. The wet soil
conditions are represented through the minimum LST (Tsmin) along the wet border
when the soil wetness is high and approaching 1. A TVDI can be established by hav-
ing TVDI = 1 at the dry edge and TVDI = 0 at the wet edge. The dry edge has limited
water availability, while the wet edge has unlimited water access.
The mathematical relation is as follows:

Ts − Tsmin
TVDI = , (13.11)
Tsmax − Tsmin

where Ts is the LST, Tsmin is the minimum LST in the triangle and represents the wet
edge, and Tsmax is the maximum LST observed for a given NDVI and signifies the
dry edge in the triangle.
Soil Moisture Using Optical Remote Sensing and Ground Measurements 253

Dry bare soil

Dry
point
Tmax

Dry border
Vegetation canopy
Land surface
temperature

Water stress
Dry

Wet
Wet
point

Tmin
Wet border
Vegetation canopy
Saturated bare soil Adequate moisture

0.0 Normalized difference vegetation index 1.0

FIGURE 13.6 TVDI representation.

Tsmin = a1 + b1NDVI, where a1 and b1 are coefficients of the regression equation for
the wet edge of the moisture condition and Tsmax = a2 + b2NDVI, where a2 and b2 are
coefficients of the regression equation for the dry edge. The TVDI broadly considers
the changes between vegetation and surface temperature.
Four important parameters, dryness, wetness, NDVI, and LST, can be perceived
from the triangle. A dry edge will develop when the LST tends to increase against
minimum NDVI. Similarly, wet edge develops when the LST tends to decrease against
maximum NDVI (Figure 13.6). As the LST decreases, the vegetation increases, the
higher the NDVI value becomes, and the lower the LST will be with the exception
of small variations. The small changes in the LST with higher NDVI values point to
the wetness of the soil in vegetation. According to Carlson (2007), variations in LST
only reveal the dryness and wetness. The drier conditions can easily be indicated by
steeper Ts–NDVI slopes (Goetz 1997). The scatterplot of Ts–NDVI has a negative
relation for most of the pixels found in the triangle. The dry edge and the wet edge
were estimated on the basis of pixel information from the area that is large enough
to represent the range of surface moisture condition and from sandy soil to a fully
vegetated surface. A plot of an NDVI pixel against a corresponding Ts is shown in
Figure 13.7 for wet and dry edges. Tsmax (dry edge) and Tsmin (wet edge) were observed
for intervals of NDVI extracted from Ts–NDVI space as shown in Figure 13.7.
The performance of satellite-based soil moisture retrieved through both RSSMM
and TVDI was assessed using root mean square error (RSME) and index of agree-
ment (Willmott 1981) d:
254 Remote Sensing for Sustainability

325
Max. land surface temperature (Kelvin)

NDVI vs. Tsmax (May, 2011)

320 R2 = 0.95, y = –21.824x + 317.7

315

310

305

300
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
(a)

306
Min. land surface temperature (Kelvin)

NDVI vs. Tsmin (May, 2011)

305 R2 = 0.61, y = 4.3394x + 300.47

304

303

302

301

300
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
(b) NDVI

FIGURE 13.7 Relationship between (a) maximum Ts and NDVI (Tsmax—dry edge) and
(b) minimum Ts and NDVI (Tsmin—wet edge).

∑
N
(Oi − Ei )2
RSME = i =1
(13.12)
N

∑
N
(Oi − Ei )2
d = 1− i =1
, (13.13)
∑ ( E −O + O −O)
N 2
i i
i =1

where N is the number of observations and Oi and Ei are the observed and estimated
values, respectively. The d index values vary between 0 and 1, indicating the low
and high relationship between observed and estimated values, respectively. Willmott
Soil Moisture Using Optical Remote Sensing and Ground Measurements 255

(1981) developed a d index, known as Willmott’s index of agreement, which is used

for validation in regression and prediction models. The model is perfect when d = 1
and RSME = 0.

13.3 RESULTS AND DISCUSSION

The RSSMM was used to analyze the soil moisture at the pixel level. The RSSMM-
derived results were then correlated with soil moisture level calculated by gravi-
metric measurements. It is clear that results derived from RSSMM and ground
measurement have a significant correlation (Figure 13.8). The relationship between
RSSMM and gravimetric (ground) measurements is statistically significant (P <
0.05) and correlation coefficients Rland ( 2
)
surface depth (RSSMM) were R02−15 cm (RSSMM) = 0.51,
R15−30 cm (RSSMM) = 0.60, and R30− 45 cm (RSSMM) = 0.57.
2 2

The effective soil depth for soil moisture measurement has been the subject of
debate in scientific literature. For example, Ghulam et al. (2004) proves that at a
10-cm depth, the visible and NIR spectral space have a close relationship, whereas
Guo et al. (1997) reported that satellite data have a correlation with soil moisture at a
depth of 20 cm. Liu et al. (1997) stated that the effective depth for estimation of soil
moisture through visible NIR is 10 cm. Carlson et al. (1995) calculated soil water
content at two depths: 0–15 and 15–30 cm. On the basis of these contradictory argu-
ments, field measurements were conducted at depths of 0–15, 15–30, and 30–45 cm to
aid our methods based on NIR–red reflectance data and TVDI.
It is evident from the results that the satellite estimates (i.e., RSSMM) and
field-measured (i.e., gravimetric soil moisture) data are in agreement (Figure 13.8
and Table 13.1). The impact of vegetation on soil moisture was not considered
in RSSMM; hence, the results present mixed information on soil and vegetation.
This poses a limitation on the precision of estimated results together with the
uncertainty of field measurements attributed to sample transportation as wind and
other metrological factors have a strong effect on soil moisture conditions. The
results are acceptable and the developed method can be applied to soil moisture
monitoring.

13.3.1 tS–ndvi SpAce (combinAtion of lSt And ndvi)

A plot of NDVI pixels against the corresponding surface temperature (Ts) presents a
triangular shape. A linear regression was applied to obtain the dry edge (Tsmax) and
wet edge (Tsmin) (Figure 13.7). A strong negative and positive relationship was found
in Tsmax and Tsmin observations, respectively. The dry edge has a negative relationship,
indicating the decrease in LST(max) when the NDVI increases. The wet edge has a
positive relationship, signifying the decrease in LST(min) with NDVI. The AT was
developed using Ts–NDVI space. These wet and dry edges were used to calculate
TVDI.
The analysis of the scatterplot (Figure 13.7) shows a stronger correlation with R2
between 0.61 and 0.95, indicating that dry and wet edges are adequately represented
by a linear equation (Figure 13.7). The results are in agreement with previous work
of Holzman et al. (2014), Patel et al. (2009), and Sandholt et al. (2002).
 256

30 30 30
In situ vs. RSSMM (0–15 cm) In situ vs. RSSMM (15–30 cm) In situ vs. RSSMM (30–45 cm)
25 R2 = 0.51, RMSE = 3.5%, y = 77.321x – 43.019 25 R2 = 0.60, RMSE = 3.6%, y = 95.594x – 53.529 25 R2 = 0.57, RMSE = 4.0%, y = 100.37x – 56.434

20 20 20

15 15 15

10 10 10

5 5 5

In situ measurement (%)

0 0 0

–5 –5 –5
0.55 0.60 0.65 0.70 0.75 0.80 0.55 0.60 0.65 0.70 0.75 0.80 0.55 0.60 0.65 0.70 0.75 0.80
RSSMM RSSMM RSSMM

FIGURE 13.8 Linear relation between gravimetric and satellite-based soil moisture measurements at different depths (0–15, 15–30, and 30–45 cm).
Remote Sensing for Sustainability
Soil Moisture Using Optical Remote Sensing and Ground Measurements 257

TABLE 13.1
Error Statistic of RSSMM and TVDI in Comparison with Gravimetric Soil
Moisture Measurements
RSSMM Relation with Gravimetric Measurements R2 RSME Willmott’s Index (d)
0–15 cm 0.51 3.5% 0.93
15–30 cm 0.60 3.6% 0.95
30–45 cm 0.57 4.0% 0.94
TVDI Relation with Gravimetric Measurements R2 RSME Willmott’s Index (d)
0–15 cm 0.75 2.4% 0.97
15–30 cm 0.70 3.0% 0.96
30–45 cm 0.63 3.8% 0.95

Figure 13.7 illustrates that the slope (−21.82) of the dry edge tends toward the neg-
ative side as compared to the slope (4.33) of the wet edge. This difference is attrib-
uted to the fraction of vegetation cover and LST as the dry edge was determined by
observing maximum LST against an interval on NDVI values from Ts–NDVI space.
The negative slope associated with the dry edge may be the result of evapotranspira-
tion and bare soil surface with confining water condition.
Hence, the dry and wet edges have a high correlation (R2 = 0.95 and R2 = 0.61,
respectively). The relationship showed that the LST was higher in the area with low
NDVI. The relationship with the dry edge shows a decrease in maximum LST with
increasing NDVI. The relationship with the wet edge is positive, showing an increase
in NDVI with an increase in minimum LST. The variation shows the rise and decline
in both the LST and NDVI. This shows that when soil moisture is high, the absorp-
tion of solar energy is used for evaporation. When soil moisture is low, the bare soil
becomes drier rapidly and there is less evaporation, and as a result, absorption is
consumed in LST. Hence, when surface temperature is high, the moisture in the root
layer is consumed to maintain a high transpiration rate.

13.3.2 SpAtiAl vAriAtion of tvdi

With dry and wet edges, the TVDI image was obtained to infer the pattern of wet-
ness and dryness. By applying Tsmin and Tsmax (extracted from Ts–NDVI) in Equation
13.11, the TVDI was obtained. Dry and very dry (TVDI > 0.6) conditions are located
in bare areas with more soil mainly in the eastern region (Figure 13.9). Moist areas
having low and very low (TVDI < 0.5) values are noticeable in the western region
(Figure 13.9). The influence of moisture was clearly visible in the study area. Moist
areas had low TVDI values where the vegetation cover is robust (Figure 13.1).
Dry condition was present in the eastward region of the study area owing to the
high TVDI values. These high TVDI values in the eastern region show a drier con-
dition as compared to the western region of the study area with predominantly irri-
gated cultivation. On the whole, the western region of the study area has low TVDI
values (0.1–0.5), which allow surface wetness as a result of the high vegetation cover
 258

69°15´0˝E 69°45´0˝E 70°0´0˝E 70°15´0˝E

Spatial variation of
temperature vegetation dryness index (TVDI)
N

25°45´0˝N
25°30´0˝N
25°15´0˝N
Temperature vegetation dryness index (TVDI)
0–0.1 0.2–0.3 0.4–0.5 0.6–0.7 0.8–0.9
km
0 3.75 7.5 15 22.5 30 0.1–0.2 0.3–0.4 0.5–0.6 0.7–0.8 0.9–1
Remote Sensing for Sustainability

FIGURE 13.9 Spatial pattern of TVDI in the study area.

Soil Moisture Using Optical Remote Sensing and Ground Measurements 259

and moisture retention capacity of soil. The extremely low TVDI values (<0.2) were
observed along the water bodies/channels. TVDI was found very effective for deter-
mining the surface wetness (moisture) and dryness.

13.3.3 compAriSon of tvdi with grAvimetric meASurementS

To prove the effectiveness of the methodology, TVDI and gravimetric measurements
were analyzed in the study area. The efficiency of TVDI as an index for assess-
ing soil moisture was compared with gravimetric measurements (Figure 13.10).
This illustrates the relationship between TVDI and gravimetric measurements
at various depths (0–15, 15–30, and 30–45 cm) for a field campaign (May 2011).
The result shows a significantly negative correlation between TVDI and gravi-
metric measurements across the respective depths during the field campaign.
(
The correlation Rland 2
)
surface depth (TVDI) was significant (P < 0.05) at all three depths
( R0−15 cm (TVDI) = 0.75, R152 −30 cm (TVDI) = 0.70, and R30
2 2
− 45 cm (TVDI) = 0.63). A significant
relationship was observed at 0–15 cm. This relationship was better than those for
other depths.
Previous studies have compared the TVDI and direct estimation of soil moisture
content. According to Sandholt et al. (2002), the relationship between TVDI and soil
moisture yields a coefficient of determination ranging from 0.23 to 0.81. Wang et al.
(2004) have found a significant negative correlation (R2 = 0.35 − 0.68) between TVDI
from NOAA-AVHRR and surface soil moisture. Patel et al. (2009) have shown that
TVDI has a significant correlation with gravimetric soil moisture at a depths of 15,
30, and 45 cm. Chen et al. (2011) also describe that a significantly negative rela-
tionship exists between the TVDI and gravimetric soil moisture measurement at
different soil depths. However, the relationship at 10–20 cm depth (R2 = 0.43) was
closest. Holzman et al. (2014) reported that the TVDI showed a strong negative cor-
relation with soil moisture measurement with R2 values ranging from 0.61 to 0.83. In
this study, the relationship is significant for depths of 0–15 and 15–30 cm, which is
also evident for some other studies (Chen et al. 2011; Holzman et al. 2014). Hence,
these results reveal the ability of the TVDI to reflect spatial variation of soil mois-
ture (Figure 13.9). It is observed that the TVDI better explains the variance in soil
moisture in upper soil layers as compared to the deeper layers. Moreover, the TVDI
results were found satisfactory for estimation of soil moisture.
The results show that the soil moisture (%) can be estimated by using linear
regression models. The linear regression models developed for the soil moisture (%)
are as follows:

0 to 15 cm: Moisture (%) = −31.04TVDI + 24.21 (13.14)

15 to 30 cm: Moisture (%) = −36.58TVDI + 29.09 (13.15)

30 to 45 cm: Moisture (%) = −38.94TVDI + 30.98. (13.16)

The spatial variation of soil moisture (%) at three different depths is shown in
Figure 13.11.
 260

30 30 30
In situ vs. TVDI (0–15 cm) In situ vs. TVDI (15–30 cm) In situ vs. TVDI (30–45 cm)
R2 = 0.75, RMSE = 2.4%, y = –31.04x + 24.21 R2 = 0.70, RMSE = 3.0%, y = –36.58x + 29.09 R2 = 0.63, RMSE = 3.8%, y = –38.94x + 30.96
25 25 25

20 20 20

15 15 15

10 10 10

5 5 5

In situ measurement (%)

0 0 0

–5 –5 –5
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8
TVDI TVDI TVDI

FIGURE 13.10 Linear regression between TVDI and gravimetric soil moisture at different depths (0–15, 15–30, and 30–45 cm).
Remote Sensing for Sustainability
 Soil moisture (%) Soil moisture (%)

3
6
9
5
8
1
4
4
8
6
0
4
8
2

12
12

0–
3–
6–
–1
–1
–2
–2
0–
4–
–1
–2
–2
–2
–3

9–
8–

12
15
18
21
12
16
20
24
28
(a) (b)

N
Soil moisture (%)

5
1
5
0
5
0
5
7
W E

0–
5–
–1
–2
–2
–3
–3
–3

10
15
20
25
30
35
S

0 12.5 25 50 75 100
(c) km
Soil Moisture Using Optical Remote Sensing and Ground Measurements

FIGURE 13.11 Spatial distribution of soil moisture (%) at soil depths (a) 0–15 cm, (b) 15–30 cm, and (c) 30–45 cm using TVDI and regression equations.
261
262 Remote Sensing for Sustainability

RSSMM and TVDI were found to be appropriate methods for estimating soil
moisture. Willmott’s Index (Willmott 1981) and RMSE were used to determine the
performance of both methods.
It can be observed from Table 13.1 that the statistical relationship between
TVDI and gravimetric measurements is significant, with less RSME than RSSMM.
Moreover, it is clear from Willmott’s Index that validation of TVDI with gravi-
metric measurements is more satisfactory than that of RSSMM. RSSMM shows
potential for estimating surface soil moisture. However, TVDI is a combination of
LST and NDVI (Ts –NDVI space), and it has been noted that LST and NDVI pres-
ent considerable change in land surface characteristics between wet edge (Tsmin)
and dry edge (Tsmax). Hence, TVDI has an advantage over RSSMM in that it con-
siders both vegetation (NDVI) and LST. Comparing RSSMM and TVDI, it is
found that the latter is an effective and a more appropriate method in determining
soil moisture.

13.4 CONCLUSION
In this study, both RSSMM and TVDI have been used for estimation of soil
moisture. It was observed that the TVDI method, which considers LST and veg-
etation index relationship, can be used effectively and appropriately for the esti-
mation of soil moisture. The main objective of this study was to investigate soil
moisture estimation using RSSM and TVDI, which were developed using NIR
and red reflectance and a combination of Landsat TM–driven LST and NDVI,
respectively. The soil moisture derived from both methods was compared with
gravimetric measurements. The relationship of TVDI and RSSMM with gravi-
metric measurements was significant at depths of 0–15, 15–30, and 30–45 cm.
The two models developed in this study delivered relatively reliable estimates
of soil moisture at different soil depths. RSSMM presents diverse information
of soil and vegetation; therefore, disintegration of the diverse pixel information
may produce more accurate results. It is also concluded that TVDI extracts soil
moisture and is useful for frequently studying soil moisture over a large area.
The TVDI has a good correlation with field-collected soil moisture (R 2 > 0.70).
The results of the study demonstrate that the TVDI is a more effective and com-
prehensive method for soil moisture estimation. Moreover, rigorous examina-
tion with concurring field data to the TVDI with temporal and spatial variation
may improve the present approach. The inferring of surface moisture status on a
regular basis can help in the monitoring of early drought, agriculture, and water
management.

ACKNOWLEDGMENTS
We are thankful to glovis.usgs.gov for creating and disseminating the data. We are
thankful to Sindh Agriculture University Tandojam for assistance with the field cam-
paign and laboratory provisions. We also express our sincere thanks to the Institute
of Space Technology (Department of Remote Sensing and Geoinformation Science)
for funding this research study.
Soil Moisture Using Optical Remote Sensing and Ground Measurements 263

REFERENCES
Albergel, C., Dorigo, W., Balsamo, G., Muñoz-Sabater, J., de Rosnay, P., Isaksen, L., &
Wagner, W. (2013). Monitoring multi-decadal satellite earth observation of soil moisture
products through land surface reanalyses. Remote Sensing of Environment, 138, 77–89.
Amiri, R., Weng, Q., Alimohammadi, A., & Alavipanah, S. K. (2009). Spatial–temporal
dynamics of land surface temperature in relation to fractional vegetation cover and
land use/cover in the Tabriz urban area, Iran. Remote Sensing of Environment, 113(12),
2606–2617.
Carlson, T. (2007). An overview of the “triangle method” for estimating surface evapotranspi-
ration and soil moisture from satellite imagery. Sensors, 7(8), 1612.
Carlson, T. N., Gillies, R. R., & Perry, E. M. (1994). A method to make use of thermal infrared
temperature and NDVI measurements to infer surface soil water content and fractional
vegetation cover. Remote Sensing Reviews, 9(1–2), 161–173.
Carlson, T. N., Gillies, R. R., & Schmugge, T. J. (1995). An interpretation of methodologies
for indirect measurement of soil water content. Agricultural and Forest Meteorology,
77(3), 191–205.
Carlson, T. N., Rose, F. G., & Perry, E. M. (1984). Regional-scale estimates of surface mois-
ture availability from GOES infrared satellite measurements. Agronomy Journal, 76(6),
972–979.
Chauhan, N., Miller, S., & Ardanuy, P. (2003). Spaceborne soil moisture estimation at high
resolution: A microwave-optical/IR synergistic approach. International Journal of
Remote Sensing, 24(22), 4599–4622.
Chen, J., Wang, C., Jiang, H., Mao, L., & Yu, Z. (2011). Estimating soil moisture using
Temperature–Vegetation Dryness Index (TVDI) in the Huang-huai-hai (HHH) plain.
International Journal of Remote Sensing, 32(4), 1165–1177.
Crow, W. T., Berg, A. A., Cosh, M. H., Loew, A., Mohanty, B. P., Panciera, R., de Rosnay, P.,
Ryu, D., & Walker, J. P. (2012). Upscaling sparse ground-based soil moisture observa-
tions for the validation of coarse-resolution satellite soil moisture products. Reviews of
Geophysics, 50(2).
Ekstrand, S. (1996). Landsat TM-based forest damage assessment: Correction for topographic
effects. Photogrammetric Engineering and Remote Sensing, 62(2), 151–162.
Evett, S., Heng, L., Moutonnet, P., & Nguyen, M. (2008). Field Estimation of Soil Water
Content: A Practical Guide to Methods, Instrumentation, and Sensor Technology.
IAEA: Vienna.
FAO. (2001). Special Report: FAO/WFP Crop and Food Supply Assessment Mission to Pakistan.
Gao, S., Zhu, Z., Liu, S., Jin, R., Yang, G., & Tan, L. (2014). Estimating the spatial distri-
bution of soil moisture based on Bayesian maximum entropy method with auxiliary
data from remote sensing. International Journal of Applied Earth Observation and
Geoinformation, 32, 54–66.
Ghulam, A., Qin, Q., Zhu, L., & Abdrahman, P. (2004). Satellite remote sensing of groundwa-
ter: Quantitative modelling and uncertainty reduction using 6S atmospheric simulations.
International Journal of Remote Sensing, 25(23), 5509–5524.
Gillies, R., Kustas, W., & Humes, K. (1997). A verification of the ‘triangle’ method for obtain-
ing surface soil water content and energy fluxes from remote measurements of the
Normalized Difference Vegetation Index (NDVI) and surface e. International Journal of
Remote Sensing, 18(15), 3145–3166.
Goetz, S. (1997). Multi-sensor analysis of NDVI, surface temperature and biophysical vari-
ables at a mixed grassland site. International Journal of Remote Sensing, 18(1), 71–94.
Guo, N., Cheng, T., Lei, J., & Yang, L. (1997). Estimating farmland soil moisture in eastern
Gansu Province using NOAA satellite data. Journal of Applied Meteorological Science
(in Chinese), 8, 212–218.
264 Remote Sensing for Sustainability

Holzman, M., Rivas, R., & Piccolo, M. (2014). Estimating soil moisture and the relationship
with crop yield using surface temperature and vegetation index. International Journal of
Applied Earth Observation and Geoinformation, 28, 181–192.
Huete, A., Hua, G., Qi, J., Chehbouni, A., & Van Leeuwen, W. (1992). Normalization of mul-
tidirectional red and NIR reflectances with the SAVI. Remote Sensing of Environment,
41(2), 143–154.
Jackson, T. J. (1993). III. Measuring surface soil moisture using passive microwave remote
sensing. Hydrological Processes, 7(2), 139–152.
Jawson, S. D., & Niemann, J. D. (2007). Spatial patterns from EOF analysis of soil mois-
ture at a large scale and their dependence on soil, land-use, and topographic properties.
Advances in Water Resources, 30(3), 366–381.
Kustas, W., & Norman, J. (1996). Use of remote sensing for evapotranspiration monitoring
over land surfaces. Hydrological Sciences Journal, 41(4), 495–516.
Li, X., & Dong, W. (1996). Methods research on monitoring drought by using remote sensing
and GIS. Remote Sensing Technology and Application, 11(3).
Li, Z., Wang, Y., Zhou, Q., Wu, J., Peng, J., & Chang, H. (2008). Spatiotemporal variability of
land surface moisture based on vegetation and temperature characteristics in Northern
Shaanxi Loess Plateau, China. Journal of Arid Environments, 72(6), 974–985. doi:
http://dx.doi.org/10.1016/j.jaridenv.2007.11.014
Liang, S., Fallah-Adl, H., Kalluri, S., JáJá, J., Kaufman, Y. J., & Townshend, J. R. (1997). An
operational atmospheric correction algorithm for Landsat Thematic Mapper imagery
over the land. Journal of Geophysical Research: Atmospheres (1984–2012), 102(D14),
17173–17186.
Liu, P., Zhang, L., Kurban, A., Chang, P., Li, L., & Polat Zhao, B. (1997). A method for moni-
toring soil water contents using satellite remote sensing. Journal of Remote Sensing,
1(2), 135–138.
Liu, S., Roberts, D. A., Chadwick, O. A., & Still, C. J. (2012). Spectral responses to plant
available soil moisture in a Californian grassland. International Journal of Applied
Earth Observation and Geoinformation, 19, 31–44.
Markham, B. L., & Barker, J. (1986). Landsat MSS and TM post-calibration dynamic ranges,
exoatmospheric reflectances and at-satellite temperatures. EOSAT Landsat Technical
Notes, 1(1), 3–8.
Nichols, S., Zhang, Y., & Ahmad, A. (2011). Review and evaluation of remote sensing meth-
ods for soil-moisture estimation. Journal of Photonics for Energy, 028001–028017.
Patel, N., Anapashsha, R., Kumar, S., Saha, S., & Dadhwal, V. (2009). Assessing potential of
MODIS derived temperature/vegetation condition index (TVDI) to infer soil moisture
status. International Journal of Remote Sensing, 30(1), 23–39.
Price, J. C. (1980). The potential of remotely sensed thermal infrared data to infer surface soil
moisture and evaporation. Water Resources Research, 16(4), 787–795.
Price, J. C. (1982). Estimation of regional scale evapotranspiration through analysis of satel-
lite thermal-infrared data. IEEE Transactions on Geoscience and Remote Sensing, (3),
286–292.
Price, J. C. (1990). Using spatial context in satellite data to infer regional scale evapotranspira-
tion. IEEE Transactions on Geoscience and Remote Sensing, 28(5), 940–948.
Prihodko, L., & Goward, S. N. (1997). Estimation of air temperature from remotely sensed
surface observations. Remote Sensing of Environment, 60(3), 335–346.
Qin, J., Yang, K., Lu, N., Chen, Y., Zhao, L., & Han, M. (2013). Spatial upscaling of in-situ
soil moisture measurements based on MODIS-derived apparent thermal inertia. Remote
Sensing of Environment, 138, 1–9. doi: 10.1016/j.rse.2013.07.003
Rajasekar, U., & Weng, Q. (2009). Application of association rule mining for exploring the
relationship between urban land surface temperature and biophysical/social parameters.
Photogrammetric Engineering & Remote Sensing, 75(4), 385–396.
Soil Moisture Using Optical Remote Sensing and Ground Measurements 265

Richardson, A. J., & Weigand, C. (1977). Distinguishing vegetation from soil background
information. Photogrammetric Engineering and Remote Sensing, 43(12).
Sandholt, I., Rasmussen, K., & Andersen, J. (2002). A simple interpretation of the surface
temperature/vegetation index space for assessment of surface moisture status. Remote
Sensing of Environment, 79(2), 213–224.
Santos, W. J. R., Silva, B. M., Oliveira, G. C., Volpato, M. M. L., Lima, J. M., Curi, N.,
& Marques, J. J. (2014). Soil moisture in the root zone and its relation to plant vigor
assessed by remote sensing at management scale. Geoderma, 221, 91–95.
Singh, R., Oza, S., Chaudhari, K., & Dadhwal, V. (2005). Spatial and temporal patterns of sur-
face soil moisture over India estimated using surface wetness index from SSM/I micro-
wave radiometer. International Journal of Remote Sensing, 26(6), 1269–1276.
Sobrino, J. A., Jiménez-Muñoz, J. C., & Paolini, L. (2004). Land surface temperature retrieval
from LANDSAT TM 5. Remote Sensing of Environment, 90(4), 434–440.
Srivastava, H. S., Patel, P., Manchanda, M., & Adiga, S. (2003). Use of multiincidence angle
RADARSAT-1 SAR data to incorporate the effect of surface roughness in soil moisture
estimation. IEEE Transactions on Geoscience and Remote Sensing, 41(7), 1638–1640.
Stisen, S., Sandholt, I., Nørgaard, A., Fensholt, R., & Jensen, K. H. (2008). Combining the tri-
angle method with thermal inertia to estimate regional evapotranspiration—Applied to
MSG-SEVIRI data in the Senegal River basin. Remote Sensing of Environment, 112(3),
1242–1255.
Vereecken, H., Huisman, J., Pachepsky, Y., Montzka, C., Van Der Kruk, J., Bogena, H.,
Weihermüller, L., Herbst, M., Martinez, G., & Vanderborght, J. (2014). On the spatio-
temporal dynamics of soil moisture at the field scale. Journal of Hydrology, 516, 76–96.
Wang, C., Qi, S., Niu, Z., & Wang, J. (2004). Evaluating soil moisture status in China using the
temperature–vegetation dryness index (TVDI). Canadian Journal of Remote Sensing,
30(5), 671–679.
Wang, H., Li, X., Long, H., Xu, X., & Bao, Y. (2010). Monitoring the effects of land use and
cover type changes on soil moisture using remote-sensing data: A case study in China’s
Yongding River basin. Catena, 82(3), 135–145.
Wang, K., Li, Z., & Cribb, M. (2006). Estimation of evaporative fraction from a combination
of day and night land surface temperatures and NDVI: A new method to determine the
Priestley–Taylor parameter. Remote Sensing of Environment, 102(3), 293–305.
Wang, L., Qu, J., Zhang, S., Hao, X., & Dasgupta, S. (2007). Soil moisture estimation using
MODIS and ground measurements in eastern China. International Journal of Remote
Sensing, 28(6), 1413–1418.
Wang, L., & Qu, J. J. (2009). Satellite remote sensing applications for surface soil moisture
monitoring: A review. Frontiers of Earth Science in China, 3(2), 237–247.
Willmott, C. J. (1981). On the validation of models. Physical Geography, 2(2), 184–194.
Xin, J., Tian, G., Liu, Q., & Chen, L. (2006). Combining vegetation index and remotely sensed
temperature for estimation of soil moisture in China. International Journal of Remote
Sensing, 27(10), 2071–2075.
Zhan, Z., Qin, Q., Ghulan, A., & Wang, D. (2007). NIR-red spectral space based new method
for soil moisture monitoring. Science in China Series D: Earth Sciences, 50(2), 283–289.
Section IV
Remote Sensing
for Sustainable Energy
 14 Earth Observation
and Its Potential to
Implement a Sustainable
Energy Supply
A German Perspective
Thomas Esch, Markus Tum, and Annekatrin Metz

CONTENTS
14.1 Remote Sensing in Support of Energy Policy .............................................. 269
14.2 Geodata for the Energy Sector— Examples of Satellite-Based Analysis ..... 270
14.2.1 Potential Analysis: Bioenergy........................................................... 270
14.2.2 Mapping of Agricultural Cultivation Patterns .................................. 272
14.2.3 Determination of Population Structures and Site Conditions
for Construction of Heating Networks.............................................. 272
14.3 Conclusions ................................................................................................... 275
References .............................................................................................................. 275

14.1 REMOTE SENSING IN SUPPORT OF ENERGY POLICY

Today, satellite sensors provide digital recordings of the earth’s surface in a spa-
tial resolution of approximately 1 km (e.g., SPOT-VEGETATION, MODIS, MERIS,
Proba-V) to less than 1 m (z. B. WorldView, QuickBird, IKONOS). The temporal
repetition cycle varies from daily to monthly surveys. The Earth observation technol-
ogy has steadily evolved in recent years from a beginning characterized by a rather
experimentally imprinted alignment toward operational services, which ensures a
long-term, regular, and quality-assured provision of spatial data and higher-value
information products. On the one hand, these data and products can be recorded
while targeted on request for specific applications, periods, or regions as needed, or
on the other hand, these can be implemented in a systematic and comprehensive mon-
itoring of the Earth’s surface. Thus, satellite-based Earth observation provides prom-
ising applications to implement a sustainable energy supply. Remote sensing–aided
applications such as wind field analysis and the levying of irradiation data or the
determination of solar surface potentials are commercially exploited and have been
available for several years and are firmly established in existing planning processes.

269
270 Remote Sensing for Sustainability

Nevertheless, the need for the provision of missing or supplementation of existing

geodata and geoinformation products to support the energy turnaround in Germany
has been formulated repeatedly by representatives from policy, management, plan-
ning, the private sector, and academia.

14.2 GEODATA FOR THE ENERGY SECTOR—

EXAMPLES OF SATELLITE-BASED ANALYSIS
Looking over the formulated demand for spatially and thematically enhanced spatial
data in support of sustainable energy supply, the collection of geospatial data from
satellite image analysis especially with regard to the potential assessment, site selec-
tion, conflict prevention of interest, and monitoring of impacts and trends must be
regarded as particularly beneficial. This applies to the comprehensive determina-
tion of the biomass volume in the context of site assessment or the parcel-related
acquisition of agricultural crops for the estimation of regional straw potentials.
Furthermore, the use of remote sensing techniques can also be used for the detec-
tion, visualization, and documentation of developments and trends of, for example,
the transformation of landscapes in the context of energy policy decisions or the
growing energy consumption and increasing use conflicts in the wake of steadily
progressing land use by settlements and transport infrastructure. The following three
exemplary recent research and development works of the department’s land surface
of the Earth Observation Center at the German Aerospace Center (DLR) will be
presented: the modeling of sustainable bioenergy potentials, the acquisition of agri-
cultural growing patterns, and the accounting of settlement structures with regard
to their suitability for the construction of heating networks. All these applications
are aimed at supporting land management, which is geared toward the promotion of
sustainable energy supply.

14.2.1 Potential analysis: Bioenergy

The Biosphere Energy Transfer Hydrology (BETHY/DLR) model of the DLR is
driven by high spatial and temporal resolution remote sensing data and derived
parameters such as leaf area index (LAI) and is used for the quantification of biomass
and bioenergy potentials. As an input data set for the evaluation described below,
global LAI data, which are available in the form of 10-day composite time series as
well as information on land cover and land use, derived from the Global Land Cover
2000 (GLC2000) were used. In addition to these remote sensing data, BETHY/DLR
also requires diverse meteorological information on air temperature, precipitation,
wind speed, and cloud cover. These data are provided by the European Center for
Medium-Range Weather Forecast in an appropriate format. Furthermore, BETHY/
DLR uses information on the dominant soil type from the Harmonized World Soil
Database and a digital elevation model of the Shuttle Radar Topography Mission.
In vegetation models such as BETHY/DLR, plant growth is parameterized in a
way that, in a first step, the biochemical processes of photosynthesis are modeled at
the leaf level. This is followed by an extrapolation of both the structure of the plant
Earth Observation and Its Potential to Implement Energy Supply 271

and the interaction (e.g., energy flows, water circulation) included in the calculations
between soil, vegetation, and atmosphere from the leaf level to the inventory level.
As a result, the model initially provides the carbon amount absorbed by the vegeta-
tion per unit area and time out of the atmosphere—the gross primary production.
Since each plant emits carbon back to the atmosphere through autotrophic respira-
tion in the form of CO2, in the balance sheet, less carbon in the form of biomass is
bound in the plant than was originally recorded. This output of the model is called
Net Primary Productivity (NPP). With the use of conversion factors, NPP can be
converted into biomass and energy potentials. For this purpose, conversion factors
such as the relationship of grain to straw or root to shoot and the ratio of aboveground
to underground biomass are also needed, as well as information on the specific water
content of the dry biomass and its energy yield per kilogram (lower heating value).
Additionally, estimates of competing uses are likewise required (e.g., soil fertiliza-
tion by straw, entry into stables, etc.).
In Figure 14.1, agricultural and forestry bioenergy potentials for Germany in
2012 are illustrated. Energy potentials of straw are locally a factor of 10 lower
than the potential growth of wood; as for the agricultural potential, because of
competing uses, only a small proportion of the straw (20%) is considered to be
usable for energy production. In contrast, 80% of forest growth is considered to
be theoretically available. For 2012, the overall energy potential was calculated
to be 572 petajoules (PJ) for agriculture and 1938 PJ for forestry. A detailed
description of all input data and the methodology for modeling of the NPP and
conversion of NPP in agricultural and forest energy potentials can be found in
Wißkirchen et al. (2013) and Tum et al. (2013).

Wood energy Straw energy

[TJ/km2/a] [TJ/km2/a]
20 2

0 0
100 km

FIGURE 14.1 Energy potential of wood growth (forestry) and straw (agriculture) in
TJ/km2/a for 2012.
272 Remote Sensing for Sustainability

14.2.2 MaPPing of agricultural cultivation Patterns

In connection with the policy on climate change, the efforts to reduce CO2 consump-
tion and the implementation to a sustainable energy supply have been added as a new
component in addition to aspects such as food security, food quality, and sustainable
management in view of the agricultural production in Germany and the increased
use of bioenergy raw materials. This contributed input to the funding policy with
regard to biogas plants and the Renewable Energies Act (EEG). The opportunity to
obtain co-financing of biogas power plants through the EEG has greatly affected the
cultivation of crops in the past decade, for example, in the form of an increase in the
acreage of silage maize (DMK 2013). The new support measures for the cultivation
of energy raw materials also drive increased reallocation of grassland in areas used
for farming. In addition to the loss of arable land for food production, the environ-
mental impact of large-scale cultivation of maize and other energy crops should be
evaluated critically.
The sometimes rapid processes of change in the cultural landscape bear in mind
that a continuous and timely monitoring of intensity of use and acreage develop-
ment is needed to promptly identify trends and unfavorable regional developments
in terms of sustainable management to be able to effectively counteract such adverse
developments based on this information. Satellite remote sensing provides an ideal
base, because it enables spatially and temporally precise analysis, which can also be
easily integrated into spatial information systems and with other data, such as sta-
tistical information or geometries of cadastral surveying. Moreover, meaningful use
potentials can be determined in order to determine, for example, areas with exploit-
able straw shares by the cultivation of grain or spatial potential through enhanced
incurred hedge trimming at national care measures. In addition, relevant information
for optimized positioning can be used for planning biogas plants.
Facing agriculture-related evaluations, the value of satellite remote sensing
becomes more apparent particularly with the combined analysis of spectral informa-
tion and phenological development curves (e.g., from vegetation indices). Esch et al.
(2014) used high-resolution multi-seasonal recordings of Sensors LISS-3 (23.5 m
spatial resolution) and AWiFS (56 m spatial resolution) of the satellite IRS-P6 to
parcel-based classification of grassland and main crops of arable crops. The aim of
this approach is to provide a methodology that allows for quick and flexible acquisi-
tion of cultivation patterns on predetermined areas of interest. The result of such
analysis is shown in Figure 14.2.

14.2.3 DeterMination of PoPulation structures anD site conDitions

forconstruction of Heating networks
The increased use of efficient energy technologies is indispensable in order to achieve
climate change goals as well as the establishment of a sustainable energy supply.
The expansion of local and district heating networks, especially in combination with
plants for combined heat and power as well as the increased use of renewable district
heating, is of central importance here. In order to evaluate the potential for estab-
lishing a wired heat supply, an analysis of settlement structural conditions and the
Earth Observation and Its Potential to Implement Energy Supply 273

(a)

Classification
211 (Crops)
216 (Maize)
220 (Beet)
232 (Rapeseed)
400 (Pasture)
No agriculture
(b) 0 2km

FIGURE 14.2 Use of multiseasonal satellite imagery (a) to determine the agricultural cul-
tivation pattern (b).

resulting potential for development is needed. An important element is the provision

of current, nationwide heat power (energy) potential, which can be used to address
the physical–structural location conditions for district heating. The heat power
potential is calculated by the ratio of annual heat energy demand of the building and
investment costs to provide the heat needed. The corresponding characteristic value
represents the annual exploitable potential by the supply quantities of heat (kWh/a)
per invested monetary unit (euros). The specific heat energy consumption value
depends on the type and use of the building, the age of the building, and the climatic
conditions. The estimation of costs for heat infrastructure is based on an analysis of
the lengths and associated costs for main supply and service lines, which in turn can
be modeled on an evaluation of the road network and the location of buildings. The
heat demand is made up of the building volume and the specific heat demand value,
while the investment costs include all expenses for distribution, connecting cables,
house transfer stations, and savings for boilers.
Figure 14.3 shows the result of a Germany-wide analysis of heat power potential
that has been identified through a combination of top-down and bottom-up approach
as methods of digital image analysis and geographic information systems (DLR
2011). The top-down approach is used during the estimation of the heat demand
on the basis of data on the housing stock and the climatic situation. The building
stock is determined by Infas-geodata (INFAS 2015) of the building type (one/two-
family houses, small and large apartment buildings) and building age (nine peri-
ods of construction) as well as information from the Federal Statistical Office for
building use (residential buildings, nonresidential buildings). Climatic conditions
274 Remote Sensing for Sustainability

300,000 400,000 500,000 600,000 700,000 800,000 900,000

5,300,000 5,400,000 5,500,000 5,600,000 5,700,000 5,800,000 5,900,000 6,000,000 6,100,000 Potential
kWh/
≤2
>2–2.5
>2.5–3
>3–3.5
>3.5–4
>4–4.5
>4.5–5
>5–5.5
>5.5
Existing
networks
>10 km

FIGURE 14.3 Heat power potential of communities in Germany in 2011, without consider-
ing the cost of a conventional heat supply.

can be displayed using the degree days provided by the German Weather Service
(http://www.dwd.de). The calculation of the wireline investment costs is bottom-
up involving data of the Authoritative Topographic–Cartographic Information
System (accessible at http://www.atkis.de) on the situation of human settlements
and the course of the road network (determining main supply and service lines)
and makes use of information about the existing buildings (location, area) extracted
from Digital Topographic Maps (DTK 25), which can be obtained at http://www
.adv-online.de. Although data on the situation, floor space, and even the volume of
buildings are available nationwide in the form of spatial data (e.g., house outlines
or 3D building model LoD1), the high cost of ~100,000 to 350,000 euros currently
prevents users to access them. The evaluation results show that the thermal power
potential in addition to the local heat demand is highly dependent on the length of
heat distribution. This is closely linked to the settlement structure. Therefore, core
cities and their surrounding particularly exhibit high values, while rural areas have
relatively low potential. For example, in the southern parts of Germany, relatively
favorable structures are available, whereas the northeast has relatively unfavorable
structures.
Earth Observation and Its Potential to Implement Energy Supply 275

14.3 CONCLUSIONS
Satellite-based remote sensing offers versatile opportunities for the provision of geo-
information for energy-related issues. Recent results and experience show that Earth
observation can make important contributions in particular with regard to potential
calculation, monitoring of current availability, and the monitoring of changes and
trends.
Advantages over established approaches and data sets in this connection arise
in particular with regard to timeliness, coverage, comparability, spatial detail,
and update costs. Nevertheless, an operational provision of the necessary remote
sensing–based geoinformation products in the form of reliable services is central
for commercial production and ultimately for success in the planning and man-
agement sector. Cornerstones in this context ensure a high spatial, temporal, and
qualitative continuity; transparent and reliable procurement costs and modalities;
and the exploitation of synergies (e.g., with existing spatial data held by the cadas-
tral survey). However, unclear responsibilities, requirements, and user requirements
currently complicate targeted developments. Against this background, there is a par-
ticular need for the coordination and pooling of research and development activities
between the different actors of planning/management, industry, and science. A net-
worked and concerted approach enables the effective development and utilization of
existing expertise and synergies. At the same time, a close connection to the GMES/
Copernicus environment should be ensured in order to meet the requirements for
data continuity and service delivery.

REFERENCES
DLR—Deutsches Zentrum für Luft- und Raumfahrt e. V. 2011. Potenzialanalyse zum Aufbau
von Wärmenetzen unter Auswertung siedlungsstruktureller Merkmale. Endbericht
(BMVBS/BBSR FKZ 3004775). www://elib.dlr.de/76816/1/Waermenetzpotenzial
_DLR_Endbericht_final.pdf (accessed September 14, 2015).
DMK—Deutsches Maiskomitee e. V. 2013. Bedeutung des Maisanbaues in Deutschland. www
.maiskomitee.de/web/public/Fakten.aspx/Statistik/Deutschland (accessed September 14,
2015).
Esch, T., Metz, A., Keil, M., Marconcini, M. 2014. Combined use of multi-seasonal high
and medium resolution satellite imagery for parcel-related updating of cropland and
grassland. International Journal of Applied Earth Observations and Geoinformation,
28C:230–237.
INFAS: Institut für angewandte Sozialwissenschaften GmbH. https://www.infas.de/ (accessed
December 17, 2015).
Tum, M., Günther, K.P., McCallum, I., Kindermann, G., Schmid, E. 2013. Sustainable bioen-
ergy potentials for Europe and the globe. Geoinformatics & Geostatistics: An Overview,
S1.
Wißkirchen, K., Tum, M., Günther, K.P., Niklaus, M., Eisfelder, C., Knorr, W. 2013. Quantifying
the carbon uptake by vegetation for Europe on a 1 km2 resolution using a remote sensing
driven vegetation model. Geoscientific Model Development 6:1623–1640.
 15 Use of Nighttime
Imaging Data to Assess
Decadal Trends in
Energy Use in China
Yanhua Xie and Qihao Weng

CONTENTS
15.1 Introduction .................................................................................................. 277
15.2 Some Background on NTL-Based Estimation of Energy Consumption ...... 278
15.3 Study Area and Data Sets ............................................................................. 279
15.4 Methodology ................................................................................................. 281
15.4.1 Data Preprocessing ........................................................................... 282
15.4.2 Estimation of Pixel-Based EC .......................................................... 282
15.4.3 Analysis of Spatiotemporal Pattern of EC........................................ 283
15.5 Results........................................................................................................... 285
15.5.1 Pixel-Based EC for Mainland China ................................................ 285
15.5.2 Spatiotemporal Pattern of EC at Multiple Scales ............................. 287
15.5.3 National and Regional Contributions of Urban Areas to EC ........... 290
15.6 Discussion and Conclusions.......................................................................... 293
References .............................................................................................................. 294

15.1 INTRODUCTION
In general, energy refers to the ability to do work and can be in the form of elec-
tricity, thermal, chemical, mechanical, gravitational, nuclear, radiant, sound, and
motion. The sources of energy can be fossil (e.g., petroleum, coal, natural gas, wood,
etc.), renewable (wind, solar, geothermal, hydrogen, etc.), and fissile (e.g., uranium,
thorium, etc.) (Bilgen 2014). Being one of the three crucial themes (i.e., energy con-
sumption, urbanization, and carbon dioxide emission) in the 21st century, energy
has caught much attention as it is essential for socioeconomic development for all
countries (Al-mulali et al. 2012; Weng 2013). For the past decades, attention exerted
on energy issues in developing countries (e.g., China and India) has been especially
intense as these countries are currently experiencing fast urbanization. The leap of
their economies has caused rapid growth of energy consumption, and this trend is
expected to continue in the near future (Bilgen 2014). However, if the unsustain-
able energy consumption (e.g., overexploitation of nonrenewable resources and air

277
278 Remote Sensing for Sustainability

pollution caused by fossil fuels) continues, environmental problems will definitely

deteriorate (Adams and Shachmurove 2008; Ahmad et al. 2015; Bloch et al. 2015;
Cherni and Kentish 2007). A better knowledge of the spatiotemporal patterns of
energy consumption at multiple scales is thus a critical step for the sustainable devel-
opment of energy as it provides policy makers with the guidelines for energy produc-
tion, distribution, and management.

15.2 SOME BACKGROUND ON NTL-BASED

ESTIMATION OF ENERGY CONSUMPTION
A major challenge for analyzing multiscale spatiotemporal patterns of energy con-
sumption is the lack of detailed data in a temporally and spatially consistent man-
ner, and this situation is more severe in developing countries (Parshall et al. 2009).
Traditionally, the primary source of energy consumption data is census data with
the interval of years and spatial resolution of administrative units (e.g., country,
province), hardly satisfying the requirements of spatiotemporal analyses at multiple
scales (especially for local scales such as urban level).
Remotely sensed nighttime light (NTL) imagery from the Operational Linescan
System (OLS) of the Defense Meteorological Satellite Program (DMSP) has been
proven effective in the estimation of electricity consumption (EC) in a spatially explicit
and consistent manner, because it provides frequently repeated records of EC for out-
door lighting (Amaral et al. 2005; Cao et al. 2014; Chand et al. 2009; Elvidge et al.
2001; He et al. 2012, 2014; Letu et al. 2010; Lo 2002; Welch 1980). Meanwhile, the new
generation of NTL image from the Visible Infrared Imaging Radiometer Suite onboard
the Suomi National Polar-Orbiting Partnership satellite has shown an enhanced rela-
tionship between EC and NTL at the sub-country scale (Shi et al. 2014). NTL data from
DMSP-OLS are especially attractive given its long time archive (1992 to the present).
The potential use of DMSP NTL images for characterizing EC pattern was first
noted in the 1980s by Welch (1980). Since then, a number of studies have demon-
strated the relationship between DMSP-OLS NTL imagery and EC at national, state/
provincial, and county scales. For instance, Elvidge et al. (1997, 2001) reported the
country-scale relationships between lit area and EC with 21 and 200 countries, respec-
tively; Letu et al. (2010) also demonstrated a country-scale relationship between EC
and saturation-corrected DMSP NTL intensity for 10 Asian countries; Amaral et al.
(2005) modeled EC in the Brazilian Amazon from a lighted area at the municipal
level; Lo (2002) established a logarithmic relationship between EC and lit area for 35
Chinese capital cities. In recent years, the examination of spatiotemporal dynamics of
EC at different scales has been conducted since the availability of time-series DMSP-
OLS NTL images. For instance, by using DMSP-OLS NTL data from 1993 to 2002,
Chand et al. (2009) characterized spatiotemporal changes in EC patterns in the major
cities and states of India; He et al. (2012, 2014) examined EC pattern at the county
level in Mainland China using DMSP-OLS NTL data during 1995–2008 and 2000–
2008, respectively; to detect the spatiotemporal pattern of EC at the sub-county level,
Cao et al. (2014) proposed a top-down method to model pixel-based EC from 1994 to
2009 in China, using NTL, population density, and GDP as the independent variables.
Use of Nighttime Imaging Data to Assess Decadal Trends 279

Despite a number of research showing the capacity of NTL as an indicator of EC

with varying degrees of success, most of them focused on global-, continental-, or
national-level estimations. However, increasing municipal authorities and planners
are addressing energy issues within the context of local climate and sustainabil-
ity initiatives, requiring adequate supply of small-scale energy data (Parshall et al.
2009). Additionally, each study focused on the analyses at a single scale, ignoring
the importance of multiscale analyses, which may offer potentials for better under-
standing the process over the interaction between socioeconomic developments and
EC. Meanwhile, although studies have been undertaken to examine the relation-
ship between NTL and EC, there is limited assessment from the current literature
about whether NTL can be a good indicator of overall energy consumption (Wu et
al. 2014; Xie and Weng 2016). According to our initial assessment, a significantly
positive relationship exists between national NTL and overall energy consumption
even though it includes both EC and energy consumed in other forms such as coal
and natural gas, which cannot be directly detected by the DMSP-OLS sensor. In this
chapter, we characterized the decadal trend (2000 to 2012) of EC at sub-country
scales (i.e., provincial, prefectural, and urban scale) in Mainland China by using
time-series DMSP-OLS NTL images. The same procedure could be followed for the
spatiotemporal analyses of overall energy consumption. The results further provided
a spatially explicit way to evaluate the relationship between China’s urbanization and
its EC in the first decade of the 21st century.

15.3 STUDY AREA AND DATA SETS

Mainland China was selected as the study area, where a rapid urbanization occurred
since the economic reform in 1978, especially after the mid-1990s (Wang 2014).
Specifically, urbanization in China accelerated for the first decade of the 21st cen-
tury (Huang et al. 2015). Population size reached 1.34 billion at the end of 2010,
and the urbanization rate of population increased from 36.22% to 51.27% between
2000 and 2011 (Wang 2014), with an annual increase of 3.21%. The rapid economic
growth associated with urbanization steadily promoted the demand of energy,
with the national consumption of electricity increased by 3.6 times, from 1361.78
billion kWh in 2000 to 4953.50 billion kWh in 2012.
To capture regional differences, the study area was divided into four regions
(eastern: Beijing, Fujian, Guangdong, Hebei, Jiangsu, Shandong, Shanghai, Tianjin,
and Zhejiang; northeastern: Heilongjiang, Jilin, and Liaoning; central: Anhui, Henan,
Hubei, Hunan, Jiangxi, and Shanxi; and western: Chongqing, Gansu, Guangxi,
Guizhou, Inner Mongolia, Ningxia, Qinghai, Shaanxi, Sichuan, Xinjiang, Xizang,
and Yunnan). Table 15.1 shows that, within the overall EC across Mainland China,
the amount consumed in eastern China increased steadily from 47.24% to 50.79%
from 2000 to 2006, but this value reduced to 48.08% in 2012. For western China, the
proportion remained stable until 2005 and started to increase from 21.85% in 2005
to 25.75% in 2012. The ratio of EC in northeastern China dropped dramatically, from
10.89% in 2000 to 6.79% in 2012. However, the ratio was around 19.50% for central
China at all times.
280 Remote Sensing for Sustainability

TABLE 15.1
National EC and the Percentage Consumed in Each Region
EC (Billion kWh) Eastern (%) Central (%) Northeastern (%) Western (%)
2000 1361.78 47.24 19.66 10.89 22.21
2001 1468.90 47.91 19.84 10.33 21.92
2002 1638.61 48.87 19.66 9.67 21.80
2003 1888.02 49.62 19.53 9.22 21.63
2004 2174.88 49.94 19.28 8.81 21.97
2005 2474.71 50.74 19.14 8.26 21.86
2006 2833.88 50.79 18.97 7.90 22.34
2007 3253.25 50.35 19.42 7.53 22.70
2008 3431.62 50.08 19.39 7.51 23.02
2009 3655.48 49.95 19.30 7.36 23.39
2010 4192.83 49.61 19.31 7.25 23.83
2011 4694.78 48.54 19.40 7.02 25.04
2012 4952.50 48.08 19.37 6.79 25.76

Source: China Statistical Yearbook 2001–2013 (http://www.stats.gov.cn/tjsj/ndsj/). The original EC data

were reported by each province. Thus, they were reorganized by the authors into national and
regional statistics.

The data sets used included DMSP-OLS NTL imageries, MODIS 16-day enhanced
vegetation index (EVI) composites (MOD13A2), Gridded Population Density data
(GPWv3), statistical EC data, and administrative boundary (Table 15.2). All image-
based data sets were from 2000 to 2012 except for population density data for 2000,
2005, and 2010 because of their availability. The National Geophysical Data Center
(NGDC) provided publicly available DMSP-OLS NTL imageries (last access in
January 2015) with cloud-free composite, which contained persistent lights from

TABLE 15.2
Data Sets Used and Descriptions
Data Sets Period Data Source
DMSP-OLS NTL data 2000–2012 NOAA-NGDC (http://ngdc.noaa.gov/eog/dmsp.html)
MODIS 16-day EVI 2000–2012 USGS (https://lpdaac.usgs.gov/)
composites (MOD13A2)
Gridded population 2000, SEDAC (http://sedac.ciesin.columbia.edu)
density map (GPWv3) 2005, 2010
Electricity consumption 2000–2012 China Statistical Yearbook (http://www.stats.gov.cn/)
data (provincial level)
Electricity consumption 2001, Urban Statistical Yearbook of China (http://www.stats.gov.cn/)
data (district level) 2005, 2010
Administrative 2015, SHP Global Administrative Areas (http://www.gadm.org/)
boundary format
Use of Nighttime Imaging Data to Assess Decadal Trends 281

human activities. The digital number (DN) of pixels ranged from 0 to 63, with a
spatial resolution of nearly 1 km (at the equator). MODIS vegetation index data
(MOD13A2) were acquired from the United States Geological Survey (USGS), with
the temporal resolution of 16 days and the spatial resolution of 1 km. EVI images
were selected instead of normalized difference vegetation index images as they tend
to saturate at suburban areas where dense vegetation and high NTL values may exist
(Zhang et al. 2013). A Gridded Population Density map (GPWv3) was created by
SEDAC (NASA Socioeconomic Data and Application Center), providing globally
consistent and spatially explicit human population information and data. The grid
cell resolution was ~5 km at the equator. Additionally, the statistical data of province-
level EC and EC of urban districts (i.e., Shixiaqu in Chinese) at the prefecture-level
were collected from China Statistical Yearbook and the Urban Statistical Yearbook
of China, respectively. Furthermore, the administrative boundary file was acquired
from Global Administrative Areas.

15.4 METHODOLOGY
The method included four major steps: preprocessing, estimating gridded EC and
assessing the accuracy of estimations, mapping urban dynamics, and analyzing spa-
tiotemporal pattern of EC at multiple scales (Figure 15.1).

DMSP/OLS GPWv3 pop MOD13A2

NTL density EVI
(a)
Reprojecting, resampling Reprojecting and Reprojecting and yearly
and intercalibration resampling maximize

Combining three data sets Calibrating province-level

Thresholding mapping
to obtain corrected NTL models

Urban extent Estimating gridded EC

(c) Pixel-based EC
from 2000–2012
(d)
Analyzing spatiotemporal
Accuracy assessment
pattern of EC

Administrative EC of urban
(b)
boundary district

FIGURE 15.1 The flowchart of this study, with (a) preprocessing, (b) estimating pixel-based
EC and accuracy evaluation, (c) mapping urban dynamics, and (d) analyzing spatiotemporal
pattern of EC at multiple scales.
282 Remote Sensing for Sustainability

15.4.1 Data PreProcessing

Image-based data sets were processed to the same projection (i.e., Lambert Azimuthal
Equal Area) and spatial resolution (i.e., 1 km). Meanwhile, yearly maximum com-
posite was computed for EVI:

EVImax = max (EV1, EV2,…EVI23), (15.1)

where EVImax is the yearly maximum EVI of a total of 23 16-day composites for
each year.

15.4.2 estimation of Pixel-BaseD ec

In this chapter, a pixel-based EC was estimated instead of using administrative bound-
ary as the unit so that EC at multiple scales could be estimated through aggregation.
To obtain better gridded EC, DMSP-OLS NTL should be enhanced to eliminate the
two notorious drawbacks (i.e., lacking onboard calibration system, and having a satu-
ration effect and a blooming effect in urban cores and peri-urban areas, respectively).
Thus, intercalibration was first applied to eliminate inconsistent DNs between NTL
imageries. The method proposed by Elvidge et al. (2009) was adopted by assum-
ing Mauritius, Puerto Rico, and Okinawa (in Japan) as the invariant regions (Wu
et al. 2013). To eliminate the saturation effect of NTL in urban cores and to account
for human activities in the unlit pixels, population density data were used (popula-
tion density data of 2000, 2005, and 2010 were used to adjust NTL for 2000–2002,
2003–2007, and 2008–2012, respectively) followed by EVI adjustment (Meng et al.
2014; Zhang et al. 2013):

NTL cal = w * NTL int * ln(PD + 1) (15.2)

NTLadj = (1 − EVImax) * NTLcal, (15.3)

where NTLint, NTLcal, and NTLadj are intercalibrated, population density–calibrated,

and population- and EVI-adjusted NTL DN value, respectively; PD refers to popu-
lation density; and w is the weight for lighted and unlit pixel. w equaled to 1.0 for
lighted pixels and 0.34 for unlit areas based on previous research on the gap between
the detected electricity access in unlit and lighted area (Doll and Pachauri 2010;
Meng et al. 2014). The adjustment using EVI was based on the assumption that veg-
etation cover and human activities were inversely correlated (Zhang et al. 2013). For
the pixels with EVI < 0.1 or EVI > 0.9, NTLadj was set to 0, based on the assumption
that those regions were usually deserts, glaciers, water bodies, or dense forests with
few human activities.
The province-level statistical EC data were used to calibrate the relationship
between EC and NTLadj owing to the difficulty of obtaining statistical data at sub-
province levels. Additionally, to absorb the errors that might be introduced by
Use of Nighttime Imaging Data to Assess Decadal Trends 283

intercalibration and to capture the yearly difference of the relationship between EC

and NTLadj, the model was calibrated for each year:

ECpro = a * NTLpro + b, (15.4)

where ECpro and NTLpro refer to the EC and the sum of NTLadj at the province level,
respectively, and a and b are the coefficients. We then constructed pixel-based time-
series EC from 2000 to 2012 based on the assumption that the calibrated provincial
models were applicable to pixels:

EC yi = a y * NTL yi , (15.5)

where y refers to year index and i is the index of pixel. Further, normalization was
conducted to regulate the total EC at the provincial level:

k yp = EC yp ∑ EC ,
p
yi (15.6)

where ECyp is the statistical EC for province p in year y, ∑ EC

p
yi is the total esti-
mated EC for each province, and k yp refers to the normalization factor. Finally, grid-
ded EC was estimated by

EC ypi = k yp * EC yi . (15.7)

A per-pixel evaluation was not possible because of the lack of high-resolution

reference EC. Because of this, previous research on the estimation of gridded EC
did not include the process of accuracy evaluation (Cao et al. 2014; Zhao et al. 2012).
To validate the results of pixel-based EC in this chapter, we used the prefecture–
city level statistical EC data as the reference. Specifically, EC by urban district (i.e.,
Shixiaqu in Chinese) at the prefectural level was used.

15.4.3 analysis of sPatiotemPoral Pattern of ec

To detect the spatiotemporal pattern of EC at different scales, pixel-based EC was
aggregated to the provincial and prefectural scale. Additionally, to study the rela-
tionship between EC and urbanization, estimated gridded EC was aggregated to
urban areas. Instead of using administrative units such as Prefectural City and
County to define urban environment, we referred to built-up area as urban areas
because (1) administrative units cover not only urban extent but also peri-urban
and rural areas, and (2) the use of administrative boundaries to define urban areas
is problematic as they are designed for political purposes, failing to capture urban-
ization dynamics and its impact on EC (Meng et al. 2014). Among the methods
proposed to map regionally and globally consistent urban extents from DMSP-OLS
284 Remote Sensing for Sustainability

NTL data, including iterative unsupervised classification (Zhang and Seto 2011),
SVM-based classification (Cao et al. 2009; Pandey et al. 2013), and threshold tech-
niques (Liu et al. 2012; Zhou et al. 2014), we adopted the technique of threshold
proposed by Liu et al. (2012) with the consideration of its simplicity and reasonable
accuracy (Zhou et al. 2015) and the study area.
Global and local Moran’s I indices were utilized to characterize the spatial pat-
tern of EC at different scales. The global Moran’s I index shows the nationwide
spatial correlation of EC, while the local one mainly reflects the heterogeneity of EC
between neighbor areas (Anselin 1995, 1996). The equation of global Moran’s I is

I=
N ∑ ∑ w (x − x )(x − x ) ,
i j
ij i j

(∑ ∑ w ) ∑ (x − x )
(15.8)
2
ij i
i j i

where N is the sample size, wij is the matrix of spatial weight, xi and xj are EC for the
ith and jth unit at the scale under consideration, and x refers to the average value of x.
The value of global Moran’s I index ranges from −1 to 1. Negative values indicate
negative spatial autocorrelation and the inverse for positive values. Further, local
Moran’s I was applied to explore where high or low EC concentrations occurred. The
equation of local Moran’s I is

Ii =
( xi − x )
si2 ∑ j , j ≠i
 wij ( xj − x )  , (15.9)

where

s 2
=
∑ j , j ≠i
( xj − x )2
− x2 . (15.10)
i
N −1

Local Moran’s I describes four types of local spatial autocorrelation. The positive
value of Ii indicates a high EC surrounded by similarly high values of EC or a low EC
surrounded by low values of EC, while the negative value of Ii implies a low value
surrounded by high values or a high value surrounded by low values.
Further, the index proposed by He et al. (2012) was used to examine the temporal
pattern of EC. The formula of the index is

n n n

n ∑ xi ECi − ∑ ∑ EC xi i

SLOPE = i =1 i =1 i =1
2 , (15.11)
n  n 
n ∑
i =1
xi2 − 
 ∑ i =1
xi 

Use of Nighttime Imaging Data to Assess Decadal Trends 285

where n is equal to 13, representing the number of years between 2000 and 2012, xi
refers to numbers 1 to 13 with increments of 1, and ECi is the EC in the ith province,
prefectural city, or urban area in year i. The negative value of SLOPE indicates a
decreasing trend of EC and the inverse trend for positive SLOPE.

15.5 RESULTS
15.5.1 Pixel-BaseD ec for mainlanD china
Table 15.3 shows the calibrated yearly model with R2 all above 0.84. The value
of coefficient a for province-level models increased remarkably and steadily from
0.0215 in 2000 to 0.0536 in 2012, indicating the increasing trend of EC intensity
per unit of NTL. Using the calibrated models in Table 15.3, we estimated pixel-
based EC for Mainland China from 2000 to 2012. Figure 15.2 presents the estimated
result from 2000 to 2012 with a temporal step of 4 years. It needs to note that the
introduction of EVI greatly enhanced variation and reduced the blooming effect in
the estimated EC maps. Visually, the overall pattern of EC in Figure 15.2a through
d was almost identical, with some regions having remarkably high EC, such as the
Beijing–Tianjin–Tangshan metropolitan region, the Yangtze River Delta, the Pearl
River Delta, the Sichuan Basin, and most of the capital cities. It is also shown in
Figure 15.2 that the urban cores tended to have a higher intensity of EC than their
suburban and peri-urban regions owing to more intensive economic activities (e.g.,
service industry) in Chinese urban cores. The highest per-pixel EC in urban cores
could be as high as 38.5, 39.8, 50.8, and 58.4 million kWh for 2000 to 2012, respec-
tively. Nevertheless, much less electricity was consumed in rural regions and non-
EC regions could be forests, deserts, and water bodies. Regionally, urban areas in
eastern China used more electricity than other regions, while the majority of western
China had less access to electricity during the past decade. Additionally, it is shown
in Figure 15.2 that Shanghai has the most intense urban EC. The overall pattern of
EC showed that urban areas had experienced the most increase in EC across China
(e.g., rural regions).

TABLE 15.3
Yearly Regression Result for Provincial-Level Models (with P = 0.0000 for All
Regressions)
Year a R2 Year a R2
2000 0.0215 0.8519 2007 0.0491 0.9092
2001 0.0236 0.8413 2008 0.0478 0.9168
2002 0.0264 0.8836 2009 0.0463 0.8492
2003 0.0301 0.9095 2010 0.0524 0.8744
2004 0.0335 0.9046 2011 0.0514 0.9081
2005 0.0387 0.8791 2012 0.0536 0.8933
2006 0.0434 0.9040
286 Remote Sensing for Sustainability

Beijing Beijing

0 20 km 0 20 km

N N

0 500 km 0 500 km
38.5589 39.8403
0 million kWh 0 million kWh
No data No data
Chengdu Pearl River Delta Shanghai Chengdu Pearl River Delta Shanghai

(a) 0 10 km 0 50 km 0 50 km
(b) 0 10 km 0 50 km 0 50 km

Beijing Beijing

0 20 km 0 20 km

N N

0 500 km 0 500 km
50.8310 58.4246
0 million kWh 0 million kWh
No data No data
Chengdu Pearl River Delta Shanghai Chengdu Pearl River Delta Shanghai

(c) 0 10 km 0 50 km 0 50 km (d) 0 10 km 0 50 km 0 50 km

FIGURE 15.2 Estimated pixel-based EC in 2000 (a), 2004 (b), 2008 (c), and 2012 (d).

Figure 15.3 presents the comparison between the estimated and reference EC
for 2001, 2005, and 2010. Although relatively strong correlations (R2 = 0.80) were
obtained by using all samples (Figure 15.3a), the slopes of the regressions dem-
onstrated larger urban EC against EC of urban districts. Meanwhile, there were
obvious outliers. These could be partly attributed to the different definition of
urban district and urban area. For example, urban district does not include small
urban areas within counties in China. Nevertheless, the gap between the estimated
and statistical EC and the degree of overestimation were smaller for the capital
cities, with R2 of 0.94 (Figure 15.3b). The result in Figure 15.3 indicated that it was
feasible to allocate China’s grid-level EC using its province-level statistical data
and DMSP-OLS NTL data, or at least for prefecture-level estimation of highly
urbanized regions.
 Use of Nighttime Imaging Data to Assess Decadal Trends 287
The statistical EC of urban district (million kWh)
140,000 140,000

2001 (244 samples) 2001 (25 samples)

120,000 2005 (257 samples) 120,000 2005 (27 samples)
2010 (255 samples) 2010 (26 samples)
100,000 100,000

80,000 80,000
y = 0.9379x
R2 = 0.9472
60,000 y = 0.8018x 60,000
R2 = 0.7711
40,000 40,000

20,000 20,000

0 0
0 20,000 40,000 60,000 80,000 100,000 120,000 140,000 0 20,000 40,000 60,000 80,000 100,000 120,000 140,000
(a) (b)

FIGURE 15.3 Scatterplot of estimated EC of urban area versus reference EC of urban dis-
trict at the prefecture level, using samples across China (a), and samples of capital city of each
province (b) in 2001, 2005, and 2010.

15.5.2 sPatiotemPoral Pattern of ec at multiPle scales

Figure 15.4 shows significantly positive spatial autocorrelations of EC at multiple
scales. However, the spatial autocorrelation showed different trends for different
scales. Specifically, the global Moran’s I of provincial EC remained stable from 2000
to 2012 around 0.18, while the value of urban EC increased steadily from 0.12 in
2000 to 0.27 in 2012 by an annual rate of increase of 0.0141 with an R2 of 0.92, which
means urban EC became more clustered during the past decade. For prefectural-
level EC, the global Moran’s I increased from 0.25 in 2000 to 0.41 in 2007 and
remained around 0.40 until 2012.

0.6
Provincial level
Prefectural level
0.5 Urban EC at prefectural level
y = –0.002x2 + 8.2106x – 8245.7
0.4 R2 = 0.9952
Global Moran’s I

0.3 y = 0.0141x – 28.1

R2 = 0.9243

0.2

y = 0.0002x – 0.1215
0.1
R2 = 0.0038

0
1999 2001 2003 2005 2007 2009 2011 2013

FIGURE 15.4 Global Moran’s I of EC at multiple scales.

288 Remote Sensing for Sustainability

Local Moran’s I indices in 2000 and 2012 were calculated and categorized into
four groups: high–high cluster, low–low cluster, low–high cluster, and high–low clus-
ter (Figure 15.5). For the provincial level in 2000, only Shandong showed a significant
high–high cluster; Xinjiang, Gansu, and Yunnan belonged to the low–low cluster,
all of which were located in western China; Fujian and Anhui were surrounded by
provinces with high EC, while only one province was detected as a high–low cluster
(i.e., Sichuan). However, change obviously occurred in western China as Gansu and
Yunnan no longer belonged to the low–low cluster in 2012 (Figure 15.5b). At the pre-
fectural level in 2000 (Figure 15.5c), three high–high clusters were found, all of which

(a) (b)

High–high N High–high N
Low–low Low–low
Low–high Low–high
High–low 0 500 km High–low 0 500 km

(c) (d)

High–high N High–high N
Low–low Low–low
Low–high Low–high
High–low 0 500 km High–low 0 500 km

(e) (f)

High–high N High–high N
Low–low Low–low
Low–high Low–high
High–low 0 500 km High–low 0 500 km

FIGURE 15.5 Spatial clustering of EC at multiple scales: provincial level in 2000 (a) and
2012 (b), prefectural level in 2000 (c) and 2012 (d), and urban EC at prefectural level in
2000 (e) and 2012 (f).
Use of Nighttime Imaging Data to Assess Decadal Trends 289

appeared in eastern China, specifically in the Beijing–Tianjin–Tangshan region, the

Yangtze River Delta, and the Pearl River Delta; low–low clusters were identified in
western China and in Anhui and Jiangxi provinces, especially the prefectural cities
in Xinjiang, Xizang, and Qinghai; cities such as Chongqing and Nanning were sur-
rounded by cities with low EC; six cities were surrounded by neighbors with high EC,
including four prefectural cities around Chongqing (i.e., Tongren in Guizhou prov-
ince, Shennongjia in Hubei province, and Guangan and Suining in Sichuan province),
Chengde in Hebei province, and Xiamen in Fujian province. Two more high–high
clusters were identified comparing Figure 15.5d and c, including Yantai, Linyi, and
Zaozhuang in Shandong province and Lianyungang in Jiangsu province. Meanwhile,
the size of low–low clusters shrank in western China but increased in central China,
including some prefectural cities in Hubei and Hunan. Additionally, some high–low
clusters appeared in northeastern, western, and central China, including Hulunbeier
in Inner Mongolia, Harbin in Heilongjiang province, Ili Kazakh and Changji in
Xinjiang, and Wuhan in Hubei province. Another significant change was that cit-
ies that belonged to the low–high cluster were all located in eastern China in 2012.
For prefectural-level urban EC, the high–high cluster almost appeared in the same
region but with smaller clusters as that in Figure 15.5c (Figure 15.5e). However, the
high–high cluster in the Beijing–Tianjin–Tangshan region was larger than the other
two clusters, which increased from 2000 to 2012, especially for Yangtze River Delta
(Figure 15.5e and f). High–low clusters can be identified in western China, such as
Chongqing, Chengdu, Lanzhou, Guiyang, and Nanning, all of which are the capital
cities of each province. This indicated that urban EC in capital cities was significantly
higher than that in cities around them in western China. This further implied the
unbalanced development in western China. The low–low spatial clusters were mainly
located in western China, yet the size of the cluster shrank; the low–high clusters were
mainly located in eastern China, around high–high clusters. However, urban EC in
central and northeastern China did not show statistically significant clusters.
Figure 15.6 shows the temporal variations of China’s EC at different scales from
2000 to 2012. The calculated SLOPE index was categorized into five groups by
using Natural Breaks (Jenks) in ArcMap 10.1: slow, relatively slow, moderate, rela-
tively rapid, and rapid. It is revealed in Figure 15.6a and b that provinces that expe-
rienced rapid growth of EC were Shandong, Jiangsu, and Guangdong, all of which
are located in eastern China; Inner Mongolia, Hebei, Henan, and Zhejiang demon-
strated a relatively rapid growth of EC; most of the central provinces showed a mod-
erate growth of EC except for Jiangxi. Particularly, a slow growth of EC was found
at the provincial level for four municipal cities—Beijing, Shanghai, Tianjin, and
Chongqing, mainly because of their smaller amount of EC than some provinces such
as Zhejiang, Jiangsu, and Guangdong and their relatively highly urbanized status at
the beginning of the 21st century (Figure 15.6b). Prefectural cities that experienced
moderate to rapid growth of EC were either located in coastal regions or belonged
to capital cities; the majority of western and northeastern cities and more than half
of central cities detected a slow or relatively slow growth of EC (Figure 15.6c). The
growth of prefectural-level urban EC demonstrated a similar pattern as prefectural-
level EC (Figure 15.6d), highlighting capital cities and three metropolitan regions
(i.e., the Beijing–Tianjin–Tangshan region, the Yangtze River Delta, and the Pearl
290 Remote Sensing for Sustainability

N
Slow
Relatively slow
Moderate
Relatively rapid 0 500 km
Rapid
(a)
4500
Billion kWh

2000 2001 2002 2003 2004 2005 2006

4000
2007 2008 2009 2010 2011 2012
3500
3000
2500
2000
1500
1000
500
0
Jilin
Fujian

Hebei
Jiangsu

Shanghai
Tianjin

Anhui
Henan
Hubei
Hunan
Jiangxi
Shanxi

Gansu
Guangxi
Guizhou

Ningxia
Qinghai
Shaanxi
Sichuan

Yunnan
Inner Mongolia
Beijing

Zhejiang

Xinjiang
Xizang
Guangdong

Shandong

Heilongjiang

Liaoning

Chongqing

Eastern Central
Western
(b) Northeastern

FIGURE 15.6 Temporal growth of EC at different scales: (a) provincial level, (b) provincial
EC. (Continued)

River Delta). This further demonstrated the unbalanced development between four
regions and capital cities and other prefectural cities.

15.5.3 national anD regional contriButions of urBan areas to ec

Figure 15.7 shows a steady increase of urban extent during the past decade. The area
increased from approximately 3.85 × 104 km2 in 2000 to 8.83 × 104 km2 in 2012,
which occupied approximately 0.4% and 0.9% of the total area in Mainland China,
Use of Nighttime Imaging Data to Assess Decadal Trends 291

N
Slow
Relatively slow
Moderate
Relatively rapid 0 500 km
Rapid Capital city of each province
(c)

N
Slow
Relatively slow
Moderate
Relatively rapid 0 500 km
Rapid Capital city of each province
(d)

FIGURE 15.6 (CONTINUED) Temporal growth of EC at different scales: (c) prefectural

level, and (d) urban EC at prefectural level for China from 2000 to 2012.

respectively. Regionally, urban extent in eastern China increased from 1.77 × 104 km2
in 2000 to 3.70 × 104 km2 in 2012, occupying approximately 2.0% and 4.2% of the total
area in this region, respectively; in central China, urban area increased from 8.7 ×
103 km2 to 2.04 × 104 km2 during the period, with an urbanization rate of 0.9% and
2.0%, respectively; the urbanization rate of northeastern China grew from approxi-
mately 0.6% to 1.4% (from 4.65 × 103 km2 to 1.07 × 104 km2, out of 7.92 × 105 km2);
urban extent nearly tripled in the western part from 7.41 × 103 km2 to 2.02 × 104 km2
out of 6.65 × 106 km2, occupying 0.1% and 0.3% of the total western region.
292 Remote Sensing for Sustainability

Beijing–Tianjin–Hebei
metropolitan region

Northeast

West

Central
N
0 500 km East

Urban area
Urban area in 2000
Expansion from 2000 to 2006
Expansion from 2006 to 2012
No data
Sichuan Basin Pearl River Delta

Yangtze River Delta

FIGURE 15.7 Urban expansion from 2000 to 2012.

Nationally, the share of electricity consumed in urban areas had increased con-
tinuously from 324 to 1310 billion kWh from 2010 to 2012 (Figure 15.8a), at an
annual growth rate of 12.35%. The ratio of urban EC to the total EC grew from
23.80% to 26.45%, given its coverage of less than 1% of the total area. For total
EC in urban areas across Mainland China, eastern China annually consumed more
than 55%, but with a slightly decreasing trend of 0.11% per year (Figure 15.8a). The
decreasing trend was also observed for northeastern China, but with a faster decreas-
ing rate of 0.33% per year. The central and western regions experienced a growth of
EC in urban areas during the period. However, EC in urban areas of western China
increased with a faster rate than in the central region (0.32% vs. 0.12%). The results
imply the accelerated urbanization in central and western China during the past
decade and, hence, higher EC of these two regions.
The ratio of electricity consumed in urban areas showed an increasing trend for the
four regions (Figure 15.8b), reflecting the growing urbanization rate across Mainland
China. For electricity consumed in each region, eastern and northeastern China exhibited
 Use of Nighttime Imaging Data to Assess Decadal Trends 293

Central Western Central Western Northeastern Eastern

Northeastern Eastern 35
EC in urban area
70 1600 33
y = –0.1104x + 277.94 y = 0.2686x – 509.81
60 R2 = 0.0773 1400 31 R2 = 0.3602
1200 29
50

% Consumed
y = 2E–96e0.1165x 27 y = 0.0159x – 3.1284
% Consumed

1000 R2 = 0.0033

Billion kWh
40 R2 = 0.9893
25
800
30 23 y = 0.2807x – 539.57
y = 0.1187x – 219.98 600
R2 = 0.3745 21 R2 = 0.4735
20 y = 0.1941x – 371.9
400 19
y = 0.3189x – 623.91 R2 = 0.4529
10 200
y = –0.3272x + 665.95 R2 = 0.78 17
R² = 0.809
0 0 15
2000 2002 2004 2006 2008 2010 2012 2000 2002 2004 2006 2008 2010 2012
(a) (b)

FIGURE 15.8 National and regional contribution of urban areas to EC in Mainland China
from 2000 to 2012: (a) EC in Chinese urban areas and the percentage consumed in each
region and (b) the percentage of EC in urban areas for each region.

the highest ratio of EC in urban areas, followed by central and western China. The ratio
remained essentially constant for eastern China, but the increasing rate of urban EC for
central, northeastern, and western China was 0.28%, 0.27%, and 0.19%, respectively.

15.6 DISCUSSION AND CONCLUSIONS

This chapter analyzed the decadal spatiotemporal pattern of EC across Mainland
China at the provincial, prefectural, and urban scales by using time-series DMSP-
OLS NTL images and auxiliary data sets. The results showed different spatio-
temporal patterns of EC, both nationally and regionally. Specifically, the results
demonstrated a moderate to rapid growth of EC for coastal regions at all scales,
while a slow growth for the majority of western and northeastern cities and part of
central China (except for the capital cities). The results of global Moran’s I showed
significantly positive spatial autocorrelations of EC with an increasing trend during
the past decade, especially for EC at the prefectural level and urban EC. Moreover,
the results of local Moran’s I showed the phenomenon of high–high clusters in
coastal regions, but high–low and low–low clusters in western China.
The disparity of economic development was a main factor for regional differences
in EC, especially for the huge gap between western and eastern China. However, the
results implied that the gaps between the western, central, and eastern regions were
narrowing during the past decade. These changes may be attributed to the develop-
ment policies applied to central and western China. For instance, in 2004, the Rise
of the Central China Plan started to coordinate regional growth in central China (Tan
2015). To create regional economic development centers in western China, the Western
Development Strategy was launched in 2000, calling for aid in infrastructure construc-
tion, natural resource exploitation, and the establishment of market liberalization poli-
cies from the central government (Tan 2015). Urbanization, including the intensification
294 Remote Sensing for Sustainability

of urban cores and urban expansion, was one of the major factors responsible for the
increase of EC in China. In western and northeastern China, the growth of energy-
related industries and heavy industries was the main reason for urbanization, while
the expansion of light and technology- and labor-intensive industries was the major
contributing factor for urbanization in eastern and central China (Su et al. 2014). For
the sustainable development of energy, main efforts in Chinese cities should be focused
on the optimization of industrial structures and the improvement of efficiency in using
electricity, due to the fact that electricity production is coal dominated but with inef-
ficient generating units in China (Cherni and Kentish 2007). Meanwhile, it is also
essential to establish a strong legal system and effective institutions to protect electric-
ity market competition and promote appropriate incentives for electricity efficiency
(Ngan 2010). Moreover, to cope with the fast-growing demand of electricity in China,
renewable energy such as solar and wind resources should be exploited, especially for
western and northeastern China with abundant supply of such energy sources.
This chapter offers several potential topics for further research. For example, to
improve the accuracy of pixel-based EC estimation, exploring alternative methods to
better eliminate the saturation and blooming effects in time-series DMSP/OLS NTL
data is preferable. Although the application of EVI increased the variation of EC in
urban regions in this study, the assumption that vegetation cover and urban surface were
inversely correlated may weaken the contribution of urban areas to total EC given the
higher DNs in urban areas than other regions. Except for developing methods to improve
the quality of DMSP-OLS NTL data, incorporating other factors (e.g., GDP and topo-
graphic variables) into the model for NTL-based EC estimation may help better predict
EC at multiple scales (Xie and Weng 2016). Additionally, a better method to extract
urban dynamics, which reduces the overestimation of large urban areas and the under-
estimation of small ones, is required to better study the contribution of urban areas to
total EC. To this extent, a cluster-based method proposed by Zhou et al. (2014), which
estimated the optimal threshold for each cluster according to its size and the mean mag-
nitude of NTL, can be more promising (Zhou et al. 2015). However, the incomparability
of DMSP-OLS NTL DNs from different years and satellites may limit the usage of a cal-
ibrated model to other years and regions. Further studies on extending the cluster-based
thresholding method to the temporal domain may provide some inspiring discoveries.

REFERENCES
Adams, F. G., and Y. Shachmurove. 2008. Modeling and forecasting energy consumption in
China: Implications for Chinese energy demand and imports in 2020. Energy Economics
30 (3):1263–1278. doi: 10.1016/j.eneco.2007.02.010.
Ahmad, S., G. Baiocchi, and F. Creutzig. 2015. CO2 emissions from direct energy use of urban
households in India. Environmental Science and Technology 49 (19):11312–11320. doi:
10.1021/es505814g.
Al-mulali, U., C. N. B. C. Sab, and H. G. Fereidouni. 2012. Exploring the bi-directional long
run relationship between urbanization, energy consumption, and carbon dioxide emis-
sion. Energy 46 (1):156–167.
Amaral, S., G. Câmara, A. M. V. Monteiro, J. A. Quintanilha, and C. D. Elvidge. 2005.
Estimating population and energy consumption in Brazilian Amazonia using DMSP
night-time satellite data. Computers, Environment and Urban Systems 29 (2):179–195.
Use of Nighttime Imaging Data to Assess Decadal Trends 295

Anselin, L. 1995. Local indicators of spatial association—LISA. Geographical Analysis 27

(2):93–115.
Anselin, L. 1996. The Moran scatterplot as an ESDA tool to assess local instability in spatial
association. Spatial Analytical Perspectives on GIS 111:111–125.
Bilgen, S. 2014. Structure and environmental impact of global energy consumption. Renewable
and Sustainable Energy Reviews 38:890–902. doi: 10.1016/j.rser.2014.07.004.
Bloch, H., S. Rafiq, and R. Salim. 2015. Economic growth with coal, oil and renewable
energy consumption in China: Prospects for fuel substitution. Economic Modelling 44:
104–115.
Cao, X., J. Chen, H. Imura, and O. Higashi. 2009. A SVM-based method to extract urban areas
from DMSP-OLS and SPOT VGT data. Remote Sensing of Environment 113 (10):2205–
2209. doi: 10.1016/j.rse.2009.06.001.
Cao, X., J. Wang, J. Chen, and F. Shi. 2014. Spatialization of electricity consumption of China
using saturation-corrected DMSP-OLS data. International Journal of Applied Earth
Observation and Geoinformation 28:193–200. doi: 10.1016/j.jag.2013.12.004.
Chand, T. R. Kiran, K. V. S. Badarinath, C. D. Elvidge, and B. T. Tuttle. 2009. Spatial charac-
terization of electrical power consumption patterns over India using temporal DMSP-
OLS night-time satellite data. International Journal of Remote Sensing 30 (3):647–661.
doi: 10.1080/01431160802345685.
Cherni, J. A., and J. Kentish. 2007. Renewable energy policy and electricity market reforms in
China. Energy Policy 35 (7):3616–3629.
Doll, C. N. H., and S. Pachauri. 2010. Estimating rural populations without access to electric-
ity in developing countries through night-time light satellite imagery. Energy Policy 38
(10):5661–5670.
Elvidge, C. D., K. E. Baugh, E. A. Kihn, H. W. Kroehl, E. R. Davis, and C. W. Davis. 1997.
Relation between satellite observed visible-near infrared emissions, population, eco-
nomic activity and electric power consumption. International Journal of Remote
Sensing 18 (6):1373–1379.
Elvidge, C. D., M. L. Imhoff, K. E. Baugh, V. R. Hobson, I. Nelson, J. Safran, J. B. Dietz,
and B. T. Tuttle. 2001. Night-time lights of the world: 1994–1995. ISPRS Journal of
Photogrammetry and Remote Sensing 56 (2):81–99.
Elvidge, C. D., D. Ziskin, K. E. Baugh, B. T. Tuttle, T. Ghosh, D. W. Pack, E. H. Erwin, and M.
Zhizhin. 2009. A fifteen year record of global natural gas flaring derived from satellite
data. Energies 2 (3):595–622.
He, C., Q. Ma, T. Li, Y. Yang, and Z. Liu. 2012. Spatiotemporal dynamics of electric power
consumption in Chinese Mainland from 1995 to 2008 modeled using DMSP/OLS stable
nighttime lights data. Journal of Geographical Sciences 22 (1):125–136.
He, C., Q. Ma, Z. Liu, and Q. Zhang. 2014. Modeling the spatiotemporal dynamics of electric
power consumption in Mainland China using saturation-corrected DMSP/OLS night-
time stable light data. International Journal of Digital Earth 7 (12):1–22.
Huang, Q., C. He, B. Gao, Y. Yang, Z. Liu, Y. Zhao, and Y. Dou. 2015. Detecting the 20 year
city-size dynamics in China with a rank clock approach and DMSP/OLS nighttime data.
Landscape and Urban Planning 137:138–148.
Letu, H., M. Hara, H. Yagi, K. Naoki, G. Tana, F. Nishio, and O. Shuhei. 2010. Estimating
energy consumption from night-time DMPS/OLS imagery after correcting for satura-
tion effects. International Journal of Remote Sensing 31 (16):4443–4458.
Liu, Z., C. He, Q. Zhang, Q. Huang, and Y. Yang. 2012. Extracting the dynamics of urban expan-
sion in China using DMSP-OLS nighttime light data from 1992 to 2008. Landscape and
Urban Planning 106 (1):62–72.
Lo, C. P. 2002. Urban indicators of China from radiance-calibrated digital DMSP-OLS
nighttime images. Annals of the Association of American Geographers 92 (2):
225–240.
296 Remote Sensing for Sustainability

Meng, L., W. Graus, E. Worrell, and B. Huang. 2014. Estimating CO2 (carbon dioxide) emis-
sions at urban scales by DMSP/OLS (Defense Meteorological Satellite Program’s
Operational Linescan System) nighttime light imagery: Methodological challenges and
a case study for China. Energy 71:468–478. doi: 10.1016/j.energy.2014.04.103.
Ngan, H. W. 2010. Electricity regulation and electricity market reforms in China. Energy
Policy 38 (5):2142–2148.
Pandey, B., P. K. Joshi, and K. C. Seto. 2013. Monitoring urbanization dynamics in India using
DMSP/OLS night time lights and SPOT-VGT data. International Journal of Applied
Earth Observation and Geoinformation 23:49–61.
Parshall, L., K. Gurney, S. A. Hammer, D. Mendoza, Y. Zhou, and S. Geethakumar. 2009.
Modeling energy consumption and CO2 emissions at the urban scale: Methodological
challenges and insights from the United States. Energy Policy 38 (9):4765–4782.
Shi, K., B. Yu, Y. Huang, Y. Hu, B. Yin, Z. Chen, L. Chen, and J. Wu. 2014. Evaluating the
ability of NPP-VIIRS nighttime light data to estimate the gross domestic product and the
electric power consumption of China at multiple scales: A comparison with DMSP-OLS
data. Remote Sensing 6 (2):1705–1724. doi: 10.3390/rs6021705.
Su, Y., X. Chen, Y. Li, J. Liao, Y. Ye, H. Zhang, N. Huang, and Y. Kuang. 2014. China’s 19-year
city-level carbon emissions of energy consumptions, driving forces and regionalized
mitigation guidelines. Renewable and Sustainable Energy Reviews 35:231–243.
Tan, M. 2015. Urban growth and rural transition in China based on DMSP/OLS nighttime light
data. Sustainability 7 (7):8768–8781. doi: 10.3390/su7078768.
Wang, Q. 2014. Effects of urbanisation on energy consumption in China. Energy Policy
65:332–339.
Welch, R. 1980. Monitoring urban population and energy utilization patterns from satellite
data. Remote Sensing of Environment 9 (1):1–9.
Weng, Q. 2013. What is spacial about global urban remote sensing? In Global Urban
Monitoring and Assessment through Earth Observation, edited by Qihao Weng, 1–12.
Boca Raton, FL: CRC Press/Taylor & Francis.
Wu, J., S. He, J. Peng, W. Li, and X. Zhong. 2013. Intercalibration of DMSP-OLS night-time
light data by the invariant region method. International Journal of Remote Sensing 34
(20):7356–7368. doi: 10.1080/01431161.2013.820365.
Wu, J., Y. Niu, J. Peng, Z. Wang, and X. Huang. 2014. Research on energy consumption
dynamic among prefecture-level cities in China based on DMSP/OLS nighttime light.
Geographical Research 33 (4):625–634 (In Chinese).
Xie, Y., and Q. Weng. 2016. World energy consumption pattern as revealed by DMSP-
OLS nighttime light imagery. GIScience & Remote Sensing 53 (2):265–282. doi:
10.1080/15481603.2015.1124488.
Zhang, Q., C. Schaaf, and K. C. Seto. 2013. The vegetation adjusted NTL urban index: A new
approach to reduce saturation and increase variation in nighttime luminosity. Remote
Sensing of Environment 129:32–41.
Zhang, Q., and K. C. Seto. 2011. Mapping urbanization dynamics at regional and global scales
using multi-temporal DMSP/OLS nighttime light data. Remote Sensing of Environment
115 (9):2320–2329.
Zhao, N., T. Ghosh, and E. L. Samson. 2012. Mapping spatio-temporal changes of Chinese
electric power consumption using night-time imagery. International Journal of Remote
Sensing 33 (20):6304–6320.
Zhou, Y., S. J. Smith, K. Zhao, M. Imhoff, A. Thomson, B. Bond-Lamberty, G. R. Asrar, X.
Zhang, C. He, and C. D. Elvidge. 2015. A global map of urban extent from nightlights.
Environmental Research Letters 10 (5):054011.
Zhou, Y., S. J. Smith, C. D. Elvidge, K. Zhao, A. Thomson, and M. Imhoff. 2014. A cluster-
based method to map urban area from DMSP/OLS nightlights. Remote Sensing of
Environment 147:173–185. doi: 10.1016/j.rse.2014.03.004.
 16 Support of Wind
Resource Modeling
Using Earth Observation
A European Perspective
on the Status and
Future Options
Thomas Esch, Charlotte Bay Hasager,
Paul Elsner, Janik Deutscher, Manuela Hirschmugl,
Annekatrin Metz, and Achim Roth

CONTENTS
16.1 Background ................................................................................................... 297
16.2 Extraction of Land Surface Parameters in Support of Wind Resource
Modeling ....................................................................................................... 299
16.2.1 Land Cover and Land Use-Related Geoinformation ........................ 299
16.2.2 Topography, Surface Morphology, and Surface Roughness ............. 303
16.2.3 Multitemporal Aspects Describing Process Dynamics ....................304
16.3 Perspectives for Improved EO-Derived Parameters for Wind Resource
Modeling .......................................................................................................304
16.3.1 2D Land Cover Mapping .................................................................. 305
16.3.2 3D Topography Characterization......................................................308
16.3.3 4D Surface Dynamics Mapping ....................................................... 310
16.4 Conclusion .................................................................................................... 311
References .............................................................................................................. 312

16.1 BACKGROUND
The fact that the power production of wind turbines is sensitive to the mean wind
speed approaching the turbine means that small errors in wind speed modeling will
translate to much larger uncertainties in the power output of wind energy installations.
The mean wind speed is in turn highly variable and sensitive to properties of the
underlying land surface. Wind resource modeling is performed by using dedicated

297
298 Remote Sensing for Sustainability

aerodynamic microscale models for limited areas as well as numerical weather pre-
diction (NWP) for the meso- to continental scale. In general, two types of land sur-
face description inputs are needed for this:

1. A digital elevation model (DEM) that describes the variation in terrain

elevation
2. A set of surface parameters affecting the wind flow

For (1), the Shuttle Radar Topography Mission (SRTM) DEM is one of the most com-
monly used descriptions in both microscale and mesoscale models. However, the
accuracy of SRTM data is too limited for built-up areas and areas with tall vegeta-
tion, where the building or forest height, rather than the terrain, is captured (e.g., Sun
et al. 2008).
Regarding (2), the traditionally most central parameter for wind resource model-
ing is the land surface roughness (z0), which determines the effect of the friction of
the surface on the wind profile. Under ideal conditions, z0 can be determined from
wind measurements (e.g., Dellwik and Jensen 2005; Mölder and Lindroth 1999).
Choudbury and Monteith (1988) as well as Raupach (1994, 1995) related z0 and the
displacement height parameter, which takes the height of vegetation into account,
to physical properties such as height and density of the underlying vegetation. On
the basis of such relationships, Tian et al. (2011) demonstrated that remote sensing
techniques can be used to calculate z0 for large areas. Some studies have also dem-
onstrated the potential of Synthetic Aperture Radar (SAR) for mapping of z0 (Bidaut
et al. 2006). SAR sensors deliver a backscattering coefficient that is strongly affected
by the composition of the ground cell and by its structure. However, these remote
sensing–based approaches are rarely used in the wind resource community.
In current operational models, surface roughness and other surface parameters
are often assigned for a certain land cover class via best-practice, well-established
tabular values (Troen and Petersen 1989). Such classification is commonly based on
either the Global Land Cover 2000 (GLC2000) database of the Joint Research Centre
(JRC) (http://forobs.jrc.ec.europa.eu/products/glc2000/products), the fine-resolution
European CORINE Land Cover (CLC) map (Hasager et al. 2003), the MODIS land
cover product (Friedl et al. 2002), or the ESA GlobCover (Arino et al. 2008).
A simplified microscale approach is used, for example, in the commercial WAsP
(Wind Atlas Analysis and Application Program) model, which is the wind energy
industry standard PC software for bankable wind resource assessment and siting of
wind turbines and wind farms. There are currently more than 4600 users in over
110 countries that use WAsP for all steps from analysis of wind and terrain effects to
estimation of wind farm production. In WAsP, the roughness layer is usually not suf-
ficiently accurate from CLC and look-up table alone and needs to be complemented
by costly and time-consuming site visits.
Recent developments in microscale Reynolds’ Averaged Navier Stokes (RANS)
equation models as well as Large Eddy Simulation (LES) models allow for a direct
simulation of the effects of vegetation as well as solid objects on the flow (Lopes
et al. 2013; Sogachev and Panferov 2006), thereby short-cutting the need for land
Support of Wind Resource Modeling Using Earth Observation 299

surface classification and uncertainties regarding the assignment of parameter val-

ues. For this advanced type of model, the dimensions of solid obstacles, as well
as the height and density of tall vegetation, are necessary input parameters since
they determine the development of the wind flow over smaller scale variations in
the landscape at around 5 m resolution (Dellwik et al. 2013). The applicability of
these models is severely limited by the lack of reliable land surface input data, and
published studies are often limited to highly idealized cases (Dupont and Brunet
2008).
In NWP models used in wind resource applications, an accurate description of
the elevation and surface roughness is important (Jiménez and Dudhia 2012; Santos-
Alamillos et al. 2013) but not as critical as in aerodynamic models. To a certain
extent, errors in the output from NWP models that are attributed to errors in sur-
face properties can be corrected at the local scale by coupling them to aerodynamic
models (Badger et al. 2014). However, an accurate depiction of the state of the land
surface (i.e., its temperature and soil moisture content) is critical for an accurate
simulation of the state of the planetary boundary layer (PBL) (Findel and Eltahir
2003; Hong et al. 2009; Refslund et al. 2013). It has also been established that an
accurate PBL is critical for an accurate depiction of mesoscale circulations (Taylor
et al. 2007) and the wind profile (Kelly and Gryning 2010), and ultimately these are
important for an accurate wind resource assessment. On the other hand, the truthful
modeling of the state of the land surface is highly dependent on the characteristic of
the land surface itself and their variations in time, as well as the atmospheric forcing
(Case et al. 2008, 2014; Chen et al. 2007; Hong et al. 2009).

16.2 EXTRACTION OF LAND SURFACE PARAMETERS

IN SUPPORT OF WIND RESOURCE MODELING
Atmospheric models are employed for a wide range of operational services, includ-
ing wind resource assessment. Despite the heterogeneity of such applications, all
share very similar parameterization requirements. Standard inputs are information
about surface roughness, land cover type, leaf area index (LAI), vegetation fraction,
and phenological state. From a remote sensing point of view, these basic information
requirements are split into the following methodological focuses:

1. Land cover– and land use–related geoinformation (2D)

2. Topography, surface morphology, and surface roughness (3D)
3. Multitemporal aspects describing process dynamics (4D)

16.2.1 Land Cover and Land Use-reLated Geoinformation

Land has considerable heterogeneity because of the existence of different land cover
types such as bare soil, water, urban land, trees, and snow, which vary over small
areas. This surface variability not only determines the microclimate but also affects
mesoscale atmospheric circulation (Weaver and Avissar 2001; Yang 2004). Accurate
representation of the land surface is therefore important to precisely model the effect
300 Remote Sensing for Sustainability

of the land surface. Remotely sensed derived global land cover products, such as
Global Land Cover Characteristics (GLCC; Loveland et al. 2000), University of
Maryland land cover classification (UMD; Hansen and Reed 2000), and GLC2000
(Bartholomé and Belward 2005), were implemented into various land surface
schemes and climate models. In the new Global Wind Atlas released in 2015 (http://
www.irena.org and http://globalwindatlas.com/), the GlobCover and MODIS land
cover data are merged for improved characterization of the land surface roughness
as shown in Figure 16.1.
Another example of using two different land cover maps is based on USGS and
CLC land use data for a region in Denmark (see Figure 16.2a and b). The WRF
model has been used to calculate the wind speed at different heights (10, 40, 60, 80,
100, and 160 m) for the period January 10, 2014 until September 12, 2014. The out-
put frequency is every 10 min. The WRF model is run as in Floors et al. (2015) but

(a)

500 km i
GlobCover classes
210 220 200 140 150 11 14 130 180 30 20 160 120 170 190 40 50 60 70 90 100 110 230
MODIS classes
0 15 16 10 None 6 7 12 11 14 9 None 13 2 4 1 3 5 8 None

(b)

500 km
Roughness from landuse type
Water 0.0 Snow 0.0004 Bare 0.005 Herb veg. 0.03 Sparse 0.05 Crops 0.1 Woody veg. 0.2 Mosaic 0.3 Shrubs 0.5 BI shrubs 0.6 Urban 1.0 Forest 1.5

FIGURE 16.1 The Global Wind Atlas land cover classes in GlobCover and MODIS with
corresponding color coding (a) and derived roughness map for the Mediterranean area (b).
(From the Global Wind Atlas, supported by EUDP with explanation from Dr. Neil Davis.
http://globalwindatlas.com/. With permission.)
Support of Wind Resource Modeling Using Earth Observation 301

WRF, Domain 3, ∆x = 1.5 km WRF, Domain 3, ∆x = 1.5 km

6°E 7°E 8°E 9°E 10°E 6°E 7°E 8°E 9°E 10°E

57°30´N 57°30´N 57°30´N

57°N 57°N 57°N

56°30´N 56°30´N 56°30´N

56°N 56°N 56°N

55°30´N 55°30´N 55°30´N

55°N 55°N 55°N

6°E 7°E 8°E 9°E 10°E 6°E 7°E 8°E 9°E 10°E
(a) (b)
160

120
WRM setup
Height

CORINE
80 USGS

40

1.50 1.75 2.00 2.25 2.50

RMSE
(c)

FIGURE 16.2 USGS (a) and CLC (b) land use data for a region in Denmark. The RMSE
between the measured and modeled wind speed at the Høvsøre meteorological mast and WRF
model using USGS and CLC land cover as input. The results are for the period January 10,
2014 until September 12, 2014 (c). (Courtesy of RUNE Project, supported by ForskEL with
explanation from Dr. Rogier Floors.)

with ERA interim boundary conditions used and the horizontal resolution at 12, 4,
and 1.333 km for domains 1, 2, and 3, respectively. The land use categories from the
CLC data were reassigned to USGS categories using the method described in Pineda
et al. (2004). The results on root mean square error (RMSE) between the measured
and modeled wind speed at the tall meteorological tower at Høvsøre located in a flat
coastal landscape near the North Sea (Peña et al. 2015) are shown in Figure 16.2c.
At around 90 m above ground and upward, the results are comparable, while at lower
levels, the higher-resolution CLC map yields the best results.
More detailed atmospheric/wind resource models could benefit from more
detailed land cover information. A list of established land cover maps and products
that would be available for Europe, for instance, is provided in Table 16.1.
Within Europe (EEA, 39 member countries), the CLC project is producing land
use cartography in the form of detailed descriptions of land occupation and fea-
tures, at an original scale of 1:100,000. It used 44 classes of the three-level CORINE
302 Remote Sensing for Sustainability

TABLE 16.1
List of Land Surface Parameter Maps Available for Europe
Resolution or
Name Product Type MMU/Coverage Provider
Corine CLC Land cover with more than 30 Minimum Mapping Unit EEA
types covering most of (MMU) 25 ha (vector),
Europe EEA’s 39 countries
GIO Forest Tree cover density (0%–100%) 20 m, EEA’s 39 countries EEA
Forest type (Coniferous/
deciduous/mixed)
GIO Imperviousness Degree of impervious-ness 20 m, EEA’s 39 countries EEA
(0%–100%)
GIO Grassland Presence of permanent 20 m, EEA’s 39 countries EEA
grassland (binary layer)
GIO Wetland Wetland presence 20 m, EEA’s 39 countries EEA
(binary layer)
GIO Water Water presence (binary layer) 20 m, EEA’s 39 countries EEA
JRC Forest/Non-forest Classes: Forest, non-forest, 25 m, EU27 plus AL, BA, JRC
map clouds, no data CH, HR, MN, MK, MR,
LI, RS
GlobCover 22 land cover classes 300 m, global ESA
GlobCover (bimonthly Average surface reflectance 300 m, global ESA
surface reflectance) values over a 2-month period
Urban Atlas for Europe Land use map MMU 0.25 ha for urban EEA
classes, 1 ha for rural
classes (vector), 305
cities throughout Europe
Geoland2 BioPar LAI, FCover, FAPAR Variable Geoland
(Biophysical) project
GLCC Land cover 24 classes ~1 km, global USGS
GLC2000 Land cover 22 classes ~1 km, global JRC
EU-Hydro Water bodies MMU: 1 ha (vector), EU EC
MODIS MCD12Q1 17 land cover classes 500 m, global USGS
MODIS MCD15A2 LAI and FPAR every 8 days 1 km, global USGS

nomenclature, for all European countries, with a definition of 25 ha. However, for
the use of surface occupation cartography like the CLC in wind flow models, it is
necessary to assign an equivalent characteristic roughness length to each different
classification or land cover class. The CLC cartography has been adapted for its
use in typical wind resource assessment studies, namely, for use with the WAsP
model software. According to the particular terrain characteristics, the 44 different
CLC classes were grouped into 14 roughness classes. The grouping was not always
obvious. One has to take into account that the main goal of the CLC project is to
classify the landscape and land cover to facilitate policymaking and environmental
Support of Wind Resource Modeling Using Earth Observation 303

management at a European scale, and not the particular roughness of the surface
with respect to its effect on atmospheric flow.
The application of CLC for dynamic roughness length simulation shows some
important shortcomings. For example, CLC lacks some very important parameters:
(a) the CLC class forest misses a reliable roughness parameter and an accurate forest
borderline location and type description (hard cut or smooth transition); (b) the CLC
class agriculture, specifically the permanent agricultural systems such as vineyards,
olive groves, and fruit plantations, are very diverse and need a subdivision plus infor-
mation on the direction of the planting rows; (c) the CLC class semi-urban is a very
heterogeneous class in terms of surface roughness and lacks details such as build-
ing density and building height. These examples illustrate that the CLC data sets or
low-resolution raster data sets have limited suitability for atmospheric/wind resource
model parameterization. Inaccuracies and errors that exist at this level will invari-
ably flow into the atmospheric/wind resource models and limit their performance.

16.2.2 topoGraphy, sUrfaCe morphoLoGy, and sUrfaCe roUGhness

Relief information is freely available on a global scale but with relatively low resolu-
tion (3 to 1 arc sec, ~90 to 30 m). Table 16.2 provides the established products, their
coverage, and sources. Most of these products, in particular the ones derived from
spaceborne missions, refer to the height of the surface plus the coverage (top of the
canopy). One can distinguish between a digital surface model (DSM), providing the
top of a surface including the canopy, and the DEM, which refers to the land surface,
typically with a height above sea level.
Surface morphology includes the detection of linear strips of woody vegetation
such as hedgerows. It has been in the focus of several studies as they are impor-
tant elements for landscape ecology as well as biodiversity. Besides their ecological
functions as habitat, shelter, or corridor for specific species, these linear landscape
elements serve as protection against soil erosion and as a biogeochemical barrier
(Ducrot et al. 2012). Thus far, methods based on high-resolution and very high–
resolution optical imagery have been developed to detect linear landscape fea-
tures such as hedgerows (Aksoy et al. 2008, 2010; Ducrot et al. 2012; Vannier and

TABLE 16.2
List of Available Digital Elevation Data Available for Europe
Product Type Spatial Resolution Provider
GTOPO30 Elevation 30 arc sec USGS
~1 km
SRTM C-band Elevation Near-global (60°N to 56° S) ~1 arc sec, with NASA/NGA
the exception of Central Asia ~3 arc sec
SRTM X-Band Elevation Near-global (60°N to 56° S) ~1 arc sec, DLR
coverage not continuous: gaps in image strips
ASTER GDEM Elevation ~30 m NASA and METI
EU-DEM Elevation 30 m EC
304 Remote Sensing for Sustainability

Hubert-Moy 2008). To date, only Ducrot et al. (2012) have compared the results based
on multitemporal optical data with multitemporal TSX data, achieving more than
80% overall accuracies for the optical analysis but only 75% for the SAR-based analy-
sis. A detailed review on techniques for extracting linear features from Earth obser-
vation (EO) imagery is given by Quackenbush (2004). The landscape roughness is
closely related to morphology.

16.2.3 mULtitemporaL aspeCts desCribinG proCess dynamiCs

Remotely sensed data collected by instruments such as AVHRR, Medium Resolution
Imaging Spectrometer (MERIS), SPOT-Vegetation, and MODIS provide near-daily
global observations of vegetation dynamics at 300 m to 8 km spatial resolution.
Exploiting these dense time series, several remote sensing algorithms have been
developed over the past decades. Most algorithms use as input time series of veg-
etation indices such as the normalized difference vegetation index (NDVI) or the
enhanced vegetation index (EVI) that are derived from multispectral satellite data
(e.g., Huete et al. 2002).
Multitemporal data can be used not only to investigate, for example, the pheno-
logical stage and changes of the land cover and surface but also for the improvement
of existing land cover maps. On the basis of the filtered/noise-reduced time series,
different algorithms have been applied to identify key phenological phases of veg-
etation. Algorithms based on user-defined thresholds depending on land cover types
were analyzed, for example, by Schwartz et al. (2002) and by White and Nemani
(2006). Algorithms based on significant and rapid increases in remotely sensed sig-
nals were developed, for example, by Moulin et al. (1997). Further algorithms deter-
mine phenological phases of vegetation by fitting of functions to the remotely sensed
time series data (Beck et al. 2006; Zhang et al. 2003). The operationally generated
MODIS Land Cover Dynamics product belongs to this group of algorithm, which
identifies phenophase transition dates based on logistic functions fit to time series
of the EVI (Huete et al. 2002). Taking into account dynamic processes can clearly
improve the quality of the input data for wind modeling and thus also influence the
final results.

16.3 PERSPECTIVES FOR IMPROVED EO-DERIVED

PARAMETERS FOR WIND RESOURCE MODELING
The EO data sets currently in operational use by the wind resource modeling com-
munity in Europe do not reflect recent advances in EO. New and near-future mis-
sions and data sets offer enormous potential for improved data quality, resolution,
and frequencies. Most notably, these are the ESA/Copernicus Sentinel missions and
the TerraSAR-X and TanDEM-X missions for the German Aerospace Center (DLR).
In this section, we will outline beyond state-of-the-art improvements and break-
throughs that can be expected from these missions in the context of atmospheric/
wind resource modeling parameters. The new sensors and missions provide comple-
mentary capabilities for beyond state-of-the-art improvement on the derivation of
land surface parameters for atmospheric/wind resource modeling.
Support of Wind Resource Modeling Using Earth Observation 305

16.3.1 2d Land Cover mappinG

The land cover detail as represented in CLC or in low-resolution raster-based land
cover maps is a generalization of the actual land cover, aggregating multiple land
cover types into one object or pixels. Within these objects or pixels, a large hetero-
geneity in roughness length that is not reflected in the aggregated object is present.
Significant improvements are already achievable by integrating existing data sets
that have been developed very recently.
This includes the high-resolution land cover layers (HRL) data that provide
additional details within the CLC objects. In this way, large agricultural areas
(as represented in CLC) now include pixels representing small villages (HRL
Imperviousness), small forest patches (HRL Forest), water and wetland areas (HRL
Water and HRL Wetlands), and grassland areas (HRL Grasslands), all at a 20-m
resolution (Figure 16.3). The heterogeneity within forests (species and density) can
also be analyzed based on the HRL forest layer. The EEA Urban Atlas data can be
integrated to create a more detailed view of the most densely populated areas. The
Urban Atlas represents the land use/land cover within the functional urban areas
of Europe with a mapping detail of 0.25 ha for the artificial areas and 1 ha for non-
artificial areas. Compared to the low detail of the CLC data, especially in urban
areas (which are very heterogeneous areas), the UA data offer an enormous upgrade
in mapping detail (Figure 16.4).
The development of the European Sentinel satellite program offers substantial
opportunities for further innovation. The objective of the Sentinel program is to
meet the data requirements of the joint ESA/European Commission (EC) initiative
Copernicus (previously known as Global Monitoring for Environment and Security)
on an operational level. However, because of its advanced concepts and mission
design, Sentinel has also significant potential to fulfill the data needs of the wind
resource modeling community. The Sentinel-1, Sentinel-2, and Sentinel-3 missions

Continuous urban fabric Discontinuous urban fabric Broadleafed forest Coniferous forest
Industrial or commerical units Water Imperviousness Water
Broadleafed forest Coniferous forest Mixed forest

(a) (b)

FIGURE 16.3 Comparison of CLC product (a) and combined HRL layers: HRL
Imperviousness, HRL Forest, and HRL Water (b).
 306

Continuous urban fabric Fruit trees and berry plantations Continuous urban fabric Port areas
Discontinuous urban fabric Pastures Discontinuous dense urban fabric Airports
Industrial or commerical units Complex cultivation patterns Discontinuous medium density urban fabric Mineral extraction and dump sites
Road and rail networks and associated land Land principally occupied by agriculture Discontinuous low density urban fabric Construction sites
Port areas Broadleafed forest Discontinuous very low density urban fabric Land without current use
Airports Mixed forest Isolated structures Green urban areas
Mineral extraction sites Moors and healthland Industrial, commercial, public, military and private units Sports and leisure facilities
Green urban areas Transitional woodland-shrub Fast transit roads and associated land Agricultural areas
Sport and leisure facilities Water courses Other roads and associated land Forest and semi-natural areas
Non-irrigated arable land Water bodies Railways and associated land Water
Vineyards

(a) (b)

FIGURE 16.4 Comparison of CLC 2006 product (a) and European Urban Atlas layer of 2006 (b) for Strasbourg, France.
Remote Sensing for Sustainability
Support of Wind Resource Modeling Using Earth Observation 307

are of particular interest in the context of wind modeling parameterization, and we

will briefly review each of the missions.
The concept of the Sentinel-1 mission is a constellation of two satellites orbiting
180° apart that carry a C-band SAR. The first of these satellites (Sentinel-1A) has
been launched in April 2014, and the launch for Sentinel-1B is scheduled for 2016.
While the current repeat cycle of Sentinel-1A is 12 days, the revisit time will increase
to 6 days with both satellites in orbit. Sentinel-1 can be operated in several modes.
The Sentinel-2 mission is again a constellation of two identical satellites that
have the key objective of collecting multispectral high-resolution data in the optical
domain of the electromagnetic spectrum. The definition of the requirements of the
Sentinel-2 program builds on the experience of previous multispectral campaigns
such as the Landsat program and, in particular, the SPOT family (Satellite Pour
l’Observation de la Terre) (Drusch et al. 2012; ESA 2014). Sentinel-2A was launched
in June 2015 and Sentinel-2B is scheduled to be put in orbit in the second half of
2016. Their advantages compared to operational multispectral missions will be
(a) higher spectral resolution with 13 bands in the visible, near-infrared, and short-
wave infrared region; (b) high spatial resolution ranging from 10 to 60 m; (c) an
unprecedented temporal resolution with a revisit time of <5 days; and (d) a dedicated
ground segment that aims to rapidly process the acquired data and provide products
to the user community in near-real time.
Sentinel-3 represents Europe’s global land- and ocean-monitoring mission that
will provide 2-day global coverage EO data. Sentinel-3A and 3-B are scheduled for
launch early 2016 and 2017, respectively. One of Sentinel-3 mission’s main objectives
is to measure land surface color with high-end accuracy and reliability in support
of ocean forecasting systems and for environmental and climate monitoring. The
Sentinel-3 Ocean and Land Colour Instrument (OLCI) is based on heritage from
Envisat’s MERIS. With 21 bands, compared to the 15 on MERIS, it has a design
optimized to minimize sun glint and a resolution of 300 m over all surfaces.
Wind resource modeling could benefit from the improved capabilities via better
information on land use change, forest cover, photosynthetic activity, and phenol-
ogy. Of all Sentinel missions, however, Sentinel-2 promises to be particularly use-
ful for wind resource modeling activities, as its data allow the development of land
cover map production systems that should be able to update the information monthly
at a global scale. The temporal dimension will allow distinguishing classes whose
spectral signatures are very similar during long periods of the year. The increased
spatial resolution will enable one to work with smaller minimum mapping units.
Additionally, the spectral bands are increased in number and decreased in width.
With new image bands in the red-edge and specific bands for vegetation analysis and
atmospheric analysis, Sentinel-2 images will show an increase in the separability of
land cover classes, combined with an easier atmospheric correction.
It should therefore become feasible to update the land cover information with new
data from a consistent source to move from mapping to monitoring. This monitor-
ing will focus primarily on the vegetation evolution during a year/season by pro-
viding bio-geophysical parameters such as LAI, NDVI, green vegetation fraction,
leaf chlorophyll content, and leaf cover. These parameters are a direct input for a
more detailed land cover classification, allowing for diversification within otherwise
308 Remote Sensing for Sustainability

aggregated areas such as agricultural fields and forested areas, thereby facilitating
improved roughness estimation, taking into account the time of year. Besides being
an input for land cover mapping, these parameters show a direct interrelation with
atmospheric/wind resource models; for example, the green vegetation fraction is
important for land surface heat flux calculation in coupled land–atmospheric mod-
els. Combining these parameter values with land cover data will provide a better
estimation/understanding on the Earth’s influences on the atmosphere.

16.3.2 3d topoGraphy CharaCterization

One of the challenges to applying DEMs to modeling approaches—in particular on
the microscale level—has been the absence of the availability of a standard DEM at
a sufficient level of spatial detail. Low-resolution DEMs such as the STRM C-Band
(30 m with the exception for Central Asia where the resolution is limited to 90 m)
version or the one derived from stereo-processing of the Advanced Spaceborne
Thermal Emission and Reflection Radiometer (ASTER) data at 30 m spacing are
widely available free of charge, but offer limited use for detailed modeling. The
reason is not only the relatively coarse spatial resolution but also the incomplete
global coverage in case of SRTM or the varying local quality owing to the number
of images per point in case of ASTER (Jacobsen 2013).
With the advent of scientifically and commercially available DEMs at higher
resolutions such as the TerraSAR-X (TSX) and TanDEM-X (TDX) DEM at 12 m
resolution, significant progress can be made in applying this data set to wind-related
modeling. The outstanding characteristics of TDX are the provision of a real global
and homogenous data set and the improved spatial and vertical resolution. The
global coverage not only comes along with high-precision elevation information in
areas above 60° latitude that were not covered by SRTM before. It also eases the
transferability to other sites.
Figure 16.5 demonstrates the relevance of an increased spatial resolution. A
coarse resolution levels out details and smoothens terrain features such as slopes and
ridges. Smaller-sized obstacles can only be detected from high-resolution DEMs.
The corresponding elevation expresses the top of trees and other land surface ele-
ments rather than the ground level. Figure 16.6 shows the increased level of detail
that can be achieved with the TDX DEM. A farmhouse, surrounded by hedges and
a line of trees, is clearly visible, and the corresponding height values can directly
be measured in the DEM. The arrow indicates the location of the farm in the DEM.
For the assessment of land cover and forestry parameters, DSMs generated by
stereo imagery, SAR interferometry, or airborne laser scanning (ALS) can support
atmospheric flow modeling both at the mesoscale and the microscale. EO-based
DSMs provide information on important surface parameters such as edge morphol-
ogy, forest borderline location and type, local obstacle properties, terrain slope,
height of canopy, and canopy density.
At the mesoscale level, the land cover classifications (CLC, HRL layers) can be
improved through the integration of height-derived parameters. This leads to sub-
stantial improvements in non-flat areas such as vineyards, orchards, and shrubby
areas for which the calculation of the wind shear effect is inaccurate when solely
Support of Wind Resource Modeling Using Earth Observation 309

(a) (b)

FIGURE 16.5 Digital elevation data of Mt. Etna in SRTM (90 m) (a) and new TanDEM-X
(12 m) DEM (b).

(a) (b)

FIGURE 16.6 Color-shaded TanDEM-X DEM (a) for Winnipeg area, Canada, and zoom to
farm house surrounded by hedges and trees via Google Earth (b).

based on the roughness length. Mesoscale models such as the widely used Weather
Research and Forecasting (WRF) model can include the option of a canopy layer
separated from the ground, such that the displacement height can be directly used as
input to the model, but suitable values for the model parameters for the most com-
mon forest types need to be provided.
At the microscale level, the modeling community moves toward using flow
models (RANS and LES models) in which the forest canopy is resolved. This cre-
ates a need for supplying the necessary surface parameters to the wind modelers.
Boudreault et al. (2015) developed a method by which raw data from aerial Lidar
scans of forested areas are transformed into gridded forest heights for a CFD model
at a forested site. Furthermore, 3D approaches can be used to assess the attributes of
the forest borders (e.g., hard cut vs. smooth transition from tall trees to non-forest)
and 3D canopy structure/canopy density distribution.
310 Remote Sensing for Sustainability

The development of 3D approaches can be based on different sources according

to availability, for example, optical (multi-)stereo data such as Pleiades, GeoEye and
PRISM, radar from Sentinel-1, TSX/TDX, and airborne Lidar data.
Since the early years of SAR remote sensing, stereo-radargrammetric tech-
niques have been applied to SAR image pairs (Toutin and Gray 2000). It has been
demonstrated in previous studies that X-band radar data from COSMO-SkyMed
and TerraSAR-X missions can provide useful information on forest cover and for-
est parameters (Deutscher et al. 2013). These sensors are able to collect images
with a ground sampling distance down to 0.75 m in Spotlight mode at various look
angles. In addition, they deliver imagery with very precise pointing accuracy so
that remote regions where no reference data, that is, ground control points, are
available can also be mapped and processed. One option to develop a DSM for for-
est assessment is based on the approaches by Raggam et al. (2010) and Perko et al.
(2010) where 3D surface reconstruction was performed by stereo-radargrammetry
in Europe. Perko et al. (2011) transferred the multi-image matching concept for
digital surface modeling based on optical satellite images (Raggam 2006) to radar
data and incorporated the SAR-specific image geometry for forest parameters in
European test sites.

16.3.3 4d sUrfaCe dynamiCs mappinG

The developing Copernicus Sentinel program also offers unprecedented data in
terms of spectral, spatial, and, most significantly, temporal resolution. Revisit times
of a few days will, for the first time, allow it to resolve highly dynamic phenologi-
cal processes. This in turn will make it possible to parameterize atmospheric/wind
resource models with near-real time and spatially explicit data of phenological stage
and LAI at high spatial resolution.
One of Sentinel-3 mission’s main objectives is to measure land surface color with
high-end accuracy and reliability. Data from the Sentinel-3 OLCI could be utilized
and coupled with the sea and land surface temperature (SST/LST) data with daily
revisit times allowing for synergy products (e.g., phenological stage mapping).
Based on time series from future Sentinel-3 imagery and optionally other satel-
lites (MODIS, Proba-V, MERIS), fully automatic, spatially explicit, wall-to-wall
phenological models covering the whole EU could be developed by such an
approach. The result of such phenological models would be maps that include
entry dates of selected key phenological parameters. Separately for each vegetation
type and biogeographic region, the temporal trajectories of vegetation index time
series could then be analyzed on the basis of phenological field observation data.
In addition to different algorithms/models, different features that are derived from
the satellite data could also be analyzed, such as the cumulative vegetation index
values. Following cross-validation, this would identify the best-suited algorithm /
model and derive quality parameters. The best suited algorithms/models and fea-
ture combinations could then be applied over the whole monitoring period at a
wall-to-wall basis.
The high revisit time of the fully operational Sentinel-1 and Sentinel-2 missions,
combined with their large swath, will provide a high probability for a cloud-free
Support of Wind Resource Modeling Using Earth Observation 311

coverage at key dates throughout the vegetation phase. This is of paramount impor-
tance since (i) it is mandatory for an accurate analysis of different vegetation types,
and (ii) it enables the assessment of the intra-annual variability of the various land
cover classes as well as the surface roughness parameters.
Sentinel-1 and Sentinel-2 time series data will enable new approaches for a detailed
land cover classification that will further improve the currently available CLC clas-
sification scheme and can in particular be tailored to the data needs of increasingly
sophisticated wind resource models. This will provide high thematic and spatial
detail compared to the currently available CLC2006 database or the HRL layers.
In particular, it will effectively discriminate between different vegetation types and
enable the estimation of surface roughness and vegetation stand height. Moreover,
the temporal variability of land cover classes and surface roughness can be analyzed
and provided for wind resource modeling.
Exploitation of high-resolution DSMs such as TDX and stereo data from high-
resolution satellites (Pléiades, Cartosat, TerraSAR, COSMO-SkyMed) will offer the
opportunity to quantify the temporal dynamics of 3D surface parameters. This will
allow the assessment of parameters such as forest growth, forest change detection
(loggings, wind fall), changes in forest stock, and forest density. Quantification of,
for example, forest growth can be used for computing the future aerodynamic flow
at a potential wind turbine site.

16.4 CONCLUSION
The previous sections gave an overview on the potential of remote sensing data to
support wind resource modeling especially through improved input parameteriza-
tion regarding the state and characterization of the land surface. Three different
topics in this relation have been discussed: land cover information (2D) from vari-
ous remote sensing sources, DEM information (3D) at high spatial resolution, and
the opportunities of monitoring changes in the state of the surface, which affects its
roughness (4D). For all three topics, the currently used data sets and methods were
briefly described and the related shortcomings were analyzed. Finally, the possibili-
ties provided by new sensors (e.g., Sentinels, TerraSAR-X/TanDEM-X) and initia-
tives, such as Copernicus, are outlined. In conclusion, there is a wide amount of
promising possibilities waiting to be tested for the specific needs in wind resource
modeling. For 2D land cover information, existing high-resolution layers will need to
be merged and integrated, possibly supported by specific classification of Sentinel-2
data. In terms of 3D data, now existing satellite-based DEMs at a resolution of 12 m
can be used showing features such as hedges, forest borders, buildings, and so on in
order to improve the modeling of surface roughness. Finally, the Sentinel-2 data are
expected to bring additional benefit to the 4D component, as frequent update will
allow monitoring the roughness situation over time, giving a more realistic scenario
than just a snapshot assessment.
There is currently a substantial underexploitation of remote sensing data in the
context of wind energy modeling. This chapter demonstrated that significant prog-
ress has been made in the past decade in terms of spaceborne data collection, rep-
resented by a range of new missions that collect global land cover and elevation
312 Remote Sensing for Sustainability

information in unprecedented spatial, spectral, and temporal resolution. This is

complemented by the increased availability of high-resolution ALS data. Yet, most
operational wind resource models that are used in the wind energy sector still have
to rely on data products that represent derivatives of data sets from sensors that have
been put into orbit one or two decades ago.
There is hence an urgent need for a coherent effort that brings together both the
remote sensing and wind modeling communities to develop bespoke downstream
services for the wind energy sector, for example, within the framework of the
European Copernicus program. Such an endeavor would not only constitute a major
step forward in adding value to remote sensing data. It can also be expected that the
new downstream data product will substantially improve wind energy planning and
the subsequent commercial viability of wind energy projects.

REFERENCES
Aksoy, S., G. Akcay, G. Cinbis and T. Wassenaar (2008): Automatic mapping of linear woody
vegetation features in agricultural landscapes. Proc. IGARSS, July 6–11, 2008, Boston,
Massachusetts, pp. IV, 403–406.
Aksoy, S., G. Akcay and T. Wassenaar (2010): Automatic mapping of linear woody veg-
etation features in agricultural landscapes using very high resolution imagery. IEEE
Transactions on Geoscience and Remote Sensing, 48(1), 511–522.
Arino, O., P. Bicheron, F. Achard, J. Latham, R. Witt and J. L. Webere (2008): GlobCover
The most detailed portrait of Earth. ESA Bulletin-European Space Agency, 136, 24–31.
Badger, J., H. Frank, A. N. Hahmann and G. Giebel (2014): Wind climate estimation based
on mesoscale and mircoscale modelling: Statistical–dynamical downscaling for wind
energy applications. Journal of Applied Meteorology and Climatology, 53, 1901–1919.
Bartholomé, E., and A. S. Belward (2005): GLC2000: A new approach to global land cover
mapping from Earth observation data. International Journal of Remote Sensing, 26(9),
1959–1977.
Beck, P., C. Atzberer, K. Høgda, B. Johansen and A. Skidmore (2006): Improved monitoring
of vegetation dynamics at very high latitudes: A new method using MODIS NDVI.
Remote Sensing of Environment, 100, 321–334.
Bidaut, S., T. Ranchin and L. Wald (2006): Mapping the aerodynamic roughness length using
SAR images. ENSMP 300740, Paris, pp. 145–150.
Boudreault, L.-E., A. Bechmann, L. Taryainen, L. Klemedtsson, I. Shendryk and E. Dellwik
(2015): A LiDAR method of canopy structure retrieval for wind modeling of heteroge-
neous forests. Agricultural and Forest Meteorology, 201, 86–97.
Case, J. L., W. L. Crosson, S. V. Kumar, W. M. Lapenta and C. D. Peters-Lidard (2008):
Impacts of high-resolution land surface initialization on regional sensible weather fore-
casts from the WRF model. Journal of Hydrometeorology, 9(6), 1249–1266.
Case, J. L., F. J. LaFontaine, J. R. Bell, G. J. Jedlovec, S. V. Kumar and C. D. Peters-Lidard (2014):
A real-time MODIS vegetation product for land surface and numerical weather prediction
models. IEEE Transactions on Geoscience and Remote Sensing, 52(3), 1772–1786.
Chen, F., K. W. Manning, M. A. LeMone, S. B. Trier, J. G. Alfieri, R. Roberts, M. Tewari,
D. Niyogi, T. W. Horst, S. P. Oncley, J. B. Basara and P. D. Blanken (2007): Description
and evaluation of the characteristics of the NCAR high-resolution land data assimilation
system. Journal of Applied Meteorology and Climatology, 46(6), 694–713. doi:10.1175
/JAM2463.1.
Choudbury, B., and J. Monteith (1988): A four-layer model for the heat budget of homogeneous
land surfaces. Quarterly Journal of the Royal Meteorological Society, 114, 373–398.
Support of Wind Resource Modeling Using Earth Observation 313

Dellwik, E., F. Bingöl and J. Mann (2013): Flow distortion at a dense forest edge. Quarterly
Journal of the Royal Meteorological Society, 140 (679), 676–686; doi:10.1002/qj.2155.
Dellwik, E., and N. Jensen (2005): Flux–profile relationship over a fetch-limited beech forest.
Boundary-Layer Meteorology, 115, 179–204.
Deutscher, J., R. Perko, K. Gutjahr, M. Hirschmugl and M. Schardt (2013): Mapping tropi-
cal rainforest canopy disturbances in 3D by COSMO-SkyMed Spotlight InSAR-stereo
data to detect areas of forest degradation. Remote Sensing, 5, 648–663. doi:10.3390
/rs5020648.
Drusch, M., U. Del Bello, S. Carlier, O. Colin, V. Fernandez, F. Gascon, B. Hoersch, C. Isola,
P. Laberinti, P. Martimort, A. Meygret, F. Spoto, O. Sy, F. Marchese and P. Bargellini
(2012): Sentinel-2: ESA’s optical high-resolution mission for GMES operational ser-
vices. Remote Sensing of Environment, 120, 25–36, ISSN 0034-4257.
Ducrot, D., A. Masse and A. Ncibi (2012): Hedgerow detection in HRS and VHRS images
from different sources (optical, radar). In: Proc. IGARSS, July 22–27, 2012, Munich,
Germany, pp. 6348–6351.
Dupont, S., and Y. Brunet (2008): Edge flow and canopy structure. Boundary-Layer
Meteorology, 126, 51–71.
European Space Agency (ESA) (2014): The Copernicus Space Component: Sentinels Data
Product List. Accessible at https://spacedata.copernicus.eu/documents/12833/14143
/Sentinel_Products_List_Issue1_Rev1_Signed.pdf (July 2016).
Findel, K., and E. A. B. Eltahir (2003): Atmospheric controls on soil moisture—Boundary
layer interactions. Part II: Feedbacks within the Continental United States. Journal of
Hydrometeorology, 4, 570–583.
Floors, R. R., A. Peña and S.-E. Gryning (2015): The effect of baroclinicity on the wind in the
planetary boundary layer. Royal Meteorological Society. Quarterly Journal, 141(687),
619–630. doi:10.1002/qj.2386.
Friedl, M. A., D. K. McIver, J. C. F. Hodges, X. Y. Zhang, D. Muchoney, A. H. Strahler, C. E.
Woodcock, S. Gopal, A. Schneider, A. Cooper, A. Baccini, F. Gao and C. Schaaf (2002):
Global land cover mapping from MODIS: Algorithms and early results. Remote Sensing
of Environment, 83, 287–302.
Hansen, M. C., and B. Reed (2000): A comparison of the IGBP DISCover and University of
Maryland 1 km global land cover products. International Journal of Remote Sensing,
21(6–7), 1365–1373.
Hasager, C. B., N. W. Nielsen, N. O. Jensen, E. Bøgh, J. H. Christensen, E. Dellwik and
H. Søgaard (2003): Effective roughness calculated from satellite-derived land cover
maps and hedge-information used in a weather forecasting model. Boundary-Layer
Meteorology, 109, 227–254.
Hong, S., V. Lakshmi, E. E. Small, F. Chen, M. Tewari and K. W. Manning (2009): Effects of
vegetation and soil moisture on the simulated land surface processes from the coupled
WRF/Noah Model. Journal of Geophysical Research, 114(D18), D18118.
Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao and L. G. Ferreira (2002): Overview of
the radiometric and biophysical performance of the MODIS vegetation indices. Remote
Sensing of Environment, 83(1–2), 195–213.
Jacobsen, K. (2013): DEM Generation from High Resolution Satellite Imagery, PFG 5(2013),
483–493.
Jiménez, P. A., and J. Dudhia (2012): Improving the representation of resolved and unresolved
topographic effects on surface wind in the WRF model. Journal of Applied Meteorology
and Climatology, 51(2), 300–316. doi:10.1175/JAMC-D-11-084.1.
Kelly, M. C., and S.-E. Gryning (2010): Long-term mean wind profiles based on similarity
theory. Boundary-Layer Meteorology, 136(3), 377–390.
Lopes, A. S., J. Palma and V. J. Lopes (2013): Improving a two-equation turbulence model for
canopy flows using large-eddy simulation. Boundary-Layer Meteorology, 149, 231–257.
314 Remote Sensing for Sustainability

Loveland, T. R., B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. W. M. J. Yang and

J. W. Merchant (2000): Development of a global land cover characteristics database
and IGBP DISCover from 1 km AVHRR data. International Journal of Remote Sensing,
21(6–7), 1303–1330.
Mölder, M., and A. Lindroth (1999): Thermal roughness length of a boreal forest. Agricultural
and Forest Meteorology, 98–99, 659–670.
Moulin, S., L. Kergoat, N. Viovy and G. Dedieu (1997): Global-scale assessment of vegeta-
tion phenology using NOAA/AVHRR satellite measurements. Journal of Climate, 10,
1154–1170.
Peña, A., R. R. Floors, A. Sathe, S.-E. Gryning, R. Wagner, M. Courtney, X. G. Larsén, A. N.
Hahmann and C. B. Hasager (2015): Ten years of boundary-layer and wind-power
meteorology at Høvsøre, Denmark. Boundary-Layer Meteorology. doi:10.1007/s10546
-015-0079-8.
Perko, R., H. Raggam, J. Deutscher, K. Gutjahr and M. Schardt (2011): Forest assessment
using high resolution SAR data in x-band. Remote Sensing, 3, 792–815.
Perko, R., H. Raggam, K. Gutjahr and M. Schardt (2010): Deriving forest canopy height mod-
els using multi-beam TerraSAR-X imagery. In: Proceedings of 8th European Conference
on Synthetic Aperture Radar, Aachen, Germany, June 7–10, 2010; pp. 568–571.
Pineda, N., O. Jorba, J. Jorge and J. M. Baldasano (2004): Using NOAA AVHRR and SPOT
VGT data to estimate surface parameters: Application to a mesoscale meteorological
model. International Journal of Remote Sensing, 25(1).
Quackenbush, L. J. (2004): A review of techniques for extracting linear features from imagery.
Photogrammetric Engineering and Remote Sensing, 70(12), 1383–1392.
Raggam, H. (2006): Surface mapping using image triplets—Case studies and benefit assess-
ment in comparison to stereo image processing. Photogrammetric Engineering and
Remote Sensing, 72, 551–563.
Raggam, H., K. Gutjahr, R. Perko and M. Schardt (2010): Assessment of the stereo-radargram-
metric mapping potential of TerraSAR-X multibeam spotlight data. IEEE Transactions
on Geoscience and Remote Sensing, 48, 971–977.
Raupach, M. R. (1994): Simplified expressions for vegetation roughness length and zero
plane displacement height as functions of canopy height and area index (research note).
Boundary-Layer Meteorology, 71, 211–216.
Raupach, M. R. (1995): Vegetation–atmosphere interaction and surface conductance at leaf,
canopy and regional scales. Agricultural and Forest Meteorology, 73(3), 151–179.
Refslund, J., E. Dellwik, A. N. Hahmann, M. J. Barlage and E. Boegh (2013): Development of
satellite green vegetation fraction time series for use in mesoscale modeling: Application
to the European heat wave 2006. Theoretical and Applied Climatology, 1–16.
Santos-Alamillos, F. J., D. Pozo-Vázquez, J. A. Ruiz-Arias, V. Lara-Fanego and J. Tovar-
Pescador (2013): Analysis of WRF model wind estimate sensitivity to physics
parameterization choice and terrain representation in Andalusia (Southern Spain).
Journal of Applied Meteorology and Climatology, 52(7), 1592–1609. doi:10.1175/JAMC
-D-12-0204.1.
Schwartz, M. D., B. C. Reed and M. A. White (2002): Assessing satellite-derived start-of-
season measures in the conterminous USA. International Journal of Climatology, 22,
1793–1805.
Sogachev, A., and O. Panferov (2006): Modification of two-equation models to account for
plant drag. Boundary-Layer Meteorology, 121, 229–266.
Sun, G., K. Ranson, D. Kimes, J. Blair and K. Kovacs (2008): Forest vertical structure from
GLAS: An evaluation using LVIS and SRTM data. Remote Sensing of Environment,
112, 107–117.
Support of Wind Resource Modeling Using Earth Observation 315

Taylor, C. M., D. J. Parker and P. Harris (2007): An observational case study of mesoscale
atmospheric circulations induced by soil moisture. Geophysical Research Letters,
34(15).
Tian, X., Z. Y. Li, C. van der Tol, Z. Su, X. Li, Q. S. He, Y. F. Bao, E. X. Chen and L. H. Li
(2011): Estimating zero-plane displacement height and aerodynamic roughness length
using synthesis of LiDAR and SPOT-5 data. Remote Sensing of Environment, 115,
2330–2341.
Toutin, T., and L. Gray (2000): State-of-the-art of elevation extraction from satellite SAR data.
ISPRS Journal of Photogrammetry and Remote Sensing, 55, 13–33.
Troen, I., and E. L. Petersen (1989): European Wind Atlas. Riso National Laboratory, Roskilde,
Denmark. 656 pp.
Vannier, C., and L. Hubert-Moy (2008): Detection of wooded hedgerows in high resolution
satellite images using an object-oriented method. In: Proc. IGARSS, July 6–11, 2008,
Boston, Massachusetts, pp. IV, 731–734.
Weaver, C. P., and R. Avissar (2001): Atmospheric disturbances caused by human modification
of the landscape. Bulletin of the American Meteorological Society, 82, 269–281.
White, M. A., and R. R. Nemani (2006): Real-time monitoring and short-term forecasting of
land surface phenology. Remote Sensing of Environment, 104, 43–49.
Yang, Z.-L. (2004): Modeling land surface processes in short-term weather and climate stud-
ies. In: Observations, Theory, and Modeling of Atmospheric Variability (ed. X. Zhu),
World Scientific Series on Meteorology of East Asia, Vol. 3, World Scientific Publishing
Corporation, Singapore, pp. 288–313.
Zhang, X. Y., M. A. Friedl, C. B. Schaaf, A. H. Strahler, J. C. F. Hodges, F. Gao, B. C. Reed
and A. Huete (2003): Monitoring vegetation phenology using MODIS. Remote Sensing
of Environment, 84(3), 471–475.
 17 Assessing Solar
Energy Potential and
Building Energy Use
in Indianapolis Using
Geospatial Techniques
Yuanfan Zheng and Qihao Weng

CONTENTS
17.1 Introduction .................................................................................................. 317
17.2 Building Solar Energy Potential Estimation................................................. 318
17.3 Building Energy Use Simulation .................................................................. 320
17.4 Case Studies .................................................................................................. 321
17.4.1 Study Area and Data ......................................................................... 321
17.4.2 Case Study I: Assessing Solar Energy Potential in Indianapolis ..... 321
17.4.2.1 Building Reconstruction: 3D City Model .......................... 322
17.4.2.2 Annual Solar Energy Yield Estimation ............................. 323
17.4.2.3 Results ................................................................................ 326
17.4.3 Case Study II: Correlation Analysis between Building Energy
Use and Building Attributes in Indianapolis .................................... 327
17.4.3.1 Building Energy Consumption Simulation
in Indianapolis ................................................................... 328
17.4.3.2 Correlation Analysis .......................................................... 329
17.5 Summary and Discussion ............................................................................. 348
References .............................................................................................................. 349

17.1 INTRODUCTION
By 2030, urbanized areas will expand to provide homes for 81% of the world’s population,
with the majority of the population increase coming from developing countries (Weng
2015). Continued urbanization will bring impacts to the environment, and there is a
rapidly growing need for consideration of sustainable cities. Buildings are the major
component in the urban environment, so accurate and timely spatial information about
buildings (and associated attributes) in urban areas is needed as the basis to assist
decision making in understanding, managing, and planning the continuously changing
environment (Weng 2010). Regarding obtaining the information from buildings, the

317
318 Remote Sensing for Sustainability

traditional field survey approach is costly in time and labor; on the other hand, the
remote sensing technique can monitor the earth surface in a very high temporal and
spatial resolution, which has a great potential to extract building information automati-
cally from the urban area. Geographic Information Systems (GISs) have the strong
ability to gather and analyze the information of urban buildings by using geospatial
analysis technique. This chapter provides two case studies to present the most recent
researches on buildings under the topic of sustainable city development, which includes
the building solar potential assessment and correlation analysis of building annual
energy use and building attributes, using the city of Indianapolis as an example.

17.2 BUILDING SOLAR ENERGY POTENTIAL ESTIMATION

Renewable energy systems (RESs), which include but are not limited to wind
power, hydropower, solar energy, and geothermal energy, have become vital parts
of future energy use resources because fossil fuels are declining but demands for
energy keep growing. Solar energy has become a fast-growing energy source in
the last few years. Compared to other renewable sources of energy, solar energy
has the following advantages: it is inexhaustible, it can drastically reduce energy-
related greenhouse gases to help limit climate change, it is relatively well spread
over the globe (Philibert 2011), and it is not subject to geological, climatologic,
and morphologic conditions. The largest demands of solar energy are coming from
urban areas because more than half of the world’s population lives in the urban
environment (Weng 2015). Solar energy absorption equipment can be ground-
mounted arrays close to where the energy is used or directly mounted on rooftops
(SEIA 2014). There are three primary technologies by which solar energy is
commonly harnessed: photovoltaic (PV), concentrating solar power, and heating
and cooling systems (SEIA 2014). In the dense urban areas, the PV equipment are
encouraged to be mounted on the rooftops. First, it is not realistic to allocate a large
land to ground-mount them. Moreover, buildings are the largest energy consumers
in the urban area today (Philibert 2011), and the PV modules mounted on the roofs
can directly transfer the solar energy absorbed from sunlight into electricity and
sold to the electric companies.
Not all buildings are suitable for installing PV modules on their roofs, such as
historical buildings or buildings shaded by surrounding objects for most of the time.
Therefore, there is a need to evaluate the solar potential for each building. A completed
database including information on geographic location, height, volume, and owner-
ship for each building is essentially important to estimate its solar potential. Building
detection and reconstruction by using the remote sensed data and GIS technique has
become a quite popular research topic for the last decade, especially after the LiDAR
data became available because such data can reflect the absolute height of the ground
feature and are not affected by solar shadows. Numerous approaches have been done
for building boundaries detection and building reconstruction using LiDAR and other
data sources (Haala et al. 1998; Henn et al. 2012; Huang et al. 2013; Lafarge et al.
2007; Miliaresis and Kokkas 2007; Sohn and Dowman 2007) in the last two decades.
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 319

The building detection studies separated buildings from other features based
on a series of hypotheses, which are based on the spatial, spectral, and textural
characteristics of buildings. Sohn and Dowman (2007) presented an approach
for automatic extraction of building footprints in a combination of the LiDAR
data with IKONOS imagery pan-sharpened multispectral bands. They achieved
a detection accuracy of 90.1% and an overall accuracy of 80.5%. Miliaresis and
Kokkas (2007) used a region-growing algorithm to extract building class from
the LiDAR digital elevation models (DEMs) using the geomorphometric segmen-
tation principles. The interpretation of the spatial distribution of clusters that is
revealed by K-means classification method allows for identification of building
class (Miliaresis and Kokkas 2007).
The three-dimensional (3D) building reconstruction studies became more and
more popular since the 1990s. Haala et al. (1998) combined ground plans of build-
ings and LiDAR data to build the 3D building models for the application of virtual
reality. Lafarge et al. (2007) reconstructed buildings from DEMs and associ-
ated rectangular building footprints based on a Bayesian approach. The Bayesian
approach can automatically find the best-fit roof primitives in the predefined library
to represent given blocks of LiDAR point clouds within the building footprints.
Huang et al. (2013) presents a statistical approach to automatic 3D building roof
reconstruction from LiDAR point clouds. The selection of roof primitives was
driven by a variant of Markov Chain Monte Carlo technique with specific jump
mechanism. Three-dimensional models were built for 21 buildings and 1 city block
in Hannover, Germany. Henn et al. (2012) approached model selection among five
roof prototypes using supervised machine learning methods, which achieved a clas-
sification accuracy of 95% for 6696 roofs.
Once a completed GIS-based building polygon database showing the clear
boundaries is available, a geospatial technique can be used to calculate the
annual solar energy yield for each building footprint. Many previous studies
were focusing on the solar energy yield estimation based on application of GIS-
based radiation models, LiDAR data, and extracted building footprints. Schuffert
(2013) presented an automatic approach to extract suitable single roof planes from
LiDAR data and assigned regional irradiation data to each extracted roof sur-
face. A shadow analysis was also combined to obtain the realistic solar energy
values since the shadowing effect might significantly affect the incoming solar
radiation. Lukač et al. (2013) rated the solar potential of building roofs using
the method that combines extracted urban topography from LiDAR data with
the pyranometer measurement of global and diffuse solar irradiances. They also
created multiresolution shadowing to complete the proposed method. Tooke et
al. (2011) quantified the diurnal and seasonal impact of trees on solar radiation
yield. Their results indicated that trees on average reduce 38% of the total solar
radiation received by residential building rooftops. A point obstruction stacking
approach was designed by Tooke et al. (2013) to model irradiance on the walls of
multiple buildings by integrating contiguous data sets of surrounding urban form
and topography with building footprints.
320 Remote Sensing for Sustainability

17.3 BUILDING ENERGY USE SIMULATION

Globally, buildings are responsible for more than 40% of energy demand and
contribute more than 30% of CO2 emissions (Tooke 2014). Building energy use is
a major component of anthropogenic heat discharge. Zhou et al. (2011) estimated
the relationship between remotely sensed anthropogenic heat discharge and building
energy use, and the result suggests that there is an obvious consistency in terms
of spatial distribution. The high anthropogenic heat discharge occurs in the dense
residential and commercial areas, where the highest building energy use can be
found.
In order to get a better understanding of the current urban energy balance and
propose suggestions for sustainable city development, monitoring and analysis of
the building energy consumption are vital. There are two major building energy
use estimation approaches: top-down and bottom-up approaches (Heiple and Sailor
2008). The top-down approaches rely on extensive historical annual energy con-
sumption data and are typically applied to a larger area with a relatively coarse scale
(Heiple and Sailor 2008; Sailor and Lu 2004). The bottom-up approaches, on the
other hand, simulate hourly consumption of different types of energies for each sin-
gle building (Ward and Choudhary 2014). In order to examine the building energy
use for the purpose of energy-efficient building design in future sustainable cities,
the bottom-up approach should be chosen, since it can provided energy use informa-
tion for each building. The building energy simulation basically requires the input of
many attributes of buildings, such as geographic location, climate conditions, floor
space, floor numbers, building prototypes, electricity systems, and occupancy status.
Ward and Choudhary (2014) developed a methodology for the analysis of building
energy retrofits for a diverse set of buildings. The methodology requires selection
of appropriate building simulation tools based on the nature of the principal energy
demand (Ward and Choudhary 2014). Wang et al. (2015) used a bottom-up approach
to model residential heating energy consumption for China’s hot summer–cold win-
ter climatic region. They considered the occupant behaviors besides other building
characteristics.
Remote sensing technique provides the data for large spatial and temporal scale
monitoring. The GIS technique, on the other hand, can model individual building
energy consumption and analyze the correlation between energy use and factors
that might affect it. Results from various studies that combined remote sensing and
GIS techniques and showed the climate, geographic location, building attributes,
and occupant behavior can all contribute to the building energy consumption.
Hemsath and Bandhosseini (2015) indicate that both the vertical and horizontal
geometric proportion are equally sensitive to the material aspects related to build-
ing energy use, and the impact of the building geometry’s effect on energy use
depends on geographic locations. Premrov et al. (2015) pointed out that the total
annual energy demand for heating and cooling has a stronger correlation with
building shape factor if buildings are located in a cold climate with a lower solar
potential. A statistical method applied by Aksoezen et al. (2015) showed that there
is a strong interdependence between energy consumption, compactness, and build-
ing age.
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 321

17.4 CASE STUDIES

In this section, two case studies are presented. The first case study assesses the solar
energy potential for commercial and residential buildings in an urban area using
geospatial techniques. The second case study examines the relationship between
building energy use and some building attributes.

17.4.1 Study AreA And dAtA

The city of Indianapolis, which is the capital of Indiana, was chosen for both case
studies. The study area was 3.78 km2, which covered the Central Business District of
Indianapolis and some nearby residential areas. It is characterized by tall and complex
commercial buildings and low apartment houses with different roof shapes.
Three data sources were used in this study: LiDAR data, building footprint
polygons, and Marion County Assessor’s parcel data. In the United States, LiDAR
data are available nationwide. One of the most famous public data inventories is
the United States Interagency Elevation Inventory (USIEI), which is a collaborative
effort of the National Oceanic and Atmospheric Administration (NOAA) and the
US Geological survey, and it is updated annually. Building footprints are available
in almost all major cities in the United States, which are stored as GIS Shapefiles.
The Marion County Assessor’s parcel data are building Shapefile polygons with
information on prototypes and ages that can be linked to the building footprint for
building energy use simulation. The high-resolution orthophoto with 0.1 m spatial
resolution was used as supplemental data.
LiDAR data digital surface model (DSM) and DEM were both rasterized from
last return LiDAR point clouds, which were acquired from the Indiana Spatial Data
Portal for 2009. They have the same spatial resolution of 0.91 m. The LiDAR normal
height model (NHM), which represents absolute height information of objects above
the ground surface, was calculated as follows:

NHM = DSM − DEM. (17.1)

Building footprints data, which were provided by the city government of

Indianapolis, were acquired in the year 2013. They were used to mask out the
non-building features, such as tall trees in the LiDAR NHM. There are totally 411
building footprints in the study area. The building footprints less than 10 m2 were
removed from this study. Buildings having zero or negative height values were also
removed since building footprints were created after the LiDAR data were obtained.

17.4.2 CASe Study I: ASSeSSIng SolAr energy PotentIAl In IndIAnAPolIS

Simulation of annual solar energy yield was first performed for building roofs that
were reconstructed as a 3D city model from LiDAR DEM and building footprint.
The shadowing effect from the surrounding objects was considered in the simulation
since it will affect the total solar energy yield. Finally, the suitability of PV module
installation was rated for all reconstructed roof segments based on their annual solar
322 Remote Sensing for Sustainability

energy yield. The objectives of this case study were (1) to demonstrate a method to
estimate the solar potential of building roofs in the densely urban area and (2) to
create a complete 3D GIS-based city model that contains information of each
building in the study area such as height, volume, total floor area, compactness,
solar energy yield, and ownership for future use.

17.4.2.1 Building Reconstruction: 3D City Model

We reconstructed the commercial and residential buildings to a 3D city model. The
3D city model, which represents the geometry of urban environment, was used as
input for the annual solar energy yield simulation in the following stage. Compared
to the 2D building footprints, 3D city models with prototypical roofs are rarely avail-
able, but they are more valuable in applications of urban planning, environmental
modeling, and particularly of solar radiation calculations or noise emission simula-
tion (Henn et al. 2012) and urban structure analysis. LiDAR NHM data were used
to calculate the height of each building, and the GIS-based building footprints were
used to define the ground boundary of each building.
There are three steps in building reconstruction in this case study: step-edge
detection, boundary refinement, and 3D building modeling. Not all buildings have
a flat roof; many commercial buildings have substructures on their roofs. Direct use
of those box-shaped models to represent buildings with complex and irregular roof
shapes may cause loss of roof information and may lead to inaccurate reconstruction.
Therefore, the step edge detection on building roofs has to be performed first. Those
step edges may either be the boundaries of roof substructures or the transition with
large variation in height between two parts in the rooftop (Figure 17.1). Ignoring step
edges would lead to inaccurate reconstruction of building roofs.
Sobel’s edge detection method (Sobel and Feldman 1968) was first used to find out
the possible step edges on roofs. Figure 17.2 presents the result of applying the Sobel

FIGURE 17.1 Building with step edges on its roof.

Assessing Solar Energy Potential and Building Energy Use in Indianapolis 323

FIGURE 17.2 Step edge detection on a building roof.

edge detector to the LiDAR NHM of a selected building. The step edge detection can
split some building footprints into many small polygons, which causes an overseg-
mentation problem. Therefore, refinement was carried out after the step edge–based
segmentation. The purpose of refinement is to merge the small polygons into their
adjacent larger polygons and simplify the boundaries of building footprints. An area
threshold (30 m2) was used to merge the cells with their adjacent larger cells. If there
were multiple adjacent cells, the one sharing the longest boundary would be chosen.
The Bend Simplify algorithm developed by Wang (1996) was used to simplify the
boundaries into straight lines since it often produces results that are more faithful
to the original and more aesthetically pleasing (Wang 1996). A classification for
building footprints by their areas was performed before the simplification since foot-
prints in different sizes require different simplification tolerance values. The same
simplification tolerance may cause the small footprints to lose their essential shapes,
while large footprints still keep the extraneous bends. A total of six groups of build-
ing footprints were classified: (1) less than 400 m2, (2) 400 to 1000 m2, (3) 1000 to
2000 m2, (4) 2000 to 4000 m2, (5) 4000 to 10,000 m2, and (6) greater than 10,000 m2.
Mean height of pixels within the boundaries of each building segment was calcu-
lated using LiDAR data. This research used box-shaped buildings, which extruded from
the 2D building segments based on the mean height value to create the 3D city model.

17.4.2.2 Annual Solar Energy Yield Estimation

The annual solar energy yield for individual building roof segments is the most
important indicator for their solar suitability rating in this study. Since the regional
irradiation data used in Schuffert’s (2013) study are not always available in every
city, it would be good to propose a solar energy simulation method, which calculates
the instant solar energy yield based on the angular motion of the sun.
The incoming solar radiation emitted by the sun has three major components
when it reaches the earth’s surface: direct radiation, diffuse radiation, and reflect
radiation. Direct radiation is the largest component of total radiation, which travels
on a straight line from the sun down to the surface of the Earth. Diffuse radiation
is solar radiation that reaches the Earth’s surface after being scattered from direct
324 Remote Sensing for Sustainability

radiation. Reflected radiation is reflected from surface features, which constitutes

only a small proportion of total radiation.
In this case study, a solar radiation map was generated by the Area Solar Radiation
tool in the ArcGIS Spatial Analyst toolbox, using the LiDAR NHM and latitude of
the study area as inputs. The simulation was based on the hemispherical viewshed
algorithm developed by Rich et al. (1994) and further developed by Fu and Rich
(2002). The total annual solar energy amount was calculated as the global radiation
for each pixel in the solar radiation map. The global radiation was calculated as the
sum of direct and diffuse radiation using the following equation:

Iglo = Idir + Idif, (17.2)

where Iglo is the global radiation, Idir is the direct radiation, and Idif is the diffuse
radiation. The reflect radiation was not considered in this study, because according
to Lukač et al. (2013), the reflected radiation only contributed a little to the total
amount of global radiation. The solar radiation tool in ArcGIS involves four steps for
solar energy calculation:

1. Calculate an upward-looking hemispherical viewshed, such as a fisheye

view, at each observation point location.
2. Calculate the direct radiation based on the overlay of the viewshed and
direct sun map.
3. Calculate the diffuse radiation based on the overlay of the viewshed and
diffuse sun map.
4. Repeat the same calculation process for every location in each simulation time.

The input latitude for this case study was defined as 39.77 for Indianapolis. The
simulation time was set for every 14 days in the year 2010. Since the course of sun
changes continuously over time in a few days, it is not necessary to calculate each
single day (Schuffert 2013). The hourly interval within a day was set to 2 h. The
sky size was set to 200, which is the highest resolution for the viewshed, sky map,
and sun map grids. The slope and aspect input type was set to “from_DEM”. Slope
means steepness, aspect means the slope, and aspect grids are calculated from the
input raster. The calculation direction was set to 32, which means the highest degree
of complexity. Zenith division was set to 8, which means relative to zenith. Azimuth
division was set to 8, which means relative to north. For a diffuse model type, uni-
form sky was the setting representing the incoming diffuse radiation, which is the
same from all sky directions. Both the diffuse proportion and transmittivity propor-
tion were set according to the clear sky condition for all times, which is the most
probable weather condition in Indianapolis.
The instant direct irradiance (Idir) for a given pixel can be calculated using the
following equation:

I dirθ,α = SConst ∗ β m (θ) ∗ SunDurθ,α ∗ SunGapθ,α ∗ cos( AngIn θ,α ), (17.3)

Assessing Solar Energy Potential and Building Energy Use in Indianapolis 325

where θ is the zenith angle, α is the azimuth angle, S Const is the solar constant
(1367 w/m2), β is the atmospheric transmissivity, m(θ) is the relative optical path
length, SunDurθ,α is the time duration represented by the sky sector, SunGapθ,α is the
gap fraction, and AngInθ,α is the angle of incidence between the centroid of the sky
sector and the axis normal to the surface.
The instant diffuse irradiance (Idif ) for a given pixel can be calculated using the
following equation:

I difθ,α = Rglb ∗ Pdif ∗ Dur * Skygapθ,α ∗ Weight θ,α ∗ cos( AngIn θ,α ), (17.4)

where Rglb is the global normal irradiance, Pdif is the proportion of the global normal
radiation flux that is diffused, Dur is the time interval used for analysis, SkyGapθ,α is the
proportion of visible sky for the sky sector, Weightθ,α is the proportion of diffuse irra-
diance originating in a given sky sector relative to all sectors, and cos(AngInθ,α) is the
angle of incidence between the centroid of the sky sector and the intercepting surface.
The instant solar energy yield in all time nods was calculated and summed up
to obtain the daily solar energy yield map. All daily solar energy yield maps were
summed up to obtain the annual solar energy yield map.
Roof can be shadowed by its surrounding objects such as other buildings and
trees that are taller. In order to obtain the realistic estimation of annual solar energy
yield, a shadowing effect should be considered because it can affect the absorption of
direct radiation. The shadow balance for a pixel at a given time was calculated using
the following equation:

S = 255.0 * ((cos(Zenith_rad) * cos(Slope_rad)) + (sin(Zenith_rad)

* sin(Slope_rad) * cos(Azimuth_rad − Aspect_rad))), (17.5)

where Zenith_rad and Azimuth_rad are the zenith and azimuth angles of the
illumination source at a given location and given time, which can be obtained in
the NOAA solar calculator website (http://www.noaa.gov/). The zenith angle can
be obtained by subtracting the altitude angle from 90°. Slope_rad is the slope value
at the given pixel, and Aspect_rad is that of the aspect value. In order to match the
simulation time matched in the annual solar energy yield simulation, a subset of days
in each 14 days was chosen. In each chosen day, the sun position was calculated for
six representative time nods: 8 a.m., 10 a.m., 12 p.m., 2 p.m., 4 p.m., and 6 p.m. The
instant shadow balance maps in six time nods were summed up to obtain the daily
shadow balanced map, and all daily shadow balanced maps were summed up to
obtain the annual shadow balance map pixels within the study area were classified
into three categories based on their annual shadow balance values: always-shaded
zone, half-shaded zone, and unshaded zone. This classification result was applied
to annual direct irradiance value in each pixel to obtain the realistic annual solar
energy yield simulation result using the following equation:

I glo = I dir ∗ SC + I dif , (17.6)

326 Remote Sensing for Sustainability

where SC is the shadowing coefficient. SC was set to 0.1 for the always-shaded
zone, because pixels in this zone can hardly receive direct irradiance owing to the
shadowing effect. SC was set to 0.5 for the half-shaded zone, which means that
pixels in this zone receive partial amount of direct solar irradiance. SC was set to
1 for the unshaded zone, which means pixels in this zone receive full amount of
direct solar irradiance. The annual solar energy yield for a specific reconstructed
roof segment was assigned using the mean value of annual global solar irradiance
values of all pixels within it.

17.4.2.3 Results
Figure 17.3 shows the annual solar energy yield in downtown Indianapolis; the
warmer colors represent larger values of annual global solar radiation absorption,
while cooler colors represent smaller values of annual global solar radiation
absorption. The solar potential index for each roof segment can be calculated using
the following equation:

SPi = (SEi)/(SEmax), (17.7)

where SPi is the solar potential index, SEi is the annual solar energy yield for the
given roof segment, and SEmax is the roof segment with the highest annual solar
energy yield in the study area. All the roof segments were assigned to five categories
according to their solar potential index, as presented in Figure 17.4.
Figure 17.4 shows the result of a building roof’s PV suitability with a view of the
3D city model after assigning the solar potential index into the roof segments’ attri-
bute table. Purple and blue represent the roof segments that are unsuitable or with
low suitability for PV module installation, green represents the roof segments with
medium level of suitability, and yellow and red represent the roof segments with high

Annual global
solar irradiance
(Unit: watts hours
per square meter)
High: 2707375

Low: 847

FIGURE 17.3 Annual global solar irradiance in Indianapolis.

Assessing Solar Energy Potential and Building Energy Use in Indianapolis 327

SPi
<0.20 Unsuitable
0.20–0.35 Low suitability
0.35–0.50 Medium suitability
0.50–0.70 High suitability
<0.70 Very high suitability

FIGURE 17.4 Assessment of building PV suitability in downtown Indianapolis with a 3D

city model.

or very high suitability. With the 3D city model, it can be determined that the com-
mon characteristics of roof segments with lower PV suitability (represented by blue
and purple colors) are located in the downtown area, surrounded by higher objects,
and have a smaller floor space area. Shading by the higher surrounding objects for
most of the time is probably the reason. Since they have a small floor space area, a
large portion of them are shaded. Roof segments with higher PV suitability (repre-
sented by red and yellow colors) usually have greater heights than the surrounding
objects or are located at a distance from the surrounding objects, making sure that
they are not shaded for most of the time.

17.4.3 CASe Study II: CorrelAtIon AnAlySIS between buIldIng

energy uSe And buIldIng AttrIbuteS In IndIAnAPolIS
Buildings are a major component in urban areas, and building energy use accounts for
a large proportion of anthropogenic heat. Understanding the characteristic of energy
use for different types of buildings is vital for current sustainable city research. The
objectives of this case study were to analyze the correlation between building energy
use and their attributes in different levels, to find out the most significant contributor
among these attributes, and to propose an energy-saving suggestion based on the
results.
328 Remote Sensing for Sustainability

17.4.3.1 Building Energy Consumption Simulation in Indianapolis

Energy consumption from commercial and residential buildings in downtown
Indianapolis was simulated using the eQUEST software, which is an energy
simulation tool designed to model the annual energy cost for a single building.
The methodology was illustrated in Figure 17.5. The simulation requires several
inputs related to the buildings such as building shape, floor area, floor numbers,
and prototype. LiDAR NHM and building footprint were used to determine the
floor numbers through dividing the height of the building by the floor-to-floor
height. In order to obtain a more accurate estimation, eave height should be used
for buildings with non-flat roofs, rather than using the top height, mean height, or
height for major parts of the roofs. A classification of flat and non-flat roof was
performed, based on a slope threshold. The percentage of pixels with a slope value
lower than 5° in each cell was calculated, and the cells with at least 40% of pixels
having a slope value lower than 5° would be recognized as flat roofs. The reason
for setting the threshold to 5° rather than 0° is that many flat roofs contain a few
substructures on their top. For building roofs classified as non-flat roofs, inner

LIDAR NHM Building footprint Parcel data

Roof classification

Non-flat Flat
roof roof
Parcel sub-footprint

Eave height Mean height

Floor number Area Shape Prototype

Energy simulation in EQUEST

Annual electricity consumption Annual fuel consumption

FIGURE 17.5 Methodology of annual building energy use simulation.

Assessing Solar Energy Potential and Building Energy Use in Indianapolis 329

buffer zones were created around the boundary of building footprints to extract the
eave area and calculate the eave height.
For individual flat roofs, the mean height of the LiDAR NHM pixels within it was
used to estimate the floor numbers. The same height value was assigned to parcel
sub-footprints within the same building footprint. The floor-to-floor height was set
differently for buildings with different prototypes, according to a standard defined
by the US Department of Energy.
Building footprint polygon and parcel polygon were overlaid to create a new poly-
gon layer. Polygons in the new layer maintain the original boundary of building foot-
prints as their outside boundary as well as contain the parcel boundaries within the
building footprint. One building footprint may have multiple parcel sub-footprints,
which can be explained by the fact that some commercial buildings are shared by
different owners for different purposes of use.
A total of 10 building prototypes, which include 8 for commercial and 2 for
residential buildings, were determined according to the parcel data for each par-
cel building sub-footprint. Each single parcel sub-footprint obtained a unique
setting of parameters for simulation according to their own characteristics.
Assigning building prototype in the first step of simulation would automati-
cally affect the default setting of several major parameters such as weekday and
weekend schedule, internal load (e.g., light and equipment use), activity area
allocation, and cooling and heating equipment characteristics. Specific meteoro-
logical information will be retrieved and will contribute to the simulation based
on the input of location setting and year for simulation. With the combination of
prototypes, related parameters, area, shape, number of floors, age, location, and
retrieved meteorological data, the annual electricity and annual fuel consump-
tion were simulated for each parcel sub-footprint. Indianapolis and the year 2014
were chosen for the location setting and year of simulation, respectively. The
energy consumption for each building was calculated by summing up energy
consumption for all parcels within it. Figure 17.6 shows the amount of annual
fuel and electricity consumption for each commercial and residential building in
downtown Indianapolis.

17.4.3.2 Correlation Analysis

This section provides a correlation analysis to examine the relationship of energy
use and building attributes including ground area, total floor area, height, surface
area, compactness, aspect ratio, and orientation. The ground area represents the
area of individual GIS building footprint. The total floor area represents the total
floor space of all stories in the building, which was calculated by multiplying
the ground floor area by floor numbers. The floor numbers were calculated by
dividing the building height by the floor-to-floor height of the building, which
is generally 7.5 ft in the case of residential buildings and 10 ft for commercial
buildings (Chun 2010). Surface area (SA) represents the outer exposed building
surface, including the roof surface and walls, can be calculated by the following
equation:

SA = A + P * h, (17.8)
330 Remote Sensing for Sustainability

(a) (b)
Less than 70,000 Less than 70,000
70,000–200,000 70,000–200,000
200,000–1,000,000 200,000–1,000,000
1,000,000–2,000,000 1,000,000–2,000,000
2,000,000–5,000,000 2,000,000–5,000,000
Greater than 5,000,000 Greater than 5,000,000

FIGURE 17.6 Amount of annual fuel consumption (a) and electricity (b) consumption (kilo-
watt hour) for each commercial and residential building in downtown Indianapolis.

where A denotes ground area, P is the perimeter of building footprint, and h is

the height of the building. Compactness is the ratio of building surface to volume,
and is a proxy for thermal energy emission from a building’s exposed surface
(Chun 2010). The aspect ratio is the proportional relationship between building
width and height. Orientation, which is the main angle of building, ranges from
−90° to 90°. The values −90°, 0°, and 90° denote that the longer side of the
building points to the west, north, and east, respectively. Figure 17.7 presents the
average amount of annual electricity and fuel consumption for all commercial
and residential buildings in the study area. Since a large contrast between the
amounts of annual energy consumption in commercial and residential buildings
can be observed, correlation analysis will be performed to these two major types
of buildings separately.
The relationships between building energy consumption and their area are pre-
sented in Figures 17.8 and 17.9 as scatterplots. It can be found out that building
energy use has a positive correlation with building ground area and total floor area.
In order to avoid the impact from building area, the relationship of the other attri-
butes and energy use per square foot is examined. Figures 17.10 and 17.11 present
the relationship between annual energy consumption per square foot and build-
ing height, surface area, compactness, aspect ratio, and orientation in scatterplots.
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 331

Average annual consumption

of fuel (kWh)
1,800,000
1,600,000
1,400,000
1,200,000
1,000,000
800,000
600,000
400,000
200,000
0
(a) Commercial Residential

Average annual consumption

of electricity (kWh)
1,200,000
1,000,000
800,000
600,000
400,000
200,000
0
(b) Commercial Residential

FIGURE 17.7 Average amount (kilowatt hour) of annual consumption of fuel (a) and
electricity (b) for commercial and residential buildings in downtown Indianapolis.

Table 17.1 defines the correlation between energy use and all attributes using Pearson
product–moment correlation coefficients.
A very strong correlation can be observed between building energy use and total
floor area for both commercial and residential buildings with a correlation from 0.889
to 0.977. Relatively lower correlations are observed for the relationship between energy
use and ground area, which indicates that energy use might have a positive correlation
with building height. As Table 17.1 suggests, for commercial buildings, height has
stronger positive correlations of 0.647 and 0.638 with fuel and electricity consumption
per square foot, respectively, compared to residential buildings. Although it appears
that height has a weak correlation with residential building energy use, a higher posi-
tive correlation between height and ground area explains the fact that residential build-
ing energy use has a higher correlation with total floor area than with ground area. It
can also be found out that area has a stronger impact on energy use than height. It is
also worth noting that energy use per square foot has a negative correlation with com-
pactness for commercial buildings, indicating that commercial buildings with com-
plicated shapes consume less energy than others with simple shapes and similar sizes
in Indianapolis. No significant correlations can be observed between the energy use
of buildings and their attributes, including surface area, aspect ratio, and orientation.
332 Remote Sensing for Sustainability

40,000,000

Annual fuel consumption (kWh)

30,000,000
for commercial buildings

20,000,000

10,000,000

0 200,000 400,000 600,000 800,000 1,000,000 1,200,000

(a) Ground area (ft2)

1,000,000
Annual fuel consumption (kWh)

800,000
for residential buildings

600,000

400,000

200,000

0 2000 4000 6000 8000 10,000

(b) Ground area (ft2)

FIGURE 17.8 Relationship between annual fuel consumption and ground area (a and b) for
commercial and residential buildings. (Continued)
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 333

40,000,000

Annual fuel consumption (kWh)

for commercial buildings
30,000,000

20,000,000

10,000,000

0 1,000,000 2,000,000 3,000,000 4,000,000 5,000,000 6,000,000

(c) Building total area (ft2)

1,000,000
Annual fuel consumption (kWh)

800,000
for residential buildings

600,000

400,000

200,000

0 20,000 40,000 60,000 80,000 100,000 120,000

(d) Building total area (ft2)

FIGURE 17.8 (CONTINUED) Relationship between annual fuel consumption and building
total area (c and d) for commercial and residential buildings.
334 Remote Sensing for Sustainability

30,000,000

(kWh) for commercial buildings 25,000,000

Annual electricity consumption

20,000,000

15,000,000

10,000,000

5,000,000

0 200,000 400,000 600,000 800,000 1,000,000 1,200,000

(a) Ground area (ft2)

600,000

500,000
Annual electricity consumption
(kWh) for residential buildings

400,000

300,000

200,000

100,000

0 2000 4000 6000 8000 10,000

(b) Ground area (ft2)

FIGURE 17.9 Relationship between annual electricity consumption and ground area (a and
b) for commercial and residential buildings. (Continued)
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 335

30,000,000

(kWh) for commercial buildings 25,000,000

Annual electricity consumption

20,000,000

15,000,000

10,000,000

5,000,000

0 1,000,000 2,000,000 3,000,000 4,000,000 5,000,000 6,000,000

(c) Building total area (ft2)

600,000

500,000
Annual electricity consumption
(kWh) for residential buildings

400,000

300,000

200,000

100,000

0 20,000 40,000 60,000 80,000 100,000 120,000

(d) Building total area (ft2)

FIGURE 17.9 (CONTINUED) Relationship between annual electricity consumption and

building total area (c and d) for commercial and residential buildings.
336 Remote Sensing for Sustainability

250

square foot for commercial buildings

Annual fuel consumption (kWh) per
200

150

100

50

0 100 200 300 400

(a) Building height (ft)

120
Annual fuel consumption (kWh) per
square foot for residential buildings

100

80

60

40

20

0 20 40 60 80 100 120
(b) Building height (ft)

FIGURE 17.10 Relationship between annual fuel consumption per square foot and building
height (a and b) for commercial and residential buildings. (Continued)
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 337

250

square foot for commercial buildings

Annual fuel consumption (kWh) per 200

150

100

50

0 250,000 500,000 750,000 1,000,000 1,250,000

(c) Building surface area (ft2)

120
Annual fuel consumption (kWh) per
square foot for residential buildings

100

80

60

40

20

0
0 20,000 40,000 60,000
(d) Building surface area (ft2)

FIGURE 17.10 (CONTINUED) Relationship between annual fuel consumption per

square foot and building surface area (c and d) for commercial and residential buildings.
(Continued)
338 Remote Sensing for Sustainability

250

square foot for commercial buildings

Annual fuel consumption (kWh) per 200

150

100

50

0.0 0.2 0.4 0.6

(e) Compactness

120
Annual fuel consumption (kWh) per
square foot for residential buildings

100

80

60

40

20

0.05 0.10 0.15 0.20 0.25

(f ) Compactness

FIGURE 17.10 (CONTINUED) Relationship between annual fuel consumption per square
foot and compactness (e and f) for commercial and residential buildings. (Continued)
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 339

250

square foot for commercial buildings

Annual fuel consumption (kWh) per 200

150

100

50

0 5 10 15 20
(g) Aspect ratio

120
Annual fuel consumption (kWh) per
square foot for residential buildings

100

80

60

40

20

1.0 1.5 2.0 2.5 3.0 3.5 4.0

(h) Aspect ratio

FIGURE 17.10 (CONTINUED) Relationship between annual fuel consumption per square
foot and aspect ratio (g and h) for commercial and residential buildings. (Continued)
340 Remote Sensing for Sustainability

250

Annual fuel consumption (kWh) per

square foot for residential buildings
200

150

100

50

–100 –50 0 50 100

(i) Orientation

120
Annual fuel consumption (kWh) per
square foot for residential buildings

100

80

60

40

20

–20 0 20 40 60 80 100
(j) Orientation

FIGURE 17.10 (CONTINUED) Relationship between annual fuel consumption per square
foot and orientation (i and j) for commercial and residential buildings.
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 341

200

per square foot for commercial buildings

Annual electricity consumption (kWh)
150

100

50

0 100 200 300 400

(a) Building height (ft)

100
per square foot for residential buildings
Annual electricity consumption (kWh)

80

60

40

20

0 20 40 60 80 100 120
(b) Building height (ft)

FIGURE 17.11 Relationship between annual electricity consumption per square foot and
building height (a and b) for commercial and residential buildings. (Continued)
342 Remote Sensing for Sustainability

200

per square foot for commercial buildings

Annual electricity consumption (kWh)
150

100

50

0 250,000 500,000 750,000 1,000,000 1,250,000

(c) Building surface area (ft2)

100
per square foot for residential buildings
Annual electricity consumption (kWh)

80

60

40

20

0 20,000 40,000 60,000

(d) Building surface area (ft2)

FIGURE 17.11 (CONTINUED) Relationship between annual electricity consumption per

square foot and building surface area (c and d) for commercial and residential buildings.
(Continued)
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 343

200

per square foot for commercial buildings

Annual electricity consumption (kWh)
150

100

50

0.0 0.2 0.4 0.6

(e) Compactness

100
per square foot for residential buildings
Annual electricity consumption (kWh)

80

60

40

20

0.05 0.10 0.15 0.20 0.25

(f ) Compactness

FIGURE 17.11 (CONTINUED) Relationship between annual electricity consumption per

square foot and compactness (e and f) for commercial and residential buildings. (Continued)
344 Remote Sensing for Sustainability

200

per square foot for commercial buildings

Annual electricity consumption (kWh)
150

100

50

0 5 10 15 20
(g) Aspect ratio

100
per square foot for residential buildings
Annual electricity consumption (kWh)

80

60

40

20

1.0000 1.5000 2.0000 2.5000 3.0000 3.5000 4.0000

(h) Aspect ratio

FIGURE 17.11 (CONTINUED) Relationship between annual electricity consumption per

square foot and aspect ratio (g and h) for commercial and residential buildings. (Continued)
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 345

200

per square foot for commercial buildings

Annual electricity consumption (kWh)
150

100

50

–100 –150 0 50 100

(i) Orientation

100
per square foot for residential buildings
Annual electricity consumption (kWh)

80

60

40

20

–20.0000 0.0000 20.0000 40.0000 60.0000 80.0000 100.0000

(j) Orientation

FIGURE 17.11 (CONTINUED) Relationship between annual electricity consumption per

square foot and orientation (i and j) for commercial and residential buildings.
346 Remote Sensing for Sustainability

TABLE 17.1
Analysis of the Relationship between Energy Use and Building Attributes
Using a Correlation (Pearson Product–Moment Correlation Coefficients)
Technique
Commercial Buildings Residential Buildings
F vs. ground area 0.893 0.722
E vs. ground area 0.860 0.589
F vs. total area 0.903 0.977
E vs. total area 0.889 0.937
Ground area vs. height 0.076 0.659
FU vs. height 0.647 0.495
FU vs. surface area 0.3 0.411
FU vs. compactness −0.412 −0.251
FU vs. aspect ratio −0.077 −0.18
FU vs. orientation 0.15 −0.094
EU vs. height 0.638 0.226
EU vs. surface area 0.28 0.12
EU vs. compactness −0.388 −0.024
EU vs. aspect ratio −0.107 −0.002
EU vs. orientation 0.093 −0.196

Note: E, annual electricity consumption; F, annual fuel consumption; FE, annual electricity
consumption per square foot; FU, annual fuel consumption per square foot.

Correlations between energy use and supplementary attributes, including build-

ing age and prototypes, are also examined. The values of energy consumption per
square foot were divided by the floor number to get rid of the influence of build-
ing height. Figure 17.12 indicates that there is a positive correlation (0.558) between
annual fuel consumption per square foot in a single floor and residential building
age, indicating that older residential buildings tend to have higher annual fuel con-
sumption. This might be due to the fact that space heating equipment in newer resi-
dential buildings are more efficient. Moreover, the newer residential buildings have
a better design of building shells for the purpose of energy saving. Table 17.2 pres-
ents the average annual energy use (fuel and electricity) per square foot in a single
floor for eight types of commercial buildings. It is worth noting that compared to
other subtypes, office buildings have significantly higher annual fuel and electricity
consumption per unit floor area. The possible reason is that office buildings have a
higher density of people and more equipment than other buildings such as comput-
ers, printers, and space cooling and heating facilities. The other reason is that in US
cities, office buildings in the downtown area tend to keep most of the internal lights
on for the whole night.
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 347

100

Annual fuel consumption (kWh)

per square foot in a single floor
80

for residential buildings

60

40

20

0 20 40 60 80 100 120
(a) Age

50
(kWh) per square foot in a single
Annual electricity consumption

40
floor for residential buildings

30

20

10

0
0 20 40 60 80 100 120
(b) Age

FIGURE 17.12 Relationship between annual fuel consumption per square foot in single
floor and building age (Pearson’s r = 0.558) (a) and between annual electricity consumption
per square foot in single floor and building age (Pearson’s r = 0.331) (b).
348 Remote Sensing for Sustainability

TABLE 17.2
Average Annual Energy Use per Square Foot in Single Floor
for Eight Subtypes of Commercial Buildings
Types Fuel (kWh) Electricity (kWh)
Community center 19.56 15.52
Conference/convention center 26.70 18.84
Hotel 17.92 9.31
Office 51.51 48.29
Religious 9.23 4.75
Retail 9.76 5.61
School 11.34 5.25
Theater 6.49 3.97

17.5 SUMMARY AND DISCUSSION

This chapter provided two case studies focusing on sustainable city development,
using Indianapolis as an example. The first case study overall demonstrated a
method to estimate the solar potential of building roofs in densely populated urban
areas. The second case study examined the correlation between building energy use
and their attributes.
In the first case study, the accuracy of solar energy yield simulation can be
improved if meteorological data are available since several input parameters, such
as diffuse proportion, transmittivity proportion, and shadowing effect, are related
to instant sky conditions. Seasonal-based analysis can also be performed since
Indianapolis usually experiences heavy snow in the winter. One limitation of this
case study is that it applied the same shadowing analysis algorithm for buildings
and vegetation without considering the transmittivity of the vegetation. Moreover,
regarding building wall material data, reflected radiation should be considered in the
simulation if there are walls made of materials with high reflectivity. A completed
3D building model with physical attributes such as height, volume, total floor area,
annual solar energy yield, shadow balance, PV suitability, and ownership has poten-
tial applications that can be used by urban planners and local solar energy companies
since it will provide a multistage decision-making scheme regarding PV suitabil-
ity. With the ownership parcel data, historical preserved buildings and government
buildings can be removed from the list in the first stage, which can also save a lot
of calculation time. Buildings with low annual solar energy yield can be removed
in the next stage. In the final stage, urban planners can make predictions for each
chosen roof surface (e.g., how many kilowatt hours of solar energy can be absorbed
theoretically per square meter annually). Local solar energy companies and local
governments will be able to determine the profits incurred for each particular build-
ing, each neighborhood and district, and the entire city if the cost of PV installation
is considered.
Assessing Solar Energy Potential and Building Energy Use in Indianapolis 349

In the second case study, the significant contributors to building energy use were
determined to be area and age for residential buildings and area, height, compact-
ness, and prototype for commercial buildings. For the purpose of saving energy,
attention should be given to the management of office buildings since such struc-
tures consume more energy than others. Many office buildings can immediately
save energy using some short-term solutions, such as turning off some of the lights,
computers, office equipment, and space heaters during the night. Long-term solu-
tions should also be provided, such as replacing the current lights with energy-saving
lights to reduce internal loads. In order to save energy in residential buildings, the
refurbishment of older buildings should be advocated. For example, the traditional
space heating facilities in such houses can be replaced by energy-saving ones. Future
studies on building energy use should consider more potential factors such as the
income level and the number of residents.

REFERENCES
Aksoezen, M., Daniel, M., Hassler, U., and Kohler, N. 2015. Building age as an indicator for
energy consumption. Energy and Buildings 87: 74–86.
Chun, B. S. 2010. Three-Dimensional City Determinants of the Urban Heat Island: A Statistical
Approach. PhD diss., Ohio State University.
Fu, P., and Rich, P. 2002. A geometric solar radiation model with applications in agriculture
and forestry. Computers and Electronics in Agriculture 37: 25–35.
Haala, N., Brenner, C., and Anders, K. H. 1998. 3D urban GIS from laser altimeter and 2D
map data. International Archives of Photogrammetry and Remote Sensing, Remote
Sensing and Spatial Information Science 32: 339–346.
Heiple, S., and Sailor, D. J. 2008. Using building energy simulation and geospatial modeling
techniques to determine high resolution building sector energy consumption profiles.
Energy and Buildings 40: 1426–1436.
Hemsath, T. L., and Bandhosseini, K. A. 2015. Sensitivity analysis evaluating basic building
geometry’s effect on energy use. Renewable Energy 76: 526–538.
Henn, A., Groger, G., Stroh, V., and Plumer, L. 2012. Model driven reconstruction of roofs
from sparse LiDAR point clouds. ISPRS Journal of Photogrammetry and Remote
Sensing 76: 17–29.
Huang, H., Brenner, C., and Sester, M. 2013. A generative statistical approach to automatic 3D
building roof reconstruction from laser scanning data. ISPRS Journal of Photogrammetry
and Remote Sensing 79: 29–34.
Lafarge, L., Descombes, X., Zerubia, J., and Deseilligny, M. 2007. Structural approach for
building reconstruction from a single DSM. IEEE Transactions on Pattern Analysis and
Machine Intelligence 6: 135–147.
Lukač, N., Zlaus, D., Seme, S., Zalik, B., and Stumberger, G. 2013. Rating of roofs surfaces
regarding their solar potential and suitability for PV systems, based on LiDAR data.
Applied Energy 102: 803–812.
Miliaresis, G., and Kokkas, N. 2007. Segmentation and object-based classification for the
extraction of the building class from LIDAR DEMs. Computers & Geosciences 33:
1076–1087.
National Oceanic and Atmospheric Administration, http://www.noaa.gov/ (accessed September
20, 2015).
Philibert, C. 2011. Renewable Energy Technologies Solar Energy Perspectives. OECD/IEA.
350 Remote Sensing for Sustainability

Premrov, M., Leskovar, V. Z., and Mihalic, K. 2015. Influence of the building shape on the energy
performance of timber-glass buildings in different climatic conditions. Energy 1–11.
Rich, P., Dubayah, R., Hetrick, W., and Saving, S. 1994. Using Viewshed models to cal-
culated intercepted solar radiation: Applications in ecology. American Society for
Photogrammetry and Remote Sensing, Technical Paper 524–529.
Sailor, D., and Lu, L. 2004. A top-down methodology for developing diurnal and seasonal
anthropogenic heating profiles for urban areas. Atmospheric Environment 17:
2737–2748.
Schuffert, S. 2013. An automatic data driven approach to derive photovoltaic-suitable roof
surfaces from ALS data. Urban Remote Sensing Event 267–270.
SEIA. Solar energy technologies solutions for today’s energy needs, http://www.seia.org/sites
/default/files/Solar%20Energy%20Technologies%20Overview%2011-13-2014.pdf.
Sobel, I. and Feldman, G. 1968. A 3 × 3 isotropic gradient operator for image processing.
Presented at a talk at the Stanford Artificial Project.
Sohn, G., and Dowman, I. 2007. Data fusion of high-resolution satellite imagery and LiDAR
data for automatic building extraction. ISPRS Journal of Photogrammetry & Remote
Sensing 62: 43–63.
Tooke, T. R. 2014. Building energy modeling and mapping using airborne LiDAR. PhD diss.,
The University of British Columbia.
Tooke, T. R., Coops, N. C., and Christen, A. 2013. A point obstruction stacking (POSt)
approach to wall irradiance modeling across urban environments. Building and
Environment 60: 234–242.
Tooke, T. R., Coops, N. C., Voogt, J. A., and Meitner, M. J. 2011. Tree structure influences on
rooftop-received solar radiation. Landscape and Urban Planning 102: 73–81.
U.S. Department of Energy, Building Energy Codes Program, https://www.energycodes.gov
/development/commercial/90.1_models (accessed October 15, 2015).
Wang, Z. 1996. Manual versus Automated Line Generalization. ESRI* White Paper Series
94–106
Wang, Z., Zhao, Z., Lin, B., Zhu, Y., and Ouyang, Q. 2015. Residential heating energy
consumption modeling through a bottom-up approach for China’s hot summer–cold
winter climatic region. Energy and Buildings 109: 65–74.
Ward, R., and Choudhard, R. 2014. A bottom-up energy analysis across a diverse urban
building portfolio: Retrofits for the buildings at the Royal Botanic Gardens, Kew, UK.
Building and Environments 74: 132–148.
Weng, Q. 2010. Remote Sensing and GIS Integration Theories, Methods, and Applications.
McGraw-Hill.
Weng, Q. 2015. Remote Sensing for Urbanization in Tropical and Subtropical Regions—Why
and What Matters. CRC Press.
Zhou, Y., Weng, Q., Gurney, K., Shuai, Y., and Hu, X. 2011. Estimation of the relationship
between remotely sensed anthropogenic heat discharge and building energy use. ISPRS
Journal of Photogrammetry and Remote Sensing 67: 65–72.
Index
Note: Page numbers ending in “f” refer to figures. Page numbers ending in “t” refer to tables.

A C
Advanced Spaceborne Thermal Emission and China
Reflectance Radiometer (ASTER), electricity consumption in, 280t, 283–293,
138, 211, 308 285t, 286f, 287f, 293f
Advanced Very High Resolution Radiometer energy use in, 277–294
(AVHRR), 104, 111, 197, 211, 304 urban expansion in, 292f, 293–294
Aerosol models, 175, 200–201 Chlorophyll estimation, 177, 178t, 183–184, 183f
Aerosol optical depth (AOD), 195–204, 196t Chlorophyll maps, 107–110, 109f
Aerosol scatterings, 174–176 Climatic gradients
Agricultural bioenergy potentials, 271–272, 271f arid climatic gradients, 229–240
Agricultural cultivation patterns, 272, 273f ecosystem transitions and, 229–233
Airborne LiDAR, 16, 18, 21, 116, 125, 128 erosion, 237–240, 238f, 239f
Air quality studies, 193–205 of Judean Mountains, 230–233, 230f, 240
Arid climatic gradients, 229–240; see also life-form compositions, 235–237, 236f, 240
Climatic gradients remote sensing techniques, 231–240
Atmospheric correction algorithms, 174–176, semiarid regions, 229–235, 230f, 246
174f vegetation cover, 231–235, 233t, 234f
Clustering accuracy, 46, 55–57, 56f
B Coal mining, 61–76; see also Land subsidence
Community capital, 4–5
Belleville, Illinois Community sustainability, 4–6
deformation comparisons, 73–74, 73f Compactness of cities, 21–22
land subsidence monitoring, 65–71
time series deformation, 67–68, 69f–70f D
Biodiversity, 13, 17–23, 104, 115, 122
Bioenergy, 270–272, 271f Defense Meteorological Satellite Program/
Biophysical composition index (BCI) Operational Linescan System (DMSP/
calculation of, 49 OLS), 45, 278–283, 286, 294
land surface temperature and, 46, 49–51 Deforestation, 22–23
mapping procedures, 46 Digital elevation model (DEM), 16
stable time series and, 50 Disease damage, 145–158, 146f, 148t–149t;
Biosphere Energy Transfer Hydrology (BETHY/ see also Forest damage
DLR) model, 270–271
Buildings E
3D city model, 322–323, 326–327, 327f
attributes of, 317–320, 327–349 Earth observation (EO)
case studies, 321–349 of air quality, 193–205
correlation analysis, 329–348, 330f–345f, in spatial planning, 29–42, 34f
346t, 347f, 348t of sustainable energy, 269–275
energy use of, 317–320, 327–349, 330f–345f, in urban planning, 29–42
346t, 347f, 348t wind resource modeling, 297–312
energy use simulation, 320, 328–329, 328f Ecosystem transitions, 229–233
fragmentation, 19–20 Electricity consumption (EC); see also
impervious surfaces, 18–20 Energy use
in Indianapolis, 318–349 in China, 280t, 283–293, 285t, 286f, 287f,
reconstruction of, 322–323, 322f, 323f 293f
solar energy potential, 318–319 estimation of, 278–279, 282–283, 285t, 286f,
step-edge detection of, 322–323, 323f 294

351
352 Index

in Indianapolis, 328f, 329–331, 330f, 331f, remote sensing of, 145–158, 146f, 148t–149t
334f–345f, 346t, 347f, 348t spatial characteristics of, 151
at multiple scales, 287–288, 287f, 288f, 290f, spatial patterns of, 155
291f spectral characteristics of, 150–151
spatiotemporal pattern of, 283–290 temporal characteristics of, 151–152
Endangered species Forestry bioenergy potentials, 271, 271f
fine spatial resolution imagery, 121–141 Fragmentation, 15, 19–20; see also Open spaces
grassland management and, 104–105, Fuel consumption, 318, 328f, 329–330,
112–114, 114f 330f–345f, 346t, 347f, 348t; see also
habitats for, 121–141 Energy use
in Hawaii, 121–141
study areas, 124–125 G
tree species and, 121–141
Energy policy, 269–270 Gap filling method, 48–50
Energy sector, 270–272 Geographic information system (GIS)
Energy supply technology, 6, 20, 125, 318–322,
agricultural cultivation patterns, 272, 273f 329–330
bioenergy, 270–272, 271f Geostationary Operational Environmental
energy policy, 269–270 Satellite Aerosol/Smoke Product
geodata for, 270–272 (GASP), 195–196, 196t
heating networks, 272–275, 274f Geostationary Operational Environmental
satellite sensors and, 269–275 Satellite (GOES), 195–197, 212–217,
sustainable energy supply, 269–275 222, 226
Energy use Global chemistry and transport model (CTM), 200
annual consumption, 329–331, 330f–345f, Global ozone chemistry aerosol radiation and
346t, 347f, 348t transport (GOCART) model, 200
building attributes and, 317–320, 327–349 Grassland health
in China, 277–294 assessment of, 107–112
correlation analysis, 329–348, 330f–345f, biophysical properties, 107
346t, 347f, 348t chlorophyll maps, 107–110, 109f
data sets for, 279–280, 280t climatic factors affecting, 111–112, 112f, 113f
decadal trends in, 277–294 disturbance of, 110–111, 110f, 111f
in Indianapolis, 317–349, 328f, 330f–345f, indicators of, 107–110
346t, 347f, 348t leaf area index, 105, 107, 108f, 116
methodology for, 281–285, 281f mapping habitats, 112–114, 114f
nighttime light imagery in, 277–294 Grassland management
results of, 285–292 case studies, 103–117
simulation of, 320, 328–329, 328f challenges in, 115–117
solar energy potential, 317–349 climatic factors, 111–112, 112f, 113f
study area, 279–280 grassland disturbance of, 110–111, 110f, 111f
Environmental degradation, 4–5 habitats for endangered species, 104–105,
Erosion, 237–240, 238f, 239f 112–114, 114f
Evapotranspiration (ET), 111, 212–213, 216, 226, health assessment, 107–112
243, 245, 257 health indicators, 107–110
image acquisition for, 115
F image processing for, 115
logistics for, 116–117
Forest damage monitoring changes, 104–105
challenges concerning, 156–157 prairie grasses, 103–117, 106f
compound disturbances, 157–158 remote sensing for, 103–117
consistent monitoring of, 157 snow mapping, 116
by diseases, 145–158, 146f, 148t–149t soil moisture mapping, 116
early warning of, 156 species-level monitoring, 105, 115
by insects, 145–158, 146f, 148t–149t topography mapping, 116
monitoring of, 145–158 vegetation mapping, 114–116
remote sensing algorithms for, 152–156 Gravimetric measurements, 244, 248–249, 256f,
remote sensing characteristics for, 149–152 259–260, 260f
Index 353

Greensboro, North Carolina Impervious surfaces

land changes in, 90f, 91f assessment of, 30–33, 33t
land cover classification for, 87t–89t diversity of, 18
urbanization in, 81–98 fragmentation and, 19–20
urbanization process analysis of, 92–94 heat hazards and, 217, 220–226, 221f
urban sustainability and, 83f land surface temperature and, 217, 220–226,
Green vegetation cover (GVC), 230–237, 233t, 221f
234f, 240; see also Vegetation cover latent flux over, 222–225, 224f, 225f
Ground heat flux, 218–220, 219t, 308; see also measurement of, 18–19, 30–33, 33t, 222
Heat hazards percentage of, 30–31, 32f, 33t, 215–217
Guiyang, Guizhou, China for urban planning, 30–33
land changes in, 90f, 91f in urban sustainability, 18–20
land cover classification for, 85–89, 86t Indianapolis, Indiana
urbanization in, 81–98, 83f 3D city model, 322–323, 326–327, 327f
urbanization process analysis of, 91–92 building reconstruction, 322–323, 322f, 323f
case studies, 321–349
H correlation analysis in, 329–348, 330f–345f,
346t, 347f, 348t
Habitats; see also Grassland management energy use in, 317–349, 328f, 330f–345f, 346t,
for endangered species, 104–105, 112–114, 347f, 348t
114f, 121–141 energy use simulation, 320, 328–329, 328f
in Hawaii, 121–141 heat hazards and, 213–216, 214f, 222, 226
mapping, 112–114, 114f land surface temperature and, 213–216, 214f,
Hawaii tree species, 121–141; see also Tree 222, 226
species solar energy potential in, 317–349
Heat hazards study area, 321–327
arid climatic gradient and, 229–240 Industrialization, 81–84
data preparation, 217 InSAR monitoring
data processing for, 216–219 of land subsidence, 61–76
data sets for, 215–216 methods of, 61–69
evapotranspiration estimation, 212–213, 216 study area, 63–71, 64f
ground heat flux, 218–220, 219t, 308 for urban planning, 61–76
impervious surfaces and, 217, 220–226, validation of, 73–74, 73f
221f Insect damage, 145–158, 146f, 148t–149t;
land surface temperature and, 211–226 see also Forest damage
linear spectral mixture analysis, 215
methodology flowchart, 216, 216f J
moisture model for, 218–220
Judean Mountains, 230–233, 230f, 240
monitoring, 211–226
results on, 220–226
soil moisture and, 222 K
soil surface and, 220–226, 221f Kernel fuzzy C-means (KFCM), 46, 48, 50–52
study area of, 213–216, 214f, 222, 226
thermal infrared images, 211–213, 215–216 L
vegetation cover and, 220–226, 221f
Heating networks, 272–275, 274f Land cover
Horizontal expansion, 94–95, 95f changes in, 22–23
deforestation of, 22–23
I temporal patterns of, 89
urban sustainability and, 22–23
IKONOS value ranges for, 171, 173t
characteristics of, 123–124, 124t wind resource modeling and, 299–303, 300f,
classification accuracy, 130t, 131f, 133f–136f, 301f, 302t, 305–308, 305f, 306f
138–140 Landsat imagery
for classifying tree species, 123–140, 123f for forest monitoring, 153, 157
comparisons of, 131f, 133f–136f, 139–140 for grassland management, 110–111, 110f,
scatterplots, 129, 132f 111f, 115
354 Index

for heat hazards monitoring, 211, 230, 238 Line-of-sight (LOS) displacement, 65–69, 75
of impervious surfaces, 30–31, 216–217 Line-of-sight (LOS) velocity, 65–69, 68f, 72f
time series imagery, 45–57
of urban growth, 45–57 M
Land subsidence
InSAR monitoring of, 61–76 Mediterranean regions, 229–240
PSI processing, 62, 65–69 Microwave remote sensing, 244–245
SBAS processing, 62, 65–69 Moderate Resolution Imaging Spectroradiometer
spatial patterns of, 67–72 (MODIS)
temporal patterns of, 67–72 aerosol optical depth and, 195–197, 196t,
underground mines and, 61–76 201–204
for urban sustainability, 61–76 of cloud coverage, 211
Land surface parameters fire product and, 199–200
2D land cover mapping, 299, 305–308 thermal infrared images, 215–216
3D topography characterization, 299, of water quality, 166–167, 173–176
308–310, 309f Modified NDWI (MNDWI), 170f, 171–173, 172f,
4D surface dynamics mapping, 299, 310–311 173t
land cover and, 299–308, 300f, 301f, 302t, Multiangle Imaging SpectroRadiometer (MISR),
305f, 306f 195, 196t, 200–201
surface morphology, 303–304 Multi-Angle Implementation of Atmospheric
surface roughness, 303–304 Correction (MAIAC), 195, 196t,
topography mapping, 303–304 197–199, 202
wind resource modeling and, 299–311
Land surface temperature (LST) N
biophysical composition index and, 46, 49–51
calculation of, 49 Nighttime light (NTL) imagery
data preparation, 217 background on, 278–279
data processing for, 216–219 data sets for, 279–280
downscaling, 217, 218f energy use and, 277–294
evapotranspiration estimation, 212–213, 216 methodology for, 281–285
ground heat flux, 218–220, 219t, 308 results of, 285–292
heat hazards and, 211–226 study area, 279–280
hourly LST, 216, 222–225, 223f, 225f, 226 NIR–red spectral reflectance space, 244–246,
impervious surfaces and, 217, 220–226, 221f 249–250, 250f, 255
instantaneous LST, 222, 224f Normalized difference vegetation index (NDVI)
latent flux of, 222–225, 224f, 225f climatic gradients and, 111–113, 111f, 112f, 113f
linear spectral mixture analysis, 215 for forest damage, 150–155
mapping procedures, 46 for heat hazards, 212–216
methodology flowchart, 216, 216f for impervious surfaces, 30–31
moisture model for, 218–220 nighttime light imagery and, 281
monitoring, 211–226 for soil moisture, 245–249, 252–262, 253f,
results on, 220–226 254f
soil moisture and, 222, 244–246, 252–257, for wind resource modeling, 304–307
262 Normalized difference water index (NDWI),
soil surface and, 220–226, 221f 169–173, 170f, 172f, 173t
spatial variability of, 222, 223f
study area of, 213–216, 214f, 222, 226 O
temporal variability of, 222, 223f
thermal infrared images, 211–213, 215–216 Object-based image analysis (OBIA), 139–140
vegetation cover and, 220–226, 221f Open spaces
Leaf area index (LAI), 105, 107, 108f, 116, fragmentation of, 15, 19–20
270–271 impervious surfaces, 18–20
LiDAR systems, 16, 18, 21, 116, 200, 318–324, in urban sustainability, 18–20
328–329 Optical remote sensing
Life-form compositions, 235–237, 236f, 240 description of, 13–16
Linear spectral mixture analysis (LSMA), ground measurements, 243–262
215–216 of soil moisture, 243–262
Index 355

P parameters for, 3–24

satellite sensors, 13–16, 14t
Pakistan spatial resolution, 13, 15–18
case study in, 243–262 spectral resolution, 15, 18
ground measurements in, 243–262 systems for, 13–16, 14t
optical remote sensing in, 243–262 temporal resolution, 15
study area of, 246, 247f for urban sustainability, 3–24
Particular matter for water quality monitoring, 163–184, 165t
aerosol optical depth data, 195–204, 196t Remote sensing soil moisture model (RSSMM),
annual mean surface data, 204–205, 204f 246, 249–255, 256f, 256t, 257t, 262
applications for, 195–200 Remote sensing technology
case study, 203–205 accessibility of, 39–42
concentration prediction of, 195–199 applications of, 6, 16–17, 17t, 21–24, 39–40,
fire data, 199–200 104–107, 140, 150
health effects from, 193–194 availability of, 39–42
LIDAR data, 200 case studies, 103–117
methods for, 200–203 climatic gradients and, 231–240
remote sensing for, 195–200 future of, 42
Pearl River Delta, China in grassland management, 103–117
case study, 46–57, 47f implementation of, 6
mapping procedures, 48–52, 48f measurements by, 16–17
urban growth in, 45–57, 53f, 54f
Persistent scatterer interferometry (PSI)
processing, 62, 65–69 S
Photovoltaic (PV) modules, 318, 321–322, Satellite-based remote sensing, 269–275
326–327, 327f, 348 Satellite-derived land surface temperature,
Pixel-based classification, 139–140 211–226; see also Land surface
PM2.5 temperature
aerosol optical depth data, 195–204, 196t Satellite sensors; see also Remote sensing data
annual mean surface data, 204–205, 204f aerosol scatterings, 174–176
applications for, 195–200 atmospheric correction for, 174–176, 174f,
case study, 203–205 182–183, 183f
concentration prediction of, 195–199 for monitoring water quality, 163–184, 165t
fire data, 199–200 ocean-color sensors, 164–169
health effects from, 193–194 spectral bands of, 166–167, 166f
LIDAR data, 200 spectral reflectance patterns, 171, 173f
methods for, 200–203 systems for, 13–16, 14t
remote sensing for, 195–200 TOA reflectance, 174–176, 174f
Prairie ecosystem water-leaving reflectance, 174–176, 174f
biophysical properties, 107 water quality parameters, 177–182
grassland management, 103–117 Semiarid regions, 229–235, 230f, 246; see also
mixed-grass prairies, 103–115, 106f Climatic gradients
remote sensing in, 103–117 Small baseline subset (SBAS) processing, 62,
spatial patterns of, 107 65–69
Smoothing method, 49–50
R Soil moisture
evaporation of, 222
Remote sensing algorithms extraction of, 250, 251f
change detection algorithm, 155–156 field sample collections, 248–249, 248f
classification algorithm, 153–154 gravimetric measurements, 244, 248–249,
statistical regression algorithm, 154–155 256f, 259–260, 260f
thresholding algorithm, 152–153 heat hazards and, 222
Remote sensing data land surface temperature and, 222, 244–246,
application of, 13–23 252–257, 262
availability of, 13–16 modeling sketch of, 251f
for detecting forest damage, 145–158 NIR–red spectral reflectance space, 244–246,
optical remote sensing, 13–16, 243–262 249–250, 250f, 255
356 Index

normalized difference vegetation index, Sustainable energy supply

245–249, 252–262, 253f, 254f agricultural cultivation patterns, 272, 273f
optical remote sensing of, 243–262 bioenergy, 270–272, 271f
Pakistan case study, 243–262 Earth observation of, 269–275
remote sensing data, 246–248 energy policy, 269–270
remote sensing soil moisture model, 246, energy potential, 271–272, 271f
249–255, 256f, 257t, 262 energy sector, 270–272
spatial variability of, 244–246, 257–262, 261f geodata for, 270–272
study area, 246, 247f heating networks, 272–275, 274f
temperature vegetation dryness index and, implementation of, 269–275
245–246, 252–262, 253f, 257t, 258f, satellite sensors and, 269–275
260f, 261f Synthetic aperture radar (SAR), 16, 18, 35, 37f,
temporal variability of, 244–245, 262 41, 62–66
Soil moisture index, 245–249, 252–262, 253f,
254f T
Soil moisture model, 218–220, 244–246, 251f
Soil surface Temperature vegetation dryness index (TVDI),
heat hazards and, 220–226, 221f 245–246, 252–262, 253f, 257t, 258f,
land surface temperature and, 220–226, 221f 260f, 261f
latent flux over, 222–225, 224f, 225f Temporal patterns, 67–72, 89
Solar energy potential Temporal resolution, 15, 46, 57, 104–105
assessment of, 317–349 Thermal infrared (TIR) images, 211–213,
estimation of, 318–319, 321–327 215–216, 244–247
in Indianapolis, 317–349, 326f Time series clustering, 46, 56f, 57
photovoltaic modules, 318, 321–322, 326–327, Time series Landsat imagery, 45–57
327f, 348 Top of atmosphere (TOA) reflectance, 174–176,
Solar irradiance, 212, 217, 319, 326–327, 326f 174f
Spatial network, 39, 40f Tree species
Spatial patterns classification accuracy, 127–129, 131f, 138–140
of disease, 147 classification comparisons, 137–140
of forest disturbances, 155 classification methods, 124–129, 130t
of land subsidence, 67–72 classification of, 121–141
of prairie ecosystem, 107 discriminant analysis of, 124–128, 137–138, 140
Spatial planning, 29–42, 34f endangered bird species and, 121–141
Spatial resolution, 13, 15–18, 31–33, 41–42 examples of, 125, 126f
Spectral resolution, 15, 18, 41–42 field data on, 125
Stable time series, 50, 52 fine spatial resolution imagery, 121–141
St. Louis, Missouri object-based classification, 132f, 137f, 139–140
land subsidence monitoring, 63–71, 64f, 72f pixel-based classification, 132f–136f, 137f,
line-of-sight displacement, 65–69, 75 139–140
line-of-sight velocity, 65–69, 68f, 72f random forest algorithm, 124, 127–128,
time series deformation, 67–68, 69f–70f, 72f 132–138, 137f, 140
Surface moisture model, 218–220 remote sensing data for, 125
Surfaces, impervious, 18–20; see also study areas, 124–125
Impervious surfaces support vector machine algorithm, 124,
Sustainability 127–128, 137–138, 140
community capital, 4–5
community sustainability, 4–6 U
definition of, 4–5
indicators for, 6–13, 7f, 9t, 16–24, 17t Underground mines, 61–76; see also
Sustainability indicators (SIs) Land subsidence
application of, 8–9 Urban air quality studies, 193–205
criteria for, 8–10, 9t Urban compactness, 21–22
definition of, 6–8 Urban expansion, 292f, 293–294
explanation of, 6–13 Urban growth
hierarchical framework of, 7–8, 7f assessment of, 45–57, 56f
for urban sustainability, 11–13, 16–24, 17t characteristics of, 52–53
Index 357

clustering accuracy, 46, 56f, 57 underground mines and, 61–76

impact of, 45–46, 317–318 urban planning and, 61–76
Landsat imagery of, 45–57
mapping, 33–38, 34f, 36f, 37f, 38t V
mapping procedures, 48–52, 48f
in Pearl River Delta, 46–57, 47f, 48f, 53f, Vegetation cover
54f climatic gradients and, 231–235, 233t, 234f
settlement pattern, 38–39 green vegetation cover, 230–237, 233t, 234f, 240
Urbanization heat hazards and, 220–226, 221f
centralization and, 95–97 land surface temperature and, 220–226, 221f
characteristics of, 81–82 latent flux over, 222–225, 224f, 225f
comparisons, 81–98 Vegetation mapping, 114–116, 122–124, 123f
data collection on, 84–89 Vertical expansion, 94–95, 95f
decentralization and, 95–96 Visible Infrared Imaging Radiometer Suite
development and, 94–97 (VIIRS), 195, 196t, 199
in Greensboro, 81–98, 83f, 87t–89t, 90f, 91f
in Guiyang, 81–98, 83f, 86t, 90f, 91f W
horizontal development, 94–95, 95f
impact of, 45–46, 317–318 Water quality
industrialization and, 81–84 algorithm for, 182–184, 183f
land changes and, 89–91, 90f atmospheric corrections, 174–176, 174f,
land cover classification for, 85–89, 86t–89t 182–183, 183f
mapping, 33–38, 34f chlorophyll estimation, 177, 178t, 183–184, 183f
market driven planning and, 96–97 delineation of, 169–172, 172f
process analysis of, 91–94 monitoring, 163–184
satellite images of, 85–89 optical properties, 181–182
settlement pattern, 38–39 satellite sensors for, 163–184, 165t
study areas, 83–84, 83f TOA reflectance, 174–176, 174f
study methods, 84–89 water delineation, 169–172, 172f
vertical development, 94–95, 95f water-leaving reflectance, 174–176, 174f
Urban planning Water quality parameters
Earth observation in, 29–42 biogeochemical parameters, 177–180, 178t
impervious surfaces in, 30–33 bio-optical model, 177
implications for, 74–75 optical properties, 181–182
land subsidence for, 61–76 from remote sensing data, 177–182, 178t
mapping urban growth, 33–38 Weighted mean patch size (WMPS), 19–20
urban sustainability and, 61–76 Wind farms, 298, 308, 309f
Urban sustainability Wind resource modeling
assessment of, 16–24, 17t background on, 297–298
biodiversity and, 23 Earth observation and, 297–312
coal mines and, 61–76 future options, 297–312
compactness in, 21–22 land cover and, 299–303, 300f, 301f, 302t,
deforestation and, 22–23 305–308, 305f, 306f
environmental indices for, 3–24 land surface parameters, 299–311
explanation of, 3–5 multitemporal aspects of, 304
impervious surfaces in, 18–20 surface morphology, 303–304, 303t
implications for, 74–75 surface roughness, 303–304, 303t
indicators for, 11–13 topography mapping, 303–304, 303t
InSAR monitoring of, 61–76 Wind turbines, 297–298, 311
land cover and, 22–23 WorldView-2 (WV2)
land subsidence for, 61–76 characteristics of, 123–124, 124t
measurements of, 16–24, 17t classification accuracy, 130t, 131f, 133f–136f,
open spaces in, 18–20 138–140
remove sensing data for, 3–24 for classifying tree species, 121–140, 123f
sustainability indicators for, 11–13, 16–24, 17t comparisons of, 131f, 133f–136f, 139–140
township layouts, 21–22 for measuring impervious surfaces, 31, 33t
transport systems in, 20–21 scatterplots, 129, 132f