Journal of

Plant Ecology Satellite remote sensing of

VOLUME 9, NUMBER 6,
PAGES 649–671 grasslands: from observation to
December 2016

doi:10.1093/jpe/rtw005
management
Advance Access publication
2 February 2016

available online at Iftikhar Ali1,4,*, Fiona Cawkwell1, Edward Dwyer3, Brian Barrett4

www.jpe.oxfordjournals.org
and Stuart Green2
1
Department of Geography, University College Cork, Cork, Ireland
2
Spatial Analysis Unit, Teagasc, Dublin, Ireland
3
EurOcean - European Centre for Information on Marine Science and Technology, Lisbon, Portugal
4
School of Geographical and Earth Sciences, University of Glasgow, Glasgow, UK
*Correspondence address. Department of Geography, University College Cork, Cork, Ireland. Tel: +353-(0)21-
4902517; Fax:+353-(0)21-427-1980; E-mail: [email protected]

Abstract

Aims are likely to continue to be the most robust method for deriving grass-
Grasslands are the world’s most extensive terrestrial ecosystem, and land information, however these require more high quality calibration
are a major feed source for livestock. Meeting increasing demand and validation data. We found that the hypertemporal satellite data are
for meat and other dairy products in a sustainable manner is a big widely used for time series generation, and particularly to overcome
challenge. At a field scale, Global Positioning System and ground- cloud contamination issues, but the current low spatial resolution of
based sensor technologies provide promising tools for grassland and these instruments precludes their use for field-scale application in
herd management with high precision. With the growth in avail- many countries. This trend may change with the current rise in launch
ability of spaceborne remote sensing data, it is therefore important of satellite constellations, such as RapidEye, Sentinel-2 and even the
to revisit the relevant methods and applications that can exploit this microsatellites such as those operated by Skybox Imaging. Microwave
imagery. In this article, we have reviewed the (i) current status of imagery has not been widely used for grassland applications, and a
grassland monitoring/observation methods and applications based better understanding of the backscatter behaviour from different phe-
on satellite remote sensing data, (ii) the technological and meth- nological stages is needed for more reliable products in cloudy regions.
odological developments to retrieve different grassland biophysical The development of hyperspectral satellite instrumentation and analyti-
parameters and management characteristics (i.e. degradation, graz- cal methods will help for more detailed discrimination of habitat types,
ing intensity) and (iii) identified the key remaining challenges and and the development of tools for greater end-user operation.
some new upcoming trends for future development.
Keywords: remote sensing, agricultural grassland, grassland
Important Findings
biomass, pasture management, grazing intensity
The retrieval of grassland biophysical parameters have evolved in
recent years from classical regression analysis to more complex, effi- Received: 13 August 2015, Revised: 5 January 2016, Accepted: 8
cient and robust modeling approaches, driven by satellite data, and January 2016

2008; Pokluda et al. 2012; Punjabi et al. 2013; van Swaay 2002).
BACKGROUND From an agricultural perspective, grasslands provide the cheapest
Global grasslands feed source for the livestock industry, however they contribute
Grasslands are one of the most prevalent and widespread land both directly and indirectly to climate change through the emis-
cover vegetation types, covering 31.5% of the global landmass sion of greenhouse gases (FAO 2014). As a result, a restriction
(Latham et al. 2014). After forests, grasslands are the largest ter- on a maximum level of grassland intensification (animal stock-
restrial carbon sink (Anderson 1991; Derner and Schuman 2007) ing) is required in order to minimize the environmental risks
and, as such, they play a vital role in regulating the global car- (Soussana and Lemaire 2014). During the period of 1994–2012,
bon cycle (Franzluebbers 2010; Scurlock and Hall 1998), as global permanent pasture cover declined by approximately 1%
well as supporting plant and animal biodiversity (Bergman et al. from 3395 × 106 to 3359 × 106 ha (FAOSTAT 2014), as a result of

© The Author 2016. Published by Oxford University Press on behalf of the Institute of Botany, Chinese Academy of Sciences and the Botanical Society of China.
All rights reserved. For permissions, please email: [email protected]

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 650 Journal of Plant Ecology

urbanization, overgrazing (Piñeiro et al. 2006b; Han et al. 2008), large scale monitoring, quantification and prediction(Gao 2006)
industrial development (Wang et al. 2008), intensive management of different phenomena (e.g. land use and land cover, biodiversity,
practices and climate change (Thorvaldsson et al. 2004). Grassland impacts of climate change) occurring on the surface of the Earth at
degradation results in increased carbon emissions. It has seri- varying spatial and temporal resolutions(Nordberg and Evertson
ous repercussions for society (Cardinale et al. 2012), and leads to 2003). The integration of multispectral and multitemporal remote
more complex interactions between grassland ecosystems, man- sensing data with local knowledge and simulation models has
agement practices and climate change. These human activities, been successfully demonstrated as a valuable approach to identi-
coupled with unfavorable environmental conditions, are major fying and monitoring a wide variety of agriculturally related char-
causes of changes in the productivity of grasslands (Xu et al. 2008). acteristics (Yiran et al. 2012; Oliver et al. 2010). In the context of
global food security and to avoid food shortages, estimated yield
Definition and distribution of managed
production prior to harvest is needed for planners and decision
grasslands
makers and remote sensing platforms are increasingly recognized
Three distinct categories of managed grasslands are recognized: as essential tools for this task (Boschetti et al. 2007). An early and
accurate indication of a decrease in fodder production is especially
• Human-generated pastures/meadows/grasslands or im-
important for agriculture-dependent developing economies, how-
proved grasslands: These grasslands are typically created
ever, to date, little work has been undertaken on grass-based food
by the conversion of natural landscapes (e.g. forests) into
security. Recently, Svoray et  al. (2013) has published a detailed
pastures or grassland paddocks (Foley et  al. 2005; Hill
review on remote sensing of rangelands, so this review focused
2004). These grasslands are intensively managed in or-
on managed grasslands and pastures for their greater relevance to
der to maximize production (dairying, meat, wool), e.g.
agriculture, livestock and the concept of precision farming from
through regular application of fertilizer, intensive grazing,
space (precision agriculture).
cutting of silage for winter-feeding and reseeding every
few years. Improved grasslands are widely found in parts Objectives and scope of the review
of Northern Europe, New Zealand and Australia.
This review will review the application of satellite remote sens-
• Highly managed natural grasslands: In this category, natu-
ing for grassland and its transition from grassland mapping to
ral grasslands are modified and managed to support inten-
grassland/pasture monitoring and management. The aims of
sive grazing for the livestock industry, e.g. the semi-im-
this review are to examine the extent of satellite remote sens-
proved natural grasslands of eastern Australia, and fescue
ing applications in the field of grasslands and pastures, and to
prairie of Alberta, Canada (Breymeyer 1990; Hill 2004).
identify the contemporary trends and future potential of these
• Rangelands: Based on their species composition, range-
data and methods. The main objectives of this article are:
lands are different from pastures due to the presence of
native herbaceous/shrubby vegetation which are a feed • to provide an overview of satellite remote sensing (optical
source for both domestic and wild herbivores, e.g. tallgrass and microwave) technological and methodological devel-
prairies (e.g. North American Great Plains), steppes, desert opments to retrieve different grassland biophysical param-
shrublands, shrub woodlands and savannas. Management eters and management characteristics
of rangelands is solely through controlling the number of • to identify trends and gaps in the work done to date re-
grazing units and length of the grazing season. sulting in recommendations for future research and op-
erational systems.
Figure  1 gives an overview of grasslands as a proportion of
land cover, with the major managed pastures, grasslands and
rangelands areas (Hill 2004) of the world highlighted. APPROACHES TO GRASSLAND
Grassland monitoring and feasibility of remote MONITORING
sensing technologies Grassland monitoring approaches are broadly categorized into
Grassland monitoring, either through in situ field observation or two groups: (i) ground-based and (ii) remote sensing meth-
remote sensing, requires data on the current status of the grass ods. The term ‘grassland management’ in the context of this
and of the potential offered by the immediate environment, such research includes weed control, removing dead plants, mow-
as soil, weather and human activities. The current status of the ing, clipping, assessment of biomass and growth rate, extent,
grass includes aspects such as sward height, biomass, quality, grazing length and utilization of grassland (incorporating ele-
phenological stage, productivity level, species composition and ments of herd management) (Hybu Cig Cymru 2008).
change in each of these since a previous recording stage (earlier
in the same season or in a previous season). In situ methods, from Ground-based measurements for validation of
visual analysis to techniques such as a rising plate meter, to cutting remotely sensed data
and laboratory analysis, can be extremely informative at a local Ground-based grass monitoring techniques heavily depend
scale, but they are labor intensive and not feasible for large-scale on an infrastructure, which includes in situ data collection sta-
coverage. Remote sensing and modeling approaches allow for tions, measurement devices and frequent field surveys (Del

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 651

Figure 1  : overview of the global extent of pastures/grasslands [Modified from Foley et al. (2005), grey boxes are the major managed pastures,
grasslands and rangelands areas (Hill 2004)].

Pozo et  al. 2006). Current methods used for the retrieval of ence Vegetation Index (NDVI) in order to investigate the
grassland biophysical parameters and other management- within-field variability in biomass and assess the potential
related information include: for the application of NDVI for pasture management activi-
ties. They found that NDVI showed a good correlation with
• Visual: visual assessment by human eye (expert or farmer),
biomass (r2 = 0.68) and with the results from the rising plate
this method is spatially sparse with limited performance
meter (r2 = 0.54), however as noted by Todd et al. (1998),
for different management strategies (Newnham 2010).
possible relationships between such indices and the vegeta-
• Cut and dry (clipping): grass harvested from the paddock
tion biomass are influenced by the ground-based sampling
is dried and weighed to get the dry matter (DM) yield, as
methods, e.g. biomass can be underestimated due to the
well as a laboratory assessment of grass quality and nutri-
presence of non-photosynthetically active plant material.
ent status (Xie et al. 2009).
• Rising plate meter (RPM): both mechanical and electronic Table  1 gives the summary comparison of different ground-
plate meters work on the principle of a plate rising up and based methods.
down the shaft taking measurements of grass height (Cas- While these ground-based methods are very useful for
tle 1976; Hakl et al. 2012; Hejcman et al. 2014). This meth- grassland monitoring on a local scale, and for providing val-
od is most commonly used for accurate biomass and grass ues for model development and calibration of ex situ data,
height estimation at a point but is very time intensive. they are subjective, time consuming and are only feasible (or
• Field spectrometry: reflectance spectra are collected using a applicable) for small scale assessment (Xu et al. 2008).
spectrometer held at waist height and are calibrated against For remote sensing studies, high quality ground truth data
in situ samples, with species discriminated using local field are of great importance for cross validation and algorithm
data or spectral libraries. Based on the reflectance at red and training. All these ground-based methods are equally appli-
near infrared wavelengths, vegetation indices (VIs) are cal- cable for this purpose, and data collected using these methods
culated, from which biophysical parameters such as above have proven very useful. For example, forest inventory, crop
ground biomass and leaf area index can be retrieved (Flynn yield and grassland (Xu et al. 2008) data collected in the past is
et  al. 2008; Psomas et  al. 2011a). Flynn et  al. (2008) used currently being used by the remote sensing scientists for forest
a ground-based sensor to calculate the Normalized Differ- change detection and development of yield estimation models.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 652 Journal of Plant Ecology

Table 1:  comparison among ground-based methods

Methods Scale Benefits Limitations Category

Visual Field/paddock farm Fast and cheap Need specific expertise, vague Non-destructive
estimation
Clipping More accurate than visual assessment Time consuming if large number of Destructive
samples are required
RPM Easy to operate and cheap Time consuming Non-destructive
Field spectrometry Information on other biophysical Trained operator and post-processing Non-destructive
parameters can also be retrieved is required

Remote sensing methods at this wavelength. Hyperspectral remote sensing data, which
As highlighted in the field spectrometry section, measure- records a larger number of wavelength bands, therefore offer
ment of the reflectance at visible and infrared wavelengths the opportunity of defining new VIs that can be tailored to a
can enable discrimination of different grassland species and particular species and/or parameter application (Clevers et al.
status. These principles are equally applicable for local scale 2007).
mapping and monitoring from optical sensors mounted on Although increased spectral resolution offers significant
eddy covariance towers, unmanned aerial vehicles, aircraft benefits to resolving species composition at a single point in
and spaceborne platforms. It is these spaceborne platforms time, it is recognized that a time series of imagery acquired
that can collect data at spatial scales from 25 cm to 1 km, for through the growing season provides maximum information
regional, national and global studies, that are the focus of this on yields and management. Phenological stages of grasslands
review. The last 20–30  years have seen many technological can progress rapidly during the growing season as a function
developments that enable economically cost effective, statisti- of factors including weather, germination, management strat-
cally reliable and consistent, and operationally robust tools for egies, grazing pressure/intensity, hydrological processes and
remote monitoring of grassland sites and acquisition of data nutrient input. Huang and Geiger (2008) demonstrated that
on their behaviour. inclusion of grass phenological stages increased the accuracy of
mapping grass cover, and Butterfield and Malmström (2009)
Optical remote sensing showed that understanding of grassland dynamics could be
Discrimination of different terrestrial ecosystem types and improved through looking at biomass-NDVI relationships
measurement of their productivity primarily relies on VI at different phenological stages. An increased temporal fre-
that combine reflectance values at two or more wavelengths, quency of image acquisition is advantageous in countries with
selected to accentuate particular features of the spectral sig- cloud-dominated climates where multiple overpasses fail to
nature, such as greenness, water content or light use effi- generate an image of the ground. O’Connor et al. (2012) high-
ciency (Song et al. 2013). Given the similar composition, and lighted the benefits offered by a dense time series of 10-day
therefore spectral signature, of many grassland sites, data at composites for mapping spatial variability in vegetation sea-
multiple wavelengths allows more robust characterization of sonality in Ireland, with landcover classes separated on the
grassland species and their biophysical parameters. This has basis of their start of season greening. The benefits of timely
been facilitated by the trend in recent years for satellite sen- imagery are recognized for yield estimation from crops (Morel
sors to record a higher number of carefully selected wave- et al. 2014), and with an increased number of spaceborne sen-
lengths, e.g. the yellow band of Worldview-3is designed to sors available in a constellation, there is an increased potential
detect ripening or dying plants. The red edge, where there to acquire more frequent, cloud-free imagery coincident with
is a rapid increase in reflectance from the red to NIR reflec- key stages in the grass growth season.
tance (around 680–730 nm), has been shown to have a strong There is typically an inverse relationship between the
correlation with the grass chlorophyll content of the canopy frequency of image acquisition and the swath width of the
(r = 0.93) and the leaves (r = 0.86) (Pinar and Curran 1996). sensor and its spatial resolution, which results in the sensors
Inclusion of measurements made in a red-edge channel are that acquire daily images doing so at resolutions of 300–1000
thus a reliable indicator of foliar chlorophyll content and m. While this may be sufficient for large rangeland areas, it is
vegetation stress (Dawson and Curran 1998), and are also often too coarse for imaging intensively managed grasslands,
useful for assessment of plant chlorophyll concentration, and where the pasture paddock size is smaller than the sen-
leaf area index and therefore nutritional status (Filella and sor resolution cell, inconsistency and discrepancies with in situ
Penuelas 1994). With the launch of RapidEye, the first high- data validation arise in averaging and aggregation during up
resolution multispectral satellite system that operationally and down scaling for multi sensor data integration (Hill 2004).
provides a red edge channel, Schuster et al. (2012) reported Due to the small size of many managed agricultural grassland
a higher classification accuracy for managed grassland types paddocks, access to a high spatial resolution imagery is essen-
than could be achieved without inclusion of measurements tial in determining inter- and intra- field variations. Figure 2

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 653

Figure  2:  spatial resolution comparison (false color composite: R  =  NIR, G  =  RED, B  =  GREEN) among QuickBird (A), RapidEye (B) and
Landsat-8 (C) covering a managed natural grassland conservation site in the west of Berlin, Germany (Courtesy: Dr Michael Förster).

shows the false colour composite of a managed grassland area Japanese ALOS and ALOS-2 L-band instruments, and European
where small-scale differences in growth are more evident in Space Agency ASAR and Sentinel-1 C-band platforms have a
the 2.4 m Quickbird image than the 6.5 m RapidEye image, lower spatial resolution but the longer wavelength can be more
and almost impossible to detect in the 30 m Landsat-8 scene. sensitive to vegetation volume, as shown by Barrett et al. (2014)
A number of high and very high-resolution sensors have in discriminating between grassland types in Ireland. A number
been launched in the last 10 years, which enable such intrafield of studies have been undertaken to compare the sensitivity of the
variations to be detected, and when multiple identical instru- different wavelengths to vegetation conditions (Gao et al. 2013;
ments are in a constellation a time series of cloud-free imagery Inoue et al. 2002), with Metz et al. (2012) demonstrating how the
can be maintained. However, the scale of imaging remains most accurate discrimination of European Natura 2000 protected
a very complex and dynamic topic in the context of remote sites and high nature value habitats could be achieved with com-
sensing, with Wu and Li (2009) and Quattrochi and Goodchild bined use of a TerraSAR-X and Radarsat-2 time series. In addition
(1997) providing more detailed discussion on this topic. to using different wavelengths for different applications, the dif-
ferent polarimetric acquisition capabilities can be exploited, e.g.
Microwave remote sensing Voormansik et  al. (2013) used a TerraSAR-X dual polarimetric
The use of optical instruments for vegetation mapping is common SAR time series to detect grassland cutting practices, and Buckley
practice, with a good understanding of the relationship between and Smith (2010) used a combination of multi angle Radarsat-2
reflectance and biophysical information, however it is limited to quad-polarization images, demonstrating improved grassland
periods when the target is illuminated by the sun under cloud- classification results when compared with the individual inci-
free conditions. In recent years, there has been a growing interest dence angles.
in the potential offered by microwave spaceborne instruments However, a number of limitations have constrained the
which measure the strength of the backscattered signal from the work done in the microwave domain, predominantly the dif-
surface under almost all weather and light conditions, allowing ficulty of distinguishing the signal response associated with
frequent repeat measurements throughout the growing season. vegetation cover from moisture and acquisition conditions.
While the number of wavelengths utilized by active microwave The inherent speckle of SAR imagery also requires processing
instruments is relatively limited, synthetic aperture radar (SAR) that reduces the spatial resolution, and thus can lose some of
instruments offer a number of different acquisition modes, with the detail that may be present at the scale at which the image
different polarizations, incidence angles and orbital directions is acquired. To overcome these limitations and derive con-
(ascending/descending). The backscatter signal from vegetated clusive results have typically required intensive ground-based
surfaces is a function of the soil surface, the radar system and the measurements (Moran et al. 1997).
biophysical parameters of the scatterers in the vegetation that can Several studies have been carried out to compare the out-
influence the depth to which the radar wave penetrates. Different puts from optical and microwave instruments. Smith and
theoretical approaches have been developed to interpret the Buckley (2011) did a comparative analysis of Radarsat-2 and
backscatter signal, for example the water cloud model in which Landsat-5 TM for the classification of cultivated crops, summer
the total backscatter signal comprises components from the soil, fallow, improved and native grassland. Even though the clas-
vegetation and attenuation (Attema and Ulaby 1978). A num- sification accuracy for Radarsat-2 (kappa = 0.65) was less than
ber of SAR instruments have been launched during the 21st that for Landsat-5 TM (kappa  =  0.81), due to the backscat-
century that have allowed advancement of microwave remote tering similarities between native and improved grasslands, it
sensing of vegetation phenology, e.g. TerraSAR-X, with a very was able to successfully discriminate between the cultivated
high resolution (up to 1 m) X-band sensor, and the COSMO– crops and grasslands. By contrast, in a recent study Dusseux
SkyMed constellation of four X-band platforms which were used et al. (2014) reported classification results of fully polarimet-
by Hajj et al. (2014) to investigate the sensitivity of radar signals to ric Radarsat-2 (98% accuracy) that outperformed the optical
soil moisture and vegetation within irrigated grassland plots. The imagery (SPOT-5 and Landsat-5 TM, 81% accuracy).

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 654 Journal of Plant Ecology

It is apparent that there have been many developments in object and pixel classification approaches for classifying
the use of remote sensing for vegetation monitoring, map- Buffel grass in Mexico from satellite imagery, and found that
ping and management in recent years, with a number of determining objects on the basis of their contiguity allowed
reviews dedicated to specific aspects of agricultural and eco- for more accurate results. Decision trees permit data from dif-
system practices (Atzberger 2013; Shoshany et al. 2013). In an ferent sources to be included to aid distinguishing between
early review paper, Tappan (1982) highlighted some topics for grassland classes and also to preclude some misclassification
future research using remote sensing for grassland applica- opportunities, as Dubinin et al. (2010) showed with a multi-
tions, e.g. biomass estimation, instrument calibration and use sensor approach to assess annual burned areas in the grass-
of high spatial and temporal resolution satellite platforms. To lands of southern Russia, and Wang et al. (2010) discriminated
date, however, available reviews on grasslands have focused between warm and cold season grasslands in the USA from
either on a site-specific approach (Trotter 2013), or on just ASTER data with an overall accuracy of 80%. Peña-Barragán
classification and mapping of grasslands (Booth and Tueller et al. (2011) developed a hybrid classification strategy, com-
2003; Svoray et al. 2013; Xie et al. 2008). The following review bining object-based image analysis with a decision tree (DT)
broadens this focus to address some of the issues raised by including information on textural features and phenology,
Tappan (1982) on spaceborne remote sensing within grass- to classify ASTER imagery of California. While some of the
land environments, and the transition from grassland classifi- 13 classes were very reliably classified with accuracies of
cation/mapping to grassland management. 95%, others remained problematic with only a 50% chance
of being correctly labelled. A  hybrid classification approach
was also adopted by Masocha and Skidmore (2011) to map
REMOTE SENSING OF MANAGED an invasive species in part of southern Zimbabwe. Artificial
neural network (ANN) and support vector machine (SVM)
GRASSLANDS AND PASTURES approaches gave accuracies of 71% and 64%, respectively,
Classification but after incorporating the information from a GIS expert
The motivation for grassland mapping includes distinguish- system the accuracies increased to 83% and 76%, respec-
ing different grassland ecologies that may reflect management tively. In addition to mapping different grassland ecologies
practices, grassland degradation and estimation of grassland or species, classification approaches have also been used to
productivity trends over time. Data (and/or derived products) assess grassland use intensity and to monitor changes over
from Landsat TM/MSS, SPOT, AVHRR, MODIS and RapidEye time. Tovar et  al. (2013) used object-based classification of
sensors amongst others have been most commonly used for Landsat imagery of Peru to analyze trends in land use and
the purpose of land cover classification and land cover change land cover from 1987 to 2007, with an overall accuracy of
mapping, including grass-based habitats such as rangelands, 80.3%, showing an annual decrease in the spatial extent of
pastures and meadows. Many of the studies have been under- the Jalca grasslands of 1.5%.
taken using optical rather than SAR sensors, which reflects Many grassland studies have been conducted at a local
their longer history of operation, the importance of the red scale using high spatial resolution imagery, but the same
and NIR bands for vegetation discrimination, and the avail- methods can be applied to a national or regional scale using
ability of data at a range of resolutions, including submeter for coarser spaceborne imagery (e.g. MODIS). In a recent study,
field scale work and 1 km for global mapping. Nitze et al. (2015) established the value and consistency of a
Discriminating between grassland types is usually achieved machine learning algorithm for the classification of improved
using either statistical, object-oriented or machine learning and semi-improved grasslands in Irelandfrom a 9-year MODIS
classification approaches. The maximum likelihood classifi- time series of NDVI and enhanced vegetation index (EVI) VIs.
cation approach was widely used until the 1990s, with typi- In order to optimize the data acquisition period, the impor-
cal overall classification accuracies in the range 70–90%. For tance of different features was considered in this study, with
example, Toivonen and Luoto (2003) mapped grasslands in the authors concluding that to achieve an accuracy of more
Finland from Landsat data with an overall accuracy of 89%, than 90%, only 6–10 images are required per year.
although the classification accuracy was as low as 63% for In general, optical sensors have been preferred to SAR
the semi-natural grassland class. Similarly, Jadhav et  al. sensors for classification of grasslands, exploiting the mul-
(1993) achieved an overall accuracy for grassland mapping tispectral information acquired at the shorter wavelengths.
in India of 82%, and Baldi et al. (2006) distinguished South For example, Price et al. (2002) conducted a comprehensive
American grasslands with accuracies of 90–95%. While some study to compare the use of Landsat TM and ERS-2 C-band
studies using these statistical classifiers performed very well, SAR data in order to discriminate different grassland types
in general the complexity of grasslands and the spectral simi- under different treatments in eastern Kansas. In this study,
larity of different grassland types limits the value of these Landsat TM and ERS-2 were used to discriminate between
approaches. Furthermore, these statistical approaches have a the cold and warm season grass species, with discriminant
limited capability to determine boundaries between different analysis showing that both types can be distinguished, with
natural grassland ecologies. Brenner et al. (2012) compared an accuracy of 90.1% using Landsat TM data, but only 73.2%

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 655

using ERS-2 SAR data. Three management strategies were VIs. Barrett et  al. (2014) highlighted the value of machine
also classified, with an accuracy of 70.4% (Landsat TM) and learning classifiers for discriminating different grassland types
39.4% (ERS-2 SAR). The last step in this study was the com- using multisensor C and L-band SAR data.
bined use of Landsat TM and ERS-2 SAR data, and it was In summary, classification of grassland types and forma-
found that the SAR contribution to the discrimination of the tions using satellite remote sensing data has been success-
grassland types was statistically significant. In another study, fully applied using different classifiers and sensors in different
Smith and Buckley (2011) used Radarsat-2 C-band polari- regions of the world. Table 2 highlights a number of studies
metric SAR data in order to discriminate improved grasslands, that have been done since 2000 using spaceborne remote sens-
native grasslands and agriculture crops, and again Radarsat-2 ing data for mapping different aspects of grasslands around
classification results were less accurate than the Landsat TM the world. The majority of these studies are from optical sen-
(Kappa coefficient: Radarsat-2  =  0.65, Landsat TM  =  0.81). sors, emphasizing their suitability for vegetation mapping and
Interestingly however, the latest generation of high resolution the availability of high resolution optical data (Franke et  al.
SAR sensors, such as TerraSAR-Xand ALOS-2, show greater 2012), as well as a good understanding of the relationships
potentialfor information retrieval from grassland pastures at between the data and biophysical plant parameters.
smaller scales, allowing changes in surface roughness and
moisture, typical of different grassland regimes, to be better Biomass estimation
detected. Wang et al. (2013) compared satellite imagery from Gao (2006) addressed the difficulties and importance of
three different SAR (X, C and L-band) sensors and showed remote sensing-based quantification of grassland properties.
that X-band SAR data has the highest correlation with the For example, (i) the date of image acquisition and ground

Table 2:  grassland mapping/classification using satellite remote sensing data (examples from literature are grouped according to the
classifiers used)

Classifiers Examples Advantages Disadvantages

Unsupervised Gu et al. (2013); Wen et al. (2010) It is simple and easy to implement. Does not consider the spatial
Training (prior knowledge) data is relationships in the data and spectral
not required for classification. It is classes do not represent the on
unbiased, as clustering is purely based ground features. Post-classification
on pixel values interpretation can be very time
consuming
Maximum likelihood Baldi et al. (2006); Miehe et al. (2011); Until recently it was the most In the case of large data sets
Reiche et al. (2012); Toivonen and Luoto popular and widely used supervised classification is extremely slow.
(2003); Weiers et al. (2004) classification approach. The pixels are Classification results can be biased
classified based on their probability for small training samples. Normally
of belonging to a class and if the distributed data assumption is not
probabilities are not same for each always true, and this might result in
class ‘weight factors’ can be specified. misclassification
It is accurate for normally distributed
datasets and considers variability in
the data
Object based classification Brenner et al. (2012); Franke et al. (2012); It can utilize the spatial information High computational cost. Accuracy
Peña-Barragán et al. (2011); Tovar et al. (i.e. shape, size, colour, compactness) depends on segmentation process for
(2013) of high resolution data and provide example scale selection, which is not
high accuracy well defined
Principal component analysis Hill et al. (2005, 1999) Reduces the data dimensionality and Assumes multi-temporal data
enhances the key features in the are highly correlated, and makes
data. The new ‘components’ might very strong assumptions that the
detect the variations/changes directions with the largest variance
contain most of the information
Decision tree Dubinin et al. (2010); Peña-Barragán et al. Simple to understand and to Decision-tree learners can create
(2011); Wang et al. (2010); Wen et al. (2010) interpret. Trees can be visualized. over-complex trees that do not
Requires little data preparation. Fast generalize the data well and trees can
and able to handle both numerical be biased if some classes dominate
and categorical data
Machine learning Filippi and Jensen (2006); Lawrence et al. Often much more accurate than Data-driven methods need a lot of
(2004); Masocha and Skidmore (2011) human-crafted rules as they are data labelled data, requiring extensive
driven. Automatic method to search ground truth datasets. Typically
for hypotheses explaining data. require some programming
Flexible and can be applied to any knowledge
learning task. Rich interplay between
theory and practice, with improved
results as datasets increase

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 656 Journal of Plant Ecology

truth collection must be the same or very close to each other, et al. (2003) found a high correlation (r2 = 0.74) between NDVI,
(ii) samples must be selected randomly, (iii) a sufficient num- derived from Landsat TM and field spectrometer measurements,
ber of samples (at least 30) is needed, (iv) the use of Global and the percentage of grass cover. By contrast, An et al. (2013)
Positioning System during ground truth collection so that in used biweekly AVHRR NDVI values to predict above ground net
situ measurements and corresponding pixels correctly overlie primary production (ANPP) in a tall grass prairie system, but
each other and (v) if the grassland is highly dynamic then their model, validated by in situ measurements, was less able to
high temporal resolution satellite time series should be used predict year-to-year ANPP variations (r2 = 0.54), with the coarse
instead of a single image. Methods for remote sensing of grass resolution (1 km), and thus the influence of mixed pixels, a
yield estimation can be broadly grouped into three strategies: possible explanation for this low value of coefficient of deter-
development of yield estimation regression models based on mination. As plant phenology is highly influenced by interan-
different satellite driven VIs, use of different machine learn- nual changes in temperature and precipitation, Lee et al. (2002)
ing algorithms (e.g. ANN, SVM), and combined use of remote investigated the influence of climatic variation on plant phe-
sensing driven vegetation parameters and biophysical simula- nology in Inner Mongolia by analysing a 9-year (1982–1990)
tion models (e.g. WOFOST, Lingra). AVHRR NDVI time series and monthly mean temperature and
precipitation. However, they reported little or no change in
Vegetation index-based regression models phenological response during this period, which could again be
Remote sensing of biomass estimation has been undertaken attributed to the low spatial resolution of the imagery.
for many years, and numerous studies show a good correla- A major challenge in the use of VIs to assess vegetation
tion between in situ measurements and VIs derived from satel- parameters is to minimize the influence of external factors
lite data (Anderson et al. 1993; Wylie et al. 1991). Boschetti et al. and to maximize the sensitivity of the relationship between
(2007) assessed pasture production in an alpine region using VIs and biophysical parameters. Many authors have tried to
field spectrometry and Landsat-7 imagery, with integration of find the most suitable subset of VIs (e.g. those for best estima-
these data, via regression analysis, supporting assessment of tion of biomass for a particular type of vegetation), with some
pasture production. Ullah et  al. (2012) used MERIS data and advocating a move away from the index-based approach.
analysed different VIs for the estimation of grassland biomass in Even though many researchers have established significant
the northern Netherlands, where NBDI [normalized band depth relationships between VIs and vegetation parameters in the
index (Mutanga and Skidmore 2004)] produced better results context of a single study, many such models are site or sea-
than the more conventional VIs [NDVI, soil-adjusted vegetation son specific, and the successful transferability from one site
index SAVI, and Transformed SAVI (TSAVI)]. Xu et  al. (2008) to another is variable. Based on the combined use of field
tested three different regression models using MODIS-derived spectroradiometer data and satellite driven indices, Boschetti
NDVI and ground measurements of grass yield for the estima- et al. (2007) concluded that log-transformed regression analy-
tion of grass production in China, where more than 8000 sam- sis between soil-adjusted VIs and fresh biomass show higher
ples were collected from 17 grassland dominant provinces and correlation than a ratio vegetation index or NDVI. Likewise,
regions, with the best correlation shown for an exponential Ullah et al. (2012) showed that band depth analysis outper-
relationship (linear r2 = 0.671, power r2 = 0.794 and exponen- formed the use of traditional VIs when they modeled veg-
tial r2  =  0.805). In the north-eastern province of China, Zha etation parameters and spectral values by simple linear

Table 3:  grassland yield estimation using satellite remote sensing data (examples from literature are grouped according to the models/
methods applied)

Models/methods Sensor Examples

Linear regression Landsat TM/MSS/ETM+, IRS, SPOT VEGETATION, SPOT 4/5, Bradford et al. (2005); Han (2001); He et al. (2009);
Hyperion, NOAA/AVHRR Kurtz et al. (2010); Loris and Damiano (2006); Prince
(1991); Psomas et al. (2011b); Verbesselt et al. (2006);
Williamson and Eldridge (1993); Wylie et al. (2002)
Exponential regression Landsat TM, MODIS Huang et al. (2013); Xu et al. (2007); Xu et al. (2008)
Optimal regression model MODIS, Landsat TM, NOAA/AVHRR Jianlong et al. (1998); Yu et al. (2010)
Power regression MODIS Xu et al. (2007); Xu et al. (2008)
Logrithmic regression ERS-SAR, IRS, SPOT-5 Vescovo and Gianelle (2008); Moreau and Le Toan
(2003)
Advantages The principal advantage of empirical modelling is its simplicity, availability, interpretability and acceptance among
the scientific community
Disadvantages In nonlinear dynamic environment, the data from chaotic systems do not correspond to the strong assumptions
of a linear model. These models do not have a physical basis and mostly used for site-specific analysis or model
development

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 657

regression and stepwise multiple linear regression (MLR), and data over different spatial scales for yield estimation (Hansen
continuum removed spectra—normalized reflectance spectra and Jones 2000). Some authors (Brilli et al. 2013; Maselli et al.
used to compare individual absorption features—were used to 2006, 2013) have explored the potential application of the
calculate band depth parameters. parametric model C-Fix, a Monteith type parametric model
Table  3 presents a summary of several studies conducted driven by temperature, radiation and fraction of Absorbed
since 1990 on grass yield estimation derived using vegeta- Photosynthetically Active Radiation (fAPAR), for the estima-
tion index-based approaches, with many of the better results tion of gross primary productivity of grasslands, olive groves
achieved at a local to regional scale. and forests in Italy. Parameters derived from satellite data
and ground measurements are combined in order to simulate
Machine learning models the total production. Maselli et al. (2013) compared the effi-
ANN models belong to a powerful class of empirical modeling ciency of C-Fix and the BIOME-BGC biogeochemical model
with the capability of computing, predicting and classifying for grassland productivity, demonstrating that the paramet-
data and are more versatile than linear regression models. ric model performed better, with a root mean square error
The use of machine learning algorithms for estimating crop of 49.7 g DM m−2y−1 compared to 85.4 g DM m−2y−1 for the
yields, e.g. corn (Panda et  al. 2010; Serele et  al. 2000) and BIOME-BGC model.
rice (Ji et al. 2007) has been widely reported, however, only In summary, regression models based on VIs have predomi-
a limited number of studies have been described for their nantly been used for grass yield estimation. Machine learn-
application to estimation of grassland above-ground biomass ing algorithms are proving to be powerful tools for grassland
(dry matter) (Ali et al. 2014, 2015). Xie et al. (2009) compared classification, but still need to be further developed for grass
the performance of ANN and MLR for above-ground grass- yield estimation (Mountrakis et al. 2011). The fusion of mul-
land biomass in the Xilingol River Basin, Inner Mongolia. tisource data into biophysical simulation models also requires
Topographic, vegetation index and spectral information from further research in order to better exploit their suitability and
Landsat ETM+ were used as input data, with ANN gener- transferability.
ating a better yield estimation than the MLR (r2  =  0.817,
Grazing management
RMSE = 42.36%, compared to r2 = 0.591, RMSE = 53.20%).
In another study, Yang et al. (2012) used a back propagation Grazing impacts
ANN algorithm for grassland yield estimation based on five Degradation in grasslands and rangelands is a very complex
VIs derived from MODIS satellite data, with NDVI and SAVI and dynamic phenomenon caused by natural and anthro-
showing the best fit with the in situ sample biomass. Once pogenic activities (Paudel and Andersen 2010), which can
again, the ANN models were more accurate (R2 = 0.56–0.71) be assessed at a small scale by an expert opinion or visual
than the statistical models (R2 = 0.54–0.68). evaluation, however, for national or global scale evalua-
Mountrakis et  al. (2011) comprehensively reviewed the tion the use of remote sensing technology is a more feasi-
application of SVM in satellite remote sensing applications ble approach. Tueller (1989) first described the application of
but its use for biomass estimation is not discussed. A limited aerial photography and satellite imagery to support manage-
number of studies have applied SVM to biomass assessment ment of rangeland resources, but the quality and quantity of
from satellite imagery (Jachowski et al. (2013), for mangrove satellite imagery available at the time proved a limiting factor.
ecosystems), but there is no reference to it being used for Tueller did however predict that within 20 years the major-
grassland biomass. The potential of SVM for grassland bio- ity of required management information would be avail-
mass estimation was established by Clevers et al. (2007) with able from satellite imagery, a prediction realised by Munyati
a band shaving algorithm to identify highly correlated bands and Makgale (2009) who used a time series of Landsat TM
in airborne hyperspectral data and thus develop the most pre- imagery to map and quantify degraded rangeland in South
dictive band ratio. With the development of new hyperspec- African communal grazing lands. Pickup et  al. (1994) first
tral satellite instruments, the potential for powerful species used satellite data for the assessment of land degradation
and site specific indices will be enhanced. by combining image derived vegetation cover index values
and spatial models of grazing density determined as a func-
Simulation models tion of distance from a watering point. Trends in rangeland
For indirect vegetation biomass estimation, simulation mod- degradation (Pickup et  al. 1998) were also identified from
eling techniques are used, whereby remote sensing data are imagery, with a vegetation cover model built from multitem-
used as an input variable or substitute for vegetation parame- poral remote sensing data in order to distinguish between
ters. In order to better understand the growth mechanism and natural and human impacts on degradation. With a longer
spatial variability of grasslands, meteorological data driven time series of Lands at data to derive locations of persistent
models have been used to simulate and predict the grass- ground cover, Bastin et al. (2012) demonstrated that it is also
growth rates (Barrett et al. 2005; Bouman et al. 1996; Moore possible to discriminate between natural and human induced
et al. 1997; Woodward 2001). The precision of these models grazing effects on ground cover in Queensland. Other stud-
heavily depends on their ability to incorporate multisource ies have also exploited multitemporal datasets for degradation

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 658 Journal of Plant Ecology

assessment (Paudel and Andersen 2010), mapping and quan- heating for longer, and, as snow has a high albedo, it can eas-
tification of degraded areas at different scales (Alves Aguiar ily be analysed from image data. Studies show that at high
et  al. 2010), and in combination with GIS technologies to latitudes where the vegetation is tall, dense snow melts ear-
investigate changes in grassland cover (Zheng et al. 2011). lier (Loranty et al. 2011; Marsh et al. 2010) compared to the
Remote sensing technology is not only useful for the iden- short vegetation. In response to Hein’s (2006), findings Retzer
tification of degraded areas, but also for mapping, monitoring (2006) reported that high resilience after drought may be due
and quantifying restoration of such degraded land after the to the precipitation dynamics not because of high intensity
implementation of corrective measures. A ban on grazing was grazing as suggested by Hein (2006).
imposed in Ningxia province of China in 2003 to decrease Careful consideration of sampling scale is very important in
degradation, and in a recent study Li et al. (2013) used Lands remote sensing studies, and needs to be determined according
at images to map the positive outcomes of this ban, with to the application. Yang et  al. (2011) tested the significance
59.41% restoration reported between 1993 and 2011. Huang of measured biophysical parameters (canopy cover, height
et  al. (2013) also successfully demonstrated how such tech- and LAI) to find the difference between grazed and ungrazed
niques could be used to effectively evaluate trends in deg- sites, where for canopy height, and ratio of photosyntheti-
radation after the implementation of restoration programs cally active and non-active vegetation cover, the difference
using AVHRR (1982–2003) and MODIS (2000–2008) remote was significant. Among the various spectral VIs, red and
sensing images. NIR-based measures showed the most significant correlation
In summary, remotely sensed imagery has been success- with canopy height. This analysis was based on single dates
fully used for detecting degradation and recovery of grass- and suggests the use of multitemporal remote sensing data
land areas. More research is needed to fully explore the data for evaluating pre and post-grazing vegetation conditions.
from newly launched high resolution SAR sensors because in A combination of remote sensing and GIS models can be used
degraded areas grass cover is sparse with open soil, and more for the evaluation and classification of study sites based on
work is required in order to better understand the backscatter their suitability for grazing (Bozkurt et al. 2011).
response from such sites. In grassland management and the livestock business,
grazing capacity and intensity are the key factors that need
Assessment of grazing capacity and intensity to be monitored consistently in order to optimize the feed-
Grazing management strategies are directly linked to factors ing resources. Information extracted from satellite remote
including grazing intensity, length of grazing period, grazing sensing has been shown to be useful forest imating graz-
regimes, stocking rate and elevation (Bradley and O’Sullivan ing capacity—the maximum number of animals that can be
2011; Vermeire et al. 2008; Volesky et al. 2004), and vary from sustained in a given area of pasture in a year—and inten-
area to area in order to meet livestock grazing management sity, which is required for nutritional planning of livestock.
goals. Grazing intensity has the most influence on grassland For the assessment of short-term grazing capacity at pad-
productivity, and overgrazing can cause grassland degradation dock level, Phillips et al. (2009) developed a model based on
(Boddey et al. 2004) with some studies showing that light to remote sensing and ground-based data on cattle nutrition.
moderate grazing intensity practices can enhance grassland They observed the underestimation of grazing capacity by
productivity under certain environmental conditions (Luo the model and suggest additional testing of the model and
et al. 2012). Remote sensing approaches can be used to moni- at multiple sites. Along with additional testing at multiple
tor livestock grazing (Feng and Zhao 2011) at light to moderate sites, use of very high resolution data (e.g. GeoEye-2: 1.35
intensity (Yang et al. 2012; Yang and Guo 2011). Kawamura m, WorldView-3: 1.24 m) might be valuable to correct this
et al. (2005a) used NDVI derived from remote sensing data for anomaly. Wu et  al. (1996) proposed a physical model for
the quantification (R2 = 0.77–0.83) of grazing distribution in simulating productivity in grazing ecosystems, with Bénié
Inner Mongolian grasslands. In another study, Numata et al. et al. (2005) developing the model further to include remote
(2007) used Landsat TM data in order to analyse the impact sensing and socioeconomic parameters in order to simulate
of grazing intensity on a pasture’s biophysical features, with the available biomass or carrying capacity with an accuracy
remotely sensed non-photosynthetic vegetation showing the of 80%. The use of remote sensing data becomes a chal-
highest correlation with grazing intensity (r2  =  0.70) com- lenge in applications where the underlying target area is
pared to the other measured biophysical features, e.g., above composed of sparse vegetation and highly reflective soil. In
ground biomass, canopy height and water content. order to overcome this problem, Edwards et al. (1999) pro-
Consistent and frequent monitoring of the effects of graz- posed a geometric optical model based on low resolution
ing intensity is crucial in arid, semi-arid and commercial graz- satellite imagery whose output is a series of change maps
ing pasture areas, as grazing intensity influences the grassland that can be used to estimate the final vegetation cover.
ecosystem (Röder et al. 2008) both in a positive and a nega- A  very high correlation between observed and estimated
tive manner. An example of apositive influence is given by vegetation cover was reported (r2 = 0.837), but even though
Cohen et al. (2013) for a high latitude, intensively grazed area, the approach was quite useful no further applications of this
where late snow melt means the surface is protected from approach can be found. Similarly, no reference to SAR data

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 659

for assessment of grazing capacity and intensity is evident collection about the physical and/or chemical properties of the
in the literature. target vegetation in order to assist decision making through the
In summary, identification of grazing capacity and intensity use of predictive tools and forecasting models. For satellite-based
is required in order to avoid overgrazing and degradation, but precision agriculture, the spatial resolution, satellite revisit fre-
there is a very fine distinction between normal grazing and quency and number of spectral bands are the key factors that
overgrazing, and in order to better understand this transition are related to the acquisition of a dense time series for consistent
the use of very high resolution optical data, SAR data, and a monitoring at a farm or paddock scale. Much of the work done
combination of both needs further investigation. to date on this subject has been focused on croplands using
field spectrometry (Gutiérrez et al. 2008; Prabhakar et al. 2011;
Pasture quality and status Zhang et al. 2003), airborne imagery (Epinat et al. 2001; Erives
Grazing capacity depends not only on the grassland spatial and Fitzgerald 2005) and satellite data (De Benedetto et al. 2013;
extent but also on the quality of grass, which is directly linked López-Lozano et al. 2010; Nahry et al. 2011; Thenkabail 2003),
to livestock feeding. While the potential of remote sensing- and it is only very recently that grassland management and
based classification and mapping of grassland quality has been precision farming has been considered. The ‘Pastures From Space
long recognized (Giraed et al. 1990), only a limited number of (http://www.pasturesfromspace.csiro.au)’ project in Australia
studies have been done on grassland quality assessment using is one of the most prominent, and has developed a dedicated
this approach. The range of data used varies between coarse grassland/pastures tool to deliver near real-time information
(Kawamura et al. 2005b; Si et al. 2012), medium (Kawamura (e.g. biomass, growth rate) at the farm and paddock level using
et al. 2005b) and high (Guo et al. 2005; Si et al. 2012) spatial high and medium resolution satellite remote sensing(Donald
resolution. Studies show that the leaf area index (LAI) is con- et al. 2004; Edirisinghe et al. 2011; Henry et al. 2004). The tech-
sidered as more appropriate for the assessment of grassland niques were developed and validated in Western Australia over
health, biomass and plant water content than the satellite a 5-year period, and then transferred and verified in Southern
derived NDVI (Guo et  al. 2005). In a recent study, Falldorf Australia, and the project is providing online (web and also
et  al. (2014) developed a remote sensing-based tool called software based) pasture growth rate at weekly regional and
the Lichen Volume Estimator (LVE) to assess winter pasture paddock scales. Schellberg et al. (2008) wrote a detailed review
quality (in terms of volume) by using a 2D Gaussian regres- focusing on precision agriculture of grasslands, in which they
sion model based on a Normalized Difference Lichen Index discuss the applications of different remote sensing techniques
(NDLI  =  MIR−NIR/MIR+NIR) and Normalized Difference for the monitoring of physical, chemical and area-based grass-
Moisture Index (NDMI  =  NIR−MIR/NIR+MIR). The authors land properties for farm-related decision-making.
concluded that LVE could become an important tool to assist Pasture growth rate is a biophysical property (monitored
in prediction of winter grazing areas for reindeer and caribou as kg dry matter/ha per day) which is related to how much
herds at one location, and with further field studies it could grass grows on a daily basis and is an important driving fac-
become more widely applicable. Multispectral remote sensing tor for feed budgeting related decisions. Apart from manage-
data has also been used in combination with in situ data (Zerger ment practices, climatic factors also influence the growth rate
et al. 2011) and models such as the radiative transfer model of grasses (Thorvaldsson et  al. 2004). There is no precipita-
PROSAIL (Quan et al. 2015; Si et al. 2012) for the assessment tion component in the C-Fix model (as discussed in Section
of vegetation/grassland condition and quality. The inversion Simulation models) but the Australian ‘Pastures From Space’
(Si et al. 2012) of the PROSAIL model and MERIS reflectance model differs by including precipitation as well a slight use
data (single biome approach) has great potential to estimate efficiency (LUE) models (Hill et al. 2004; Piñeiro et al. 2006a),
the grassland LAI (R2 = 0.70) and canopy chlorophyll content data integration (Hill et al. 1996; Moore et al. 1999) and clas-
(R2 = 0.61). Hill (2013) simulated ESA Sentinel-2 (high reso- sification (Vickery et al. 1997) tools for growth rate prediction
lution optical sensor) data and showed that VIs based on these (Donald et  al. 2010), monitoring and mapping. Multisource
bands can be used for the identification of vegetation states in (e.g. Landsat, SPOT, MODIS, AVHRR, Hyperion) remote
grassland and savannas. sensing data with different spatial resolutions were used to
In summary, pasture quality and status are directly related successfully assess the growth rate at different spatial scales
to grassland management. Detailed investigations on the use (Donald et al. 2004; Henry et al. 2004).
of hyperspectral remote sensing data are required, and to ‘Pastures From Space’ is an effective tool for near real-time
exploit the large number of bands different VIs at different monitoring at farm and paddock level in order to better manage
wavelengths can be calculated in order to retrieve multiple the feed resources for livestock industries, but currently repre-
vegetation parameters. sents the only operational system designed specifically for pas-
tures. Schellberg and Verbruggen (2014) discuss the delay in
Pasture growth rate assessment transferring techniques developed for arable land to grassland,
To meet the increasing demand for food, optimization of agri- although there is scope for the successful implementation of
cultural production and effective resource management are emerging technologies such as precision agriculture in a vari-
critical. Precision agriculture involves real or near real-time data ety of environments. After the successful implementation and

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 660 Journal of Plant Ecology

validation of the ‘Pastures From Space’ project, in 2003 Fonterra the implementation of these directives, mapping, reporting
(http://www.fonterra.com/global/en) formed a partnership and monitoring on the status of nature conservation sites
with CSIRO in order to explore its potential in New Zealand has been a key research topic. Over the time remote sensing
dairy farming and pasture monitoring, and various studies methodologies and techniques have become more sophisti-
have been done since then (Ausseil et al. 2011; Dymond et al. cated, especially for synoptic data acquisition, and are now
2006; Edirisinghe et al. 2012; Mata et al. 2010). being successfully used for fast, reliable and consistent map-
In conclusion, both airborne and spaceborne remote sens- ping of habitats and species (Nagendra 2001; Nagendra and
ing data are being used to collect real time (or near real time) Gadgil 1999). Most conservation sites, including grasslands,
information on pasture yields and growth rates. Based on sat- are small in size, therefore, very high-resolution imagery is
ellite remote sensing data, decision support systems can be required to monitor them, and some of the very high-reso-
developed for farm-related management decisions. lution spaceborne instruments with a short revisit time of a
few days, launched within the last decade have been proven
Transhumance suitable for this application (Schuster et al. 2015).
In mountainous regions, there is an annual cycle of livestock The nature conservation sites are monitored using both mul-
migration to the higher elevation pastures in warm seasons and tispectral optical and SAR imagery, and increasingly a com-
return to lower altitudes for the rest of the year, with a concurrent bination of both. Optical sensors have a long legacy of use in
cycle of high grazing intensity and pressure. Such transhumance, identification of location and changes in habitats (Velazquez et al.
or herd mobility, is one of the key components for sustainable 2008), knowledge and object based classification mapping of
use of these upland resources (Sitters et al. 2009) that are highly Natura 2000 species (Förster et al. 2008, 2012) and for assessing
sensitive to environmental changes, and for that reason it is climatic influences on Natura 2000 habitats (Förster et al. 2014).
essential to monitor their land cover dynamics (Morán-Ordóñez Multitemporal high resolution RapidEye data have proven par-
et al. 2011). Satellite imagery has considerable potential to detect ticularly useful in deriving phenological vegetation dynamics
and map land use, their corresponding effects on livestock feed from time series imagery, where at least three acquisition dates
resources and feed deficit management strategies (Mekasha et al. within a year are available (Franke et al. 2012). Since the launch
2014). Butt et al. (2011) used a MODIS NDVI time series from of the very high-resolution TerraSAR-X and COSMO-SkyMed
2000 to 2010 in order to evaluate the gradient of rangeland phe- SAR sensors, protected sites can also be monitored using radar
nology with respect to the changing latitude and its effects on imagery, with recent studies by Ali et  al. (2013) and Schuster
the direction and timing of livestock movement in the Sudano– et al. (2011) demonstrating the potential of both sensors for suc-
Sahelian region in West Africa. A  double logistic function was cessfully identifying grassland management practices in pro-
adapted to fit the NDVI trajectories derived from 1 km resolution tected sites. Although, the combined use of SAR and optical data
MODIS data, and a strong dependency of vegetation phenology has not yet been explored in detail, Ali et al. (2013) highlighted
on altitude was found. In another study, Sulieman and Elagib the potential use of both data sources for cross validation.
(2012) used multitemporal remote sensing data to map the effects Vanden Borre et  al. (2011a, 2011b) conducted a detailed
of climate, land use and land cover changes along three different review of the legal requirements for Natura 2000 habitat
livestock seasonal migration routes in eastern Sudan. A  major monitoring requirements and practices, and how remote
conversion from natural vegetation cover to agricultural land is sensing is being used to fulfil this task. In order to enhance
reported along with the significant increase in climate warming the utilization of remote sensing technology, field experts and
[based on 68 years (1941–2009) of climate date, e.g., tempera- conservation site managers suggested that the prime focus
ture, rainfall and aridity index]. Dedicated efforts are being made must be on data standardization, development of user-friendly
to fully detect and map the transhumance corridors using both products, method validation and knowledge sharing. Since
remote sensing and geospatial analysis approaches (Trans 2014). their review, work has been ongoing to resolve these issues,
In brief, the potential of remote sensing to trace corridors of e.g., Schröder et al. (2013) stress the need for pre-validation
seasonal movement of herds has been established. More work of Earth observation products for Natura 2000 sites before
needs to be done in order to exploit the use of high resolution delivery. On the other hand, Nieland et al. (2012) are working
optical and radar imagery in order to fully uncover the impact on an ontological approach for the integration of classifica-
of these seasonal movements on vegetation phenology. tion methodologies in order to overcome the issues of scale
and the transferability of methodologies. While these stud-
Remote sensing of nature conservation ies address all conservation sites, the challenges raised apply
grassland sites equally to grasslands, and the need for common data stand-
For the maintenance of biodiversity in Europe, the European ards and methods, and accessible products for a range of end-
Union has legislated a legal policy framework that includes users are of relevance to all aspects of grassland management.
the Habitats and Birds Directives (EEC 1979, 1997), which To sum up, the applicability of multispectral and multi-
describe the types of habitat (e.g., grassland, forest or meadow temporal remote sensing data for both monitoring and map-
types) whose existence is in danger (Natura 2000) and needs ping of grassland conservation sites has been demonstrated.
to be preserved by the member states (Ali et al. 2013). Since More research is required to overcome the limitations of site

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 661

specific methodologies (Schuster et al. 2011) in order to make a challenge to build a sufficiently long time series for machine
them more robust and standardised. These sites are typically learning applications, especially for multitemporal analysis.
small in size, so high resolution hyperspectral remote sens- For example, remote sensing data from Landsat and MODIS
ing data can be used to better explore species compositions. are available for longer periods of time, but in situ or inven-
Application of SAR data in cloudy conditions is equally feasi- tory data are available only for selected sites, which limits the
ble as demonstrated by Ali et al. (2013). national or global scale evaluation using these methods.
Until recently, optical sensors were considered the best data
source for mapping and monitoring small-scale variations within
OPERATIONAL AND TECHNICAL and among the fields due to their high spatial and spectral
resolution. However, following the launch of high-resolution
CHALLENGES microwave radar remote sensing satellites (e.g. TerraSAR-X,
Overall, in the domain of remote sensing the research focus COSMO-SkyMed, Radarsat, Sentinel-1) the application domain
for classification and retrieval of biophysical parameters is now of radar sensors has widened. For instance, the TerraSAR-X
shifting towards the application of machine learning algo- Staring Spotlight acquisition mode can acquire images with a spa-
rithms. Object-based image classification presents a paradigm tial resolution of up to 0.25 m every 11 days. Achieving this high
shift to gain a new perspective on image classification and resolution from space can further support precision agriculture
better follow the boundaries of natural vegetation elements. developments, especially for areas under persistent cloud cover.
In object-based classification, segmentation scale and classifi- High-resolution radar remote sensing data with an improved
cation accuracy are strongly linked (Liu and Xia 2010), and temporal resolution will help to monitor crop health and will
careful selection of segmentation scale is required. Machine provide a mechanism for timely crop yield estimation, while in
learning strategies are becoming more widely used within the case of grasslands it can be used for monitoring grassland man-
remote sensing community, and methods like random forest agement practices as shown in Fig. 3. Spatial resolution is a cru-
and extremely randomized trees are now widely evident in the cial component in remote sensing applications, especially for
literature (Barrett et al. 2014). In future, approaches such as quantitative scientific analysis, and as Fig. 3 demonstrates inter
deep learning and data assimilation will provide more insight and intra paddock/field variations can be detected using radar
into the integration of multisource remote sensing data for data, highlighting different agricultural states. Using high-res-
complex and dynamic environmental systems. These meth- olution sensors (e.g. TerraSAR-X, Radarsat, COSMO-SkyMed),
ods are based on supervised learning, and thus training data it is possible to detect many management-related activities, e.g.,
are required for classification and parameter retrieval applica- grazing herds, hedges and cultivation. It is also possible to trace
tions. Machine learning algorithms are data driven and their the identify poorly performing patches of the field using multi-
performance is highly influenced by the number of features, temporal acquisitions, but the major challenge and limitation
sample size and data pre-processing steps. Until recently it was remains in the high data acquisition cost of the highest spatial

Figure 3:  TerraSAR-X Staring Spotlight colour composite (R: 08-06-2014, G: 19-06-2014, B: 11-07-2014) of Teagasc Curtin Farm. Potential of
very high-resolution microwave radar (TerraSAR-X) data: (A) Monitoring of hedges and individual tree count, (B) furrow/plough lanes, (C)
possibility to detect the location of grazing herds if they are standing close to each other and (D) inter and intra paddock variation.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 662 Journal of Plant Ecology

resolution sensors, and their small area coverage. Currently, uniform standardized and transferable approach for monitoring
most of the radar remote sensing sensors (with some excep- farms at different geographical scales. For the transferability of
tions) are single or dual channel, and polarization limited to methods, it is very important to have a uniform input dataset,
two directions, but as the technology matures further future and one potential solution for this could be the development
radar sensors will potentially provide additional information of a new ontology-based data collection and standardization
for more reliable methods for agricultural monitoring. framework as undertaken by the biology community (Bard and
Thus, for both optical and radar remote sensing the major Rhee 2004). Additionally, the remote sensing community must
limitation is the compromise between spatial resolution and continue to advocate the launch of follow-up missions of imag-
spatial coverage. For instance, TerraSAR-X Staring Spotlight ing satellites in order to ensure long-term consistent monitoring.
mode has the highest illumination time and spatial resolution There is a need to train and educate the end users (farmers,
(up to 0.25 m) but the smallest swath size (4 km (width) × 3.7 land manages and policy makers) about the potential applica-
km (length)), compared to the Spotlight mode (spatial resolu- tions of satellite remote sensing, and with standardized meth-
tion up to 2 m: 10 km (width) × 10 km (length)). A similar ods, this is more achievable. Current technical and scientific
comparison is true for WorldView and MODIS, where high deliverables (e.g., project reports, scientific publications)
spatial resolution is achieved at the cost of swath size. output from many research projects further discourage com-
Farmers in every region follow different management strate- munication between the data providers and the end users.
gies, i.e., amount of fertilizer, use of pesticides, grazing season One option could be for scientists to develop more portable
length, and measuring units (kg/tonne dry matter per hectare, (i.e., WebGIS) and accessible (mobile apps) solutions, which
kg/tonne dry matter per acre). With this diversification in man- are readily available to the end users (e.g. PastureFromSpace
agement practices there are challenges in building a robust and Australian project). The benefits offered by remote sensing
transferable classification and reporting scheme (Fig.  4 gives scientists working with those in the agricultural community
an overview of different remote sensing techniques and their will not only help to generate more business, but also to
potential scope and limitations). In future, as more sensors widen the scope and application domain that can be achieved
are launched it is important for the community to develop a through the use of imaging satellites.

Figure 4:  an overview of grassland monitoring technologies with their limitations and scope.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 663

DISCUSSION AND FUTURE PROSPECTS thus overcome cloud contamination issues, but they cannot
be applied for field level mapping and monitoring in many
In conclusion, grasslands are one of the most widespread land- countries due to the coarse resolution whereby the pixel size
cover types found globally, and they need to be monitored at is greater than field size.
multiple scales (gobal, regional, national, paddock) depend- To achieve the maximum benefit from satellite remote
ing on the nature of the information required. Given the sensing for grassland-related activities a number of issues
small-scale coverage of traditional ground-based methods of have been identified.
grassland monitoring, satellite remote sensing approaches are Classification is a classical application of satellite images,
likely to be a significant contributor to the future operational and currently the focus is shifting from statistical to machine
studies. Different sensor specifications are required depending learning approaches, due to their ability to better identify the
on the application scale, e.g., for global scale applications a relative importance of different inputs as well as learn from
sensor with large spatial coverage and coarse resolution (i.e., repeated use. Classification of grassland types and formations
MODIS, AVHRR) would be sufficient. In the case of managed using satellite remote sensing data has been tested by using dif-
grassland-related applications (at paddock scale), sensors with ferent classifiers and sensors in different regions of the world.
high spatial and temporal (GeoEye: 1.35 m, 3 days; RapidEye: In addition to local, regional and national scales, an acceptable
6.5 m, 5.5 days; QuickBird: 2.4 m, 1–3.5 days) resolution are classification accuracy using medium resolution (Landsat TM/
the preferred choice. During the growing season, temporal ETM+) data has been achieved at the global scale (Gong et al.
resolution is very important and plays a critical role in near 2013). However such an approach is very data and computa-
real-time monitoring of phenological stages, and when com- tionally intensive. Individually, machine learning and object-
bined with very high spatial resolution imagery, inter- and based classification methods perform very well but, in future,
intra- field variations can be detected. Thus, despite some these two approaches may be further integrated to exploit the
instrument biases (Yang et al. 2013), satellite sensors currently benefits of each, for example, a random forest random field
present the best option for long term, large scale, objective (RF)2 classifier (Payet and Todorovic 2010). The literature
and repeatable studies. suggests that random forest and extremely randomized trees
Optical sensors are more appropriate for grass monitoring classifiers have the best potential and offer improved classi-
and mapping compared to radar sensors (Price et  al. 2002; fication results for grassland identification, but further work
Smith and Buckley 2011). At present, given the difficulty in on these methods is needed to validate new high resolution
relating radar backscatter to grassland properties (Hajj et al. optical and SAR data and explore the transferability of these
2014), but this may change with the advent of very high res- methods.
olution fully polarimetric SAR sensors. Different VIs derived Maximum separability of spectrally similar classes, such
from optical remote sensing data correlate well with differ- as different grassland types, can be achieved with a larger
ent vegetation biophysical parameters, but the biggest chal- number of narrowband images, but currently the scope of
lenge to the use of optical imagery is cloud contamination spaceborne hyperspectral remote sensing is very limited due
and atmospheric noise. Data cleaning, by filtering or use of a the fact that Hyperion is the only operational satellite. More
cloud mask to remove noisy pixels is widely undertaken, but detailed analysis is still to be done on the potential for grass-
is very sensor specific and location dependent. The conser- land mapping and monitoring from spaceborne hyperspectral
vation of image information and removal of noisy signals is data, but this is unlikely to progress prior to the launch of
complex, and in order to construct a long time series of reli- EnMAP which has 244 spectral bands (scheduled for 2017).
able values, the most commonly used approaches are time- The use of hyperspectral data for grassland classification using
series composites and the integration of multisensor data. machine learning classifiers has not been fully explored but
However, the latter approach is hindered by variable instru- studies using airborne hyperspectral remote sensing data
ment biases, spectral response signals and spatial resolutions. (Chan and Paelinckx 2008; Darvishzadeh et  al. 2011; Yang
Poorly designed data fusion algorithms that assimilate differ- and Everitt 2010) suggest the potential and feasibility of
ent datasets might also result in high uncertainty in the final the application of spaceborne hyperspectral remote sensing
output. On the other hand, modeling approaches driven by data for grassland mapping. In future this might result in a
satellite remote sensing have proven to be a robust method paradigm shift in sensor development from multispectral to
for deriving grassland information, but the availability of hyperspectral constellations.
high quality validation data to accurately calibrate the model The advantage of using fully polarimetric SAR data over
can be a limiting factor as it requires a collection of sufficient dual and single polarizations in terms of improvement in clas-
high quality validation samples at large scales both expen- sification performance is well established (Lee et al. 2001). The
sive and laborious. Careful selection of sensors (especially in inconsistencies reported in the literature (Dusseux et al. 2014;
terms of spatial and temporal resolution) for data acquisition Smith and Buckley 2011) indicate that SAR polarimetry appli-
is also very important, for example, frequently acquired and cations to grasslands still require more detailed investigation as
freely available hypertemporal remote sensing data (e.g., an understanding of SAR polarimetry theory matures and the
MODIS) are widely used to generate time composites and availability of spaceborne fully polarimetric data increases. In

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 664 Journal of Plant Ecology

coming years, especially after the launch of SAOCOM-1/2 (an combination of SAR and optical data has been less widely
Argentinian constellation of two L-band SAR sensors sched- reported, although this will increase as more SAR missions
uled for launch in 2015)  and the RADARSATConstellation become operational in the coming years.
mission (three Canadian C-band SAR sensors, scheduled for
launch in 2018), a better understanding of the potential for
fully polarimetric SAR data to analyze the back scattering
SUMMARY
behaviour of different habitat types at different polarizations To conclude this review:
will be possible. As a result, a more reliable delivery of grass-
• Satellite remote sensing can be used for the retrieval of
land products in cloudy regions should be possible.
grassland biophysical parameters, including biomass, qual-
The application of very high resolution data for remote
ity, growth, land cover, degradation, grazing capacity, as
sensing-based precision agriculture approaches to grassland is
well as mapping and monitoring for conservation and
now evolving to the same level of maturity as experienced by
management.
arable agriculture. As more very high-resolution sensors are
• Optical sensors have been most widely used given the
launched and work is done on data standarization more reli-
good understanding between reflectance and vegetation
able operational satellite-based grassland management tools
properties and the difficulty in relating radar backscatter
are expected. Furthermore, operational tools that are simple
to grassland biophysical properties, but this may change
to understand and operate for non-experts, such as websites
with the advent of very high-resolution fully polarimetric
or mobile applications that retrieve information from a dedi-
SAR sensors.
cated data center server could become a more common prac-
• The use of hyperspectral data for grassland classification
tice across precision agriculture for all land cover types.
using machine learning classifiers has not been fully ex-
Much of the research that has been done on grasslands has
plored but studies using airborne hyperspectral remote
exploited multitemporal datasets, with relatively few long
sensing data suggest the potential and feasibility of the ap-
term studies done except those which could exploit informa-
plication of spaceborne hyperspectral remote sensing data
tion content from Landsat or MODIS datasets. Additionally,
for grassland mapping, and with future hyperspectral sen-
hypertemporal time series that are optimized to minimize
sors this potential may be realised.
the computational load, can enhance grassland classification,
• The application of very high-resolution data for remote
especially where there are rapid or distinctive phenological
sensing-based precision agriculture approaches to grass-
changes through the growing season. To have a consistent
land is now evolving to the same level of maturity as ex-
time series of data over many years to track long term changes
perienced by arable agriculture, but more work needs to
in land cover, and especially for operational purposes, a com-
be done on communicating the benefits and opportunities
mitment to continuity missions is required. MODIS is pro-
of space to the farming community.
viding free data at different spatial scales for more than a
• Hypertemporal time series that are optimized to minimize
decade. Suomi National Polar-orbiting Partnership (NPP)
the computational load, can enhance grassland classifica-
equipped with five sensors including Visible Infrared Imaging
tion, especially where there are rapid or distinctive phe-
Radiometer Suite (VIIRS) was launched in 2011. Spacecraft
nological changes through the growing season
orbits the Earth 14 times a day. VIIRS has the spatial resolution
of 375 and 750m for Imagery and Moderate resolution bands
respectively. NPP VIIRS data will be used to expand upon the ACKNOWLEDGEMENTS
MODIS applications to land, ocean and air quality. The VIIRS
data will also be freely available to the public unlike Rapideye We thank Teagasc Walsh Fellowship Program for funding this
research. The authors would like to thank the anonymous reviewers
and Quickbird hyper-spatial data, which is not easily accessi-
for their valuable comments and suggestions to improve the quality
ble and are expensive for developing countries and large scale
of the article.
applications. Sentinel-2 will also provide a comparable dataset Conflict of interest statement. None declared.
to the Landsat and SPOT missions in the optical part of the
spectrum, and at radar wavelengths Sentinel-1 will provide
C-band SAR data following ERS1/2 and ENVISAT ASAR, and REFERENCES
a TerraSAR-X2 launch is planned in 2016 as a follow-up mis- Ali I, Cawkwell F, Green S, et al. (2014) Application of statistical and
sion of TerraSAR-X (Janoth et al. 2012). machine learning models for grassland yield estimation based on
Despite the complexity of grassland ecosystems, this a hypertemporal satellite remote sensing time series. Presented at
review has demonstrated that satellite remote sensing tech- the Geoscience and Remote Sensing Symposium (IGARSS), 2014
nologies have been proven as effective tools for monitoring, IEEE International, 5060–3.
mapping and quantifying different grassland types and bio- Ali I, Greifeneder F, Stamenkovic J, et al. (2015) Review of machine
physical parameters. Use of optical remote sensing data is the learning approaches for biomass and soil moisture retrievals from
most prevalent in the literature, while the use of SAR or a remote sensing data. Remote Sens 7:16398–421.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 665

Ali I, Schuster C, Zebisch M, et  al. (2013) First results of monitor- Bouman BAM, Schapendonk AHCM, Stol W (1996) Description of
ing nature conservation sites in Alpine region by using very high the Growth Model LINGRA as Implemented in CGMS, Quantitative
resolution (VHR) X-Band SAR data. IEEE J Sel Top Appl Earth Obs Approaches in Systems Analysis; no. 7. AB-DLO, Wageningen.
Remote Sens 6:2265–74. Bozkurt Y, Uzun N, Dogan C, et al. (2011) Grassland evaluation based
Alves Aguiar D, Adami M, Fernando Silva W, et  al. (2010) Modis on GIS model and Remote Sensing data for beef cattle grazing. In
time series to assess pasture land. Presented at the Geoscience and Pötsch EM, Krautzer B, Hopkins A (eds). Grassland Farming and
Remote Sensing Symposium (IGARSS), 2010 IEEE International, Land Management Systems in Mountainous Regions. Proceedings of the
2123–6. 16th Symposium of the European Grassland Federation, Gumpenstein,
Anderson G, Hanson J, Haas R (1993) Evaluating landsat thematic Austria. Agricultural Research and Education Center (AREC),
mapper derived vegetation indices for estimating above-ground Raumberg-Gumpenstein, 181–3.
biomass on semiarid rangelands. Remote Sens Environ 45:165–75. Bradford JB, Hicke JA, Lauenroth WK (2005) The relative impor-
Anderson JM (1991) The effects of climate change on decomposition tance of light-use efficiency modifications from environmental
processes in grassland and coniferous forests. Ecol Appl 1:326–47. conditions and cultivation for estimation of large-scale net primary
An N, Price KP, Blair JM (2013) Estimating above-ground net primary productivity. Remote Sens Environ 96:246–55.
productivity of the tallgrass prairie ecosystem of the Central Great Bradley BA, O’Sullivan MT (2011) Assessing the short-term impacts
Plains using AVHRR NDVI. Int J Remote Sens 34:3717–35. of changing grazing regime at the landscape scale with remote
Attema EPW, Ulaby FT (1978) Vegetation modeled as a water cloud. sensing. Int J Remote Sens 32:5797–813.
Radio Sci 13:357–64. Brenner JC, Christman Z, Rogan J (2012) Segmentation of landsat the-
Atzberger C (2013) Advances in remote sensing of agriculture: con- matic mapper imagery improves buffelgrass (Pennisetum ciliare) pas-
text description, existing operational monitoring systems and ture mapping in the Sonoran Desert of Mexico. Appl Geogr 34:569–75.
major information needs. Remote Sens 5:949–81. Breymeyer AI (1990) Managed Grasslands: Regional Studies, 1st edn.
Ausseil A-G, Dymond J, Dynes R, et al. (2011) Estimating pasture qual- Amsterdam, New York: Elsevier Science.
ity using Landsat ETM+: application for the greenhouse gas inven- Brilli L, Chiesi M, Maselli F, et  al. (2013) Simulation of olive grove
tory of New Zealand. Presented at the International Symposium on gross primary production by the combination of ground and multi-
Remote Sensing for the Environment, Sydney, Australia. sensor satellite data. Int J Appl Earth Obs Geoinformation 23:29–36.
Baldi G, Guerschman JP, Paruelo JM (2006) Characterizing fragmen- Buckley JR, Smith AM (2010) Monitoring grasslands with radarsat 2
tation in temperate South America grasslands. Agric Ecosyst Environ quad-pol imagery. Presented at the Geoscience and Remote Sensing
116:197–208. Symposium (IGARSS), 2010 IEEE International, 3090–3.
Bard JBL, Rhee SY (2004) Ontologies in biology: design, applications Butt B, Turner MD, Singh A, et  al. (2011) Use of MODIS NDVI to
and future challenges. Nat Rev Genet 5:213–22. evaluate changing latitudinal gradients of rangeland phenology in
Barrett B, Nitze I, Green S, et  al. (2014) Assessment of multi-tem- Sudano-Sahelian West Africa. Remote Sens Environ 115:3367–76.
poral, multi-sensor radar and ancillary spatial data for grasslands Butterfield HS, Malmström CM (2009) The effects of phenology on
monitoring in Ireland using machine learning approaches. Remote indirect measures of aboveground biomass in annual grasses. Int J
Sens Environ 152:109–24. Remote Sens 30:3133–46.
Barrett PD, Laidlaw AS, Mayne CS (2005) GrazeGro: a European Cardinale BJ, Duffy JE, Gonzalez A, et al. (2012) Biodiversity loss and
herbage growth model to predict pasture production in perennial its impact on humanity. Nature 486:59–67.
ryegrass swards for decision support. Eur J Agron 23:37–56. Castle ME (1976) A simple disc instrument for estimating herbage
Bastin G, Scarth P, Chewings V, et  al. (2012) Separating grazing and yield. J Br Grassl Soc 31:37–40.
rainfall effects at regional scale using remote sensing imagery: Chan JC-W, Paelinckx D (2008) Evaluation of Random Forest and
a dynamic reference-cover method. Remote Sens Environ 121:443–57. Adaboost tree-based ensemble classification and spectral band
Belsky AJ (1986) Does herbivory benefit plants? A review of the evi- selection for ecotope mapping using airborne hyperspectral
dence. Am Nat 127:870–92. imagery. Remote Sens Environ 112:2999–3011.
Bénié GB, Kaboré SS, Goïta K, et  al. (2005) Remote sensing-based Clevers J, van Der Heijden G, Verzakov S, et  al. (2007) Estimating
spatio-temporal modeling to predict biomass in Sahelian grazing grassland biomass using SVM band shaving of hyperspectral data.
ecosystem. Ecol Model 184:341–54. Photogramm Eng Remote Sens 73:1141.
Bergman K-O, Ask L, Askling J, et  al. (2008) Importance of boreal Cohen J, Pulliainen J, Ménard CB, et  al. (2013) Effect of reindeer
grasslands in Sweden for butterfly diversity and effects of local and grazing on snowmelt, albedo and energy balance based on satellite
landscape habitat factors. Biodivers Conserv 17:139–53. data analyses. Remote Sens Environ 135:107–17.
Boddey RM, Macedo R, Tarré RM, et al. (2004) Nitrogen cycling in Darvishzadeh R, Atzberger C, Skidmore A, et  al. (2011) Mapping
Brachiaria pastures: the key to understanding the process of pas- grassland leaf area index with airborne hyperspectral imagery:
ture decline. Agric Ecosyst Environ 103:389–403. a  comparison study of statistical approaches and inversion
Booth DT, Tueller PT (2003) Rangeland monitoring using remote of radiative transfer models. ISPRS J Photogramm Remote Sens
sensing. Arid Land Res Manag 17:455–67. 66:894–906.
Boschetti M, Bocchi S, Brivio PA (2007) Assessment of pasture pro- Dawson TP, Curran PJ (1998) Technical note. A  new technique for
duction in the Italian Alps using spectrometric and remote sensing interpolating the reflectance red edge position. Int J Remote Sens
information. Agric Ecosyst Environ 118:267–72. 19:2133–9.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 666 Journal of Plant Ecology

De Benedetto D, Castrignano A, Diacono M, et al. (2013) Field par- Feng XM, Zhao YS (2011) Grazing intensity monitoring in Northern
tition by proximal and remote sensing data fusion. Biosyst Eng China steppe: integrating CENTURY model and MODIS data. Ecol
114:372–83. Indic 11:175–82.
Del Pozo A, Ovalle C, Casado MA, et  al. (2006) Effects of grazing Filella I, Penuelas J (1994)The red edge position and shape as indica-
intensity in grasslands of the Espinal of central Chile. J Veg Sci tors of plant chlorophyll content, biomass and hydric status. Int J
17:791–8. Remote Sens 15:1459–70.
Derner JD, Schuman GE (2007) Carbon sequestration and range- Filippi AM, Jensen JR (2006) Fuzzy learning vector quantization for
lands: a synthesis of land management and precipitation effects. J hyperspectral coastal vegetation classification. Remote Sens Environ
Soil Water Conserv 62:77–85. 100:512–30.
Donald GE, Edirisinghe A, Henry DA et al. (2004) Pastures from space: Flynn ES, Dougherty CT, Wendroth O (2008) Assessment of pasture
validation of predictions of pasture growth rates. http://www.pub- biomass with the normalized difference vegetation index from
lish.csiro.au/?act=view_file&file_id=SA0401077.pdf. active ground-based sensors. Agron J 100:114.
Donald G, Gherardi S, Edirisinghe A, et al. (2010) Pasture growth rate Foley JA, DeFries R, Asner GP, et al. (2005) Global consequences of
for individual paddocks can be accurately predicted real-time from land use. Science 309:570–74.
MODIS imagery, climate and soil data. Presented at the Australian Förster M, Frick A, Walentowski H et al. (2008) Approaches to utilis-
Society of Animal Production, Armidale, 611–5. ing QuickBird data for the monitoring of NATURA 2000 habitats.
Dubinin M, Potapov P, Lushchekina A, et al. (2010) Reconstructing Commun Ecol 9:155–68.
long time series of burned areas in arid grasslands of southern Förster M, Schmidt T, Schuster C, et al. (2012) Multi-temporal detec-
Russia by satellite remote sensing. Remote Sens Environ 114:1638–48. tion of grassland vegetation with RapidEye imagery and a spectral-
Dusseux P, Corpetti T, Hubert-Moy L, et  al. (2014) Combined use temporal library. Presented at the Geoscience and Remote Sensing
of multi-temporal optical and radar satellite images for grassland Symposium (IGARSS), 2012 IEEE International, 4930–33.
monitoring. Remote Sens 6:6163–82. Förster M, Zebisch M, Wagner-Lücker I, et al. (2014) Remote sensing-
Dymond JR, Shepherd JD, Clark H, et al. (2006) Use of VEGETATION based monitoring of potential climate-induced impacts on habitats.
satellite imagery to map pasture quality for input to a methane In Rannow S, Neubert M (eds). Managing Protected Areas in Central
budget of New Zealand. Int J Remote Sens 27:1261–8. and Eastern Europe Under Climate Change, Advances in Global Change
Edirisinghe A, Clark D, Waugh D (2012) Spatio-temporal modelling Research. Netherlands: Springer, 95–113.
of biomass of intensively grazed perennial dairy pastures using Franke J, Keuck V, Siegert F (2012) Assessment of grassland use
multispectral remote sensing. Int J Appl Earth Obs Geoinformation intensity by remote sensing to support conservation schemes. J
16:5–16. Nat Conserv 20:125–34.
Edirisinghe A, Hill MJ, Donald GE, et  al. (2011) Quantitative map- Franzluebbers AJ (2010) Soil organic carbon in managed pastures of
ping of pasture biomass using satellite imagery. Int J Remote Sens the southeastern United States of America. In FAO (ed). Grassland
32:2699–24. Carbon Sequestration: Management, Policy and Economics. Proceedings
Edwards MC, Wellens J, Al-Eisawi D (1999) Monitoring the grazing of the Workshop on the Role of Grassland Carbon Sequestration in the
resources of the Badia region, Jordan, using remote sensing. Appl Mitigation of Climate Change. Rome, Italy, 163–75.
Geogr 19:385–98. Gao J (2006) Quantification of grassland properties: how it can ben-
EEC (1997) The Council of the European Communities, “Council efit from geoinformatic technologies? Int J Remote Sens 27:1351–65.
Directive 97/62/EC of 27 October 1997 adapting to technical Gao T, Xu B, Yang X, et al. (2013) Using MODIS time series data to
and scientific progrss directive 92/43/EEC on the conservation of estimate aboveground biomass and its spatio-temporal variation in
natural habitats and of wild fauna and flora”. Off J Eur Commun Inner Mongolia’s grassland between 2001 and 2011. Int J Remote
pp. 42–65. Sens 34:7796–810.
EEC (1979) The Council of the European Communities, “Council Giraed CM, Benoit M, Vaubernier ED, et al. (1990) SPOT HRV data to
Directive 79/403/EEC of 2 April 1979 on the conservation of wild discriminate grassland quality. Int J Remote Sens 11:2253–67.
birds.” Off J Eur Commun pp. 1–18. Gong P, Wang J, Yu L, et al. (2013) Finer resolution observation and
Epinat V, Stein A, de Jong SM, et al. (2001) A wavelet characteriza- monitoring of global land cover: first mapping results with Landsat
tion of high-resolution NDVI patterns for precision agriculture. Int TM and ETM+ data. Int J Remote Sens 34:2607–54.
J Appl Earth Obs Geoinformation 3:121–32. Guo X, Zhang C, Wilmshurst JF, et al. (2005) Monitoring grassland
Erives H, Fitzgerald GJ (2005) Automated registration of hyper- health with remote sensing approaches. Prairie Perspect 8:11–22.
spectral images for precision agriculture. Comput Electron Agric Gutiérrez PA, López-Granados F, Peña-Barragán JM, et  al. (2008)
47:103–19. Logistic regression product-unit neural networks for mapping
Falldorf T, Strand O, Panzacchi M et al. (2014) Estimating lichen vol- Ridolfia segetum infestations in sunflower crop using multitempo-
ume and reindeer winter pasture quality from Landsat imagery. ral remote sensed data. Comput Electron Agric 64:293–306.
Remote Sens Environ 140:573–9. Gu Y, Wylie BK, Bliss NB (2013) Mapping grassland productivity with
FAO (2014) The Role of Livestock in Climate Change. http://www.fao.org/ 250-m eMODIS NDVI and SSURGO database over the Greater
agriculture/lead/themes0/climate/en/ Platte River Basin, USA. Ecol Indic 24:31–6.
FAOSTAT (2014) Food and Agriculture Organization Of The United Hajj ME, Baghdadi N, Belaud G, et al. (2014) Irrigated grassland mon-
Nations: STATISTICS DIVISION. http://faostat.fao.org/site/377/ itoring using a time series of TerraSAR-X and COSMO-SkyMed
DesktopDefault.aspx?PageID=377#ancor X-Band SAR Data. Remote Sens 6:10002–32.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 667

Hakl J, Hrevušová Z, Hejcman M, et  al. (2012) The use of a rising Inoue Y, Kurosu T, Maeno H, et al. (2002) Season-long daily meas-
plate meter to evaluate lucerne (Medicago sativa L.) height as an urements of multifrequency (Ka, Ku, X, C, and L) and full-
important agronomic trait enabling yield estimation. Grass Forage polarization backscatter signatures over paddy rice field and
Sci 67:589–96. their relationship with biological variables. Remote Sens Environ
Han G, Hao X, Zhao M, et  al. (2008) Effect of grazing intensity on 81:194–204.
carbon and nitrogen in soil and vegetation in a meadow steppe in Jachowski NRA, Quak MSY, Friess DA, et al. (2013) Mangrove bio-
Inner Mongolia. Agric Ecosyst Environ 125:21–32. mass estimation in Southwest Thailand using machine learning.
Han L-C (2001) A method of modifying error for non-synchronicity Appl Geogr 45:311–21.
of grass yield remote sensing estimation and measurement. Int J Jadhav RN, Kimothi MM, Kandya AK (1993) Grassland mapping/
Remote Sens 22:3363–72. monitoring of Banni, Kachchh (Gujarat) using remotely-sensed
Hansen JW, Jones JW (2000) Scaling-up crop models for climate vari- data. Int J Remote Sens 14:3093–103.
ability applications. Agric Syst 65:43–72. Janoth J, Gantert S, Koppe W, et al. (2012) TerraSAR-X2 - Mission
Hein L (2006) The impacts of grazing and rainfall variability on the overview. Presented at the Geoscience and Remote Sensing
dynamics of a Sahelian rangeland. J Arid Environ 64:488–504. Symposium (IGARSS), 2012 IEEE International, 217–20.
Hejcman M, Sochorová L, Pavlů V, et al. (2014) The Steinach Grassland Jianlong L, Tiangang L, Quangong C (1998) Estimating grassland
Experiment: soil chemical properties, sward height and plant spe- yields using remote sensing and GIS technologies in China. New Z
cies composition in three cut alluvial meadow after decades-long J Agric Res 41:31–8.
fertilizer application. Agric Ecosyst Environ 184:76–87. Ji B, Sun Y, Yang S, et al. (2007) Artificial neural networks for rice
Henry DA, Edirisinghe A, Donald GE, et al. (2004) “Pastures from Space” yield prediction in mountainous regions. J Agric Sci 145:249–61.
- Quantitative Estimation of Pasture Biomass and Growth Rate Using Kawamura K, Akiyama T, Yokota H, et al. (2005a) Quantifying graz-
Satellite Remote Sensing. Precision Agriculture Center, University of ing intensities using geographic information systems and satel-
Minnesota, Department of Soil, Water and Climate, 1246–54. lite remote sensing in the Xilingol steppe region, Inner Mongolia,
He Y, Guo X, Wilmshurst JF (2009) Reflectance measures of grassland China. Agric Ecosyst Environ 107:83–93.
biophysical structure. Int J Remote Sens 30:2509–21. Kawamura K, Akiyama T, Yokota H, et al. (2005b) Comparing MODIS
Hill MJ (2013) Vegetation index suites as indicators of vegetation state vegetation indices with AVHRR NDVI for monitoring the forage
in grassland and savanna: an analysis with simulated SENTINEL 2 quantity and quality in Inner Mongolia grassland, China. Grassl
data for a North American transect. Remote Sens Environ 137:94–111. Sci 51:33–40.
Hill MJ (2004) Grazing agriculture - Managed Pasture, Grassland and Kurtz DB, Schellberg J, Braun M (2010) Ground and satellite based
Rangeland. In Ustin SL (ed). Manual of Remote Sensing, Remote Sensing assessment of rangeland management in sub-tropical Argentina.
for Natural Resource Management and Environmental Monitoring. New Appl Geogr 30:210–20.
York: Wiley International, 449–530. Latham J, Cumani R, Rosati I, et al. (2014) Global Land Cover SHARE
Hill MJ, Donald GE, Hyder MW, et al. (2004) Estimation of pasture (GLC-SHARE): Database Beta-Release Version 1.0-2014. htttp://www.
growth rate in the south west of Western Australia from AVHRR glcn.org/downs/prj/glcshare/GLC_SHARE_beta_v1.0_2014.pdf.
NDVI and climate data. Remote Sens Environ 93:528–45. Lawrence R, Bunn A, Powell S, et al. (2004) Classification of remotely
Hill MJ, Donald GE, Vickery PJ, et al. (1996) Integration of satellite sensed imagery using stochastic gradient boosting as a refinement
remote sensing, simple bioclimatic models and GIS for assessment of classification tree analysis. Remote Sens Environ 90:331–6.
of pastoral development for a commercial grazing enterprise. Aust Lee J-S, Grunes MR, Pottier E (2001) Quantitative comparison of
J Exp Agric 36:309–21. classification capability: fully polarimetric versus dual and single-
Hill MJ, Ticehurst CJ, Lee J-S, et al. (2005) Integration of optical and polarization SAR. IEEE Trans Geosci Remote Sens 39:2343–51.
radar classifications for mapping pasture type in Western Australia. Lee R, Yu F, Price KP, et al. (2002) Evaluating vegetation phenological
IEEE Trans Geosci Remote Sens 43:1665–81. patterns in Inner Mongolia using NDVI time-series analysis. Int J
Hill MJ, Vickery PJ, Furnival EP, et al. (1999) Pasture land cover in Remote Sens 23:2505–12.
eastern Australia from NOAA-AVHRR NDVI and Classified Landsat Li J, Yang X, Jin Y, et al. (2013) Monitoring and analysis of grassland
TM. Remote Sens Environ 67:32–50. desertification dynamics using Landsat images in Ningxia, China.
Huang C, Geiger EL (2008) Climate anomalies provide opportunities Remote Sens Environ 138:19–26.
for large-scale mapping of non-native plant abundance in desert Liu D, Xia F (2010) Assessing object-based classification: advantages
grasslands. Divers Distrib 14:875–84. and limitations. Remote Sens Lett 1:187–94.
Huang L, Xiao T, Zhao Z, et  al. (2013) Effects of grassland restora- López-Lozano R, Casterad MA, Herrero J (2010) Site-specific man-
tion programs on ecosystems in arid and semiarid China. J Environ agement units in a commercial maize plot delineated using very
Manage 117:268–75. high resolution remote sensing and soil properties mapping.
Huang N, He J-S, Niu Z (2013) Estimating the spatial pattern of soil Comput Electron Agric 73:219–29.
respiration in Tibetan alpine grasslands using Landsat TM images Loranty MM, Goetz SJ, Beck PSA (2011) Tundra vegetation effects on
and MODIS data. Ecol Indic 26:117–25. pan-Arctic albedo. Environ Res Lett 6:024014.
Hybu Cig Cymru MPW (2008) Grassland Management. http://hcc- Loris V, Damiano G (2006) Mapping the green herbage ratio of grass-
mpw.org.uk/publications/farming_and_industry_development/ lands using both aerial and satellite-derived spectral reflectance.
grassland_management/. Agric Ecosyst Environ 115:141–9.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 668 Journal of Plant Ecology

Luo G, Han Q, Zhou D, et al. (2012) Moderate grazing can promote Bahurutshe communal grazing lands, South Africa. Int J Remote
aboveground primary production of grassland under water stress. Sens 30:3649–68.
Ecol Complex 11:126–36. Mutanga O, Skidmore AK (2004) Hyperspectral band depth analy-
Marsh P, Bartlett P, MacKay M, et al. (2010) Snowmelt energetics at sis for a better estimation of grass biomass (Cenchrus ciliaris) meas-
a shrub tundra site in the western Canadian Arctic. Hydrol Process ured under controlled laboratory conditions. Int J Appl Earth Obs
24:3603–20. Geoinformation 5:87–96.
Maselli F, Argenti G, Chiesi M, et al. (2013) Simulation of grassland Nagendra H (2001) Using remote sensing to assess biodiversity. Int J
productivity by the combination of ground and satellite data. Agric Remote Sens 22:2377–400.
Ecosyst Environ 165:163–72. Nagendra H, Gadgil M (1999) Satellite imagery as a tool for monitor-
Maselli F, Barbati A, Chiesi M, et al. (2006) Use of remotely sensed ing species diversity: an assessment. J Appl Ecol 36:388–97.
and ancillary data for estimating forest gross primary productivity Nahry AHE, Ali RR, Baroudy AAE (2011) An approach for precision
in Italy. Remote Sens Environ 100:563–75. farming under pivot irrigation system using remote sensing and
Masocha M, Skidmore AK (2011) Integrating conventional classifiers GIS techniques. Agric Water Manag 98:517–31.
with a GIS expert system to increase the accuracy of invasive spe- Newnham G (2010) Improved Methods for Assessment and Prediction of
cies mapping. Int J Appl Earth Obs Geoinformation 13:487–94. Grassland Curing. Satellite Based Curing Methods and Mapping - Final
Mata G, Edirisinghe A, Waugh D, et al. (2010) Algorithm improve- Report: Project A1.4. http://www.bushfirecrc.com/sites/default/
ments to deliver higher accuracy of prediction of pasture cover files/managed/resource/a1-4_final_report_0.pdf
from satellite data in Waikato Dairy Farms. Presented at the Nieland S, Förster M, Kleinschmit B (2012) A remote sens-
Australasian Dairy Symposium, Lincoln, New Zealand, 225–9. ing biodiversity monitoring ontology. In Gensel J, Josselin
Mekasha A, Gerard B, Tesfaye K, et  al. (2014) Inter-connection D, Vandenbroucke D (eds). Proceedings of the AGILE’2012
between land use/land cover change and herders’/farmers’ International Conference on Geographic Information Science. Avignon,
livestock feed resource management strategies: a case study France: Springer, pp. 312–5.
from three Ethiopian eco-environments. Agric Ecosyst Environ Nitze I, Barrett B, Cawkwell F (2015) Temporal optimisation of image
188:150–62. acquisition for land cover classification with Random Forest and
Metz A, Schmitt A, Esch T, et al. (2012) Synergetic use of TerraSAR-X MODIS time-series. Int J Appl Earth Obs Geoinformation 34:136–46.
and Radarsat-2 time series data for identification and charac- Nordberg M-L, Evertson J (2003) Monitoring change in mountain-
terization of grassland types - a case study in Southern Bavaria, ous dry-heath vegetation at a regional scale using multitemporal
Germany. Presented at the Geoscience and Remote Sensing landsat TM Data. Ambio 32:502–9.
Symposium (IGARSS), 2012 IEEE International, 3560–3.
Numata I, Roberts DA, Chadwick OA, et al. (2007) Characterization of
Miehe G, Miehe S, Bach K, et al. (2011) Plant communities of central pasture biophysical properties and the impact of grazing intensity
Tibetan pastures in the Alpine Steppe/Kobresia pygmaea ecotone. J using remotely sensed data. Remote Sens Environ 109:314–27.
Arid Environ 75:711–23.
O’Connor B, Dwyer E, Cawkwell F, et al. (2012) Spatio-temporal pat-
Moore AD, Donnelly JR, Freer M (1997) GRAZPLAN: Decision support terns in vegetation start of season across the island of Ireland using
systems for Australian grazing enterprises. III. Pasture growth and the MERIS Global Vegetation Index. ISPRS J Photogramm Remote
soil moisture submodels, and the GrassGro DSS. Agric Syst 55:535–82. Sens 68:79–94.
Moore AD, Vickery PJ, Hill MJ, et al. (1999) Combining satellite data Oliver YM, Robertson MJ, Wong MTF (2010) Integrating farmer
with a simulation model to describe spatial variability in pasture knowledge, precision agriculture tools, and crop simulation model-
growth at a farm scale. Aust J Exp Agric 39:285–300. ling to evaluate management options for poor-performing patches
Moran MS, Inoue, Y, Barnes EM (1997) Opportunities and limitations in cropping fields. Eur J Agron 32:40–50.
for image-based remote sensing in precision crop management. Panda SS, Ames DP, Panigrahi S (2010) Application of vegetation
Remote Sens Environ 61:319–46. indices for agricultural crop yield prediction using neural network
Morán-Ordóñez A, Suárez-Seoane S, Calvo L, et al. (2011) Using pre- techniques. Remote Sens 2:673–96.
dictive models as a spatially explicit support tool for managing cul- Paudel KP, Andersen P (2010) Assessing rangeland degradation using
tural landscapes. Appl Geogr 31:839–48. multi temporal satellite images and grazing pressure surface model
Moreau S, Le Toan T (2003) Biomass quantification of Andean wet- in Upper Mustang, Trans Himalaya, Nepal. Remote Sens Environ
land forages using ERS satellite SAR data for optimizing livestock 114:1845–55.
management. Remote Sens Environ 84:477–92. Payet N, Todorovic S (2010) (RF)2—Random Forest Random Field.
Morel J, Todoroff P, Bégué A, et  al. (2014) Toward a satellite-based In Lafferty JD, Williams CKI, Shawe-Taylor J, et al. (eds). Advances
system of sugarcane yield estimation and forecasting in small- in Neural Information Processing Systems 23. Curran Associates, Inc.,
holder farming conditions: a case study on Reunion Island. Remote 1885–93.
Sens 6:6620–35. Peña-Barragán JM, Ngugi MK, Plant RE, et  al. (2011) Object-based
Mountrakis G, Im J, Ogole C (2011) Support vector machines crop identification using multiple vegetation indices, textural fea-
in remote sensing: a  review. ISPRS J Photogramm Remote Sens tures and crop phenology. Remote Sens Environ 115:1301–16.
66:247–59. Phillips R, Beeri O, Scholljegerdes E, et al. (2009) Integration of geo-
Munyati C, Makgale D (2009) Multitemporal Landsat TM imagery spatial and cattle nutrition information to estimate paddock graz-
analysis for mapping and quantifying degraded rangeland in the ing capacity in Northern US prairie. Agric Syst 100:72–9.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 669

Pickup G, Bastin GN, Chewings VH (1998) Identifying trends in land Schröder C, Garcia F, Buck O, et  al. (2013) Validation of habitat
degradation in non-equilibrium rangelands. J Appl Ecol 35:365–77. assessment services in Natura 2000 sites: streamlining the role
Pickup G, Bastin GN, Chewings VH (1994) Remote-sensing-based of stakeholders from user needs to conservation status evalu-
condition assessment for nonequilibrium rangelands under large- ation. In Jekel T, Car A, Strobl J, et  al. (eds). GI_Forum 2013
scale commercial grazing. Ecol Appl 4:497–517. – Creating the GI Society – Conference Proceedings. Wichmann,
Salzburg, 548–51.
Pinar A, Curran PJ (1996) Technical Note. Grass chlorophyll and the
reflectance red edge. Int J Remote Sens 17:351–7. Schuster C, Ali I, Lohmann P, et al. (2011) Towards detecting swath
events in TerraSAR-X time series to establish NATURA 2000 grass-
Piñeiro G, Oesterheld M, Paruelo JM (2006a) Seasonal variation in
land habitat swath management as monitoring parameter. Remote
aboveground production and radiation-use efficiency of temperate
Sens 3:1308–22.
rangelands estimated through. Remote Sens Ecosyst 9:357–73.
Schuster C, Förster M, Kleinschmit B (2012) Testing the red edge
Piñeiro G, Paruelo JM, Oesterheld M (2006b) Potential long-term
channel for improving land-use classifications based on high-res-
impacts of livestock introduction on carbon and nitrogen cycling
olution multi-spectral satellite data. Int J Remote Sens 33:5583–99.
in grasslands of Southern South America. Glob Change Biol
12:1267–84. Schuster C, Schmidt T, Conrad C, et al. (2015) Grassland habitat map-
ping by intra-annual time series analysis – Comparison of RapidEye
Pokluda P, Hauck D, Cizek L (2012) Importance of marginal habitats
and TerraSAR-X satellite data. Int J Appl Earth Obs Geoinformation
for grassland diversity: fallows and overgrown tall-grass steppe as
34:25–34.
key habitats of endangered ground-beetle Carabus hungaricus. Insect
Conserv Divers 5:27–36. Scurlock JMO, Hall DO (1998) The global carbon sink: a grassland
perspective. Glob Change Biol 4:229–33.
Prabhakar M, Prasad YG, Thirupathi M, et al. (2011) Use of ground
based hyperspectral remote sensing for detection of stress in cotton Serele CZ, Gwyn QHJ, Boisvert JB, et al. (2000) Corn yield predic-
caused by leafhopper (Hemiptera: Cicadellidae). Comput Electron tion with artificial neural network trained using airborne remote
Agric 79:189–98. sensing and topographic data. Presented at the Geoscience and
Remote Sensing Symposium (IGARSS), 2000 IEEE International,
Price KP, Guo X, Stiles JM (2002) Comparison of Landsat TM and
384–6.
ERS-2 SAR data for discriminating among grassland types and
treatments in eastern Kansas. Comput Electron Agric 37:157–71. Shoshany M, Long D, Bonfil D (2013) Remote sensing for sustainable
agriculture. Int J Remote Sens 34:6021–3.
Prince SD (1991) Satellite remote sensing of primary production:
comparison of results for Sahelian grasslands 1981–1988. Int J Sitters J, Heitkönig IMA, Holmgren M, et al. (2009) Herded cattle and
Remote Sens 12:1301–11. wild grazers partition water but share forage resources during dry
years in East African savannas. Biol Conserv 142:738–50.
Psomas A, Kneubühler M, Huber S, et  al. (2011b) Hyperspectral
remote sensing for estimating aboveground biomass and for Si Y, Schlerf M, Zurita-Milla, R, et al. (2012) Mapping spatio-temporal
exploring species richness patterns of grassland habitats. Int J variation of grassland quantity and quality using MERIS data and
Remote Sens 32:9007–31. the PROSAIL model. Remote Sens Environ 121:415–25.
Punjabi GA, Chellam R, Vanak AT (2013) Importance of native grass- Smith AM, Buckley JR (2011) Investigating RADARSAT-2 as a tool
land habitat for den-site selection of Indian foxes in a fragmented for monitoring grassland in western Canada. Can J Remote Sens
landscape. PLOS ONE 8:e76410. 37:93–102.
Quan X, He B, Li X, et  al. (2015) Estimation of grassland live fuel Song C, Dannenberg MP, Hwang T (2013) Optical remote sens-
moisture content from ratio of canopy water content and foliage ing of terrestrial ecosystem primary productivity. Prog Phys Geogr
dry biomass. IEEE Geosci Remote Sens Lett 12:1903–7. 37:834–54.
Quattrochi DA, Goodchild MF (1997) Scale in Remote Sensing and GIS. Soussana J-F, Lemaire G (2014) Coupling carbon and nitrogen cycles
Boca Raton, Boston, London, New York, Washington DC: Lewis for environmentally sustainable intensification of grasslands and
Publishers; CRC Press. crop-livestock systems. Agric Ecosyst Environ 190:9–17.
Reiche M, Funk R, Zhang Z, et  al. (2012) Application of satellite Sulieman HM, Elagib NA (2012) Implications of climate, land-use
remote sensing for mapping wind erosion risk and dust emission- and land-cover changes for pastoralism in eastern Sudan. J Arid
deposition in Inner Mongolia grassland, China. Grassl Sci 58:8–19. Environ 85:132–41.
Retzer V (2006) Impacts of grazing and rainfall variability on the Svoray T, Perevolotsky A, Atkinson PM (2013) Ecological sustainabil-
dynamics of a Sahelian rangeland revisited (Hein, 2006))—new ity in rangelands: the contribution of remote sensing. Int J Remote
insights from old data. J Arid Environ 67:157–64. Sens 34:6216–42.
Röder A, Udelhoven T, Hill J, et al. (2008) Trend analysis of Landsat-TM Tappan G (1982) A Review of Remote Sensing and Grasslands
and -ETM+ imagery to monitor grazing impact in a rangeland eco- Literature. http://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/19
system in Northern Greece. Remote Sens Environ 112:2863–75. 820016679.pdf.
Schellberg J, Hill MJ, Gerhards R, et al. (2008) Precision agriculture Thenkabail PS (2003) Biophysical and yield information for precision
on grassland: applications, perspectives and constraints. Eur J farming from near-real-time and historical Landsat TM images. Int
Agron 29:59–71. J Remote Sens 24:2879–904.
Schellberg J, Verbruggen E (2014) Frontiers and perspectives on research Thorvaldsson G, Bjornsson H, Hermannsson J (2004) The influence
strategies in grassland technology. Crop Pasture Sci 65:508–23. of weather on early growth rate of grasses. Icel Agric Sci 4:65–73.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 670 Journal of Plant Ecology

Todd SW, Hoffer RM, Milchunas DG (1998) Biomass estimation on Wang C, Jamison BE, Spicci AA (2010) Trajectory-based warm sea-
grazed and ungrazed rangelands using spectral indices. Int J Remote son grassland mapping in Missouri prairies with multi-temporal
Sens 19:427–38. ASTER imagery. Remote Sens Environ 114:531–9.
Toivonen T, Luoto M (2003) Landsat TM images in mapping of semi- Wang X, Ge L, Li X (2013) Pasture monitoring using SAR with
natural grasslands and analysing of habitat pattern in an agricul- COSMO-SkyMed, ENVISAT ASAR, and ALOS PALSAR in Otway,
tural landscape in south-west Finland. Int J Geogr 181:49–67. Australia. Remote Sens 5:3611–36.
Tovar C, Seijmonsbergen AC, Duivenvoorden JF (2013) Monitoring Wang Y, Zhou G, Jia B (2008) Modeling SOC and NPP responses of
land use and land cover change in mountain regions: an example meadow steppe to different grazing intensities in Northeast China.
in the Jalca grasslands of the Peruvian Andes. Landsc Urban Plan Ecol Model 217:72–8.
112:40–9. Weiers S, Bock M, Wissen M, et  al. (2004) Mapping and indicator
Trans (2014) TRANS: Transhumance, Natural Resources, and Conflict in the approaches for the assessment of habitats at different scales using
Sahel. http://lcccrsp.org/focusregions/west-africa/trans/ (25 March remote sensing and GIS methods. Landsc Urban Plan 67:43–65.
2014, date last accessed). Wen Q, Zhang Z, Liu S, et al. (2010) Classification of grassland types
Trotter M (2013) PA Innovations in Livestock, Grazing Systems and by MODIS time-series images in Tibet, China. IEEE J Sel Top Appl
Rangeland Management to Improve Landscape Productivity and Earth Obs Remote Sens 3:404–9.
Sustainability. http://search.informit.com.au/documentSumma Williamson HD, Eldridge DJ (1993) Pasture status in a semi-arid
ry;dn=410039721522825;res=IELHSS (2 March 2015, date last grassland. Int J Remote Sens 14:2535–46.
accessed).
Woodward SJR (2001) Validating a model that predicts daily growth
Tueller PT (1989) Remote sensing technology for rangeland manage-
and feed quality of New Zealand dairy pastures. Environ Int
ment applications. J Range Manag 42:442–53.
27:133–7.
Ullah S, Si Y, Schlerf M, et al. (2012) Estimation of grassland biomass
Wu H, Li B-L, Stoker R, et al. (1996) A semi-arid grazing ecosystem
and nitrogen using MERIS data. Int J Appl Earth Obs Geoinformation
simulation model with probabilistic and fuzzy parameters. Ecol
19:196–204.
Model 90:147–60.
Vanden Borre J, Haest B, Lang S, et  al. (2011a) Towards a wider
Wu H, Li Z-L (2009) Scale issues in remote sensing: a review on anal-
uptake of remote sensing in Natura 2000 monitoring: streamlin-
ysis, processing and modeling. Sensors 9:1768–93.
ing remote sensing products with users’ needs and expectations.
Presented at the 2011 2nd International Conference on Space Wylie BK, Harrington JA, Prince SD, et  al. (1991) Satellite and
Technology (ICST), 1–4. ground-based pasture production assessment in Niger: 1986–1988.
Vanden Borre J, Paelinckx D, Mücher CA, et al. (2011b) Integrating Int J Remote Sens 12:1281–300.
remote sensing in Natura 2000 habitat monitoring: prospects on Wylie B, Meyer D, Tieszen L, et al. (2002) Satellite mapping of sur-
the way forward. J Nat Conserv 19:116–25. face biophysical parameters at the biome scale over the North
van Swaay CAM (2002) The importance of calcareous grasslands for American grasslands: a case study. Remote Sens Environ 79:266–78.
butterflies in Europe. Biol Conserv 104:315–8. Xie Y, Sha Z, Yu M (2008) Remote sensing imagery in vegetation
Velazquez J, Förster M, Kleinschmit B (2008) GIS and remote sens- mapping: a review. J Plant Ecol 1:9–23.
ing for Natura 2000 monitoring in Mediterranean biogeographic Xie Y, Sha Z, Yu M, et al. (2009) A comparison of two models with
region. In Car A, Griesebner G, Strobl J (eds). Proceedings of the Landsat data for estimating above ground grassland biomass in
Geoinformatics Forum Salzburg. Wichmann, Heidelberg, 268–75. Inner Mongolia, China. Ecol Model 220:1810–8.
Verbesselt J, Somers B, van Aardt J, et  al. (2006) Monitoring her- Xu B, Yang XC, Tao WG, et  al. (2008) MODIS‐based remote sensing
baceous biomass and water content with SPOT VEGETATION monitoring of grass production in China. Int J Remote Sens 29:5313–27.
time-series to improve fire risk assessment in savanna ecosystems.
Xu B, Yang X, Tao W, et al. (2007) Remote sensing monitoring upon
Remote Sens Environ 101:399–414.
the grass production in China. Acta Ecol Sin 27:405–13.
Vermeire LT, Heitschmidt RK, Haferkamp MR (2008) Vegetation
Yang C, Everitt JH (2010) Mapping three invasive weeds using air-
response to seven grazing treatments in the Northern Great Plains.
borne hyperspectral imagery. Ecol Inform 5:429–39.
Agric Ecosyst Environ 125:111–9.
Yang J, Gong P, Fu R, et al. (2013) The role of satellite remote sensing
Vescovo L, Gianelle D (2008) Using the MIR bands in vegetation indi-
in climate change studies. Nat Clim Change 3:875–83.
ces for the estimation of grassland biophysical parameters from
satellite remote sensing in the Alps region of Trentino (Italy). Adv Yang X, Guo X (2011) Investigating vegetation biophysical and spec-
Space Res 41:1764–72. tral parameters for detecting light to moderate grazing effects: a
case study in mixed grass prairie. Cent Eur J Geosci 3:336–48.
Vickery PJ, Hill MJ, Donald GE (1997) Satellite derived maps of pas-
ture growth status: association of classification with botanical com- Yang X, Guo X, Fitzsimmons M (2012) Assessing light to moderate
position. Aust J Exp Agric 37:547–62. grazing effects on grassland production using satellite imagery. Int
Volesky JD, Schacht WH, Richardson DM (2004) Stocking rate and J Remote Sens 33:5087–104.
grazing frequency effects on Nebraska Sandhills meadows. J Range Yang X, Wilmshurst JF, Fitzsimmons M, et  al. (2011) Can satellite
Manag 57:553. imagery evaluate the pre-condition of a grazing experiment?
Voormansik K, Jagdhuber T, Olesk A, et al. (2013) Towards a detection Prairie Perspect Geogr Essays 14:45–50.
of grassland cutting practices with dual polarimetric TerraSAR-X Yang X, Xu B, Yunxiang J, et al. (2012) On grass yield remote sensing
data. Int J Remote Sens 34:8081–3. estimation models of China’s Northern Farming-Pastoral Ecotone.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018
 Ali et al.     |     Satellite remote sensing of grasslands 671

In Lee G (ed). Advances in Computational Environment Science, Advances condition and restoration potential of eucalypt grassy woodlands.
in Intelligent and Soft Computing. Springer, Berlin, 281–91. Landsc Urban Plan 102:226–33.
Yiran GAB, Kusimi JM, Kufogbe SK (2012) A synthesis of remote Zhang M, Qin Z, Liu X, et al. (2003) Detection of stress in tomatoes
sensing and local knowledge approaches in land degradation induced by late blight disease in California, USA, using hyperspec-
assessment in the Bawku East District, Ghana. Int J Appl Earth Obs tral remote sensing. Int J Appl Earth Obs Geoinformation 4:295–310.
Geoinformation 14:204–13. Zha Y, Gao J, Ni S, et al. (2003) A spectral reflectance-based approach
Yu L, Zhou L, Liu W, et al. (2010) Using remote sensing and GIS tech- to quantification of grassland cover from Landsat TM imagery.
nologies to estimate grass yield and livestock carrying capacity of Remote Sens Environ 87:371–5.
alpine grasslands in Golog Prefecture, China. Pedosphere 20:342–51. Zheng S, Meng K, Sun Y (2011) Grassland degradation in the Western
Zerger A, Mcintyre S, Gobbett D, et  al. (2011) Remote detection of of Songnen plain. Presented at the 2011 19th International
grassland nutrient status for assessing ground layer vegetation Conference on Geoinformatics, 1–5.

Downloaded from https://academic.oup.com/jpe/article-abstract/9/6/649/2623732

by guest
on 21 February 2018