Automation in Construction 132 (2021) 103916

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

Review

A review on flood management technologies related to image processing

and machine learning
Hafiz Suliman Munawar a, Ahmed W.A. Hammad a, *, S. Travis Waller b
a
School of Built Environment, University of New South Wales, Kensington 2052, NSW, Australia
b
School of Civil and Environmental Engineering, University of New South Wales, Kensington 2052, NSW, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: Flood management, which involves flood prediction, detection, mapping, evacuation, and relief activities, can be
Flood management improved via the adoption of state-of-the-art tools and technology. Focusing on ways to mitigate floods and
Image processing provide a quick response after floods is critical to ensuring fatalities are minimized, along with reducing envi­
Machine learning
ronmental and economic damages. In the literature, techniques from different domains including remote sensing,
Disaster management
machine learning, image processing and data analysis have been explored to manage different tasks related to
flood management. This study proposes a new framework that categorizes the recent research that has been
conducted on flood management systems. The framework addresses the following significant research questions:
(1) What are the major techniques deployed in flood management? (2) What are the phases of flood management
which existing studies tend to focus on? (3) What are the systems that are proposed to tackle problems related to
flood management? (4) What are the research gaps identified in the literature when it comes to deploying
technology for flood management? A classification framework for flood management has been proposed to group
the various technologies reviewed. Lack of hybrid models, which combine image processing and machine
learning, for flood management was observed. In addition, the application of machine learning-based methods in
the post-disaster scenario was found to be limited. Thus, future efforts need to focus on combining disaster
management knowledge, image processing techniques and machine learning tools to ensure effective and holistic
disaster management across all phases.

1. Introduction are deemed crucial for understanding the occurrences of floods.

Expansion of commercial and residential areas has led to the develop­
The cutting-edge technologies in recent years have changed the way ment of communities in the coastal areas and river basins, which are
that the world operates. Like every other field, the area of disaster among the regions naturally prone to flood risks [6]. Additionally, land
management is increasingly moving towards the use of modern tech­ use and critical infrastructure are important factors that determine the
nology. Both developing and developed countries are constantly facing extent of floods and the extent of their propagation.
flood risks [1]. Predictive analysis into future flood risks reveals that the The true extent to which climate change and urbanization impact the
increasing impact of climate change [2], along with insufficient pre­ levels of floods remains to be explored. Increased urbanization has
paredness in many regions around the world for flood events is likely to induced flood plains among the residential areas in many regions around
result in a historical level of flood-related damages [3]. Flood instances the world, thus resulting in increased flood risks [8]. Various studies
around the world that have been on the rise [1], necessitate the need for have focused on analysing flood risks in a country or region [9]. Yet, as
identifying effective ways of risk management in the face of disasters. technology progresses, developments in the data-driven models and
The worldwide loss of lives, crops, infrastructure and economic re­ their contributions to the field of flood risk management are worthy of a
sources can all be linked with flooding events [3]. Wang et al. [4] esti­ thorough analysis.
mated that the risks and losses associated with floods are higher than Currently, the risk of floods is mapped on a global scale using tech­
that of any other climatic hazard. The increased instances of floods are nology like satellite imagery and remote sensing [10]. Such imple­
linked to an increase in population and climate change [5]. Both factors mentation on a local level has not yet been successful, given that floods

* Corresponding author.
E-mail addresses: [email protected] (H.S. Munawar), [email protected] (A.W.A. Hammad), [email protected] (S.T. Waller).

https://doi.org/10.1016/j.autcon.2021.103916
Received 22 November 2020; Received in revised form 13 August 2021; Accepted 19 August 2021
Available online 9 September 2021
0926-5805/© 2021 Elsevier B.V. All rights reserved.
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

are a very dynamic process and the rate of imagery capture via satellites approaches are somehow related in that they share certain techniques
is very slow. There are limitations associated with analysing, capturing and algorithms and iii) there is a clear indication in the field for the need
and understanding the full scale of flood risks using just remotely sensed to automate the prediction, detection, and management process when it
images [10]. Besides, floods are known to cause significant damage to comes to floods. Emphasis is placed on studies that deploy more recent
critical infrastructure, impacting roads, bridges and the telecommuni­ approaches from fields like AI, machine learning, image processing and
cation network, thus making it difficult to reach people stuck in flooded computer vision, more precisely, the following major questions are
areas [11]. According to a report “Financial Management of Flood Risk” addressed in the research classification framework proposed:
published by the Organization for Economic Cooperation and Devel­
opment on flood-related economic loss in the year 2016, there was a 1. What are the major techniques being used for flood management?
total of 40 billion US dollars that the world lost owing to floods in 2. On which phases of disaster management do the existing studies tend
various regions of the world. Failure of telecommunication systems to focus?
during floods will vastly impact flood map generation, thus hampering 3. How the problems related to flood management are being tackled?
risk management and rescue operations [12]. To increase the level of 4. What are the identified research gaps, when it comes to the use of
disaster preparedness, information related to flood risks needs to be technology for flood management?
collected on a local level. With the recent development in technology, it
has now become possible to integrate computer models with remote To answer these questions, a rigorous process of literature search and
sensing which makes it possible to simulate dynamic events like a flood analysis is conducted.
in space and time continuously [13–17]. One of the techniques that have This study is structured as follows: The next section discusses the
prominently developed over the years is that which relies on image methodology adopted to collect the research articles for the review and
processing and machine learning. This review aims to therefore track the the strategy used for article retrieval. Section 3 presents the classifica­
recent developments in disaster management approaches that rely on tion framework devised for this study. Section 4 presents the results of
image processing and machine learning. Analysing the existing tech­ this review, answering the major research questions in detail. Section 5
niques for identifying flooded areas along with the gaps in methods discusses the various gaps in the literature that were identified in this
currently used will help in developing and proposing a model for research based on the limitations of the technology used and the limited
enhancing flood analysis. A crucial point that needs attention is the lack application of the technique concerning the disaster management cycle.
of focus on post-disaster management systems in the literature that links Section 6 discusses the major achievements of the study and possible
with existing methods used for detecting floods way [18]. This study will future directions.
aim to ensure that the latest technology is reviewed and analysed for
enhancing the post-disaster management systems. A novel classification 2. Materials and methods
framework has been developed to categorize the studies according to the
disaster management phase that they are dealing with and the domain of To answer the review questions raised above, a systematic literature
the technology being used in them. The framework will accurately review procedure is adopted which relies on 3 main steps: first, relevant
identify for which phase is the technology designed and to which literature is collected using online search engines including Springer,
domain does it belong. Once these factors are identified, each technique Google Scholar, Scopus, Association for Computing Machinery (ACM),
is studied further in detail to analyse its method, application, outcomes, Science Direct and others. For this purpose, specific keywords were
and drawbacks. adopted when conducting the literature review search. Semantic
Currently, the focus of flood risk management systems is mainly on searches were utilized to retrieve articles that are most relevant to these
predicting floods and creating maps for determining the areas prone to keywords. Second, the articles were filtered after assessing them for
disaster [19]. Techniques such as artificial intelligence (AI) - based al­ relevance. This involved reading and examining the abstracts to deter­
gorithms machine learning and computer vision have been proposed mine the relevance of the research. Third, the selected research articles
[20] to overcome many limitations of the traditional methods [21]. were classified into two categories: image processing and machine
Several surveys have been conducted on flood management techniques. learning-based techniques. Approaches adopted in developing and
These include a review on flood forecasting technologies [22], using developed countries were considered to highlight the existing ap­
machine learning for flood prediction [20] systematic literature reviews proaches utilized for predicting, detecting, and managing floods
on using big data for disaster management [23] and flood mapping and worldwide. Traditional methods being used for flood management,
assessment systems [24]. However, these classical review papers seem to including Global Positioning System (GPS) and Global Information
target only flood prediction, risk assessment and forecasting technolo­ System (GIS), were examined, along with complex image processing and
gies to map areas that are prone to disaster [19], and they lack in ana­ machine learning-based approaches. Third, an integrative category that
lysing the technologies being used to detect flood water, locate affected involves combining image processing techniques with machine learning
people and deal with after-effects of the disaster. Besides, most of these algorithms was proposed as an alternative to improve the detection and
do not consider the latest techniques based on machine learning and rescue operations. The solution will be a conceptual framework
image processing [25]. Past work had focused on the traditional comprising both image processing and machine learning techniques to
methods for flood management which includes satellite imaging, remote manage floods more efficiently.
sensing and hyper-ion imaging [26]. Image processing methods For the keywords adopted, terms like “flood management tech­
currently adopted are rarely discussed in the existing literature. Also, niques”, “Image processing-based flood management”, “flood manage­
there is a lack of focus on the later phases of the disaster management ment using machine learning”, “post-flood management” and “flood
process like disaster response and recovery, which are imperative given detection using machine learning”. Databases including Google Scholar,
that it is not always possible to accurately predict natural disasters [27]. Science Direct, ACM, Springer and Scopus were used to carry out the
This study will present a systematic review of the literature by covering: search process. Initially, separate keywords were adopted; these were
i) disaster management in the pre and post-disaster phases, focusing on then combined using key phrases like “flood management methodolo­
the use of recent state-of-the-art technologies based on machine gies”. Such search phrases and keywords were designed carefully to
learning, image processing and artificial intelligence ii) comprehensive exhaust the database and find the most relevant research articles for the
framework for the classification of the studies. study that has been utilized in review-based studies [28–30]. The
The importance of reviewing machine learning and image processing selected articles were written in English. Only research papers were
techniques in flood management is as follows: i) these approaches have considered and all the articles in the form of editorials, comments, let­
developed drastically with the introduction of technology; ii) both ters, or notes were dismissed.

2
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

The screening process for filtering the articles identified is displayed 70

in Fig. 1. A four-step screening process was implemented based on an
60 65
assessment criterion that was developed and justified by researchers to
reduce the bias regarding the subjectivity of the retrieved articles as 50
50
proposed by Tranfield [31]. First, the results are limited to articles dis­ 40
cussing “floods” and “flood management”. Second, results are filtered to 30
incorporate the technology aspect in flood management. Third, rele­
vance to the domains of image processing and machine learning is 20
considered in filtering out the papers. Fourth, the search results were 10 15
10
restricted based on language, document type, time, and duplication of 0
the search results. These limits are collectively represented by the Image Processing Machine Both ML and IP Other
“Categorization” block. As the figure depicts, after each screening pro­ Learning
cess, the resultant articles became less in number. Finally, a total of 140
articles were retrieved. Image Processing Machine Learning Both ML and IP Other
When filtering the articles, a condition instilled was that the research Fig. 2. Distribution of articles retrieved.
was not to be published before 2010. This ensures that the latest
research in the last decade is captured. The absence of desired keywords
3. Classification framework
in the abstract and written language other than English was filtered. The
distribution of retrieved articles among different categories is shown in
In this survey, a classification framework is proposed that is based on
Fig. 2. Table 1 shows the number of articles retrieved through various
a set of identified research fields in flood management related to pre-
search engines using keywords and search phrases. A total of 140 studies
disaster and post-disaster phases. The techniques are further classified
were found related to our research. Out of these 140 studies, 50 focused
based on the domain to which they belong. The identified technological
on the use of machine learning for predicting floods, 65 studies explored
domains include i) image processing, ii) machine learning and iii)
the use of image processing for analysing flood risks, 15 studies utilized
hybrid models. Fig. 3 illustrates the proposed classification framework,
remote sensing methods, including GPS and GIS, for predicting floods,
showing the different categories and sub-categories chosen to group the
and only 10 studies focused on the use of machine learning and image
retrieved research articles. As shown in this figure, the identified studies
processing for post-disaster management [32].
are based on either pre or post-disaster phase. Hence, these are the main
categories to group the techniques. The selected studies either propose

Fig. 1. The screening process for filtering of 140 articles.

3
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

Table 1 (SAR) and Unmanned Aerial Vehicles (UAVs) imagery. The machine
Articles retrieval through search engines. learning models used in the pre and post-disaster phases are Artificial
Search Engine Keyword/Search Phrase Articles Duplicate Neural Networks (ANN), Support Vector Machine (SVM), adaptive
Retrieved neuro-fuzzy inference system (ANFIS), Wavelet Neural Network (WNN)
Google Scholar, “Flood Management Tech OR 151 2 and Multilayer Perceptron (MLP) models. Hybrid techniques include the
ACM, Science Flood Management Technology integration of machine learning models such as SVM along with edge
Direct OR Flood management technique detection and using ANN and SVM models along with pixel-based clas­
OR flood management using sification on images.
technology”
English Language only 100 0
Document type: article or book 4. Results
chapter or abstract or proceedings
paper (1) What are the major techniques deployed in flood management?
Timespan: 2010–2019
Several activities that are carried out during the flood management
No Duplicates
Scopus “flood management using image 27 4 process have been discussed in the reviewed articles. The main tech­
processing OR image processing- niques can be classified as i) Flood Prediction; ii) Flood Detection; iii)
based flood management OR use of Flood Mapping; and iv) Flood Hazard or Risk Assessment. Flood pre­
image processing for flood
diction deals with forecasting the occurrence of flood events. This in­
management”
English Language only 10 0
volves using data related to rainfall, monitoring changes in the river
Document type: article or book water levels, knowledge about the moisture produced by storm events,
chapter or abstract or proceedings data about the drainage basin of rivers like soil moisture, vegetation
paper levels, temperature and other factors. Such factors are monitored
Timespan: 2010–2019
throughout the region, to report about risks of any future flood event
No Duplicates
Springer “flood management using machine 40 5 and its possible severity. These activities are grouped under the pre-
learning OR using machine disaster phase. Flood detection involves monitoring water levels and
learning for flood management OR issuing warnings when a flood is detected. Currently, this is done using
machine learning-based flood
remote sensing and satellite data to detect river flood levels in real-time.
management”
Document type: article or book 30 0
Flood mapping deals with highlighting areas that are prone to flood or
chapter or abstract or proceedings are already flooded. The generated maps can be used to carry out flood
paper risk analysis and to locate the flooded regions, find stranded people and
Timespan: 2010–2019 determine the paths to reach them. Flood risk or hazard assessment can
No Duplicates
be considered as a flood mitigation procedure that involves estimating
Total Articles Retrieved 140
the risk of flood events from all flooding mechanisms to determine what
actions are to be performed to deal with floods.
Each category is mapped to some appropriate analytic classes, as
listed in Table 2. The collected research articles were first classified
based on the disaster management phase to which they belong. These
phases can be classified into four categories. The first one deals with
flood mitigation which takes place before a disaster occurs. The tech­
niques listed in this category involve risk assessment, flood hazard
analysis and flood prediction. The second phase deals with disaster
preparedness. The third phase is labelled as disaster response which
occurs in the post-disaster scenario and relates to rescue and relief ser­
vices. The fourth and last phase deals with overcoming the effects of
disaster for the rehabilitation of people and recovering the damage. The
techniques belonging to each of these phases are classified based on
their use of AI, image processing and machine learning-based methods.
As there were many overlaps found in the literature between different
disaster management phases, the selected articles were generally
distributed into two categories which are: Pre-disaster and post-disaster.
After identifying the phase, the techniques presented in the study were
grouped into either machine learning, image processing and hybrid
categories. Many studies employing remote sensing approaches like
satellite imagery, GIS and GPS integrated either image processing, ma­
chine learning or both areas. Such studies were grouped into a category
labelled as Hybrid. The image processing category consists of studies
discussing techniques based on edge detection, object detection and
pixel analysis for flood management. The machine learning category
contains studies using trained statistical models for mapping and
Fig. 3. Framework for classification of technologies. detecting flooded regions, making predictions to forecast future flood
events or flood risks and assessment of the damage caused by a flood.
an image processing method, a machine learning model, or a combi­ The major aim of this categorization was to classify the articles based on
nation of both. Therefore, the subcategories are image processing ma­ the fields to which the proposed respective algorithm adopted belongs
chine learning and hybrid. The major methods and technologies used and to identify and then classify each study to its appropriate flood
from the domain of image processing for flood management include management phase. Using the proposed framework, the tasks belonging
edge detection, pixel analysis, remote sensing, Synthetic Aperture Radar to each phase of the disaster management process can be tackled in the
light of the most recent approaches, using image processing, machine

4
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

Table 2
Distribution of studies among pre-and post-disaster phases.
Category Image Processing Machine Learning Hybrid

Pre-Disaster: Edge Remote SAR UAV Hyperion GIS GPS ANN MLP ANFIS WNN SVM N/A
Disaster Detection Sensing Imagery Imaging Imaging
Mitigation & [34] [35]; [37] [38]; [39] N/A N/A N/A [40]; [44]; [46]; [49] [51]
Disaster [36]; [41–43] [45] [47] [50]
Preparedness [10] [48]
Post-Disaster: [52] [11]; N/A [54]; [56]; [57]; [60] [61]; [64–66]; N/A [67,68]
Disaster [53] [55] [58]; [59] [62]; [63]
Response & [10] [63]
Recovery

learning or both. As shown in Table 2, it is evident that image processing regarding flood susceptibility [70]. Machine learning methods use their
techniques focused on both pre and post-disaster phases while the ma­ own experience and pre-existing data to automatically learn and make
chine learning models focused only on the pre-disaster phase, as these decisions. For flood management, such systems involve the usage of
techniques were dealing with only flood risk analysis, prediction and trained prediction models that can quickly predict the risk of flood from
forecasting. Fig. 4 shows a graphical representation of the distribution of the given hydraulic data and weather parameters [71].
selected research articles among different categories. (3) What are the systems that are proposed to tackle problems related to
A comprehensive look at the different machine learning methods flood management?
that have been reported in the literature for flood predictions is shown in
(Fig. 5).
4.1. Cyber-physical systems
Fig. 5 clearly shows that there has been an increase in the usage of
machine learning tools for predicting floods. Refinements in these
Cyber-physical systems are components that integrate different sci­
methods are being proposed continuously, for instance, Cui [33] used a
entific theories and engineering disciplines, including IoT, cybernetics,
linear regression model to monitor the changes in the water levels while
embedded systems, distributed control, sensor networks, control theory,
integrating data visualization capabilities and a Canadian Digital
AI, Big data and systems engineering. Such systems can make decisions
Elevation Model (CDEM) to simulate and visualize the changes in water
and operate independently. Recent technologies, including mobile
levels [33]. Such refinements are being introduced to enhance the fea­
phones, the internet of things, artificial intelligence, big data, machine
tures while improving the overall accuracy and precision of the flood
learning, social media, robotics, and blockchain have all assisted in the
prediction system.
prediction, planning and response to floods. Searching for ways to
(2) What are the phases of flood management which existing studies tend
harness technologies in the most effective manner when it comes to
to focus on?
flood disaster management is a growing area of research [23]. The use of
The examined literature on flood management systems can be clas­
AI, IoT and Big Data for flood management has seen an explosion in
sified according to whether the systems are adopted in the pre-or post-
recent years [41,42,48]. Such systems rely on existing data from pre­
disaster phases. The systems targeting the pre-flood phase tend to focus
vious disasters to develop an informed analysis of flood disaster risks
on flood mitigation, planning, risk assessment and hazard analysis tasks
and public safety.
while the systems dealing with post-disaster flood management focus on
The internet of things (IoT) is a technology that makes use of radio-
flood detection, mapping, damage assessment and evacuation planning.
frequency identification (RFID) to integrate with the physical world.
The techniques proposed for both the pre-and post-disaster flood man­
Using sensor technology and the scanning features of IoT, data collection
agement problem belong to the areas of image processing and machine
and storage becomes more streamlined [72]. Collected data from sensors
learning. Here it is also important to differentiate between image pro­
generate data nodes, which can then be utilized for analysing flood risks
cessing techniques and machine learning. The difference is that image
and flood patterns [73]. IoT enhances the management of information
processing procedures carry out some operations on an input image to
systems, especially in the case of machine-to-machine interactions.
get some information from it or to simply improve or enhance it. For
Artificial intelligence can help in capitalizing on the weather predictions
flood management, these procedures are used to extract information
and climate change science is a great way [74]. Applying remote sensors
regarding the occurrence of floods from an input image of a flood-prone
and drones for meteorological purposes can help improve the data
or flooded area. For instance, edge detection methods have been applied
generated by collecting images of the affected region that are difficult to
to determine the water levels in various water bodies to issue warnings
reach by relief workers and ensure long term planning for tackling po­
tential flood disasters [9]. For example, the identification of potential
flood risks can help in preparing a particular region by generating
30
effective flood diversion tactics or population evacuation routes ahead
25 of time [15]. Such measures of resilience are the direct product of cyber-
No. of Arcles

20 physical systems’ applications. Prediction capacity for meteorological

15 departments will be greatly enhanced with the integration of AI for
predicting and communicating disaster risks and developing mitigation
10
strategies as a result [41]. Similarly, big data provides an immense op­
5 portunity in terms of developing flood disaster risk preparedness, miti­
0 gation, and response strategies [75]. The concept of big data and the
IP ML Hybrid tools that are involved in sorting, storing and analysing big data can be
Categories employed for developing frameworks that help in detecting the level of
risk of disasters and also to prepare for post-disaster management [28].
Mining information from previous disasters, government organisations
Pre-Disaster Post-Disaster
and first responders can provide insights that help to forecast future
Fig. 4. Distribution of the selected studies (reflecting categories and phases) - incidents. Combined with satellite imagery and real-time data from the
IP (Image Processing) and ML (Machine Learning). UAV, big data analytics will allow critical areas to be surveyed and

5
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

250
Machine Learning Techniques over the Years
ANN MLP ANFIS WNN SVM
200
No. of arcles

150

100

50

0
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019
Years
Fig. 5. Rate of use of different machine learning tools for flood prediction over the years.

evaluated. For example, a region has been flooded and understanding indicate that model systems will be extended to future applications in
how much area is affected, useful benchmarks can be obtained for disaster management.
mapping out flood-prone areas and decisions can be made where to store Artificial intelligence holds great potential in the analysis of post-
key resources beyond the range of typical inundation. disaster damage and preparing for the next disaster by improving
Integrating IoT and RFID for analysing data related to the urban infrastructure. Property damages related to harsh weather can be stud­
environment and incorporating it with a standard information system ied using AI as it can analyse the physical environment and improve
has been explored [72]. For instance, in the US, infrastructure was predictions of possible damages in the case of future disasters. Storms,
analysed to help identify ways in which disaster risks could be mini­ strong winds, and hail can be predicted with the use of AI [20]. Various
mized [19]. The development of sensors that can detect fire risks and machine learning algorithms have been developed for predicting severe
create an alert is deemed an appropriate warning system that can in­ weather conditions. Weather forecasts are one area of predicting di­
crease the preparedness for dealing with flood disasters [19]. In [5], the sasters and the data collected from forecasts can be combined with data
infrastructure of German cities prone to flood damage was analysed and related to infrastructure for identifying the possible risks and damages.
combined with data sets related to river water levels, weather conditions Management of water sources to prevent floods is a topic of great in­
and previous flood instances, to manage flood risks. Such cross-sectional terest among researchers working on disaster management. The use of
data can then be utilized through big data for the analysis of flood risks. disruptive technology in managing flood risks holds great potential.
Although such examples may still be considered proof of concept, results

Table 3
Pre-Disaster Flood Management Techniques.
Technique Method Limitation Authors

Image Applying edge detection to determine water surface levels of an urban Manually choosing parameters for edge detectors; not good results for low [34]
Processing region. contrast regions in images
Integration of data assimilation with data sensing to get 3D reports of the Limited tracking time as UAVs are battery operated [36]
flood-affected region.
Remote sensing measures the level of the water up to decimeter level and Expensive; requires special training to analyse the images [35]
does a real-time transmission
Narrow spectral bands used for classification; data is compared to known Lack of recent research based on this technology. [10]
flooding data to reduce false detections
Knowledge about the location of water bodies and water levels is Less number of algorithms for real-time flood risk management; Increased [37]
gathered. SAR facilitates a real-time assessment of flood-affected areas. need for satellite programming and training
Segmenting of image pixels and then clustering them as per objects Need extensive satellite programming [72]
recognized in pixels
Using UAVs to examine morphological changes and coastal dynamics to The limited battery life of UAVs [38]
estimate flood risks
Machine ANN model used for simulation of water flows at various locations in the Increased hardware dependence of ANN algorithm; requires parallel [41]
Learning river. processing
ANN-BPNN model for flood predictions. [42]
ANN and SVMs for predicting flood events [51]
Used adaptive neuro FIS (ANFIS) which combines ANN and FIS for [48]
creating a more robust model for flood predictions.
Levenberg-Marquardt method (LM) [73]
Combined wavelet analysis with ANN
WNN along with enhanced time-series data used for analysing data Use of gradient descent learning approach by the traditional WNN to [50];
related to river flow, flash floods and rainfall-runoff update the weight parameters, making the training process slow. [49]
Use of MLP for water elevation level prediction Redundancy and inefficiency in the fully connected network due to the [44]
presence of too many parameters
Training of ANFIS model using genfis2 and genfis3 Computationally extensive in the presence of large inputs [8]
The fuzzy logic-based system is used for forecasting river water levels. Need to do extensive testing of the system through hardware to validate [46]
the fuzzy knowledge-based system

6
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

4.2. Pre-disaster management systems between them. The third method involved using graph-based segmen­
tation to determine the water levels. Results showed that the canny edge
The techniques reviewed here are based on flood risk assessment, detector yielded the best performance among the three methods with an
prediction, and forecasting. These approaches are used to analyse areas, accuracy of 98.77%. Lin et al. [77], used image processing along with a
water levels in water bodies and monitoring different climatic factors to photogrammetric approach to find water levels from water bodies. The
predict the flood occurrence, map the areas prone to flood and estimate flooding line was detected using the Hough Transform-Canny method.
the severity of the possible flood event. Table 3 summarizes the pre- By applying photogrammetry along with DEM, these lines were further
disaster flood management techniques, which are reviewed in the sub­ used to determine the water levels in the input images. The overall ac­
sequent sections. curacy was calculated as 88.05%. Hasan et al., 2016 developed an op­
The following subsections provide a detailed discussion on each pre- tical gauge for enhanced water level measurement. The sensors work by
disaster flood management technique adopted. capturing the image from the target site and selecting a Region of In­
terest (ROI) from the image. After that edge detection is applied because
4.2.1. Image processing for pre-disaster management systems the differences in pixel intensities in an image correspond to changes in
Image processing techniques have been widely used to deal with depth, material and orientation [78]. Finally, a Hough transform is
floods. The most common use of such methods in the pre-disaster phase applied to determine the longest straight line ([79]).
is to monitor the level of water in dams and rivers to forecast any
possible flood. For instance, Skakun et al. [74] presented a flood hazard 4.2.1.2. Remote sensing. The water level of a wide area can be measured
and risk assessment model by calculating a relative flood inundation in consecutive events. Analysing the change in water level can help in
(RFI) value to generate a flood hazard map. For any flooding incident, the easy prediction of flood events. Remote sensing measures the ac­
pixel analysis has been done to classify each pixel in either “water” or curacy of water up to decimeter level and shows real-time transmission
“no water”. Using this analysis, a maximum flood extent image is [35]. This technology is also adopted for the forecasting of floods, as
generated which is further processed using a mathematical approach to carried out in [80,81]. Image processing technique like image segmen­
calculate the RFI value that represents flood risk [74]. Robertson and tation has been utilized on the remotely sensed images to determine the
Chan carried out aerial image segmentation to perform flood hazard formation of a flood event and analyse the flood status at any time [36].
assessments. The proposed model used colour based segmentation, Image segmentation methods separate pixels from an image that
texture classification using entropy and image gradients to segment the represent a visual component. All pixels in one segment have the same
flood-prone areas from the images [75]. Nirupama and Simonovic [76] visual attributes as colour and texture. Using this method, the fluctua­
carried out flood risk analysis by performing the classification of image tions in water levels, elevation, and disturbances related to floods are
pixels using a maximum likelihood method. It uses the variance/ determined to generate real-time flood warnings. Geoscience Australia
covariance between classes to determine a posterior probability estimate has been working on developing newer techniques for flood mapping
indicating the pixel class [76]. [82]. These efforts led to successful flood mapping in 2008 by the use of
Pre-flood activities including flood risk assessment, hazard analysis the Landsat-5 TM technique [26].
and mapping of flood-prone regions have been carried using techniques Rapid response in the case of disasters has been deemed as possible
like pixel segmentation and edge detection. The various image acquisi­ using remote sensing techniques. This is due to the reliance on remote
tion approaches which have been discussed in the literature for flood sensing techniques on satellite images which make it easy to gather data
management include remote sensing, SAR, and UAV. Each of these from different regions in hours or a few days (Bates, 2004). However,
image capturing technologies has been discussed in the subsequent barriers exist in this method as the use of satellite imaging is not possible
sections along with the image processing method adopted to process the across the globe since many of the regions are devoid of satellite receiver
acquired images. stations. There is also an added disadvantage of having to train satellites
to obtain and store flood data, calling for the increased need for satellite
4.2.1.1. Edge detection. Edge detection is an image processing method programming [26,83]. These barriers can prove to be major setbacks in
that is used to determine the edges of object boundaries in an input the case of emergencies such as floods. The networks based on gauge
image. The differences in brightness are used to find these edges. This sensors provide one-dimensional physical parameter measurements.
technique is widely used for image segmentation and extraction tasks. It While dynamic information about the site can be gained through the
is used to monitor the water levels at dams and rivers to ensure that application of visual sensors, thus providing actual field data for making
water balance does not reach dangerous levels. Constant monitoring of an important decision. The availability of a co-flood image improves the
water levels in water bodies is therefore crucial to avoid any flood-like detection and refinement of the inundated area. Data from river gauges
emergency. can facilitate the estimation of maximum inundation. In case river gauge
In terms of flood management, edge detection has been used to data is not available ancillary data from media, photos and videos can be
detect a horizontal water line or the height of a dam [34]. This helps in used. In semi-automation mapping, pre-flood images must be considered
separating the water image from other objects in the surrounding. A [80]. As permanent water bodies may be wrongly classified as flooded
warning system was established using several image processing tech­ regions. This can be solved by masking the permanent water bodies. The
niques such as ROI and Otsu-thresholding. The water surface level of any climatic conditions and presence of clouds can limit the availability of
region can be calculated by processing the captured image. Overall, an the co-flood images. Furthermore, the small, inundated areas cannot be
accuracy of 96% was observed using this system. Zhang et al. [52] mapped by low-resolution satellites and areas affected by small streams
experimented with three different image processing techniques for the have a short co-flood time [36].
real-time monitoring of floods. These methods include canny edge
detection, region growing and image segmentation using graph cut. 4.2.1.3. Synthetic aperture radar (SAR)
Canny edge detector finds the boundaries where a difference in intensity 4.2.1.3.1. Image acquisition. SAR is a specific radar that is used to
levels between two regions is detected. It works by first transforming an generate two dimensional or three-dimensional object reconstruction. It
image into a grayscale and applying a Gaussian filter to remove noise. consists of transmitter and receiver nodes built in the form of antennas
The change in intensity is detected by calculating the gradient of the to transfer and receive signals. SAR is found to be a reliable tool for
image. After that, non-maxima suppression is done to retain only edge monitoring floods on rivers with less than 1 km in width. These sensors
pixels from the image. The region growing method is used to divide the provide different acquisition modes, with each mode linked to resultant
pixels in an image into several parts depending on the similarity image resolution and area of the flooded region. SpotLight, StripMap or

7
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

ScanSAR mode can be acquired by RADARSAT, TerraSAR-X and Cos­ analyse risks without the need for working with the physical constraints
moSkyMed. The SpotLight mode obtains high-resolution images for or implications that hinder flood risk analysis. Floods being natural di­
small areas; the StripMap mode provides a good exchange between the sasters are a complex phenomenon to be modelled using traditional
resolution and the affected area size; while for large area size the statistical and mathematical methods. During the last two decades,
ScanSAR mode is applicable [37]. However, challenges may occur while machine learning methods have significantly contributed to the devel­
using an appropriate sensor, data acquisition and interpretation of the opment of advanced systems for predicting floods with enhanced per­
images for extracting information about flood events. formance and less expensive solutions. The machine learning methods
B. Image Processing for SAR Imagery. offer a wide range of advantages which include being quick, cheap, high
Different SAR based image classification methods have been performing and easy to validate. Hybridization of standard models along
explored and image classifiers have been developed over the years to with the introduction of new ones has become a common practice
enhance flood mapping through image processing. Segmenting of image among researchers for flood prediction problems. Over the years, several
pixels and then clustering them as per objects recognized in the different different algorithms such as ANN, support vector regression (SVR) and
pixels has been observed to be a possible technique for accurately esti­ support vector machine (SVM) have been shown to yield reliable results
mating the land region and the flooded regions. The water levels from for flood predictions [84,85]. Different algorithms and systems of ma­
SAR images are measured using image processing techniques and are chine learning can be combined to build stronger flood prediction
monitored to issue warnings before the occurrence of a disaster [38]. models.
SAR was applied successfully for mapping floods and water detection of Most of the machine learning methods have focused on the flood
urban areas during high wind velocity [38]. Flood forecasting in Severn prediction and forecasting problem hence they were classified under the
and Avon rivers, in the UK using SAR was found to be feasible to work as pre-disaster phase. In the domain of flood risk management, systems
a standalone method [39]. focusing on hydrological predictions remains to be a huge challenge for
researchers. More recently, advancements in Artificial Intelligence (AI)
4.2.1.4. Unmanned aerial vehicle (UAV) have introduced many powerful technologies to model hydrological
4.2.1.4.1. Image acquisition. A UAV is a battery-operated drone, that events like floods. Among these tools, ANN and SVM have been suc­
is being rapidly used to gather image data, to assist in disaster man­ cessfully used in a wide range of case studies related to flood prediction.
agement tasks. Fig. 6 shows a UAV controlled by a ground control sta­ One method is to gather rainfall data and daily water levels to test and
tion, which is responsible for assigning surveillance missions to UAVs, train a machine learning model like ANN [86]. This model has been built
monitoring them, and managing the captured data. A UAV can provide to predict the water levels after a time interval of 24, 48 and 72 h. Ruslan
fast image capture and transmission services to monitor the affected [87] used data based on water levels at three upstream river locations to
region, transmit images to the ground station which can be further train an ANN model. A Neural Network (NN) inverse model was inte­
analysed and processed by the system to make further decisions that grated with the output to improve the results [87]. Shi et al. [88]. used
could assist in the disaster recovery and relief process. rainfall and river flow data to train an SVM classifier to make predictions
4.2.1.4.2. Image processing for UAV imagery. In the last couple of regarding river flow change and peak flow within 48 h [88].
years, the suitability of UAVs for analysing flood risks has been explored
through different studies. The advantages of UAVs include the ability to 4.2.2.1. ANN. The ANN models have been regarded as the best ones for
effectively analyse morphological changes that occur due to varying developing flood risk prediction models [40]. ANN model is pro­
coaster dynamics and changing urban environments [40]. Combining grammed to mimic the human’s brain way of learning. The relationship
UAV photogrammetry with the Global Navigation Satellite System between inputs and outputs is recognized and learned by an ANN model
(GNSS) results in capturing highly enriched spatial data. The accuracy of during training. According to the latest research [43], ANN has shown
the data is reported to match the levels of the American Society for promising results when dealing with a flood-related crisis by predicting
Photogrammetry and Remote Sensing (ASPRS) standards. However, the streamflow. Also, the ANN model has proved to be beneficial for making
major limitation of UAVs is associated with their battery life [40]. This flood-related predictions as it requires just one input variable. Thus in
can be overcome by route optimization and resource allocation using such conditions where there is insufficient information related to the
algorithms [40]. problem, ANN proves to be a viable approach [41]. An ANN model was
formulated to simulate flows at a certain location in the river reach,
4.2.2. Machine learning for pre-disaster management systems based on flow at upstream locations [41]. Studies have shown that ANNs
The machine learning methods can use pre-existing flood data and have higher speed and accuracy than many of the conventional models

Fig. 6. Application of cyber-physical systems for flood management.

8
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

and tools used previously. However, there are certain barriers to the use where g(j) is the output value for the node ‘j’ in the hidden layer; gj is the
of ANNs as well which mainly include difficulty in data handling, mother wavelet function; wij is weight connecting the input layer and
parameter tuning, prediction of precipitation and slow response to hidden layer; qj is the shift factor, and pj is the stretch factor for hj [91].
training when gradient-based images are used. Among the different The WNN has proven to be useful in accurate flood modelling due to
classes of ANNs, the backpropagation ANN (BPNN) is the quickest and the potential of enhanced time-series data [49]. This helps in analysing
most powerful tool for predicting floods [42]. Typical artificial neural data related to river flow, flash floods and rainfall-runoff [50].
networks work in the forms of layers i.e. input layer, hidden layer, and
output layer. Several neurons in the output layer are defined by the
4.2.2.5. SVM. Another popular method for analysing floods is SVM.
application. The output neuron signal can be represented as:
The SVM is a method that works based on statistical learning and risk
( )
∑k minimization rule [92]. The SVM training models assign binary linear
Y = g(net) = g wi ai (1) classifiers that reduce classification errors to a minimum while maxi­
i=1
mizing geometric margin through an inverse method of problem-
Neuron output is represented by Y, wi represent the weighted vector, solving. A high dimensional space is created by the training data. This
a is the input signal and g(net) is the transfer function where the net is a space is divided into positive and negative instances by a hyperplane. To
scalar product of weight is and input with an expression as: classify new instances, their location in this space concerning the hy­
perplane is determined. Fig. 7 shows an SVM hyperplane.
net = (w1 a1 ) + (w2 a2 ) + … + (wn an ) (2)
In recent years, a regression tool has also been associated with SVM,
which is known as SVR (support vector regression). This method has
4.2.2.2. MLP. One of the most popular tools for flood prediction is the been used to conduct a flood risk assessment [51]. SVM got its name
MLP model which can be classified as a feed-forward ANN. It has an because in mathematics the training examples to the non-zero co­
input layer, output layer and two or more layers of trainable weights. efficients are called support vectors. Formulas of the hyperplane are
The MLP uses backpropagation for the training of the multi-layered given as:
network [89]. Some of the crucial features which make MLP an
important and suitable tool for flood predictions include its simplicity, w*x + b = 0 (7)
multiple layers, and nonlinear activation. The back-percolation algo­
where ‘w’ is the hyperplane vector, ‘x’ is the magnitude of a vector and
rithm is used for analysing and removing errors, making MLP a popular
‘b’ is a non-zero coefficient. The formula of the two margins is expressed
tool for hydrologists [45]. MLP was used to predict floods using rainfall
below. The distance between these margins with the hyperplane needs
time series data and water levels in a weir [90]. Mathematically MLP can
to be maximized. Thus, the distance of both these margins with the
be written as:
( ) hyperplane is the same i.e. ±1.
∑ k
y=g wj qj + a (3) w*x + b = 1 (8)
j=1
w*x + b = − 1 (9)
where wj and qj are the weighted and input vectors. a represents the bias
and g is the transfer function.
4.3. Flood management during a flood event
4.2.2.3. ANFIS. Artificial neural networks combined with fuzzy infer­
ence form an ANFIS. It is a fuzzy system that uses a learning method Flood management during the flood event is critical particularly
based on a neural network to find its parameters. There are two inputs during the first 72 h of a disaster, as preserving human lives is at stake
‘x’ and ‘y’ and one output ‘z’. A rule base contains two if-then based and rescue missions need to be expedited at the earliest [38]. The main
rules, which are given below: issue during a flood event is the loss of communication and reduced
efficiency of the first responders’ team due to a lack of situational
Rule − 1 : If x is M1 and y is N1 ; ➝z = a1 x + b1 y + c1 (4) awareness of the disaster [39]. Aerial assessment of the affected region is
the most efficient way to analyse the extent of damage and to be aware
Rule − 2 : If x is M2 and y is N2 ; ➝z = a2 x + b2 y + c2 (5)
of the progression of the disaster. Research has shown some promising
results in terms of developing a UAV network to help facilitate the search
where M1, M2, N1 and N2 are nonlinear parameters and a1, a2, b1, b2, c1
and rescue mission [67]. The use of UAVs allows the detection of
and c2 are linear parameters.
structures that have been impacted by the flood, giving real-time up­
Fuzzy logic-based systems are also quite popular and have been used
dates of the extent of damage caused to the structures and transport
for forecasting river water levels [46]. Rainfall-runoff water modelling
infrastructure [39]. This will help rescue teams to identify the safest
has also been carried out using the fuzzy interface system (FIS) [47].
route to the destination and the number of victims stranded [42,57]. The
Short term rainfall predictions and flood identifications are possible
damage can be assessed through different methods such as images taken
using a slightly modern FIS approach, known as the adaptive neuro FIS
by the UAV, UAV video inspection and structural health monitoring
(ANFIS), which combines ANN and FIS for creating a more robust model
techniques [57]. The disaster preparedness can take place before the
for flood predictions [48].
flood event via cyber-physical systems, historical data, satellite imagery
of before and after disaster events and big data analytics, thus preparing
4.2.2.4. WNN. WNN is a combination of the Wavelet Transformation
the authorities for any unforeseen events [59].
(WT), wavelet regression models and the feed-forward neural network
After constructing the safest route to the target region, the distri­
(FFNN) methods. For the input signal sequence =a1,a2,…, an, the output
bution of aid among the affected people is a major challenge. It is
of the hidden layer is calculated as:
essential to resolve technical issues and optimize resource allocation and
∑
n vehicle routing to enhance the efficiency of the relief operation.
wij ai − qj
Resource allocation and vehicle routing have been major areas of
g(j) = gj i=1 (6)
pj research when dealing with such a problem. Different models have been
previously presented to transport relief goods to the victims during a
j = 1, 2, …., m disaster [56]. The properties of various routing problems and the logistic
method to plan humanitarian relief and rehabilitation have been

9
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

UAV

Ground Control Staon

Image Acquision

User

Flooding
Fig. 7. A flood monitoring system using UAVs.

Table 4
Post- Disaster Flood Management Techniques.
Method Limitation Authors

Image Feature extraction from images to detect flood levels. Results highly depend on image segmentation results [93]
Processing Target recognition of landmarks using image segmentation [94]
Detection and labelling of objects within an image to find flooded areas. Manually choosing parameters for edge detectors; not good [11]
Target recognition of landmarks from aerial images using edge detection and Hough results for low contrast regions in images [95]
transform, to do flood-related analysis
Use of GIS to determine exit routes and transport facilities. Many cases of destruction of telecommunication systems [96]
Collecting knowledge about a geographic area in the form of datasets using GIS and during floods. [61]
creating maps of exit routes and transportation in case of emergency.
Using GPS for flood risk management to carry out guided evacuation and rescue Ambiguity in mapping the exact position; Unavailability of [97]
activities. the internet during the flood crisis
Transmission of location and time information to a GPS receiver [63]
Use of optical and radar imagery to measure water level which helps in defining the Expensive; requires special training to analyse the images [98]
scope of a flooded area
Knowledge about the location of water bodies and water levels is gathered. SAR Less number of algorithms for real-time flood risk [37]
facilitates a real-time assessment of flood-affected areas. management.
Modifications in the type of sensors being used for SAR have been proposed to Increased need for satellite programming and training [99]
overcome cloud cover issues.
SAR is used for flood mapping and to determine the extent of floods in a given region [100]
Use of SAR to detect floods in areas covered by vegetation and seemingly closed spaces. [83]
Segmenting of image pixels and then clustering them as per objects recognized in pixels [78]
Satellite imaging to analyse areas affected by disaster Satellite imaging is not possible across the globe; Many [101]
regions are devoid of satellite receiver station
Narrow spectral bands used for classification; data is compared to known flooding data Lack of recent research based on this technology. [82]
to reduce false detections
Successful flood mapping using Landsat-5 TM technique [26]
Using UAVs to capture high-resolution spatial images in complex landscapes and Limited tracking time as UAVs are battery operated [58]
transmitting to the servers. [102]
[59]
Combining UAV photogrammetry with the Global Navigation Satellite System (GNSS) [39]
to get highly enriched spatial data.
UAVs are presented as a cost-effective solution for real-time mapping of floods. [56]
UAV data acquisition: Integration of data assimilation with data sensing to get 3D [36]
reports of the flood-affected region.
Images captured through UAV are analysed for detection, segmentation, and [103]
classification of flooded areas. [57]
Application of image processing techniques on UAV images to determine the spread of [104]
flood and identify routes that can help in rescue activities.
Use of hyper-ion imaging for analysing and mapping floods, using spectral bands for Not well researched to deal with flood management problems [60]
classification
Machine Landmarks detection on UAV images along with SVM classifier to detect images of Increased sensitivity of the landmark detection algorithm to [105]
Learning flooded regions noise
Pixel-based classification on remotely sensed images using SVM and ANN to separate Increased noise levels in satellite images [106]
flooded and non-flooded areas

10
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

reviewed [57–59]. One of the most crucial aspects during a flood event is the gathered data an evacuation path map is built and an escape plan is
the determination of the shortest routes to the affected areas to reduce developed [69].
cost, maximize the vehicle capacity and minimize the time spent in In the next subsections, the applicability of the most common image
visiting the target sites [59]. Image processing techniques such as edge processing techniques for flood management will be discussed.
detection along with ANN could be used to assess the historic and
existing data for the relief team to analyse the affected region and 4.4.1.1. Edge detection. Edge detection techniques have been exten­
allocate the UAV wisely. The aim is to cover all the affected regions and sively used in the literature to detect various objects from images such as
reach out to a maximum number of people who are affected by the roads, buildings, and houses. The extracted objects can be analysed
disaster. The overall expenditures of a disaster relief organization are using image processing to determine if they have been affected by a
related to vehicles cost of services, management, and transportation disaster. The images can be labelled for different objects and charac­
cost. These costs can be reduced by determining an optimum route for teristics, such as the boundary of flooded areas, reflection from the sky,
the vehicles and locating the target sites. clouds or any nearby object and water edges by utilizing the segmen­
tation capability of the edge detection algorithm that detect disconti­
4.4. Post-disaster flood management nuities in the grey level of pixels [52]. The analysed image can be further
processed to identify true positive results for flooded areas and differ­
In this section, we discuss techniques from the fields of image pro­ entiate images of floods from those images that did not depict flooded
cessing and machine learning, that are focused on the post-disaster areas [11]. Target recognition from aerial images has been done using
phase, that is, after the occurrence of flood events. The post-disaster edge detection and Hough Transform to find rescue routes [112].
phase deals with events related to disaster response and recovery. Detecting roads, runways and bridges are necessary for identifying
Hence, the techniques reviewed here are based on timely mapping of routes that could be used for reaching areas prone to disasters. Canny,
flooded regions, determination of water levels and dealing with evacu­ Sobel, Harris and Deriche are different types of edge detection algo­
ation operations. These approaches are used to provide an immediate rithms that can play important role incorrect detection of bridges, road
response to flood events by locating the flooded regions and and dams [113]. A description of these edge detectors has been provided
commencing the rescue operations by identifying rescue routes and in Table 5.
transport services in an area. Table 4 summarizes the post-disaster, flood
management techniques that are discussed in the subsequent sections. 4.4.1.2. UAV imagery
4.4.1.2.1. Image acquisition. UAVs are extremely cost-effective
4.4.1. Image processing related to post-disaster flood management platforms for quick deployment and mapping of floods in real-time
Image processing involves performing certain operations on an [56]. The use of images captured from cameras on UAVs is deemed as
image to extract useful information from it. Various techniques from this an effective approach in flood mapping and monitoring [57,115]. UAVs
domain including object detection, edge detection and image segmen­ have now replaced the traditional data capturing tools like satellite and
tation are being widely used to deal with natural disasters. High- GPS for doing coverage of the flood-affected regions. They are being
resolution satellite images are processed using image processing tech­ used to capture data from flood-affected regions in the post-disaster
niques to identify the areas affected by disasters, do damage assessment, period instead of using traditional data capturing tools such as satel­
and locate the transport routes. The estimation of floods is carried out by lite imaging and GPS based monitoring. They have the edge of being
using the segmentation [75] and patch-based approach for studying the quick in gaining high-resolution image data and transmitting it to their
images gathered from flood-affected areas. In a segmentation approach, respective servers [58] so that it can be analysed to detect flood-related
the flooded and non-flooded areas are classified to perform flood map­ damage or floodwater levels. Satellite images on the other hand are low
ping [75] whereas in a patch-based approach image patches from the in quality and are not readily available in the post-flood period. Using
pre and post-flood satellite imagery have been compared [107]. Some of
the approaches reported in the literature include the detection of
buildings near the river banks, roads [105] and bridges [95]. All these Table 5
objects provide insights about the presence or absence of flooding in Most Common Edge Detection Techniques.
each area either by comparing the state of these landmarks in both pre- Author Technique Method Application in Flood
and post-disaster images or by doing their damage assessment. Feature Management

matching approaches including brute force matching, Flann matching [52] Canny Locate edges by finding zero- Detection of flood in real-
and SURF detector can be used to compare different pre and post-flood crossings along the edge time by analysing river
normal directions (non- water levels.
landmarks to conclude the occurrence of the flood [108]. Damage
maximum suppression):
assessment from key infrastructure has been done using feature
∂2 (G*I)
extraction and morphological operations [109]. Image processing is ∂n 2
=0
being used extensively by different countries that have a radar system [114] Harris Finds the difference in Feature detection from
intensity ‘I’ for a displacement pre-and post-flood
for capturing satellite images [110]. Floods in the USA have successfully
of (p,q) in all directions. images. These features
been analysed with remotely sensed image data [10]. In particular, were analysed further by
images were analysed for detecting flood-related damages in an area analysing the texture of
[111] by using post-disaster satellite imagery. Histogram equalization the features.
method was used to enhance contrast and a simple linear iterative [99] Sobel A 3 × 3 convolution mask is Water lines are detected
applied to the image. The from SAR images to
clustering (SLIC) was applied to form segments of uniform super-pixels. differences are calculated at classify water regions.
For each superpixel, 62 texture features are applied to determine vari­ the centre pixel of the mask.
ations in texture indicating damage. Once damaged roads and routes are N/A Deriche 1. Noise removal N/A
detected the rescue teams can be guided effectively to reach the affected 2. Determination of the
direction of gradient &
people to carry out the post-disaster relief missions. A route estimation
magnitude
approach has been proposed by Kafi and Gibril [69] where a network of 3. Non-maximum
(Emergency-urgent-communication Mobile Terminal) EMTs is con­ suppression
structed. A (Mobile Ad-hoc Network) MANET is used by an EMT to 4. Using two thresholds to
transmit its location information to other EMTs. The data about the calculate hysteresis
thresholds
motion of the people in a location is collected by EMTs using GPS. Using

11
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

UAVs provides high flexibility due to their ability to fly with varying
sensors and the ease of configuring these devices as per the required data
[102]. These abilities particularly come into use when a complex land­
scape is to be analysed like in the case of flood-affected environments
[59]. In the post-flood images, flood maps have been generated to
analyse the extent of water elevation using a digital elevation model
(DEM) along with knowledge about the topography of the land [10].
4.4.1.2.2. Image processing on UAV imagery. In the case of floods, the
instances of damages due to rising water levels can be analysed over
time and safety routes can be identified for rescue purposes. Image
processing has been used for analysing the UAV based images of flooded
areas to determine the spread of flood and identify roads and paths that
can help in evacuations and reaching people who are stranded due to
flooding [116]. Popescu et al. worked on locating and detecting flooded
areas from images using a segmentation approach. Various properties of
images including texture and colour were considered and a clustering
mechanism was applied to these features to identify the classes. Finally,
a similarity criterion was defined for the segmentation of flooded areas
from images. This algorithm yielded an accuracy of 98.87% in detecting Fig. 8. SVM Hyperplane.
the flooded regions from images [117]. Tuna [118] investigated the
ability of UAVs to determine rescue routes in the post-flood period. The
system consists of a set of UAVs, each having a computer on-board
running sub-system that carries out communication, navigation and Capture SAR
Pre-processing
Calculate pixel
control [118]. Using this computer along with the UAV controller, the images threshold values ‘T’
goal of the acquisition of communication routes and infrastructure is
achieved. Casado [119] carried out a post-flood damage assessment with
UAV imagery. A photogrammetric process was applied to achieve digi­
tized orthoimages of the flood-affected land. Each image was then
divided into 10x10m squares to determine the flood impact features T<=42
[119]. Recently, UAV based images have been used for rescue planning
after the occurrence of floods [104]. The concept behind the use of flood
information was that floods cannot be stopped merely by making pre­
dictions, but the prediction-based data can be utilized for reaching
Flood Detected
people in the case of emergency and carrying out rescue related tasks Water Pixels
once floods have occurred.
Fig. 9. Flood detection in SAR images.
4.4.1.3. SAR imagery
4.4.1.3.1. Image acquisition. SAR has been used in post-disaster satellite. Thus, regardless of its high accuracy and real-time monitoring,
image acquisition for further analysis using image processing tech­ it does not fit as the best technology for urban flood prediction. How­
niques. These sensors have now emerged as the most used image ever, it works best for large water bodies including oceans and rivers
acquisition resources to map flooded regions as they can provide [36].
coverage during day and night and in all weather conditions. The
number of SAR satellites in orbits are rapidly increasing some of these 4.4.1.4. Hyper-ion imaging
satellites are Sentinel-1 Constellation of Small Satellites for Mediterra­ 4.4.1.4.1. Image acquisition. Hyperion imaging relies on laser tech­
nean basin Observation (COSMO SkyMed), RADARSAT-2 and TerraSAR- nology to capture in-depth images of the landscapes providing a
X [77]. These satellites ensure immediate mapping and detection of comprehensive view at a single scan using hyperspectral sensors. The
floods in a post-flood emergency. first Hyperion platform EO-1 was launched in the year 2000 which was
4.4.1.3.2. Image processing on SAR imagery. Anusha and Bharathi able to capture data within the spectral range of 400 to 2500 nm using a
[54] worked on post-flood mapping of flood-affected regions using the 30-m GSD device [120]. These sensors now record radiance within 5-15
SAR imagery. Images of the target region are captured through SAR nm spectral bands which are narrow, adjacent and very large in the
which are passed through noise filters to remove noise and get a smooth amount [121]. Air-based hyperspectral sensors deliver highly promising
image. After this, thresholding is applied to separate the flooded areas outcomes, as they can provide a high spatial and spectral resolution
from the non-flooded areas in the SAR image. Flooded and non-flooded [122]. They are also not much impacted by atmospheric disturbances.
regions are classified by using pixel-based thresholds. The idea is to One drawback is that they do not regularly cover large regions as done
classify the pixels which are less than or equal to the threshold value ‘T’ by the space-based sensors and satellites.
(where T = 42), as water pixels. The rest of the pixels are categorized as B. Image Processing for Hyper-ion Imagery.
non-flooded. Fig. 8 shows this process of finding water pixels from the It was first adopted for flood mapping and it was observed that this
SAR images to identify floods. (See Fig. 9.) algorithm can also be used for analysing and mapping localized floods
Data Assimilation techniques facilitated the real-time integration of [60]. In the realm of flood detection, the Hyperion imaging algorithm
SAR-derived water levels and developed forecast models for disasters uses spectral bands for classification which are narrow and compares
[55]. The integration of sensing data with Data Assimilation provided data to known flooding data to remove the chances of false indication by
3D reports of the flood used for the prediction of the flood as well as detection of wet regions such as lakes and rivers.
organizing the warning system for the flood. A problem faced by this
technology is its inability to measure the long-term and real-time water 4.4.1.5. Geographic information system (GIS). GIS has become indis­
level at fixed points. This is because of the orbital cyclic movement of the pensable to disaster management and response. Geographic knowledge

12
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

and an understanding of a local area help in devising a quick and installed on various points in flood plains, riverbanks, and other flood-
effective emergency plan [123]. In the case of floods, the local routes prone areas. Each node transmits the flood status data to a portal
and public transport systems are likely to be affected most negatively managed by the company [89]. Arthur et al. [130] proposed a model to
[124]. GIS can be put to use in emergency cases, where telecommuni­ detect flooded areas. Data was collected from Tweets of the disaster
cation and power supply system’s failure resulting in the unavailability affected people. By analysing the text of the tweets, the GPS locations of
of internet services in the affected area [96]. According to studies car­ the users and their hometowns were determined. Using this data, the
ried out so far, having image datasets depicting the routes of an area locations having ongoing flood situations were detected. Integration of
before a disaster can prove to be extremely useful as these datasets can GIS and GPS has been reported to help analyse roadblocks and safe
help in generating maps in times of emergencies. Even in the absence of routes for reaching people stranded in floods [69]. However, the limi­
pre-existing data, maps of flooded areas could be created by gaining the tations existing in the use of GPS make it unlikely that large scale rescue
recent data related to the geographical location but this would take more operations could be carried out successfully using this technology in
time, as it requires the application to get data in real-time [61]. real-time flood risk management. As reported in the literature [93,121]
As an example of the applicability of GIS, it was put to effective use there are a few disadvantages associated with the GPS based estimations
by the local government in New Orleans during Hurricane Katrina [7]. It for disaster management. First, these estimations have a certain degree
was observed that the use of GIS enabled the local government to carry of ambiguity, as the exact position of objects cannot be mapped due to
out mapping of the affected area in a quick manner. [62]. This was done an approximate error of 15 m for every 3 Km [93]. Second, during
by mapping the transportation data, available routes, and highways to emergencies, the lack of satellite imaging and network can lead to a loss
enable the rescue of the stranded people. However, not all the transport of information regarding the flooded areas [111]. In the cases, the use of
available in the region could be used due to the lack of information GPS becomes non-feasible and unreliable [63]. Hence, it is important to
about the vehicles that could be readily used to rescue people. In this look for alternative methods or approaches that can reduce the error
situation, GIS was utilized to overcome the transport problem, as it can rates currently being encountered in flood risk analysis. To reduce the
determine the transport routes available in an area. The GIS also pro­ precision error rates associated with the GPS results, a relative position
vides monitoring capabilities using its graphical display which enables algorithm has been developed for detecting the extent of damage and
the user to visualize the information regarding spatial data and precisely identifying the level of support and rescue operations to be carried out
locate the geographical position of the vehicles [125]. Iqbal [126] [63]. This algorithm creates a new functional model using single and
generated a flash flood map by leveraging satellite imagery along with double differences in the readings of GPS.
GIS. An analytical hierarchical process was used to find the weight of the
effect of a set of flood causative factors to achieve a flood hazard index 4.4.1.7. Hybrid techniques. The use of stand-alone machine learning
(FHI). Samuel R. J. [63] used GPS sensor nodes to get relative positions techniques to deal with a post-disaster crisis is found to be rare in
of objects like buildings from both pre and post-disaster periods which literature. Research is now being focused on developing hybrid models
are then combined with GIS data to estimate changes in these objects to for flood predictions [102]. These models are based on combining two or
perform damage assessment [63]. more machine learning techniques for predicting floods by creating
systems that use the best features of different methods. Over the years,
4.4.1.6. Global positioning systems (GPS). Currently, the risk manage­ there has been a significant increase in the use of hybrid methods over
ment techniques around the world are mainly focused on carrying out single method-based machine learning tools [20].
guided evacuation and rescue of the flood-prone areas [97]. These ef­ Newer algorithms and updated technology are constantly being
forts are carried out using guidance from the GPS based analysis proposed for object recognition. The constant growth in machine
methods. Flood management based on the use of GPS has been suggested learning and image processing can help in determining newer ways for
in numerous studies [69,126]. GPS has been used in post-disaster relief managing real-time data and then using it for disaster management.
operations, by providing three-dimensional coordinates of a person’s Disaster preparedness and post-disaster management can both be car­
location to help him evacuate a disaster-affected area in minimum time. ried out efficiently in the future by using UAVs for analysing disaster-
This technology has been used to provide information to scientists about prone or affected areas and identifying safe routes in those regions.
flood-prone lands. The emergency response team can perform rescue Aerial images can be captured from flood-prone areas to detect land­
operations effectively through GPS. After gaining location information, marks like bridges and roads, to determine rescue routes in case of a
the data is combined with images and a geographical model for exam­ flood [110]. Emergency evacuations can follow the initial analysis and
ining the flood risks [64,96]. The resulted data can provide information rescue teams could be guided through the information gathered and
about flood-related parameters such as expected flood depth and processed in real-time through the combination of UAVs and the image
coverage within the geographical area. The use of GPS has been pro­ processing algorithms. In the future, these techniques and technologies
posed and carried out to some extent for flood relief and rescue missions. can be further enhanced to achieve higher accuracy of land monitoring
Beckheinrich reported an accuracy of 82% while monitoring water and object recognition.
levels of the Mekong river situated in Vietnam. It was concluded that an Machine learning algorithms have been used with data analysis in
error in the coherency of estimated values has been recorded which can the pre-disaster phase to make predictions related to the future occur­
be attributed to the roughness on the water surface. An empirical mode rence of flood events, their severity and possible damage [105]. How­
decomposition method was proposed to correct the observed water level ever, in the post-disaster period, techniques based on machine learning
results up to a few centimetres [117]. Yoshida et al. [66] developed a have been used along with image processing to map flooded regions and
GPS based system to stimulate the movement of people in a flood- make decisions related to relief and rescue activities. Such methods have
affected area to assist the evacuation plans. An approach called Monte been categorized under the Hybrid class [120].
Carlo is presented that uses sparse spatial GPS data to model the flow of Syifa et al. [106] used artificial intelligence approaches on images
people in both pre-and post-flood conditions. Ghorbanzadeh et al. [108] captured by remote sensing. Image datasets containing pre and post-
conducted a study to support the flood control strategies in the Hima­ disaster images were used to train machine learning models to detect
layas. The water surface elevation data along with GPS location and floods [106]. The aim was to accurately map the flooded regions in the
information about the land usage of the flood plain was used to simulate aftermath of a flood caused by the collision of a dam in Brazil. The pre-
the water flow in the river, produce flood risk warnings and generate and post-flood images were first captured by using Landsat-8 and
flood inundation maps. A flood inundation sensor was developed by Sentinel-2 satellites. These images were used to train machine learning
Intellisense Systems that uses GPS location along with a node network models. The dimensions of the flooded regions from these images were

13
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

estimated by using both ANN and SVM classifiers to do pixel-based telecommunication methods get destroyed during flooding. The appli­
classification on images. The water clogged regions and land were cation of image processing and machine learning techniques could prove
segmented providing a clear picture regarding the availability of avail­ to be beneficial for disaster risk management. The use of these integrated
able routes in the disastrous area. technologies can gather real-time information, increase the real-time
Landmarks detection from images has been combined along with response, and make future predictions even for a short duration. This
machine learning algorithms to assist in flood management [105]. Sig­ will help the relief workers to prepare for the oncoming disaster. The
nificant landmarks like buildings, roads and houses were detected from images from the disaster sites will enable the relief workers to develop
images using edge detection. The extracted landmarks were then used to the maps, find the safest route to the disaster area and carry out the
train an SVM classifier to build a prediction model that could classify a mission timely. Using UAVs in this process will prove to be of great help
new input image as flooded or non-flooded. This method has yielded an as it will ensure easy access to the areas which are humanely impossible
accuracy of 90% in classifying a new set of images. to reach also this will provide a clearer view of the areas which can help
Object detection from images is done by using machine learning in predicting the estimated destruction at a place and a great help in the
models trained to extract features that can help in detecting flood levels, rescue of victims. This can surely decrease the casualties. Application of
is crucial in successfully detecting the presence of flooding and then UAV swarm in different locations of the flood-affected area will assist in
analysing ways through which rescue missions and routes for evacua­ gathering the images of the flood region. The swarm knowledge can be
tions could be approached [93]. used for developing flood maps and assess the damage of the region
(4) What are the research gaps identified in the literature when it comes to (Fig. 10). Route optimization and floor plan simulation could help in
deploying technology for flood management? locating the shortest route for escaping the disaster region. Application
of the shortest path algorithm will help in devising the best evacuation
5. Discussion plan and guide the relief teams. The developed models and techniques
can be applied for different scenarios to handle the disaster. Thus, these
Currently, there are research gaps in machine learning usage for technologies can help the authorities in disaster preparedness, planning
flood prediction and post-flood management. Only a few studies have and decision making and can work in dynamic environments. A better
been carried out which have explored the use of machine learning understanding and application of these technologies by the emergency
techniques for developing a flood predicting model [13]. The robustness departments can fasten up the relief operations.
of such models and the ability to train quickly for analysing hydrological A combination of disaster management knowledge, image processing
data make the machine learning methods extremely useful for predicting techniques and machine learning tools can help in identifying a way
floods [35]. These data sources include variables such as rainfall, pre­ forward. Since many concepts and steps overlap when trying to study
cipitation, soil moisture, water levels, river inflow, run-off water, disaster management, image processing, and machine learning inde­
streamflow, river flood, flood frequency, flash floods, peak flow, pendently, it might prove useful in the long run to merge these concepts
groundwater level, storm surge and rainfall stages [106]. However, it is and come up with a way to use flood risk information for post-flood
important to note that there can be different types of floods, stemming rescue and relief missions.
from various sources such as increased soil moisture or long term stream
flow, therefore making it tough to make accurate flood predictions in the 6. Conclusion
longer run [105]. Analysing the existing tools and techniques being used
for flood management reveals that there are gaps between technologies This study is based on the content analysis method proposed by
that need to be filled. These gaps are mainly related to a lack of align­ Nguyen [122]. Using this methodology, a systematic literature review
ment in using existing tools and technologies for post-disaster relief. The was conducted to examine 140 of the most recent research articles from
use of UAVs, image processing and machine learning tools have all been the years 2010–2020, on the application of recent technology in the
carried out for flood risk management [57]. However, this use is domain of flood management. A novel classification framework was
restricted mainly with mapping floods and analysing flood risks from the formulated for the categorization of various studies conducted for flood
point of predicting floods. management in the domains of image processing, machine learning or
In the past few years, machine learning algorithms have been tested both. This study answers four key research questions: (1) What are the
rigorously to conclude that the approach is highly successful in flood major techniques deployed in flood management? (2) What are the
detection. An accuracy level of up to 90% has been reported for the phases of flood management which existing studies tend to focus on? (3)
machine learning algorithms tested so far. Majorly, algorithms have What are the systems that are proposed to tackle problems related to
been tested for binary classifications, where the only option is to classify flood management? (4) What are the research gaps identified in the
flooded versus unflooded areas [38]. Recovering from a disaster requires literature when it comes to deploying technology for flood manage­
developing a reliable system that can aid immediate communication ment? By addressing these questions, this review has highlighted the
about flooded areas and help in increasing public safety, rescue, and state-of-the-art technology that has been used in different phases of the
security measures. The use of machine learning approaches can enhance disaster management lifecycle, along with the limitations of each
existing systems and possibly build better systems for emergency man­ technique.
agement [69]. The focus of the classification is mainly on the use of image pro­
The proposed model is to ensure that modern technology could be cessing and machine learning techniques adopted for flood prediction,
aligned to be used in the case of post-flood management. Since floods flood hazard or risk mapping, flood detection and flood inundation
cannot entirely be avoided or stopped, knowledge about flood-prone mapping. The major systems used to address the issues related to flood
areas and flood risks could be used for developing a method that management include systems using image processing techniques such as
could aid in pre-and post-disaster relief work. In a pre-disaster scenario, edge detection, segmentation and pixel analysis and systems incorpo­
UAVs can gather images of the areas that are prone to floods. The in­ rating machine learning models such as ANN, SVM, MLP and WNN. The
formation about major landmarks, dams, road infrastructure, bridges, most common image acquisition approaches used are UAV imaging,
etc. can be collected. This data can prove to be useful for clustering and SAR, and remote sensing. The existing techniques from both image
scanning in the future. The UAVs based on their internal navigation processing and machine learning domains tend to focus on both pre and
system (INS) can easily locate such areas and can be assigned to carry post-disaster phases. A classification framework has been proposed to
out any relief activity during flood events [85–87]. GIS has always group the various technologies reviewed for flood management. The
played a great role in reaching out to flood-stricken areas and providing identified research gaps include the lack of application of hybrid models,
relief to the people. However, the major limitation is that most of the which combine image processing and machine learning, for flood

14
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

Fig. 10. Integrated approach for flood management.

management. Also, recently, the use of machine learning-based methods processing, machine learning and related techniques for disaster man­
is found to be rare while dealing with the post-disaster crisis. This calls agement has been studied for over a decade. Some significant technol­
for improvement in the post-disaster management process by the inte­ ogies from beyond the specified range of periods in this research can also
gration of AI for automation. The use of AI-enabled big data for flood be explored and evaluated for their applicability in the current era. In
management is also an appealing area of research for the future. the machine learning category, further subcategories may be introduced
One limitation of this review is that an interpretative classification of focusing on unsupervised learning-based methods, reinforcement
the research articles has been carried out in this research as issues learning and deep learning techniques. Furthermore, models for pre­
regarding bias in subjectivity may be raised. This is a well-known dicting the recession of floods on calculating length of recovery and
concern of the content analysis methodologies [101]. Another issue reconstruction could be explored in future studies.
may be the period selected for the study, that is between 2010 and 2020.
This duration has been chosen to focus on the most recent and advanced
technologies for flood management. However, the use of image

15
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

Declaration of competing interest [20] R. De, O. Werneck, I.C. Dourado, S.G. Fadel, S. Tabbone, R. Da, S. Torres, Graph-
based early-fusion for flood detection, in: 2018 25th IEEE International
Conference on Image Processing (ICIP), IEEE, 2018, pp. 1048–1052, https://doi.
The authors confirm that this manuscript, “A Review on Flood org/10.1109/ICIP.2018.8451011.
Management Technologies Related to Image Processing and Machine [21] R.W. Tiner, M.W. Lang, V.V. Klemas, Remote Sensing of Wetlands, CRC Press,
Learning”, has not been published elsewhere and is not under consid­ 2015, https://doi.org/10.1201/b18210.
[22] M. Tanoue, Y. Hirabayashi, H. Ikeuchi, Global-scale river flood vulnerability in
eration by another journal. All authors have approved the manuscript the last 50 years, Sci. Rep. 6 (2016) 36021, https://doi.org/10.1038/srep36021.
and agreed with the submission to Automation in Construction. [23] M. Syifa, S.J. Park, A.R. Achmad, C.-W. Lee, J. Eom, Flood mapping using remote
sensing imagery and artificial intelligence techniques: a case study in
Brumadinho, Brazil, J. Coast. Res. 90 (2019) 197, https://doi.org/10.2112/SI90-
References 024.1.
[24] S. Supratid, T. Aribarg, S. Supharatid, An integration of stationary wavelet
[1] Q. Zou, J. Zhou, C. Zhou, L. Song, J. Guo, Y. Liu, The practical research on flood transform and nonlinear autoregressive neural network with exogenous input for
risk analysis based on IIOSM and fuzzy α-cut technique, Appl. Math. Model. 36 baseline and future forecasting of reservoir inflow, Water Resour. Manag. 31
(2012) 3271–3282, https://doi.org/10.1016/j.apm.2011.10.008. (2017) 4023–4043, https://doi.org/10.1007/s11269-017-1726-2.
[2] Q. Zhang, N. Jindapetch, R. Duangsoithong, D. Buranapanichkit, Investigation of [25] J. Sulaiman, S.H. Wahab, Heavy rainfall forecasting model using artificial neural
image processing based real-time flood monitoring, in: 2018 IEEE 5th network for flood prone area, in: Lecture Notes in Electrical Engineering,
International Conference on Smart Instrumentation, Measurement and Springer, 2018, pp. 68–76, https://doi.org/10.1007/978-981-10-6451-7_9.
Application (ICSIMA), IEEE, 2018, pp. 1–4, https://doi.org/10.1109/ [26] K. Sowmya, C.M. John, N.K. Shrivasthava, Urban flood vulnerability zoning of
ICSIMA.2018.8688775. Cochin City, southwest coast of India, using remote sensing and GIS, Nat. Hazards
[3] Q. Zhang, N. Jindapetch, D. Buranapanichkit, Investigation of image edge 75 (2015) 1271–1286, https://doi.org/10.1007/s11069-014-1372-4.
detection techniques based flood monitoring in real-time, in: 16th International [27] S. Skakun, N. Kussul, A. Shelestov, O. Kussul, Flood hazard and flood risk
Conference on Electrical Engineering/Electronics, Computer, assessment using a time series of satellite images: a case study in Namibia, Risk
Telecommunications and Information Technology (ECTI-CON) 2019, IEEE, 2019, Anal. 34 (2014) 1521–1537, https://doi.org/10.1111/risa.12156.
pp. 927–930, https://doi.org/10.1109/ECTI-CON47248.2019.8955273. [28] R. Sivanpillai, B.K. Jones, R.M. Lamb, Accessing satellite imagery for disaster
[4] S. Zakaria, M.R. Mahadi, A.F. Abdullah, K. Abdan, Aerial platform reliability for response through the international charter: lessons learned from the 2011 US
flood monitoring under various weather conditions: a review, in: Lecture Notes in Midwestern floods, Space Policy 42 (2017) 54–61, https://doi.org/10.1016/j.
Geoinformation and Cartography, 2019, pp. 295–314, https://doi.org/10.1007/ spacepol.2017.08.003.
978-3-030-05330-7_12. [29] C. Shukun, Z. Heng, Z. Xiangbo, Y. Qiujuan, A. Changsheng, Software and
[5] P.-S. Yu, T.-C. Yang, S.-Y. Chen, C.-M. Kuo, H.-W. Tseng, Comparison of random hardware platform design for open-CNC system, in: IEEE International
forests and support vector machine for real-time radar-derived rainfall Symposium on Knowledge Acquisition and Modeling Workshop 2008, IEEE,
forecasting, J. Hydrol. 552 (2017) 92–104, https://doi.org/10.1016/j. 2008, pp. 139–142, https://doi.org/10.1109/KAMW.2008.4810444.
jhydrol.2017.06.020. [30] C. Shu, T.B.M.J. Ouarda, Regional flood frequency analysis at ungauged sites
[6] B. Yu, Y. Zhu, T. Wang, Y.B. Zhu, Research on the 10e-minute rainfall prediction using the adaptive neuro-fuzzy inference system, J. Hydrol. 349 (2008) 31–43,
model for debris flows, Adv. Water Sci. 26 (2015) 347–355, https://doi.org/ https://doi.org/10.1016/j.jhydrol.2007.10.050.
10.14042/j.cnki.32.1309.2015.03.006. [31] J. Ryu, E.J. Yoon, C. Park, D.K. Lee, S.W. Jeon, A flood risk assessment model for
[7] T. Yoshida, K. Hiroi, Y. Yamagata, D. Murakami, Verification on evacuation of companies and criteria for governmental decision-making to minimize hazards,
flood disaster by using Gps: case study in Mabi, Japan 2018, in: IGARSS Sustainability 9 (2017) 2005, https://doi.org/10.3390/su9112005.
2019–2019 IEEE International Geoscience and Remote Sensing Symposium, IEEE, [32] X. Shan, J. Zhou, F. Xiao, Support vector machine method for multivariate density
2019, pp. 5633–5635, https://doi.org/10.1109/IGARSS.2019.8898574. estimation based on copulas, in: International Conference on Internet Computing
[8] M.A. Witherow, C. Sazara, I.M. Winter-Arboleda, M.I. Elbakary, M. Cetin, K. and Information Services 2011, IEEE, 2011, pp. 140–143, https://doi.org/
M. Iftekharuddin, Floodwater detection on roadways from crowdsourced images, 10.1109/ICICIS.2011.41.
in: Computer Metehods in Biomechanics and Biomedical Engineering: Imaging & [33] S. Seuring, A review of modeling approaches for sustainable supply chain
Visualization 7, 2019, pp. 529–540, https://doi.org/10.1080/ management, Decis. Support. Syst. 54 (2013) 1513–1520, https://doi.org/
21681163.2018.1488223. 10.1016/j.dss.2012.05.053.
[9] I.R. Widiasari, L.E. Nugroho, Widyawan, Deep learning multilayer perceptron [34] G. Schumann, P.D. Bates, M.S. Horritt, P. Matgen, F. Pappenberger, Progress in
(MLP) for flood prediction model using wireless sensor network based hydrology integration of remote sensing–derived flood extent and stage data and hydraulic
time series data mining, in: 2017 International Conference on Innovative and models, Rev. Geophys. 47 (2009), RG4001, https://doi.org/10.1029/
Creative Information Technology (ICITech), IEEE, 2017, pp. 1–5, https://doi.org/ 2008RG000274.
10.1109/INNOCIT.2017.8319150. [35] G. Schumann, Preface: remote sensing in flood monitoring and management,
[10] G.K. Wedajo, LiDAR DEM data for flood mapping and assessment; opportunities Remote Sens. 7 (2015) 17013–17015, https://doi.org/10.3390/rs71215871.
and challenges: a review, J. Rem. Sens. GIS 06 (2017) 2015–2018, https://doi. [36] S. Sankaranarayanan, M. Prabhakar, S. Satish, P. Jain, A. Ramprasad,
org/10.4172/2469-4134.1000211. A. Krishnan, Flood prediction based on weather parameters using deep learning,
[11] Y. Wang, Z. Li, Z. Tang, G. Zeng, A GIS-based spatial multi-criteria approach for J. Water Clim. Change 11 (2020) 1766–1783, https://doi.org/10.2166/
flood risk assessment in the Dongting Lake region, Hunan, Central China, Water wcc.2019.321.
Resour. Manag. 25 (2011) 3465–3484, https://doi.org/10.1007/s11269-011- [37] N.M. Robertson, T. Chan, Aerial image segmentation for flood risk analysis, in:
9866-2. 2009 16th IEEE International Conference on Image Processing (ICIP), IEEE, 2009,
[12] G. Wang, L. Guo, H. Duan, Wavelet neural network using multiple wavelet pp. 597–600, https://doi.org/10.1109/ICIP.2009.5413865.
functions in target threat assessment, Sci. World J. 2013 (2013) 1–7, https://doi. [38] M. Rivas Casado, T. Irvine, S. Johnson, M. Palma, P. Leinster, The use of
org/10.1155/2013/632437. unmanned aerial vehicles to estimate direct tangible losses to residential
[13] D. Wagenaar, A. Curran, M. Balbi, A. Bhardwaj, R. Soden, E. Hartato, G. Mestav properties from flood events: a case study of Cockermouth following the Desmond
Sarica, L. Ruangpan, G. Molinario, D. Lallemant, Invited perspectives: how storm, Remote Sens. 10 (2018) 1548, https://doi.org/10.3390/rs10101548.
machine learning will change flood risk and impact assessment, Nat. Hazards [39] F. Raymond, B. Wilhelm, S. Anquetin, The role of precipitation for high-
Earth Syst. Sci. 20 (2020) 1149–1161, https://doi.org/10.5194/nhess-20-1149- magnitude flood generation in a large mountainous catchment (upper Rhône
2020. River, NW European Alps), in: Hydrology and Earth System Sciences Discussions,
[14] M. Valipour, M.E. Banihabib, S.M.R. Behbahani, Comparison of the ARMA, 2019, pp. 1–26, https://doi.org/10.5194/hess-2019-100.
ARIMA, and the autoregressive artificial neural network models in forecasting the [40] A.B. Ranit, P.V. Durge, Different techniques of flood forecasting and their
monthly inflow of Dez dam reservoir, J. Hydrol. 476 (2013) 433–441, https:// applications, in: International Conference on Research in Intelligent and
doi.org/10.1016/j.jhydrol.2012.11.017. Computing in Engineering (RICE) 2018, IEEE, 2018, pp. 1–3, https://doi.org/
[15] K. Uddin, M.A. Matin, F.J. Meyer, Operational flood mapping using multi- 10.1109/RICE.2018.8509058.
temporal Sentinel-1 SAR images: a case study from Bangladesh, Remote Sens. 11 [41] B. RAM, A.S. KOLARKAR, Remote sensing application in monitoring land-use
(2019) 1581, https://doi.org/10.3390/rs11131581. changes in arid Rajasthan, Int. J. Remote Sens. 14 (1993) 3191–3200, https://doi.
[16] G. Tuna, B. Nefzi, G. Conte, Unmanned aerial vehicle-aided communications org/10.1080/01431169308904433.
system for disaster recovery, J. Netw. Comput. Appl. 41 (2014) 27–36, https:// [42] M. Rahnemoonfar, R. Murphy, M.V. Miquel, D. Dobbs, A. Adams, Flooded area
doi.org/10.1016/j.jnca.2013.10.002. detection from uav images based on densely connected recurrent neural
[17] P.R.J. Trim, An integrative approach to disaster management and planning, networks, in: IGARSS 2018–2018 IEEE International Geoscience and Remote
Disaster Prev Manag 13 (2004) 218–225, https://doi.org/10.1108/ Sensing Symposium, IEEE, 2018, pp. 1788–1791, https://doi.org/10.1109/
09653560410541812. IGARSS.2018.8517946.
[18] J. Transon, R. D’Andrimont, A. Maugnard, P. Defourny, Survey of hyperspectral [43] M.R. Rahman, P.K. Thakur, Detecting, mapping and analysing of flood water
earth observation applications from space in the sentinel-2 context, Remote Sens. propagation using synthetic aperture radar (SAR) satellite data and GIS: a case
10 (2018) 157, https://doi.org/10.3390/rs10020157. study from the Kendrapara District of Orissa State of India, Egypt. J. Remote Sens.
[19] D. Tranfield, D. Denyer, P. Smart, Towards a methodology for developing Space Sci. 21 (2018) S37–S41, https://doi.org/10.1016/j.ejrs.2017.10.002.
evidence-informed management knowledge by means of systematic review, Br. J. [44] M. Rahman, N. Chen, A. Elbeltagi, M.M. Islam, M. Alam, H.R. Pourghasemi,
Manag. 14 (2003) 207–222, https://doi.org/10.1111/1467-8551.00375. W. Tao, J. Zhang, T. Shufeng, H. Faiz, M.A. Baig, A. Dewan, Application of
stacking hybrid machine learning algorithms in delineating multi-type flooding in

16
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

Bangladesh, J. Environ. Manag. 295 (2021) 113086, https://doi.org/10.1016/j. [69] A.K. Lohani, N.K. Goel, K.K.S. Bhatia, Improving real time flood forecasting using
jenvman.2021.113086. fuzzy inference system, J. Hydrol. 509 (2014) 25–41, https://doi.org/10.1016/j.
[45] S. Raghavendra, P.C. Deka, Support vector machine applications in the field of jhydrol.2013.11.021.
hydrology: a review, Appl. Soft Comput. 19 (2014) 372–386, https://doi.org/ [70] A.K. Lohani, R. Kumar, R.D. Singh, Hydrological time series modeling: a
10.1016/j.asoc.2014.02.002. comparison between adaptive neuro-fuzzy, neural network and autoregressive
[46] M. Pregnolato, A. Ford, S.M. Wilkinson, R.J. Dawson, The impact of flooding on techniques, J. Hydrol. 442–443 (2012) 23–35, https://doi.org/10.1016/j.
road transport: a depth-disruption function, Transp. Res. Part D: Transp. Environ. jhydrol.2012.03.031.
55 (2017) 67–81, https://doi.org/10.1016/j.trd.2017.06.020. [71] S.-W. Lo, J.-H. Wu, F.-P. Lin, C.-H. Hsu, Visual sensing for urban flood
[47] P. Prasad, V.J. Loveson, B. Das, M. Kotha, Novel ensemble machine learning monitoring, Sensors 15 (2015) 20006–20029, https://doi.org/10.3390/
models in flood susceptibility mapping, Geocarto Int. (2021) 1–23, https://doi. s150820006.
org/10.1080/10106049.2021.1892209. [72] C. Liu, Y. Li, GIS-based dynamic modelling and analysis of flash floods
[48] D. Popescu, L. Ichim, T. Caramihale, Flood areas detection based on UAV considering land-use planning, Int. J. Geogr. Inf. Sci. 31 (2017) 481–498, https://
surveillance system, in: 2015 19th International Conference on System Theory, doi.org/10.1080/13658816.2016.1207774.
Control and Computing (ICSTCC), IEEE, 2015, pp. 753–758, https://doi.org/ [73] Y.-T. Lin, M.-D. Yang, J.-Y. Han, Y.-F. Su, J.-H. Jang, Quantifying flood water
10.1109/ICSTCC.2015.7321384. levels using image-based volunteered geographic information, Remote Sens. 12
[49] E.J. Plate, Flood risk and flood management, J. Hydrol. 267 (2002) 2–11, https:// (2020) 706, https://doi.org/10.3390/rs12040706.
doi.org/10.1016/S0022-1694(02)00135-X. [74] S. Li, K. Ma, Z. Jin, Y. Zhu, A new flood forecasting model based on SVM and
[50] J.M. Peters, R.S. Schumacher, Mechanisms for organization and echo training in a boosting learning algorithms, in: 2016 IEEE Congress on Evolutionary
flash-flood-producing mesoscale convective system, Mon. Weather Rev. 143 Computation (CEC), IEEE, 2016, pp. 1343–1348, https://doi.org/10.1109/
(2015) 1058–1085, https://doi.org/10.1175/MWR-D-14-00070.1. CEC.2016.7743944.
[51] D. PANAGOULIA, Artificial neural networks and high and low flows in various [75] L. Li, H. Xu, X. Chen, S.P. Simonovic, Streamflow forecast and reservoir operation
climate regimes, Hydrol. Sci. J. 51 (2006) 563–587, https://doi.org/10.1623/ performance assessment under climate change, Water Resour. Manag. 24 (2010)
hysj.51.4.563. 83–104, https://doi.org/10.1007/s11269-009-9438-x.
[52] K. Owusu Twum, M. Abubakari, Cities and floods: a pragmatic insight into the [76] J. Langhammer, T. Vacková, Detection and mapping of the geomorphic effects of
determinants of households’ coping strategies to floods in informal Accra, Ghana, flooding using UAV photogrammetry, Pure Appl. Geophys. 175 (2018)
Jàmbá J. Disast. Risk Stud. 11 (2019) 1–14, https://doi.org/10.4102/jamba. 3223–3245, https://doi.org/10.1007/s00024-018-1874-1.
v11i1.608. [77] P. Lamovec, T. Veljanovski, M. Mikoš, K. Oštir, Detecting flooded areas with
[53] Y.O. Ouma, R. Tateishi, Urban flood vulnerability and risk mapping using machine learning techniques: case study of the Selška Sora river flash flood in
integrated multi-parametric AHP and GIS: methodological overview and case September 2007, J. Appl. Remote. Sens. 7 (2013), 073564, https://doi.org/
study assessment, Water 6 (2014) 1515–1545, https://doi.org/10.3390/ 10.1117/1.JRS.7.073564.
w6061515. [78] P. Lal, A. Prakash, A. Kumar, P.K. Srivastava, P. Saikia, A.C. Pandey,
[54] C. OTTLE, M. STOLL, Effect of atmospheric absorption and surface emissivity on P. Srivastava, M.L. Khan, Evaluating the 2018 extreme flood hazard events in
the determination of land surface temperature from infrared satellite data, Int. J. Kerala, India, Rem. Sens. Lett. 11 (2020) 436–445, https://doi.org/10.1080/
Remote Sens. 14 (1993) 2025–2037, https://doi.org/10.1080/ 2150704X.2020.1730468.
01431169308954018. [79] V. Klemas, Remote sensing of floods and flood-prone areas: an overview, J. Coast.
[55] Gabarró Olmedo, Martínez González-Gambau, Turiel Ballabrera-Poy, Res. 314 (2015) 1005–1013, https://doi.org/10.2112/JCOASTRES-D-14-
Fournier Portabella, Lee, E. Olmedo, et al., Seven years of SMOS sea surface 00160.1.
salinity at high latitudes: variability in Arctic and Sub-Arctic regions, Remote [80] Ö. Kişi, A combined generalized regression neural network wavelet model for
Sens. 11 (8) (2019) 940, https://doi.org/10.3390/rs11080940. monthly streamflow prediction, KSCE J. Civ. Eng. 15 (2011) 1469–1479, https://
[56] N. Nirupama, S.P. Simonovic, Increase of flood risk due to urbanisation: a doi.org/10.1007/s12205-011-1004-4.
Canadian example, Nat. Hazards 40 (2007) 25–41, https://doi.org/10.1007/ [81] V.Y. Kishorbhai, N.N. Vasantbhai, AON: a survey on emergency communication
s11069-006-0003-0. systems during a catastrophic disaster, Proc. Comput. Sci. 115 (2017) 838–845,
[57] T. Nguyen, J. Han, D.-C. Park, Satellite image classification using convolutional https://doi.org/10.1016/j.procs.2017.09.166.
learning, in: AIP Conference Proceedings, 2013, pp. 2237–2240, https://doi.org/ [82] K.M. Kafi, M.B.A. Gibril, GPS application in disaster management : a review GPS
10.1063/1.4825984. application in disaster management : a review, Asian J. Appl. Sci. 4 (2018), pp.
[58] R. Narayanan, V.M. Lekshmy, S. Rao, K. Sasidhar, A novel approach to urban 2321–0893.
flood monitoring using computer vision, in: Fifth International Conference on [83] K.E. Joyce, S.E. Belliss, S.V. Samsonov, S.J. McNeill, P.J. Glassey, A review of the
Computing, Communications and Networking Technologies (ICCCNT), IEEE, status of satellite remote sensing and image processing techniques for mapping
2014, pp. 1–7, https://doi.org/10.1109/ICCCNT.2014.6962989. natural hazards and disasters, in: Progress in Physical Geography: Earth and
[59] N. Najibi, N. Devineni, Recent trends in the frequency and duration of global Environment 33, 2009, pp. 183–207, https://doi.org/10.1177/
floods, Earth Syst. Dyn. 9 (2018) 757–783, https://doi.org/10.5194/esd-9-757- 0309133309339563.
2018. [84] A.R. Joshi, I. Tarte, S. Suresh, S.G. Koolagudi, Damage identification and
[60] D. Munasinghe, S. Cohen, Y.-F. Huang, Y.-P. Tsang, J. Zhang, Z. Fang, assessment using image processing on post-disaster satellite imagery, in: 2017
Intercomparison of satellite remote sensing-based flood inundation mapping IEEE Global Humanitarian Technology Conference (GHTC), IEEE, 2017, pp. 1–7,
techniques, JAWRA J. Am. Water Resour. Assoc. 54 (2018) 834–846, https://doi. https://doi.org/10.1109/GHTC.2017.8239286.
org/10.1111/1752-1688.12626. [85] S. Jonkman, R. Dawson, Issues and challenges in flood risk management&amp;
[61] M. Motta, M. de Castro Neto, P. Sarmento, A mixed approach for urban flood mdash;editorial for the special issue on flood risk management, Water 4 (2012)
prediction using machine learning and GIS, Int. J. Disast. Risk Reduction 56 785–792, https://doi.org/10.3390/w4040785.
(2021) 102154, https://doi.org/10.1016/j.ijdrr.2021.102154. [86] J.M. Jones, S.T. Gille, H. Goosse, N.J. Abram, P.O. Canziani, D.J. Charman, K.
[62] F. Mohammadimanesh, B. Salehi, M. Mahdianpari, B. Brisco, M. Motagh, Multi- R. Clem, X. Crosta, C. de Lavergne, I. Eisenman, M.H. England, R.L. Fogt, L.
temporal, multi-frequency, and multi-polarization coherence and SAR backscatter M. Frankcombe, G.J. Marshall, V. Masson-Delmotte, A.K. Morrison, A.J. Orsi, M.
analysis of wetlands, ISPRS J. Photogramm. Remote Sens. 142 (2018) 78–93, N. Raphael, J.A. Renwick, D.P. Schneider, G.R. Simpkins, E.J. Steig, B. Stenni,
https://doi.org/10.1016/j.isprsjprs.2018.05.009. D. Swingedouw, T.R. Vance, Assessing recent trends in high-latitude southern
[63] I. McCallum, W. Liu, L. See, R. Mechler, A. Keating, S. Hochrainer-Stigler, hemisphere surface climate, Nat. Clim. Chang. 6 (2016) 917–926, https://doi.
J. Mochizuki, S. Fritz, S. Dugar, M. Arestegui, M. Szoenyi, J.-C.L. Bayas, P. Burek, org/10.1038/nclimate3103.
A. French, I. Moorthy, Technologies to support community flood disaster risk [87] C.L. Johnson, S.J. Priest, Flood risk management in England: a changing
reduction, Int. J. Disast. Risk Sci. 7 (2016) 198–204, https://doi.org/10.1007/ landscape of risk responsibility? Int. J. Water Resour. Develop. 24 (2008)
s13753-016-0086-5. 513–525, https://doi.org/10.1080/07900620801923146.
[64] D.C. Mason, R. Speck, B. Devereux, G.J.-P. Schumann, J.C. Neal, P.D. Bates, Flood [88] K. John Samuel Raj, K. Naresh Kumar Thapa, R. Balakrishnan, Post disaster
detection in urban areas using TerraSAR-X, IEEE Trans. Geosci. Remote Sens. 48 damage estimation using integrated GPS sensor network &amp; GIS, in: 2016
(2010) 882–894, https://doi.org/10.1109/TGRS.2009.2029236. International Conference on Wireless Communications, Signal Processing and
[65] D. Marmanis, J.D. Wegner, S. Galliani, K. Schindler, M. Datcu, U. Stilla, Semantic Networking (WiSPNET), IEEE, 2016, pp. 1234–1238, https://doi.org/10.1109/
segmentation of aerial images with an ensemble of CNNS, in: ISPRS Annals of the WiSPNET.2016.7566333.
Photogrammetry, Remote Sensing and Spatial Information Sciences. III–3, 2016, [89] A. Irimescu, D. Mihailescu, G. Stancalie, Sentinel data for flood disaster
pp. 473–480, https://doi.org/10.5194/isprs-annals-III-3-473-2016. monitoring and assessment: case studies in Romania, in: Science and Practice for
[66] P. Mallikarjuna, C.H. Suresh Babu, A.J.M. Reddy, Rainfall—runoff modelling an Uncertain Future, Budapest University of Technology and Economics, 2021,
using artificial neural networks, ISH J. Hydraulic Eng. 15 (2009) 24–33, https:// https://doi.org/10.3311/FloodRisk2020.7.5. Online.
doi.org/10.1080/09715010.2009.10514928. [90] F. Ip, J.M. Dohm, V.R. Baker, T. Doggett, A.G. Davies, R. Castaño, S. Chien,
[67] X. Lu, B. Mao, B. Dong, Comment on “an efficient and stable hydrodynamic model B. Cichy, R. Greeley, R. Sherwood, D. Tran, G. Rabideau, Flood detection and
with novel source term discretization schemes for overland flow and flood monitoring with the Autonomous Sciencecraft Experiment onboard EO-1, Remote
simulations” by Xilin Xia et al, Water Resour. Res. 54 (2018) 621–627, https:// Sens. Environ. 101 (2006) 463–481, https://doi.org/10.1016/j.rse.2005.12.018.
doi.org/10.1002/2017WR021563. [91] U.G. Inyang, E.E. Akpan, O.C. Akinyokun, A hybrid machine learning approach
[68] P. Lu, L. Wang, Z. Niu, L. Li, W. Zhang, Prediction of soil properties using for flood risk assessment and classification, Int. J. Comput. Intell. Appl. 19 (2020)
laboratory VIS–NIR spectroscopy and Hyperion imagery, J. Geochem. Explor. 132 2050012, https://doi.org/10.1142/S1469026820500121.
(2013) 26–33, https://doi.org/10.1016/j.gexplo.2013.04.003. [92] J.A. Klemens, R.G. Harper, J.A. Frick, A.P. Capparella, H.B. Richardson, M.
J. Coffey, Patterns of organochlorine pesticide contamination in Neotropical

17
H.S. Munawar et al. Automation in Construction 132 (2021) 103916

migrant passerines in relation to diet and winter habitat, Chemosphere 41 (2000) [108] S.H. Elsafi, Artificial neural networks (ANNs) for flood forecasting at Dongola
1107–1113, https://doi.org/10.1016/S0045-6535(99)00561-5. Station in the River Nile, Sudan, Alexandria Eng. J. 53 (2014) 655–662, https://
[93] L. Hashemi-Beni, J. Jones, G. Thompson, C. Johnson, A. Gebrehiwot, Challenges doi.org/10.1016/j.aej.2014.06.010.
and opportunities for UAV-based digital elevation model generation for flood-risk [109] L.K. da Cunha, Exploring the benefits of satellite remote sensing for flood
management: a case of Princeville, North Carolina, Sensors. 18 (2018) 3843, prediction across scales, Univ. Iowa (2012), https://doi.org/10.17077/etd.
https://doi.org/10.3390/s18113843. es9ooi4m.
[94] Q. Li, H. Xiong, F. Zhang, Broadcast revocation scheme in composite-order [110] C. Polykretis, C. Chalkias, M. Ferentinou, Adaptive neuro-fuzzy inference system
bilinear group and its application to attribute-based encryption, Int. J. Secur. (ANFIS) modeling for landslide susceptibility assessment in a Mediterranean hilly
Netw. 8 (2013) 1, https://doi.org/10.1504/IJSN.2013.055036. area, Bulletin of Engineering Geology and the Environment 78 (2019)
[95] F. Han, J. Qin, H. Zhao, J. Hu, A general transformation from KP-ABE to 1173–1187, https://doi.org/10.1007/s10064-017-1125-1.
searchable encryption, Futur. Gener. Comput. Syst. 30 (2014) 107–115, https:// [111] J.-F. Crétaux, A. Arsen, S. Calmant, A. Kouraev, V. Vuglinski, M. Bergé-Nguyen,
doi.org/10.1016/j.future.2013.09.013. M.-C. Gennero, F. Nino, R. Abarca Del Rio, A. Cazenave, P. Maisongrande, SOLS: a
[96] A.C. Hall, G.J.P. Schumann, J.L. Bamber, P.D. Bates, M.A. Trigg, Geodetic lake database to monitor in the near real time water level and storage variations
corrections to Amazon River water level gauges using ICESat altimetry, Water from remote sensing data, Adv. Space Res. 47 (2011) 1497–1507, https://doi.
Resour. Res. 48 (2012), https://doi.org/10.1029/2011WR010895. org/10.1016/j.asr.2011.01.004.
[97] B.B. Gupta, M. Quamara, An overview of Internet of Things (IoT): architectural [112] S. Coveney, K. Roberts, Lightweight UAV digital elevation models and
aspects, challenges, and protocols, Concurr. Comput. 32 (2020) 4946, https:// orthoimagery for environmental applications: data accuracy evaluation and
doi.org/10.1002/cpe.4946. potential for river flood risk modelling, Int. J. Remote Sens. 38 (2017)
[98] C.A. Guimarães Santos, G.B.L. da Silva, Daily streamflow forecasting using a 3159–3180, https://doi.org/10.1080/01431161.2017.1292074.
wavelet transform and artificial neural network hybrid models, Hydrol. Sci. J. 59 [113] R. Costache, Q.B. Pham, E. Sharifi, N.T.T. Linh, S.I. Abba, M. Vojtek, J. Vojteková,
(2014) 312–324, https://doi.org/10.1080/02626667.2013.800944. P.T.T. Nhi, D.N. Khoi, Flash-flood susceptibility assessment using multi-criteria
[99] L. Guanter, H. Kaufmann, K. Segl, S. Foerster, C. Rogass, S. Chabrillat, T. Kuester, decision making and machine learning supported by remote sensing and GIS
A. Hollstein, G. Rossner, C. Chlebek, C. Straif, S. Fischer, S. Schrader, T. Storch, techniques, Remote Sens. 12 (2019) 106, https://doi.org/10.3390/rs12010106.
U. Heiden, A. Mueller, M. Bachmann, H. Mühle, R. Müller, M. Habermeyer, [114] K. Beven, J. Hall, Applied Uncertainty Analysis for Flood Risk Management,
A. Ohndorf, J. Hill, H. Buddenbaum, P. Hostert, S. van der Linden, P. Leitão, Imperial College Press, 2014, https://doi.org/10.1142/p588.
A. Rabe, R. Doerffer, H. Krasemann, H. Xi, W. Mauser, T. Hank, M. Locherer, [115] D. Bandyopadhyay, J. Sen, Internet of things: applications and challenges in
M. Rast, K. Staenz, B. Sang, The EnMAP spaceborne imaging spectroscopy mission technology and standardization, Wirel. Pers. Commun. 58 (2011) 49–69, https://
for earth observation, Remote Sens. 7 (2015) 8830–8857, https://doi.org/ doi.org/10.1007/s11277-011-0288-5.
10.3390/rs70708830. [116] M. Azeem, F. Ullah, M.J. Thaheem, S. Qayyum, Competitiveness in the
[100] S. Grimaldi, Y. Li, V.R.N. Pauwels, J.P. Walker, Remote sensing-derived water construction industry: a contractor’s perspective on barriers to improving the
extent and level to constrain hydraulic flood forecasting models: opportunities construction industry performance, J. Construct. Eng. Manag. Innov. 3 (2020)
and challenges, Surv. Geophys. 37 (2016) 977–1034, https://doi.org/10.1007/ 193–219, https://doi.org/10.31462/jcemi.2020.03193219.
s10712-016-9378-y. [117] M. Aslam, Flood management current state, challenges and prospects in Pakistan:
[101] O. Ghorbanzadeh, H. Rostamzadeh, T. Blaschke, K. Gholaminia, J. Aryal, A new a review, Mehran Univ. Res. J. Eng. Technol. 37 (2018) 297–314, https://doi.org/
GIS-based data mining technique using an adaptive neuro-fuzzy inference system 10.22581/muet1982.1802.06.
(ANFIS) and k-fold cross-validation approach for land subsidence susceptibility [118] M. Ashrafi, L.H.C. Chua, C. Quek, X. Qin, A fully-online Neuro-Fuzzy model for
mapping, Nat. Hazards 94 (2018) 497–517, https://doi.org/10.1007/s11069- flow forecasting in basins with limited data, J. Hydrol. 545 (2017) 424–435,
018-3449-y. https://doi.org/10.1016/j.jhydrol.2016.11.057.
[102] M. Gerke, H.-J. Przybilla, Accuracy analysis of photogrammetric UAV image [119] R. Arthur, C.A. Boulton, H. Shotton, H.T.P. Williams, Social sensing of floods in
blocks: influence of onboard RTK-GNSS and cross flight patterns, the UK, PLoS One 13 (2018), e0189327, https://doi.org/10.1371/journal.
Photogrammetrie - Fernerkundung - Geoinformation. 2016 (2016) 17–30, pone.0189327.
https://doi.org/10.1127/pfg/2016/0284. [120] M. Arslan, A.-M. Roxin, C. Cruz, D. Ginhac, A review on applications of big data
[103] A. Gebrehiwot, L. Hashemi-Beni, G. Thompson, P. Kordjamshidi, T. Langan, Deep for disaster management, in: 2017 13th International Conference on Signal-Image
convolutional neural network for flood extent mapping using unmanned aerial Technology & Internet-Based Systems (SITIS), IEEE, 2017, pp. 370–375, https://
vehicles data, Sensors 19 (2019) 1486, https://doi.org/10.3390/s19071486. doi.org/10.1109/SITIS.2017.67.
[104] S. Frigerio, L. Schenato, G. Bossi, M. Mantovani, G. Marcato, A. Pasuto, Hands-on [121] N. Anusha, B. Bharathi, Flood detection and flood mapping using multi-temporal
experience of crowdsourcing for flood risks. An android mobile application tested synthetic aperture radar and optical data, Egypt. J. Remote Sens. Space Sci. 23
in Frederikssund, Denmark, Int. J. Environ. Res. Public Health 15 (2018) 1926, (2020) 207–219, https://doi.org/10.1016/j.ejrs.2019.01.001.
https://doi.org/10.3390/ijerph15091926. [122] B. Anbalagan, C. Valliyammai, #ChennaiFloods: leveraging human and machine
[105] J. Fernández-Hernandez, D. González-Aguilera, P. Rodríguez-Gonzálvez, learning for crisis mapping during disasters using social media, in: Proceedings -
J. Mancera-Taboada, Image-based modelling from unmanned aerial vehicle 23rd IEEE International Conference on High Performance Computing Workshops,
(UAV) photogrammetry: an effective, low-cost tool for archaeological HiPCW 2016, 2017, pp. 50–59, https://doi.org/10.1109/HiPCW.2016.17.
applications, Archaeometry 57 (2015) 128–145, https://doi.org/10.1111/ [123] B. Alizadeh Kharazi, A.H. Behzadan, Flood depth mapping in street photos with
arcm.12078. image processing and deep neural networks, Comput. Environ. Urban. Syst. 88
[106] C.A.F. Ezequiel, M. Cua, N.C. Libatique, G.L. Tangonan, R. Alampay, R. (2021) 101628, https://doi.org/10.1016/j.compenvurbsys.2021.101628.
T. Labuguen, C.M. Favila, J.L.E. Honrado, V. Canos, C. Devaney, A.B. Loreto, [124] A. Alipour, A. Ahmadalipour, P. Abbaszadeh, H. Moradkhani, Leveraging
J. Bacusmo, B. Palma, UAV aerial imaging applications for post-disaster machine learning for predicting flash flood damage in the Southeast US, Environ.
assessment, environmental management and infrastructure development, in: Res. Lett. 15 (2020), 024011, https://doi.org/10.1088/1748-9326/ab6edd.
2014 International Conference on Unmanned Aircraft Systems (ICUAS), IEEE, [125] M. Al-Fawa’Reh, A. Hawamdeh, R. Alrawashdeh, M.T. Jafar, Intelligent methods
2014, pp. 274–283, https://doi.org/10.1109/ICUAS.2014.6842266. for flood forecasting in Wadi al Wala, Jordan, in: 2021 International Congress of
[107] A. Enders, Z. Brandt, Using geographic information system technology to improve Advanced Technology and Engineering (ICOTEN), IEEE, 2021, pp. 1–9, https://
emergency management and disaster response for people with disabilities, doi.org/10.1109/ICOTEN52080.2021.9493425.
J. Disability Pol. Stud. 17 (2007) 223–229, https://doi.org/10.1177/ [126] S. Akter, S.F. Wamba, Big data and disaster management: a systematic review and
10442073070170040501. agenda for future research, Ann. Oper. Res. 283 (2019) 939–959, https://doi.org/
10.1007/s10479-017-2584-2.

18