Journal of Geomatics Vol. 15, No.

2, October 2021

Flood Inundation Mapping and Depth Modelling using Machine Learning algorithms and
Microwave data
Gyan Prakash1*, Praveen Kumar Gupta2, G Venkata Rao1 and Deva Pratap1,
1Department of Civil Engineering, National Institute of Technology Warangal, Telangana, 506004, India
2Space Applications Centre, ISRO, Ahmedabad, Gujarat, 380015, India

*Email: [email protected]

(Received: Aug 10, 2021; in final form: Oct 6, 2021)

Abstract: Flooding is one of the most devastating natural hazards that significantly impact human life and property.
During floods, monitoring and mapping flood extent is crucial in identifying the flood-affected areas and the damage
assessment. Space-based monitoring of floods can provide a systematic, spatial, timely, and impartial way to monitor
disastrous floods. The study area is a part of the Kosi River in the Bihar state. In this study, using microwave remote
sensing data (Sentinel - 1), an independent and open-source tool was developed to monitor the flooding extent and water
depth. The tool consists of a hybrid model and a floodwater depth analysis model: The hybrid model is fully automated
in which Binarization techniques and Random Forest Classifier (RFC) and K Nearest Neighbor (KNN), supervised
Machine Learning(ML) algorithms were used. Using flood inundation maps and Digital Elevation Model (DEM), the
floodwater depth analysis model (PyQGIS standalone tool) was developed to calculate the flood water depth. Supervised
classification algorithms in the hybrid model were further compared and found that the performance of both the KNN
and RFC classifiers was close enough, but the time taken by RFC was less than KNN Classifier. The model results were
compared and validated with the August 2017 flood event results over the Darbhanga district. The results of the fully
automated model have shown a deviation of 0.9% to 19% compared with the published results over the Darbhanga
district. The present study suggests that the RFC ML algorithm can classify the SAR data into flooded and non-flooded
areas. The developed tool can be used to monitor floods in near/real-time to issue warnings to the people and rescue
operations.

Keywords: Flood Inundation, Floodwater Depth, Microwave Remote Sensing, Machine Learning Algorithms,
Supervised Classification

1. Introduction serious one is that the optical data doesn't have the ability
to penetrate through the clouds (Shen et al., 2019), which
Flood is one of the most devastating natural hazards which are mainly prevalent during the monsoon season.
is caused due to excessive increase in surface runoff, Microwave sensors are the alternative sources to
heavy rainfall, rise in the riverbed, cyclones, and cloud overcome the drawbacks of optical sensors. Due to the
bursts, etc. (Singh, 2015). Among the nations in the world, usage of longwave radiation, microwave sensors can
India is one of the most flood-affected countries due to its penetrate through the clouds and detailed vegetation
unique geo-climatic conditions, precipitation patterns, coverage. Microwave sensors are independent of solar
topographic features, population growth, urbanization, radiation, and they can provide the data in all weather
industrialization, etc. (Mohanty et al., 2020). According to conditions (Chang Huang et al., 2018; Shen et al., 2019).
National Flood Commission, out of a total geographical
area of approximately 329 million hectares, about 40 Several studies have been carried out for flood inundation
million hectares are prone to floods (Sharma et al., 2016; mapping and damage assessment using microwave data
Gangwar et al., 2013). Among all the river basins in India, (Anusha and Bharathi 2020; de Groeve 2010;
Ganga and Brahmaputra river basins experience the Gouweleeuw et al. 2011; D. C. Mason et al. 2012; Matgen
highest number of floods (Mohanty et al. 2020). et al. 2007; Schumann and Moller 2015; Shen et al. 2019;
Temimi et al. 2005; Tripathi et al. 2020, 2020). (Matgen
It is essential to have information about their intensities et al. 2007) extracted the flood extent and depth of
and extents to cope with the damage caused by floods. floodwater using DEM and SAR data with the help of the
Therefore, the preparation of flood inundation maps is the HEC-RAS river flow model and reported an RMSE of
primary step for damage control and assessing a flood 41cm for flood water depth. Using image segmentation,
event (Matgen et al., 2007). Compared to in-situ (David C. Mason et al. 2012) extracted the flood
measurement, remote sensing offers practical ways to inundation maps in urban and rural areas with an accuracy
observe and monitor the surface water dynamics at of 89% and a false-positive rate of 6%. For urban flood
multiple spatial and temporal scales. There are generally pixels using TerraSAR-X, 75% of pixels were accurately
two types of remote sensing datasets are available for the identified as water, with a false positive rate of 24%.
purpose of monitoring the surface water – the optical and Tripathi et al. 2020 used the Binarization method for the
microwave remote sensing data. Optical data has been classification of MODIS and SAR data by selecting
widely used to monitor and map surface water bodies due threshold values. They reported that MODIS data had
to the high availability and suitable Spatio-temporal shown an overestimation of 21% in the flood area
resolutions (Chang Huang et al., 2019). Although the compared with SAR data. Anusha and Bharathi 2020 used
optical data is numerously used for surface water body SAR and optical data for flood mapping of the August
extraction, the data has several limitations. The most

© Indian Society of Geomatics

221
Journal of Geomatics Vol. 15, No. 2, October 2021

2017 flood in Uttar Pradesh with the help of thresholding area is 49.81m and 100m max elevation was found in
and Unsupervised classification methods. SRTM DEM. Gole and Chitale 1966 reports that the Kosi
river is built by large sediment flux, which also plays a
ML algorithms such as Support vector machine, random vital role in causing westward shifting of Kosi and
forest, K Nearest Neighbor (KNN), Decision Tree (DT), extensive flooding. Thus, the Kosi river changes its course
K-means, and iso-data (ISO) cluster have been used in of flow frequently with a 24 year frequency period and
several studies to minimize the human interference and causes a lot of damage in the Northern Bihar region (Bhatt
time taken for flood mapping (Benoudjit and Guida 2019; et al. 2010). In August 2008, the Kosi River routed to its
Campolo et al. 1999; D Amitrano 2018; Elsafi 2014; Feng old course of flow, followed by the Kosi river 100 years
et al. 2015; Schumann and Moller 2015; Shahabi et al. ago, and this flood affected over 2.3 million people in the
2020; R. Sinha et al. 2008; Tehrany et al. 2014, 2015). northern area of Bihar state (Singh et al. 2011). In the
Benoudjit and Guida 2019 developed an algorithm for present study, taking these damages into account caused
flood mapping using Sentinel 1 and Sentinel 2 data with by Kosi river flood, a part of Kosi river basin which lies in
the help of NDWI and a supervised Classifier. They Bihar of 14,861.535 Km2 areas and 112 Km long river was
reported an overall accuracy of 77 % for the rural and selected study area as shown in Figure 1.
74.7% for the urban floods. Shahabi et al. 2020 developed
an ensemble model using KNN as meta classifier and 3. Data and Methodology
Weighted base classifier for flood inundation mapping.
Thus, from the above studies, it can be interpreted that the 3.1 Data and Pre-Processing
hybrid/ensemble model results in higher accuracy than Active microwave remote sensing data in C band, dual
individual models. For automation of flood mapping tools polarization with VV and VH polarization from Sentinel-
with high accuracy results, hybrid models were developed 1 satellite was used. The SAR data over the Kosi river
in several studies (Anusha and Bharathi 2020; Matgen et basin was acquired for flood events days - from Alaska
al. 2007; Tehrany et al. 2014; Twele et al. 2016). Satellite Facility (ASF) as Ground Range Detected (GRD)
product with a spatial resolution of 10 m and temporal
The above studies were mostly done for inundation resolution of 10 to 12 days (Table 1).
mapping. Thus, there is a need for a coupled model that
can also estimate the floodwater depth along with It is found in various studies that the VH polarization band
inundation extent. Floodwater Depth Estimation Tool is more useful in separating Water and Other Land features
(FwDET) was used to estimate an approximate water (Benoudjit and Guida 2019; Matgen et al. 2007; Tavus et
depth of the flood plain (Cohen et al., 2018). In this study, al. 2019; Tripathi et al. 2020; Twele et al. 2016) based on
a fully automated coupled model approach for flood their Backscatter value, which can be derived from SNAP
mapping and depth modeling was made. tool and all the classification was done on Sigma0_VH_db
band of SAR datasets.
2. Study Area
SNAP tool was used for pre-processing of SAR data such
North Bihar faces heavy damages due to floods in the Kosi as radiometric calibration, speckle filtering, orbit file,
river (Bhatt et al., 2010). During the last few years, the geometric correction etc. as shown in Figure 2.
Kosi River has changed its flow course by 150 Km and
caused damage to human lives and properties every year. Flooded pixels were identified using binarization
For more than five decades, flood control management has techniques as used in (Tripathi et al. 2020) by applying
been working for this basin but continues to bring harm threshold values as a trial and error process. These
through its devastating floods every year (R. Sinha et al. threshold values can be estimated from the histogram
2008). The geomorphological properties of the Kosi River shown in Figure 3 (b). There are two peaks can be seen in
have a significant role in these extensive floods. Kosi the histogram in Figure 3(b). Thus, it can be interpreted
flows through the slopes of the Himalayas in Tibet and the that the high peak shows other land features, and the low
Southern slopes in Nepal. After that, it enters into Indian peak shows water. Water mask band was created using the
region (Kosi River). following math in the "band math" in snap tools.

In Himalayan region, only it has three tributaries, Arjun, If σo_VH < th then 255*( σo_VH < th) else 0*( σo_VH > th &&
Tamur and Sun-Kosi. Three gauge/discharge stations σo_VH !=0) (1)
along the Kosi River, namely, Barahkshetra, Birpur, and
Baltara, were used by Central Water Commission (CWC), Similar equation was used for the VV band. Further, the
India. In which Barahkshetra and Birpur show higher peak water mask generated using the above equation was
discharges than Baltara for the same return period. The checked and found 78% similarity with a published global
annual average discharge at Kosi was found 2236 m3/s, the water mask data of the Kosi River basin. While this
average monsoon discharge 5156 m3/s being almost five published Global water mask (Pekel, J F., et al. 2016)
times higher than the non-monsoon discharge 1175 m3/ s shows a 50 % probability of flood extent, as shown in
huge difference the river vulnerable to extensive flooding Figure 4. Pre-processing of DEM data is performed using
(R. Sinha et al. 2008). The average elevation in the study QGIS software.

222
Journal of Geomatics Vol. 15, No. 2, October 2021

Figure 1. Study area

Table 1. Data used

Data Type/Format Source Acquisition Spatial Temporal Use Satellite
Date Resolution Resolution
GRD-HD ESA Open 24/07/2020 10m 12 days Flood Sentinel 1
with VH VV Access hub 11/08/2017, 10m 12 days Mapping Sentinel 1
Polarization and 23/08/2017,
in IW mode Alaska 04/09/2017,
SAR Satellite 16/09/2017
Facility
DEM Raster USGS 30m Flood water SRTM
depth
estimation

Figure 1. Flowchart of pre-processing

Figure 3. (a) SNAP tool interface for SAR pre-
processing (b) Backscatter value histogram

223
Journal of Geomatics Vol. 15, No. 2, October 2021

based on the concept of the binarization method and

supervised ML algorithms. Binarization includes the
selection of threshold backscatter values for water bodies
and flooded areas using histogram-based thresholds and
published global surface water mask data as ground truth.
Further, the flooded and the non-flooded area gets
separated as shown in Figure 3(a), top panel window is
water mask, the lower one is pre-processed data and
histogram of the pixel value is seen in Figure 3(b). There
are two peaks can be seen in the histogram in Figure 3(b).
Thus, it can be interpreted that the taller peak shows other
land features, and the smaller peak shows water.

It was found that the pixel value of the water surface was
very close to the value same as under the second smaller
peak of the histogram. The threshold value for water in this
study area varies from -19db to -22db, and using a "band
math" tool in SNAP software water mask was created by
applying equation (1) in Sigma0_VH_db band image
shown in Figure 6(b). This water mask was updated using
Figure 4. Surface water mask (source: global surface
a published global surface water mask (Pekel, J F., et al.
water product)
2016). Further, the water mask band was used as ground
truth data for RFC and KNN supervised classification.
3.2 Methodology
The main objective of this study is to develop an
Water mask had labeled data of flooded area as 255 and
automated model which will have the potential to provide
non-flooded area as 0 shown in Figure 6 (b). The labeled
flood inundation extent and water depth in near real-time.
information is used as training datasets for the machine
Automated flood mapping and water depth estimation was
learning models. N_estimator (RFC parameter) was set to
done in two stages. In the first stage, flood inundation
a value of 100 whereas six neighbors were selected in the
extent was estimated using a hybrid model, which was
KNN algorithm. Models estimated inundations are shown
developed using Machine learning-based supervised
in Figure 8.
classifiers (mainly RFC and KNN). In the second stage,
flood extents maps were used along with DEM to estimate
floodwater depth maps using the PyQGIS tool based on
FwDET (Cohen et al. 2018). The methods used in this
study are shown in Figure 5 as a flowchart; further
description of the method is discussed below.

Figure 6. (a) Water Mask (b) Sigma0_VH_db band

image

3.2.2 Floodwater Depth Analysis Model

Floodwater depth was estimated with the help of an
inundation map and DEM using PyQGIS script and
FwDET tool developed by (Cohen et al. 2018). This tool
was applied by taking input from the hybrid flood
inundation model output. Further, to get the elevation
Figure 5. Schematic representation of approach value of pixels in inundation maps was converted into
adopted for flood water depth estimation vector form of polygons. These polygons consist of
flooded areas and non-flooded areas. Flooded area
3.2.1 Hybrid Flood Inundation Model (Classification polygon with grid id of 255 was extracted and merged into
Using Water Mask and Supervised Machine Learning a single polygon. So that, these inundated area polygons
Algorithm) were further converted into polylines, which can serve as
This model was developed to get a fully automated the flood area extent boundary line. The elevation value
approach for flood inundation mapping. The model is for these boundary lines was extracted from DEM and a

224
Journal of Geomatics Vol. 15, No. 2, October 2021

surface interpolated within this boundary line using grow 4. Result and validation
distance tool as in QGIS. The interpolated surface zones.
Flood water depth was found after subtracting the surface 4.1 Hybrid model based inundation
created within flood extent with DEM in raster format, and The random forest model results flood inundation map,
each pixel shows the flood water depth at that location in which shows 5432 Km2 under non-flooded area and 2517
meters. The structure of this depth estimation tool is shown Km2 as flooded area. The KNN model estimated 5892
in Figure 7. Km2 as non-flooded area and 2057 Km2 under flooded
area. The algorithms, KNN and RFC, show nearly the
same flood extent and demonstrated an accuracy of 0.9719
and 0.9726, respectively with ground truth data used in
this study. This classification report shows that both
Classifiers perform well with a very little difference in
performance, but the time taken by KNN classifiers was
28hr, whereas the RFC algorithm takes only 6hr. The
confusion matrix, as shown in Figure 10, was generated to
asses the classification results.

The confusion matrix shows the performance of KNN and

RFC in identifying the flooded and non-flooded area, e.g.,
in Figure 9 model based on the RFC classifier method
Figure 7. Flow chart of Flood Water Depth Analysis identifies 28114541 flooded pixels and 11968 number of
Model pixels confused as a flooded pixel. On the other hand,
57936144 pixels were considered non-flooded pixels and
The concept of floodwater depth estimation is constructed confused in 1990969 pixels as flooded areas. A similar
from the property of water, i.e., as compared with result was found in RFC classifiers.
boundary elevation value, water surface also shows the
same elevation value with its boundary of extent. This A floodwater depth map was generated using the algorithm
concept helps in determining the flood water depth of the for water depth estimation discussed above, and it is shown
inundated areas in a river basin. The classified maps show in Figure 7. Flood extent from the machine learning
flood extent in the study area, and a floodwater depth map approach was used to prepare the Flood depth map shown
was generated using this tool, as shown in Figure 10. in Figure 10. A total area of 2515 Km2 was found
inundated in the flood water depth map using an RFC-
based flood inundation map. While 145 Km2 area of the
inundated part was found overestimated as it shows 0 m
flood water depth and further spatial variation of flooded
water is shown in table 2.

Figure 8. Flood inundation map using RFC (left) and KNN (right) based hybrid model

225
Journal of Geomatics Vol. 15, No. 2, October 2021

4.2 Validation
Further validation of these models results was done with
the flood mapping of Darbhanga district 2017 Tripathi et
al. 2020. SAR data of 11/08/2017, 23/08/2017,
04/09/2017, and 16/09/2017 dates were used to map
inundation using the Binarization technique. The flood
inundation map generated by Tripathi et al. 2020 and
hybrid model based inundation in the present study were
compared for validation. On August 23, 2017, heavy
runoff was calculated using TRMM and IMD rainfall
products found in (Tripathi et al. 2020).

According to a published study, most inundation area was

observed on August 23, 2017. Similarly, model results
Figure 9. Confusion Matrix of Model also show the most flood inundated area on August 23, as
shown in Figure 11. Flooded and the non-flooded area
Table 2. Flood water depth Information (26/07/2020) were also calculated. Published data shows
Depth (m) Area (Km2) underestimated whereas model-derived area shows
0-0.25 m 814.1034 slightly over estimated results. Flood maps estimated
0.25-0.5m 468.1819 using the model can be rectified using a water depth map
0.5-1m 504.8314 derived using floodwater depth tool. Floodwater depth
1 - 2m 386.8695 maps for different dates are shown in Figure 11.
2-3m 126.88
The automated model shows the flooded area as 104.6,
3-5m 60.27397 820, 536 Km2, and 371.4 Km2 on 11/08/2017, 23/08/2017,
5-10m 9.004411 04/09/2017, and 16/09/2017, respectively. When
>10 m 0.098399 compared with previous studies, these flooded areas found
that the flooded area estimated using this model is slightly
overestimated, varying from 14.9% to 70% before
rectification was done.
To calculate the depth of flooded water, DEM was used to
extract the elevation values. Flood water depth
information of the study area is shown in the table. Water
depth was estimated and found that in the study area
during August and September month there was a max
water level of 13 meters, as also shown in table 3. There
were also some pixels showing a 0-meter depth of water,
and these pixels were considered a non-flooded area which
was confused by the model as a flooded area. After
removing 0 m pixels from the flooded area, it was found
that the final Flood Inundation map, along with flood
water depth information, was deviating from 0.9% to
19.33 % respective to the published flood map as provided
in the table 3.

Thus, depth map accuracy depends on the resolution of

DEM. In this study, DEM of 30m spatial resolution was
used, and in any case, when a single pixel of 30 meters
covers an area where some portion is inundated, and the
rest is dry land, then it might happen that the site will be
classified as flooded or non-flooded zone thus it could
affect the accuracy of the model.

From this study, it can be suggested that coupled model

addressing water extent and depth is highly useful to
analyze flood events. Figure 11 shows varying flood water
Figure 10. Flood Depth Map using Flood water depth depth over Darbhanga district in August and September
analysis model 2017.

226
Journal of Geomatics Vol. 15, No. 2, October 2021

Figure 11. Flood water depth map of Darbhanga 2017 flood

Table 3 Floodwater depth of Darbhanga district 2017 flood

Date 0-1m 1-5m 5-10m 10-13m Total Area Total Percentage
depth(Km2) depth(K depth( (Km2) (Km2) Inundated Deviation from
2 2
m) Km ) depth (Hybrid Area in previous
Model) (Tripathi et al. study(Tripathi et
2020) al. 2020)
11/08/2017 20.62 0.051 0 0 21 36 13.88 %
23/08/2017 488.50 70.609 0.626 0.0364 560 554 0.90%
04/09/2017 316.66 41.48 0.388 0.0453 358 330 8.64%
16/09/2017 205.93 18.39 0.023 0 224 188 19.33%

227
Journal of Geomatics Vol. 15, No. 2, October 2021

5. Conclusions Alexandria Engineering Journal, 53(3), 655–662.

https://doi.org/10.1016/j.aej.2014.06.010
In this study, different methods of flood inundation
Feng Q., J . Gong, J. Liu and Y. Li (2015). Flood mapping
mapping were used using active Microwave C band SAR
based on multiple endmember spectral mixture analysis
data. Because the SAR C band having a higher
and random forest classifier-the case of yuyao, China.
wavelength can penetrate dense clouds and even thin
Remote Sensing, 7(9), 12539–12562.
vegetative canopy, thus even in dense cloudy conditions,
the SAR image provides precise information. Machin Gole C. V., and S. V. Chitale (1966). Inland Delta Building
learning classification algorithms, e.g., K- nearest Activity of Kosi River. Journal of the Hydraulics Division,
neighbor and Random Forest classifiers, were used, and it 92(2), 111–126.
was found that the Random Forest classifier gives better
results. Inundated surface estimates were also compared Gouweleeuw B., C. Ticehurst, P. Dyce, J. P. Guerschman,
with published data over the Darbhanga District and found A. I. J. M. Van Dijk and P. Thew (2011). An experimental
satellite-based flood monitoring system for southern
in good agreement. Subsequently, estimated inundation
Queensland, Australia. 34th International Symposium on
along with DEM was used to estimate water depth using
Remote Sensing of Environment - The GEOSS Era:
PyQGIS standalone tool. As mentioned above, all the
Towards Operational Environmental Monitoring.
methods used in the study were packaged to develop an
independent and open-source tool to monitor the flooding Peke, J.F., A. Cottam, N. Gorelick and A. S. Belward
extent and water depth. This tool consists of a hybrid (2016) High-resolution mapping of global surface water
model for inundation extents and a Flood water depth and its long-term changes. Nature 540, 418–422.
analysis model. The present study suggests that the RFC
ML algorithm can be used to classify the SAR data into Mason D. C., G. J. P. Schumann, J. C. Neal, J. Garcia-
flooded and non-flooded areas. The developed tool can be Pintado and P. D. Bates (2012). Automatic near real-time
used for monitoring floods in near real-time for rescue selection of flood water levels from high resolution
operations. The greatest advantage of this tool is that it Synthetic Aperture Radar images for assimilation into
works independently. ML algorithm for classification and hydraulic models: A case study. Remote Sensing of
its automation for flood mapping makes this tool usable Environment, 124, 705–716.
on any platform. The tool could be most helpful for Mason David C., I. J. Davenport, J. C. Neal, G. J. P.
monitoring flood damages and their effect. Schumann and P. D. Bates (2012). Near real-time flood
detection in urban and rural areas using high-resolution
References synthetic aperture radar images. IEEE Transactions on
Anusha N., and B. Bharathi (2020). Flood detection and Geoscience and Remote Sensing, 50(8), 3041–3052.
flood mapping using multi-temporal synthetic aperture
Matgen P., G. Schumann, J. B. Henry, L. Hoffmann and L.
radar and optical data. Egyptian Journal of Remote
Pfister (2007). Integration of SAR-derived river
Sensing and Space Science, 23(2), 207–219.
inundation areas, high-precision topographic data and a
Benoudjit A., and R. Guida. (2019). A novel fully river flow model toward near real-time flood management.
automated mapping of the flood extent on sar images using International Journal of Applied Earth Observation and
a supervised classifier. Remote Sensing, 11(7). Geoinformation, 9(3), 247–263.
Bhatt C. M., G. Srinivasa Rao, P. Manjushree and V. Mohanty, M. P., S. Mudgil and S. Karmakar (2020). Flood
Bhanumurthy (2010). Space based disaster management of management in India: A focussed review on the current
2008 Kosi floods, North Bihar, India. Journal of the Indian status and future challenges. International Journal of
Society of Remote Sensing, 38(1), 99–108. Disaster Risk Reduction, 49(July 2019), 101660.
Campolo M., P. Andreussi and A. Soldati (1999). River Schumann G. J. P., and D. K Moller (2015). Microwave
flood forecasting with a neural network model, 35(4), remote sensing of flood inundation. Physics and Chemistry
1191–1197. of the Earth, 83–84, 84–95.
Cohen S., G. R. Brakenridge, A. Kettner, B. Bates, J. Shahabi H., A. Shirzadi, K. Ghaderi, E.Omidvar, N. Al-
Nelson, R. McDonald, Y. F. Haung, D. Munasinghe and J. Ansari, J. J. Clague, M. Geertsema, K. Khosravi, Ata
Zhang (2018). Estimating Floodwater Depths from Flood Amini, S. Bahrami, O. Rahmati, K. Habibi, A.
Inundation Maps and Topography. Journal of the Mohammadi, H. Nguyen, A. M. Melesse, B. B. Ahmad
American Water Resources Association, 54(4), 847–858. and A. Ahmad. (2020). Flood detection and susceptibility
mapping using Sentinel-1 remote sensing data and a
Amitrano D, G. D. Martino, A. Iodice, D. Riccio and G.
machine learning approach: Hybrid intelligence of
Ruello (2018). Unsupervised rapid flood mapping using
bagging ensemble based on K-Nearest Neighbor classifier.
Sentinel-1 GRD SAR images. IEEE Trans Geosci Remote
Remote Sensing, 12(2).
Sens, 56(6), 3290–3299.
Shen X., D. Wang, K. Mao, E. Anagnostou and Y. Hong
de Groeve, T. (2010). Flood monitoring and mapping
(2019). Inundation extent mapping by synthetic aperture
using passive microwave remote sensing in Namibia.
radar: A review. Remote Sensing, 11(7), 1–17.
Geomatics, Natural Hazards and Risk, 1(1), 19–35.
Singh S. K., A.C. Pandey and M.S. Nathawat (2011).
Elsafi S. H. (2014). Artificial Neural Networks ( ANNs )
Rainfall variability and spatio temporal dynamics of flood
for flood forecasting at Dongola Station in the River Nile ,
inundation during the 2008 Kosi flood in Bihar State,

228
Journal of Geomatics Vol. 15, No. 2, October 2021

India. Asian Journal of Earth Sciences, 4(1), 9–19. Journal of Hydrology, 512, 332–343.
Singh D.A. (2015). Floods in India: Cause and Control Tehrany M. S., B. Pradhan, S. Mansor, and N. Ahmad
(2015). Flood susceptibility assessment using GIS-based
Sinh R., G.V. Bapalu, L. K. Singh and B. Rath, B. (2008).
support vector machine model with different kernel types.
Flood risk analysis in the Kosi river basin, north Bihar
Catena, 125, 91–101.
using multi-parametric approach of Analytical Hierarchy
Process (AHP). Journal of the Indian Society of Remote Temimi M., R. Leconte, F. Brissette and N. Chaouch
Sensing, 36(4), 335–349 (2005). Flood monitoring over the Mackenzie River Basin
using passive microwave data. Remote Sensing of
Sivanpillai R., K. M. Jacobs, C. M. Mattilio and E.V.
Environment, 98(2–3), 344–355.
Piskorski (2020). Rapid flood inundation mapping by
differencing water indices from pre- and post-flood Tripathi G., A. C. Pandey, B. R. Parida, and A. Kumar
Landsat images. Frontiers of Earth Science 2020 15:1, (2020). Flood Inundation Mapping and Impact
15(1), 1–11. Assessment Using Multi-Temporal Optical and SAR
Satellite Data: a Case Study of 2017 Flood in Darbhanga
Tavus B., S. Kocaman, H. Nefeslioğlu and C.
District, Bihar, India. Water Resources Management,
GÖKÇEOĞLU (2019). Flood Mapping Using Sentinel-1
34(6), 1871–1892.
SAR Data: A Case Study of Ordu August 8 2018 Flood.
International Journal of Environment and Geoinformatics, Twele A., W. Cao, S. Plank, and S. Martinis (2016).
6(3), 333–337. Sentinel-1-based flood mapping: a fully automated
processing chain. International Journal of Remote Sensing,
Tehrany M. S., B. Pradhan and M. N. Jebur (2014). Flood
37(13), 2990–3004.
susceptibility mapping using a novel ensemble weights-of-
evidence and support vector machine models in GIS.

229