The document summarizes a study that used Sentinel-1 synthetic aperture radar (SAR) images to automatically detect flooded areas in Aqqala, Iran following heavy rainfall and flooding in March-April 2019. The study applied the Otsu thresholding algorithm on eight Sentinel-1 images to determine an optimal threshold value for separating flooded pixels from other land covers. Thresholding the images using this value delineated the flooded areas over time. Validation against high-resolution images found the flooded areas were accurately detected with overall accuracies over 90%, confirming the applicability of the automatic Otsu thresholding method for flood mapping.
2. 1 3
models provide timely flood monitoring; however,
they need accurate inputs (such as discharge and
weather data). This limits their applications because
weather observations and discharge data are rarely
available over the impacted region. Remote sensing
images have been extensively used in flood-related
studies. Space-borne sensors provide images from
the Earth’s surface with adequate spatial and tem-
poral resolutions for environmental studies (Roy
et al., 2017). Flood maps extracted from satellite
images are one of the main components in flood haz-
ard assessment. Satellite images provide near-real-
time flood monitoring. The use of satellite images in
flood studies is normally limited to passive sensors
operating in the optical and microwave range of the
electromagnetic spectrum (Giustarini et al., 2015).
Optical sensors can be affected by the presence of
clouds; therefore, the use of optical sensors is lim-
ited in cloudy and rainy weather (Fu et al., 2020).
In case a flood is caused by heavy rainfall, the per-
sistent cloud cover makes it impossible to acquire
cloud-free satellite images during the flood event.
Passive microwave sensors operate at longer wave-
lengths which are not affected by weather condi-
tion. They provide images with a short revisit time
(i.e., once or twice a day); however, their coarse
spatial resolution bounds their application in flood
mapping, especially on a regional and local scale
(Oliveira et al., 2019). Synthetic Aperture Radar
(SAR) sensors overcome these two main limitations;
first, they can penetrate clouds; consequently, they
acquire images in all weather conditions, independ-
ent from solar radiation at day and night. Second,
the complicated processing technology in SAR sys-
tems allows high spatial resolution regardless of the
high altitude of the satellite. SAR signals are sensi-
tive to the geometrical structure, surface roughness,
and moisture content of the target (Maître, 2013).
SAR sensors can be used in flood mapping (Rah-
man & Thakur, 2018). However, their applications
have been confined mainly due to the complicated
processing of SAR images and limited accessibil-
ity of these images. The increased access to SAR
images and the development of commercial software
encourage more studies based on SAR images. The
emergence of the Sentinel satellite series and free
access to their images accelerate the SAR images’
application in Earth science. In the past few years,
radar images have been frequently used in mapping
and monitoring hydrological parameters (Voigt
et al., 2009; Anusha et al., 2020). SAR sensors have
capability to acquire images in all atmospheric con-
ditions. This characteristic turns them into a reliable
source in flood-stricken area mapping (Haruyama
& Shida, 2008; Mason et al., 2014). Radar images
acquired before, during, and after a flood event
enable users to monitor the event and consequently
estimate the damage (Ety et al., 2020; Rahman,
2006). SAR-based flood mapping generally benefit
from the joint use of thresholding and classification
algorithms. The flooded area’s dominant scattering
mechanism is surface scattering; therefore, these
regions are depicted in dark tones in SAR images.
Different scattering mechanisms make it possible to
delineate flooded areas from other land covers. The
threshold value could be computed based on either
scene’s global histogram (Landuyt et al., 2018; Lu
et al., 2014; Manjusree et al., 2012) or a non-linear
fitting algorithm (Martinis et al., 2009; H. Cao
et al., 2019; Liang & Liu, 2020). Various classifica-
tion methods have been used; object based (Aldous
et al., 2020; Martinis & Twele, 2010; Mason
et al., 2012), texture based (Dasgupta et al., 2018;
Elfadaly et al., 2020; Ouled Sghaier et al., 2018),
region growing based (Giustarini et al., 2012;
Matgen et al., 2011), fuzzy logic based (Grimaldi
et al., 2020; Pulvirenti et al., 2013). A more com-
prehensive literature review of SAR-based flood
mapping can be found in Martinins (2010). This
study focuses on automatic thresholding algorithms
because it is computationally simple and more accu-
rate compared to some of the mentioned approaches
(Liang & Liu, 2020). Regarding the acquisition
limitation of optical images in the cloudy sky, only
Sentinel-1 images have been considered for flood
mapping (Twele et al., 2016). Sentinel-1 operates
at C-band and has started its mission in 2014. Since
then, it provides free SAR images with high temporal
revisit time to the users’ community. Its images have
been widely used in flood-related studies (Anusha &
Bharathi, 2019; Boni et al., 2016; Chini et al., 2019;
Liang & Liu, 2020). Although several research
works have utilized the Otsu thresholding algorithm
in flood mapping, such efforts are rarely investi-
gated in Iran. This has motivated the application
of an automatic thresholding method to Sentinel-
1time series images in the delineation of flooded
areas in Iran.
Environ Monit Assess (2021) 193: 248
Page 2 of 17
248
3. 1 3
The north part of Iran received heavy rainfall
in March and April 2019 and more than 10 cit-
ies experienced flooding. Aqqala city and neigh-
boring villages were the worst affected areas.
The steady and heavy rain in this arid region
caused heavy damage. The main objective of this
research is to detect the extent of the flooded
area of this particular flood event based on the
analysis of a series of Sentinel-1 images acquired
before, during, and after the flooding. An optimal
threshold value has been calculated to delineate
water-covered areas from the remaining land. This
threshold has been automatically computed using
Otsu thresholding algorithms in Google Earth
Engine. Google Earth Engine is the main plat-
form used to process the Sentinel images in this
research. Google Earth Engine offers an exciting
tool for flood mapping because it saves time and
accelerates image processing. The findings of this
research provide insights into the exploitation of
SAR images in flood mapping based on an auto-
matic thresholding method. The effectiveness
of the implemented method has been validated
against the high-resolution satellite images.
Materials and methods
Study area and flood event
Aqqala is located in the northern part of Golestan
province in Iran, near the border with Turkmeni-
stan. Aqqala county covers approximately 1840
km2
and is located at the latitude 36° 54′ 53″ to
37° 27′ 13″ N and the longitude 54° 13′ 5″ to 54°
Fig. 1 Study area
Environ Monit Assess (2021) 193: 248 Page 3 of 17 248
4. 1 3
51′ 27″ E (Fig. 1). The elevation of the study area
ranges from 47 m below mean sea level (BMSL)
to 81 m above mean sea level (AMSL) (Fig. 2) and
the slope varies between 0 and 57 degrees (Fig. 3).
Aqqala county’s population stands at approxi-
mately 132,000 in 2017 (Plan and Budget Organi-
zation of Iran 2017).
Data
In this study, Sentinel-1 images covering the
study area in descending orbit have been obtained.
Level-1 Ground Range Detected (GRD) prod-
ucts have been acquired. GRD products are
multi-looked and projected from slant range to
Fig. 2 Elevation map
Fig. 3 Slope map
Environ Monit Assess (2021) 193: 248
Page 4 of 17
248
5. 1 3
the ground range by the European Space Agency.
These images consist of square pixels with a mini-
mized speckle effect. The image’s swath width is
equal to 250 km. In total, eight multi-date scenes
have been selected to make the flood monitor-
ing possible: one pre-flood scene, six during
flood scenes, and one post-flood scene (Table 1).
In March 2019, the heavy rainfall caused flood-
ing in Aqqala county. Aqqala county was flooded
from 17 March 2019 to 25 May 2019. In this study,
only backscattering values in VV polarization have
been examined for the automatic extraction of
flood-affected areas. Anusha and Bharathi (2019)
and Liang and Liu (2020) reported VV polariza-
tion has the potential in flood mapping because the
co-polarized VV band has stronger backscattering
intensities in comparison to the cross-polarization
VH band. Image processing procedure and the
Otsu thresholding algorithm have been performed
in the Google Earth Engine platform. The GEE
codes for processing the Sentinel images and
Otsu thresholding procedure are provided in the
“Code availability” section. The average monthly
precipitation (2019) and average annual precipi-
tation are presented in Figs. 4 and 5. This region
received 176.6 mm precipitation in March 2019.
This amount of precipitation was 117.9 mm and
120.14 mm higher than the 10-year and 20-year
average precipitation in March in Aqqala county,
respectively. More than 75% of monthly rainfall
was received in 5 days (17–21 March 2019). From
2010 to 2019, the maximum annual precipitation
occurred in 2019 (i.e., 666.3 mm), 236.3 mm and
256.3 mm higher than the 10-year and 20-year
average annual precipitation in Aqqala county (Iran
Meteorological Organization). Shuttle Radar Topo-
graphic Mission (SRTM) digital elevation model
(DEM) was also used for images terrain correction.
The extent and distribution of detected flooded
areas have been investigated with respect to eleva-
tion, slope, and depression data.
Methodology
This section describes the general workflow of
the proposed method for mapping flooded areas
(Fig. 6). In the first step, eight Sentinel-1 GRD
images are acquired. Then, the pre-processing
steps were performed. This includes (1) apply
orbit file, (2) thermal noise removal, (3) radio-
metric calibration, (4) speckle filtering, (5) terrain
correction, and (6) conversion to dB. In the next
step, the Otsu automatic thresholding method was
applied to each image to find the optimum thresh-
old. Based on the computed threshold, each image
Fig. 4 Average monthly precipitation (2019)
Table 1 Image dataset
Satellite/sensor Image captured Mode Processing level Acquisition date Polarization
(single/dual)
Swath width Spatial resolution
Sentinel-1A Pre- flood IW Level-1 GRD 2019.03.11 Single-VV 250 km 10 × 10 m
Sentinel-1A During the flood IW Level-1 GRD 2019.03.23 Single-VV 250 km 10 × 10 m
Sentinel-1A During the flood IW Level-1 GRD 2019.04.04 Single-VV 250 km 10 × 10 m
Sentinel-1A During the flood IW Level-1 GRD 2019.04.16 Single-VV 250 km 10 × 10 m
Sentinel-1A During the flood IW Level-1 GRD 2019.04.28 Single-VV 250 km 10 × 10 m
Sentinel-1A During the flood IW Level-1 GRD 2019.05.10 Single-VV 250 km 10 × 10 m
Sentinel-1A During the flood IW Level-1 GRD 2019.05.22 Single-VV 250 km 10 × 10 m
Sentinel-1A Post -flood IW Level-1 GRD 2019.06.03 Single-VV 250 km 10 × 10 m
Environ Monit Assess (2021) 193: 248 Page 5 of 17 248
6. 1 3
was segmented into flooded and non-flooded
areas. Independent validation pixels were selected
from high-resolution Google Earth images. These
validation datasets were used to assess the accu-
racy of the extracted flooded areas. All these steps
were coded and run in the Google Earth Engine
platform.
Sentinel‑1 images pre‑processing
Sentinel-1 images have been pre-processed in Google
Earth Engine (Cloud Platform) based on the Sentinel-1
toolbox. First, the satellite orbital correction was per-
formed. This operation precisely adjusts the satellite
orbital parameters. Accurate satellite position and its
velocity are needed for this step (Filipponi, 2019). Satel-
lites precise orbit are calculated after a few days and are
available days-to-weeks after the generation of the prod-
uct in the European Space Agency webpage (https://qc.
sentinel1.eo.esa.int/aux_poeorb) (Elfadaly et al., 2020).
Thermal noise emerges randomly over the image and
causes difficulties in SAR image perception. Thermal
noise removal is more essential in a cross-polarized
channel compared to the co-polarized channel, because
cross-polarized channels have lower backscattered
power in comparison to the co-polarized channels. The
thermal noise removal was done by calculating a noise
look-up table which is available with Sentinel-1 level 1
image (Park et al., 2017). The European Space Agency
provides thermal noise information for each scene,
which are included in a Sentinel-1 SAFE format as an
independent XML file (Park et al., 2017). In the next
step, radiometric calibration was done, and the back-
scattering coefficients have been computed (Elfadaly
et al., 2020). Radiometric calibration is not necessary
for quantitative interpretation; however, it is a manda-
tory step in multi-temporal approaches (Hajduch, 2018).
SAR images are affected by speckle that gives a salt and
pepper appearance to the images. Speckles affect the
radiometric resolution and cause difficulties in images’
interpretation and classification. Multi-looking and spa-
tial filtering are extensively used to minimize the speckle
Fig. 5 Average annual
precipitation (2010–2019)
Fig. 6 Flowchart of the proposed method
Environ Monit Assess (2021) 193: 248
Page 6 of 17
248
7. 1 3
effect. GRD images have already been multi-looked.
Therefore, in this paper, the Lee filter has been applied
to further reduce the speckle. Lee filter is one of the
most used filters in speckle suppression of SAR images
(Lee & Pottier, 2017). Many studies reported Lee fil-
ter’s efficiency in various applications (Y. Brombacher
et al., 2020; Li et al., 2018; Salameh et al., 2020; Slagter
et al., 2020; Zeng et al., 2020). Then, the terrain correc-
tion was done using SRTM to simulate SAR images. For
that, the image and DEM were co-registered (WRAP
function) to minimize the image twists. Finally, the new
value for each pixel in its new position was interpolated
by a bi-linear interpolation method. In the last step of
pre-processing, the digital numbers were converted into
the backscattering coefficients in dB (Eq. 1) (Richards,
2009):
Otsu automatic thresholding method
Image segmentation is one of the basic techniques in
image processing, comprehension, and description.
Among all segmentation techniques, the threshold-
ing segmentation method is one of the most popu-
lar algorithms widely used in image segmentation
(Kahaki et al., 2020). The basic idea of thresholding
is to select an optimal grey-level value for separating
objects of interest from the background in an image.
The single threshold value is derived from an image
histogram. It operates well, in case the histogram is
bimodal and has a deep and sharp valley between
two peaks representing the object and background,
respectively, so that the threshold value is chosen at
the bottom of the valley (Vijay & Patil, 2016). Image
thresholding segments it into two classes: 1 and 0.
One corresponds to the object in interest and 0 corre-
sponds to the background. The gray scale image will
be converted into a binary image. In many studies,
the threshold value between flooded and non-flooded
areas has been determined by the trial-and-error
method (Rahman & Thakur, 2018). The trial-and-
error procedure is subjective and time-consuming
(Tong et al., 2018). However, the automatic methods
overcome these shortcomings and improve the speed
and accuracy of the delineation process. There are
many automatic binary thresholding techniques in
image segmentation and pattern recognition (Sezgin
(1)
𝜎◦
dB = 10logabs(DN)
& Sankur, 2004; Wunnava et al., 2020). Among
them, Otsu is one of the most practical thresholding
methods. Otsu has proved to be an effective approach
in flood delineation in different types of satellite
images, especially SAR images (Du et al., 2014; N.
Li et al., 2014; Pan et al., 2020; Zhang et al., 2020).
In this study, the Otsu automatic algorithm has been
used to separate the flooded and non-flooded areas
in a series of Sentinel-1 images via the Google Earth
Engine. Otsu automatic thresholding is an iterative
method that finds the optimum threshold by exam-
ining all possible values. It maximizes the between-
class variance of the two segments and minimizes the
within-class variance (Otsu, 1979). The pixel values
will be normalized into [a, b] where −1 ≤ a < b ≤ 1,
and the pixel can be divided into two classes of C1
with the range of [t, a] and C2 with the range of [b,
t], where t is the threshold value. Optimal threshold-
ing parameters are shown in Eq. 5. Optimal threshold
(t) is characterized by the inter-class variance of C1
(e.g., non-flooded area) and C2 (e.g., flooded area)
using Eqs. 2–4:
where σ2
is the inter-class variance of C1 and C2, M
is the average value of the indexed image, Pc1 and Pc2
are the placement possibilities of a pixel in C1 and C2
classes, and Mc1 and Mc2 are the average values of C1
and C2 pixels.
The calculated threshold value was applied to each
scene. The image was segmented into flooded (water)
and non-flooded (non-water) areas via the Google
Earth Engine platform.
Analysis of topographic characteristics
SRTM (Survey, 2015) was used to calculate the
slope and surface depression. DEM is digitally
filled to calculate the surface depression (Brychta
et al., 2020; Safanelli et al., 2020). For that,
(2)
𝜎2
= Pc1 ×
(
Mc1 − M
)2
+ Pc2 ×
(
Mc2 − M
)2
(3)
M = Pc1 × Mc1 × Pc2 × Mc2
(4)
Pc1 + Pc2 = 1
(5)
t ∗= ArgMaxa<t<b
{
Pc1 × (Mc1 − M)2
+ Pc2 × (Mc2 − M)2
}
Environ Monit Assess (2021) 193: 248 Page 7 of 17 248
8. 1 3
neighboring elevations located outside the depres-
sion polygons are interpolated. The difference
between original and filled DEM is considered to be
surface depression (Branton & Robinson, 2020):
Accuracy assessment
The extracted flooded and non-flooded areas were
validated against independent validation datasets.
The validation datasets were selected from high-
resolution satellite images available in Google Earth.
The validation points should be well-distributed over
the entire image. Therefore, many efforts have been
done to select the validation pixels from all over
the scene. Approximately, a total of 400 validation
points have been selected on each image by a simple
random sampling method. Then, confusion matri-
ces have been built for each image. The confusion
matrix is frequently used to describe the performance
of binary or multi-class classification (Olofsson
et al., 2014). In the confusion matrix, the number of
rows and columns is equal to the number of classes
(here 2). Kappa coefficient, overall, producer, and
user accuracies have been calculated based on the
confusion matrix. The overall accuracy shows the
percentage of validation pixels classified correctly
(Morales-Barquero et al., 2019). Kappa coefficient
is the ratio of agreement between the classified
image and reference data (Foody, 2020). Producer
and user accuracies represent the accuracies of each
class (Stehman, 2009). In this case, producer accu-
racy of water class shows the percentage of water
pixels in the output image that is classified correctly.
User accuracy of this class shows the percentage of
water pixels on the output image is actually covered
by water (Stehman & Foody, 2019).
Results and discussion
Figure 5 shows the pre-flood, during the flood, and post-
flood images of the study area. Pixels covered by water
have low backscattering values and depicted as dark
areas in the image. Bare land, farmland, and built-up
areas have medium to high backscattering values, shown
in moderate gray to white (Fig. 7). On 23 March 2019
(when the event started), the dark pixels were increased
Surface depression = Filled DEM − Original DEM
compared to 11th March. Then, the water level began to
decline. This continued until 28 April 2019, when the
water came up again (Fig. 8). Permanent water surfaces
(such as dams) in the northern part of the study area
constantly have low backscattering values, even in the
pre-flood image. Their backscattering values are similar
to the flooded pixels. These areas were masked out to
minimize the challenges caused by the backscattering
similarities of these areas and flooded areas.
The Otsu method has been applied to each image
separately to find the optimum threshold value to
separate flooded areas from non-flooded areas. The
flooded areas have been shown in Fig. 9. The results
show that there were not many water surfaces except
dams in the study area. The flood started on 23 March
and covered the surrounding lands. The inner parts
of the city were also flooded because the Gorganrood
river passes through the city. Some days later, the
water level of surrounding areas started to decline.
However, the inundations entered the agricultural
lands and covered a major part of them. On 28 April,
the flood level increased once again. Special reporting
committee on Iran floods 2019 outlined that the latter
flood was caused by melting snow in high lands. The
temperature rise accelerated snowmelt. The snowmelt
water entered the dam and excess water overflowed
onto the land. Then, the floodwater started to recede,
and the flood ended on May 22. Almost equal thresh-
old values (i.e., −14.90) have been calculated in all
scenes.
All available Sentinel-1 images acquired during the
flood period have been used. This provided more accu-
rate identification of flood damage in the mentioned
period. Descriptive statistics of images (minimum,
maximum, mean, variance, and coefficient of varia-
tion) have been presented in Table 2. The least and the
most diverse backscattering values belong to the pre-
flood image (CV
=
0.21) and the during-flood (second)
image (CV
=
0.38), respectively. This showed that the
backscattering values are near the mean value in the
pre-flood image. However, backscattering values are
spread out over the wider range in flood images. The
post-flood image has a wider range of backscatter-
ing values in comparison to the pre-flood image. Otsu
algorithm is highly capable of distinguishing the edge
between land and water surfaces. Surprisingly, the com-
puted thresholds did not differ too much, and the maxi-
mum difference among computed threshold values was
only 0.07 dB. Therefore, −14.9 dB was applied as the
Environ Monit Assess (2021) 193: 248
Page 8 of 17
248
9. 1 3
optimum threshold, i.e., pixels with values higher than
−14.9 dB will be considered as non-flooded areas, and
accordingly, flooded areas have backscattering values
lower than −14.9 dB in the study area.
According to Fig. 10, in the pre-flood image
(11 March 2019), 30
km2
of the region was cov-
ered by water. The area has increased to 162.5
km2
on 23 March 2019. On 4 April 2019, 11 days later,
the flood-affected area increased to 236
km2
. The
Fig. 7 Multi-temporal Sentinel-1 images
Fig. 8 Profile of the SAR
backscatter (dB) variations
over the flood event
Environ Monit Assess (2021) 193: 248 Page 9 of 17 248
10. 1 3
maximum flood extent was observed on 16 April
2019 (i.e., 250
km2
). Then, the water level started to
decline, and on 28 April 2019, the flood covered the
area of 228.5
km2
. This decreasing trend continued.
The flood covered 223.4 and 78.5
km2
of the region
on 10 May 2019 and on 22 May 2019, respectively.
Eventually, on 3 June 2019, the flood-affected area
declined to 76.8
km2
.
The flood map was superimposed with DEM
and the slop map to analyze the distribution of the
flooded area. An analysis of the superimposed image
indicates 35.83% of the flooded areas are dispersed
at an elevation of 24 to 14 m BMSL and 42.06% at
an elevation of 14 m BMSL to the mean sea level
(0 m) (Fig. 11). The overlay of flooded areas with
the slope layer showed that 82.14% of flooded areas
Fig. 9 Detected flooded area overlaid on Sentinel-1 images
Table 2 Statistic summary
of Sentinel-1 images
Threshold (dB) CV Variance Mean Max Min Date ID
−14.89 0.21 8.29 −13.25 18.16 −48.07 2019 Mar 11 1
−14.89 0.38 19.95 −11.68 29.80 −51.94 2019 Mar 23 2
−14.91 0.31 16.72 −12.92 26.08 −58.46 2019 Apr 04 3
−14.91 0.27 13.01 −12.99 23.37 −50.73 2019 Apr 16 4
−14.88 0.31 16.52 −13.04 24.14 −51.63 2019 Apr 28 5
−14.87 0.24 11.14 −13.59 15.26 −47.73 2019 May 10 6
−14.93 0.28 13.52 −13.13 26.26 −50.07 2019 May 22 7
−14.94 0.29 16.24 −13.68 23.67 −54.60 2019 Jun 03 8
Environ Monit Assess (2021) 193: 248
Page 10 of 17
248
11. 1 3
are located at a slope of 0 to 3 degrees (Fig. 12). The
depressions are dispersed over the Aqqala county
(Fig. 13). The superimposed image of the flood map
with the depression layer presents that 62.17% of
flooded areas are located at depressions.
Validation
The images have been divided into the flooded
and non-flooded classes by applying the computed
threshold. It is important to verify how well the
calculated threshold delineates the flooded area.
To achieve this goal, the national flood report has
been carefully reviewed, and validation pixels
have been selected on the high-resolution Google
Earth images. A total of 400 validation pixels
have been selected by a simple random sampling
method. The confusion matrices have been com-
puted based on the comparison of binary images
and the independent validation datasets. Based on
the confusion matrices, kappa coefficient, over-
all accuracy, user accuracy, and producer accu-
racy have been computed. The comparison results
revealed that the calculated thresholds were effi-
cient in flooded area delineation (Table 3). The
best result belongs to the image dated 23 March
2019 with an overall accuracy of 96.2 and a kappa
coefficient of 95.4. Otsu performs quite well in
differentiating flooded and non-flooded areas in
pre-flood, during-flood, and post-flood images. It
was reported that the Otsu thresholding algorithm
is one of the most efficient among automatic
thresholding approaches (Liang & Liu, 2020;
Zeng et al., 2020).
Fig. 10 Changes of the
flood extent
Fig. 11 Percentage of flooded areas at different elevations Fig. 12 Percentage of flooded areas at different slopes
Environ Monit Assess (2021) 193: 248 Page 11 of 17 248
12. 1 3
Discussion
In this research, multi-temporal Sentinel-1 images
have been exploited to monitor the severe and
dynamic flood event in the north of Iran, March 2019.
The flooded area was detected by the Otsu threshold-
ing algorithm using the GEE platform. Few studies
have been focused on automatic algorithms in SAR-
based flood mapping; however, optical/passive micro-
wave sensors and the joint use of them have been
widely explored. Sentinel-1 provides valuable high
spatial resolution images with short revisit time,
which makes it suitable for flood mapping.
Fig. 13 Depression map
Table 3 Accuracies of the
threshold images
Date Class Producer accu-
racy (%)
User accuracy
(%)
Overall
accuracy
Kappa
coeffi-
cient
2019 Mar 11 Flooded 90.1 91.2 92.8 91.1
Non-flooded 89.2 89.9
2019 Mar 23 Flooded 95.2 96.4 96.2 95.4
Non-flooded 94.1 94.8
2019 Apr 04 Flooded 94.2 94.9 95.7 95
Non-flooded 93.2 93.8
2019.Apr.16 Flooded 93.6 92.8 93.7 91.9
Non-flooded 92.1 92.5
2019 Apr 28 Flooded 94.3 95 95.1 94.2
Non-flooded 94.1 93.7
2019 May 10 Flooded 93 93.1 93.3 92.4
Non-flooded 92.9 92.2
2019 May 22 Flooded 93.9 93.4 94.2 93.1
Non-flooded 93.1 92.5
2019 Jun 03 Flooded 94.1 93.6 94.8 93.7
Non-flooded 93.5 93.8
Environ Monit Assess (2021) 193: 248
Page 12 of 17
248
13. 1 3
High overall accuracies proved that automatic
thresholding algorithms are efficient in the delineation
of flood-affected areas from SAR images. Our results
are in line with the results of previously conducted
researches (Liang & Liu, 2020; Zeng et al., 2020).
Zeng et al., (2020) concluded that the threshold-based
method based on Otsu’s algorithm performs well with
a promising overall accuracy of 89.83%. Liang & Liu,
(2020) compared four different thresholding algorithms
for flood detection, and their results showed that the
Otsu thresholding algorithm was one of the best thresh-
olding algorithms with an overall accuracy of 98.12%.
The Otsu algorithm perfectly distinguished the
flooded area from the non-affected built-up area
Fig. 14 Water surfaces detected by the Otsu algorithm in four subsets of Sentinel-1 images (Subsets are shown in Fig. 1)
Environ Monit Assess (2021) 193: 248 Page 13 of 17 248
14. 1 3
(Fig. 14), mainly due to the high backscattering dif-
ferences between the water and built-up area. Water’s
dominant scattering mechanism is surface scattering
resulted in low backscattering values; however, built-
up areas have high backscattering values because of
double-bounce scattering. The algorithm also per-
forms well in the detection of water surfaces such
as the lake (Fig. 14). The extent of the flooded area
has been underestimated in the narrow river chan-
nel (Fig. 14). Sentinel-1 has 10-m spatial resolution,
which limits its application in differentiation between
the flooded and non-flooded areas in the narrow part
of the channel. Speckle filtering mostly results in
the spatial resolution’s degradation; this negatively
affects the flooded area’s detection. The algorithm
also fails to detect all flooded areas in farmlands. This
could be explained by the backscattering value varia-
tion in this particular land cover class.
Conclusions
Flood is one of the most devastating natural hazards
that cause massive economic and human loss all
around the world. A flood occurs as overflow water
as a result of heavy rain, rapid melting of snow, and
dam failure. Although optical images served as an
efficient tool to monitor processes on the Earth, they
have limitations for monitoring floods caused by long
rainfall. Optical wavelengths cannot penetrate the
clouds. However, clouds seem transparent in long
wavelength in the microwave part of the spectrum.
SAR images contain unique information in cloud-
prone areas. In this paper, the usage of Sentinel-1
images has been considered to monitor the severe
flood event in the north of Iran during spring 2019.
The images have been segmented into the flooded and
non-flooded areas by applying the specific thresh-
old value determined by the Otsu algorithm. The
reported accuracies were high, indicating the effi-
ciency of the applied method. Although this flood
event was highly dynamic and a series of pre-flood,
during flood, and post-flood imageries have been
used in this study, a unique threshold value has been
applied. This will simplify flooded area mapping. An
automatic thresholding method has also accelerated
the process. This is of great value because accurate
and on-time flood mapping is crucial. The presented
threshold shall be further examined in other flood
events for generalization purposes. The exploitation
of Sentinel-1 images highlights the application of the
presented research. Sentinel-1 full archive is freely
available from its mission start, and its spatial and
temporal resolution makes it suitable for timely flood
mapping. The research’s findings demonstrate that
SAR images can effectively be used in flood inun-
dation mapping. This is of critical importance in the
calibration and validation of flood inundation models.
This will help the authorities to make proper deci-
sions in disaster time. It is recommended that classi-
fication methods such as machine learning and arti-
ficial intelligence techniques be considered for flood
mapping in future studies. The contribution of polari-
metric decomposition techniques in SAR-based flood
mapping should be investigated.
Acknowledgements The authors acknowledge the Google
Earth Engine for providing Sentinel-1 images and computation
capabilities.
Data availability Data are available upon reasonable request.
Code availability The code is available online at https://code.
earthengine.google.com/6dcd097df2c9e9b821858e5046a8b21d
Declarations
Conflict of interest The authors declare no competing inter-
est.
References
Aldous, A., Schill, S., Raber, G., Paiz, M. C., Mambela, E.,
Stévart, T. J. R. S. i. E., et al. (2020). Mapping complex
coastal wetland mosaics in Gabon for informed ecosystem
management: use of object‐based classification.
Anusha, N., & Bharathi, B. (2019). Flood detection and flood
mapping using multi-temporal synthetic aperture radar
and optical data. The Egyptian Journal of Remote Sensing
and Space Science.
Anusha, N., Bharathi, B. J. T. E. J. o. R. S., & Science, S.
(2020). Flood detection and flood mapping using multi-
temporal synthetic aperture radar and optical data. 23(2),
207–219.
Atanga, R. A. J. I. j. o. d. r. r. (2020). The role of local com-
munity leaders in flood disaster risk management strategy
making in Accra. 43, 101358.
Barredo, J. I. (2007). Major flood disasters in Europe: 1950–
2005. Natural Hazards, 42(1), 125–148.
Environ Monit Assess (2021) 193: 248
Page 14 of 17
248
15. 1 3
Bioresita, F., Puissant, A., Stumpf, A., & Malet, J.-P. J. R. S.
(2018). A method for automatic and rapid mapping of
water surfaces from sentinel-1 imagery. 10(2), 217.
Boni, G., Ferraris, L., Pulvirenti, L., Squicciarino, G., Pierdicca,
N., Candela, L., et al. (2016). A prototype system for flood
monitoring based on flood forecast combined with COSMO-
SkyMed and Sentinel-1 data. IEEE Journal of Selected Top-
ics in Applied Earth Observations and Remote Sensing, 9(6),
2794–2805.
Bourenane, H., Bouhadad, Y., & Tas, M. (2018). Liquefaction
hazard mapping in the city of Boumerdès, Northern Alge-
ria. Bulletin of Engineering Geology and the Environ-
ment, 77(4), 1473–1489.
Branton, C., & Robinson, D. T. J. W. (2020). Quantifying
topographic characteristics of wetlandscapes., 40(2),
433–449.
Brombacher, J., Reiche, J., Dijksma, R., & Teuling, A. J.
(2020). Near-daily discharge estimation in high latitudes
from Sentinel-1 and 2: a case study for the Icelandic
Þjórsá river. Remote sensing of Environment, 241, 111684.
Brychta, J., Brychtová, M. J. S., & Research, W. (2020). -Pos-
sibilities of including surface runoff barriers in the slope-
length factor calculation in the GIS environment and its
integration in the user-friendly LS-RUSLE tool.
Cao, C., Xu, P., Wang, Y., Chen, J., Zheng, L., & Niu, C.
(2016). Flash flood hazard susceptibility mapping using
frequency ratio and statistical index methods in coalmine
subsidence areas. Sustainability, 8(9), 948.
Cao, H., Zhang, H., Wang, C., & Zhang, B. (2019). Operational
flood detection using Sentinel-1 SAR data over large areas.
Water, 11(4), 786.
Chapi, K., Singh, V. P., Shirzadi, A., Shahabi, H., Bui, D. T., Pham,
B. T., et al. (2017). A novel hybrid artificial intelligence
approach for flood susceptibility assessment., 95, 229–245.
Chini, M., Pelich, R., Pulvirenti, L., Pierdicca, N., Hostache,
R., & Matgen, P. (2019). Sentinel-1 InSAR coherence to
detect floodwater in urban areas: Houston and Hurricane
Harvey as a test case. Remote Sensing, 11(2), 107.
Dasgupta, A., Grimaldi, S., Ramsankaran, R., Pauwels, V. R.,
& Walker, J. P. (2018). Towards operational SAR-based
flood mapping using neuro-fuzzy texture-based approaches.
Remote sensing of Environment, 215, 313–329.
Du, Z., Li, W., Zhou, D., Tian, L., Ling, F., Wang, H., et al.
(2014). Analysis of Landsat-8 OLI imagery for land sur-
face water mapping. Remote sensing letters, 5(7), 672–681.
Elfadaly, A., Abate, N., Masini, N., & Lasaponara, R. J. R. S.
(2020). SAR Sentinel 1 imaging and detection of palaeo-
landscape features in the Mediterranean area. 12(16), 2611.
Ety, N. J., Chu, Z., & Masum, S. M. J. Q. I. (2020). Monitoring
of flood water propagation based on microwave and opti-
cal imagery.
Filipponi, F. Sentinel-1 GRD preprocessing workflow. In Multidis-
ciplinary Digital Publishing Institute Proceedings, 2019 (Vol.
18, pp. 11, Vol. 1)
Foody, G. M. J. R. S. o. E. (2020). Explaining the unsuitability
of the kappa coefficient in the assessment and comparison
of the accuracy of thematic maps obtained by image clas-
sification. 239, 111630.
Fu, W., Ma, J., Chen, P., & Chen, F. (2020). Remote sensing
satellites for digital Earth. In Manual of Digital Earth (pp.
55–123): Springer, Singapore.
Giustarini, L., Chini, M., Hostache, R., Pappenberger, F., & Matgen,
P. J. R. S. (2015). Flood hazard mapping combining hydrody-
namic modeling and multi annual remote sensing data., 7(10),
14200–14226.
Giustarini, L., Hostache, R., Matgen, P., Schumann, G.J.-P.,
Bates, P. D., & Mason, D. C. (2012). A change detec-
tion approach to flood mapping in urban areas using Ter-
raSAR-X. IEEE Transactions on Geoscience and Remote
Sensing, 51(4), 2417–2430.
Grimaldi, S., Xu, J., Li, Y., Pauwels, V. R., & Walker, J. P.
(2020). Flood mapping under vegetation using single
SAR acquisitions. Remote sensing of Environment, 237,
111582.
Hajduch, G. (2018). Masking “no-value” pixels on GRD
products generated by the Sentinel-1 ESA IPF. European
Space Agency Paris.
Haruyama, S., & Shida, K. (2008). Geomorphologic land clas-
sification map of the Mekong Delta utilizing JERS-1 SAR
images. Hydrological Processes: An International Journal,
22(9), 1373–1381.
Kahaki, S. M., Nordin, M. J., Ahmad, N. S., Arzoky, M.,
Ismail, W. J. N. C., & Applications (2020). Deep convo-
lutional neural network designed for age assessment based
on orthopantomography data. 32(13), 9357–9368.
Kundzewicz, Z. W. (2008). Flood risk and vulnerability in
the changing climate. (p. 39). Annals of Warsaw Univer-
sity of Life Sciences-SGGW.
Landuyt, L., Van Wesemael, A., Schumann, G.J.-P., Hostache, R.,
Verhoest, N. E., & Van Coillie, F. M. (2018). Flood mapping
based on synthetic aperture radar: an assessment of estab-
lished approaches. IEEE Transactions on Geoscience and
Remote Sensing, 57(2), 722–739.
Lee, J.-S., & Pottier, E. (2017). Polarimetric radar imaging:
from basics to applications: CRC press.
Li, N., Wang, R., Liu, Y., Du, K., Chen, J., & Deng, Y. (2014).
Robust river boundaries extraction of dammed lakes in
mountain areas after Wenchuan Earthquake from high
resolution SAR images combining local connectivity and
ACM. ISPRS journal of photogrammetry and remote
sensing, 94, 91–101.
Li, Y., Martinis, S., Plank, S., & Ludwig, R. (2018). An auto-
matic change detection approach for rapid flood mapping
in Sentinel-1 SAR data. International Journal of Applied
Earth Observation and Geoinformation, 73, 123–135.
Liang, J., & Liu, D. (2020). A local thresholding approach to
flood water delineation using Sentinel-1 SAR imagery.
ISPRS journal of photogrammetry and remote sensing,
159, 53–62.
Long, S., Fatoyinbo, T. E., & Policelli, F. (2014). Flood extent
mapping for Namibia using change detection and thresh-
olding with SAR. Environmental Research Letters, 9(3),
035002.
Lu, J., Giustarini, L., Xiong, B., Zhao, L., Jiang, Y., & Kuang,
G. (2014). Automated flood detection with improved
robustness and efficiency using multi-temporal SAR data.
Remote sensing letters, 5(3), 240–248.
Maître, H. (2013). Processing of Synthetic Aperture Radar
(SAR) images: John Wiley & Sons.
Manjusree,P.,Kumar,L.P.,Bhatt,C.M.,Rao,G.S.,&Bhanumurthy,
V. (2012). Optimization of threshold ranges for rapid flood
inundation mapping by evaluating backscatter profiles of high
Environ Monit Assess (2021) 193: 248 Page 15 of 17 248
16. 1 3
incidence angle SAR images. International Journal of Disas-
ter Risk Science, 3(2), 113–122.
Marchi, L., Borga, M., Preciso, E., & Gaume, E. (2010).
Characterisation of selected extreme flash floods in
Europe and implications for flood risk management.
Journal of Hydrology, 394(1–2), 118–133.
Martinez, J.-M., & Le Toan, T. (2007). Mapping of flood
dynamics and spatial distribution of vegetation in the
Amazon floodplain using multitemporal SAR data.
Remote sensing of Environment, 108(3), 209–223.
Martinis, S. (2010). Automatic near real-time flood detection
in high resolution X-band synthetic aperture radar satel-
lite data using context-based classification on irregular
graphs. lmu,
Martinis, S., & Twele, A. (2010). A hierarchical spatio-temporal
Markov model for improved flood mapping using multi-
temporal X-band SAR data. Remote Sensing, 2(9), 2240–2258.
Martinis, S., Twele, A., & Voigt, S. (2009). Towards operational
near real-time flood detection using a split-based automatic
thresholding procedure on high resolution TerraSAR-X
data. Natural Hazards & Earth System Sciences, 9(2).
Mason, D. C., Davenport, I. J., Neal, J. C., Schumann, G.J.-
P., & Bates, P. D. (2012). Near real-time flood detection
in urban and rural areas using high-resolution synthetic
aperture radar images. IEEE Transactions on Geoscience
and Remote Sensing, 50(8), 3041–3052.
Mason, D. C., Giustarini, L., Garcia-Pintado, J., & Cloke,
H. L. (2014). Detection of flooded urban areas in high
resolution Synthetic Aperture Radar images using dou-
ble scattering. International Journal of Applied Earth
Observation and Geoinformation, 28, 150–159.
Matgen, P., Hostache, R., Schumann, G., Pfister, L., Hoffmann,
L., & Savenije, H. (2011). Towards an automated SAR-based
flood monitoring system: lessons learned from two case stud-
ies. Physics and Chemistry of the Earth, Parts A/B/C, 36(7–
8), 241–252.
Morales-Barquero, L., Lyons, M. B., Phinn, S. R., & Roelfsema,
C. M. J. R. s. (2019). Trends in remote sensing accuracy
assessment approaches in the context of natural resources.
11(19), 2305.
Oliveira, E. R., Disperati, L., Cenci, L., Gomes Pereira, L.,
& Alves, F. L. J. R. S. (2019). Multi-Index Image Dif-
ferencing Method (MINDED) for Flood Extent Estima-
tions. 11(11), 1305.
Olofsson, P., Foody, G. M., Herold, M., Stehman, S. V.,
Woodcock, C. E., & Wulder, M. A. J. R. S. o. E. (2014).
Good practices for estimating area and assessing accu-
racy of land change. 148, 42–57.
Otsu, N. (1979). A threshold selection method from gray-
level histograms. IEEE Transactions on systems, man,
and cybernetics, 9(1), 62–66.
Ouled Sghaier, M., Hammami, I., Foucher, S., & Lepage, R.
(2018). Flood extent mapping from time-series SAR
images based on texture analysis and data fusion. Remote
Sensing, 10(2), 237.
Pan, F., Xi, X., & Wang, C. J. R. S. (2020). A comparative
study of water indices and image classification algo-
rithms for mapping inland surface water bodies using
Landsat imagery., 12(10), 1611.
Park, J.-W., Korosov, A. A., Babiker, M., Sandven, S., &
Won, J.-S. (2017). Efficient thermal noise removal for
Sentinel-1 TOPSAR cross-polarization channel. IEEE
Transactions on Geoscience and Remote Sensing, 56(3),
1555–1565.
Penning-Rowsell, E., Floyd, P., Ramsbottom, D., & Surendran,
S. (2005). Estimating injury and loss of life in floods: a
deterministic framework. Natural Hazards, 36(1–2),
43–64.
Pulvirenti, L., Pierdicca, N., Chini, M., & Guerriero, L.
(2013). Monitoring flood evolution in vegetated areas
using COSMO-SkyMed data: the Tuscany 2009 case
study. IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing, 6(4), 1807–1816.
Rahman, M. R. (2006). Flood inundation mapping and damage
assessment using multi-temporal RADARSAT and IRS 1C
LISS III Image. Asian Journal of Geoinformatics, 6(2),
11–21.
Rahman, M. R., & Thakur, P. K. (2018). Detecting, mapping
and analysing of flood water propagation using synthetic
aperture radar (SAR) satellite data and GIS: a case study
from the Kendrapara District of Orissa State of India. The
Egyptian Journal of Remote Sensing and Space Science,
21, S37–S41.
Richards, J. A. (2009). Remote sensing with imaging radar
(Vol. 1): Springer.
Roy, P., Behera, M., & Srivastav, S. (2017). Satellite remote
sensing: sensors, applications and techniques. Springer.
Safanelli, J. L., Poppiel, R. R., Ruiz, L. F. C., Bonfatti, B.
R., Mello, F. A. d. O., Rizzo, R., et al. (2020). Terrain
analysis in Google Earth Engine: a method adapted for
high-performance global-scale analysis. 9(6), 400.
Salameh, E., Frappart, F., Turki, I., & Laignel, B. (2020).
Intertidal topography mapping using the waterline
method from Sentinel-1 & -2 images: the examples of
Arcachon and Veys Bays in France. ISPRS journal of
photogrammetry and remote sensing, 163, 98–120.
Sezgin, M., & Sankur, B. (2004). Survey over image thresholding
techniques and quantitative performance evaluation. Journal
of Electronic imaging, 13(1), 146–166.
Slagter, B., Tsendbazar, N.-E., Vollrath, A., & Reiche, J.
(2020). Mapping wetland characteristics using tempo-
rally dense Sentinel-1 and Sentinel-2 data: a case study in
the St. Lucia wetlands, South Africa. International Jour-
nal of Applied Earth Observation and Geoinformation,
86, 102009.
Stehman, S. V. J. I. J. o. R. S. (2009). Sampling designs for
accuracy assessment of land cover. 30(20), 5243–5272.
Stehman, S. V., & Foody, G. M. J. R. S. o. E. (2019). Key
issues in rigorous accuracy assessment of land cover
products. 231, 111199.
Survey, U. J. U. G. (2015). Shuttle radar topography mission
(SRTM) 1 Arc‐Second global.
Tong, X., Luo, X., Liu, S., Xie, H., Chao, W., Liu, S., et al.
(2018). An approach for flood monitoring by the com-
bined use of Landsat 8 optical imagery and COSMO-
SkyMed radar imagery. ISPRS journal of photogramme-
try and remote sensing, 136, 144–153.
Environ Monit Assess (2021) 193: 248
Page 16 of 17
248
17. 1 3
Twele, A., Cao, W., Plank, S., & Martinis, S. (2016). Sentinel-
1-based flood mapping: a fully automated processing
chain. International Journal of Remote Sensing, 37(13),
2990–3004.
Vijay, P. P., & Patil, N. J. J. f. R. (2016). Gray scale image
segmentation using OTSU thresholding optimal
approach. 2(05).
Voigt, S., Martinis, S., Zwenzner, H., Hahmann, T., Twele,
A., & Schneiderhan, T. Extraction of flood masks using
satellite based very high resolution SAR data for flood
management and modeling. In RIMAX Contributions
at the 4th International Symposium on Flood Defence
(ISFD4), 2009: Deutsches GeoForschungsZentrum GFZ
Wunnava, A., Naik, M. K., Panda, R., Jena, B., & Abraham, A.
J. A. S. C. (2020). An adaptive Harris hawks optimization
technique for two dimensional grey gradient based multilevel
image thresholding., 95, 106526.
Yariyan, P., Janizadeh, S., Van Phong, T., Nguyen, H. D., Costache,
R., Van Le, H., et al. (2020). Improvement of best first decision
trees using bagging and dagging ensembles for flood probabil-
ity mapping., 34(9), 3037–3053.
Zeng, Z., Gan, Y., Kettner, A. J., Yang, Q., Zeng, C., Braken-
ridge, G. R., et al. (2020). Towards high resolution flood
monitoring: an integrated methodology using passive
microwave brightness temperatures and Sentinel syn-
thetic aperture radar imagery. Journal of Hydrology, 582,
124377.
Zhang, W., Hu, B., & Brown, G. S. J. W. (2020). Automatic
surface water mapping using polarimetric SAR data for
long-term change detection., 12(3), 872.
Publisher’s Note Springer Nature remains neutral with regard
to jurisdictional claims in published maps and institutional
affiliations.
Environ Monit Assess (2021) 193: 248 Page 17 of 17 248