Flood Mapping in the Coastal Region of Bangladesh Using Sentinel-1 SAR Images: A Case Study of Super Cyclone Amphan

Floods are triggered by water overflow into drylands from several sources, including rivers, lakes, oceans, or heavy rainfall. Near realtime (NRT) flood mapping plays an important role in taking strategic measures to reduce flood damage after a flood event. There are many satellite imagery based remote sensing techniques that are widely used to generate flood maps. Synthetic aperture radar (SAR) images have proven to be more effective in flood mapping due to its high spatial resolution and cloud penetration capacity. This case study is focused on the super cyclone, commonly known as Amphan, stemming from the west Bengal-Bangladesh coast across the Sundarbans on 20 May 2020, with a wind speed between 155kmph-165 kmph gusting up to 185 kmph. The flooding extent is determined by analyzing the pre and post-event synthetic aperture radar images, using the change detection and thresholding (CDAT) method. The kmresults showed an inundated landmass of 2146kmon 22 May 2020, excluding Sundarban. However, the area became 1425kmabout a week after the event, precisely on 28 May 2020km. This persistency generated a more severe and intense flood, due to the broken embankments. Furthermore, 13 out of 19 coastal districts were affected by the flooding, while 8 were highly inundated, including Bagerhat, Pirojpur, Satkhira, Khulna, Barisal, Jhalokati, Patuakhali and Barguna. These findings were subsequently compared with an inundation map created with a validation survey immediately after the event and also with the disposed location using a machine learning-based image classification technique. Consequently, the comparison showed a close similarity between the inundation scenario and the flood reports from the secondary sources. This circumstance envisages the significant role of CDAT application in providing relevant information for an effective decision support system.


INTRODUCTION
Cyclonic storm surge-related inundation appears common in Bangladesh's coastal regions, due to the location within the world's largest tidedominated delta, termed Ganges-Brahmaputra-Meghna (GBM). However, there is a general occurrence mainly during early summer (April-May) and late rainy season (October-November) (Debsarma, 2009, Dasgupta et al., 2010. Bengal bay plays a very significant role in developing these cyclones, with approximately 7% targeting the coast (Ali, 1999). This condition generated a more severe effect, as 53% of the global cyclonerelated deaths reportedly occurred in Bangladesh (Debsarma, 2009). Also, cyclonic storm surge causes disastrous floods due to a shallow continental shelf (Murty, Flather & Henry, 1986), coastal funneling shape (Das,1972) and triangular structure in the northeast corner of the Meghna Estuary (Flierl & Robinson, 1972, Haider et al., 1991. Therefore, the resulting flood intensity greatly expanded during high currents from the tide-surge interactions (Debsarma, 2009).
The super cyclone triggered a landfall over the West Bengal-Bangladesh coast on 20 May 2020, between 10:00-12:00, UTC. This is the first Amphan recorded after the Odisha occurrence in 1999 (RSMC-Tropical Cyclones, 2021). A postevent survey conducted by Area Cyclone Warning Centre (ACWC), Kolkata, reported lowlying areas in the south and north Pargana and the adjoining regions of east Medinipur in west Bengal received 4.6m high storm surge during the flood (RSMC-Tropical Cyclones, 2021). As a consequence, 2.6 million persons in 19 districts (second administrative boundary) were affected, alongside a total of 76 Upazilas (third administrative boundary) with 34 inundated regions. Under these circumstances, 550 km of roads and embankments have been damaged (Needs Assessment Working Group, 2020). Flooding is a widespread phenomenon in lowland countries, including Bangladesh. Therefore, an immediate supply of disaster relief to the affected areas becomes very necessary, including proximate real-time flood maps. These charts are also necessary for the planning and policymaking stages. However, the water levelbased illustrations are unable to provide an accurate spatial extent of the overflow (Huang, Chen & Wu, 2014). Also, constructing a flood map using field survey data immediately after a severe flood event appears challenging and unsuitable for massive landmass.
Conversely, satellite-based flood maps tend to provide adjacent real-time representations with various spatial resolutions over time and higher accuracy (Singha et al., 2020). These techniques are divided into two categories. The first refers to flood mapping using a passive optical sensor, while the second employs the active option. Passive optical sensors use solar reflectance and are unable to capture images under cloud cover, but the active devices rely on signals capable of penetrating the barrier. This competence offers the material suitability for flood mapping (Long, Fatoyinbo & Policelli, 2014, Clement, Kilsby & Moore, 2018, e. g the synthetic aperture radar (SAR). Furthermore, the water surfaces and flooded areas appears relatively dark due to low backscattering signals in the SAR images. This condition forms a separate behavior in the water surfaces compared to other landforms. However, flood detection using SAR imagery is accompanied with certain limitations. For instance, during a flood event, windy conditions potentially develop particular water surface roughness. This circumstance creates various challenges in classifying the areas (Brisco et al., 2009), where radar shadows in SAR images are mistakenly categorized as flooded regions (Mason et al., 2010).
Numerous studies have also employed SAR images to map and monitor various flood events in Bangladesh. In addition, RADARSAT snapshots were used to explore the flooded areas for varying occurrences in other regions.
This study also adopted the Sentinel-1 SAR images, with high spatial resolution ranging from 5×5m to 20×40m based on the acquisition mode and processing level, although the temporal resolution was 6 days (ESA, 2012). The CDAT method, described as fast and straightforward, was used to detect the flooded segments (Cian, Marconcini & Ceccato, 2018). Subsequently, the image processing and SAR analysis were achieved using Google Earth Engine (GEE). The advantage of this cloud computing platform is that the applied Sentinel-1 dataset and other satellite resources are also stored. These diagrams are radiometrically calibrated and terrain corrected (Clement, Kilsby & Moore, 2018) to ensure cost-effectiveness, in terms of data storage and computation time. Therefore, the purpose of this study is to determine the spatial extent of inundation caused by the super cyclone, Amphan, in Bangladesh's coastal regions. Figure 1 represents the research location, including the original flood track. Under this analysis, single pre-event and double post-event SAR images were evaluated.

METHODOLOGY
This study encompasses flooded area detection using satellite images, followed by verification and validation with secondary data. Also, highresolution C-band SAR diagrams from the Sentinel-1 SAR GRD dataset were acquired for pre and post-Amphan cyclone events. These illustrations were obtained on 16 May 2020 for pre-cyclone, while 22and 28th featured the postcyclone effect. Image processing for detecting flooded areas was conducted using the GEE platform. The Sentinel-1 satellite employed 4 acquisition modes, termed strip map (SM), interferometric wide swath (IW), extra-wide swath (EW) and wave (WV). However, only the IW was used for land surface capture (Torres et al., 2012), although Sentinel-1obtains images in VH (vertical transmit and horizontal receive) and VV (vertical transmit and vertical receive) polarization in IW acquisition mode. Several researches showed that co-polarized (VV and HH) pictures have advantages over crosspolarized (VH and HV), in terms of flooded area identification (Twele et al., 2016). Previous studies suggested that VV samples are less prone to wrong classification (Clement, Kilsby & Moore, 2018), but this study solely adopted the VV option. Meanwhile, the CDAT methods were used to detect flooded areas, as proposed by Long et al. (Long, Fatoyinbo & Policelli, 2014), due to its proven reliability in earlier related studies (Long, Fatoyinbo & Policelli, 2014, Clement, Kilsby &Moore, 2018, including an analysis in Bangladesh between 2014-2018 (Singha et al., 2020). As a consequence, more accurate and superior flood maps are obtained (Matgen et al., 2011). Furthermore, pre and post-cyclone images were extracted from the Sentinel-1 GRD dataset by filtering based on date, acquisition mode and GEE polarization. The reference sample (R) is the non-flooded image acquired before the cyclone (16 May 2020). Proper selection of reference representations plays a significant role in detecting the flooded areas using SAR data (Hostache, Matgen &Wagner, 2012). Generally, the driest period is preferred for the reference image collection (Clement, Kilsby & Moore, 2018, Singha et al., 2020. Average rainfall of 200mm is observed depending on the CHIRPS data (Hassan et al., 2020). In this reported study, the period was selected before the Amphan event to avoid heavy rainfall in the research location. Therefore, the reference images indicate realistic wetlands, while presenting the flooded areas. Also, the post-cyclone effects were specified after the landfall for days 1 and 7.
Subsequently, a filter was applied to the reference and flood images to eliminate noise. The famous Lee filter was employed with a window size of 7×7 and a separate image was generated, using the variations in the backscattering differences between the pre-and post-cyclone captures.
In the various illustrations, the flooded areas appeared dark while non-flooded, or regions with no changes were grey.
A threshold value was applied to the difference image to extract only the areas with additional inundation probability. The flooded pixels (PF) was identified using equation (1), as formulated by Long et al. (Long, Fatoyinbo & Policelli, 2014): Where, μ[D] and σ[D] is the mean and standard deviation of the difference image(D), and fc is a coefficient with an assumed optimal value of 1.5.
Under this threshold criterion, the pixels are considered flooded with value less than the mean pixel minus 1.5 times the image standard. Therefore, flooded pixels were detected for both post-flood images on 22 and 28 May, with subsequent map generation. These plots were manually analyzed using GIS tools to obtain the flooding scenario at the district and Upazila levels.
The collection of ground observation data in the affected areas was deferred due to the COVID-19 lockdown. However, the result accuracy was validated using two separate secondary data sources because of the reliability, either as actual field data or validated sources from previous related studies. In the first step, the validation is achieved by comparing the total inundated areas based on a machine learning classification technique, termed support vector machines (SVM), in addition to using SAR imagery (Hassan et al., 2020). This is followed by the use of an inundation map for the post-Amphan event, based on primary field data from the Needs Assessment Working Group (NAWG) survey, Bangladesh. The methodology used in this study is illustrated using a workflow diagram in Figure  2.

The flooded area by CDAT method
Amphan storm surge, alongside heavy rainfall, instigated the flooding. The affected regions were detected using the CDAT method and were further analyzed with a GIS tool to identify the sites, after preparing the flood extent map with the SAR data. This analysis showed 2 an inundated landmass of 2146 2 on 22 May 2020, a day after the event, excluding the Sundarban. Among the 19 coastal districts, 13 were affected due to the flood. Table 1 Figure 4 shows Betagi, Bakerganj, Tungipara, Kanthalia, Nalchity, Rajapur, Paikgacha, Mirzaganj, Bhandaria, Kawkhali, Mathabaria, Zianagar as the most affected Upazilas with over 25% inundation.
Precisely on 28 May, virtually a week after the Amphan event, the total inundation area reduced to 1429 2 2 , with no significant change in most districts. Meanwhile, the percentages of inundation areas remain unaltered in Bagerhat, Satkhira, Khulna, Patuakhali, Jessore and Lakshmipur, due to the broken embankments. Generally, 76 km embankments have been reportedly damaged by the cyclone Amphan (Needs Assessment Working Group, 2020). These structures were supposed to prevent water influx into the inland during high tides. Table 1 and Figure 3 show the changes in the inundation area on 22 and 28 May, respectively, indicating the flood severity. Conversely, flood water abated significantly in Pirojpur, Barisal, and Jhalokati, but demonstrated an increase in the inundation area for Barguna, Gopalganj, and Narail, due to a higher river discharge. As earlier mentioned, the extreme precipitation instigated the excessive river overflow, resulting in a more severe post-Amphan situation in certain regions.   Figure  5, most of the areas were known to retain the flood a week after the occurrence, similar to a scenario in Figure 6. Furthermore, the majority of the Upazilas demonstrated minimal reductions in inundation areas or no recorded relevant changes. Virtually, the entire Upazilas in Jhalokati and Pirojpur showed a significant decline in inundation area percentage. Figures  5(g) and 5(l) reflect a rapid floodwater withdrawal in these regions, compared to the remaining affected Upazilas.   Figure 5 shows the creation of two inundation maps based on the flood area calculation on 22and 28 May. These regions are represented in the percentage of the total affected Upazila and the maps represented the spatial and temporal extent of the flood. However, the flood water became very persistent after crossing the Sundarbans, along the mangrove forest. This circumstance is easily comprehended by closely comparing these two inundation maps.

Validation
In the validation of image processing results, comparisons were formed with both secondary datasets, including the NAWG field survey, and a machine learning approach, known as SVM using the SAR data. Table 2 compares the findings of these two inundation studies in the affected districts. The affected area variation for several regions was calculated. Jhalokati obtained the maximum percentile variation at 6.9%, while Jessore was 5.1%. These two locations were both evaluated on 22 May 2020, but the remaining districts, ranged between 0.5-2.5%. Furthermore, most territories showed similar results from these two distinct processes.
The present research results were also compared with another data source, an inundation map ( Figure 6) prepared from the validation survey by Needs Assessment Working Group (NAWG), Bangladesh. This illustration showed the severity of the union-level flooding scenario due to the Amphan cyclone. The unions with inundation areas above 50% were presented with red, while the ranges of 1-25% and 25-50% denoted dark and light orange, respectively. Subsequently, the union-level inundation data was compared to the flood map generated for 22 May 2020 (Figure 7), with the flooded areas represented in blue. The level of inundation in the affected unions was very similar in both maps and the reference illustration was prepared using the primary data from the validation survey. Therefore, the study findings positively reflected the actual flood conditions.

CONCLUSION
The purpose of this study was to construct a flood inundation map due to Amphan cyclone, using SAR data. This effort is possibly achieved by applying a methodology that combines change detection and thresholding (CDAT) methods. Subsequently, the inundated areas were analyzed with geospatial tools.
2 The results showed an inundated landmass of 2146 2 , excluding the Sundarban. In addition, Bagerhat, Pirojpur, Satkhira, Khulna, Barisal, Jhalokati, Patuakhali, and Barguna obtained the most coverage. Therefore, the flood magnitude tends to aggravate the prevalent situation, despite declining to 1429 2 after seven 2 days. The floods were reportedly caused by heavy rainfall triggered, leading to several structural breaches. A total of 550 km of roads and embankments were relatively damaged, according to a field survey (Needs Assessment Working Group, 2020). The analysis results were subsequently compared with an inundation map. This diagram was designed using a validation survey by NAWG, Bangladesh, immediately after the Amphan event. Also, the present study was assessed with a previous related investigation using SAR data based on a machine learning technique, SVM. In this regard, a reasonable match was obtained, and the study findings are assumed to offer viable guidance for future decision support systems.

DISCLAIMER
The authors declare no conflict of interest.

AVAILABILITY OF DATA AND MATERIALS
All data are available from the corresponding author.

ACKNOWLEDGMENTS
This research was supported by the funds provided for the first author undergoing a Master of Science program in Civil Engineering, Chittagong University of Engineering and Technology (CUET), Bangladesh. The authors are grateful to BWDB, CPA, and NAWG for providing the necessary bathymetric, meteorological, and validation data, respectively. from Bangladesh. Remote Sensing,11(13