Flood Extent Mapping From SAR Images by a Dynamic Threshold: A Case Study of 2022 Compound Flood in Northeastern Bangladesh

Satellite images have gained considerable traction for near-real-time and rapid flood mapping because they are agnostic to flooding mechanisms and their drivers, hence applicable to any floods, whether pluvial, fluvial, or both (i.e., compound flood). However, flood mapping using satellite images is challenging due to cloud cover and shadows, particularly in regions where floods typically occur during monsoon seasons with frequent and extensive cloud cover (e.g., South Asian countries). Generally, optical images (e.g., Landsat) are used alone or in combination with Synthetic Aperture Radar (SAR) images (e.g., Sentinel-1) under the multisource flood mapping (MSFM) framework for rapid flood detection. However, clouds often affect optical images, limiting their ability to provide reliable flooding information. Additionally, the MSFM requires diligent (and often manual) identification of a pre-flood image, which may be challenging when automatically delineating multiple floods. Moreover, using a unique permanent water body for flood delineation may be erroneous for regions like northeastern Bangladesh, where the water extent climatology varies significantly for monsoon months (when floods usually occur). In contrast, this work presents a rapid, simple, and comparable thresholding algorithm for flood delineation using Sentinel-1 SAR images, where backscatter coefficient values are used to identify the watered pixels, and the flooded pixels are determined by comparing them with the monthly climatology of water extent. The method is applied to map the surface water and flood extents for catastrophic back-to-back flood events in the northeastern part of Bangladesh in 2022 caused by heavy local precipitation and upstream discharges (i.e., compound flood), leveraging the cloud based computing capabilities of Google Earth Engine (GEE). The results (i.e., surface water and flood extent maps) are comparable to those obtained using the recently developed sophisticated MSFM and existing SAR-based methods. Besides, our thresholding algorithm provides reliable flood maps for instances where MSFM miscalculates flood extent due to significant cloud cover. The technique developed in this study will enable decision makers and emergency responders to obtain precise information rapidly during time-sensitive flooding events, supporting informed decision-making for resource allocation and rescue planning.