Flooded area monitoring using SAR image-based water body detection technique

Flood is one of main water disasters and causes damage to human life and property. The spatial and temporal disproportion of precipitation due to recent climate change causes flood worldwide every year. As the severity of flood rises, accurate monitoring of flooded areas is being essential for preparation and adaptation. Due to the wide area-occurring characteristics of flood, the use of remote sensing is effective for detecting of flooded areas. In a SAR image, surface of water body appears smooth, so the backscattering coefficient value is generally low. Conversely, surface of non-water body is rough, so the backscattering coefficient value is high. It is possible to divide water body and non-water body by using the characteristics of the backscattering coefficient and specific threshold value. However, the histogram of the backscattering coefficients around rivers where flood occurs most often has a multi-modal distribution, so there is a limit in detecting water bodies using a threshold value only. In this study, a histogram-based multi-threshold method, an AI-based K-means clustering method, and an object segmentation-based Chan-Vese method were used to detect water bodies before and after floods in Sentinel-1 SAR images. The water/non-water body classification image from the Sentinel-2 optical image was used for validation. If SAR images with high spatial and temporal resolution will be available, it is expected that efficient water disaster management will be possible through near real-time detection of flooded areas.