2025 Volume 6 Issue 1 Pages 403-416
In Japan, river overflows and landslides caused by typhoons and heavy rainfall frequently occur. To rapidly assess disaster-affected areas, satellite Synthetic Aperture Radar (SAR) imagery is widely utilized. However, high-resolution SAR data are typically costly, limiting accessibility. The use of freely available SAR imagery offers a viable alternative for disaster assessment and infrastructure management. Existing studies have proposed methods for estimating inundation areas based on the decrease in SAR backscatter coefficients. However, these approaches face challenges in urban areas where backscatter coefficients do not necessarily decrease during flooding, making inundation detection difficult. To enable large-scale flood assessment applicable to both urban and non-urban regions, a more robust method is required.
In this study, we propose a novel approach that first extracts inundation areas using backscatter coefficients and then detects flooded zones in urban regions by analyzing coherence values with deep learning techniques. Our findings indicate that this method can effectively estimate urban inundation areas that conventional backscatter-based approaches fail to capture.
