Flood Detection Utilizing Sentinel-1 Synthetic Aperture Radar Data: A Comparative Analysis of U-Net and DeepLabV3 Architectures with Varied Encoder Configurations

Abstract

Flood detection is essential for mitigating the devastating impacts of natural disasters, especially in flood-prone regions where timely intervention can save lives and resources. This study proposes an innovative approach to flood detection. It leverages Synthetic Aperture Radar (SAR) imagery from Sentinel-1 satellites, which offers robust, all-weather monitoring for large geographic areas. This research paper explores the use of deep learning techniques to enhance the accuracy of flood detection using SAR data.  The paper further performs  comparative analysis of two state-of-the-art deep learning architectures namely U-Net and DeepLabV3, both of which are optimized for pixel-level segmentation tasks. By experimenting with varied encoder configurations, including ResNet, MobileNet, DenseNet and applying them to SAR data, this study evaluates the accuracy, precision, recall, and computational efficiency of each model in identifying flooded regions. The findings highlight critical differences in how these architectures and encoder configurations handle noisy radar data, offering insights into the most effective model-encoder combinations for flood detection. This research contributes to the advancement of automated flood monitoring systems and serves as a basis for further improvements in disaster response technologies.