Harnessing Satellite Imagery for Effective Flood Detection Using Convolution Neural Network

Earth Observation has become essential for monitoring geosciences and human activity. With the increasing availability of data, Artificial Intelligence (AI) algorithms have delivered significant advancements in remote sensing. However, weather-related disasters like hurricanes and floods, often accompanied by dense cloud cover, limit the effectiveness of optical imaging. Synthetic Aperture Radar (SAR), which penetrates clouds, offers a robust alternative. This study presents a novel dataset of co-registered optical and SAR image time series for flood event detection. By integrating spatial and temporal information using a ResNet-GRU architecture, we overcome the challenges posed by limited labeled SAR data. Our method achieves state-of-the-art accuracy, demonstrating resilience in detecting floods under adverse conditions. Performance metrics indicate substantial improvements in disaster response capabilities, emphasizing the practical utility of SAR in flood monitoring. This work provides critical insights for researchers and practitioners in leveraging SAR and optical data for disaster management.