Evaluating Flood Susceptibility in Coastal Regions of China Using Deep Learning Models and Multispectral Remote Sensing Data

For successful disaster management, risk assessment, and long-term growth in coastal areas that are at risk, it is essential to have accurate and up-to-date flood maps. We examine sophisticated deep learning models for distinguishing the meanings of flood-affected areas in China's most flood-prone coastal regions, including the Yangtze River Delta, Pearl River Delta, Eastern Coastal China, and the Bohai Sea region. We utilize a high-resolution aerial imagery dataset with manually annotated flood masks to evaluate the effectiveness of an Attention U-Net architecture designed explicitly for binary flood-water delineation. We utilize essential segmentation measures, including Intersection over Union (IoU), F1-score, accuracy, and recall, to evaluate the model's performance. The findings of the experiments suggest that the Attention U-Net is effective in identifying the difference between flooded and non-flooded areas, and it performs well across a wide range of geographic and land cover settings. The system shown here holds considerable promise for working with multi-sensor data fusion methodologies, making flood monitoring in complex coastal areas easier and faster. This research provides valuable insights for enhancing flood mapping using remote sensing, which will enable China's coastal megadeltas to respond more effectively to disasters and mitigate risks.