Coastal inundation, often triggered by tropical cyclones, presents a significant compound hazard resulting from the combined effects of storm surges, riverine flooding, and intense rainfall. Rapid and accurate inundation mapping is critical for timely disaster management and emergency response in coastal regions. This study proposes a novel deep learning model designed for rapid coastal inundation mapping, utilizing dual-polarization and bitemporal Sentinel-1 synthetic aperture radar (SAR) imagery. The model introduces two key innovations aimed at enhancing feature extraction and interaction, thereby improving the accuracy of inundation detection.

In this research, coastal flooding events from two tropical cyclones were analyzed, resulting in the collection of 5,039 image pairs after extensive preprocessing and quality control. Of these, 2,784 pairs were used for training, 696 for validation, and 1,559 for testing the model. The proposed model demonstrated strong performance, achieving an intersection over union (IOU) score of 79.44%, surpassing the accuracy of state-of-the-art flood detection models.

The results highlight the model’s potential for real-world application in emergency management scenarios, where rapid and reliable coastal flooding detection is essential. By providing more accurate and timely inundation maps, this approach could significantly enhance disaster response efforts, contributing to better protection of life and property in vulnerable coastal areas. The study underscores the importance of leveraging advanced deep learning techniques and SAR data to address the growing challenges posed by climate-related coastal hazards.