Performance Comparison of Water Body Detection from Sentinel-1 SAR and Sentinel-2 Optical Imagery Using Attention U-Net Model

As global warming accelerates greenhouse gas emissions, the frequency and severity of abnormal weather events such as floods and droughts are increasing, complicating disaster management and amplifying socio-economic damage. In response, effective strategies for mitigating water-related disasters and proactively addressing climate change are essential, which can be achieved through the use of satellite imagery. This study aims to compare the water body detection performance of Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 optical imagery using the Attention U-Net model. Through this comparison, the study seeks to identify the strengths and limitations of each satellite imagery type for water body detection. A 256 × 256-pixel patch dataset was developed using multi-temporal imagery from the Han River and Nakdong River basins to reflect seasonal variations in water bodies, including conditions during wet, dry, and flood seasons. Additionally, the study evaluates the impact of data augmentation techniques on model performance, emphasizing the need to select augmentation methods that align with the specific characteristics of SAR and optical data. The results demonstrate that Sentinel-1 SAR imagery exhibited stable performance in detecting large water bodies, achieving high precision in defining water boundaries (Intersection over Union [IoU]: 0.964, F1-score: 0.982). In contrast, Sentinel-2 optical imagery achieved slightly lower accuracy (IoU: 0.880, F1-score: 0.936) but performed well in detecting complex water boundaries, such as those found in wetlands and riverbanks. While data augmentation techniques improved the performance of the Sentinel-1 SAR dataset, they had only a marginal effect on Sentinel-2 optical imagery, aside from slight improvements in boundary detection under new environmental conditions. Overall, this study underscores the importance of threshold and satellite imagery integration for water body monitoring. It further emphasizes the value of selecting appropriate data augmentation techniques tailored to the characteristics of each dataset. The insights from this study offer guidance for developing enhanced water resource management strategies to mitigate the impacts of climate change.