Accurate reservoir area data are essential for effective water resource management, yet traditional field surveys often face labor and logistical challenges. In this study, we evaluated the Geospatial Segment Anything Model (GeoSAM) in conjunction with high-resolution KOMPSAT-3/3A satellite imagery for reservoir delineation in the Korean Peninsula. Our experiments demonstrate that GeoSAM consistently achieves high accuracies (85.95–97.10%), surpassing the conventional normalized difference water index-based extraction method, which averaged 93.74%. Moreover, GeoSAM maintains robust performance under challenging conditions—such as frozen reservoirs, shadowed areas, and cloudy environments—by incorporating additional point prompts. These findings underscore the potential of GeoSAM to advance remote sensing applications in water resource management, particularly for small- and medium-sized urban areas.



Keywords: remote sensing, deep learning, Segment Anything Model, image segmentation, water body extraction, reservoir monitoring, KOMPSAT-3/3A