Surface Water Extraction by Multidimensional Feature Fusion of Sentinel-1/2 and Random Forest: A Five-Year Analysis in Jiangsu, China

Surface water extraction in regions with complex terrain is often hindered by low accuracy and frequent misclassification. To mitigate these challenges, this paper proposes a high-precision water extraction framework based on the GEE platform. The framework centers on the RF classifier and integrates multi-polarization scattering features derived from Sentinel-1 SAR data, multispectral water indices obtained from Sentinel-2 imagery, and slope constraints extracted from the DEM. The joint use of multi-source features and topographic constraints improves the delineation of surface water boundaries, particularly in complex terrain and under various interference conditions. Jiangsu Province was selected as the experimental study area. Experimental results indicate that the proposed framework achieves a Kappa above 0.9 and a PA exceeding 90%, reflecting stable and reliable surface water extraction performance. In addition, the framework is capable of characterizing the spatiotemporal variations of surface water in Jiangsu, thereby supporting large-scale and high-frequency monitoring of surface water resources in complex environments. Further experiments suggest that the framework can be generalized and applied to other regions. Our code is provided at https://github.com/wumingxing5/Surface-Water-Extraction-by-Multidimensional-Feature-Fusion-of-Sentinel-1-2.