Amir Naghibi

Accurate and continuous monitoring of inland surface‌ water dynamics is vital for sustainable water resource management, climate impact assessment, and ecological planning. This study introduces a machine learning-based remote sensing-based data-fusion framework that improves Sentinel-3 radar altimeter (SRAL) estimations of water surface elevation (WSE) by integrating ERA5 reanalysis climate variables (precipitation, temperature, and evapotranspiration). Specifically, the Random Forest (RF) model was employed for its robustness and ability to capture nonlinear relationships among diverse geospatial variables. Using Lake Winnebago (LW, United States) as a test site and 101 Sentinel-3A observations, we designed six scenarios to evaluate model performance against in situ gauge records (2016-2024): (0) median SRAL WSE values alone; (1) RF with 41 virtual-station elevations only; (2-4) RF with the addition of 1climate variable at a time; and (5) RF with all 3variables. The best result RF with all 3climate inputs reduced RMSE from 0.47m (scenario 0) to 0.08m and increased R² from 13.65 % to 73.68 %. These findings underscore the dominant role of environmental variables in WSE modeling and demonstrate that fusing SRAL with ERA5 via RF can significantly enhance hydrological monitoring of small inland water bodies.