Climate change intensifies the hydrological cycle, and reservoir responses to meteorological forces are vital for water sustainability, but long-term quantitative assessments remain scarce. In this study, we extracted the water surface area of Xinfengjiang Reservoir from 1990 to 2024 using multisensor Landsat/Sentinel imagery and a random forest classifier (RFC) on the Google Earth Engine (GEE) platform. Furthermore, we established an interpretable modeling framework that combined random forest regression with Shapley additive explanations (RFR-SHAP) to elucidate the nonlinear influences of meteorological drivers on multidecadal reservoir dynamics. The RFC model achieved high accuracy (OA = 96.90%, KC = 0.95) and efficiency (processing time = 1.17s per image), enabling the generation of a reliable 35-year time series dataset. Further analysis of the underlying mechanisms revealed that total precipitation (TP) was the most significant driving factor for the hydrological dynamics of reservoirs, with mean |SHAP| values for annual and monthly assessments recorded at 3.33 (16.15%) and 4.32 (20.36%), respectively. The synergistic and antagonistic effects of meteorological variables constituted the nonlinear adaptation mechanisms of reservoir systems to climate forcing. These findings quantify the hydrological dynamics and the driving mechanisms of meteorological variables on the water surface area of Xinfengjiang Reservoir, providing a basis for proactive reservoir operations based on precipitation forecasts.
Huang et al. (Fri,) studied this question.