Abstract: To quantify the relative contributions of climate change and human activities to terrestrial water storage anomalies (TWSA) in China, we develop an integrated assessment framework that combines GRACE/GRACE-FO satellite observations, multi-source climatic factors, and an improved water balance model. Machine learning is used to fill the data gap between 2017 and 2018, and an improved gradient boosting machine model is introduced to reconstruct climate-driven TWSA based on precipitation, air temperature, evapotranspiration, and other factors, achieving high fitting accuracy across China’s nine major river basins (correlation coefficients all exceed 0.9). The results indicate that terrestrial water storage in China showed an overall declining trend during 2005—2020, with the most pronounced decreases occurring in the Haihe River (− 14.64 mm/a) and Huaihe River (− 11.74 mm/a) basins, while the Yangtze River and Pearl River basins exhibited increasing trends. Contribution analysis reveals that the influence of human activities has continued to intensify in arid and semi-arid regions, with their contribution in the Yellow River Basin reaching nearly 90% by 2017, whereas TWSA in humid regions is mainly driven by climate change. These findings provide a quantitative basis for identifying key driving factors and supporting regional water resources management.