As a critical water regulatory region and ecological barrier in the middle reaches of the Yangtze River, understanding sub-basin scale Terrestrial Water Storage Anomaly (TWSA) is crucial for regional water resources management and early warning of droughts and floods in the Dongting Lake Basin. However, the relatively short and discontinuous records from the Gravity Recovery and Climate Experiment (GRACE) and GRACE-Follow On (FO) satellite missions limit long-term hydrological analysis. This study reconstructed monthly TWSA from 1980 to 2024 at the sub-basin scale by integrating multi-source data, including GRACE/GRACE-FO satellite observations, global reanalysis (MERRA-2), the Global Hydrological Model (WGHM), and meteorological data (precipitation and temperature). Methodologically, the Seasonal-Trend decomposition using Loess (STL) method was applied to isolate different temporal components of the input variables, which were then reconstructed using two machine learning approaches: the Generalized Regression Neural Network (GRNN) and the Long Short-Term Memory (LSTM) network. Results demonstrate that the GRNN method consistently outperformed LSTM across most sub-basins. The GRNN model driven by MERRA-2 data achieved the most accurate reconstruction, demonstrating strong agreement with the GRACE-derived TWSA in both phase and amplitude. Among the sub-basins, the Xiangjiang River Basin, which has the largest spatial extent, showed the highest accuracy, whereas the Lishui River Basin, being the smallest sub-basin, presented the largest uncertainty due to increased data noise at the smaller scale. Long-term trend analysis revealed a significant increasing trend in TWSA (+1.57 mm/a) for the entire basin and the sub-basins over the past 45 years, which is consistent with the increasing trend in precipitation (+0.87 mm/a) and reservoir water storage (+1.12 mm/a). Short-term TWSA dynamics often exhibited an inverse trend against evapotranspiration. Notably, sharp declines in TWSA were detected during major historical drought years (1985, 2003, 2011, 2022), aligning strongly with negative precipitation anomalies and extreme climatic conditions.Although the terrestrial water storage in the Dongting Lake Basin shows an increasing trend, droughts still occur frequently, indicating that the water resource situation in the basin is becoming increasingly severe. The reconstructed TWSA time series provides scientific supports for sub-basin scale water resources management and prediction of extreme hydrological events in the Dongting Lake Basin.