Abstract: Accurate and reliable predictions of TWSA over the Yangtze River Basin are crucial for achieving its efficient and sustainable water resource management. This study utlizes a method combining seasonal adjustment and a nonlinear autoregressive neural network (NARNN-SA), leveraging the autocorrelation of TWSA changes in the Yangtze River Basin for prediction. Two tests were designed to evaluate the effectiveness of this method by comparing against three models: AutoRegression (AR), Seasonal AutoRegressive Integrated Moving Average (SARIMA), and a nonlinear autoregressive neural network without seasonal adjustment (NARNN). Results demonstrate that the introduction of seasonal adjustment significantly improved the predictive performance of the NARNN-SA model, which achieved high consistency with GRACE observations, with a correlation coefficient of 0.89, a Nash-Sutcliffe Efficiency (NSE) of 0.77, and a root mean square error (RMSE) of 1.43 cm during the testing phase. Additionally, NARNN-SA successfully predicted TWSA changes in the headwaters, eastern, and western regions of the Yangtze River Basin and effectively bridged the data gap between the GRACE and its successor, GRACE-FO missions.