In order to enhance the predictability of seasonal streamflow, numerical weather prediction was coupled with the Xin’anjiang model, the distributed hydrological soil vegetation model (DHSVM), and the long short-term memory model (LSTM) for ensemble forecasting of monthly streamflow in the Jiaojiang River Basin from 2012 to 2020. Three different methods, namely, equal weighting, unequal weighting, and BP neural network-based weighting, were employed to fuse the outputs from the three models. Comparison was made between the fused forecasts and the optimal forecasts of single models. The results indicate that the BP neural network fusion method significantly enhances the forecasting accuracy, demonstrating superior performance over other methods. Notably, this method substantially extends the effective forecast lead time across all four distinct seasons (spring, summer, autumn, and winter), thereby providing more reliable hydrological predictions for water resources management and utilization in the basin.