Abstract: To improve the accuracy and reliability of streamflow forecasting, nine ensemble runoff prediction schemes were developed for the Qingjiang River basin by combining two precipitation forecasts （ECMWF and NCEP） with two hydrological models （GR4J and XAJ）. A generator-based post-processing method （GPP） was applied to correct systematic biases in model precipitation. The performance of each scheme was comprehensively evaluated from deterministic, probabilistic, and uncertainty perspectives. Results indicate that the GPP method significantly enhanced precipitation quality, thereby improving streamflow forecasting accuracy—particularly for the NCEP product, which exhibited larger initial errors. Integrating high-accuracy precipitation forecasts with multiple hydrological models further improved the robustness and reliability of ensemble runoff predictions. During validation against a typical flood event in 2020, all schemes accurately captured the flood peak timing at 1- and 3-day lead time. Notably, the ensemble-mean peak flows from the multi hydrological model integration schemes closely matched the observed peak.