Abstract:Based on CN05.1 precipitation data and the SWAT (Soil and Water Assessment Tool Model) model, a deterministic runoff forecast was carried out in Ganjiang River Basin. On this basis, the characteristic of prediction error was analyzed and provided a probabilistic streamflow forecasting using the HRD (Heterogeneous Residual Distribution) method, which takes the heterogeneous distribution characteristic of errors into account. By using the residual standardization module and Gaussian anamorphosis module, the influence of streamflow on errors is eliminated, then temporal structure of Gaussian errors is constructed, and the probability distribution of errors is estimated, thereby achieving probabilistic streamflow forecasting. The results of study on hydrological and meteorological data in 1963-2013 indicate: (1) The SWAT model has good precision in simulating daily runoff at Waizhou station with Nash–Sutcliffe Efficiency bigger than 0.84 and flood volume error and the average annual flood peak error are both within 10%. (2) The HRD model can effectively consider the impact of streamflow on forecasting errors and provide a reasonable probabilistic forecasting with containing ratio coefficient above 0.66. Moreover, both median and mean forecasts are superior to the results of deterministic prediction, with the benchmark efficiency above 0.64. This study can provide new ideas for high-precision runoff forecasting, and has important theoretical and practical significance for water resource management and flood disaster prevention and mitigation in the basins.