Abstract:The daily precipitation records of 59 rain gauges over the Huaihe River Basin and the ensemble mean forecasts of 24 h cumulated precipitation of the reforecast data generated by Global Forecast System (GFS) during the period of 1 January 1981 to 31 December 2003 are employed to construct a probability forecast model which can produce ensemble forecast based on conditional meta Gaussian distribution. The experiment to show how well the model works is done in four catchments of Huaihe River Basin. The result shows that the root mean square error (RMSE) of the ensemble forecasts generated by the new method introduced here reduces remarkably for all catchments and seasons. Especially the RMSE reduces by 3.11 mm in the stream between Bengbu Station and Hongze Lake in June. The brier skill score (BSS) of the ensemble forecasts of four catchments is between 0.16 and 0.61, suggesting that the ensemble forecasts can be used throughout the whole year. The ensemble forecasts in summer of the stream between Dapoling and Xixian Station are always lower than the observations in varying degrees when the threshold of daily mean areal precipitation (MAP) is between 0.00 mm and 7.82 mm, but when the threshold is 18.12 mm, the ensemble forecast in summer shows better reliability. Known from the percentile comparison between the ensemble forecasts and the observations, the ensemble forecast can capture more precipitation events of small magnitude MAP but is weak in forecasting MAP of large magnitude over 30 mm.