Abstract: The rainfall field forecasted by the ECMWF (European Centre for Medium-range Weather Forecasts) model from January of 2012 to March of 2015 in Anhui province was tested using TS (Threat Score),forecasting bias (BIAS),ETS (Equitable Threat Score) after removing random events,and TSS (True Skill Statistic) score methods in order to understand the forecasting performance of the ECMWF model and improve the correction skills.The results show that,in general,the forecasting performance of the ECMWF model in Anhui province is stable and exhibits a little inter-annual variation.There is a spatial distribution feature with high in the south part and low in the north part of the ETS value.All amounts of rainfall in each meteorological station are forecasted excessively.Based on the classification test in the amount of rainfall prediction,it indicates that the forecasting score of sprinkle is significantly higher than that of the other rainfall,while the forecasting bias of sprinkle is higher and the excessive forecasting phenomenon is serious.The model forecasting ability for the rainstorm within 72 h is very limited and the model has no forecast score for the rainstorm beyond 72 h.The accuracy forecasting rate of clear or rain within 48 h in spring,autumn and winter are above 88% with a small room to correct.The rate of false-alarm for forecasting of rainfall in summer and in spring,autumn and winter beyond 168 h is above 60%,and this needs to be corrected accordingly.The rate of missing report of rainfall in autumn is higher than that in the other seasons,especially for the rainfall forecast beyond 120 h.The amount of rainfall in each season is forecasted excessively,especially in summer.The rate of false-alarm of rainfall within 0.1-0.7 mm to the grid prediction is higher.This false-alarm rate can be improved significantly after correction,but it increases the risk of missing report rate.

Key words: ECMWF model, Test, Rainfall, Threat Score (ETS), Equitable Threat Score (ETS), Forecasting bias