Multi-model Ensemble Member Forecast And Verification Of Precipitation Over China

Based on the daily 24-168 h ensemble precipitation forecasts over China from April 1 to September 30 in 2016 from the global ensemble models of ECMWF,JMA,UKMO,NCEP and CMA taken from the TIGGE archives,and the hourly merged precipitation product over China as the observation,a study on multi-model ensemble forecasts has been conducted.Different metrics have been used to evaluate the forecasting ability of different models of their precipitation forecasts.Then the measure of forecast challenge(MFC),which combines various forecast errors,and the predictability horizon diagram index(PHDX),which verifies the forecast process through the multiple lead time when the precipitation approaches,have give the more accurate and comprehensive estimate on the forecast models.Finally the result of multi-model ensemble forecasts before and after calibration by frequency matching method has been discussed in order to improve the results of prediction of precipitation based on numerical weather forecast date.The ability for precipitation forecasts of the five ensemble models is different.The root mean squared error(RMSE)and mean absolute error(MAE)of each model increase,while the anomaly correlation coefficient(ACC)and threat score decrease with the increasing of lead time.Among all ensemble model,the forecasting ability of ECMWF is superior to JMA and UKMO,and the ability of NCEP and CMA is poor.Different from current metrics,MFC verifies the ensemble forecasts through the aspect of forecast deviation and uncertainty information.For an ensemble model,the smaller MFC value means the higher forecasting capability.Furthermore,the predictability horizon diagram index by applying MFC,verifies the ensemble forecasts from the whole lead time of prediction.A positive value indicates the skillful forecast ability when the precipitation reaches and the ensemble forecast gradually approaches the observation,while the negative value is associated with the poor forecast skill.Comparing the differences from MFC and PHDX of the precipitation forecasts of ECMWF and JMA,the results show that the forecasting skill of ECMWF are indeed better than JMA.The precipitation forecast of ensemble mean method(EMN)can barely improve the result of single model.The precipitation of multi-model super ensemble(SUP)andbias-removed ensemble mean(BREM)have a slight improvement over ECMWF,which reduce the forecast error of each lead time by 0.2 mm compared with single model.What's more,the result of Kalman filter(KF)has the beat forecast skill among all multi-model ensemble forecasts,which can produce the more accurate precipitation of the area of precipitation.The results show that the precipitation forecasts calibrated by the frequency matching method,which improve the TS of light and heavy rain,can effectively reduce the problem of the filtering effect caused by ensemble mean forecast.Therefore,the multi-model ensemble precipitation after calibration based on ensemble members is more effective in precipitation forecasts than the traditional method,which is closer to the observational data both in terms of the prediction of the precipitation area and category and make the spatial distribution of the precipitation forecast more accurate.