A Study On Multi-Model Ensemble Forecast Of Precipitation Using Categorized Rainfall

Based on the ensemble forecasts of 1-10 day daily accumulated precipitation of 123 days(from May to August in 2013)from the China Meteorological Administration(CMA),the European Centre for Medium-Range Weather Forecasts(ECMWF),the Japan Meteorological Agency(JMA),the National Centers for Environmental Prediction(NCEP)and the UK Met Office(UKMO)in the TIGGE datasets,and the hourly merged precipitation product over China as the observed data,a multi-model ensemble forecast study on daily precipitation in China has been conducted.Firstly,the precipitation forecast of each model was tested by using observation data to evaluate the precipitation forecast error and the forecasting ability of different models for different rainfall levels.Then Based on the categorized rainfall,a multi-model ensemble forecast has been conducted by using the method of Bias-removed ensemble mean(BREM)and Super ensemble(SUP)and compared with the multi-model ensemble forecast without categorized rainfall.Finally,the Bayesian model average method(BMA)was used to conduct the precipitation probability forecast research,and the BMA method based on the categorized rainfall was further analyzed for the correction effect of precipitation forecast,so that a more reasonable probabilistic forecast product of precipitation can be obtained.Different foreca,sting centers in TIGGE data has different forecasting ability for precipitation.With the extension of the lead time of forecast,the TS scores and ACC of precipitation forecast in each center gradually decreased,while the RMSE increased continuously.ECMWF had the highest ACC during the the lead time of 1-10 days,and the RMSE was the smallest.The TS scores of all the five forecasting centers showed that JMA has the best forecasting techniques for light rain and ECMWF has the best forecasting skills for moderate and heavy rains.BREM can effectively reduce the system error of model and enhance the anomaly correlation coefficient with the observed data,so the accuracy of forecast was better than the best single-mode ECMWF.However,the precipitation forecast of SUP was not stable and the RMSE was even larger than the optimal single model forecast in some lead time.Results show that the forecasts of BREM and SUP after the categorized rainfall are more accurate than the ones after direct multi-model ensemble In particular,the forecast of BREM based on the categorized rainfall is an optimal forecasting method because the root mean square error in each lead time is less than others and the ACC is bigger than others.Compared with single-model forecast,the RMSE decrease by 20%,8%,11%,17%,11%respectively.In addition,the TS scores of this method is superior to the optimal single-model forecast results in light rain,moderate rain,and heavy rain.The BMA method can provide the probability forecast of precipitation and the probability density function.The BMA method based on categorized rainfall is more accurate in predicting the precipitation,and it can improve the accuracy of the deterministic forecast by reasonably optimizing the distribution of the weight coefficients of each model;THE forecast of BMA without categorized rainfall can accurately predict the area of precipitation and the location of the rain band,but its deterministic forecast results are relatively smaller than the observed data,While the forecast of BMA with categorized rainfall is better than the ones of the former and optimal single-mode because of the higher ACC and lower RMSE.