Research On Predictability Of Ensemble Foecasting For Landing Tropical Cyclone Precipitation

The severe meteorological disasters such as gale,precipitation and storm surge brought by tropical cyclones(TC)have a tremendous impact on the economic development and public life of coastal areas.Compared with statistical forecasting and traditional numerical forecasting methods,ensemble forecasting is an important means for TC forecasting developed in recent years.However,it is more concentrated on the research of intensity and path forecasting of TC,and lacks of the specific analysis of predictability of TC precipitation forecasting.Therefore,it is urgent to carry out TC ensemble forecast error diagnosis analysis and revision technology research to deepen understanding of the mechanisms affecting TC precipitation,improve the quality and level of TC precipitation ensemble forecast,and meet the urgent need of meteorological business units to prevent and reduce disasters.This article objectively compares the cumulative precipitation forecast products from the TIGGE(the THORPEX interactive grand global ensemble)dataset and the precipitation data of the tropical rainfall measurement satellite TRMM/3B42RT,and uses a variety test indicators of quantitative precipitation forecasts(QPFs)and probabilistic QPFs(PQPFs)to evaluate the European ECMWF,the United States NCEP,China CMA three forecasting centers on TC precipitation ensemble forecasting capabilities,It also analyzes the correlation of TC precipitation errors between different forecasting ages,and develops a statistical correction technique for TC precipitation forecasting considering different member forecasting levels.The main conclusions are as follows:First of all,similar to the ensemble forecast features of other factors,the TC precipitation ensemble mean forecasting is better than the corresponding control forecast,and with the increase of forecasting time,the ensemble mean forecasting ability of TC precipitation declines.As far as frequency bias are concerned,with the prolongation of forecasting time and the increase of precipitation threshold(from light rain to large rain),the ensemble mean forecasting of TC precipitation from the three forecast centers has changed from overestimate to underestimate.Through further comparison with non-TC precipitation forecasting evaluating results,it is found that although ensemble forecasting still has lower forecasting levels for torrential rain,the forecasting capacity of the forecasting mode for medium and large rain is better than that for light rain,which may be due to TC precipitation ensemble forecasting have a serious overestimation of light rain.In addition,with the increase of forecasting time,the discriminational ability of each forecasting center of different threshold precipitation is reducing,mainly reflecting in the underestimation of precipitation.At the same time,the TC precipitation forecasts of all centers are all under-discrete,and the specific manifestation is that the degree of underdispersion after 48 hours increases with the increase of forecasting time.Furthermore,the ensemble mean forecasting of different centers differ in the ability to predict TC precipitation.In general,ECMWF has the best forecasting effect.NCEP and CMA have different forecasting effects on different forecasting period.That is,NCEP is superior to CMA in short-term forecasting,and CMA is better than NCEP in medium-term forecasting.For different thresholds of TC precipitation,the best forecast for middle rain is ECMWF,and for other grades of precipitation(light rain,large rain,torrential rain),the best forecast is NCEP.In addition,ECMWF distinguishes different threshold precipitation better than the other two centers.Secondly,the PQPFs of TC precipitation have different characteristics with the ensemble mean forecast.With the increase of precipitation threshold,the PQPFs shows the trend of increasing first and then decreasing,that is,the PQPFs skills of moderate rain and large rain is high,but for light rain and torrential rain is low.The probable reason is that the reliability of ensemble forecasting for the light rain forecast is poor,and on the other hand,the ensemble forecasting model has poor resolution for torrential rain.In addition,there are differences in probabilistic indicators between different center TC precipitation forecast products.Overall,ECMWF performed optimally,while NCEP was superior to CMA in short-term and CMA was better than NCEP in long-term predictions,this was consistent with the analysis of the ensemble mean foecasting.For different threshold of precipitation,it was found that forecast of light rain of NCEP was the best,and which of CMA was the worst,while forecast of moderate to torrential rain of ECMWF was the best,and which of NCEP was worst,which was caused by the difference in resolution and reliability of different center TC precipitation probability forecast.Thirdly,using MVL method to analyze the growth characteristics of TC precipitation forecasting errors,we find that the forecasting error have a linear increase in the short-term(1-4 days)forecasting,that is,there is a linear correlation between the errors of pre-precipitation forecasts and post-precipitation forecasts.In 1-4 days,members with good pre-precipitation forecasts are still at good levels of post-precipitation forecast.Based on this characteristic,a new calculation scheme for the ensemble mean is established by means of an arithmetic weighted average.The weights of different members of the scheme are objectively estimated by RMSE,MAE,and Partial Least Squares Regression(PLS),respectively.Comparing with the traditional and other ensemble mean forecasting,it is found that the new scheme significantly improves the traditional method in the.short term,especially in the 48h,and in the long term the new program has a lower perform than the traditional method.This is consistent with the TC precipitation forecast error growth characteristics.Therefore,this paper establishes a ensemble forecast post-processing method for TC precipitation.Specifically,for ECMWF,a correction method based on RMSE is used to revise the forecast within 144h,and use the traditional ensemble mean forecasting over 144h;for CMA,the forecast within 48h is corrected using the PLS,and the 48h-144h forecast revision is based on the MAE-based correction method,exceeding 144h.The forecast uses the traditional ensemble mean forecasting as the forecast result;for the NCEP,it uses the PLS to correct the forecast results within 240h.