Precipitation Forecast And Verification In Guangdong-Hong Kong-Macao Great Bay Area Based On Multi-model Ensemble

The Guangdong-Hong Kong-Macao Greater Bay Area has a humid subtropical marine climate.This area is abundant with moisture for most of the time during the year.Precipitation frequently occurs during the flooding season every year,sometimes accompanied by thunderstorms.Frequent heavy rainfall has caused great inconvenience to people's routine travel.At the same time,the safety of local residents' lives and properties are threatened somehow.Timely and accurate forecast of heavy rainfall is not only a challenge facing the whole world,but also an urgent problem to be solved to guarantee a safe daily journey and life of the residents in the Greater Bay Area.This study applies the numerical prediction model outputs from the European Centre for Medium-range Weather Forecasts Model(ECMWF)with a resolution of 0.1°,the Japanese Meteorological Agency Model(JMA)with a resolution of 0.5°,and the Global/Regional Assimilation and Prediction System(GRAPES)with a resolution of 0.09°.This study considered the characteristics of the numerical prediction models and explored multi-model ensemble methods to forecast the rainfall in Guangdong province.The multi-model ensemble methods are usually better than a single numerical prediction model as they can avoid the problem of prediction uncertainty,which may lead to forecasting bias.The hourly observation data from approximately 2,500 automatic weather stations in Guangdong Province were used to validate the forecast,and the Guangdong-Hong Kong-Macao Greater Bay Area was selected as the verification area to evaluate the models' forecast performance of the future 24-hour accumulated rainfall during the flooding season in 2018.The traditional threat score(TS),equal threat score(ETS),and fraction skill score(FSS)were used to assess the models' rainfall forecasting performance.The ensemble model of the Probability Matched Ensemble Mean method(PME)was built based on ECMWF,JMA,and GRAPES.Based on the prediction characteristics of the three model members,the PME is further improved by using the step-by-step optimization method to form the optimized integrated PME(OPT_PME)with better prediction results.Results showed that all models perform well in forecasting light rain.Based on the TS score of rainfall forecast for all rainfall levels(light,moderate,heavy and torrential rainfall),the average rainfall forecasting ability for the multi-model superensemble method of PME has been improved by 1.8%,4.7%,13.7%,comparing to ECMWF,JMA and GRAPES,respectively;the multi-model superensemble method of OPT_PME's average rainfall forecasting ability has been improved by 3.2%,6.1%,15.3%,comparing to the three original models,respectively.In addition,based on the FSS score,the average rainfall forecasting ability of PME has been improved by 3.4%,11.6%,7%,respectively,comparing to the three original models;the average rainfall forecasting ability of OPT_PME has been improved by 2.9%,11.0%,6.5%,respectively,comparing to the three original models of ECMWF,JMA and GRAPES.