Study On The Improvement Of The DSAEF?LTP Model By Introducing Typhoon Self Factors And Existing Parameters

Tropical cyclones(TCs)are high-impact weather events in China.The precipitation of TCs has caused heavy casualties and economic losses many times.However,the forecast ability for TC-related precipitations still needs further improvement.Ren et al.(2020)proposed a dynamical-statistical-analog ensemble forecast model for landfall tropical cyclones precipitation(DSAEF?LTP model),the forecast ability of the first version of the DSAEF?LTP model close to that of the dynamical models(ECMWF,GFS,T639).However,only two factors of TC track and landfall season are considered in this version of the DSAEF?LTP model,without taking the similarity of other factors affecting TC precipitation into consideration.Besides,two ensemble methods----ensemble mean and maximum value----in this model result in high rate of misses and false alarms,respectively.What's more,though the simulation and prediction experiments in South China are carried out,there is no depth analysis of individual cases.This paper aims at further improving the forecast performance of this model mainly through these three aspects.Firstly,new values of similarity region parameters were added to the DSAEF?LTP model.A simulation experiment was carried out for the case of "LEKIMA".It was found that the forecast performance of the DSAEF?LTP model was improved by adding intensity similarity.Compared with the dynamical models(ECMWF,GFS,GRAPES,SMS-WARMS,RMAPS),the TSsum of the DSAEF?LTP model is only smaller than that of the GFS model.Although DSAEF?LTP does not clearly outperform these dynamical models over every specific region,it does a better job overall in producing reasonable rainfall predictions across the entire domain,especially in southern regions.And it can capture the small range of large precipitation.It was also found that the similarity region of the DSAEF?LTP model is too small,too far east and south.Thus,new values of similarity region parameters were added to the DSAEF?LTP model,which further improved the forecast performance of this model.There are two ensemble methods,mean and maximum,in the DSAEF?LTP model originally.These two methods have high rates of misses and false alarms,respectively.Thus,five new ensemble methods [optimal percentile,fuse,probability matching mean,equal difference-weighted mean,and TSAI(tropical cyclone track Similarity Area Index)-weighted mean] were added to the DSAEF?LTP model.Ten TCs that made landfall over China in 2018 were used to study the simulated accumulated precipitation.Results show that the overall performance of optimal percentile(the 90 th percentile)ensemble method is the best,with the rate of false alarms lower than that of the original ensemble methods.The TC translation speed similarity was also introduced into the DSAEF?LTP model.The TC translation speed is classified by statistics,so as to select the historical TCs whose translation speed is similar to that of the target TC.The results show that K-means clustering algorithm is a reasonable method to classify the minimum TC translation speed in the process.After completing all the improvements of the DSAEF?LTP model,the simulation and forecast experiments for South China were carried out.The results show that each improvement can make better simulation.The best forecast performance appears when the track similarity,new similarity region values and new ensemble methods are considered.Compared with the dynamical models,the DSAEF?LTP model performs better at the precipitation forecast for TCs with typical tracks.The DSAEF?LTP model is more prone to false alarms,while dynamical models are more prone to misses.The structure of TCs and the environmental conditions affecting TC precipitation are not included in this study.But in general,the simulation and prediction ability of the DSAEF?LTP model has been improved.