Evaluation And Correction Of Typhoon Intensity Forecast By Global Ensemble Forecast System

In the 2010 s,the development of the global ensemble forecast system(EPS)has brought about a simultaneous improvement in the forecast accuracy and uncertainty of tropical cyclone(TC)track forecast.However,the application of EPS in TC intensity prediction is still very limited.The main reason is that we still lack a sufficient understanding of EPS's TC intensity forecasting capabilities,and we have not established an effective EPS interpretation and application method for forecasters' reference.In response to this situation,this article selects five global EPS TC intensity forecast products provided by the THORPEX Interactive Grand Global Ensemble(TIGGE)database,and conducts a five-year(2015-2019)evaluation from two aspects: ensemble mean and probability forecast.On this basis,the European Centre for Medium-Range Weather Forecasts EPS(ECMWF-EPS)with the best comprehensive performance was selected to analyze the possible influence factors of TC intensity forecast errors.On this basis,the linear stepwise regression,neural network method and translation method are combined to develop a new TC intensity forecast correction scheme.Notable error features include the underestimation of the TC intensity by ensemble mean forecast and the under-dispersion of the probability forecasts.Improvements of 22.4?63.9% and 17.2?35.0% were obtained for the root mean square errors(RMSE)and Brier Scores(BS)respectively for different lead times up to 168 h in the five years.Positive forecast skill appeared in the most recent two years(2018-2019)at 120 h or later as compared with the climatology forecasts.However,there is no obvious improvement for the intensity change forecasts during the 5-year period,with abrupt intensity change remaining a big challenge.The probability forecasts shows no skill either for strong TCs at all the lead times.Among the five EPSs,ECMWF-EPS ranks the best for the intensity forecast,while National Centers for Environmental Prediction Global Ensemble Forecast System(NCEP-GEFS)ranks the best for the intensity change forecast,according to the evaluation for ensemble mean.As for the probability forecast evaluation,ECMWF-EPS ranks the best at lead times shorter than 72 h,while NCEP-GEFS ranks the best later on.Based on the above evaluation results,ECMWF-EPS is selected as a representative,and the possible relationship between the ensemble mean intensity forecast error and the characteristics of TC and the environmental field are analyzed,in order to gain a deeper understanding of the characteristics of ECMWF-EPS TC intensity prediction and lay the foundation for its interpretation and application.The results show that the relative error of the ensemble mean intensity forecast has obvious regional characteristics.When the TC is located at a higher latitude(greater than 30°N),the forecast tends to be stronger,and when the TC is located at a lower latitude(less than 25°N),the forecast tends to be weaker.The TC intensity and the relative(absolute)error are significantly negatively(positive)correlated at shorter lead times,while the long-term lead time is the opposite;ECMWF-EPS tends to be stronger for the initial weaker TC at shorter lead times;the initial TC intensity relative error has significant persistence characteristics in the short lead time effect;when the initial environmental pressure is large,the absolute error of the TC intensity forecast is usually large;when the maximum possible intensity is strong,the forecast TC intensity tends to be stronger;vertical wind shear and intensity forecast errors are only significantly correlated at shorter lead times;the greater the dispersion of ensemble forecasts,the weaker TC intensity of ensemble mean forecasts,and the greater the absolute error,and vice verse.On the basis of correlation factor analysis,the stepwise regression method was used to select factor to establish a deep neural network(DNN)correction model of the TC intensity forecast deviation.Assuming that the ensemble dispersion is constant,the established deviation correction model and translation method are combined to propose an improved scheme for ECMWF-EPS TC intensity probability prediction.Independent sample(2020)evaluation shows that the improvement rate of the intensity probability forecast is 21.1%(24 h)and 16.6%(48 h).Therefore,the new TC intensity correction scheme established by this research has a good reference value for improving the 24 and 48 h intensity forecasting capabilities.