The ensemble forecasting of tropical cyclone intensity based on EnKF data assimilation and biascorrection

The techniques of ensemble forecasting based on EnKF(Ensemble Kalman Filter) and biascorrection are applied to predict the tropical cyclone intensity by using MM5 model. Adopting the AnthesKuo, Grell, and BettsMiller cumulus parameterization schemes, and highresolution Blackadar, BurkThompson, and MRF PBL process parameterization schemes, nine groups model configuration are designed, and45 , 60 and 75min forecasts are conducted for each situation. With the “mirror imaging method”, 18 different initial conditions are obtained. Taking the “Rankine vortex” as observational data and the 18 different initial conditions as the background ensemble, the EnKF data assimilation with EnSRF arithmetic are then carried out. Utilizing the 18 data assimilation results as the ensemble forecasting initial fields and taking 9 different model configurations, 72 h forecast is carried out. Two experiments are designed: one is the ensemble forecasting based on EnKF data assimilation, which is used to compare with nonassimilation ensemble forecasting, and 6 typhoon cases occurred in 2004 are selected. The other uses the biascorrection method to modify the ensemble forecasting results, in which 16 typhoon cases occurred in 2003 and 2004 are selected. So we can discuss the impact of bias correction. The results of the first experiment show that because the MM5 model has deficiencies in intensity forecast, the ensemble forecasting system with data assimilation has made little improvement compared with nonassimilation. The results of the second experiment show that, as can be seen from the RSS index, the biascorrected intensity forecast makes great improvement compared with that of uncorrected. On the average, the absolute error of intensity forecast is reduced obviously than that of control prediction in every forecast time. The error for 48 and 72h are 18 hPa， and 11 hPa respectively, which minishes 5 hPa forecast error at 48h. By bias correction, the error of intensity ensemble forecast sample mean is reduced, and the PDF mean is closer to the theoretical value. Therefore, the quality of the intensity ensemble forecast results, particularly forecast probability, is increased. So the sample variance is reduced and a more reasonable intensity probability forecasting is obtained.