基于集合Kalman滤波数据同化的热带气旋路径集合预报研究

The Ensemble Forecasting of Tropical Cyclone Track Based on Ensemble Kalman Filter Data Assimilation

构建了一个基于集合Kalman滤波数据同化的热带气旋集合预报系统,通过积云参数化方案和边界层参数化方案的9个不同组合,采用MM5模式进行了不同时间的短时预报。对预报结果使用“镜像法”得到18个初始成员,为同化提供初始背景集合。将人造台风作为观测场,同化后的结果作为集合预报的初值,通过不同参数组合的MM5模式进行集合预报。对2003-2004年16个台风个例的分析表明,初始成员产生方法能够对热带气旋的要素场、中心强度和位置进行合理扰动。同化结果使台风强度得到加强,结构更接近实际。基于同化的集合路径预报结果要优于未同化的集合预报。使用“镜像法”增加集合成员提高了预报准确度,路径预报误差在48小时和72小时分别低于200 km和250 km。

The technique of ensemble forecasting based on Ensemble Kalman Filter （EnKF） data assimilation is applied to the problem of tropical cyclone track prediction using MM5 model. Adopting the Anthes-kuo, Grell and Betts-Miller cumulus parameterization schemes, High-resolution Blackadar, Burk-Thompson and MRF PBL process parameterization schemes to design 9 groups model configuration, 45-, 60- and 75-minute forecasts are conducted for each situation. With the ＂mirror imaging method＂, 18 different initial conditions are obtained. Taking the ＂Rankine vortex＂ as observation 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 with 9 different model configuration, 72-hour forecast is simulated. Two experiments are designed. One is non-assimilation ensemble forecasting, in which bogus typhoon is directly joined and 6 typhoon cases in 2004 are selected. The other is the ensemble forecasting based on EnKF data assimilation, in which 16 typhoon cases in 2003 and 2004 are selected. There are three methods, full ensemble average, cluster average and select average, in ensemble average. The results show that because of no adjoint processing, EnKF data assimilation method is more efficient than that of 4D-VAR, and with the assimilation, the intensity of typhoon becomes stronger and its central position is corrected. By disturbing the background fields, the disturbances of the initial position, intensity and structure of typhoon occur. The results of the first experiment show that the error of the typhoon position with data assimilation is less than non-assimilation, especially is the cases of number 0419 typhoon. The results of the second experiment show that the position error of clustering average method is the smallest among the three methods. On the average, the error for 24-hour prediction is within 130 km, the errors for 48-hour and 72-hour predictions are less than 200 km and 250 km respectively. In every forecast time, there are at least more than half cases whose errors are less than that of control prediction. The results indicate that the accuracy of the forecast position is much more improved by using ensemble forecast method based on EnKF than that of control forecast in the cases of recurved track type and the cases with weaker intensity. One application of ensemble forecasting is to estimate the probability distribution of weather phenomenon. As for tropical cyclone, the most important thing the authors concern is that what the observational typhoon track is within the range of the forecast tracks. Since the moving route of tropical cyclone is a two-dimensional problem with direction and speed, any analysis of its ensemble forecasting results should include two aspects. The results of the statistical analysis of 16 typhoon cases show that for every forecast time, more than 65% cases include moving direction and all cases at the 54th hour. For moving speed, it is more than 50% cases that include observational values at every forecast time and more than 90% cases at middle prediction time. The prediction of directions and speeds is successful on the whole. The experiments show that ensemble forecast method can improve the forecast veracity of the tropical cyclone track, and the method suggests in this paper based on EnKF data assimilation is an effective approach.