摘要:以新一代全球/区域多尺度通用同化与数值预报系统−热带气旋路径数值预报系统(global/regional assimilation and prediction system-tropical cyclone model, GRAPES-TCM)为试验模式,采用组合不同的物理参数化方案(MP)方法和随机全倾向扰动(STTP)方法,生成反映模式不确定性的集合成员,在此基础上设计包含6个成员的3种集合方案,方案1和方案3的成员分别用MP方法和STTP方法生成,方案2的成员同时采用MP和STTP方法生成,用3种集合方案对1109台风“梅花”进行了36次72h的集合预报试验。结果显示:对于路径预报,3种集合方案中预报效果最好的是方案3,其次为方案2,最差的是方案1;对于强度预报,方案1和方案2的预报效果差异不大,都远好于方案3。方案2和方案3的路径预报与强度预报都好于控制试验的预报,方案1的路径预报好于大部分成员的预报,强度预报好于所有成员的预报。3种方案的路径离散度都偏小,方案3偏小最多,其次为方案2;方案3的强度离散度也过于偏小,是3种方案中最小的,方案1和方案2的强度离散度在积分前期明显偏小,积分后期则有偏大的趋势,其中方案2的强度离散度大于方案1。与国内外8个业务数值模式的预报结果比较,对于路径预报,方案1优于5个业务模式的预报,方案2和方案3则优于除欧洲数值以外的7个业务模式的预报;对于强度预报,方案1和方案2优于所有8个业务模式的预报,方案3优于6个业务模式的预报。总体而言,3种集合方案的路径和强度预报都表现出优于确定性预报的预报能力,相对于各业务数值模式都表现出一定的预报优势,具有业务应用的价值,其中同时应用STTP和MP方法的方案2对台风的综合预报效果是最优的。%The GRAPES-TCM (global/regional assimilation and prediction system-tropical cyclone model) is used to make ensemble prediction experiments for typhoon Muifa (1109) in 2011. Three kinds of ensemble schemes are designed for the experiments. Every scheme has six ensemble members, which reflect the uncertainty of the model. The method of multiple physics (MP) is used to form the members of scheme 1. The method of stochastic total tendency perturbation (STTP) is used to form the members of scheme 3. Both the MP method and the STTP method are used to form the members of scheme 2. Thirty-six experiments are made and the integration time is 72 h. The experiment results are as follows. In the three ensemble schemes, the track prediction of scheme 3 is the best, that of scheme 2 is the second, and that of scheme 1 is the worst. The intensity prediction of scheme 1 is close to that of scheme 2. They are both much better than that of scheme 3. The track and intensity predictions of scheme 2 and scheme 3 are better than those of their control experiments. The track prediction of scheme 1 is better than that of its most members. The intensity prediction of scheme 1 is better than that of all of its members. The track dispersions of the three schemes are all small. In the three schemes, the track dispersion of scheme 3 is the smallest and that of scheme 2 is the second. They are both very small. The intensity dispersion of scheme 3 is also too small and is the smallest in the three schemes. In the early integration, the intensity dispersions of scheme 1 and scheme 2 are obviously small. In the late part of the integration, they are overall a little large. The intensity dispersion of scheme 2 is larger than that of scheme 1. Compared with the predictions of the eight domestic and abroad operational numerical weather prediction (NWP) models, the track prediction of scheme 1 is better than that of five operational models, and the track predictions of scheme 2 and scheme 3 are better than that of seven operational models (not including the ECMWF). The intensity predictions of scheme 1 and scheme 2 are better than that of all eight operational models, and the intensity prediction of scheme 3 is better than that of six operational models. Overall, the track and intensity predictions of the three ensemble schemes all outperform those of the deterministic prediction. They all show certain superiority to the predictions of the eight operational models. The three ensemble schemes all have the potential of operational application. Considering the composite prediction effect, scheme 2 based on both MP and STTP methods is the best among the three schemes.