中文摘要：

采用最新发布的区域气候模式RegCM3．1，分别引入Grell-AS、Grell-FC、Kuo和MIT-Emanuel 4种积云对流参数化方案，对1991年5～7月江淮地区暴雨进行模拟试验。从各月降水量的模拟来看，Grell—AS方案较好地模拟出了江淮暴雨的强度，尤其是对7月降水的强度和位置的模拟与观测非常接近；Grell—FC方案能较好地反映7月江淮暴雨的强度和位置，但对华南地区降水模拟一直显著偏强；Kuo方案能模拟大尺度降水的情况，但对强对流性降水模拟偏弱；MIT-Emanuel方案较好地反映了5月降水的空间形势，但各月降水的模拟都比观测强。从各区月平均降水的对比发现，Grell-AS和Kuo方案的模拟要优于Grell-FC和MIT-Emanuel方案。从降水的南北变动来看，Grell-AS方案较好地刻画了江淮地区雨带的强度和南北变动。高低空环流形势和整层水汽通量的分析表明，造成MIT-Emanuel方案降水模拟偏强的主要原因与水汽输送偏强有关。对4种积云对流参数化方案进行集合，其结果表明，物理过程集合方法能有效地减小物理过程参数化的不确定性对模拟结果的影响。

英文摘要：

By using Grell-AS, Grell-FC, Kuo and MIT-Emanuel cumulus convection parameterization schemes, a new version of the regional climate model RegCM3.1 has been used to simulate heavy rains from May to July 1991 over the Yangtze-Huaihe valley. In the simulation of monthly rainfall, Grell-AS scheme well simulated the intensity of heavy rains over the Yangtze-Huaihe valley, especially the intensity and spatial distribution in July. Grell-FC scheme well simulated the spatial distribution and intensity of heavy rains over the Yangtze-Huaihe valley in July, however it always obviously overestimated the rainfall over east and south China. Kuo scheme was able to simulate the large-scale rainfall, but it often underestimated the convectional heavy rains. MIT-Emanuel scheme well simulated the spatial distribution in May, but the simulated monthly rainfalls were often heavier than the observations. The schemes of Grell-AS and Kuo exhibited better performance than the schemes of Grell-FC and MIT-Emanuel in the simulation of monthly mean precipitation in various subdivisions. The scheme of Grell-AS well simulated the in tensity and variations of rain belts over the Yangtze-Huaihe valley. The analysis of high and low level wind and water vapor flux in whole levels indicated that the stronger simulation of water vapor flux caused the overestimated rainfall in MIT-Emanuel scheme. The ensemble was created by choosing four kinds of cumulus convection parameterization schemes, the results indicated that the physical process ensemble method could effectively reduce the simulated errors from the physical process parameterization uncertainty.