Abstract:A few operational models predicted the Beijing “7.21” (21 July 2012) severe rainstorm with the wrong reason that the severe convective rainstorm resulted mainly from cold front. In fact, a large amount of rainfall occurred over the warm area of the southwest of Beijing before the cold front. In this paper, after assimilating the surface and sounding conventional data three times by ensemble Kalman filter (EnKF) method, a non hydrostatic mesoscale numerical model (WRF) that has a collection of 30 members was employed to simulate the process. The comparison of a better member and a worse member reveals that the better member can successfully simulate the mesoscale convective system (MCS) in the warm sector in the central and western part of Hebei Province and the southwest of Beijing and the relatively stable system configuration, thus making the MCS fully develop over Beijing. So, it does a better simulation of the warm sector precipitation during the process of the severe rainstorm. In contrast, the worse member fails to simulate the precipitation in warm sector, and turns out to be frontal precipitation with the rain band southerly and the occurrence time lagged. The difference in these simulated results is greatly related to the corresponding simulated location of the low vortex in each ensemble member. The improvement in initial field of ensemble member by EnKF, which makes the simulated movement and development of main influence systems more accurate, is therefore the key to simulate the trigger and maintenance of the convective system in warm sector successfully.