In order to improve the weather forecast quality over the low latitude plateau regions, the wind data retrieved with VAD (Velocity Azimuth Display) method are assimilated to the WRF (Weather Research and Forecasting) model by WRF 3DVar (3 Dimensional Variational Data Assimilation System). With different assimilation schemes, a torrential rain event occurred in Yunnan Province from 00:00UTC 30 June 2009 to 00:00UTC 1 July 2009 is numerically simulated and comparatively analyzed. The results indicate that the initial wind fields of the WRF model are markedly improved by assimilating the retrieved wind data. The WRF 3DVar can availably introduce the information of the retrieved wind to the initial conditions of the regional numerical model. The assimilation of the retrieved wind data helps enhance the wind convergence and vapor transportation over the rainy area. Furthermore, the assimilation help improve quantitative precipitation forecasts. The quantitative test of the 18 hour rainfall forecast shows that forecasts are more accurate, less pretermissions, and more rational pertinence for over 250 mm precipitation in the assimilation experimentations. The higher the assimilation frequency and the longer the assimilation time is, the more obvious the influence of data assimilation on the initial fields and forecast fields of the regional model is. But long assimilation time may increase the speed of synoptic systems and the overestimate rainfall, and so the suitable selection of frequency and time is crucial in numerical experimentations.