Abstract: The information of wind fields is an important parameter in weather forecast. At present, a great deal of information from CINRAD is not really applied to numerical weather prediction (NWP), domestic studies on the application of wind information from Doppler radar in NWP are not many. In order to apply observational wind data of Doppler radar to numerical weather prediction models better, wind profiles are retrieved from single Doppler radar observation by using VAD method (Velocity Azimuth Display), the VAD method has developed for thirty years and now becomes mature. Further more the wind profile data are processed as the normal form of radiosonde observation to make up the deficiency of radiosonde observation in space-time resolution. And then the retrieved wind data are assimilated by the 3D-VAR system of GRAPES-Meso (Global and Regional Assimilation and PrEdiction System-Meso) model. With different initial fields a torrential rain occurring in the Changjiang River valley from 1200UTC on June 27, 2005 to 1200UTC on June 28, 2005 is numerically simulated and comparatively analyzed. The results indicate that the retrieved wind profile corresponds to the radiosonde observation profile very closely and accurately reflects the status and location of the wind shears at low levels. The retrieved wind profile not only has higher space-time resolution but also remedies the lack of radiosonde observation in this case. It reflects that using VAD method to retrieve wind profile from single Doppler radar is viable and of advantages. With the assimilation of the retrieved wind profile the initial wind fields of numerical experiments are remarkably improved. In the improved initial wind fields vapor transportation becomes stronger where the observed torrential rain occurs. Moreover, the intensity and areas of forecasted precipitation are improved in some ways. As a whole the amendment of the 6-hour precipitation forecast is better than that of the 24-hour. With assimilated initial fields the numerical experiments successfully simulate a rainfall center of 10—25 mm in the 6-hour forecast, which indicates that using VAD method to retrieve wind profile from single Doppler radar has a prospective future in NWP once more. On the other hand, the contrast analysis of the numerical results indicates that a difference of assimilation frequency and assimilation time window will have a different influence on a short time torrential rain forecast in the numerical assimilation experiments. In this case, the 6-hour rain forecast by assimilating the retrieved wind profile once an hour is better than the forecasts once three hours or once six hours; the 6-hour rain forecast by assimilating the retrieved wind profile for three hours is better than the forecast assimilating for six hours. That is to say, a higher assimilation frequency of retrieved wind data will lead to a better forecast, but a longer assimilation time not always leads to a better simulated result in NWP, which has a valuable guidance effect on the assimilation operation of wind data from Doppler radar.