Abstract: Microphysical variables and vertical velocity retrieved were incorporated using the nudging method into the initial data assimilation of GRAPES (Global/Regional Assimilation and Prediction System) model at each time step of the integration by adding an extra term to the prognostic equation. Simulation experiments of a torrential rain event were carried out using the Doppler radar observations of Hefei, Anhui province at 02:00 BST 5 July 2003 and the GRAPES model developed by CAMS. Different experiments were designed to investigate the effects of water vapor, liquid water and vertical velocity on Nudging and prediction results. Some conclusions were drawn as follows: Nudging technique was effective in forcing the model forecast gradually to the observation information, yielding the thermodynamically and dynamically balanced analysis field, and correspndingly, the spin-up phenomenon of the model has been to some extent removed. As viewed from the simulation results, water vapor is vital to precipitation, and a governing factor of the amount and duration of precipitation; the initial cloud water, rain water and vertical velocity determine the strength distribution of convection and precipitation at the beginning time of forecast; the horizontal wind field steers the motion of the mesoscale weather system embedded in and impacts the position of precipitation zone to a large extent. The model successfully forecasted the precipitation process within 6 hours, and the distinct characteristics of the mesoscale weather system, such as the updraft velocity in the center of convective structure is about 2.0 m/s, the maximum cloud water content of about 1.5 g/kg occurs above the level of 400 hPa and the maximum rain water content of about 3.0 g/kg below 400 hPa, and there is the notable convergence of air flow at the low level of the convective structure and the distinctive divergence in the upper level. The simulation experiments show that the influence of the initial retrieval data on prediction is weakening with the increasing of forecast time, and within the first hour of forecast, the retrieval data have an important impact on the evolution of the weather system, but its influence becomes trivial after the first 3 hours.Herein only one radar reflectivity was used, the radar observations did not cover the whole model domain, and some empirical parameters were used in the retrieval method, therefore some difference still lies between simulation and observation to certain extent, further studies on several aspects are demanded.