Using SMS-WARR and rapid refresh technique, the sensitivity of surface observation data (surface conventional observation data, automatic weather station data) to the numerical simulation of the severe convection event which occurred in Shanghai on July 31 2011 was analyzed. Four comparison experiments were designed to study the assimilation of surface observation data. The results showed that the adjustment of model initial fields varies with assimilating different surface observation data. The density of surface observation data has impacts on initial fields. The initial temperature and wind fields are obviously affected by assimilating the automatic weather station data. After assimilating conventional surface observation data and automatic weather station data, we found that the model can not only well simulate the process of this severe convection, but also simulate the surface temperature, wind field and evolution of convergence line which is more consistent with the observation. Through the analysis and comparison of the initial fileds and simulation results, we also found that the assimilation of all the observation data can improve the initial values of the model, and the observation data are integrated into the model by the mode of rapid refresh, which enhances the characterization of mesoscale structure of server convection, and improved simulation capacity of the local severe convective systems.