Abstract: The high-resolution numerical simulations and verifications of 10 convective cases that occurred in Beijing have been conducted by assimilating surface observations in a four-dimensional Variational Doppler Radar Analysis System (VDRAS) based on the rapid-refresh 4D variational assimilation (RR4DVar) technique of multi-radar observations and three-dimensional cloud-scale numerical model. Compared with the surface observations fusion scheme, the verification results show that the surface observations assimilation obviously can improve analysis results below 1 km boundary layer height, and the root mean square errors (RMSE) of simulated wind speed and wind direction are respectively reduced by 0.1 m/s and 7.2° on average, the RMSE of temperature is reduced by 0.2℃. The RMSE of wind speed is decreased by 0.5 m/s at the lowest model level of 100 m, and the error of wind speed increases with height below 3 km. The RMSEs of wind direction and temperature are respectively reduced by 15.5° and 0.4℃ at the lowest model level, and the errors of temperature are decreased at all levels below 1.5 km height. The RMSEs of 10 m surface wind speed and wind direction are respectively reduced by 0.2 m/s and 10.8°, and the error of 2 m surface temperature is also reduced. In addition, the surface observations assimilation can to a certain extent improve the 1 hour forecast of surface fields, whereas the RMSEs of regional surface temperature and wind field increase with forecast time. Combined with the detailed analysis of localized and rapidly intensified convection case that occurred in Beijing on 17 May, 2019, it is found that the surface observations assimilation can better represent the dynamical and thermodynamical characteristics in the lower atmosphere with more details, and thus improves the forecast of low-level meteorological variables. Further investigation of the local trigger mechanism of convection indicates that the interaction between the convergence line of sea breeze and the urban condition to some extent has affected the trigger and development of local convection in Beijing. This method can further improve the nowcasting of localized and rapidly intensified convection in Beijing.