Based on the de-correlation length theory of background error, the impact of the adjustment of background error horizontal resolution on 3-Dimensional Variational Assimilation (3D-VAR) of Dopplar radar data was studied. Firstly, the NMC method is used to gain the background error covariance of different horizontal resolution in torrential rain area. The structure characteristics of two background errors of different resolution and the impact of horizontal resolution on de-correlation length of background error covariance are both analyzed. Then it is applied into radar data assimilation in order to analyze the impact of background error horizontal resolution on the radar assimilation. Results show that when the horizontal resolution of the background error resolution increases from 27 km to 3 km, the lower atmosphere can reflect more meticulous information of dynamic field and the de-correlation length of dynamic field would decrease by secondary roots of the horizontal resolution, while the de-correlation of temperature field and water vapor field has no significant change. When applying them to 3D-VAR, the background with higher resolution would reflect more mesoscale information in the analysis field increments and significantly improve radar radial velocity data assimilation effect and the precipitation forecast effect in torrential rain simulation.