Abstract: The local analysis and prediction system (LAPS) is widely applied, providing fast comprehensive analysis products for forecasters and assimilating various observational information. GRAPES_Meso is a new generation of global/regional assimilation and prediction system, developed by Chinese Academy of Meteorological Sciences, and has been used in operation and research. However, LAPS itself does not provide any interface system with GRAPES_Meso.Based on the investigation of LAPS and GRAPES_Meso, the output data of LAPS are used as the initial condition of GRAPES_Meso model (GRAPES-LAPS), and compared with the original initial scheme of GRAPES_Meso model (GRAPES-3DVAR). Twenty-eight rainfall events in Southern China from 2008 to 2010 are simulated with the two initial schemes, which both assimilate data including 14 Doppler radar observations, 30 radiosonde observations and about 530 surface observations.The results show that the initial fields obtained by the GRAPES-LAPS scheme accurately represent the moisture fields and meso-scale environmental circulations. The simulated moisture and convergence at low level by the GRAPES-LAPS scheme are stronger than the results of the GRAPES-3DVAR scheme, and these initial fields are beneficial to improve the GRAPES_Meso model simulation. The mean square error (MSE) and threshold scores (TS) of two schemes of the twenty-eight rainfall events are equivalent, but ten rainstorm events are accurately simulated by using GRAPES-LAPS scheme and only five rainstorm events are accurately simulated by GRAPES-3DVAR scheme. The TS scores of moderate rain, heavy rain and rainstorm by the GRAPES-LAPS scheme are a little better than those of the GRAPES-3DVAR scheme, indicating that the GRAPES-LAPS scheme is more suitable to forecast strong rainfall events than the GRAPES-3DVAR scheme.