Abstract: There are great uncertainties in mesoscale precipitation model forecasting. To better describe the uncertainties of key parameters in the physical processes that are closely related to model precipitation forecasting,a stochastically perturbed parameterizations (SPP) scheme consisting of temporally and spatially varying perturbations of 18 key parameters in the cumulus convection,microphysics,boundary layer and surface layer parameterization schemes has been developed in the Global and Regional Assimilation and Prediction Enhanced System-Regional Ensemble Prediction System (GRAPES-REPS) of China Meteorological Administration. Sensitivities of parameter perturbations in different physical processes,spatial and temporal decorrelation scales,as well as energy evolution characteristics and ensemble prediction performance are analyzed by conducting sensitivity experiments for 10 summer days in June and July 2015. The main conclusions are as follows. Almost all the SPP experiments with parameter perturbations in selected physical parameterization schemes exhibit higher probabilistic forecasting skill and the results are better than that without SPP in the forecast of precipitation and other variables on isobaric surfaces. It is also found that perturbing parameters in the cumulus and boundary layer schemes has more significant impacts than perturbing parameters in the microphysics scheme. Furthermore,simultaneously perturbing parameters in the cumulus,microphysics,boundary layer and surface layer parameterization schemes can yield better ensemble prediction than perturbing some of the parameters in any single parameterization scheme. This result indicates that the SPP can effectively improve the probabilistic forecasting skill for mesoscale precipitation. The characteristics of energy evolution indicate that perturbing parameters in different physical processes affects energy at different levels and with different features,but overall the SPP scheme has little influence on the internal energy,kinetic energy and total energy of the atmosphere,and the energy before and after the perturbation is basically the same. By conducting sensitivity experiments on spatial and temporal decorrelation scales of random patterns,it is found that the choice of spatial and temporal decorrelation scale of random patterns has great impact on the ensemble prediction. The optimal ensemble prediction can be obtained when choosing the temporal decorrelation coefficient of 12 h and the truncated wave number of 20 in the stochastic perturbation field. In conclusion,the SPP scheme can not only effectively improve the performance of ensemble probability prediction,but also improve the ensemble prediction skill of precipitation forecasting. Therefore,it has a promising prospect for operational application and development.