Based on WRF v3.9 model, the stochastically perturbed parameterization is used to perturb MYNN boundary layer and RUC land surface process scheme parameters to simulate a heavy rainstorm in southwestern mountains of China. The optimal settings for parameter perturbations of MYNN boundary layer and RUC land surface process in mountain rainstorm ensemble forecast are explored. The main conclusions are as follows. In the random disturbance MYNN boundary layer scheme (SPPM) and RUC land surface process scheme (SPPR) parameters, the disturbance is mainly for the variables at the surface and the lower level of the model. The disturbance energy gradually develops from lower levels to higher levels in the model, and the SPPM can get greater disturbance energy than SPPR. The SPPM scheme is more sensitive to the variation of temporal correlation parameters than the spatial correlation parameters. However, as the perturbation energy of SPPR scheme is generally small, the variations of spatiotemporal correlation parameters have relatively small influence on its ensemble prediction performance. In SPPM scheme, a better ensemble prediction skill can be obtained by the temporal correlation selection for 6 h and the spatial scale selection for 70 km, while in SPPR scheme, a better ensemble prediction skill can be obtained by the temporal correlation for 6 h and the spatial scale for 50 km.