Abstract: In this study, three model perturbation schemes, namely Stochastically Perturbed Parameter scheme (SPP), Stochastically Perturbed Physics Tendencies (SPPT), and Multi-Physics process parameterization (MP), were employed to represent the model errors in the REPS (Regional Ensemble Prediction Systems). To investigate the influences of the various model perturbation schemes on typhoon forecasting, three sensitive experiments using three different combinations (EXP1: MP, EXP2: SPPT+SPP, and EXP3: MP+SPPT+SPP) of the model perturbation schemes were established based on the Weather Research and Forecasting (WRF) V4.2 model for Typhoon Kompasu (2021). The results reveal that for typhoon forecasting, ensemble forecasting experiments could simulate the process of typhoon intensification and the typhoon path, and EXP3 had the best simulation results. The path bias of EXP3 was the smallest of the three ensemble forecast experiments, with an average value of 52.8 km, while those of CTRL, EXP1, and EXP2 were 61.8, 54.4, and 65.7 km, respectively. Additionally, the perturbation energies of the three sets of ensemble prediction experiments were greater than that of the CTRL. The perturbation energy of EXP3 developed the fastest and was the largest. Moreover, the Brier scores of the three sets of experiments enhanced the forecast results compared to the CTRL. Furthermore, those of EXP3 were the most improved of the three sets of experiments, with EXP1 and EXP2 demonstrating improvements of 45% and 48.76% relative to the CTRL, while EXP3 could reach 70%. The forecasts of EXP2 and EXP1 were comparable, and EXP3 had improved the forecasts compared with EXP1 and EXP2, with improvement rates of 57.5% and 40%, respectively.