At present,a Regional Ensemble Prediction System has been developed by the Kelamayi Meteorological Bureau.The system has only adopted the initial condition perturbation of the Ensemble Transform Kalman Filter (ETKF),which causes the system lack spread.In order to improve the performance of this Ensemble Prediction System,the model perturbation method of Stochastic Physics Parameterization Tendency (SPPT) is adopted and tested.Firstly,SPPT scientific parameters for the system are determined and set using a sensitivity test on the critical parameters of SPPT.Secondly,an ensemble forecast experiment test based on the Kelamayi Ensemble Prediction System is conducted to compare the ETKF initial perturbation scheme with the combination of the ETKF initial perturbation and SPPT model perturbation (ETKF-SPPT) scheme.The results show that the ETKF method provides initial condition perturbations with dynamic structure,while the spread saturates within a short forecast lead time and decreases due to the constraint of identical Lateral Boundary Condition (LBC) for all members.Adopting the SPPT model perturbation method significantly improves the ensemble spread for all forecast lead times.Based on the ensemble verification,adding model perturbation to initial perturbation condition improves the probabilistic forecast skill with a higher forecast reliability and a smaller RMSE.Without model perturbation,ETKF’s reliability is low and its root mean square error (RMSE) is relatively large.Additionally investigated and examined is a local gale case that occurred in Kelamayi during the experimental period.Its results show that employing model perturbation significantly improves the ensemble forecast,resulting in a more accurate forecast of the local gale’s magnitude and duration.