Abstract: In this study, we developed a regional ensemble prediction system Global and Regional Assimilation and Prediction Enhanced System-Regional Ensemble Prediction System (GRAPES-REPS V3.0) with a horizontal resolution of 10 km, which was put into operation by the Numerical Weather Prediction Center of the China Meteorological Administration in 2019. The background field of the GRAPES-REPS V3.0 is the GRAPES global model. The initial perturbation and model perturbation methods used are the ensemble transform Kalman filter and the stochastic perturbed parameterization tendencies, respectively. The system was run in real time for the 2019 summer season (July through September) and its results were compared with those obtained by the GRAPES-REPS V2.0 and ECMWF global ensemble prediction systems by statistical verification and case analysis to objectively and comprehensively evaluate its precipitation forecast skill and forecast uncertainties. We also analyzed the physical mechanism responsible for the forecast uncertainties regarding meso-scale intense precipitation. The study results, which provide a basis for diagnosing regional ensemble prediction systems and developing ensemble forecast methods, can be summarized as follows: (1) GRAPES-REPS V3.0 model obtained better precipitation Equitable Threat Scores (ETS) than the GRAPES-REPS V2.0 model in terms of the forecast lead times and rainfall classes with more equal rainfall members. The probability forecast fraction skill scores were also better. As such, the precipitation forecast skills of the GRAPES-REPS V3.0 model are better overall than those of the GRAPES-REPS V2.0 model. (2) The ensemble mean precipitation bias and ETS scores of the GRAPES-REPS V3.0 for light rain and rainstorm are better than those obtained by the ECMWF global ensemble forecast system, and the probability forecast skill of the two models is comparable. (3) The case studies show that different ensemble prediction systems capture precipitation forecast uncertainties by describing the uncertainties of meso-scale physical quantities. At an initial lead time, the circulation patterns of the GRAPES-REPS V3.0 regional ensemble prediction system and the ECMWF global ensemble prediction system are similar. However, with the evolution of the forecast lead time, the GRAPES-REPS V3.0 ensemble prediction model better describes the meso-scale dynamic and thermal fields, with more accurate rainfall locations and magnitudes and a better probabilistic forecast result. (4) Compared with the ECMWF model, the ensemble members of the GRAPES-REPS V3.0 model can effectively forecast the occurrence, development, and extinction of rainfall processes. As such, the GRAPES-REPS V3.0 model shows higher skill in forecasting precipitation during the Chinese flood season.