<p>The multi-physics ensemble method is a widely used method to represent the prediction uncertainty arising from model errors and has great potential to improve precipitation forecasts. This study performs multi-physics ensemble simulations of ten heavy precipitation cases in South Korea and evaluates the performance of the ensemble mean. The multi-physics ensemble is generated from 27 different combinations of cloud microphysics, planetary boundary layer, and radiation schemes. In the prediction of 24-h accumulated precipitation amount, the ensemble is underdispersive for most cases, indicating that the prediction uncertainty is only partly represented by the ensemble. The overall performance of the ensemble mean is better than that of any individual ensemble member. No individual ensemble member consistently shows good performance for every case. The root-mean-square error (RMSE) of the ensemble-mean prediction is smaller than the average RMSE of individual ensemble members for every case. The relative difference in RMSE exhibits a strong positive correlation with the spread-error ratio. In comparison with subset ensembles, the total ensemble shows the most stable performance. Among the three types of physics parameterization in the multi-physics suite, the cloud microphysics parameterization contributes the most and the radiation parameterization contributes the least to the ensemble spread and the ensemble-mean performance. Each physics parameterization scheme has tendencies to predict cloud and precipitation properties to be larger or smaller than those predicted by other schemes, which stem from differences in the choices of individual process parameterizations and the physics parameter values used. This systematic difference contributes to the ensemble spread of the multi-physics ensemble, which is a key factor for the ensemble-mean performance.</p>