The fundamental goal of ensemble forecasting(EF) is to estimate a forecast probability density function(PDF) of possible future states of the atmosphere from which the true state is consistently a random sample. The skills of EF is superior to the deterministic-style NWP(numerical weather prediction) where a single run is considered in long-/meso-range. However, the skills of EF in the storm-scale is inscient. So an ideal storm case is simulated by the WRF(weather research and forecasting) model V2.2 on the supercell severe local storm, occurred in central Oklahoma on 20 May 1977 to study the feasibility whether EF can be applied to the forecast for the storm-scale system. Simultaneously the influence of the different ensemble sizes on the skills of the ensemble is explored through experiments with the addition of the random initial errors to different areas. It is found that the ensemble average can work in gaining higher forecast skills and protracting the effective forecasting. Although it enhances with the increase of ensemble sizes as a whole, the improvement is saturated when the ensemble sizes comes to 5-13.