Abstract: To improve the cloud initialization, a convective-scale velocity determined by surface sensible and latent heat fluxes is introduced into the scheme as a criterion of convection to estimate cloud ice and cloud water for stratiform and convective clouds separately. Using the second generation of Shanghai Meteorological Service-WRF ADAS Rapid Refresh System, the improved cloud analysis method was tested in a severe convective case occurred on 28 April 2015 in East China with a focus on the influence of the analysis method on cloud initialization and forecast results. The results show that about 40% of the total cloud analysis grids were determined to be non-convective grids based on the convection criterion. The distribution of convective grids was similar to that of upward sensible heat flux, which displayed a diurnal variation over the land. With separate stratiform and convective grids as well as different cloud water and ice schemes, the mixing ratios of cloud ice and cloud water were significantly reduced, which led to a weakening of radical adjustment of cloud particles at the initial stage of the model integration especially over the land area. The precipitation intensity in 1 h and 6 h model forecasts using the improved cloud initialization scheme were reduced. In the cycling experiments, the area with strong precipitation tended to be smaller and the intensity was significantly weakened.