Abstract: By using the GFS data, the numerical experiments were carried out to evaluate the Kain-Fritcsh convective parameterization scheme and its improved schemes in the numerical weather prediction model GRAPES_Meso.The improvement schemes:1) Decompose the old temperature disturbance into vertical and horizontal which determined by water vapor (KF1 scheme); 2)add the vapor advection calculated by relative humidity into the original temperature disturbance (KF2 scheme); 3) Add vapor advection calculated by relative humidity into the decomposed temperature disturbance (KF3 scheme).The results show that:1) In 5.23 rainstorm, the improved schemes can reduce part of false precipitation that simulated by the original scheme, the KF1 simulated the precipitation range well, the KF2 simulated the heavy precipitation center well,and the KF3 was the best.2)All the three improved schemes have weak simulated precipitation for heavy precipitation stations, and the precipitation trend of KF1 is close to the observation,the precipitation in the later period is slightly insufficient because convective excitation is faster, the KF2 is the opposite, the KF3 is between them.Both KF2 and KF3 will adjust the convective excitation in a favorable direction at the moment of the strongest and weakest rainfall.KF1 simulates the central pressure better,KF2 simulates the path of typhoons better, and KF3 scheme has better performance than them in typhoon.The result of T-score verification indicate that the KF1 has a low score than others,the RMSE of KF1 is larger in the middle layer and smaller in the high and low layer, while the other two schemes is opposite.