Abstract:Based on wind speed of tropical cyclones from automatic weather stations in Zhejiang and GRAPES_TCM model outputs of U, V at 10 m from Shanghai Typhoon Institute during 2006-2011, verification has been carried out on different speed criteria in coastal areas of Zhejiang province. Result shows that the more the terrain simulation deviations are, the greater the mean errors of GRAPES model direct output velocity are. Wind speed verification at the stations with absolute model terrain deviations less than 10 m indicates that there exist systematic errors in the model output velocities for the leading hours of 12-72 h. Threat scores tend to descend with speed criteria ascending, owing to increasing false alarming ratio. Bias scores imply that the model tends to overforecast the winds. Analysis indicates that the model outputs issued at 08 PM perform better than those at 08 AM because of the declining missing event ratio. MOS can improve correlation coefficients between forecast and practical speeds, as well as output stability and Threat scores, moreover, to some extent, it can also help improve the model Threat scores under different wind scales, it is proved to be able to forecast for leading hours of 12 and 24 h and for wind force over 6 scale as good as the subjective one.