Abstract: In traditional statistical analyses, the vector wind field is always verified by wind speed and wind direction separately. However, assessment results of wind speed are often contrary to those of wind direction, which then makes it difficult to obtain a uniform conclusion. To solve this problem, a novel selection scheme of wind speed thresholds is defined based on the probability distribution of wind speed, which is not affected by geographical and seasonal factors and it can keep universality in different complex environments. A vector wind classification is established based on integrating wind speed classes and wind directions. Using the spatial verification technique of fraction skill score (FSS), a vector wind verification method is developed by integrating wind speed and wind direction together. Based on hourly forecast products with different resolution (10 km and 3 km) simulated by GRAPES_Meso model from 1 April 2018 to 30 April 2018, assessment results show that the randomness of the wind direction forecast will decrease with the increasing of wind speed, which indicates it's difficult to predict the wind direction of weak wind speed successfully. By the comprehensive analysis of the vector wind field, it is found that the high-resolution (GRAPES_3 km) forecasting performance has a maximum score advantage of 0.24 on the 170 km spatial scale than the lower one (GRAPES_10 km). Scores in adjacent region are highly consistent, and it does not change the evolution of the score with the time series. And therefore, a comprehensive score can be calculated and used to assess the modelling performance by averaging skill scores in each spatial scale. In this way, deficiencies of artificial definitions of spatial scale can be avoided, which guarantee the spatial verification score of vector wind better practical application value. Simultaneously, different vector wind field classification methods have an impact on evaluation results. By sensitivity analysis, the higher wind classification method can make the score more stable with moderate wind classification one, and the magnitude of comprehensive scores are basically equivalent. The lower wind classification method has a relatively low overall score due to its single classification score weight, and it leads to the weak consistency with results of higher classification method. Therefore, under conditions of computing ability, choosing an encrypted wind direction classification to obtain a vector wind classification method will help to obtain a more stable verification result and improve the convergence and stability of the comprehensive evaluation.