Abstract:

Based on the wind speed observation data of 24 wind towers in Liaoning and Jilin provinces, the linear regression method was adopted to revise wind speed forecast biases of the high-resolution mesoscale model.Firstly, the impacts of the training sample duration and rolling method on correction effectiveness were studied to determine the optimal scheme by four different calibration experiments, and the applicability of the station correction method on different underlying surfaces was synthetically analyzed.Then, the determined station correction relation from the 24 wind towers was used to correct gridded forecast wind field data, and the other 23 wind towers data were employed to assess the correction effect.The results show that the duration of the training sample has a direct impact on the correction effect.In the experiment area, the duration of 20-day for the training sample can achieve the best effect.When the training sample duration is 20 days, the correction effects of different sample selection methods are consistent.The prediction effect under various underlying surfaces can be significantly improved with the linear correcting method, and the improvement is the most obvious in the hilly area with the root mean square error (RMSE) reducing by 1.61 m·s^-1.The RMSEs in the plain and coastal areas decrease by 0.95 m·s^-1 and 0.91 m·s^-1, respectively.The overall correction experiments of the gridded wind speed data indicate that the extrapolation of the correction relation can achieve an obvious correction effect with the RMSE reducing by 0.20 m·s^-1.Therefore, the method can be effectively applied to the region where observation is scarce and will be available for modifying grid wind speed data in the future.

Key words: Numerical forecast of wind speed, Bias correction, Linear regression