Abstract:To enhance the forecast accuracy of average wind speed and direction, based on the 2 min average wind speed and direction data from surface observations provided by China Meteorological Administration from January 2021 to December 2022, as well as the 10 m wind forecasts from the ECMWF deterministic model at 24 h lead time, a two-step bias correction method tailored for multi-site wind speed and direction forecasts is developed for 662 national observation stations that are located within the range of 30°-40°N and 110°-120°E. Then this method is validated by using data from January to December 2023. The results show that proper correction of the u and v winds forecasted by model can effectively improve wind speed and direction forecast performance at individual stations. However, significant discrepancies exist in the u and v wind forecast errors among stations, and joint modeling of u and v winds tends to compound errors in wind speed and direction forecasts. Constructing simple linear regression correction models separately for u and v winds at each station can improve wind direction forecasts at most stations, but the correction capability for strong wind speeds (≥10 m·s-1) remains limited. To overcome this shortcoming, a quantile matching approach, i.e., the two step bias correction method, is applied to further correct the wind speed derived from the regression adjusted u and v winds. The validation results show that the two step bias correction method performs well in both wind speed and direction forecasting. Compared with the ECMWF model, the corrected root mean square error by this method is reduced by 18.8% for all wind speeds and by 29.6% for strong wind speeds. Moreover, this method also exhibits distinct advantages in forecasting strong winds associated with cold air and typhoons.