Research on WRF wind and temperature forecast error correction methods based on machine learning during cold waves

This study addresses the forecast errors of near-surface wind speed and temperature in complex terrain under cold wave conditions. Taking the Meihuashan area in Guizhou as a case study, numerical simulations of a strong cold wave process in January 2024 were conducted using the WRF model. Combined with meteorological observation data, the correction effects of three machine learning methods—Random Forest (RF), XGBoost, and Support Vector Machine (SVM)—on 2-meter temperature and 10-meter wind speed were compared. The results indicate that the WRF model effectively captures the overall trends of wind and temperature variations but exhibits systematic biases. Random Forest performed best on the test set, significantly reducing the RMSE for wind speed and temperature to 0.982 m/s and 0.748°C, respectively. This provides reliable technical support for power grid icing warnings.