Study on bias correction method of ECMWF surface variable forecasts based on deep learning

Wind power is affected by various meteorological conditions, including wind speed and temperature, leading to significant volatility that can impact the safety of grid operations. Numerical weather prediction (NWP) is an efficient technique for predicting wind power. To enhance the accuracy of wind power prediction, this study proposed a correction model based on convolutional neural network to reduce the error of NWP surface products. When applying the correction model to the forecasts of NWP in June 2019, the results showed a significant reduction in errors in western China. Moreover, the effect of the correction model was better than that of the correction model trained only with surface variables, after the inclusion of upper-air variables. To reduce computational effort, this study also investigated the impact of different resolution training datasets on the correction effect. The results showed that a correction model trained with low-resolution data can achieve the same effect as that trained with high-resolution data. This study supports improving the accuracy of NWP surface products and reducing the computational effort of correction models.