A Spatial Downscaling Approach for Enhanced Accuracy in High Wind Speed Estimation Using Hybrid Attention Transformer

Due to lack of a dense network of ground observations in China before 2008, the China Meteorological Administration’s Land Data Assimilation System (CLDAS) faces challenges in directly generating high-resolution and high-quality land assimilation products prior to 2008. To address this issue, this paper proposes a deep learning model based on the Hybrid Attention Transformer (HAT), aiming to improve the downscaling accuracy of high speed winds in the CLDAS2.0 10-m wind field from 6.25 to 1 km by (1) incorporating digital elevation information (DEM), (2) enhancing the loss function, and (3) employing a prediction error method. We utilized data in 2020–2021 for training and validation, and data in 2019 for testing, conducted ablation experiments to verify the effectiveness of each module, while comparing the results with those of the traditional bilinear interpolation method and the UNET model coupled with a dual cross-attention mechanism. The ablation experiment results indicate that in terms of wind speed categories, HAT with DEM performs the best for wind speeds below level 3 on the Beaufort scale, while HAT with DEM, loss function, and prediction error improvements excels for wind speeds above level 4. Specifically, for wind speeds above level 6, the HAT with all the three improvement measures achieves decent results, with mean absolute error (MAE), probability of detection (POD), and threat score (TS) of 0.825 m s⁻¹, 0.813, and 0.607, respectively, when evaluated against CLDAS3.0 as the ground truth. The model performs better in March–May and November, while its performance is the weakest in June–August; it also performs better during the day than at night and shows suboptimal performance over the plains. The model is closer to the ground truth in reconstructing the structural details of wind fields and outperforms the annual average during most high wind weather events, indicating better predictive capability and adaptability for such events. Overall, the HAT with all the three proposed improvements demonstrates significant progress in downscaling predictions of high winds and provides insights into generation of high-resolution historical meteorological gridded data.