Location-Aware Memory Enhancement for Wind Speed Super-Resolution

High-resolution wind speed data holds significant importance for wind energy assessment, urban climate modeling, and atmospheric pollution research. However, traditional reanalysis data are limited by insufficient resolution, not meeting the demands of precision applications. To address the accuracy limitations of existing deep learning methods caused by inadequate input information for complex wind speed region reconstruction, this study proposes a wind speed super-resolution method called Location-Aware Memory Enhancement (LAME). The approach innovatively decouples geographical features from model parameters, implicitly modeling site-specific wind speed characteristic distributions through self-supervised learning, while integrating multi-scale feature fusion via windowed attention mechanisms. Experimental results show that the LAME method achieves a 0.1-0.2dB Peak Signal-to-Noise Ratio (PSNR) improvement across mainstream super-resolution model frameworks. This research provides a novel approach for high-precision wind speed reconstruction in complex terrain scenarios, advancing the practical application of data-driven methods in meteorological super-resolution. The source code and datasets are available at https://github.com/USTC-AI4EEE/LAME.