A Meteorological Downscaling Method Based on Time-Frequency Processing Network

Atmospheric activity has a profound impact on human life and production, making the acquisition of accurate and high-resolution meteorological data critically important. To meet this demand, meteorological downscaling techniques have been developed. While deep learning-based downscaling methods inspired by image super-resolution (ISR) have shown promising results, directly migrating ISR algorithms to the meteorological domain often overlooks the unique characteristics of meteorological data, thereby limiting reconstruction performance. To address these issues, this paper proposes the Time-Frequency Processing Network (TFPN), which separates and processes the low- and high-frequency components of meteorological data individually, while also modeling their temporal dependencies. Comparative experiments demonstrate that TFPN outperforms existing baseline methods, validating the effectiveness of the proposed approach.