Satellite-radar-Swin-Unet: a Swin-Unet-based model for synthetic radar generation from satellite data

Based on satellite data, synthetic radar serves as a vital method for weather data completion, providing essential radar echo information for remote areas. However, many existing synthetic radar models rely on relativelysimplearchitectures, which are insufficient for effectively enhancing the accuracy of radar synthesis. To address this challenge, this paper proposes a novel model, Satellite-Radar-Swin-Unet, which utilizes Swin-Unet as the core synthesis network. The model leverages Swin-Unet’s powerful feature extraction capabilities, thanks to the transformer architecture, andits image transformation ability to synthesize multi-channel Himawari-8 satellite data into CREF radar data. Experimental results demonstrate that Satellite-Radar-Swin-Unet significantly improves the accuracy of synthetic radar, making it a promising approach for future radar data synthesis.