Data-driven Deep SwinTransUnet Enables the Prediction of 4K Future Solar Extreme Ultraviolet Images
Year: 2026
Type: article
Abstract: Solar extreme ultraviolet (EUV) images are crucial for understanding the structure and dynamics of the solar corona and predicting various space weather hazards. However, there is a significant research gap in forecasting future 4K solar EUV images, which are vital as high-quality inputs to support downstream tasks for more accurate prediction of spaceborne hazards like magnetic storms. Current mainstream approaches to space weather forecasting typically focus on predicting one or several parameters of the Sun from existing EUV images, rather than generating holistic predictions of full-disk solar images and remaining restricted to resolutions. To address the prediction of future solar EUV images, this study combines the advantages of the Swin transformer and TransUNet, proposing a new method: Data-driven Deep SwinTransUnet Enables the Prediction of 4K Future Solar Extreme Ultraviolet Images (Swin-4KEUV). Considering the strict hardware requirements of 4K images, each 4K image is segmented into nine overlapping 2K segments (with a 1024 pixel overlap) during processing. During the training process, Swin-4KEUV learns the weighted relationships among existing solar EUV images of different regions to forecast future images at 12 and 24 minute lead times. These results demonstrate the substantial potential of Swin-4KEUV for 4K solar EUV image prediction and its capability to serve as an advanced data foundation for downstream space weather applications. Furthermore, the extended forecasting capacity of the model is explored in our discussion section, evaluating multistep prediction results up to eight frames into the future with a total lead time of 96 minutes.
Language: English
Subfield: Astronomy and Astrophysics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $4,499
OpenAlex ID: W7221792706