Abstract
Heliophysics is central to understanding and forecasting space weather eventsand solar activity. Despite decades of high-resolution observations from theSolar Dynamics Observatory (SDO), most models remain task-specific andconstrained by scarce labeled data, limiting their capacity to generalizeacross solar phenomena. We introduce Surya, a 366M parameter foundation modelfor heliophysics designed to learn general-purpose solar representations frommulti-instrument SDO observations, including eight Atmospheric Imaging Assembly(AIA) channels and five Helioseismic and Magnetic Imager (HMI) products. Suryaemploys a spatiotemporal transformer architecture with spectral gating andlong--short range attention, pretrained on high-resolution solar imageforecasting tasks and further optimized through autoregressive rollout tuning.Zero-shot evaluations demonstrate its ability to forecast solar dynamics andflare events, while downstream fine-tuning with parameter-efficient Low-RankAdaptation (LoRA) shows strong performance on solar wind forecasting, activeregion segmentation, solar flare forecasting, and EUV spectra. Surya is thefirst foundation model in heliophysics that uses time advancement as a pretexttask on full-resolution SDO data. Its novel architecture and performancesuggest that the model is able to learn the underlying physics behind solarevolution.