The Sun constantly emits energy in the form of light, including extreme ultraviolet (EUV) radiation, which is primarily produced in its multi-million-degree outer atmosphere. This EUV radiation affects Earth’s space environment by increasing drag on satellites, interfering with radio signals, and reducing the accuracy of GPS systems. However, direct measurements of this radiation are limited and often unavailable during fast, short-lived events like solar flares or eruptions.

This study introduces two new models that use artificial intelligence to reconstruct solar EUV radiation accurately, efficiently, and in a way that is interpretable by humans. One model, called HelioFM, learns from satellite images of the Sun to understand how EUV radiation varies across space and time. The other model, Spectral-KAN, takes a simpler approach based on known solar spectral patterns to make fast and interpretable predictions.

By combining these two models, scientists can better model both past solar activity and current solar behavior. Trained on over a decade of solar data, including major solar storms, these models outperform traditional methods. This work moves us closer to real-time space weather forecasting that can help safeguard technology and infrastructure on Earth and in near-Earth space.