SOFIA: Intelligent analysis of solar activity for space weather forecasting

Objective

Space weather describes variable conditions in the near-Earth environment that can disrupt satellites, navigation and communication systems, aviation, and power grids, with cascading consequences for modern technologically-dependent society. The Sun is the primary driver of these disturbances, releasing bursts of radiation, energetic particles, and plasma through solar flares and coronal mass ejections. Reliable forecasts of such events are therefore critical to strengthen early-warning systems and safeguard European infrastructure, yet current approaches lack consistent parameters that capture how solar activity emerges and evolves across scales. This project integrates advanced machine learning methods to improve the description and prediction of solar activity. A central innovation is the extensive use of Variational Autoencoders (VAEs) for explainable ML-based parametrization of complex solar processes. First, VAEs will generate structured latent representations of solar active regions from multi-wavelength observations provided by NASA’s Solar Dynamics Observatory, delivering physically meaningful parameters that link magnetic morphology to the occurrence of eruptions. Weakly supervised and sequential models will then enhance interpretability and capture temporal evolution, while extensions to full-disk solar observations will integrate local and global drivers of solar variability. Finally, these data-driven representations will be coupled with geophysical models of the near-Earth environment, connecting solar drivers to response in interplanetary space and Earth’s atmosphere. By uniting physics-based and data-driven approaches, this work will lay the foundation for more accurate, reliable, and actionable space weather forecasts, strengthening resilience against solar-driven hazards.