A new AI-powered tool developed through joint research by the Southwest Research Institute (SwRI) and the National Science Foundation’s National Center for Atmospheric Research (NSF-NCAR) is paving the way for significantly longer forecasts of space weather—potentially weeks in advance instead of just hours. The breakthrough centers on PINNBARDS, a physics-informed neural network designed to link observable surface features of the Sun to complex magnetic processes occurring deep within its interior. Solar active regions, which are areas of intense magnetic activity, often give rise to powerful solar flares and coronal mass ejections (CMEs)—events that can disrupt satellites, GPS systems, power grids, and endanger astronauts. These regions do not appear randomly; they tend to form along large, warped magnetic structures known as toroidal bands. By analyzing surface magnetic data from the Solar Dynamics Observatory’s Helioseismic and Magnetic Imager (SDO/HMI), researchers were able to identify patterns in these bands and use them to infer conditions beneath the Sun’s surface. Traditional forecasting methods rely on detecting small-scale magnetic changes that only become predictive hours before an eruption. PINNBARDS, however, goes further by using a physics-informed machine learning framework to reconstruct the subsurface magnetic state of the Sun’s tachocline—a thin transition layer between the Sun’s stable radiative interior and its turbulent outer convection zone. This layer is believed to play a key role in generating the large-scale magnetic fields that eventually emerge as active regions. The model takes surface magnetogram data and works backward to infer the underlying magnetic configurations and flow patterns that drive the formation of active regions. Results from the model show strong agreement with observed surface patterns, including the spatial distribution of magnetic bulges and depressions, as well as velocity fields that indicate how material moves beneath the surface. “Understanding where and when large, flare-producing active regions will emerge is one of the biggest challenges in heliophysics,” said Dr. Subhamoy Chatterjee, an early-career scientist at SwRI and co-author of the study published in the Astrophysical Journal. “These regions are full of tangled magnetic fields, and their eruptions can trigger hazardous space weather.” Dr. Mausumi Dikpati, a senior scientist at NSF-NCAR and lead researcher on the project, emphasized the importance of the new approach: “The reconstructed subsurface states from PINNBARDS provide the initial conditions needed for forward simulations of solar magnetic evolution. This opens the door to predicting the location and timing of major active regions weeks in advance.” The ability to forecast space weather with such lead time would allow critical infrastructure operators, space agencies, and mission planners to take protective measures—shielding satellites, adjusting flight paths, or preparing power grid systems. The latitude and longitude of emerging active regions are especially important, as they determine whether solar particles will be directed toward Earth and other parts of the solar system. This research marks a significant step toward a new generation of AI-driven, physics-based forecasting tools that combine the power of machine learning with fundamental solar physics to better anticipate and prepare for extreme space weather events.