HWBRIEF
A NASA led model reads faint shifts in the Sun's acoustic waves to flag emerging active regions up to 12 hours before they appear.
Sunspots don't appear out of nowhere. Hours before they break through the Sun's surface, the magnetic flux that will become them sends faint signals through the star's acoustic field, and a NASA-led machine-learning model has learned to read those signals. The model can flag where active regions will emerge up to 12 hours before they appear, according to NASA's COFFIES team.
The team is led by Alexander Kosovichev at the New Jersey Institute of Technology with Princeton University. The model does not see the active region itself; magnetic flux rising through the solar interior is not directly observable. Instead, it reads tiny shifts in the Sun's acoustic waves caused by that hidden magnetic activity, an indirect proxy that turns an invisible process into a measurable one.
Active regions drive the severe space weather that threatens astronauts, disables satellites, and disrupts radio communications. Twelve hours of lead time is a research milestone, not yet an operational forecast, and the model was trained on 46 emerging regions for the Transformer version, with a best-case RMSE of 0.1189. The Transformer study on arXiv extends a 2024 LSTM paper that reached about five hours of lead time from the same kind of acoustic precursors, and was published in the Journal of Geophysical Research: Machine Learning and Computation. Generalizability across the full solar cycle remains open.
The COFFIES team says the next step is testing the model on a wider range of solar conditions to see whether the 12-hour head start holds beyond the dataset used so far.