NASA AI spots storm-making solar regions 12 hours early

Before a sunspot breaks through the Sun’s surface, something quieter is already happening below it: magnetic fields are rising and acoustic waves are shifting. NASA says a new AI model can detect those faint signals and predict the emergence of a storm-making active region up to 12 hours before it becomes visible.

The stakes are larger than a new mark on a telescope image. Active regions are the engines behind solar flares and coronal mass ejections, blasts of high-energy radiation and charged particles that can threaten astronauts, disable satellites, and disrupt radio communications on Earth. Today, operational forecasters at NOAA’s Space Weather Prediction Center and the U.S. Air Force mainly monitor regions that have already appeared on the solar surface.

The COFFIES team — NASA’s Consequence Of Fields and Flows in the Interior and Exterior of the Sun science center — trained the model on observations from NASA’s Solar Dynamics Observatory, with computing support from NASA’s Ames Research Center. Its sliding-window transformer architecture moves across long sequences of solar data, concentrating on recent activity while retaining broader patterns. The goal is to identify tiny drops in acoustic power and changes in magnetic fields that precede the emergence of a sunspot.

Researchers from the New Jersey Institute of Technology, Princeton University, and NASA Ames contributed to the work, published in the Journal of Geophysical Research: Machine Learning and Computation. Alexander Kosovichev of NJIT compared the signal to “a slight change in rhythm within a very noisy orchestra”: the magnetic structure cannot be seen directly while it is still inside the Sun, so the system must infer it from indirect effects.

So what, concretely? If the approach survives validation across many more known solar events, forecasters could gain several hours to locate an emerging active region before it is visible and assess the risk of later eruptions. That extra warning could help protect astronauts, satellites, and communications systems. The limit is clear: NASA says the model is promising but not yet ready for operational real-time forecasting.