EarlyDetect locates active regions of the Sun 9.24 hours before they appear

A model developed at NJIT reads the acoustic waves and magnetic field captured by NASA's solar observatory to anticipate the emergence of areas that cause solar storms.

Before an active region appears on the Sun's surface, magnetic fields rising from deep layers leave tenuous signatures in the star's acoustic waves — the field of helioseismology. EarlyDetect, presented on August 14, 2026 in the Journal of Geophysical Research: Machine Learning and Computation, learns how to recognize them.

The model cross-references hourly acoustic power maps with magnetic field measurements from the Helioseismic and Magnetic Imager on board NASA's Solar Dynamics Observatory, which samples sound waves every 45 seconds. It detects precursors on average 9.24 hours before the active region becomes visible, and outperforms standard Transformer architectures as well as previous reference approaches — at the cost of false alarms and residual late predictions.

The work is led by Jonas Tirona, an undergraduate student at the New Jersey Institute of Technology and corresponding author, with physicist Alexander Kosovichev and data scientist Mengjia Xu, in collaboration with Princeton and NASA's Ames Research Center. The challenge is operational: a few hours in advance is enough for satellite operators and power grid managers to put their facilities in safe configuration.