A new artificial intelligence model, EarlyDetect, can identify the first signs of the formation of solar active regions on average 9,24 hours before they appear on the visible surface of the Sun.
The development was carried out by researchers from the New Jersey Institute of Technology (NJIT), Princeton University, and NASA’s Ames Research Centre. The system analyzes changes in acoustic waves and the Sun's magnetic field using data from NASA's Solar Dynamics Observatory (SDO).
Active regions are zones of intense magnetic activity where sunspots occur and flares or coronal mass ejections can form. Their origin begins beneath the visible surface, making direct observation difficult.
EarlyDetect uses a transformer-based architecture that processes sequences of hourly acoustic power maps and magnetic field measurements from the Helioseismic and Magnetic Imager instrument aboard SDO.
In tests, the model outperformed standard transformers and previous methods. Interestingly, attempting to filter out 'extra' oscillations reduced accuracy: it was precisely in these weak signals that information about the earliest stages of formation was contained.
The model remains experimental for now. It cannot predict whether a detected region will lead to a flare or ejection, and it may produce false positives.
The researchers have published the Solar Active Region Emergence Dataset (SolARED) and an interactive portal, the Solar Active Region Portal, so that other scientists can test new forecasting approaches.
Early detection of active regions could, in the future, give satellite operators, communication providers, and power grids extra hours to prepare for potential disruptions. However, testing on a much larger number of events is needed before operational use.
Could EarlyDetect become part of a real space weather warning system in the future?
