AI Model Detects Hidden Solar Eruption Signs Hours Before Emergence

Researchers from the New Jersey Institute of Technology, in collaboration with Princeton University and NASA's Ames Research Center, have developed an artificial intelligence model capable of detecting precursor signals of solar active region emergence nearly nine hours before they become visible. Published in the Journal of Geophysical Research: Machine Learning and Computation, the study introduces EarlyDetect, a machine learning tool that identifies subtle acoustic and magnetic changes occurring beneath the sun's surface, offering a potential breakthrough in space weather forecasting. Solar active regions, where intense magnetic fields eventually manifest as sunspots and solar storms, begin developing below the photosphere long before they can be directly observed. This sub-surface development leaves faint signatures in acoustic waves traveling through the sun, detectable through helioseismology. EarlyDetect analyzes hourly acoustic power maps and magnetic field measurements derived from NASA's Solar Dynamics Observatory and its Helioseismic and Magnetic Imager. The model employs a Transformer architecture to recognize complex patterns in this data. The research team, led by NJIT computer scientists and solar physicists, found that removing standard data filters significantly improved prediction accuracy. Jonas Tirona, the study's corresponding author and an NJIT undergraduate researcher, noted that initial filtering techniques meant to isolate short-timescale patterns inadvertently averaged away the very faint fluctuations essential for early detection. By discarding this noise-canceling approach, the model could capture the critical signals indicating an active region's imminent rise. In testing on unseen data, the best-performing version of EarlyDetect identified precursor signals an average of 9.24 hours before active regions emerged, outperforming standard Transformer models and previous benchmarks. Alexander Kosovichev, a distinguished professor at NJIT and co-principal investigator, emphasized the difficulty of detecting these signals within the complex vibrations of the sun. Mengjia Xu, assistant professor of data science at NJIT and the project's principal investigator, highlighted that this work demonstrates the potential of advanced machine learning to open new frontiers in space weather prediction. The implications for space weather preparedness are substantial. An advance warning of active region emergence could allow satellite operators and power grid companies to implement protective measures against potential solar storm damage. However, the team cautioned that EarlyDetect is not yet operational for real-time forecasting. The model was trained on known emergence events and still generates occasional false alarms or late predictions. Furthermore, detecting the emergence of an active region does not guarantee a subsequent flare or coronal mass ejection, as many such regions do not produce major eruptions. To support further research, the team has released the Solar Active Region Emergence Dataset, a public collection of observations compiled from the Solar Dynamics Observatory, alongside an interactive platform for data exploration. Tirona expressed hope that the project will increase awareness of machine learning's role in heliophysics, viewing the current results as a foundational step toward future capabilities in predicting solar weather events.