AI Model Detects Solar Storms 9 Hours Before Impact

Most of us think about weather in earthly terms—rain, clouds, wind patterns shifting across a map. But the Sun sends its own weather our way, and we've been largely blind to it until it arrives. Space weather can paint the sky with auroras at the poles, a sight that draws crowds and cameras. What gets less attention is what happens when a solar storm hits our infrastructure: satellites go dark, communication networks fail, power grids strain under the load. For decades, scientists have struggled with a fundamental problem: how to see these storms coming before they reach us.

A team of researchers at the New Jersey Institute of Technology may have found a way forward. They've built an artificial intelligence model called EarlyDetect, and in a paper published in the Journal of Geophysical Research: Machine Learning and Computation, they describe how it can spot the warning signs of dangerous solar activity roughly nine hours before the storm arrives. The model works by analyzing data from NASA's Solar Dynamics Observatory, specifically looking at the Sun's magnetic field and the acoustic waves that constantly ripple through it. The challenge, as Dr. Alexander Kosovichev of NJIT explains, is that active regions—the places where space weather originates—begin forming beneath the Sun's visible surface, where direct observation is impossible. What researchers can detect are tiny shifts in magnetism and subtle changes in wave patterns. "It's more like detecting a slight change in rhythm within a very noisy orchestra," Kosovichev said.

EarlyDetect uses what's called Transformer architecture, the same framework that powers large language models like ChatGPT and Gemini. The researchers trained the model on years of data collected by the Helioseismic and Magnetic Imager aboard the Solar Dynamics Observatory, then tested it on solar activity it had never seen before. The results were striking: the model identified precursor signals an average of 9.24 hours before active regions became visible and produced space weather events. Dr. Mengjia Xu, the principal investigator on the project, noted that machine learning has rarely been applied to solar forecasting at this scale. "Our work shows that advanced machine learning models can open new possibilities for future space weather prediction," she said.

The stakes for this kind of early warning are enormous. History offers a sobering reminder. In September 1859, the Sun unleashed what became known as the Carrington Event, named after British astronomer Richard Carrington, who observed the solar flare on the morning of September 1. The intensity of those flares was later estimated to be equivalent to ten billion atomic bombs. The storm reached Earth within hours, and the consequences were immediate and global. Telegraph networks failed as operators received electric shocks. Telegraph machines continued operating even after their power sources had been cut. Telegraph paper caught fire. Auroras were visible worldwide. Some people believed the apocalypse was unfolding.

If such an event occurred today, the damage would be far more severe. Modern civilization depends on satellites for communications, navigation, and weather forecasting. Power grids span continents. A solar storm of Carrington Event magnitude could disable transformers, disrupt financial systems, and leave millions without electricity for months. With nine hours of warning, utilities could take protective measures. Communications companies could reroute traffic. Spacecraft operators could move satellites into safe modes. The difference between no warning and nine hours of warning could mean the difference between catastrophe and managed disruption.

EarlyDetect is not yet operational as a forecasting tool. The research is still in its early stages, and the model will need to be tested and refined further before it can be deployed operationally. But the proof of concept is there. For the first time, machine learning has shown it can detect the subtle precursors of solar storms hours before they arrive. Whether this technology will eventually become part of the standard toolkit for space weather forecasters remains to be seen, but the door has opened.