Is AI the key to predicting solar storms?

A superheated stream of plasma bursts from the Sun in this 2014 image captured by the Solar Dynamics Observatory satellite. Credit: Solar Dynamics Observatory/NASA

On Sept. 1, 1859, the most intense geomagnetic storm in recorded history paid Earth a visit. This colossal solar outburst, which led to telegraph systems catching fire and aurorae lighting up skies down to the tropics, became known as the Carrington Event, in honor of English amateur astronomer Richard Carrington, who determined that a major coronal mass ejection (CME) was to blame for these curious happenings. 

The impact of CMEs

Though powerful, solar storms pose little threat to humans or other living creatures, even when CMEs blast material straight at our planet. Rather, they are far more likely to impact electrical components and systems. This is because they generate induced currents that can flow into the electrical grid, wreaking havoc. In 1859, the world was not especially reliant on electricity, meaning that despite the spectacle, the Carrington Event was only moderately disruptive. 

Were a similar incident to occur today, however, the effects would be calamitous. And, with the risk of geomagnetic storms increasing as we approach the next solar maximum — the period of highest activity in the Sun’s 11-year cycle, expected to peak next in 2025 — the need for robust prediction and mitigation strategies is a pressing concern.

“The key impacts [of a major CME] would primarily be satellites, both radiation exposure and risk of unplanned deorbiting or positional changes; radiation exposure to aircraft over the poles; radio communication blackouts; and damage to electrical power grids and railways,” explains Andy Smith, Senior Research Fellow at Northumbria University in the United Kingdom. “The worst-case scenario would likely involve … damage to electrical infrastructure, leading to long-lasting blackouts around the world,” he adds.

Even short power outages can be deadly if people lose access to medical devices and climate control. Maybe you’ve experienced tossing a freezer full of spoiled food. Now imagine there are no working freezers at all — for months. And no access to bank accounts or advanced medical care. It would be catastrophic. 

Fortunately, a number of leading scientists, researchers, and institutions are aware of this danger and are working on ways to predict when a CME is liable to occur and how the consequences could be dampened in the wake of a particularly severe one.

The answer is AI

The best approach, many experts have suggested, could be to effectively harness the power of artificial intelligence (AI). “In principle, AI can help to accurately forecast either the conditions that can lead to problems, or the problems themselves,” says Smith. 

NASA is one of the pioneers of such AI technology. Working alongside the U.S. Geological Survey and the U.S. Department of Energy, NASA has used deep learning to build DAGGER, an AI-driven computer model that can predict geomagnetic disturbances caused by solar storms 30 minutes before they occur. 

By analyzing solar wind data from various NASA satellites, DAGGER can produce rapid, global predictions that are updated by the minute. This could help power grid operators, satellite controllers, and telecom companies prepare for — and ultimately mitigate — the effects of solar storms on critical infrastructure.

Other organizations are also exploring AI’s usefulness in this area. The Europlanet 2024 Research Infrastructure (RI) is a major project that supports planetary science in Europe by providing open access to a wealth of data from various space missions, simulations, and laboratory experiments. Utilizing this information, the Know Center and the Space Research Institute, both based in Graz, Austria, have joined forces to see whether machine learning can predict solar storm risk. 

While the development of its model is still in the early stages, initial observations published in the journal Space Weather found that such a model “might be suited for operational space weather forecasting in the future.”

“The advantage of AI is that the models run much more quickly than physics-based models, so [they] have the potential to provide more lead time,” explains Amy Keesee, Associate Professor of Physics and Astronomy at the University of New Hampshire. “One of the challenges for usable forecasts, particularly for effects like geomagnetically induced currents, is their localized nature. Having such high spatial resolution in a physics-based simulation is computationally expensive, and therefore takes longer to run.”

And when the fastest computer programs can only give 30 minutes lead time, there’s no time to waste.

Not ready yet

While AI models have the potential to provide valuable warnings about solar storms before they impact Earth and could help to minimize the repercussions, the technology is not yet perfect. Machine learning needs data to learn from, and our luck at evading another Carrington-level event means there’s no record of one in the data to teach an AI program.

“I think the reluctance to move to purely data-trained forecasts is partly because of the limited data,” says Mathew Owens, Professor of Space at the University of Reading in the United Kingdom. “Once you start trying to forecast events that were not present in the training data, the forecasts become increasingly unreliable. There’s a risk, therefore, that machine learning approaches wouldn’t be reliable for the really big storms that we worry about the most.”

Smith is also keen to note that, while the use of AI is undeniably compelling, taking a cautious, vigilant approach is important. “There will be a long phase where AI will have to prove that it is better than the current forecasting methods and is reliable, particularly given the black-box nature of some of the models. We need to have a lot of trust in the model predictions, as some mitigation procedures will be expensive or undesirable (e.g., limiting capacity in some sense).”

Indeed, protecting the power grid usually means shutting it down. While a controlled shutdown is preferable to frying the grid and having to rebuild it, it’s not without cost, both monetary and in human lives.

AI technology is becoming more impressive by the day. Should a CME on par with the Carrington Event occur, the work being done today should improve our ability to anticipate — and avoid — what, in the not too distant past, would have been a wholly unpredictable and devastating impact.