NASA and IBM’s AI model predicts solar flares 16% better than standard methods

NASA and IBM have developed an AI model called Surya that can predict what the sun will look like hours into the future, including the appearance of potentially dangerous solar flares. The breakthrough could provide crucial early warnings for space weather events that threaten satellites, power grids, and astronauts, with the model showing 16% better accuracy than standard machine learning approaches in predicting solar flares within 24 hours.

How it works: The Surya model was trained on nine years of ultra-high-resolution solar imagery from NASA’s Solar Dynamics Observatory, which captures the sun in 13 different wavelengths.

• The AI learned to identify visual patterns in solar activity and generate predictive images of what the observatory would see up to two hours in the future.
• When tested against historical data, Surya demonstrated superior performance in forecasting solar flare occurrences compared to conventional machine learning models.

Why this matters: Solar flares and coronal mass ejections can unleash high-energy particles, X-rays, and extreme ultraviolet radiation that pose serious risks to Earth-based infrastructure and space operations.

• These solar outbursts can disrupt GPS systems, communications satellites, and potentially harm astronauts and commercial airline passengers.
• Geomagnetic storms triggered by solar activity are capable of knocking out entire power grids, making accurate prediction systems critical for modern society.

What they’re saying: “I love to think of this model as an AI telescope where you can look at the sun and you can understand the moods,” says Juan Bernabé-Moreno at IBM Research Europe.

• “The power of AI is that it has the ability to learn the physics in a more roundabout way – it kind of develops an intuition for how the physics works,” explains Lisa Upton at Southwest Research Institute in Colorado.

Looking ahead: The model shows promise for predicting solar activity on parts of the sun that current instruments cannot directly observe, including the far side and polar regions.

• Surya has already demonstrated success in forecasting solar appearance as the far side rotates into view, according to Bernabé-Moreno.
• Future integrations could make the technology accessible to power grid operators and satellite constellation owners as part of comprehensive early warning systems.

The limitations: Current challenges in predicting exactly how solar activity impacts Earth remain unresolved, as there’s no direct way to observe magnetic field configurations between the sun and Earth.

• These magnetic fields determine the paths of high-energy particles traveling from the sun, making precise Earth-impact predictions difficult even with improved solar forecasting, according to Bernard Jackson at the University of California, San Diego.