NASA and IBM Develop AI to Forecast Solar Flares Before They Reach Earth

AI models developed with NASA satellite imagery are now capable of forecasting the sun’s appearance hours ahead.

“I envision this model as an AI telescope that enables us to observe the sun and grasp its ‘mood,'” states Juan Bernabe Moreno from IBM Research Europe.

The sun’s state is crucial because bursts of solar activity can bombard Earth with high-energy particles, X-rays, and extreme ultraviolet radiation. These events have the potential to disrupt GPS systems and communication satellites, as well as endanger astronauts and commercial flights. Solar flares may also be accompanied by coronal mass ejections, which can severely impact Earth’s magnetic field, leading to geomagnetic storms that could incapacitate power grids.

Bernabé-Moreno and his team at IBM and NASA created an AI model named Surya, derived from the Sanskrit word for ‘sun,’ by utilizing nine years of data from NASA’s Solar Dynamics Observatory. This satellite captures ultra-high-resolution images of the sun across 13 wavelength channels. The AI models have learned to recognize patterns in this visual data and create forecasts of how the sun will appear from future observational stations.

When tested against historical solar flare data, the Surya model demonstrated a 16% improvement in accuracy for predicting flare occurrences within the next day compared to traditional machine learning models. There is also a possibility that the model could generate visualizations of flares observable for up to two hours in advance.

“The strength of AI lies in its capacity to comprehend physics in unconventional ways. It enhances our intuition regarding physical processes,” remarks Lisa Upton at the Southwest Research Institute in Colorado.

Upton is especially eager to explore if the Surya model can aid in predicting solar activity across the sun and at its poles—areas where NASA instruments cannot directly observe. While Surya does not explicitly aim to model the far side of the sun, it has shown promise in forecasting what the sun will resemble for several hours ahead as sections rotate into view, according to Bernabe Moreno.

However, it remains uncertain whether AI models can overcome existing obstacles in accurately predicting how solar activity will influence Earth. Bernard Jackson from the University of California, San Diego, points out that there is currently no means to directly observe the magnetic field composition between the Sun and Earth, a crucial factor determining the direction of high-energy particles emanating from the star.

As stated by Bernabé-Moreno, this model is intended for scientific use now, but future collaborations with other AI systems that could leverage Surya’s capabilities may allow it to support power grid operators and satellite constellation owners as part of early warning frameworks.