Surya: IBM and NASA’s open AI predicts solar flares two hours ahead

IBM, working with NASA, has unveiled Surya—an open AI model whose name means “Sun” in Sanskrit. For the first time, researchers can analyze high-fidelity records of solar activity and forecast how it might affect Earth and space-borne technologies. The aim is practical: shield critical systems such as GPS navigation, power grids, and telecommunications from the fallout of solar storms.

Solar flares and coronal mass ejections can disable satellites, disrupt aviation navigation, cause power outages, and endanger astronauts. Forecasting has long been limited by local observations. Surya changes the picture by training on the largest dataset of its kind: nine years of imagery from NASA’s Solar Dynamics Observatory. These images are ten times larger than standard datasets, which required tailored processing approaches to handle them effectively.

Tests indicate a 16% jump in flare-classification accuracy compared with earlier methods. Just as important, the model can for the first time produce high-resolution forecast visualizations that point to where a flare may occur up to two hours in advance. Two hours may sound modest, but in operations that live on tight margins, that window can make all the difference.

Surya is part of a broader IBM–NASA effort to apply AI across Earth and space research. The partners previously introduced the Prithvi family for climate and geospatial analysis. Both Surya and the largest open solar-physics dataset are hosted on Hugging Face, making them accessible to researchers worldwide—an opening that is likely to speed up progress in space-weather studies.