New IBM–NASA AI aims to forecast solar flares before they knock out satellites or endanger astronauts

A new open-source AI model from IBM and NASA promises sharper forecasts of solar storms that can disrupt global infrastructure and endanger astronauts in orbit. Released on Hugging Face, Surya offers unprecedented tools for researchers to understand the Sun’s behaviour and protect technology on which the world economy depends

IBM and NASA have released a new open-source artificial intelligence model designed to forecast solar activity and its potential impact on Earth and space-based technology.

Named Surya, after the Sanskrit word for the Sun, the model represents a significant step in using AI to interpret high-resolution solar imagery and improve predictions of so-called “space weather” – the solar storms and flares that can disrupt telecommunications, GPS, power grids and satellites.

The model has been made available via the AI platform Hugging Face, part of a deliberate move to democratise access to advanced heliophysics tools and accelerate scientific discovery.

It is the first foundation model of its kind trained specifically on detailed solar observation data, offering researchers new capabilities in analysing the Sun’s surface and predicting its behaviour.

Although the Sun sits 93 million miles away, Solar flares and coronal mass ejections – explosive bursts of radiation and charged particles – can knock out satellites, interfere with aircraft navigation, trigger widespread power outages, and expose astronauts to dangerous levels of radiation.

According to a systemic risk scenario created by Lloyd’s, the global economy could be exposed to losses of $2.4 trillion over a five-year period, with the expected loss of $17 billion from the threat of a hypothetical solar storm.

Recent episodes have underlined the vulnerability of existing systems, from disrupted GPS signals to diverted flights and damaged satellite components.

Surya is intended to support academic research and operational planning. IBM’s Director of Research for Europe, the UK and Ireland, Juan Bernabe-Moreno, likened the model to a weather forecast but for space.

Just as we work to prepare for hazardous weather events, we need to do the same for solar storms,” he said. “Surya gives us unprecedented capability to anticipate what’s coming and is not just a technological achievement, but a critical step toward protecting our technological civilization from the star that sustains us.”

Traditional forecasting of solar weather has been limited by the incomplete nature of satellite imagery and the complexity of the Sun’s activity.

Surya addresses this by drawing on a newly released curated dataset covering nine years of high-resolution observations from NASA’s Solar Dynamics Observatory.

These images are around ten times larger than typical AI training inputs, forcing the team to develop a bespoke, multi-architecture solution capable of handling the scale while maintaining computational efficiency.

The result is a model with unprecedented spatial resolution, allowing it to capture fine-grained solar features that were previously out of reach for AI systems.

In testing, Surya has already delivered a 16 per cent improvement in classifying solar flares, compared to older methods. It is also able to generate visual predictions of flare locations up to two hours in advance, producing images that indicate where activity is expected to occur — a first in this field.

The model has been tested on multiple forecasting tasks, including flare prediction, solar wind speed estimation, and the prediction of solar extreme ultraviolet spectra and active region emergence. These are critical components in understanding how space weather develops and what its downstream effects might be on Earth.

Kevin Murphy, NASA’s Chief Science Data Officer, said the initiative was about embedding deep scientific expertise into state-of-the-art AI models. He said: “By developing a foundation model trained on NASA’s heliophysics data, we’re making it easier to analyse the complexities of the Sun’s behaviour with unprecedented speed and precision.

“This model empowers broader understanding of how solar activity impacts critical systems and technologies that we all rely on here on Earth.”

The release of Surya on Hugging Face means researchers worldwide can build upon it, adapting the foundation model to their own regional or sector-specific needs.

The aim is to create a platform for collaborative development, allowing scientists and engineers to refine solar forecasting tools and apply them to operational scenarios, from satellite operators to power grid managers.

Surya is part of a wider collaboration between IBM and NASA focused on using AI to explore Earth and the solar system. It sits alongside other models in the “Prithvi” family, which includes a geospatial model and a weather model.

IBM and NASA released the Prithvi weather model on Hugging Face in 2024, enabling scientists to generate both short- and long-term weather and climate projections using open-source AI.

Main image: Composite images of the Sun in different wavelengths, captured by NASA’s Solar Dynamics Observatory. IBM and NASA’s new open-source AI model, Surya, has been trained on nine years of such high-resolution solar data to improve forecasts of solar storms that threaten satellites, power grids and even astronauts. Photo: IBM