Nasa and IBM use artificial intelligence to combat solar digital disruption

The digital technology that drives modern society is

Space weather is a threat to the global economyand a Lloyd’s systemic risk scenario showed that $2.4tn in losses could be incurred over a 5-year period. A hypothetical solar storm would cause $17bn in economic damage.

Coronal mass ejections and solar flares can

They can damage astronauts, satellitesspacecraft, and/or spacecraft that are beyond Earth. They can cause satellite hardware to fail and damage solar panels and circuits. Solar weather can also affect airline travel due to navigational mistakes and radiation risks for passengers and airline crew. Solar weather disruptions could affect agriculture and lower food production.

Surya, a 366-million-parameter foundational model for heliophysics – the study of the Sun’s effect on the Solar System – is a model with a total of 366 million parameters. It was pre-trained with data from the Atmospheric Imaging Assembly and Helioseismic and Magnetic Imager Instruments on Nasa’s Solar Dynamic Observatory mission (SDO), which was launched in 2010 It uses self supervised learning to identify patterns within unlabelled solar data. This eliminates the need for solar experts to manually categorise thousands complex solar events. These solar images are ten times larger than the typical AI training data. Surya needed a custom technology to handle the massive dataset and maintain efficiency.

In an article discussing the model researchers from IBM said Surya had learned general-purpose representations of the sun that captured both “the fine scale variability of magnetic field and the large scale dynamics of the solar atmospheric”. The researchers claimed that the pre-training technique enabled the model to perform zero shot forecasting of solar activity. This represents a shift away from narrowly focused and task-specific models towards a more versatile, scalable approach to heliophysics.

The traditional solar weather forecast relies on partial views of the Sun’s surfaces from satellites, making accurate predictions extremely difficult. Surya overcomes this limitation by training with the largest curated dataset of heliophysics, which IBM and Nasa claim was designed to assist researchers in better studying and evaluating critical space weather predictions tasks.

Kevin Murphy, chief scientist data officer at Nasa headquarters in Washington, said that the company was advancing data-driven research by embedding Nasa’s deep scientific expertise into cutting edge AI models.

By developing a foundational model trained on Nasa heliophysics data we make it easier to analyze the complexities of Sun’s behavior with unprecedented speed and accuracy. This model enables a broader understanding of the impact of solar activity on critical systems and technologies we all rely upon here on Earth.

IBM and Nasa released Surya on hugging face to “democratise access to advanced tools for understanding solar weather and scientific research”, encouraging the development specialised applications for various regions and industries.

Juan Bernabe Moreno, director of IBM Research Europe UK and Ireland said: “This AI-model gives us unprecedented ability to anticipate what is coming. It’s not just a technical achievement, but an important step toward protecting our technology civilisation from the star which sustains us.”