A new application of AI improves early warnings to protect satellites and power grids from solar storms by providing an early warning. The technology predicts solar wind days in advance with far greater accuracy than existing methods.
This is obtained by analysing ultraviolet solar images. Solar wind is a continuous stream of charged particles released by the Sun. When these particles speed up, they can cause “space weather” events that disrupt Earth’s atmosphere and drag satellites out of orbit, damage their electrons, and interfere with power grids.
For example, in 2022, a strong solar wind event caused SpaceX to lose 40 Starlink satellites (as the BBC reported). SpaceX reported that the orbital decay on Starlink satellites was considered to be linked to a geomagnetic storm that was initiated on February 3, 2022. This demonstrates the urgent need for better forecasting.
Solar winds
A solar wind is a flow of particles that comes off the sun at about one million miles per hour and travels throughout the entire solar system. The ‘wind’ is composed of a stream of electrons and protons, with energies sufficient to escape the Sun’s gravity.
Solar winds were first proposed in the 1950s by University of Chicago physicist Eugene Parker, the solar wind is visible in the halo around the sun during an eclipse and sometimes when the particles hit the Earth’s atmosphere— as the aurora borealis, or northern lights.
Solar winds can impact on satellites. Our reliance on satellite technology for navigation, weather forecasting, telecommunications, and global connectivity means that the space weather has become a critical concern.
As an example, geomagnetic Storms can cause electrical surges in satellite systems and lead to damage or failure. In particular, solar wind can increase atmospheric drag, causing satellites to drift and potentially collide with the Earth’s surface.
Credit – Dre Erwin Photography, CC SA 4.0.
What does the AI do?
The scientists, from New York University, trained their AI model using high-resolution ultraviolet (UV) images from NASA’s Solar Dynamics Observatory, combined with historical records of solar wind.
Instead of analysing text, like the everyday AI language models, the AI system analyses images of the Sun to identify patterns linked to solar wind changes. The result is a 45 percent improvement in forecast accuracy compared to current operational models, and a 20 percent improvement over previous AI-based approaches.
Practical use
The U.S. breakthrough demonstrates how AI can solve one of space science’s toughest challenges: predicting the solar wind. With more reliable forecasts, scientists and engineers hope to better prepare for space weather events, strengthening resilience against disruptions to critical infrastructure.
The research appears in The Astrophysical Journal Supplement Series, titled “A Multimodal Encoder–Decoder Neural Network for Forecasting Solar Wind Speed at L1.”
