Researchers from NYU Abu Dhabi trained an artificial intelligence model on ultraviolet images from NASA's Solar Observatory, and were able to predict solar wind speeds up to four days in advance – with an accuracy tens of percent higher than existing models, with the potential to better protect satellites, power grids, and astronauts.
New artificial intelligence (AI) developed at NYU Abu Dhabi’s New York City campus can predict solar wind conditions up to four days before they reach Earth—a significantly higher accuracy than current methods. This capability could improve protection of satellites, power grids and astronauts from extreme space weather.
Image: An artificial intelligence system developed at NYU Abu Dhabi can predict solar wind conditions four days in advance by analyzing detailed images of the sun. The improved accuracy could help protect satellites, power grids, and astronauts from space weather. Credit: Shutterstock
The new system, described in a paper published in The Astrophysical Journal Supplement Series, learns to predict solar wind speeds up to four days in advance—a period of time that allows for better preparation for hazardous events. The researchers report that the level of accuracy they achieve is significantly higher than that of the operational forecast models currently used at space weather stations around the world.
The solar wind is a constant stream of charged particles emitted from the sun and drifting out into space. When this stream becomes particularly strong or turbulent, it can cause space weather events: changes in Earth’s magnetic field, disturbances in the upper atmosphere, the knocking of satellites out of orbit, damage to electronics in space, and disruptions to electrical and communications systems on the ground. In 2022, for example, a powerful solar wind burst caused the loss of 40 SpaceX Starlink satellites—a painful reminder of the importance of early warning.
To develop the solar wind prediction AI system, the NYU Abu Dhabi research team, led by postdoctoral researcher Dattaraj B. Dhuri and Center for Space Science (CASS) Deputy Director Shravan Hanasoge, trained a neural network on pairs of data: on the one hand, high-resolution ultraviolet (UV) images of the sun's surface and solar atmosphere, collected by NASA's Solar Dynamics Observatory (SDO); on the other hand, historical measurements of solar wind speed.
Unlike large language models (LLMs) that process text, this model “reads” detailed images of the sun. It looks for visual patterns—active regions, magnetic structures, and other features—that are linked to future changes in the solar wind. A comparison with currently operational models shows an improvement of about 45% in prediction accuracy, and about 20% improvement over previous methods that were also based on AI, but in a less rich and “multisensory” way.
According to Dori, the lead author of the paper, this is “a significant leap forward in protecting the satellites, navigation systems, and power infrastructures on which modern life depends.” He explains that combining high-resolution solar observations with artificial intelligence to predict the solar wind allows for early warnings—days in advance—that can help satellite operators, power companies, and space agencies take damage-preventing measures.
This progress highlights the growing role of artificial intelligence in exploring one of the most persistent challenges in space physics: understanding the behavior of the sun and predicting space weather. As solar wind and space weather predictions become more reliable, scientists and engineers can better plan for the response of satellite, navigation, and power systems to future storms, and strengthen the resilience of critical infrastructure on which modern society relies.
More of the topic in Hayadan:
