AI powers the future of space weather

NASA, in collaboration with IBM and research partners, has introduced Surya – a new artificial intelligence foundation model designed to analyze and forecast the Sun’s activity. Trained on nine years of continuous data from NASA’s Solar Dynamics Observatory, Surya demonstrates how advanced AI can transform heliophysics and space weather prediction.

Surya applies a foundation model architecture that learns directly from raw solar data rather than requiring extensive labeling. Its design combines spectral block layers with a long-short transformer backbone, enabling it to detect both long-term solar cycles and short-lived phenomena such as sunspots and flares.

Preliminary benchmarks indicate that Surya improves solar flare forecasting accuracy by up to 16 percent compared to previous models. It can generate visual predictions two hours in advance and forecast solar wind speeds up to four days ahead. These capabilities provide valuable tools for anticipating space weather events that can affect satellites, power grids, navigation systems, and human spaceflight.

Surya has been validated in four core research tasks:

• Active Region Emergence Forecasting – Generating 24-hour predictions of new solar regions.
• Solar Flare Forecasting – Estimating the likelihood of major flares within the next day.
• Solar Wind Speed Prediction – Forecasting solar wind conditions critical for satellites and grid stability.
• EUV Spectra Prediction – Mapping the Sun’s irradiance to better understand its impact on Earth’s atmosphere.

The model’s versatility highlights its potential not only for heliophysics but also for other scientific domains. Its architecture and methodology can be adapted for planetary science, Earth observation, and beyond.

Surya was developed under NASA’s Interagency Implementation and Advanced Concepts Team (IMPACT) at Marshall Space Flight Center as part of the agency’s broader AI for Science initiative. The project involved NASA centers, IBM Research, the Southwest Research Institute, SETI Institute, and universities across the United States.

Development required overcoming significant challenges, including memory constraints and the need to combine frequency-aware with time-series modeling. By integrating expertise from both AI engineers and heliophysicists, the team created a model capable of learning fine spatial details while also capturing long-term temporal dynamics.

Both the Surya model and its training datasets are available on Hugging Face and GitHub, ensuring open access for researchers, educators, and students worldwide. The training process was supported by the National Artificial Intelligence Research Resource (NAIRR) Pilot, an initiative led by the National Science Foundation, with computing infrastructure provided by NVIDIA and other partners.

Beyond its role in advancing heliophysics, Surya also supports sustainability by strengthening the resilience of technologies that underpin modern society. Accurate space weather forecasts help protect renewable energy infrastructure, power grids, and global communications systems from solar disruptions, reducing the risk of costly outages and environmental impact from damaged equipment. In this way, AI-driven solar forecasting indirectly contributes to the stability of energy networks and the broader goals of green technology.

Surya is part of NASA’s broader “5+1” AI strategy, which aims to develop foundation models for each of NASA’s major science domains alongside a large language model to connect them. By embedding scientific expertise directly into advanced AI systems, NASA is creating tools that accelerate discovery, improve prediction, and strengthen resilience against the risks posed by space weather.