Surya: NASA and IBM Release Largest Open-Source Heliophysics AI Model on Hugging Face

IBM and NASA have unveiled a groundbreaking open-source AI model called Surya, designed to predict solar weather and protect critical infrastructure from space-based disruptions. The powerful AI model is trained on 14 years of observations from NASA’s Solar Dynamics Observatory (SDO).

Here's a breakdown of what makes this initiative so impactful:

What Is Surya?

• Name Origin: “Surya” is Sanskrit for “Sun,” reflecting its heliophysics focus.
• Purpose: Predict solar flares, coronal mass ejections, and other solar phenomena that can disrupt satellites, GPS, power grids, and telecommunications.
• Availability: Open-source and hosted on Hugging Face.

Technical Highlights

• Foundation Model: Trained on 14 years of high-resolution solar data from NASA’s Solar Dynamics Observatory (SDO).
• Data Types: Includes solar coronal EUV images, magnetic field maps, and solar surface velocity data.
• Model Size: 366 million parameters—lightweight enough for broader deployment.

Capabilities & Performance

• Forecasting Power:
  • Predicts solar flares up to 2 hours in advance.
  • Achieved 16% improvement in flare classification accuracy over previous models.
• Use Cases:
  • Early warnings for satellite operators.
  • Infrastructure protection for energy grids and aviation.
  • Academic research in heliophysics and space weather.

Why It Matters

• Economic Risk: A major solar storm could cost the global economy up to $2.4 trillion over five years.
• Recent Events: Solar storms have already disrupted GPS, diverted flights, and damaged satellites.
• Future-Proofing: As humanity ventures deeper into space, accurate solar forecasting becomes essential for safety and continuity.

Open Science Impact

• SuryaBench Dataset: IBM and NASA also released the largest curated heliophysics dataset to support further research.
• Community Collaboration: Encourages scientists and developers to build on Surya for new applications in space weather prediction.

Summary Table

Feature	Details
Model Name	Surya
Developed By	IBM & NASA
Training Data	14 years of solar observations from NASA's SDO
Parameter Count	366 million
Forecast Window	Up to 2 hours before solar flare events
Accuracy Improvement	16% over prior models
Hosted On	Hugging Face