Thanks to Surya, a pioneering AI model co-developed by IBM and NASA, we could potentially anticipate powerful solar storms hours before they threaten our technology or astronauts in space. Surya stands out as the first open-source, foundation AI model focused on solar physics, giving scientists new tools to decode and forecast the Sun’s often unpredictable outbursts.
A Leap Forward in Space Weather Prediction
Solar storms can disrupt GPS, communications, and even the safety of astronauts. Traditionally, predicting these events has been a major challenge. NASA’s Solar Dynamic Observatory (SDO) has captured vast amounts of solar data over 15 years, but only now are we equipped to put this data to its best use.
Surya marks a turning point. Built from raw SDO data, the model analyzes and forecasts solar activity faster and more accurately than ever. Researchers can access Surya through platforms like Hugging Face, GitHub, and IBM’s TerraTorch. The accompanying SuryaBench provides open datasets and benchmarks specifically designed for solar and space weather research.
Inside Surya: How the Model Works
Processing solar data is no small feat. SDO sends high-resolution images every 12 seconds, capturing details across multiple wavelengths. Surya was trained on nine years of this rich data, harmonizing it and applying cutting-edge AI architectures.
- Model Architecture: Surya leverages a long-short vision transformer with spectral gating, ideal for analyzing complex solar imagery while reducing noise and optimizing memory use.
- Forecasting Ability: The model was trained not just to reconstruct images, but to predict what SDO would see an hour into the future. This demands a deep understanding of solar geometry, magnetic fields, and the Sun’s unique rotational patterns.
The results are impressive so far. Surya enables scientists to detect solar flares up to two hours in advance nearly doubling previous warning times. It has also improved solar flare classification accuracy by 16% compared to existing methods, representing a major advance for the field.
Making Solar Research Accessible
Predicting solar storms is more than a technical puzzle, it’s a scientific frontier. The Sun’s physics are still not fully understood but Surya helps bridge this knowledge gap by automating the extraction of patterns and events from petabytes of data, making sophisticated solar analysis accessible to a broader range of researchers.
SuryaBench enhances this accessibility by providing ready-to-use datasets and benchmarks for vital space weather tasks such as:
- Forecasting the emergence and development of active solar regions
- Predicting solar flares and coronal mass ejections (CMEs)
- Monitoring extreme ultraviolet emissions and solar wind speeds
By targeting these phenomena, Surya and SuryaBench help scientists anticipate threats to our power and communication systems, and better protect infrastructure and astronauts.
Driving Future Discovery
Surya is a catalyst for scientific innovation. By translating massive, complex datasets into actionable insights, the model connects data-driven discovery with real-world impact. Its open-source, modular design allows researchers worldwide to adapt and fine-tune the model, sparking further advances in heliophysics.
The next steps will focus on predicting how solar storms affect Earth, enabling even more precise mitigation strategies. With Surya, the scientific community is better prepared than ever to understand and respond to the Sun’s changing activity.
Empowering a New Generation of Solar Science
Surya is redefining what’s possible in solar forecasting and research. By democratizing access to sophisticated AI tools and datasets, IBM and NASA are empowering scientists everywhere to uncover new insights about our Sun and to safeguard our technology and explorers from its powerful influence.
Surya: The Open-Source AI Solar Forecasting Model