Overview

• The model was trained on more than a decade of multi-wavelength Solar Dynamics Observatory imagery totaling about 250 terabytes.
• IBM reports experiments in which Surya anticipated some events roughly two hours ahead, which it says could double typical warning horizons, though performance varies.
• NASA and IBM made the system openly available, alongside SuryaBench datasets and benchmarks to encourage community testing and development.
• Solar physicists, including ETH Zürich’s Louise Harra, warn that predicting the exact timing and intensity of eruptions remains challenging and limits operational certainty.
• Planned work focuses on fine-tuning the model and incorporating additional data to validate results and improve real-world forecasting utility.