Among the tasks Surya has been tested on are predicting solar flares, estimating solar wind speeds, forecasting solar EUV spectra and identifying the development of active regions on the Sun.
Surya’s training involved nine years’ worth of high-resolution solar observation data from NASA’s Solar Dynamics Observatory. The images used are ten times larger than standard AI training datasets, necessitating a specialised multi-architecture approach to manage the vast volume while ensuring efficiency.
This has produced a model with exceptional spatial resolution, capable of discerning solar features at scales and in contexts that have not been achieved before in large-scale AI training processes.
Surya is now accessible from today on Hugging Face, GitHub and through IBM’s TerraTorch library. Alongside Surya, SuryaBench has been made available as an open-source resource.
SuryaBench comprises carefully curated datasets and benchmarks designed to facilitate the development and assessment of applications not only for space weather prediction but also to deepen understanding of the Sun.
“Think of this as a weather forecast for space. Just as we work to prepare for hazardous weather events, we need to do the same for solar storms. Surya gives us unprecedented capability to anticipate what’s coming and is not just a technological achievement, but a critical step toward protecting our technological civilisation from the star that sustains us,” said Juan Bernabe-Moreno, Director of IBM Research Europe, UK and Ireland.
