IBM-NASA Surya open source AI model forecasts solar flares
The new AI model could improve readiness for surges in solar activities that affect communications systems such as GPS. It also affects space weather forecasts.
A new open source model from IBM and NASA will allow researchers to predict how solar activity affects Earth.
IBM and NASA on Wednesday introduced Surya, an open source foundation model designed to understand solar observation data and predict solar flares. It can produce images of where a flare will appear two hours before they occur and forecast space weather.
IBM said researchers trained the model on high-resolution solar observation data from NASA's Solar Dynamics Observatory, which is 10 times larger than the typical AI data.
Surya is available on Hugging Face and joins the Prithvi family of foundation models, which includes a geospatial model and the Prithvi weather model, released last year.
Solar flare improvements
The model addresses a need for better technological responses to solar flares and coronal mass ejections (CME). According to NASA and IBM, solar storms can affect satellites and airline travel.
The model is also part of IBM's plan to use AI technology as a tool and motivating factor for scientific discovery.
It's unclear how the model will evolve, but experts say its scope could be expanded. Surya reported a 16% improvement over the current ways scientists are trying to predict solar flares.
"Perhaps there are similar space phenomena that can be learned and predicted using translation properties of this model," said Chirag Shah, a professor in the Information School at the University of Washington, Seattle. "LLMs trained on one task can be reconfigured for another."
He added that the open source release will allow researchers worldwide to build on the model to develop specialized applications for their regions and industries.
The interesting part of this story is understanding that coronal mass ejections can take between 15 hours and several days to reach Earth.
David NicholsonAnalyst, The Futurum Group
Depending on how it works, the model can lead to a certain level of preparedness, said David Nicholson, an analyst at The Futurum Group.
"The interesting part of this story is understanding that coronal mass ejections can take between 15 hours and several days to reach Earth," Nicholson said. "If this model does a much better job of predicting the severity of solar flares, then it can help give us advanced warning so that we can prepare ourselves to some degree."
It's unclear whether having a warning will make much of a difference, Nicholson continued. He added that recently the Earth passed a solar maximum point, which creates an increase in solar activity that can affect communications and navigation systems such as GPS.
"So a model that tracks CMEs and solar flares is released right when we expect the numbers to decrease; a bit like buying a new winter coat in May or June," Nicholson said.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
