IBM and NASA have launched Surya, an innovative open-source artificial intelligence model designed to analyze high-resolution solar data and predict the impact of solar activity on Earth-based and space-faring technologies. Named after the Sanskrit word for the Sun, Surya aims to decipher and forecast solar dynamics, addressing the growing need for precise space weather prediction as reliance on technology increases.
Solar flares and coronal mass ejections can cause significant disruptions, including damage to satellites, interference with air traffic control communications, widespread power outages, and risks to astronauts. A severe solar storm could result in substantial economic damage, with one assessment suggesting losses of up to $2.4 trillion globally over five years, and an estimated $17 billion from a single event. Recent incidents have already demonstrated these tangible threats, with disruptions to GPS services, flight diversions, and satellite damage occurring due to solar events.
Juan Bernabe-Moreno, Director of IBM Research Europe, described Surya as a "weather report for space," emphasizing the importance of preparing for solar storms with the same diligence as for terrestrial weather. Surya offers an advanced capability to foresee incoming solar activity by being trained on the most extensive high-resolution heliophysics dataset available. This dataset, comprising nine years of data from NASA's Solar Dynamics Observatory, is ten times larger than typical AI training datasets, necessitating a custom multi-architecture solution for efficient processing.
Initial evaluations of Surya have shown a notable 16% improvement in solar flare classification accuracy compared to previous methodologies. The model can now visually predict solar flares, generating a high-resolution image that indicates where a flare is expected up to two hours in advance. This development, which involved overcoming substantial technical hurdles, represents a significant advancement in data-driven science, integrating NASA's scientific expertise with cutting-edge AI models.
Kevin Murphy, Chief Science Data Officer at NASA, highlighted that a foundational model trained on NASA's heliophysics data will enable unparalleled speed and precision in analyzing solar behavior. The curated heliophysics dataset for key space weather tasks is accessible through the SuryaBench collection, complementing IBM's Prithvi foundational models. By making Surya available on platforms like Hugging Face, IBM and NASA are democratizing access to advanced tools, empowering researchers worldwide to develop specialized applications and bolster global preparedness against the disruptive potential of space weather events. The model's capabilities are particularly relevant given recent events, such as the May 2024 solar storm that impacted GPS services for precision farming equipment and caused disruptions to satellite broadband.