Artificial intelligence is getting better at forecasting sun-driven threats. IBM and NASA have unveiled a groundbreaking AI model named Surya, Sanskrit for “Sun,” which represents the most advanced digital twin of the star we orbit. The model analyzes solar activity to forecast solar flares and wind faster and more accurately than existing methods, reports WIRED.com.
Surya was trained on nine years of high-resolution imagery from NASA’s Solar Dynamics Observatory (SDO), which has been photographing the Sun in multiple wavelengths every 12 seconds since 2010. This data includes temperature variation, magnetic field strength, and motion across the Sun’s surface.
What makes Surya special isn’t just its massive training set; it’s how it uses the data. The model employs long-range vision transformers, which can analyze massive spatial patterns, and spectral gating, which filters noise and reduces memory use by about 5% during processing. The model is fine-tuned to identify early flaring activity from evolving visual cues across the solar disk.
In early tests, Surya improved flare classification accuracy by 16%. More impressively, it delivered predictions up to two hours in advance, roughly double the lead-time of traditional solar predictors.
Surya isn’t just a scientific breakthrough; it’s practical. Solar storms can cause major damage: they can knock out power grids, overload satellite systems, degrade GPS, and disrupt radio communication. With longer warning times, operators of space infrastructure, aviation networks, and utilities can take actions to mitigate risk.
Surya’s framework also enables future growth: the AI architecture is adaptable to future missions and datasets (e.g., Parker Solar Probe, SOHO). The model is open source, available via HuggingFace and GitHub, allowing researchers to build tools for planetary science, Earth observation, and more.