Artificial Intelligence at the Edge of the Sun

Here’s the sun, restless and brilliant, spitting storms that can scramble GPS, blind satellites, and nudge power grids toward the edge. For decades, space-weather forecasters have watched and modeled its moods. Now NASA is adding a new co-pilot: AI.

In mid-2025 NASA and IBM unveiled Surya, a heliophysics foundation model trained on years of high-resolution observations of the Sun. Surya ingests extreme-ultraviolet imagery and other solar data to anticipate outbursts—flares and other phenomena that can trigger chain reactions from the upper atmosphere to the ground. NASA says models like this can give earlier warnings to satellite operators and help predict how changes in the Sun’s ultraviolet output ripple through Earth’s ionosphere—crucial for communications and navigation.

The bigger story is that Surya isn’t a one-off gadget; it’s part of a strategy to fuse physics-based models with machine learning. NASA’s own overview of recent heliophysics work highlights teams that pair coronagraph and heliospheric images with ML classifiers to judge whether a coronal mass ejection (CME) will be “geoeffective”—that is, actually disturb Earth’s magnetic field when it arrives. One approach, GeoCME, is emblematic: learn from past events to flag the ones most likely to cause trouble.

AI is also being pushed right to the operational front line. In 2023, NASA described an ML-powered system that acts like a tornado siren for space weather, combining satellite data with AI to forecast hazardous conditions that threaten technology and power infrastructure. The goal isn’t to replace human experts but to buy precious lead time and triage attention on the events most likely to matter.

A lot of this acceleration has come from rapid-prototyping programs at the Frontier Development Lab (FDL)—an applied-AI accelerator run with NASA partners. FDL’s Heliolab teams have tested deep-learning pipelines on EUV imagery to assess whether solar events are eruptive, and to build “virtual instruments” that can plug into foundation models like Surya. It’s an R&D relay race: build on open data, iterate with ML, and fold the best pieces back into NASA’s toolchain.

Surya itself has been open-sourced with IBM, signaling NASA’s intent to make heliophysics AI a community endeavor. IBM characterizes Surya as the first AI foundation model in this domain, designed to help predict solar weather that endangers astronauts, satellites, and terrestrial systems—and to do it faster than current methods. Opening the weights and code invites labs worldwide to fine-tune for niche tasks (say, flare nowcasting versus CME tracking) and to benchmark against shared datasets.

Meanwhile, peer-reviewed research shows why AI is so attractive here: the Sun is periodic, but not politely so. Long-short term memory (LSTM) and related architectures have proven adept at learning solar cycles and flare statistics from decades of time series, improving long-range forecasts of sunspot numbers and related indices that underpin everything from satellite-drag models to HF radio planning. That research gives NASA and partners a menu of architectures to test as they plug AI into the space-weather stack.

Why this matters now

Solar Cycle 25 has already delivered the strongest storms in years, and the next big disturbances won’t wait for cleanroom schedules. By pairing physics with pattern recognition, NASA aims to:

• Warn earlier. Minutes to hours of extra lead time help operators reorient satellites, schedule instrument safe modes, and protect astronauts.

• Target the right threats. ML classifiers help sort “spectacular but harmless” solar fireworks from CMEs that will actually couple with Earth’s magnetosphere.

• Share tools openly. Foundation models and FDL prototypes seed a broader ecosystem, so forecasting improves everywhere—academia, industry, and national agencies.

The endgame is pragmatic: fewer surprises. Better forecasts mean steadier GPS and comms, fewer satellite anomalies, and power grids that can brace before currents surge. The Sun will keep throwing curveballs; NASA’s bet is that AI can teach us to read the pitcher’s hand a little sooner.

---

Sources

1. NASA Science: “NASA, IBM’s ‘Hot’ New AI Model Unlocks Secrets of Sun” (Aug. 2025). NASA Science

2. NASA Science: “NASA Missions Help Explain, Predict Severity of Solar Storms” (July 1, 2025) — includes machine-learning GeoCME approach. NASA Science

3. NASA (Mar. 30, 2023): “NASA-enabled AI Predictions May Give Time to Prepare for Solar Storms.” NASA

4. IBM Research Blog (Aug. 20, 2025): “Introducing Surya, a new heliophysics foundation model.” IBM Research

5. Frontiers in Astronomy and Space Sciences (2025): “Forecasting long-term sunspot numbers using the LSTM-WGAN model.” Frontiers