|

IBM and NASA Just Launched Surya—the Open-Source AI That Predicts Solar Flares Before They Strike

ByInnoVirtuoso

What if we could forecast a dangerous solar flare with enough time to protect satellites, reroute flights, and safeguard power grids? Until now, space weather forecasting has struggled with limited visibility and tight timelines. Today, that changes.

IBM and NASA have unveiled Surya, an open-source AI “foundation model” trained to predict solar flares up to two hours in advance—and to show exactly where on the Sun they’re likely to ignite. It’s a first for heliophysics and a big step toward more resilient technology here on Earth.

Here’s why this matters: solar storms can knock out GPS, damage satellites, disrupt aviation, and even trigger large-scale power outages. A model that predicts flares with 16% greater accuracy than current methods gives operators a precious buffer. Two hours is enough time to move spacecraft into safe mode, adjust grid load, or keep crews and passengers away from polar routes.

In short, Surya brings cutting-edge AI to one of the toughest forecasting problems—and it’s open for anyone to use and build on.

Let’s unpack how it works, why it’s different, and what you can do with it.

What Is Surya? An AI Foundation Model for Space Weather

Surya is a generative, vision-based AI model designed specifically for solar physics. Think of it like a large language model, but for the Sun’s behavior. Instead of reading sentences, it “reads” the Sun’s surface and atmosphere across multiple wavelengths and infers what’s likely to happen next.

A few standout points: – It predicts solar flares up to two hours ahead with higher accuracy than existing approaches. – It generates visual maps (heatmaps) of where a flare is likely to erupt on the solar disk. – It’s trained on nine years of high-resolution observations from NASA’s Solar Dynamics Observatory (SDO). – It’s open-source, along with a curated dataset called SuryaBench, so researchers and operators can customize it.

Surya’s name comes from the Sanskrit word for “sun”—fitting for a model built to watch our star in unprecedented detail.

To understand the leap here, it helps to know how high the bar is. The Sun is chaotic. Its magnetic fields twist and snap in ways that are hard to model. Historically, forecasters have used physics-based models and handcrafted features (like sunspot size or magnetic complexity) to estimate flare probability. Surya learns directly from raw solar imagery—no expensive labeling required—and it does so at a scale that would choke most AI systems.

Why Predicting Solar Flares Matters (A Lot)

You might not think about the Sun much during your workday. But the modern world runs on space-dependent tech that the Sun can disrupt in minutes.

Here’s the short list: – Satellites: Flares and their associated radiation can degrade electronics, reduce solar panel efficiency, and shorten lifespan. – GPS/GNSS: Signal accuracy drops during solar disturbances, affecting aviation, shipping, precision agriculture, and logistics. – Aviation: Airlines flying polar routes face communication blackouts during high-radiation events, forcing diversions and delays. – Power grids: Extreme space weather can induce currents in long transmission lines, stressing transformers and risking blackouts. – Astronaut safety: Crews on the ISS and future lunar missions need rapid, reliable radiation warnings.

The stakes are not hypothetical. According to analyses cited by Lloyd’s of London, a worst-case space weather scenario could cost the global economy trillions of dollars over several years as infrastructure recovers and supply chains reset. While timelines and totals vary by study, the message is consistent: major solar storms have outsized economic impact. You can explore industry risk perspectives in Lloyd’s risk reports library for context and ranges of estimates: Lloyd’s Risk Reports.

Here’s the kicker: early warnings are actionable. A two-hour heads-up can: – Put satellites into “safe mode” to ride out the storm. – Reassign or delay high-risk tasks like spacecraft maneuvers or EVAs. – Reroute or reschedule polar flights to preserve communications. – Adjust grid load or transformer configurations to reduce stress.

That’s why Surya’s combination of spatial prediction (where) and short-term forecasting (when) is compelling.

For background on space weather and official alerts, see NOAA’s Space Weather Prediction Center and NASA’s Heliophysics Division.

How Surya Works (Without the Jargon)

Surya digests a firehose of solar data and turns it into short-term forecasts. At a high level, it’s three big ideas working together.

1) It learns from the Sun directly

Surya is trained on nine years of imagery from NASA’s SDO, a mission that’s been watching the Sun non-stop since 2010. SDO’s instruments capture the solar disk at 4096 x 4096 pixels, every 12 seconds, across multiple wavelengths. That’s incredibly rich: each wavelength highlights different solar layers and processes, from the lower atmosphere to the corona.

Because Surya learns from the raw image data, it avoids a huge bottleneck—manually labeling events. Instead of someone cataloging millions of flares by hand, the model uses self-supervised learning to discover patterns that correlate with flare activity. That makes it adaptable and fast to retrain when new data comes in.

2) It uses a long-short vision transformer with spectral gating

Transformers are the architecture behind many state-of-the-art AI systems (including the large models that power chatbots). Surya uses a specialized variant built for giant images and multiple wavelengths.

  • Long-short vision transformer: Imagine two sets of eyes. One looks at fine details (granulation, active-region structure). The other watches the big picture (global magnetic context) over longer time windows. Surya blends both, so it doesn’t miss tiny signals or global trends.
  • Spectral gating: The model learns how much to “listen” to each wavelength or frequency band at each moment—like a smart audio mixer for the Sun. That way, it amplifies the most informative channels for flare prediction and turns down the noise.

If that analogy helps: predicting flares is like forecasting a thunderstorm by watching turbulent clouds, wind shear, and humidity in real time. Surya observes many “weather channels” of the Sun at once, then ranks which ones matter most in the next couple of hours.

3) It scales to enormous images and relentless cadence

SDO’s 4096 x 4096 frames are about 10 times larger than what most computer vision models train on. And they arrive every 12 seconds—across multiple wavelengths. That’s a deluge.

IBM and NASA built custom data pipelines and training infrastructure to keep up. In practice, that means: – Efficient chunking of large solar images without losing context. – Temporal modeling to track evolving regions across minutes and hours. – Multi-wavelength fusion to correlate features that only “pop” in certain bands.

The payoff: Surya generates pixel-level probability maps that show where on the solar disk a flare is likely to erupt. This is new. It moves space weather from coarse “flare probability today” to visual, short-horizon “flare risk here and soon.”

For coverage of the launch and technical claims, see reporting in New Scientist, Computer Weekly, PR Newswire, and industry outlets such as MarketScreener.

How Surya Compares to Traditional Forecasting

Space weather forecasting has long mixed physics-based models, expert heuristics, and statistical classifiers trained on hand-engineered features. Surya takes a different path.

What’s traditionally used: – Sunspot classification and magnetic complexity (e.g., McIntosh or Mount Wilson classes). – Physics-based models of the solar corona and solar wind. – X-ray flux measurements (GOES) and probabilistic flare forecasts. – Region-based features derived from magnetograms and EUV images.

What Surya adds: – Direct, end-to-end learning from raw multi-wavelength imagery—no heavy feature engineering. – Spatially explicit predictions that pinpoint likely flare sites. – Short-term, two-hour forecasts with a reported 16% accuracy improvement over baseline approaches. – A foundation model approach that can be fine-tuned for related tasks (e.g., coronal hole tracking, CME precursors).

The models aren’t mutually exclusive. In fact, the future likely blends both worlds: – Physics-based models capture known mechanisms and provide interpretability. – Data-driven models exploit patterns that are hard to encode by hand. – Ensemble systems tend to be more robust across solar cycles and instruments.

If you watch space weather for operations, the practical difference is clarity and time. Seeing a heatmap that lights up a specific active region—and getting a two-hour window—lets you make concrete decisions.

For official flare classifications and alerting, NOAA’s quick primers on X-ray flares are helpful: NOAA SWPC Flare Information.

Open-Source Release: Surya and the SuryaBench Dataset

In a field where data access and reproducibility can be tricky, IBM and NASA chose an open path. Both the Surya model and the SuryaBench dataset are available for the community to download, inspect, and extend.

Where to find them: – Model and datasets: Hugging Face – Code and research assets: GitHub – NASA’s heliophysics data portals: NASA Heliophysics

Why open-source matters here: – Faster iteration: Universities, national labs, and companies can fine-tune Surya for their region or industry. – Transparency: Researchers can audit training data, test for bias, and validate performance across solar cycles. – Resilience: Multiple organizations can build tooling around the same core model, reducing single points of failure. – Education: Students and early-career researchers get hands-on with state-of-the-art heliophysics AI.

This wasn’t a small team effort either. The collaboration includes experts from IBM, NASA, Princeton University, the Southwest Research Institute, the University of Colorado Boulder, and several more research institutions. That breadth matters—space weather impacts many domains, from fundamental solar physics to grid engineering and satellite operations.

Who Should Care—and What You Can Do with Surya

Space weather is cross-industry by nature. If your operations depend on timing, navigation, comms, or high-reliability electronics, Surya is worth attention.

Here are concrete use cases by sector:

  • Satellite operators and ground stations
  • Enter safe mode or reconfigure sensitive payloads during high-risk windows.
  • Reschedule propulsion burns and delicate maneuvers.
  • Optimize downlink priorities before or after predicted flare windows.
  • Aviation and airlines
  • Reroute or delay polar flights that rely on HF comms vulnerable to flares.
  • Pre-brief crews for comms degradations and contingency procedures.
  • Coordinate with air traffic management on expected disruptions.
  • Power grid operators
  • Prepare transformers and substations for potential geomagnetically induced currents (GICs), especially if flare activity is paired with CME risk.
  • Adjust load flows or implement temporary reconfigurations to reduce stress.
  • Sync Surya signals with existing NOAA SWPC alerts for layered decision support.
  • Telecommunications and GNSS providers
  • Expect degraded positioning accuracy and plan maintenance or high-precision tasks accordingly.
  • Notify enterprise customers (e.g., precision ag, surveying) of anticipated performance impacts.
  • Space agencies and mission planners
  • Combine Surya outputs with radiation environment models for crewed mission planning.
  • Protect spacewalks (EVAs) by aligning with low-risk windows.
  • Insurers and risk managers
  • Use Surya-derived signals as part of operational risk scoring for satellite fleets and critical infrastructure.
  • Stress-test scenarios tied to solar maximum periods and high activity forecasts.

Surya’s visual predictions make it easier to move from “general risk” to “actionable risk.” That’s the difference between a headline and a playbook.

Limits, Caveats, and What’s Next

As promising as Surya is, no single model will solve space weather. A few important caveats:

  • Flares vs. CMEs: Flares are intense bursts of radiation. Coronal mass ejections (CMEs) are massive plasma eruptions that can drive geomagnetic storms at Earth. Not every flare produces a harmful CME, and vice versa. A two-hour flare forecast doesn’t automatically equal a grid-impact forecast.
  • Lead time: Two hours is a meaningful window, but some mitigations (e.g., grid coordination across regions) may need longer. Surya complements, not replaces, day-ahead and hour-ahead physics-based models.
  • Domain shift: Surya is trained on SDO data. Performance may change with different instruments or data gaps. Adapting to other observatories will require careful validation.
  • Solar cycle variability: Training across nine years is strong, but the Sun’s 11-year cycle changes activity patterns. Continuous retraining helps.
  • Interpretability: Heatmaps are intuitive, but the model’s inner logic is still statistical. Pairing Surya with explainability tools and physical models can increase trust.

What’s next: – Multi-modal fusion: Combine magnetograms, coronagraph data, and in-situ measurements to predict CME likelihood, speed, and direction. – Probabilistic ensembles: Blend Surya with physics-based models for calibrated, scenario-based forecasts. – Longer horizons: Explore architectures that push beyond two hours without sacrificing accuracy. – Operational tooling: Build dashboards, APIs, and alerting systems that plug into mission control, NOCs, and SOCs.

For European context and tools, the European Space Agency maintains a comprehensive portal on operational services and research: ESA Space Weather.

How Organizations Can Prepare Now

You don’t need to be a heliophysicist to put Surya to work. Here’s a practical checklist:

1) Subscribe to authoritative alerts – NOAA SWPC alerts and dashboards: swpc.noaa.gov – NASA mission updates for SDO: sdo.gsfc.nasa.gov – Regional services (e.g., ESA Space Weather): esa.int/Space_Weather

2) Pilot Surya in a sandbox – Download the model and SuryaBench dataset from Hugging Face and code from GitHub. – Run historical backtests on periods relevant to your operations (e.g., past high-activity months). – Evaluate impact metrics: false alarms, lead-time gains, operational cost savings.

3) Build a cross-functional playbook – Define thresholds for action (e.g., probability x duration in a region of interest). – Map actions to roles: who triggers satellite safe modes, who notifies airlines, who adjusts grid load. – Practice: run drills during active solar periods.

4) Integrate and automate – Feed Surya outputs into your existing dashboards or data lakes. – Combine with SWPC alerts and internal telemetry for layered decisions. – Log actions and outcomes to improve thresholds over time.

5) Stay iterative – As Surya evolves, retrain or fine-tune for your environment. – Share findings with the community—open science accelerates everyone.

Key Takeaways

  • IBM and NASA released Surya, an open-source AI that predicts solar flares up to two hours ahead and maps where they’ll likely occur.
  • Trained on nine years of 4K solar imagery from NASA’s SDO, Surya uses a specialized vision transformer with spectral gating to handle massive, multi-wavelength data.
  • It delivers a reported 16% accuracy boost over existing approaches and represents a shift toward spatial, short-horizon forecasting.
  • The stakes are real: solar storms threaten satellites, GPS, aviation, power grids, and astronaut safety; early warnings are operationally valuable.
  • Surya and the SuryaBench dataset are available on Hugging Face and GitHub, inviting researchers and operators to build industry-specific tools.
  • It’s not a silver bullet—pair it with physics-based models and official alerts from NOAA SWPC for robust decisions.

Frequently Asked Questions

Q: What exactly is Surya? A: Surya is an open-source AI foundation model built by IBM and NASA to predict solar flares. It learns from raw, multi-wavelength solar images and produces short-term forecasts with spatial maps of likely flare locations.

Q: How accurate is it, and what’s the lead time? A: The team reports 16% higher accuracy than existing methods, with forecasts up to two hours ahead. That lead time can be enough to protect satellites, adjust grid operations, or reroute polar flights.

Q: Can Surya predict CMEs and geomagnetic storms too? A: Surya focuses on solar flares (radiation bursts). Some flares are associated with CMEs, but not all. Predicting Earth-impacting geomagnetic storms requires CME detection, speed, direction, and coupling with Earth’s magnetosphere—often via complementary models and observations.

Q: How is Surya different from NOAA’s current forecasts? A: NOAA SWPC provides authoritative alerts using a mix of observations, physics-based models, and statistical tools. Surya is a data-driven foundation model that learns directly from raw imagery and generates spatial, short-horizon predictions. The best practice is to use Surya alongside SWPC alerts.

Q: Is the model really open-source? Where can I get it? A: Yes. The Surya model and the SuryaBench dataset are openly available on Hugging Face and GitHub. Check the repositories for documentation, sample notebooks, and licenses.

Q: What data was Surya trained on? A: Nine years of high-resolution imagery from NASA’s Solar Dynamics Observatory (SDO), which captures the Sun at 4096 x 4096 pixels across multiple wavelengths roughly every 12 seconds. Learn more about SDO here: sdo.gsfc.nasa.gov.

Q: Will Surya work with other solar observatories? A: It’s trained on SDO data, so out-of-the-box performance on other instruments may vary. Fine-tuning or domain adaptation will likely be needed to generalize across sensors and vantage points.

Q: How can satellite operators use Surya in practice? A: Integrate Surya’s heatmaps and probabilities into operations dashboards. Define risk thresholds that trigger safe-mode entry, rescheduling burns, or shielding actions. Combine with SWPC alerts and internal telemetry for confirmation.

Q: Does this reduce risk of a major “internet apocalypse” scenario? A: Surya improves short-term awareness and can mitigate certain risks by informing protective actions. But extreme space weather is a complex chain of events. Resilience still requires redundancy, hardening, and coordinated response across infrastructure providers.

Q: Where can I learn more about the risks and economics of space weather? A: Start with NOAA SWPC, NASA’s Heliophysics resources, ESA’s Space Weather, and industry analyses such as Lloyd’s Risk Reports.

The Bottom Line

Surya marks a turning point: open, AI-powered space weather forecasting that’s both visual and practical. It shifts us from rough odds to actionable maps, from passive alerts to proactive playbooks. That’s a big deal for a world leaning ever harder on satellites, timing, and electrified infrastructure.

If you work in aerospace, energy, aviation, or critical infrastructure, now’s the right time to pilot Surya alongside your existing alerts. Start small, measure impact, and build the muscle memory your team will rely on during the next solar surge.

Want more deep dives like this? Keep exploring our latest coverage on AI, climate, and resilience—or subscribe to get the next breakthrough in your inbox.

Discover more at InnoVirtuoso.com

I would love some feedback on my writing so if you have any, please don’t hesitate to leave a comment around here or in any platforms that is convenient for you.

For more on tech and other topics, explore InnoVirtuoso.com anytime. Subscribe to my newsletter and join our growing community—we’ll create something magical together. I promise, it’ll never be boring! 

Stay updated with the latest news—subscribe to our newsletter today!

Thank you all—wishing you an amazing day ahead!

Read more related Articles at InnoVirtuoso

InnoVirtuoso

Hello there! I'm passionate content writer navigating the digital landscape. With a deep love for words and an insatiable curiosity about technology.

With 10 years of experience in IT field, I write mostly about emerging tech, sharing news, informational articles and covering other topics I find interesting.

Let's craft the future of content together – one word at a time!

Browse InnoVirtuoso for more!