Pioneers Across Europe Deploy NVIDIA Earth-2 Open Models to Power Real-World Weather and Climate Prediction

NVIDIA added StormScope EU to its Earth-2 family of open weather and climate models. It's a short-term forecasting model trained on European data that uses EUMETSAT satellite observations to predict evolving cloud conditions across 16 Meteosat channels at 10-minute intervals.

Alongside it, AXA, Vaisala Xweather, TotalEnergies, Space42 and the University of Manchester described how they are adapting Earth-2 models. Global forecasts show the storm. Decisions about solar output, shipping and insured property depend on what happens over one valley in the next hour, and that's where these tools are aimed.

We're broadly enthusiastic here. Weather is one of the clearest cases where AI earns its keep: the physics is real, the data is abundant, and better short-range forecasts help grid operators, emergency planners and ordinary people. Publishing these as open models that organizations can retrain with their own data is a healthier pattern than a single black-box forecast sold by subscription.

The Manchester example is the one I keep coming back to. Researchers adapted the CorrDiff model for U.K. air-quality modeling, trained it on the Isambard-AI supercomputer, then ran it locally on desktop-class DGX Spark machines. Moving environmental modeling from national infrastructure to a lab bench widens who can do this work.

Now the caution. AXA is using FourCastNet 3 to study cyclone risk across more than 1,000 simulated seasons, beyond the historical record. That's valuable for understanding tail risk. It's also how pricing models get built, and if simulated seasons start driving premiums, the assumptions inside those simulations become a consumer issue. We'd like insurers to be transparent about where AI-generated scenarios feed underwriting.

There's a quieter risk too: AI forecasts can look confident where they're weakest. Space42's work producing uncertainty maps alongside its radar-based height estimates is the right instinct, and we'd like to see uncertainty become standard output for every one of these models, not a research extra.

Most of what's described is still exploration. TotalEnergies is evaluating solar nowcasts at two sites; Vaisala is adapting StormScope with lightning data. The direction is good. The proof is in seasons of verified forecasts.

Earth-2 is NVIDIA's platform of open AI weather and climate models and tools, including StormScope for short-term forecasting, CorrDiff for adding local detail to coarse data, and the PhysicsNeMo framework for customization. At GTC Berlin, NVIDIA showed a live StormScope EU demo built with Earthmover, refreshed as new satellite observations arrive every 10 minutes.