What was released
NVIDIA announced on its blog that StormScope EU is now the newest model in Earth-2, its family of open AI models and tools for weather and climate work. It is a regional version of the existing StormScope model, built for short-term forecasting and trained on European weather and climate data.
The model uses EUMETSAT satellite observations to predict how cloud conditions will evolve across 16 Meteosat channels, at 10-minute intervals. The source does not give a release date, licence terms or pricing for StormScope EU, so those remain unconfirmed.
How it fits together
At GTC Berlin, NVIDIA is showing a live StormScope EU pipeline in a demo built with Earthmover. Earth2Studio runs the forecasts, and Earthmover's data platform manages and serves the results, refreshing whenever new satellite observations arrive. Results are served as map tiles, so visitors can browse predictions across Europe without downloading whole datasets.
Developers can combine Earth-2 models in one workflow. StormScope handles short-term forecasting, while CorrDiff adds local detail to coarser weather data. The open source PhysicsNeMo framework supplies training and customisation tools, so teams can retrain the architectures on proprietary data or for new uses such as air-quality modelling.
Who is building on it
The post lists several organisations working with Earth-2:
- AXA uses FourCastNet 3 to study regional cyclone risk across more than 1,000 simulated seasons, to inform risk assessments and prevention efforts.
- Vaisala Xweather is combining Earth-2 with its own observations and lightning data, adapting StormScope to extend short-term warning lead time and delivering it via its SaaS offerings and AI-assistant integrations.
- University of Manchester adapted CorrDiff for UK air-quality modelling, trained it on the Isambard-AI supercomputer and ran it locally on NVIDIA DGX Spark systems.
- TotalEnergies is evaluating StormScope-based solar-radiation nowcasts at two photovoltaic sites and developing models for subhourly weather information.
- Space42 built SARDiff4Height with CorrDiff and PhysicsNeMo to estimate surface heights from a single radar image, with uncertainty maps.
What to take from it
The examples are mostly pilots and research, not finished products, and the source offers no performance figures. The commercially relevant thread is Vaisala Xweather's plan to deliver StormScope-based warnings through software and assistants, which is how most operators would consume this.