NVIDIA described European projects using Earth-2 models for satellite-based forecasting, cyclone risk analysis, air-quality modeling and solar-energy prediction.
NVIDIA’s October 21, 2026, Earth-2 update describes European organizations applying its open AI models and tools to regional forecasting and environmental research. The examples range from short-term cloud prediction to cyclone-risk analysis, air quality and solar-energy forecasting; several are still research or evaluation efforts.
A new regional model, StormScope EU, was trained on European weather and climate data. It uses EUMETSAT satellite observations to predict changing cloud conditions across 16 Meteosat channels at 10-minute intervals. At GTC Berlin, NVIDIA is demonstrating a pipeline built with Earthmover: Earth2Studio runs forecasts, and Earthmover’s platform serves results as map tiles, refreshed as new satellite observations arrive.
The company says developers can combine StormScope for short-term forecasts with CorrDiff to add local detail to coarser weather data. NVIDIA’s open-source PhysicsNeMo framework provides tools for training and customizing models with an organization’s own data and expertise.
The listed deployments have different aims and stages. AXA is using FourCastNet 3 to examine regional cyclone risk across more than 1,000 simulated seasons, with results intended to inform risk assessments. Vaisala Xweather is combining Earth-2 with its observations, lightning data and forecasting expertise; it is adapting StormScope to work with lightning data and plans to make resulting warnings available through its software services and AI-assistant integrations.
Researchers at the University of Manchester adapted CorrDiff for U.K. air-quality modeling, training it on the Isambard-AI supercomputer and running it locally on DGX Spark systems. TotalEnergies is evaluating StormScope-based solar-radiation nowcasts at two photovoltaic sites while developing weather information at subhourly intervals. Separately, Space42 developed SARDiff4Height, which uses CorrDiff and PhysicsNeMo to estimate surface heights from a single radar image and generate uncertainty maps.
Together, these examples show Earth-2 being adapted for regional applications rather than used solely for global forecasting. The work ranges from a live demonstration to exploratory forecasting and research, so the article does not present all projects as production deployments.