European Organizations Deploy NVIDIA Earth-2 Open Models for Real-World Climate and Weather AI

At GTC Berlin, NVIDIA announced that companies and research institutions across Europe are deploying open artificial intelligence models from its Earth-2 ecosystem to build customized weather forecasting, risk assessment, and environmental monitoring systems.

Alongside the announcement, NVIDIA introduced StormScope EU, the latest addition to the Earth-2 family specifically tailored for regional short-term weather predictions across Europe.

Real-Time Regional Weather with StormScope EU

Traditional global climate models provide broad projections, but localized decisions—such as managing solar farm outputs, rerouting cargo ships, or assessing insured property risks—require granular data. StormScope EU addresses regional short-term forecasting by training directly on European climate datasets.

The model ingests satellite observations from EUMETSAT across 16 Meteosat channels to predict evolving cloud dynamics at 10-minute intervals. To demonstrate these real-time capabilities at GTC Berlin, NVIDIA showcased an interactive pipeline created in partnership with data platform Earthmover. Powered by the open-source Earth2Studio framework, the system updates predictions every 10 minutes as new satellite imagery arrives and delivers interactive map tiles directly to users without requiring large dataset downloads.

Developers can integrate StormScope for nowcasting and layer it with NVIDIA’s CorrDiff model to apply generative AI super-resolution to coarser climate inputs.

Commercial Deployments in Insurance and Energy

Several major European organizations are using Earth-2 model architectures to tailor predictions using proprietary datasets:

• AXA: The insurance leader is leveraging Earth-2’s FourCastNet 3 model to evaluate regional cyclone risks beyond historical records. By simulating over 1,000 climate seasons, AXA generates risk insights to optimize customer prevention efforts and risk pricing.
• TotalEnergies: The energy company is evaluating StormScope-based solar-radiation nowcasts across two photovoltaic (PV) sites. TotalEnergies is developing subhourly weather forecasting tools to better anticipate fluctuations in solar-power generation.
• Vaisala Xweather: The weather intelligence provider is combining Earth-2 with proprietary observation data and lightning telemetry. By integrating lightning data into StormScope, Vaisala aims to extend short-term warning lead times across its Software-as-a-Service (SaaS) platforms and AI-assistant tools for transportation, maritime, and energy sectors.

Desktop Environmental Modeling and Radar Heights

Beyond commercial forecasting, research groups are applying Earth-2 tools to specialized physics and mapping challenges using the open-source NVIDIA PhysicsNeMo framework:

• University of Manchester: Researchers adapted the CorrDiff generative model for U.K. air-quality modeling. The team trained the architecture on the NVIDIA-powered Isambard-AI supercomputer and deployed it locally on desktop-class NVIDIA DGX Spark systems, bringing complex environmental modeling directly to local workstations.
• Space42: Using CorrDiff and NVIDIA PhysicsNeMo, Space42 created SARDiff4Height, a model designed to estimate surface terrain heights from a single radar image while outputting uncertainty maps to pinpoint areas that require manual analyst review.

By providing modular tools like Earth2Studio and open model architectures, NVIDIA aims to accelerate domain-specific climate solutions across Europe's public and private sectors.