Skip to content
Neural Daily

ModelsNews

StormScope EU: open short-range forecasts for Europe

At GTC Berlin NVIDIA introduced StormScope EU, the European short-range forecaster in its Earth-2 weather model family. It predicts cloud conditions from Meteosat data every 10 minutes, under an open license, and runs on your own GPUs.

By: Mikola Máté Patrik · · 3 min read

NVIDIA Earth-2 examples: AXA hurricane tracks, historical record vs AI ensemble, a Vaisala globe forecast and an Earthmover cloud map of Europe
Image: NVIDIA · Press kit · Official blog hero image · Source

What changed

At GTC Berlin NVIDIA introduced StormScope EU, the newest member of its Earth-2 open weather model family (NVIDIA blog). It is a regional version of the StormScope short-range forecasting model, trained on European weather and climate data.

The model works from EUMETSAT satellite observations. It predicts evolving cloud conditions across 16 Meteosat channels at 10-minute intervals.

Under the hood

It is available on Hugging Face as StormScope-Meteosat (model card). According to the card it is a diffusion transformer (DiT) with 2D neighborhood attention. Two checkpoints ship:

  • European grid (1120x2432): 306 million parameters,
  • EMEA grid (4672x4160): 176 million parameters.

It takes 6 time steps of Meteosat FCI data as input, at 2 km resolution at the sub-satellite point. The license is OpenMDW 1.1, which allows commercial use. Inference runs through the Earth2Studio library, and customization through the open PhysicsNeMo framework.

Who is using it

  • AXA uses FourCastNet 3 to study regional cyclone risk across more than 1,000 simulated seasons, beyond the historical record.
  • Vaisala Xweather adds lightning data to StormScope to extend short-term warning lead time.
  • The University of Manchester adapted CorrDiff for UK air-quality modeling, trained it on the Isambard-AI supercomputer and runs it on DGX Spark.
  • TotalEnergies is evaluating StormScope-based solar radiation nowcasts at two photovoltaic sites, at subhourly intervals.
  • Space42 built SARDiff4Height on CorrDiff to estimate surface heights from a single radar image, with uncertainty maps.

A live demo with Earthmover runs at the conference: StormScope EU updates as each new satellite observation arrives, and results load as map tiles without downloading whole datasets.

Why it matters

A global forecast shows the storm, but a solar farm, a shipping operator or an insurer needs a local picture within the hour. An open model of around 300 million parameters that runs on your own GPUs and can be fine-tuned on your own data targets exactly that need. NVIDIA gave no accuracy numbers for the European version in the announcement, so validate it on your own measurements before production use.

Try it

The Earth2Studio docs include forecasting examples, and StormScope and CorrDiff can be chained in one workflow: short-range forecast first, then local downscaling. AXA's case study is here.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles