StormScope EU brings ten-minute satellite nowcasting into NVIDIA's Earth-2 platform

The regional model tracks evolving cloud conditions over Europe, while partners adapt Earth-2 for insurance, energy and environmental research.

NVIDIA introduced StormScope EU as the latest addition to its Earth-2 weather-model family during GTC Berlin. The regional model uses EUMETSAT satellite observations to forecast evolving cloud conditions over Europe across 16 Meteosat channels, with ten-minute intervals.

The 21 October announcement brings a short-term forecasting tool into an ecosystem that also includes global models and regional downscaling. It arrives alongside examples of insurers, energy companies and researchers adapting those components to their own observations and operating questions.

A regional model with a live delivery path

NVIDIA and Earthmover demonstrated a pipeline using Earth2Studio to run forecasts and Earthmover to manage and serve the results. New observations trigger updates every ten minutes, while map tiles let a user explore the output without downloading the full underlying dataset.

That delivery path is a substantial part of a working forecast product. A model may calculate a new field quickly, but a customer still needs a current view, a known observation time and a way to retrieve the relevant region. Serving map tiles supports an interactive application; serving numerical arrays supports analysis and integration with other systems. The two interfaces can address different users while sharing the same forecast run.

The distinction between a global outlook and a local nowcast also shapes application design. An energy planner assessing tomorrow's demand needs a different horizon from an operator anticipating a near-term change in cloud cover. A model family gives developers components for those different tasks rather than a single forecast intended to answer everything.

Insurers and energy operators adapt the models

AXA is using FourCastNet 3 to study regional cyclone risk across more than a thousand simulated seasons. Vaisala Xweather is combining Earth-2 with its own observations, lightning data and forecasting expertise for weather-intelligence services. TotalEnergies is evaluating StormScope-based solar-radiation nowcasts at two photovoltaic sites.

These projects use different kinds of evidence. Simulated cyclone seasons expand the scenarios available to risk analysts. Lightning observations can enrich a short-term warning product. Solar-radiation nowcasts can become an input to an energy-generation estimate. Their common feature is the combination of a broadly trained model with a specific organisation's data and decisions.

For a developer building a solar application, for example, the forecast is one stage in a chain. Weather output must be aligned to the plant's location and time intervals, then combined with the installation's physical characteristics and operating record. That work determines whether a general weather capability becomes a useful operational forecast.

Environmental modelling reaches smaller systems

The University of Manchester adapted the CorrDiff architecture for U.K. air-quality modelling, training on Isambard-AI and running locally on DGX Spark. Space42 used CorrDiff and PhysicsNeMo in research estimating surface heights from a radar image, including uncertainty maps.

Those examples show that the platform's role extends beyond a weather map. Training and deployment can take place on different scales of computing, and a shared architecture can be adapted to another environmental variable.

The immediate contribution of StormScope EU is a more geographically specific short-term model and an example of how to deliver its changing output. For application teams, the next integration work is concrete: select the observation stream, preserve the timing and version of each run, and connect the forecast to the question their users actually need answered.

Sources

- [Pioneers Across Europe Deploy NVIDIA Earth-2 Open Models](https://blogs.nvidia.com/blog/europe-earth-2-open-models-weather-climate-prediction/) — NVIDIA