NVIDIA Earth-2 Adds StormScope EU for 10-Minute European Weather AI
NVIDIA expands Earth-2 in Europe with StormScope EU forecasts every 10 minutes, as AXA, Vaisala, Manchester and TotalEnergies test new climate AI tools.
Summary
NVIDIA’s October 22, 2026 Earth-2 update adds StormScope EU, a short-term model trained on European weather and climate data. Using EUMETSAT observations, it predicts changing clouds across 16 Meteosat channels every 10 minutes. At GTC Berlin, an interactive pipeline built with Earthmover uses Earth2Studio for forecasting and Earthmover’s platform to process each new observation, update results every 10 minutes and serve European predictions as map tiles without full dataset downloads.
Developers can combine StormScope with CorrDiff for localized forecasts or customize architectures through the open source NVIDIA PhysicsNeMo framework using proprietary data and expertise, including for air-quality applications. AXA uses FourCastNet 3 to assess regional cyclone risk across more than 1,000 simulated seasons and guide prevention. Vaisala Xweather combines Earth-2 with proprietary observations, lightning data and forecasting expertise to extend warning lead times for government, energy, maritime, transportation and insurance customers through software services and AI assistants. University of Manchester researchers trained CorrDiff for U.K. air quality on the NVIDIA-powered Isambard-AI supercomputer, then ran it locally on NVIDIA DGX Spark systems. TotalEnergies is evaluating StormScope solar-radiation nowcasts at two photovoltaic sites and developing subhourly weather models. Space42 built SARDiff4Height with CorrDiff and PhysicsNeMo to estimate surface heights from one radar image and produce uncertainty maps.
Positives
- StormScope EU updates predictions across 16 Meteosat channels every 10 minutes using EUMETSAT satellite observations.
- AXA is testing FourCastNet 3 across more than 1,000 simulated seasons to improve regional cyclone-risk assessments and prevention planning.
- Vaisala Xweather is adding lightning data to StormScope to extend warning lead times across five customer sectors.
- University of Manchester moved its CorrDiff air-quality model from the Isambard-AI supercomputer to local NVIDIA DGX Spark systems.
- TotalEnergies is evaluating subhourly solar forecasting at two photovoltaic sites to anticipate generation changes.
- SARDiff4Height estimates surface elevation and uncertainty from a single radar image.
Risks & concerns
- StormScope EU accuracy benchmarks are not disclosed for its forecasts across 16 Meteosat channels.
- TotalEnergies’ solar-radiation work remains an evaluation limited to two photovoltaic sites.
- AXA’s analysis extends beyond historical records through simulated seasons rather than additional observed events.
- SARDiff4Height produces uncertainty maps because some surface-height estimates require closer scrutiny.
- Production rollout dates and commercial performance results remain undisclosed for the showcased projects.