Google DeepMind's WeatherNext 3 learns from live satellite data: hourly 5 km global forecasts and up to 60% better precipitation skill
On Sept 3, 2026 Google DeepMind and Google Research released WeatherNext 3. It ingests live geostationary satellite mosaics and trains directly on station observations, producing a new global forecast every hour at up to 5 km resolution (about 5x sharper than WeatherNext 2). Precipitation CRPS improves by up to 60% against IMERG. Google calls it the most accurate global weather model on Brightband's independent live leaderboard, and it powers Search, Gemini, Maps and Earth Engine.
Key facts
- Announced Sept 3, 2026; paper arXiv 2609.03582
- Inputs: live 1-hour geostationary satellite mosaics plus historical analysis, into a single Functional Generative Network (FGN) mesh transformer
- Hourly forecasts; 5 km for key surface variables, 10 km for other surface variables, 25 km for atmospheric variables (WeatherNext 2: 25 km, 6-hourly)
- Precipitation CRPS improvement up to 60% vs IMERG, 30% vs MRMS, 10% vs rain gauges at early lead times; up to 50% more accurate precipitation forecasts a day or more ahead for users
- New clean-energy variables: 100 m wind speed, cloud cover, surface solar radiation
- Ranked top on Brightband's live Operational WeatherBench, per Google; deployed in Search, Gemini, Maps, Maps Platform Weather API, Earth Engine, BigQuery
Science result
- Field
- climate-weather / numerical weather prediction
- Problem
- Global high-resolution, rapidly updated weather forecasting from real observations
- Result
- Hourly global forecasts at up to 5 km trained on live satellite and station data, with large precipitation-skill gains
- AI system
- WeatherNext 3
- Human role
- Human-led research; AI model
- Verification
- Independent live leaderboard (Brightband) + preprint
- Status
- confirmed
What happened
DeepMind's weather model stopped depending only on physics-model reanalysis and learns directly from real-time observations, which removes the six-hour data lag of numerical weather prediction.
Why it matters
It is a step from AI emulating weather simulators to AI forecasting from raw observations, with global 5 km detail that regions without supercomputing budgets have lacked.
Changelog
- 2026-09-29: created
Related events
Sources (4)
- officialGoogle: Introducing WeatherNext 3
- paperarXiv 2609.03582: WeatherNext 3
- docsBrightband Operational WeatherBench live leaderboard
- pressTechCrunch: Google's latest AI weather model
id: 2026-09-03-weathernext-3 · updated 2026-09-29 · open in the interactive timeline