WeatherNext 3: Our most advanced global weather AI model — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google DeepMind and Google Research introduced WeatherNext 3 as their most advanced global weather AI model, with Brightband's independent live evaluations confirming it as the most accurate to date.
- WeatherNext 3 generates hourly forecasts at a native 5km resolution for surface variables like temperature and moisture — roughly 5x sharper than WeatherNext 2's 25km grid updated every 6 hours.
- The model trains directly on live, hourly geostationary satellite mosaics rather than 6-hour-lagged numerical weather prediction models, and on sparse weather station data to resolve local topography.
- Precipitation accuracy improved significantly, with the model scoring up to 60% better CRPS against NASA's IMERG satellite retrievals, 30% against MRMS, and 10% against rain gauges for early lead times.
- Renewable energy operators gain turbine-height 100-meter wind speed forecasts alongside high-resolution cloud cover and solar radiation data for wind and solar farm output planning.
- WeatherNext 3 is rolling out across Google Search, the Gemini app, Google Maps, Google Maps Platform Weather API, and Google Earth Engine starting today, offering up to 50% more accurate longer-term precipitation forecasts.
- Latin America, Africa, and Asia-Pacific — regions historically underserved by high-resolution forecasting due to supercomputing costs — now get access to 5km-resolution global forecasts trained on raw satellite observations.
Why it matters: For grid operators and renewables developers, turbine-height wind speeds and high-resolution solar radiation forecasts turn AI weather prediction into operational planning for clean energy assets. For billions in Latin America, Africa, and Asia-Pacific, the 5km resolution bypasses the supercomputing economics that kept those regions stuck with coarser forecasts.
Ask SkimNews
