Google Puts WeatherNext 3’s Hourly AI Forecasts Into Search, Maps and Gemini

Google Research and Google DeepMind have introduced WeatherNext 3, an AI weather model that generates new global forecasts every hour from live geostationary satellite data and historical analysis. Google says it will begin powering weather experiences worldwide across Search, Gemini, Maps, Maps Platform’s Weather API and Earth Engine. The practical wager is that faster refreshes and finer local detail can make a general-purpose product layer more useful for rain, wind and energy planning.

An hourly forecast built from observations

The model also trains directly on sparse weather-station observations to account for local terrain. It produces temperature and moisture forecasts on a 5-kilometer grid, other surface variables at 10 kilometers, and atmospheric measures such as wind speed at 25 kilometers. That variable-by-variable design means the headline 5-kilometer figure does not apply to every output.

Precipitation is a key test

Google reports improvements in Continuous Ranked Probability Score, a measure for probabilistic forecasts, of up to 60% against IMERG, 30% against MRMS and 10% against rain-gauge measurements at early lead times. It also says forecasts made at least a day ahead are up to 50% more accurate for precipitation, with the largest gains in places where forecasts have historically been less reliable. Those are company-reported evaluation results, not a public-safety guarantee.

One model, several routes to use it

It also adds variables designed for renewable-energy planning: 100-meter wind speeds, roughly turbine height, plus cloud cover and solar radiation. Google says those inputs can help estimate wind and solar output and match expected generation with demand. The same detail could be useful across many operational settings, but the release does not establish results for a specific grid operator, business or region.

Official guidance remains with weather agencies