Google's WeatherNext 3 Uses Satellite Data to Update Forecasts Hourly
By feeding on live satellite observations rather than delayed numerical model output, WeatherNext 3 shortens the gap between observation and forecast to one hour and sharpens resolution to 5km.
Reporting from 1 source: GIGAZINE.
Google DeepMind and Google Research announced WeatherNext 3, an AI weather model that ingests real-time geostationary satellite observations directly instead of learning from numerical model data. This removes the roughly six-hour delay in supercomputer-based data, letting it generate new forecasts hourly. Spatial resolution improves up to about five times over WeatherNext 2, down to 5km intervals for surface temperature and humidity. It is already rolling into Google Search, Gemini, and Google Maps.
Precipitation has been a weak point for global models because rain forms in narrow bands that shift quickly. WeatherNext 3 trains on NASA satellite precipitation data and renders the shape and extent of rain clouds in finer detail than WeatherNext 2, which the article illustrates with side-by-side comparisons against satellite observation.
The model is already live in Google Search, the Gemini app, and Google Maps, and forecast data is available to researchers through BigQuery, Google Earth Engine, and Google Cloud Storage. An experimental map viewer called Weather Lab exists, though it returned a not-available message when accessed from Japan at the time of writing.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.