Google's WeatherNext 3 sharpens forecasts to 5-km hourly grid

Google DeepMind and Google Research have released WeatherNext 3, an AI weather model that runs hourly global forecasts at 5-kilometer resolution for surface variables like temperature and moisture. The post, authored by "The WeatherNext team," calls it "a global weather picture roughly five times sharper than our previous model," WeatherNext 2, which ran on a 25-kilometer grid with 6-hour increments.

Google bills it as "the most advanced and accurate global weather model to date, according to independent live evaluations by Brightband," a public leaderboard for forecast systems. On precipitation, the team reports Continuous Ranked Probability Score improvements of "up to 60% against IMERG, 30% for MRMS, and 10% against rain gauge measurements." For lead times of a day or more, users "will see up to 50% more accurate precipitation forecasts."

The model breaks from pure physics simulation by ingesting a mosaic of live geostationary satellite data and training directly on sparse weather station observations. Alongside the accuracy work, it adds 100-meter wind speed predictions aimed at wind energy, plus cloud cover, solar radiation, and cyclone track outputs. Google is routing WeatherNext 3 into Google Search, the Gemini app, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine, with bulk data through BigQuery and Google Cloud Storage.

The blog post does not name individual researchers, disclose a maximum forecast horizon, or publish per-region accuracy figures.