WeatherNext 3 brings hourly updates and finer detail to AI forecasts
Google's latest weather model uses real-time satellite observations and produces hourly forecasts. Selected surface variables reach a five-kilometer grid, while other predictions use coarser resolutions.
Google DeepMind and Google Research introduced WeatherNext 3 on September 3. The model incorporates raw satellite observations to produce updated forecasts every hour, addressing a weakness of systems that struggle to absorb rapidly changing local conditions.
The detail varies by prediction. Temperature and moisture can be represented on a five-kilometer grid, other surface variables at ten kilometers, and atmospheric variables at 25 kilometers. WeatherNext 2 used a 25-kilometer grid with six-hour increments, so the new model's finest outputs represent a substantial change in both space and time.
Google says the system also improves precipitation forecasting and adds variables useful for clean energy. It is being integrated across Search, Gemini, Maps, Google Maps Platform and Cloud.
This is a useful example of AI advancing beyond chat interfaces. Resolution describes the grid, though; it does not guarantee equally accurate predictions for every location or weather event.