Google has introduced WeatherNext 3, an AI forecasting model that incorporates current satellite observations and produces a new global forecast every hour. The company says the model creates a picture with five times the spatial resolution of its previous system, improving detail around local rain and snowfall.
Traditional numerical weather prediction uses supercomputers to solve equations that simulate the atmosphere. AI systems instead learn patterns from large stores of historical observations, which can generate forecasts more quickly. WeatherNext 3 combines that learned approach with fresher observational data so it does not rely only on the conditions represented in its training set.
Google says the updated model improves precipitation forecasts in particular and can provide higher-resolution predictions with less delay. Hourly refreshes could be useful when conditions change rapidly, while the global scope allows the same system to cover regions with very different weather patterns.
The performance claims come from Google, and a sharper model does not remove uncertainty from weather prediction. Forecast quality still depends on available observations, and unusual local events can remain difficult to anticipate. WeatherNext 3 should therefore be understood as another input for forecasting services and meteorologists, not a replacement for warnings issued by public weather agencies. Its practical impact will depend on how the model performs across seasons, regions, and severe events after deployment.