Google DeepMind Launches WeatherNext 3 for High-Resolution Hourly Weather Forecasting
Google DeepMind and Google Research have introduced WeatherNext 3, a global AI weather forecasting model that ingests real-time geostationary satellite data to deliver hourly updates at resolutions as fine as 5 kilometers. The model is already integrated into Google Search, the Gemini app, Google Maps, and Earth Engine. By delivering fast, localized forecasts without the high computational cost of traditional numerical weather supercomputers, WeatherNext 3 provides high-fidelity predictions to historically underserved regions like Africa, Latin America, and Asia-Pacific. Furthermore, its specialized data on solar radiation and 100-meter wind speeds directly assists power grid operators and renewable energy planning. WeatherNext 3 achieves a 5 km spatial resolution for primary surface variables such as temperature and humidity, 10 km for other surface metrics, and 25 km for atmospheric variables like wind patterns. It combines continuous geostationary satellite imagery mosaics with sparse weather station observations to maintain physical consistency across regional terrains.
## BACKGROUND
Traditional Numerical Weather Prediction (NWP) models rely on solving complex physical equations on massive supercomputers, making high-resolution global updates computationally expensive and slow. Recent advances in machine learning allow AI models to learn atmospheric dynamics directly from satellite and sensor data, significantly accelerating forecast speed and lowering operational costs.