Google DeepMind Introduces WeatherNext 2 for Faster, More Accurate Forecasting

Google DeepMind and Google Research have introduced WeatherNext 2, a new AI-powered weather forecasting model designed to deliver faster and more precise global predictions. The system generates forecasts up to eight times faster than traditional physics-based models, offering hourly resolution and improved accuracy across nearly all weather variables and lead times.

WeatherNext 2 uses a novel approach called a Functional Generative Network, which introduces controlled noise into the model to simulate hundreds of possible weather scenarios from a single starting point. This enables the model to capture both individual weather elements and complex, interconnected systems, supporting more detailed predictions such as regional heatwaves or wind farm output.

The model’s forecast data is now accessible through Google Earth Engine and BigQuery, with an early access program available on Google Cloud’s Vertex AI platform for custom model inference. In addition, WeatherNext technology has been integrated into Google’s consumer and developer products, including Search, Gemini, Pixel Weather, and the Google Maps Platform Weather API.

The release marks an expansion of DeepMind’s work in applying AI to geospatial and environmental challenges, following previous efforts to support weather agencies with experimental cyclone prediction tools.