Google Launches WeatherNext 3: A More Accurate and Granular AI-Driven Weather Forecasting Model
Scientists at Google DeepMind and Google Research have unveiled a new artificial intelligence model for weather forecasting called WeatherNext 3. This advanced system is designed to provide more accurate, high-resolution forecasts by leveraging deep learning techniques. The model outperforms both traditional government-based systems like the US National Weather Service and ECMWF as well as other AI models from companies such as Microsoft and Nvidia in terms of accuracy on Operational WeatherBench. WeatherNext 3 improves upon its predecessor with a resolution down to 5km, enhanced rain prediction by 60%, and hourly forecasts instead of every six hours. It also incorporates real-time satellite data for more frequent updates. The model is now integrated into Google's ecosystem including Search, Maps, and Gemini, while being available on cloud platforms for broader use. The improvements are attributed to a larger neural network with 2.4 times more parameters than the previous version and targeted training methods that align forecasts closer to real-world data stations.
