Key facts
- Google's WeatherNext 3 AI model has been released, improving upon its predecessor.
- The new model demonstrates superior accuracy compared to other AI models and traditional forecasts.
- WeatherNext 3 can predict weather with a resolution of 5km and provides hourly forecasts.
- It is the first AI model to directly incorporate raw observations for high-resolution global forecasting.
- The model will be integrated into Google products like Search, Maps, and Gemini.
Google DeepMind and Google Research have unveiled WeatherNext 3, a new artificial intelligence model designed for enhanced weather forecasting. This advanced model aims to provide clearer insights into atmospheric behavior and more frequent predictions, building upon deep learning techniques that are revolutionizing meteorology.
WeatherNext 3 has demonstrated superior accuracy in tests on Operational WeatherBench, outperforming not only other deep-learning models from entities like Microsoft and Nvidia but also traditional forecasts from the US National Weather Service and the European Center for Medium-Range Weather Forecasting. The model excels in predicting key variables such as temperature, wind speed, and humidity.
Unlike conventional weather forecasts that rely on expensive and time-consuming supercomputers, WeatherNext 3 leverages machine learning to learn patterns from vast datasets, enabling faster and comparably accurate predictions. This approach addresses the chaotic nature of weather by approximating noisy physics from incomplete information.
Key advancements in WeatherNext 3 include a higher resolution of 5km, a 60% improvement in rain prediction over its predecessor, and the ability to generate hourly forecasts. The model also incorporates raw weather observations directly, a first for high-resolution global forecasting, though it still relies on national weather datasets for full assimilation. This development is part of a broader transformer revolution in meteorology, promising economic benefits for regions lacking advanced forecasting infrastructure.
Google plans to integrate WeatherNext 3 into its popular products, including Google Search, Google Maps, and Gemini. It will also be accessible to users and researchers via Google's cloud platforms. Experts like Bill Gates have highlighted AI-powered weather forecasting as a critical application of the technology, with potential benefits for agriculture in developing countries and for optimizing renewable energy projects.