Google DeepMind has introduced WeatherNext 3, a cutting-edge AI-based weather model that improves forecast accuracy, resolution, and frequency. This new model will power weather insights in Google Search, Google Maps, Gemini, and be accessible via Google's cloud services.
- Highest accuracy in AI weather forecasts tested against top models
- Predicts key weather metrics down to 5km every hour
- Integrates raw satellite data for real-time updates across Google products
What happened
Google DeepMind and Google Research have launched WeatherNext 3, an advanced AI weather model designed to offer more precise and timely weather forecasts. The model improves on prior versions by increasing resolution to 5 kilometers, enhancing rain prediction accuracy by 60%, and providing hourly forecast updates instead of every six hours.
This model will begin feeding weather information visible to users on Google Search, Google Maps, and Gemini, and will be available to researchers and developers through Google’s cloud platforms. WeatherNext 3 notably incorporates raw satellite observational data, a first for AI-driven high-resolution global weather forecasting, allowing it to update forecasts in near real-time.
Why it matters
Traditional weather forecasts rely on government supercomputers processing physics-based models which, while accurate, can be slow and costly. WeatherNext 3 leverages deep learning to offer comparable or better accuracy with faster processing and finer spatial detail, addressing limitations like broad forecast areas and low temporal frequency common to earlier AI models.
With improved resolution and real-time data integration, WeatherNext 3 provides more actionable localized weather insights such as precise rainfall forecasts and cyclone tracking. This advancement also brings AI forecasting closer to practical verification by targeting predictions to specific weather stations, enhancing trust in AI-generated weather data.
What to watch next
Observing how widely Google deploys WeatherNext 3’s updates across its consumer-facing platforms will be key, as well as how other weather AI models evolve in response. The improvement in rain forecasts and hourly updates could redefine user expectations for weather services globally.
Further innovation may focus on integrating more direct observational data inputs and expanding the granularity and types of weather variables predicted. Monitoring the model's performance compared to government and private-sector forecasts will provide insight into AI’s role in the future of meteorology.