Google DeepMind and Google Research have unveiled a new AI weather forecasting model, WeatherNext 3. It will be integrated into Google Search, Google Maps, and Gemini, and will also be available to users and researchers through the company's cloud platforms.
WeatherNext 3 outperforms previous deep learning models from Google, Microsoft, Nvidia, and the European Centre for Medium-Range Weather Forecasts, as well as traditional forecasts from the US National Weather Service. Testing was conducted on the Operational WeatherBench platform for comparing weather forecasts. The model evaluates temperature, wind speed, and humidity with greater accuracy. The new model addresses three key limitations of earlier AI models. It produces forecasts at a resolution of up to 5 km, compared to the previous 15–25 km. Precipitation assessment has improved by 60% compared to WeatherNext 2. Instead of the standard six-hour time step, the model generates hourly forecasts.
These improvements were achieved through several technical advances. WeatherNext 3 contains 2.4 times more parameters than its predecessor and has been fine-tuned for specific weather stations. The model is capable of processing data from meteorological satellites received on an hourly basis and using it as raw observational input. According to Google, this is the first AI model to directly use raw observational data to produce a high-resolution global forecast. That said, WeatherNext 3 still partially relies on national weather datasets, meaning additional work will be required for full integration of direct observations. It is noted that the speed and low cost of AI-based forecasts could be particularly valuable for regions where traditional meteorological infrastructure is too expensive.
More accurate wind, rainfall, and cloud cover forecasts will also aid in renewable energy planning. Researchers believe that the advancement of AI in meteorology is gradually bringing the forecasting challenge closer to real user needs. New models allow for faster access to detailed weather information and better preparation for adverse conditions.