Google DeepMind's WeatherNext 3 AI Delivers Hourly Global Forecasts

Google DeepMind and Google Research released WeatherNext 3 on September 3, 2026, an AI weather forecasting model designed to provide high-resolution global updates every hour. The system is being integrated across Google Search, Gemini, Maps, and Google Cloud platforms to deliver more precise, real-time environmental data to users and enterprises.

Unlike previous iterations, WeatherNext 3 incorporates raw, real-time geostationary satellite data. This shift allows the model to generate forecasts on an hourly basis, a significant increase in frequency compared to the standard six-hour increments used by traditional numerical weather prediction (NWP) models and earlier AI versions. In terms of precision, the model achieves a spatial resolution of 5km for surface variables such as temperature and moisture, 10km for other surface variables, and 25km for atmospheric variables like wind speed. Google reports that precipitation forecasting accuracy has seen a substantial boost, with a Continuous Ranked Probability Score (CRPS) improvement of up to 60% when measured against NASA's IMERG data.

The Shift to Direct Data Assimilation

Traditional weather forecasting relies on supercomputers to solve complex physics equations, a process that is computationally expensive and often results in a six-hour data lag. While earlier AI models, including GraphCast, were trained on the outputs of these NWP models, WeatherNext 3 moves toward "direct data assimilation." By training on raw satellite mosaics and sparse weather station data, the model bypasses the constraints of traditional physics-based simulations.

Ferran Alet, a staff research scientist manager at DeepMind, noted that machine learning is uniquely suited for this task because it "targets the problem we are really solving, which is approximate noisy physics from incomplete information and finite compute," allowing the system to learn patterns directly from vast datasets.

Industry and Global Impact

This increase in resolution and update frequency is critical for predicting fast-moving weather events and accounting for local topographical variations, such as those found in coastlines and valleys. The accessibility of this AI-driven approach is particularly impactful for underserved regions in Latin America, Africa, and the Asia-Pacific, where the prohibitive cost of supercomputing often makes high-resolution traditional models inaccessible.

Beyond general consumer use, the model provides specific utility for the energy sector. By offering detailed forecasts for solar radiation and 100-meter wind speeds, WeatherNext 3 provides essential data for the management of renewable energy grids. Samier Merchant, a senior staff engineer at Google, stated that this marks the first time these core variables will power a wide array of Google products.

Future Outlook

As Google rolls out WeatherNext 3 across its ecosystem, the industry will be watching how direct data assimilation performs during extreme weather anomalies compared to traditional NWP models. While the technical metrics show significant improvement in precipitation and resolution, the long-term reliability of bypassing physics-based simulations for global forecasting remains a key point of interest for the meteorological community.