Technology

Google Unveils WeatherNext 3, Boosting Forecast Accuracy in Search and Gemini

By Abner Li

Google has introduced WeatherNext 3, described as its most advanced global weather AI model to date, with immediate applications in Search, Maps, and the Gemini app. Developed by Google DeepMind and Google Research, the new system builds on WeatherNext 2, which launched in November 2025.

Unlike previous models, including WeatherNext 2, which relied on data from numerical weather prediction (NWP) systems, WeatherNext 3 draws on a network of live, global geostationary satellite data. Traditional NWP models depend on supercomputer-driven physics simulations that carry roughly a six-hour data lag, which can introduce inaccuracies for fast-changing conditions such as rainfall or surface temperature. By learning directly from real-time satellite observations, WeatherNext 3 provides a continuously updating view of the atmosphere, enabling more timely and localized forecasts.

The model produces hourly forecasts at several spatial resolutions: temperature and moisture readings at 5 kilometers, surface variables at 10 kilometers, and wind speed at 25 kilometers. This marks a significant improvement over WeatherNext 2, which operated on a 25-kilometer grid with 6-hour update intervals. Google says the new model’s output is roughly five times sharper, allowing for earlier detection of fast-developing weather events such as storms, fronts, and precipitation systems.

A major focus of the update is precipitation forecasting. WeatherNext 3 was trained using two high-quality data sources: NASA’s satellite-based Integrated Multi-satellite Retrievals for GPM (IMERG) and Google’s own global precipitation reanalysis based on satellite radar. In testing, the model achieved a Continuous Ranked Probability Score (CRPS) improvement of up to 60% compared to IMERG, 30% compared to MRMS, and 10% compared to rain gauge measurements for early forecast periods.

Google noted that these advances could particularly benefit regions in Latin America, Africa, and Asia-Pacific, which have historically lacked access to high-resolution forecasting due to the high computing costs of traditional regional models. The company said the technology could bring localized, high-fidelity forecasts to billions of people and local businesses in these areas.

The model also introduces forecasts tailored to renewable energy production, including 100-meter wind speed predictions—approximately turbine height—for more precise wind-energy output, as well as high-resolution cloud cover and solar radiation estimates to help solar farms predict energy generation.

For everyday users, Google says WeatherNext 3 will significantly improve longer-term forecasts available through Search, Maps, and the Gemini app. When planning a day or more in advance, users can expect precipitation forecasts that are up to 50% more accurate, with the largest gains occurring in regions where forecasting has traditionally been less reliable.