Google DeepMind and Google Research have officially presented the advanced WeatherNext 3 model, which, according to independent evaluations by Brightband, represents the most advanced and accurate global weather model to date. The model learns directly from real-world observations to deliver rapid, localized forecasts for weather events.

Key Features of WeatherNext 3:
Unprecedented Spatial and Temporal Precision: The model generates hour-by-hour forecasts at multiple spatial resolutions, maintaining physical balance between global winds and local topography.

Advanced Interface: Key variables such as temperature and humidity are displayed at a 5 km resolution, other surface variables at 10 km, and atmospheric variables such as wind speed at 25 km. This provides a global picture nearly five times sharper than the previous version (WeatherNext 2).

Reliance on Real-World Global Data: By utilizing a mosaic of global satellite data, WeatherNext 3 acquires new forecasts every hour, moving away from exclusive reliance on traditional numerical weather prediction (NWP) models that suffer from bias.

Weather Station Data: The model is trained directly on weather station data, helping deliver predictions across Latin America, Africa, and the Asia-Pacific regions that were previously neglected due to the high cost of supercomputers.

Renewable Energy Development: The model forecasts wind speed at a height of 100 meters (for wind turbines) along with cloud cover levels and solar radiation to assist grid operators in clean energy planning.

Precipitation Forecast Accuracy: Through training on NASA IMERG data and radar networks, the model achieves a massive improvement in rainfall forecasting, showing up to a 60% enhancement in CRPS compared to IMERG.

Availability Across Platforms:
Data can be queried via BigQuery and Earth Engine, or downloaded directly from Google Cloud Storage.

Starting today, the model will also power weather experiences across Google Search, the Gemini app, Google Maps, the Google Maps Weather API platform, and Google Earth Engine, enabling users to see rainfall forecasts that are up to 50% more accurate when planning a day or more in advance.