Google DeepMind and Google Research have introduced WeatherNext 3, a new AI weather forecasting model designed to deliver more frequent and localized predictions as weather conditions change.
The model generates forecasts every hour at up to 5-kilometer spatial resolution, while using live satellite imagery and ground-station observations as inputs. Google says this represents roughly five times higher spatial resolution than its predecessor.
WeatherNext 3 is now powering weather experiences across Google Search, Gemini, Google Maps, Google Maps Platform Weather API, and Google Earth Engine.
WeatherNext 3 updates forecasts every hour
Traditional weather forecasting relies heavily on numerical weather prediction, which uses complex physical models and substantial computing resources to simulate how the atmosphere will evolve.
WeatherNext 3 instead uses AI to produce updated predictions hourly, allowing forecasts to respond more frequently as storms, fronts, and other weather systems develop.
For users, that could translate into more timely forecasts when conditions change quickly, particularly when planning activities around precipitation or severe weather.
Higher resolution targets more localized forecasts
WeatherNext 3 can resolve surface temperature and moisture at 5 km (0.05°) and surface winds at 10 km (0.1°), according to Google.
The higher resolution is designed to provide more localized information, including in areas where conventional weather forecasting has historically been less reliable.
Google also says WeatherNext 3 improves longer-range predictions. For forecasts made a day or more in advance, users can see precipitation forecasts that are up to 50% more accurate on key benchmarks, with the largest improvements reported in regions that previously had lower forecast reliability.
AI improves precipitation forecasting
Precipitation is one of the most difficult aspects of weather forecasting because rainfall can vary significantly over relatively short distances.
Google says WeatherNext 3 cuts error metrics by up to 50% on key precipitation benchmarks compared with traditional numerical weather prediction baselines.
That improvement could make the model particularly useful for situations where knowing not just whether rain is expected, but where and when it is likely to occur, matters.
Weather data also supports clean energy
WeatherNext 3 is not limited to consumer forecasts.
The model produces specialized weather parameters that can support renewable-energy operations, including wind speeds measured at a 100-meter turbine height, multi-layer cloud coverage, and solar irradiance.
These outputs can help clean-energy producers and grid operators better account for changing weather conditions when managing generation and electricity supply.
WeatherNext 3 is already rolling out across Google services
Unlike a research model that remains confined to testing, WeatherNext 3 is already being used across several Google products, including Google Search weather experiences, Gemini, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine.
The combination of hourly updates, higher spatial resolution, and improved precipitation forecasting marks another step toward using AI to make weather information more localized and responsive.
For everyday users, the most noticeable change may simply be more useful forecasts when checking conditions for the next day. Behind those forecasts, however, WeatherNext 3 represents a broader push to use AI for weather prediction at a scale that can also support infrastructure and renewable-energy planning.
