Google has unveiled WeatherNext 3, the new version of its artificial intelligence model dedicated to weather forecast. The system aims to reduce the margin of error of traditional bulletins by integrating real-time satellite surveys, allowing data to be updated on an hourly scale and a spatial definition that is significantly higher than in the past.
The structure of the platform marks a clear difference compared to the previous generation. While WeatherNext 2 calculated at six-hour intervals based on a 25-kilometer geographic grid, the new iteration pushes the resolution up to 5 kilometers for key parameters such as temperature and humidity. A global image five times sharper which allows you to track rapidly evolving phenomena with greater accuracy, reducing the gap between real conditions and the estimates shown on users’ screens.
The enabling factor lies in the direct processing of live satellite observations, overcoming the typical waiting times of physical simulations processed on supercomputers. Leveraging fresher data streams translates into rain and snowfall estimates up to 50% more precise when calculated at least one day in advance. The performance gain is even more evident in regions without a dense network of ground detection stations, historically located outside North American and European territories.
Integration into services and support for renewable energy
The new predictive engine has already started feeding weather responses on Google Search, Google Maps, the Gemini application and several suites dedicated to developers and businesses. The Mountain View company has partnered with bodies such as the US National Hurricane Center and several Asian agencies to refine the monitoring of extreme weather events.
In parallel, the model includes specific carriers dedicated to the generation of renewable energypredicting for example the wind speed at 100 meters above sea level (an altitude equivalent to that of the hubs of wind turbines). This is a strategic function to optimize energy resources and balance consumption that has grown with the spread of data centers.
The synergy with classical physical modeling is confirmed. WeatherNext 3 continues to be trained on the parameters provided by official meteorological institutes, alongside and not replacing traditional atmospheric equations. The objective remains to offer a dynamic tool capable of reacting promptly to sudden changes in the local climate.
