Google Launches WeatherNext 3 AI Weather Model

Google has introduced WeatherNext 3, its latest artificial intelligence model designed to predict weather patterns with significantly higher precision. By integrating real-time satellite observations, the new model generates global forecasts that are five times sharper than those of its predecessor, WeatherNext 2. While the previous version provided forecasts every six hours on a 25-kilometer grid, WeatherNext 3 delivers hourly updates and can visualize key atmospheric variables, such as temperature and moisture, at a resolution of up to five kilometers.

This boost in spatial and temporal resolution is particularly effective for tracking fast-moving rain and snow systems. Google reports that the model can deliver precipitation forecasts that are up to 50 percent more accurate when predicting at least a day in advance. Furthermore, by utilizing satellite imagery, WeatherNext 3 can fill critical data gaps in regions outside of the United States and Europe that lack dense networks of ground-based rain gauges.

For practitioners in the energy sector, the model offers specialized predictions, such as wind speeds at a height of 100 meters, which corresponds to the hub height of typical wind turbines. This capability is intended to help optimize renewable energy generation as global power demands rise. Google has already integrated WeatherNext 3 into its consumer-facing products, including Google Search, Google Maps, and the Gemini assistant. Additionally, the company is collaborating with organizations like the US National Hurricane Center and meteorological agencies in Asia to refine forecasting workflows.

Despite these advancements, AI weather models are not replacing traditional physics-based simulations. WeatherNext 3 is trained on data generated by physics-based models, and meteorologists will continue to use both systems in tandem to issue public safety warnings. However, the speed of WeatherNext 3 allows it to recognize complex atmospheric patterns much faster than traditional supercomputers, which must solve resource-intensive physical equations.