Google Enhances Weather Forecast Accuracy with New AI Model

Google has introduced an upgraded AI model, WeatherNext 3, which significantly improves the accuracy of weather forecasts. The technology leverages real-time satellite observation data, allowing forecasts to be updated hourly. The previous version, WeatherNext 2, generated forecasts every six hours with a 25 km resolution, while the new model achieves a resolution of up to 5 km.

WeatherNext 3 places special emphasis on predicting precipitation, including rain and snow. According to the company, the accuracy of such forecasts for the next 24 hours has improved by 50%. The model also fills data gaps for regions with insufficient ground-based weather stations, which is particularly relevant for countries outside the U.S. and Europe.

Additionally, WeatherNext 3 is optimized for renewable energy forecasting. For example, it can predict wind speeds at 100 meters— a key parameter for wind turbine operations. This is important for Google, which is actively expanding its AI infrastructure and faces growing energy demands for its data centers.

The new model is already integrated into the company’s services, including Search, Maps, and Gemini. Google is also collaborating with the U.S. National Hurricane Center and meteorological agencies in Asia to enhance forecasts using AI. Despite these advancements, AI models still work alongside traditional physical models to ensure maximum accuracy.