Google has unveiled WeatherNext 3, an AI model that uses real-time satellite data to predict high-resolution weather every hour.

TechCrunch reported on Wednesday that WeatherNext 3 is a next-generation global weather forecasting model developed by Google DeepMind and Google Research. In an assessment by Operational WeatherBench, an independent evaluation tool operated by Brightband, it outperformed major AI and traditional weather models in combined measures including temperature, wind speed and humidity.

The biggest changes are resolution and forecast frequency. WeatherNext 2 made predictions on a 25-km grid at six-hour intervals. The new model predicts key indicators such as temperature and humidity at resolutions of up to 5 km and generates a new forecast every hour. Some surface variables are provided at 10-km resolution and atmospheric variables at 25-km resolution.

It also strengthened precipitation forecasting. Medium-range probabilistic forecasting performance based on NASA's IMERG satellite precipitation data improved by as much as 60 percent from the previous model. Google also used its own satellite-radar-based reanalysis data and actual rain gauge observations for training.

WeatherNext 3 takes direct input from geostationary meteorological satellite imagery received every hour. It was designed to reflect local topography and changes in temperature and humidity in greater detail by training on actual observations at specific weather stations as targets. After Google claimed it was the first high-resolution global AI forecasting model to directly use raw observations, rival Windborne countered that its WeatherMesh has used raw observations such as balloons since last year.

Google started applying WeatherNext 3 from Thursday to Search, Gemini, Google Maps, the Maps Platform Weather API and Earth Engine. Researchers and companies can query the data in BigQuery and Earth Engine or download it from Google Cloud Storage. Google also plans to provide forecasts for wind speed at a height of 100 metres for wind power generation, as well as solar irradiance and cloud cover, for use in renewable energy operations.