Google with WeatherNext 3 in weather forecast It incorporates live satellite data directly into the calculation. New Google artificial intelligence model produces forecasts every hour and some surface data 5 kilometer resolution reaching. By contrast, WeatherNext 2 produced forecasts at six-hour intervals with a resolution of 25 kilometers. Additionally, Google is bringing WeatherNext 3 to Search, Gemini, Maps and Google Cloud services.
WeatherNext 3 processes hourly geostationary satellite mosaics with historical atmospheric analysis in a Functional Generative Network-based mesh transformer model. Thus, in addition to temperature, humidity and wind forecasts, the model produces results specific to tropical cyclone routes and meteorological stations. Additionally, the system changes in temperature and humidity near the surface. 5 kilometersprovides 10 kilometer resolution in other surface data. In upper atmosphere variables, the resolution increases to 25 kilometers.
Google states that most of its previous AI weather models used analysis data from numerical weather prediction systems. However, since this data carries a delay of approximately six hours, it may lose its current status in rapidly changing precipitation and temperature values. WeatherNext 3 adds current satellite observations directly to the calculation every hour. In addition, the model is also trained with direct measurements from weather stations to account for local geographic details.
Google, NASA in rainfall forecasts IMERG It uses satellite measurements and global precipitation analyzes prepared by the company with satellite-radar data. Accordingly, Google is against IMERG in the CRPS criterion. up to 60 percent reported improvement. Additionally, the improvement reaches 30 percent in MRMS radar data in the United States and 10 percent in early rain gauge comparisons. Brightband, on the other hand, does not yet include WeatherNext 3 precipitation results in its Operational WeatherBench rankings due to different reference data.
WeatherNext 3 produces new forecasts every hour with current satellite data
Brightband evaluates WeatherNext 3’s surface forecasts at a resolution of 0.1 degrees and upper atmosphere forecasts at a resolution of 0.25 degrees. On the other hand, Google uses high-resolution data of the model. BigQuery, Earth Engine and offers it to researchers via Google Cloud Storage. The company also provides forecasts of wind speed at 100 meters altitude for wind power and cloudiness and solar irradiance for solar power. Agriculture, energy production, flight planning and emergency teams can also benefit from this data.
Google previously published separate studies on the route, intensity and wind characteristics of tropical cyclones within the scope of its WeatherNext studies. However, WeatherNext 3 is a direct climate model and Google focuses the system on medium-term weather forecasts. The company uses separate models such as NeuralGCM for longer-term weather and climate simulations. WeatherNext 3 forecasts reach users within Search, Gemini, Maps and Google Maps Platform.
Join Channel