Google has introduced a new AI model for weather forecasting, WeatherNext 3

Google DeepMind has introduced WeatherNext 3, a new artificial intelligence model for weather forecasting. It utilizes real-time data from geostationary satellites, updates forecasts every hour, and operates at a resolution of up to 5 kilometers.

WeatherNext 3 predicts temperature, humidity, precipitation, wind speed, and other weather indicators. Compared to the previous version, WeatherNext 2, the forecast resolution has increased fivefold—from 25 to 5 kilometers—and data updates occur every hour instead of every six hours.

A key feature of the model is its ability to work with up-to-date satellite data. WeatherNext 3 receives information from geostationary satellites and generates a new forecast every hour. This allows for faster consideration of weather changes, including storms and precipitation.

Google also trained the model on data from weather stations. This enables WeatherNext 3 to account for regional weather characteristics. This is particularly important for Latin America, Africa, and the Asia-Pacific region, where creating detailed local forecasts requires significant computational resources.

WeatherNext 3 can also forecast indicators related to renewable energy. The model calculates wind speed at an altitude of approximately 100 meters, as well as cloud cover and solar radiation levels. These data can be used to estimate energy production from solar and wind power plants.

WeatherNext 3 also provides more accurate precipitation forecasts. In medium-range global forecasts, the improvement in the CRPS metric reached up to 60% compared to IMERG, up to 30% compared to MRMS, and up to 10% compared to rain gauge data in the early stages of forecasting.

WeatherNext 3 is already being used in Google Search, Gemini, Google Maps, Google Maps Platform Weather API, and Google Earth Engine. For forecasts lasting one day or longer, the accuracy of precipitation predictions has increased by up to 50%, especially in regions where such forecasts were previously less reliable.

Source: blog.google