WeatherNext 2: Google DeepMind introduces new AI weather model
DeepMind has introduced a new AI weather model that creates fast and accurate forecasts. It requires very little computing power.
Weather models that work with artificial intelligence (AI) have recently delivered very accurate forecasts in this year's North Atlantic hurricane season. Google's AI division, DeepMind, has introduced a new AI weather model that, according to the developers, is very fast and accurate.
WeatherNext 2 can create weather forecasts with a resolution of one hour and is eight times faster than the previous version, $(LEhttps://blog.google/technology/google-deepmind/weathernext-2/:the development team announced|_blank).
The system has already been used for hurricane forecasts for weather services. Now, weather forecasts for Search, Gemini, and Pixel Weather will also be created with it. Integration into Google Maps is planned for the coming weeks.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
The model provides forecasts for two weeks regarding temperature, air pressure, and wind with an accuracy of one hour. In addition, the system calculates the trajectory of a tropical cyclone three days in advance, according to the developers. Previous systems could only do this for two days.
Many industries need accurate forecasts
The high resolution of wind, clouds, and temperature, as well as the paths of storms, are particularly important for the economy. A whole range of industries depend on it – agriculture, for example, the energy sector, or shipping.
AI-based weather models “take historical records and try to identify large-scale patterns from them to predict what will happen next. Because they don't worry about the small details, they can work much faster,” said Peter Battaglia, a senior scientist at DeepMind.
Videos by heise
Based on the current weather situation, WeatherNext 2 creates hundreds of possible scenarios. The system uses few resources: each scenario is calculated on a tensor processing unit (TPU) in just one minute, say the developers.
This performance increase is made possible by a new approach in AI modeling, the Functional Generative Network (FGN), which DeepMind $(LEhttps://arxiv.org/abs/2506.10772:introduced a few months ago|_blank).
The use of AI has revolutionized weather forecasting, as it requires very little computing power. Without AI, weather models have to be created on supercomputers.
(wpl)
