AI model Google more accurate weather forecasterBack
Google's development of AI applications continues. Google has now developed GraphCast, an AI model that predicts the weather faster and better than traditional algorithms. And it requires much less hardware.
The AI model developed by Google Deepmind uses 'graphs', pieces of information that indicate how they relate to each other. This allows them to store and process a lot of information, and makes them very suitable for describing, among other things, the weather and the atmosphere.
The GraphCast model is trained on weather observations collected over 40 years by the European Center for Medium-Range Weather Forecasts (ECMWF). In addition, satellite information and information from radar systems and other sources were used. The AI model can predict the weather up to ten days in advance. This concerns temperature, humidity, wind speed and several variables at different altitudes of the atmosphere.
An internal test showed that the AI model performed better than other commonly used algorithms. Especially if the model made predictions for different layers of air, such as the troposphere, the layer of the atmosphere directly above the Earth's surface. In addition, Google GraphCast would be able to better predict extreme weather events, such as the appearance of a cyclone.
Another advantage is that this AI model requires significantly less hardware. Current weather forecast models require a lot of hardware, because the calculations are complex and often take a very long time. Google says its new model requires a single Google Processing Unit for AI for a 10-day forecast. This saves a lot on hardware and data center capacity, and the calculations are also performed within a minute. The ECMWF site shows a live demonstration of the weather model predictions.