Google announces GraphCast: AI that forecasts the weather in seconds
GraphCast advances the use of AI for weather forecasting by providing more accurate forecasts.
Thanks to Google’s new artificial intelligence (AI) tool, GraphCast, it will be possible to have a faster and more accurate weather forecast up to ten days in advance.
The GraphCast model, based on machine learning and, according to those responsible, from the company DeepMind, «significantly outperforms» traditional systems and also serves to provide earlier warnings of extreme weather phenomena.
It predicts dozens of 10-day weather variables across the globe in less than a minute, they claim.
The details are published in the journal Science, where it is said that GraphCast – with open source – «takes an important step forward in AI for weather prediction, offering more accurate and efficient forecasts, and opening paths to support critical decision making,» say the authors, among whom are Spaniards Álvaro Sánchez González and Ferran Alet.
GraphCast is trained on decades of historical weather data to develop a model of the cause-effect relationships that regulate the evolution of weather on Earth from present to future.
This AI forecasts five variables over the entire Earth’s surface, including temperature, wind speed, wind direction, and mean sea level pressure, as well as six atmospheric qualities at each of 37 height levels, including humidity.
According to the authors, this is a significant advance in accurate and efficient weather forecasting and helps realize the promise of machine learning for modeling complex dynamic systems; however, this approach should not be considered a replacement for traditional methods; in fact, the European Centre for Weather Forecasting is already experimenting with this AI.s at each of 37 altitude levels, including humidity.
