"Leap forward"

Google AI predicts weather more accurately than ever before

The AI weather forecast was trained using data on temperature, wind speed and air pressure from 1978 to 2018.

(Bild: x.com/GoogleDeepMind)

A Google research lab says it has developed an AI-based application that can produce weather forecasts for two weeks with unprecedented accuracy and speed. GenCast shows "better forecasting capabilities" than the world's current leading model, the internet company announced on Wednesday.

The benchmark for meteorological accuracy is data from the European Center for Medium-Range Weather Forecasts (ECMWF), which produces forecasts for 35 countries. According to the London-based AI research laboratory DeepMind, GenCast's predictions exceeded the center's forecasts in 97 percent of cases, using 1320 real-life scenarios from 2019 as a test basis. The results were also published in the scientific journal "Nature".

ECMWF director Florence Rabier said the project was a first step towards incorporating AI in weather forecasting and that it was "indeed a leap forward". The AI weather forecast was trained using data on temperature, wind speed and air pressure from 1978 to 2018. It can now produce a two-week forecast in eight minutes, which currently takes several hours.

"Protect lives, avert damage, save money"
GenCast offers not only better daily forecasts, but also better forecasts for extreme weather events, according to a statement from DeepMind. In view of the increase in extreme weather events as a result of climate change, the developers point to the associated opportunities to save lives. More accurate predictions of the risks of extreme weather events could help the authorities to "protect more lives, avert damage and save money".