Google claims to have developed very reliable AI for the weather

The weather of the future. An artificial intelligence program developed by Google is now capable of predicting the weather over a period of two weeks with unequaled precision, the American giant said on Wednesday. The GenCast model invented by DeepMind, Google’s London-based AI research lab, “showed better forecasting capabilities” than the current reference model, according to the company, whose researchers published a study in the review Nature.

Trained with data spanning 1979 to 2018 and covering temperature, wind speed and atmospheric pressure, this model can produce a fifteen-day forecast in just eight minutes, compared to several hours currently. The European Center for Medium-Range Weather Forecasts currently produces forecasts for 35 countries and is considered the global benchmark in this field. But now, Google researchers note in their publication, GenCast exceeds the accuracy of the center’s forecasts in more than 97% of the 1,320 climate disasters listed in 2019 on which the two models were tested.

According to Google, GenCast “consistently outperformed” the current benchmark model in predicting extreme heat, extreme cold and high wind speeds. “More accurate forecasts of extreme weather risks can help authorities protect more lives, avoid significant property damage, and save money,” Google said. Ironic when we know that artificial intelligence also contributes to increasing the firm’s carbon footprint.

Support existing weather models

This progress is “a first step” to integrate AI into weather forecasts, responded Florence Rabier, the director general of this European reference center, the ECMWF. For now, says this specialist, this technology can be used to assist existing models. “Any method that can improve and speed up this process is absolutely welcome, at a time when climate change is exerting extreme societal pressure,” she added.

“I am convinced that weather forecasting systems that use AI will continue to gradually improve, in particular to make better forecasts of extreme climatic phenomena and their intensity, in which there is great room for improvement said David Schultz, professor of meteorology at the University of Manchester, who was not involved in the research. But these forecasting systems depend on existing models, such as the one operated by ECMWF, added the researcher. Next step: a robot weather presenter on television.