By Seth Fiegerman, Bloomberg
Google’s DeepMind has developed an artificial intelligence model that uses decades of historical weather conditions to produce accurate 10-day forecasts in less than a minute.
GraphCast outperformed the gold-standard forecasting model from the European Centre for Medium-Range Weather Forecasts on roughly 90% of metrics tested, according to a study, which was published in the journal Science on Tuesday. The model also performed well when predicting the most severe weather, the study showed.
While computer models have long been central to forecasts, the changing climate has made weather events more unpredictable and potentially devastating. DeepMind said its system is able to “predict extreme weather events further into the future than was previously possible,” including tracking the paths of cyclones and anticipating severe temperatures. The AI model relies on decades of weather data and current weather conditions to identify patterns.
“GraphCast takes a significant step forward in AI for weather prediction, offering more accurate and efficient forecasts, and opening paths to support decision-making critical to the needs of our industries and societies,” Rémi Lam, a research scientist at DeepMind, wrote in a blog post Tuesday.
In the study, DeepMind’s team said its model shouldn’t be viewed as a replacement for conventional forecasting models but rather as proof that AI-based weather prediction systems have “potential to complement and improve the current best methods.”
More stories like this are available on bloomberg.com