For the first time, AI predicted the weather more accurately than meteorologists


A meteorological artificial intelligence model from Google DeepMind called GraphCast for the first time predicted the weather more accurately than meteorologists. During the tests, it demonstrated superiority over the world's leading traditional system, which is managed by the European Center for Medium-term Weather Forecasts (ECMWF), surpassing it by 90% in 1380 indicators. The indicators include temperature, pressure, wind speed and direction, as well as humidity at various levels of the atmosphere. GraphCast is able to predict the weather 10 days in advance in less than a minute. According to the researchers, the new development is much cheaper in terms of energy consumption compared to traditional forecasting methods. For example, AI predicted Hurricane Lee in September of this year in the province of Nova Scotia in Canada in nine days – three days earlier than meteorologists warned. But so far GraphCast is considered only as an addition to traditional weather forecasting methods.