Google is preparing to revolutionize weather forecasting, obviously with AI
How many times have you cursed the weather forecast for not being able to predict even a few hours into the future? The same probably happened in Mountain View, at Google’s headquarters, given that DeepMindthe company’s artificial intelligence division, has developed a machine-learning model designed to predict the weather, and the first results promise sparks!
A study published in Science a few days ago illustrates this fact GraphCastthe model in question, which not only proved to be very faster models based on numerical calculations (seconds versus hours, to give you an idea of the orders of magnitude), but also more accurateat least in ordinary situations.
DeepMind initially trained GraphCast with global weather data between 1979 and the 2017in order to teach him the effect of all the most common variables, such as wind, temperature, atmospheric pressure etc. At the moment GraphCast uses weather estimates up to 6 hours ago, combined with current conditions, to make its weather forecasts, which tend to be better than traditional ones, except perhaps in extreme climate conditions, but it is probably just a matter of further training so that GraphCast knows how to properly include other metrics.
We will then see in some time if Google’s weather forecast, which often makes mistakes every few hours, will really become so far-sighted. What is certain is that including i climate changes into the equation will not be easy, given their rapid evolution, but this would open the doors to a much broader discussion, which we do not want to delve into now.
