Thanks to the advancement of artificial intelligence, weather forecasts will become increasingly accurate.
For this reason, Google DeepMind and Google Research have presented WeatherNext 2, its most efficient and accurate weather prediction model to date.
This model represents a notable leap over its predecessor, as it can generate forecasts up to eight times faster and with a temporal resolution of one hour.
Its greatest innovation lies in that it is capable of producing hundreds of different scenarios from a single starting pointallowing meteorological agencies to make decisions based on a much wider range of possible climate developments.
Services that will benefit from this technology include weather forecasts in Google search, Gemini artificial intelligence and, of course, Google Maps.
The key to WeatherNext 2 is an AI approach called Functional Generative Network (FGN). This system introduces “noise” into the model architecture to generate coherent variability in predictions without losing physical consistency. Thanks to this, A single input can produce multiple realistic simulations in less than a minute, versus hours for traditional physics-based models..
Plus, WeatherNext 2 is not only faster, but also more reliable. Outperforms its predecessor in 99.9% of atmospheric variablesincluding temperature, wind and humidity, and in all prediction periods between 0 and 15 days.
The arrival of this model marks a turning point in the practical application of artificial intelligence to meteorology, bringing forecasts closer to what actually happens in the real world.