Google has just iroduced a new and more advanced AI model for its weather forecasting capabilities that will be used in products such as Search, Gemini and Pixel phones.
According to The Verge, Google’s new artificial ielligence model is called WeatherNext 2, and it is able to produce weather forecasts 8 times faster than Google’s previous model, and it is also more accurate in predicting 99.9% of variables such as temperature and wind. The model can also provide hundreds of possible outcomes from a given starting poi.
Google’s new AI weather forecast is faster and more accurate than ever
According to Google, the WeatherNext 2 model, based on one of the company’s TPU chips, needs less than a minute to make a forecast. This is while traditional physics-based models usually require several hours of processing on supercomputers.
Traditional models require very high processing power; Because they try to recreate the complex physics of the atmosphere to make predictions. But artificial ielligence models try to predict future conditions using patterns extracted from past weather data.
Google has been able to make the forecasting process more efficie in WeatherNext 2 by using an approach it calls Functional Generative Network. This new approach iroduces noise, or purposeful randomization, io the model each time the input is received so that WeatherNext 2 can generate a large number of differe possible outputs and scenarios in one step.
These improvemes allow WeatherNext 2 to predict weather conditions 15 days ahead, as well as provide hourly forecasts. Google hopes these features will appeal to eerprise customers as well as regular consumers.
Akib Edin, Google Research Product Manager, said: “We found that the energy, agriculture, transportation, logistics and many other businesses are ierested in these hourly forecasts. This capability helps them make more accurate decisions related to issues that affect their business.”
