Google updates its weather forecasts with a new AI model

"The new AI model, WeatherNext 2, can generate forecasts eight times faster than Google's previous model, and is also more accurate in predicting 99.9 percent of variables like temperature or wind. WeatherNext 2 can pump out hundreds of potential outcomes from a particular starting point. It takes less than a minute using one of Google's TPU chips to make a prediction, which the company says would typically take several hours to accomplish using physics-based models on a supercomputer."

"Those conventional models are computationally intensive because they're essentially attempting to recreate the complicated physics of the atmosphere to produce forecasts. AI models, in contrast, try to discern patterns out of historical weather data in order to predict future outcomes. Google was able to streamline its process by using a strategy it calls a Functional Generative Network (FGN) in WeatherNext 2. Older AI weather models still required repeated processing to generate one forecast."

WeatherNext 2 generates forecasts eight times faster than Google's previous model and improves accuracy for 99.9 percent of variables such as temperature and wind. The model can produce hundreds of potential outcomes from a single starting point and produces a prediction in less than a minute using one of Google's TPU chips, compared with several hours for physics-based models on a supercomputer. The model uses a Functional Generative Network (FGN) to streamline processing and reduce repeated computation. AI forecasting models infer patterns from historical weather data rather than simulating atmospheric physics. Google is deploying the model across Search, Gemini, and Pixel phones.

Read at The Verge