Company forecasts, supplemented by artificial intelligence, proved to be very accurate still. AI weather models have also been able to make forecasts faster and more efficiently than traditional physics-based models. So far, Google's work in this area has been largely experimental. These predictions are now being used to argue for Google's products and services.
“We're taking it out of the lab and actually putting it in the hands of users.”
“We're taking it out of the lab and really putting it in the hands of users more than before and kind of moving away from the experimental designation because we believe that our predictions are actually quite powerful and quite useful,” Peter Battaglia, senior director of research and sustainability at Google DeepMind, said in a briefing with reporters.
new artificial intelligence model WeatherNext 2, can generate forecasts eight times faster than Google's previous model, and also more accurately predicts 99.9 percent of variables such as temperature or wind. WeatherNext 2 can identify hundreds of potential results from a given starting point. Using one of Google's TPU chips, it takes less than a minute to make a forecast, and the company says it typically takes several hours to complete the forecast using physics-based models on a supercomputer.
These traditional models are computationally intensive because they essentially try to recreate complex atmospheric physics to make predictions. AI models, on the other hand, try to identify patterns in historical weather data to predict future outcomes.
Google has been able to optimize its process using a strategy it calls Functional generative network (FGN) in WeatherNext 2. Older AI weather models still required reprocessing to produce a single forecast. FGN is more efficient because it incorporates noise—or target randomness—into the model every time it provides input, so WeatherNext 2 can generate many different possible outcomes in a single step.
These advances allow WeatherNext 2 to make forecasts up to 15 days out and generate hourly forecasts. Google expects this to be attractive to both corporate clients and individual consumers.
“We've found that energy, agriculture, transportation, logistics and customers in many other industries are very interested in these one-hour steps. It helps them make better decisions about things that impact their business,” said Aqib Uddin, product manager at Google Research.
In addition to adding WeatherNext 2 to Maps, Search, Gemini and Pixel Weather, Google is also offering early access program for clients interested in custom styling. Forecast data is also available on Google. Earth engine for geospatial analysis and Big request for large-scale data analysis.
