Google is improving the weather forecasts you see in the Pixel Weather app, Google Search, and in other locations. The Alphabet subsidiary has introduced WeatherNext 2, a weather forecasting model powered by-what else-AI. Google says that this forecasting model delivers faster and more accurate forecasts. How much faster? Google says that this “most advanced and efficient forecasting model” will generate weather forecasts eight times faster and generates four six-hour weather forecasts daily, with future predictions going out as long as 15 days.
WeatherNext 2 outperforms WeatherNext 1 when forecasting 99% of variables such as wind speed and direction, temperature, precipitation, pressure, and humidity. Using a Tensor Processing Unit (TPU), which is a specialty Application-Specific Integrated Circuit (ASIC) chip designed by Google to accelerate machine learning and AI workloads, a weather forecast can be made in less than a minute. Compare that to the hour the same forecast would take if it were made on a Supercomputer using models based on physics.
In the next few weeks, WeatherNext 2 will also take over the weather forecasting job for Google Maps. So, if you notice an improvement in the quality of the forecasts you see from the Pixel Weather app, Google Maps, the Pixel Weather app and Google Search, it could be due to the improvements made by Google in the WeatherNext 2 forecasting model.
WeatherNext 2 understands better the many different variables that make up a weather forecast. These variables, such as temperature, wind, and pressure are related and these relationships can be captured by equations. AI weather forecasting doesn’t rely on crunching data using Supercomputers. Instead, with AI, users are looking for patterns that would be expected to repeat.
With WeatherNext 2, in just one minute you could have 15 days of forecasts generated in just one minute. Overall, hundreds and thousands of forecasts can be quickly created.
Read the latest from Alan Friedman
