View post on imgur.com
Google recently launched an updated Weather experience for Pixel devices, introducing an intriguing addition named ‘Google Nowcast’. The feature provides real-time 12-hour precipitation forecasts. This enhancement signifies a significant step forward in providing users with precise and up-to-the-minute weather insights. Now, in a recent blogpost, Google shed light on the weather model that powers this new experience.
MetNet-3: Powering Nowcast on the Pixel Weather app
MetNet-3 is developed by Google Research and Google DeepMind. Building upon the foundations laid by its predecessors, MetNet and MetNet-2, this model offers users highly accurate and detailed predictions up to 24 hours in advance. But, the current implementation in the Pixel Weather app only provides information for a 12-hour period.
Leveraging a myriad of data sources, including radar, satellite, and numerical weather prediction assimilations, Google’s system continuously generates comprehensive forecasts at regular intervals, ensuring users remain well-informed about the latest weather conditions.
You can see MetNet-3 implemented in the Pixel Weather app on a card that shows up with the label as ‘starting soon.’ Here, you can scroll through the data horizontally to see the 12-hour forecast in detail. Check out the GIF below for reference:
Nowcast first started rolling out in June this year. Its reach now extends across the United States and 27 European countries. While initially available on the Pixel Tablet, Fold, and 8, it has now been extended to older Google phones as well. Moreover, there are plans to add support for it across all Android devices.
Technically, you can still take advantage of MetNet-3’s enhanced precipitation forecasts using Google Search. Searching for ‘Weather’ should bring up a suggestion like ‘Light to moderate rain from 9:30 AM continuing beyond 3:00 PM.’ It’s from here that you can see the 12-hour precipitation forecast.
Google’s MetNet-3 and the Nowcast feature represent a significant leap forward in the world of weather forecasting, promising users unparalleled accuracy and reliability in planning their day-to-day activities.
Google’s blogpost dives into further details on how the team trained the model. So if you’re interested in learning more about MetNet-3, check out the official post.