New Google Weather adds ‘Nowcast,’ bests Dark Sky with 12-hour ML predictions

The launch of the Pixel Tablet and Fold also brings a Material You redesign of Google Weather. This revamp, which will later come to other devices, features “Google Nowcast” for more accurate rain, hail, and snow predictions.

Google confirmed to us today that it will bring the new Weather experience to more devices in the future but has no details about timing just yet. At the moment, it’s taking in user feedback from the revamp’s initial launch on the Pixel Tablet and Fold to refine the design. On those two devices, it’s available in 26 languages.

• Pixel Tablet review: Uniquely Google, made possible by Android

Google Nowcast

Besides the new design for Weather, one major addition is “Google Nowcast” for more precise predictions about rain, hail, and snow happening in the next 12 hours.

These short-term forecasts are based on Multi-Radar/Multi-Sensor System (MRMS) and High-Resolution Rapid Refresh (HRRR) data from the National Oceanic and Atmospheric Administration, as well as other sources. This Nowcast data is refreshed multiple times per hour.

When there’s an active or upcoming continuous precipitation event, Google will show a new section above the hourly forecast carousel. In the example below, the “Rain starting soon” card notes, “Light to heavy rain from 8:32 AM to 12:36 PM.” The card also includes a timeline showing whether it will be light, medium, or heavy.

Google won’t surface Nowcast if there’s no rain, hail, or snow in the forecast, or if there’s uncertainty. For longer-term forecasts, Google is still using data from weather.com.

Google Nowcast is available when searching for weather in the continental US (English only) from a mobile device, including the new Weather experience on the Pixel Tablet and Fold, as well as the web.

This is somewhat reminiscent of Dark Sky and now Apple Weather providing “minute-by-minute precipitation intensity over the next hour.” However, Google’s up to 12-hour forecasts could offer a marked improvement.

In 2020, the company detailed its ML research on “nowcasting” to better forecast localized rain storms and other short-term weather events that are not well-served by previous methods.

Google’s approach uses radar data and treats weather prediction as a computer vision problem. Neural network learn about “atmospheric physics from the training examples alone, not by incorporating a priori knowledge of how the atmosphere actually works.”

One of the advantages of the ML method is that predictions are effectively instantaneous, meaning that our forecasts are based on fresh data, while HRRR is hindered by computational latency of 1-3 hours. This leads to better forecasts for computer vision methods for very short term forecasting.