Google AI better at predicting weather than conventional models

Posted on Tuesday, Jan 14 2020 @ 10:32 CET by Thomas De Maesschalck
Google AI researchers say they've developed a new convolutional neural network (CNN) that is better at nowcasting weather than traditional weather models. In a paper called "Machine Learning for Precipitation Nowcasting from Radar Images," the Google researchers explain the model can generate 0 to 6 hour forecasts that have a 1km resolution with a total latency of just 5-10 minutes (whereas traditional models have a 1-3 hours latency). Google's AI model outperformed three acclaimed weather models: High Resolution Rapid Refresh (HRRR) numerical forecast, an optical flow (OF) algorithm, and the persistence model.
Unlike traditional methods, which incorporate a priori knowledge of how the atmosphere works, the researchers used what they are calling a 'physics-free' approach that interprets the problem of weather prediction as solely an image-to-image translation problem. As such, the trained CNN? by the team?—a U-Net?—only approximates atmospheric physics from the training examples provided to it.

For training the U-Net, multispectral satellite images were used. Data collected over the continental US from the year 2017 to 2019 was used for the initial training. Specifically, the data was split into chunks of four weeks where the last week was used as the evaluation dataset while the rest of the weeks were used for the training dataset.

Via: Neowin

About the Author

Thomas De Maesschalck

Thomas has been messing with computer since early childhood and firmly believes the Internet is the best thing since sliced bread. Enjoys playing with new tech, is fascinated by science, and passionate about financial markets. When not behind a computer, he can be found with running shoes on or lifting heavy weights in the weight room.

Loading Comments