October 30, 2024

• 

Weather forecasts often become a laughing stock, not only because meteorologists are incapable of accurate predictions but also because it is difficult due to the unpredictable weather patterns amid climate change.

But deep learning technology might aid meteorologists, thanks to experts working to improve the accuracy of weather forecasts.

Global internet giant Google recently introduced how artificial intelligence, particularly machine learning, could improve weather forecasts.

Introducing “Nowcast” -- combination of now and forecast -- in Seoul on Feb. 4, a Google researcher said that machine learning allows “near instantaneous” prediction. 


Nearly instantaneous forecasts are possible as the neural network learns to approximate the atmospheric physics from training examples alone, not by incorporating prior knowledge of how the atmosphere actually changes.

Google scientist Carla Bromberg talks to reporters in Seoul via video conference on Feb. 4. (Google)

Within five to 10 minutes, the technology can predict weather for the next six hours. The same takes several hours when using existing techniques.

Carla Bromberg, who leads the Nowcast project, said that the results of Google’s research project are 10 times more accurate than the High Resolution Rapid Refresh numerical forecasts -- a model used by the US weather agency, National Oceanic and Atmospheric Administration.

Google said it has no immediate plans to use it commercially but will continue to conduct research on forecasts including flood prediction.

By Park Ga-young (gypark@heraldcorp.com)