Google Inc. could greatly improve weather predictions by using machine learning, a type of artificial intelligence technology, a scientist from the company claimed Tuesday.

Accurately predicting weather can be particularly challenging for localized events that evolve on hourly timescales, such as thunderstorms.

Google, however, said it is currently conducting research into the development of a machine learning model that addresses this challenge by making predictions that are applied to the immediate future.

Carla Bromberg, a lead research scientist at Google's AI for Social Good, said the tech company's precipitation "nowcasting" model focuses on weather forecasts within six hours.

"A significant advantage of machine learning is that inference is computationally cheap, allowing forecasts that are nearly instantaneous," the scientist told reporters in Seoul by a video conference.

Unlike conventional measurements used by weather agencies around the globe, Google's program uses physics-free approach, meaning that the neural network will learn to approximate the atmospheric physics from the training examples alone, not by incorporating a prior knowledge of how the atmosphere actually works.

The scientist said the Google's results were compared with the High Resolution Rapid Refresh numerical forecast model by the US' weather agency, National Oceanic and Atmospheric Administration.

The scientist claimed that Google's model was about 10 times more accurate than HRRR, only taking between five and 10 minutes for the forecast.

Google said it currently has no plan to commercialize the research but is sharing the research product with the general public. (Yonhap)