Facial recognition AI trained to produce better storm forecasting

19 Aug 2019

Image: © Minerva Studio/Stock.adobe.com

It could soon be easier to predict the severity of hailstorms using existing facial recognition technology.

Despite the somewhat controversial nature of facial recognition technology and its ability to read emotions, scientists have found a way to use the same AI tech to scan for something very different: hailstorms.

According to a new study from the US National Center for Atmospheric Research (NCAR) published to Monthly Weather Review, scientists have trained a deep learning model called a convolutional neural network to recognise features of individual storms that lead to the formation of hail and how large the hailstones will be.

In the past, this has proven notoriously difficult to predict, but promising results obtained in this study highlighted the importance of being able to take into account a storm’s entire structure.

“We know that the structure of a storm affects whether the storm can produce hail,” said NCAR scientist David John Gagne, who led the research. “A supercell is more likely to produce hail than a squall line, for example. But most hail forecasting methods just look at a small slice of the storm and can’t distinguish the broader form and structure.”

In the example of hail, there are a myriad of different factors that eventually contribute to a storm producing pellets of ice. For example, the air needs to be humid close to the surface, but dry higher up. Likewise, the freezing level within the clouds needs to be close to the ground and strong updrafts that keep the hail aloft to grow large are also essential.

Hail to the AI

Even if all of the necessary conditions for hail are met, the size of the hailstones can vary dramatically. But using this variant of facial recognition technology, the important understanding of a storm’s structure can be revealed.

Existing computer models for storms are limited by the mathematical complexity of representing the properties of an entire storm and the physics of its formation. Instead, the machine learning neural network is able to ingest large amounts of data, search for patterns and teach itself which storm features are crucial to accurately predict hail.

To train the system, Gagne used images of simulated storms and combined it with the data to identify if a storm is forming. The AI was then able to figure out which features of the storm correlated with the formation of hail and what size it would be.

Tracing back through its decision-making, the researchers could see that among the telltale signs were storms with winds blowing from the south-east near the surface and from the west at the top. Additionally, storms with a more circular shape are also most likely to produce hail.

“I think this new method has a lot of promise to help forecasters better predict a weather phenomenon capable of causing severe damage,” Gagne said. “We are excited to continue testing and refining the model with observations of real storms.”

Colm Gorey was a senior journalist with Silicon Republic

editorial@siliconrepublic.com