Machine learning technology used in facial recognition could predict extreme weather events responsible for billions of dollars of damage every year, according to a new study published in the Monthly Weather Review.
Storms that produce hail can have large, damaging impacts on agriculture, property, and even wildlife. Just last week, as many as 13,000 shorebirds and waterfowl were killed in a severe hailstorm in eastern Montana. Such storm events cost as much as $22 billion every year in damages to people and property, according to CBS News. More than 4,600 major hailstorms occurred in 2018, according to NOAA’s Severe Storms database, with the majority of damages being reported in the central part of the country.
But the size and severity of hailstorms are often difficult to predict. That’s where artificial intelligence technology by the National Center for Atmospheric Research (NCAR) comes into play. Rather than zooming in on the features of a face, scientists have trained a deep learning model called “convolution neural network” to pinpoint specific storm features in order to determine whether hail will be formed and, if so, how large the hailstones will be.
"We know that the structure of a storm affects whether the storm can produce hail," said NCAR scientist David John Gagne in a statement. "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."
A perfect recipe of meteorological ingredients allows for a storm to produce hailstones, but even when conditions are ripe, the size and severity of hailstones will vary depending on the path and conditions within the storm, which is collectively known as the “storm structure”.
"The shape of the storm is really important," Gagne said. "In the past, we have tended to focus on single points in a storm or vertical profiles, but the horizontal structure is also really important."
NCAR scientists presented the machine learning software with images of simulated storms paired with data about temperature, pressure, and wind speed and direction, along with simulations of hail based on those factors. The program then figured out which features correlated with whether or not it hails and how big the hailstones are.
Generally speaking, the model confirmed storm features that the team had previously linked to hailstones. However, it’s important to note a number of limitations, including the fact that simulated storms vary dramatically from actual storms. Regardless, the team says their research could eventually transition into operational use to potentially replace the complex mathematical predictions currently used.