Hail Forecasts May Be Improved Using Artificial Intelligence

Linda Lam
Published: August 21, 2019

Forecasting hail may become more accurate thanks to artificial intelligence.

A new study from the National Center for Atmospheric Research (NCAR), published in the American Meteorological Society's Monthly Weather Review, compared forecasting hail using so-called deep learning models to standard machine learning.

Specifically, researchers used the same technique used in facial recognition to train the model to search for and recognize patterns in the 3D structure of a thunderstorm that are related to hail formation.

Betty Proue holds some of the large hail that damaged her house and outbuildings on Sunday evening as residents clean up after high winds and large hail pounded Yellowstone County in Huntley, Mont., Monday, Aug. 12, 2019.
(Larry Mayer/The Billings Gazette via AP)

The deep learning technique was found to be more skillful than a previously tested form of machine learning in predicting whether hail would form and how large the resulting hailstones might be.

The techniques focused on predicting which storms would produce severe hail (hailstones at least one inch in diameter) up to an hour after the storm was detected.

Currently, because the complex nature of storms, only a small part of a given storm's most important features for predicting hail are analyzed. Artificial intelligence, however, is able to teach itself what features are key to predicting hail and can quickly analyze those features.

"Using these deep learning tools in unique ways will provide additional insight into the conditions that favor large hail, improving model predictions," said Nick Anderson, a program officer in National Science Foundation's Division of Atmospheric and Geospace Sciences, which funded the research.

"This is a creative and very useful convergence of scientific disciplines."

The size of hailstones can vary depending on the path the hail takes through a storm and the weather conditions along its path.

“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 team. “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.”

Having the ability to more accurately predict if a given storm will produce hail and how large that hail will be can be very beneficial, since large hail often results in property and crop damage, as well as injury.

(MORE: Hail: The Most Underrated Costly Weather Disaster

One of the cars damaged by a hailstorm on Monday, Aug. 6, 2018, at the Cheyenne Mountain Zoo in Colorado Springs, Colorado. At least eight people were injured by the hail.
(Colorado Springs Fire Department public information officer)

Analysis showed that the deep learning system discovered, on its own, the same factors that other research has confirmed make a storm more likely to produce severe hail.

These characteristics include storms that have a more circular shape, favorable wind shear (winds blowing from the southeast at the surface and from the west at the top) and a vertical pressure pattern that favors strong updrafts.

“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.”

Additional testing using storm observations and radar-estimated hail is needed before the technique can be used in daily forecasting.

How Hail Forms

Hail is a form of frozen precipitation that is created by thunderstorms with strong updrafts or fast air that is pulled upward.

The center of the hailstone is an ice particle. At first, the ice particle remains within the thunderstorm, often at temperatures well below freezing, due to the force of the updrafts.

However, it may grow as it collides with droplets of supercooled water (which remain unfrozen because they lack a surface on which to freeze). A hailstone is born as these supercooled water droplets freeze onto the ice particle.

Air bubbles can get trapped in the developing hailstone, leading to a more spongy texture.

The hailstone may continue to collide and coalesce with other water droplets and ice crystals until it is thrown outside the main updrafts or it becomes too large and heavy to be influenced by the updrafts and falls to the ground.

The stronger and more sustained the updrafts, the larger the hailstone can become.

Sometimes a hailstone will make several circuits into and out of a strong updraft, growing larger each time, before finally falling to the ground.

Hail size and updraft needed.

According to the National Weather Service, in order to produce dime-sized hail, a thunderstorm would need to have an updraft speed of at least 38 mph.

Wind shear (changes in wind direction and speed with height) helps to separate a thunderstorm's updrafts and downdrafts, which allows for stronger updrafts and a better chance of hail. Highly unstable conditions (relatively warm, moist air near the surface with relatively cooler, drier air aloft) also lead to stronger updrafts and larger hail.

Hail is also more likely in storms where the freezing level is relatively low, closer to the bottom of the cloud. This is especially common across the higher elevations of the High Plains, where cloud bases are higher than in the eastern U.S.

The Weather Company’s primary journalistic mission is to report on breaking weather news, the environment and the importance of science to our lives. This story does not necessarily represent the position of our parent company, IBM.