Scientists in Colorado and Wyoming are developing new ways to solve a tricky weather forecasting problem using artificial intelligence: predicting destructive hailstorms.
Hail is driving up the cost of insurance by damaging more roofs, cars, and crops, but it's notoriously hard to model.
In order for a storm to produce hail, it needs the right atmospheric ingredients, said Kristen Rasmussen, an associate professor at Colorado State University who's leading the project. They include a strong updraft, freezing temperatures, and liquid water droplets.
"It's a combination of many factors that are really hard for models to get right," said Rasmussen.
So, it's often unclear when and where hail will actually fall. Current weather systems only dispatch warnings about 30 minutes in advance.
But Rasmussen and her research team aim to transform those alerts by providing better forecasts hours or even days ahead of time.
The project, funded by the National Science Foundation, combines two AI models. The first tracks hailstorms as they're forming. It collates multiple radar sources and trains a model from AI company NVIDIA to focus on the patterns important to hail, extending forecasts from minutes to a few hours.
Rasmussen said a tool like this could help people get to safety if a big storm is approaching. For example, in 2023, dozens of people were injured when golf ball-sized hail struck concertgoers at Red Rocks Amphitheatre.
The second model looks at historic weather data and studies the key ingredients in past destructive storms to make predictions up to eight days out. The lead time could help the owner of a solar farm, for example, avoid costly damage.
"A couple-day warning can sometimes give better preparations and can allow for some of that equipment to either be moved, or sometimes you can tilt up the solar panels to withstand hail better," Rasmussen said.
Hail is the biggest cost driver of property insurance premiums in Colorado. Its destruction also contributes to the cost of insurance in Wyoming and New Mexico.
Rasmussen said these new tools are game-changing because, compared to physical radar data, they're analyzing a specific subset of data.
"With AI models, we can really hone in on specific phenomena like hail," she said.
The next step for the researchers is ground-truthing the models — testing them in the real world — to see how well they predict hail that actually falls. They'll install new high-tech sensors near a weather station in Greeley next year.
This story was produced by the Mountain West News Bureau, a collaboration between Boise State Public Radio, Wyoming Public Media, Nevada Public Radio, KUNR in Nevada, KUNC in Northern Colorado, KANW in New Mexico, Colorado Public Radio, and KJZZ in Arizona as well as NPR, with support from affiliate newsrooms across the region. Funding for the Mountain West News Bureau is provided in part by the Corporation for Public Broadcasting and Eric and Wendy Schmidt.
Copyright 2026 KUNC
