We have experienced a broadcast interruption at our transmitter site that services a large part of Central Southern Colorado. This includes the Salida region as well as the entire San Luis Valley—including Lake George, Coaldale, Del Norte and San Luis. We are actively working to assess the point of failure. Thank you for your understanding and support! Our live stream is also available online and on the RMPBS+ app.

Stream live and on-demand content now on our new app:
RMPBS+

Thank you to those who supported Rocky Mountain PBS on Colorado Gives Day!

Predicting avalanches is notoriously difficult. Could AI help?

share
Denny Schaedig digs a snowpit to collect a sample for his research. Photo: Cormac McCrimmon, Rocky Mountain PBS
NEWS
SILVERTHORNE, Colo. — Colorado is the deadliest state for avalanches.

The state suffers from a cold, shallow snowpack and windy conditions, and a growing number of people are exploring the backcountry. 

From 1950 to 2023, snow slides killed 325 people in Colorado, nearly twice as many people as the second most deadly state, Alaska. 
Like ski patrol, Colorado Department of Transportation occasionally uses explosives to preemptively trigger avalanches that could affect roads. Photo: Cormac McCrimmon, Rocky Mountain PBS
Like ski patrol, Colorado Department of Transportation occasionally uses explosives to preemptively trigger avalanches that could affect roads. Photo: Cormac McCrimmon, Rocky Mountain PBS
Predicting avalanches is notoriously difficult. Variables forecasters use to make predictions can change dramatically over short distances. For example, snowfall can vary by elevation or from one side of a ridge to another. 

Denny Schaedig, a data engineer and backcountry skier from Silverthorne, hopes to use artificial intelligence and machine learning to improve avalanche forecasts. 

Schaedig is currently developing an app called AvAI, which he hopes could help backcountry users analyze snowpack. 

At ski resorts, patrollers work to mitigate avalanche hazards by closing terrain and using tools like explosives to preemptively trigger avalanches. Backcountry users must navigate avalanche hazards by themselves.
Denny Shaedig uses an avalanche rescue probe to measure the snow depth during a late-November outing. Photo: Cormac McCrimmon, Rocky Mountain PBS
Denny Shaedig uses an avalanche rescue probe to measure the snow depth during a late-November outing. Photo: Cormac McCrimmon, Rocky Mountain PBS
Last winter, Schaedig collected 2,300 snow samples he is now using to train an artificial intelligence model. 

Data collection starts with Schaedig hiking to a snow collection site. Schaedig digs to the base of the snowpack. Then, he uses an apple corer to take a sample of the layers. 

It’s a technique Schaedig devised himself. He compared it to the way scientists study ice cores to analyze ancient climate conditions. 

Schaedig records other variables, like snow depth, slope angle and number of avalanches seen. 

Using a magnified loupe, Schaedig takes a close-up picture of the snow crystals. He also records data like the slope angle, temperature and number of avalanches seen.
Denny Shaedig hikes to a site at Loveland Pass to gather a snow sample. Photo: Cormac McCrimmon, Rocky Mountain PBS
Denny Shaedig hikes to a site at Loveland Pass to gather a snow sample. Photo: Cormac McCrimmon, Rocky Mountain PBS
So far, he’s collected almost all of the samples near Silverthorne, but he hopes to expand the number of sites where he collects data this winter. 

Schaedig hopes to train other backcountry users to collect their own data, but so far, none have signed up. 
Using a magnified-loupe, Shaedig photographs the snow crystals. Photo: Cormac McCrimmon, Rocky Mountain PBS
Using a magnified-loupe, Shaedig photographs the snow crystals. Photo: Cormac McCrimmon, Rocky Mountain PBS
Denny Shaedig’s snow study kit includes an apple corer. Photo: Cormac McCrimmon, Rocky Mountain PBS
Denny Shaedig’s snow study kit includes an apple corer. Photo: Cormac McCrimmon, Rocky Mountain PBS
Ethan Greene, director of the Colorado Avalanche Information Center, said that he is excited about the ways artificial intelligence could help his team of forecasters, but Greene urged caution about how AI is applied. 

“You have to be careful about what sort of data you have available to train it and what you are trying to predict,” he said. 

Right now, Greene said people are only able to observe a fraction of the total avalanches that occur. 

That poses a statistical problem.
Last year, Shaedig collected more than 2,000 samples. He hopes to expand his public database this year. Photo: Cormac McCrimmon, Rocky Mountain PBS
Last year, Shaedig collected more than 2,000 samples. He hopes to expand his public database this year. Photo: Cormac McCrimmon, Rocky Mountain PBS
“Data collection is heavily dependent on where you can ride a snowmobile or ski, and visibility. As a result, we document five, six, maybe 7,000 avalanches a year in Colorado. But we really have no idea how many avalanches actually happen,” Greene said.

“Our guess is that we're probably documenting, maybe 10, 20% at the most. And so if you're trying to predict avalanche activity and you're using a data set that is 10% of the event as your training data, it's really hard to predict 100% of the events that are going to happen.” 
Type of story: News
Based on facts, either observed and verified firsthand by the reporter, or reported and verified from knowledgeable sources. To read more about why you can trust the journalism of Rocky Mountain PBS, please visit our editorial standards and practices page.

Subscribe to Our Newsletters

Get trusted Colorado stories, programs, and events from Rocky Mountain PBS straight to your inbox.

Set Your Preferences >