Is Artificial Intelligence The Future of Avalanche Forecasting?

Ryan Kime | | Post Tag for AvalancheAvalanchePost Tag for Industry NewsIndustry News
avalanche debris field
Avalanche debris field; pc:bridgeportavalanchecenter

Considered by some observers as more of an art than a science, avalanche forecasting involves a unique combination of human observation, analysis, and interpretation. Perhaps even some good luck.  Even the best forecasters only reach about 75% accuracy in their predictions.  After all, they are human too and susceptible to bias and imperfection just like the rest of us. One of the biggest challenges in creating an accurate forecast lies in the fact that avalanche danger cannot be precisely measured and is therefore a matter of expert assessment (opinion).

Avalanche service warning
Sample avalanche report, pc:theadventuremedic

That was until last season when the Swiss Institute for Snow and Avalanche Research Group (SLF), a part of the National Swiss Federal Institute for Forest, Snow, and Landscape Research successfully tested a first-of-its-kind, artificial intelligence computer program to assist in its avalanche forecasts. A world leader in the field of avalanche forecasting since it took over the duties of the Swiss Army in 1945, SLF continues to push the development of avalanche forecasting.

SLF observers begin their day before 6 each morning, running from November through April when they collect measurements relayed to them from more than 120 different remote automatic weather stations located across the Swiss Alps. Observers then analyze the data, make their interpretations, and combine their opinions into a single avalanche service warning published at 5 pm each day.

SLF researchers used this new AI model as a “second opinion” against their human forecasts. Many times, the AI model coincided, sometimes, however, the AI model varied from the human prediction. This “discrepancy” between forecasts is then thoroughly reviewed looking for reasons for the variance.

avalanche warning report
Avalanche terrain, pc:theadventuremedic

SLF’s new AI model trained on their more than two decades-worth of weather and forecast information before starting last season. SLF researchers now estimate the AI program is about 75% accurate in its model predictions. SLF forecaster Frank Techel describes the AI program by saying,

The computer analyses the data in a different way than we do. That’s why it occasionally arrives at a slightly different conclusion. The computer forecast is very useful, especially for drawing exact boundaries between regions in which different danger levels prevail.”

This current AI model, however, has some limitations in that it was only suitable for dry-snow avalanches.  For next season, SLF is planning on introducing a new AI system capable of modeling wet-snow avalanches as well as snowpack stability.

Even with these amazing advances in AI-driven avalanche forecasting, SLF still believes in the value of human observation. SLF senior forecaster Gian Darms comments,

Humans not only observe but also interpret data. For example, an observer can see if snow cracks are appearing… we can anticipate things. The local dimension of assessing avalanche danger is a synthesis of information. This is something a machine simply cannot do.

And I think that is probably some great advice when it comes to your own assessment. Nothing should override your direct observation when it comes to signs of instability in the snowpack. 

For those interested in learning more about avalanche safety, or who want to access your local forecast(s) here’s a list of all the U.S. Avalanche Centers (credit: avalanche.org).  Stay safe and enjoy!

U.S Forest Service Avalanche Centers

Bridgeport Avalanche Center
Box 1000
Bridgeport, CA 93517

Bridger-Teton Avalanche Center
PO Box 424
Teton Village, WY 83025

Chugach NF Avalanche Center
P.O. Box 129
Girdwood, AK 99587

Flathead Avalanche Center
10 Hungry Horse Drive
Hungry Horse, MT 59919

Gallatin National Forest Avalanche Center
PO Box 130
Bozeman, MT 59771

Idaho Panhandle Avalanche Center
PO Box 405
Ponderay, ID 83852

National Avalanche Center
P.O. Box 130
Bozeman, MT 59771

Northwest Avalanche Center
7600 Sand Point Way NE
Seattle, WA 98115

Mt Shasta Avalanche Center
204 West Alma Street
Mount Shasta, CA 96067

Mt Washington Avalanche Center
300 Glen Road
Gorham, NH  03581

Payette Avalanche Center
PO Box 2177
McCall, ID 83638

Sawtooth Avalanche Center
206 Sun Valley Road
Sun Valley, ID  83353

Sierra Avalanche Center
10811 Stockrest Springs Rd
Truckee, CA 96161

Utah Avalanche Center
2242 West North Temple
Salt Lake City, UT 84116

West Central Montana Avalanche Center
PO Box 72
Missoula, MT 59806

State Avalanche Centers

Colorado Avalanche Information Center
325 Broadway WS1
Boulder, CO 80305

Local Nonprofit Centers

Alaska Avalanche Information Center

Central Oregon Avalanche Center
2701 NW Scandia Loop
Bend, Oregon 97701

Crested Butte Avalanche Center
PO Box 2351
Crested Butte, CO 81224

Eastern Sierra Avalanche Center
PO Box 1675
Mammoth Lakes, CA 93546

Hatcher Pass Avalanche Center
5629 Byrwood Circle
Palmer, AK 99645

Kachina Peaks Avalanche Center
PO Box 174
Flagstaff, AZ 86002-0174

Wallowa Avalanche Center
P.O. Box 324
Joseph, Oregon 97846


Related Articles

3 thoughts on “Is Artificial Intelligence The Future of Avalanche Forecasting?

Got an opinion? Let us know...