Wet Avalanche Prediction Based on Weather Patterns in the Going-to-The-Sun-Road Corridor of Glacier National Park using Machine Learning

Overview

Wet snow avalanches are incredibly hazardous, so predicting them and their destructive force size is crucial to protect lives, and this study seeks to do so by using weather data. Machine learning techniques can detect patterns in the weather that influence avalanches but aren’t widely used. This study sought to better predict wet avalanches and their destructive force than previous approaches. The study area was the Going-to-the-Sun-Road corridor in Glacier National Park, a popular attraction prone to wet avalanches. Wet slab and glide snow avalanches were categorized together due to sharing similar causes while wet loose avalanches were predicted separately. Avalanches are affected by weather conditions over several days, so a sliding window input was used. The Long Short-Term Memory (LSTM) model was chosen for classification due to its ability to learn patterns over long periods of time. If a day was classified as an avalanche - day, it was sent to an XGBoost model that was trained to predict the maximum des tructive force of any avalanche that would occur that day. This study tested different parameters for the LSTMs. The presented wet loose LSTM model predicted 90% of avalanches and had a total accuracy of 79% while the WS+GS LSTM model predicted 98% of aval anches and had a total accuracy of 80%. The XGBoost model for the wet loose avalanches had predictions close to the actual results, although the WS+GS XGBoost wasn’t as accurate. Overall, these models perform better than previous attempts and could mitigate the harmful effects of avalanches.

Competition history

• JSHS 2025 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.