Predicting Avalanche Danger With Machine Learning in NW Montana

Abstract

This study evaluates the feasibility of using machine learning to predict daily avalanche danger ratings for mountain ranges covered by the Flathead Avalanche Center in northwest Montana. A data pipeline was developed to automate retrieval of meteorological inputs from the High-Resolution Rapid Refresh (HRRR) model, simulate snowpack conditions at multiple grid points using the SNOWPACK model, and generate structured features for supervised learning. A Random Forest classifier was trained to predict daily danger ratings (levels 1-4) for the Center’s elevation bands.

Initial results show that the model successfully captured several patterns present in historical avalanche forecasts and produced consistent predictions across the training and validation data. However, when applied to a new winter season, predictive performance declined, indicating reduced generalization to unseen seasonal conditions. This outcome suggests that while the modeling framework and automated data pipeline are capable of producing operational predictions, the model remains sensitive to differences between winter seasons. Limited historical training data and strong inter-annual variability in snowpack development likely contribute to this instability. Overall, the results demonstrate that machine learning approaches can support avalanche prediction, but additional historical data and further feature development are necessary before such models could be considered reliable for operational forecasting.