Building the quantitative foundation of decision-support tools for backcountry snow avalanche risk assessment using avalanche terrain modeling, GPS tracking, and machine learning

Snow avalanches are a complex natural hazard that cause approximately 140 reported fatalities each year. Most accidents involve backcountry recreationists, and there are often signs of unstable snow or dangerous terrain that are overlooked by the groups involved. Development of decision-making aides can reduce avalanche fatalities by contributing to individuals making more informed risk management decisions. Current methods lack the ability to combine high quality avalanche terrain information with current conditions to assist backcountry users in making informed risk assessments. This thesis presents several geospatial models that can assist winter backcountry recreationists in better understanding their exposure to avalanche release areas and runout zones and develops an automated decision-making tool for professional guides to determine what terrain is appropriate for the day based on records of their past decisions. The research integrates human expert training data into model development by developing novel validation approaches in collaboration with local avalanche experts. First, I adapt existing avalanche terrain mapping methods to include forest density in detecting avalanche start zones using widely available satellite remote sensing data and develop a validation procedure to assess model performance using input from local expert guides. Second, I combine automated avalanche terrain modeling with GPS tracking and operational data from a Canadian mechanized skiing operation to develop several decision-support tools and compare their performance against real-world decisions of professional guides. These tools illustrate the decision-making process and reveal decision-making patterns of professional guides by combining avalanche terrain information with current conditions using machine learning models. Third, I contribute to the development of an open-source automated avalanche terrain exposure scale (ATES) model by updating the PRA model to account for forested terrain, applying an improved avalanche runout simulation tool, and refining the ATES classification process. Fourth, I validate the automated ATES model in western Canada by performing a grid search to reverse engineer the optimal input parameters against two ATES benchmark maps developed in collaboration with local avalanche experts. Overall, the research contributes to further develop open-source automated avalanche terrain models and provides an example of how knowledge from local experts can be integrated into decision-making model development.