Item: THE OUTDOOR ADVENTURE RISK INDEX: FORECASTING BACKCOUNTRY EVACUATION LIKELIHOOD FROM WEATHER AND VISITOR EXPOSURE
-
-
Title: THE OUTDOOR ADVENTURE RISK INDEX: FORECASTING BACKCOUNTRY EVACUATION LIKELIHOOD FROM WEATHER AND VISITOR EXPOSURE
Proceedings: Proceedings, International Snow Science Workshop, Whistler, BC, Canada, 2026
Authors:
- Matthew Sheahan [ Thompson Rivers University, BC, Canada ]
- Peter Tsigaris [ Thompson Rivers University, BC, Canada ]
- Anne Terwiel [ Thompson Rivers University, BC, Canada ]
- Yaou Hu [ Thompson Rivers University, BC, Canada ]
Date: 2026-09-28
Abstract: Backcountry risk assessment requires practitioners and operators to interpret continually changing environmental conditions alongside uneven patterns of visitor exposure. Although professional judgment remains as a requirement, the likelihood component of risk assessment is often represented categorically rather than estimated from observed outcomes. The Outdoor Adventure Risk Index (OARI) was developed as a probability-based decision support model for estimating evacuation likelihood across monitored backcountry locations and hours. OARI was developed using 511,328 location-hour observations from thirty recreation monitoring locations in Mount Revelstoke and Glacier National Parks, British Columbia, representing 771,212 recorded visitor passes between December 2016 and May 2024. Each location-hour integrates visitor exposure, weather, temporal information, location, and evacuation-relevant incident occurrence. Incidents were spatially and temporally linked to a monitored location before model fitting. An operationally constrained genetic algorithm selected ten forecast-compatible weather variables, which were incorporated into a binomial generalized linear model with a complementary log-log link. Visitor exposure entered the model through a log-exposure offset. The final model achieved a precision-recall area under the curve of 0.010679, ROC-AUC of 0.787008, and Brier score of 0.002216 on held-out historical observations. More importantly for operational prioritization, the highest-ranked 1% of location-hours contained incidents at 10.0 times the rate expected under random ranking, while the highest-ranked 5% produced a lift of 5.5. OARI is not an avalanche-hazard model and does not predict that an evacuation will occur within a particular time or place. It is a dynamic probability signal that is intended to aid in the identification of location-hours that warrant additional operational attention. The results support the feasibility of integrating environmental forecasts and visitor exposure into probabilistic decision support for backcountry safety management.
Object ID: ISSW2026_P3.14.pdf
DOI: https://doi.org/10.15788/1790099100
Language of Article: English
Presenter(s): Matthew Sheahan
Keywords: backcountry risk, evacuation likelihood, rare-event prediction, visitor exposure, decision support, search and rescue
Page Number(s): 660 - 663
-