Examining the operational use of avalanche problems with decision trees and model-generated weather and snowpack variables

Abstract

Abstract. Avalanche problems are used in avalanche forecasting to describe snowpack, weather, and terrain factors that require distinct risk management techniques. Although they have become an effective tool for assessing and communicating avalanche hazard, their definitions leave room for interpretation and inconsistencies. This study uses conditional inference trees to explore the application of avalanche problems over eight winters in Glacier National Park, Canada. The influences of weather and snowpack variables on each avalanche problem type were explored by analysing a continuous set of weather and snowpack variables produced with a numerical weather prediction model and a physical snow cover model. The decision trees suggest forecasters' assessments are based on not only a physical analysis of weather and snowpack conditions but also contextual information about the time of season, the location, and interactions with other avalanche problems. The decision trees showed clearer patterns when new avalanche problems were added to hazard assessments compared to when problems were removed. Despite discrepancies between modelled variables and field observations, the model-generated variables produced intuitive explanations for conditions influencing most avalanche problem types. For example, snowfall in the past 72 h was the most significant variable for storm slab avalanche problems, skier penetration depth was the most significant variable for dry loose avalanche problems, and slab density was the most significant variable for persistent-slab avalanche problems. The explanations for wind slab and cornice avalanche problems were less intuitive, suggesting potential inconsistencies in their application as well as shortcomings of the model-generated data. The decision trees illustrate how forecasters apply avalanche problems and can inform discussions about improved operational practices and the development of data-driven decision aids.

A quantitative module of avalanche hazard—comparing forecaster assessments of storm and persistent slab avalanche problems with information derived from distributed snowpack simulations

“…Numerous snowpack modeling studies have been carried out at Glacier National Park (e.g. Bellaire and Jamieson, 2013;Horton et al, 2020c), which is known for high-quality avalanche hazard assessments and observations. For our simulations, we feed the Canadian numerical weather prediction model HRDPS (Milbrandt et al, 2016, 2.5 km resolution) into the detailed snow cover model SNOWPACK (Bartelt et al, 2002;Lehning et al, 2002b, a) to simulate the snow stratigraphy at 100 grid point locations within the boundaries of Glacier National Park.…”

“…Operational experience and recent research has shown that there are considerable differences in how the avalanche danger rating, the CMAH, and the concept of avalanche problems are applied by avalanche forecasters (Lazar et al, 2016;Statham et al, 2018b;Techel et al, 2018;Clark, 2019;Horton et al, 2020c;Hordowick, 2022). Since these inconsistencies can lead to serious miscommunications among forecasters themselves and with the recreational backcountry community, there is a need for improving the consistency and quality of the operational use of these cornerstones of avalanche hazard assessments.…”

“…While the use of predictive models is a possible approach for addressing these challenges, training such models on the existing data sets runs the risk of perpetuating biases and inconsistencies that are contained in the human assessments. Horton et al (2020c) concluded that a more prescriptive approach might be needed to numerically predict avalanche problem characteristics in an objective way.…”

A Large-scale Validation of Snowpack Simulations in Support of Avalanche Forecasting Focusing on Critical Layers

“…Due to their ease of use and easily interpretable results, CTrees have already been used several times in snow and avalanche research for exploratory analyses (e.g. Horton et al, 2020b). While the top node of a CTree represents the most significant split that divides the entire sample, the resulting subsamples are recursively split into smaller subsamples until the algorithm cannot find any significant splits in the the response variable anymore.…”

“…There are no other data sets that describe the avalanche conditions in such a structured and consistent manner for so many seasons and over such large areas. However, since human assessments are subjective judgments that can be influenced by operational requirements and practices, they can contain biases and inconsistencies (Statham et al, 2018b;Techel et al, 2018;Horton et al, 2020b). Particularly since the Conceptual Model of Avalanche Hazard (Statham et al, 2018a) is purely qualitative and does not suggest any quantitative links between observations and hazard components (e.g., likelihood of an avalanche problem), the assessment requires substantial human judgment and expertise.…”