Statistical runout modeling of snow avalanches using GIS in Glacier National Park, Canada

doi:10.1016/j.coldregions.2008.07.006

Cold Regions Science and Technology

Volume 54, Issue 3, November 2008, Pages 183-192

https://doi.org/10.1016/j.coldregions.2008.07.006 Get rights and content

Abstract

Using models to estimate snow avalanche runout distance is useful for areas where there is a lack of historical avalanche observations and no obvious physical signs of avalanche activity. Along roadways, details of avalanche runout are often recorded; however, in Canada, backcountry areas typically used by recreationists may not have a recorded history of avalanche activity or runout distances. Knowledge of predicted runout extents mapped in Geographic Information Systems (GIS) has the potential to inform backcountry users on route selection and decision making pertaining to slopes for skiing, snowboarding, ice climbing and snowmobiling. The Rogers Pass area in Glacier National Park, British Columbia, Canada provides an ideal location for studying well documented avalanche paths that impact the Trans Canada Highway, as well as representing a backcountry area that is a popular ski touring destination in Canada. A statistical approach using the alpha–beta runout model, first developed in Norway, has been adapted for use in Rogers Pass. Topographic parameters from well-known avalanche paths along the Trans Canada Highway corridor, with a historical record of over 40 years, have been extracted with GIS and used to calibrate the alpha–beta runout model. This model is then applied to an avalanche path in the Glacier National Park backcountry. A high resolution Digital Elevation Model (DEM) was created for the study area using digital stereo photogrammetry. A comparison of model calculations using the higher resolution dataset and a lower resolution dataset did not reveal any significant difference between the model parameters.

Introduction

Determining snow avalanche runout extent is an important consideration for mapping avalanche hazard for backcountry users, transportation corridors and other human infrastructure. The runout zone is the portion of the avalanche path where large avalanches begin to slow down and deposition of snow and entrained material occurs. The threshold for mapping runout in backcountry ski operations involves identifying a runout extent for a return period of less than 10 years (Canadian Avalanche Association, 2002). Runout zones can be identified through a combination of field observations, historical records, meteorological data and analysis of aerial photos and topographic maps for vegetative and geomorphic evidence (Canadian Avalanche Association, 2002, Mears, 1992, Weir, 2002). In areas where historical observations and field evidence are lacking, estimating avalanche extent may be difficult and runout models provide an option for estimating runout. Snow avalanche runout modeling is generally accomplished with statistical models, physically based models or a combination of the two approaches. The statistical models provide an estimate for runout along the centreline of the avalanche path and are limited in that they do not indicate the lateral extent of avalanches.

In 2004 Parks Canada developed an Avalanche Terrain Exposure Scale (ATES) based upon terrain and land cover characteristics to identify avalanche susceptible areas in the backcountry (Statham et al., 2006). Using a defining set of terrain based criteria, avalanche professionals designate “simple”, “challenging” and “complex” avalanche terrain. Elements of determining the exposure scale rely upon the identification of runout. ATES is also used as part of the Avaluator trip planner that takes into account the current avalanche bulletin (Haegeli and McCammon, 2006). Varying levels of caution are recommended depending upon the level of avalanche danger and the type of terrain the user is travelling in. Runout models offer the potential for ATES-based runout mapping to aid backcountry users in reducing risk.

Dynamic models are adept at indicating velocity and impact pressures along with avalanche runout and are especially suited for analysis where defence structures would be situated or impacts to forest resources or infrastructure are concerned; however, they require estimates of friction coefficients and release mass which may be unavailable or difficult to estimate in remote areas or areas with varying terrain cover and unknown snow depth. Small variations in these parameters can lead to great discrepancies in estimating runout distance (Lied, 1998). Statistical models using regression equations with simple topographic inputs, first introduced by Bovis and Mears (1976), are able to predict maximum runout but do not provide estimates of avalanche size, speed, force or lateral extent. For the purposes of hazard mapping for recreational users, the runout predicted by statistical models is sufficient to identify the extent of avalanche paths to a backcountry user. The alpha–beta regression runout model (Bakkehoi et al., 1983, Lied and Bakkehoi, 1980, Lied et al., 1989), initially applied in Norway by the Norwegian Geotechnical Institute (NGI) and applied in other parts of the world (Fujisawa et al., 1993, Furdada and Vilaplana, 1998, Johannesson, 1998, Jones and Jamieson, 2004, Lied et al., 1995) is well suited to topographic mapping. A modified version of the Norwegian model, the runout-ratio model, has been developed for extreme value distributions. The main assumption is that extreme avalanche events follow a Gumbel distribution as opposed to an assumed normal distribution of the residuals from a regression model (McClung, 2000, McClung and Lied, 1987, McClung and Mears, 1991, McClung and Mears, 1995, McClung et al., 1989). Comparisons between the runout-ratio model and the regression model have been made using different datasets. In Iceland, runout-ratio models did not show an improvement over regression models (Johannesson, 1998). For short path avalanche data in Canada, the runout-ratio model often predicted longer runout distances than the alpha–beta model, especially for large non-exceedence probabilities (Jones and Jamieson, 2004). For this study, the alpha–beta approach is used for predicting avalanche runout.

The terrain inputs required for the regression runout model are typically taken from topographic maps and field survey measurements; however, in Europe, Geographic Information Systems and digital elevation model (DEM) data (based on data with contour intervals of 20 m) have been used to extract these parameters (Furdada and Vilaplana, 1998, Lied et al., 1989, Lied and Toppe, 1989, Toppe, 1987), with the assumption that better or higher resolution DEM data would improve the models. One part of this study examines the assumption that higher resolution DEM data improves regression runout models by running a comparison between the results obtained in an alpha–beta regression analysis between low resolution DEM data and high resolution DEM data.

Variations of the alpha–beta regression model equation have been produced for diverse mountain ranges. This study focuses on developing a regression model specifically for the Columbia Mountains, Glacier National Park, British Columbia, Canada. Utilizing an extensive historical database of observed maximum runout records of over 40 years supplemented with expert knowledge along the Trans Canada Highway (TCH) corridor. The model is applied to an avalanche path in the Connaught Creek drainage, a popular backcountry ski area, as an example of the model's applicability for predicting runout extent in lesser known areas in the park.

Section snippets

Study area

The location of the study area is in Glacier National Park (GNP), British Columbia, Canada, which is situated in the Columbia Mountains of western Canada. GNP is located along the Trans Canada Highway and the main line of the Canadian Pacific Railway, approximately 350 km west of Calgary, Alberta. This area is locally known as Rogers Pass and is located between the towns of Golden and Revelstoke, British Columbia. The highway provides easy access to backcountry skiing in the park. The terrain

Methods

Avalanche centrelines and maximum observed runout positions were digitized based upon expert knowledge from the top of the starting zone to the observed maximum runout position. Digitized linework was referenced to both the high and low resolution DEM data (Delparte, 2008, pp. 88–89). ArcGIS® v 9.2 was used to extract relevant coordinate points into a spreadsheet to perform the calculations required for the regression model.

Results and analyses

To develop an alpha–beta regression model specific to Glacier National Park from the well-known avalanche paths along the highway corridor, we initially tested correlations between the variables in Table 1 for both high and low resolution datasets. Subsequently, we tested variations of the regression model and performed significance tests to determine if there was any difference between the equations derived from the high and low resolution datasets. This included a modification of the β point

Discussion

The primary objective of this research was to produce a runout model to approximate a 10 year return period. For a 40 year observation period (L), the encounter probability (E) that an avalanche has been observed with a return period (T) of 10, 50, 100 years is 0.98, 0.55, and 0.33, respectively. The relationship between E, T and L is expressed as E = 1 − (1 − 1/T)^L (LaChapelle, 1966, McClung, 1999). Hence the return period for the modelled avalanches can be labelled 10+ years. This provides a

Summary and conclusions

Alpha–beta statistical snow avalanche runout models were developed for Glacier National Park, Canada. Models with three predictors (β, H and Hy″θ) and a simple single predictor model (β only) were derived and cross validated. Similar to other studies referenced, the simplified β only model derived for this research has a high R² value close to that of the three predictor equations. The models were derived from both high and low resolution DEM datasets. A comparison of the models derived from

Acknowledgements

The authors would like to acknowledge and thank Bruce McMahon, Senior Avalanche Officer, Parks Canada, for the many hours he contributed to identifying topographic features in the GIS for this study. We would also like to thank Dr. Tak Fung of the University of Calgary for the statistical analysis support, Janet Rose for her assistance with stereo photogrammetry and DEM generation and Laura Bakermans for sharing her findings on short avalanche paths. The Selkirk College Geospatial Research

References (32)

McClungD. et al.
Statistical and geometric definition of snow avalanche runout
Cold Regions Science and Technology
(1987)
McClungD. et al.
Extreme value prediction of snow avalanche runout
Cold Regions Science and Technology
(1991)
BakkehoiS. et al.
Calculation of snow avalanche runout distance
Annals of Glaciology
(1983)
BovisM.J. et al.
Statistical prediction of snow avalanche runout from terrain variables in Colorado
Arctic and Alpine Research
(1976)
Canadian Avalanche Association
Guidelines for Snow Avalanche Risk Determination and Mapping in Canada
(2002)
Delparte, D., 2008. Avalanche Terrain Modeling in Glacier National Park, Canada. PhD Thesis, University of Calgary,...
FujisawaK. et al.
Estimating snow avalanche runout distances from topographic data
Annals of Glaciology
(1993)
FurdadaG. et al.
Statistical predication of maximum avalanche run-out distances from topographic data in the western Catalan Pyrenees (northeast Spain)
Annals of Glaciology
(1998)
HaegeliP. et al.
Avaluator Avalanche Accident Prevention Card
(2006)
HarbitzC. et al.
On probability analysis in snow avalanche hazard zoning
Annals of Glaciology
(2001)

JohannessonT.

Icelandic avalanche runout models compared with topographic models used in other countries

JonesA.S.T. et al.

Statistical avalanche runout estimation for short slopes

Annals of Glaciology

(2004)

LaChapelleE.R.

Encounter probabilities for avalanche damage, USDA Forest Service, Miscellaneous Report 10

(1966)

LiedK.

Snow avalanche experience through 25 years at NGI

LiedK. et al.

Empirical calculation of snow-avalanche run-out distance based on topographic parameters

Journal of Glaciology

(1980)

LiedK. et al.

Snow avalanche maps for the Norwegian army

Annals of Glaciology

(1989)

Cited by (24)

Snow avalanche hazard prediction using machine learning methods
2019, Journal of Hydrology
Snow avalanches are among the most destructive natural hazards threatening human life, ecosystems, built structures, and landscapes in mountainous regions. The complexity of snow avalanche modelling has been discussed in many studies, but its modelling is not well-documented. Snow avalanche modeling in this study was done using three main categories of data, including avalanche occurrence locations, meteorological factors, and terrain characteristics. Two machine learning models, namely support vector machine (SVM) and multivariate discriminant analysis (MDA), were employed. A ratio of 70 to 30 of data was considered for calibrating and validating the models. Results indicated that both models had an excellent performance in snow avalanche modeling (area under curve, AUC > 90), although hits and misses analysis demonstrated the superior performance of MDA. Sensitivity analysis indicated that the topographic position index, slope, precipitation, and topographic wetness index were the most effective variables for modeling. A snow avalanche map indicated that the high snow avalanche hazard zone was mostly near the streams and was matched with hillsides around the water pathways. Findings of study can be helpful for land use planning, to control snow avalanche paths, and to prevent the probable hazards induced by it, and it can be a good reference for future studies on modeling snow avalanche hazards.
Geospatial probabilistic modelling for release area mapping of snow avalanches
2019, Cold Regions Science and Technology
Citation Excerpt :
In this regard, geospatial techniques are very helpful for mapping the release areas of snow because of their tremendous capability to utilize various type of topographical information and management of large volume of data (Selçuk, 2013). Geographic information system (GIS) has been worldwide used for modelling and mapping purpose of avalanche release areas (Gleason, 1994; Furdada et al., 1995; Stoffel et al., 1998; Tracy, 2001; Hebertson and Jenkins, 2003; Hendrikx et al., 2004; Gruber and Bartelt, 2007; Delparte et al., 2008; Biskupic and Barka, 2010; Barbolini et al., 2011; Campbell et al., 2012; Selçuk, 2013; Snehmani et al., 2013; Snehmani et al., 2014). The studies performed by these researchers were based on the types of the factors affecting the avalanche hazard and hazard zones of the avalanche occurrence.
This paper presents GIS-based mapping of snow avalanche release areas using probabilistic occurrence ratio (POR) model. The proposed method utilized the topographical and avalanche occurrence dataset. The avalanche occurrence topographical factors such as slope, aspect, curvature and elevation generated from ASTER-GDEM V2 whereas vegetation cover from Landsat 8 OLI were considered in the proposed model. An avalanche inventory map of avalanche occurrence locations was prepared for training and validation of the model. From avalanche inventory map, 80% of avalanche occurrence data for training and remaining 20% was used for validation of the model. Finally, avalanche release area map was generated using the POR model and classified into five categories based on the Jenks natural break method. The R_A-index and ROC-AUC methods were used for validation of the avalanche release area map. The validation accuracy of the model is confirmed as 92.41% and 89.13% as per success and prediction rates.
Comparing methods for estimating β points for use in statistical snow avalanche runout models
2014, Cold Regions Science and Technology
Citation Excerpt :
DEM error arises from coarsely spaced or inaccurate data. While DEM resolution requirements have been identified in avalanche literature (Bühler et al., 2011; Delparte et al., 2008), less focus has been placed on comparing the two methods. As DEMs are integrated into the avalanche hazard mapping process, an understanding of their limitations and how they relate to the traditional field survey becomes more important.
Snow avalanche runout estimates are core to risk assessment and mitigation for infrastructure development and transportation corridors in mountain regions. Two statistical models, the Runout Ratio model and the alpha-beta model, estimate the extreme runout position using the point where the slope of the avalanche path centerline first reduces to ten degrees (β point). In North America, the β point has traditionally been determined through a field survey of the avalanche path runout zone; however, as they become more accessible, digital elevation models (DEM) are increasingly being used to determine β as part of a preliminary review. While DEM requirements have been identified in avalanche literature, more focus is required on reviewing field error and relating the two methods.
We surveyed 53 paths in western Canada, and estimated a field error distribution for the β point with an interquartile range of ± 2% of path length and a maximum range of ± 6.5% of path length. Five DEMs were sourced with spatial resolutions ranging from 1 m to 90 m. Of these, a 10 m DEM generated the most similar β point estimates to the field survey.
Validation of extreme snow avalanches and related return periods derived from a statistical-dynamical model using tree-ring techniques
2014, Cold Regions Science and Technology
Citation Excerpt :
These models typically consist of simple statistical regressions which explain observed runout distances with various topographic covariates. While they allow an estimation of maximum runout distances, they do not, however, provide estimates of avalanche size, speed, force or lateral extent (Delparte et al., 2008). The more physical characteristics of snow avalanches can, in contrast, be gathered from dynamical models (Voellmy, 1955).
Specification of expected runout distances and related return periods are the first and most important steps for zoning in snow avalanche prone terrain. In the past, runout distances of extreme events have often been evaluated with physically- or statistical-based numerical models. More recently, the statistical-dynamical modeling approach has been put forward, as it has the advantage of providing information on avalanche velocity, pressure, and flow depth at each point along a path quantified in terms of probabilities. Most often, calibration of statistical-dynamical modeling is based on existing data from historical archives so that current events with return periods ≤ 30 yr can normally be simulated with high confidence, but uncertainty increases as soon as one want to deal with longer return periods, thus calling for validation procedures to corroborate model predictions. In this context, we used dendrogeomorphic records of trees impacted by snow avalanches in their runout zone to reconstruct past activity in two avalanche paths of the French Alps. Based on the reconstructed distribution of runout distances of 25 events and mean event frequencies, we successfully derived runout values for events with return periods of ≤ 300 yr. Comparison of relations between runout distance and return periods between dendrogeomorphic data and predictions of a locally calibrated statistical-dynamical model show good agreement. Within the classical intervals used in hazard zoning (i.e. 10–300 yr), mean and mean square errors amounted to ~ 20 and 30−45 m, respectively. These results suggest that dendrogeomorphic time series of snow avalanches can yield valuable information to anticipate future extreme events and that the employed statistical-dynamical model can be used with reasonable confidence to predict runout distances of avalanches with high return periods, despite some uncertainty inherent to the limits of both approaches.
Can we learn more from the data underlying the statistical α-α model with respect to the dynamical behavior of avalanches?
2010, Cold Regions Science and Technology
Hazard and risk assessment in avalanche-prone areas involves estimation of runout distances of potential avalanches. Methods for determination of the runout may be divided into two categories: 1) methods based on statistical approaches such as the well known α–β model or 2) methods based on numerical avalanche models such as the PCM-model or VS-type models (just to name the more traditional ones). Methods in the second group have the advantage that besides the runout distance, velocity and impact pressure distributions along the avalanche track can also be obtained, this being a requisite for meaningful risk assessments. However, the predictive power of dynamical models depends on the use of appropriate rheological models and their parameters.
In the statistical α–β model, the maximum runout distance is solely a function of topography. The runout distance equations were found by regression analysis, correlating the longest registered runout distance of several hundred avalanche paths with a selection of topographic parameters.
In this paper, we re-evaluate Norwegian and Austrian avalanche data, which served as basis for the α–β model in the respective countries, and additional avalanche data with respect to dynamical measures. As most of those avalanche data originate more or less from extreme events (i.e. avalanches with return periods of the order of 100 years), the dynamical measures may give hints about an appropriate rheology for dynamical models suitable for extreme avalanche events.
The analysis raises reasonable doubt whether the classical ansatz for the retarding acceleration of snow avalanches with additive terms involving Coulomb-friction and a velocity-squared dependency, which is used in many avalanche models, is adequate for a physically-based model. Back-calculations of runout distances using a simple block model show a discrepancy between commonly proposed parameter values (and of the underlying rheological models) and the observations.
Snow avalanche formation and dynamics
2008, Cold Regions Science and Technology

View all citing articles on Scopus

¹: Tel.: +1 403 220 7479; fax: +1 403 282 7026.

²: Tel.: +1 703 993 4687; fax: +1 703 993 1216.

View full text

Statistical runout modeling of snow avalanches using GIS in Glacier National Park, Canada

Abstract

Introduction

Section snippets

Study area

Methods

Results and analyses

Discussion

Summary and conclusions

Acknowledgements

Cold Regions Science and Technology

Cold Regions Science and Technology

Calculation of snow avalanche runout distance

Annals of Glaciology

Statistical prediction of snow avalanche runout from terrain variables in Colorado

Arctic and Alpine Research

Guidelines for Snow Avalanche Risk Determination and Mapping in Canada

Estimating snow avalanche runout distances from topographic data

Annals of Glaciology

Statistical predication of maximum avalanche run-out distances from topographic data in the western Catalan Pyrenees (northeast Spain)

Annals of Glaciology

Avaluator Avalanche Accident Prevention Card

On probability analysis in snow avalanche hazard zoning

Annals of Glaciology

Icelandic avalanche runout models compared with topographic models used in other countries

Statistical avalanche runout estimation for short slopes

Annals of Glaciology

Encounter probabilities for avalanche damage, USDA Forest Service, Miscellaneous Report 10

Snow avalanche experience through 25 years at NGI

Empirical calculation of snow-avalanche run-out distance based on topographic parameters

Journal of Glaciology

Snow avalanche maps for the Norwegian army

Annals of Glaciology