Safe Travel on the Spearhead Traverse
-Discussion-

Introduction Methodology Results Discussion References


Contents

Problems and Inaccuracies
Potential For Improvement
Additional Caveats
Utility and Immediate Applications
Acknowledgements
Further Reading


Image Courtesy of Lee Lau


Problems and Inaccuracies

Raster Resolution

     Regrettably, my analysis can only be as accurate as the data with which I started. I began with a Digital Elevation Model (DEM) of my area of focus that has a resolution of 30m, and is accurate to within 10m. This means that each cell is 30m², which entails the occlusion of any terrain variation that is smaller in magnitude than 30 m. Such occlusion of data is problematic for my analysis in several ways:
  • It fails to represent the slope at any given point within a cell with perfect accuracy. 
    • The slope for each cell is calculated by dividing the elevation difference between two cells by the length of a cell. 
    • It is a reasonably accurate measure of the average slope of a cell, but a given cell could contain a significant convexity or concavity within it that would be 'averaged' out by the DEM. 
    • This is a problem, because a difference of only a few degrees makes a big difference for an avalanche. As anyone who has watched ski patrol throw three or four bombs at a slope before it slides will tell you, it only takes one 'sweet spot' to release an entire slope.
  • It fails to identify ridgelines and terrain traps that are less prominent or shallower respectively than 30m.
    • Such ridgelines are safer travel paths, and so it would be useful to identify them for the purpose of producing a least-cost (hazard) path.
      • I tried a number of different ways to identify these smaller ridgelines:
        • I produced a Triangulated Irregular Network (TIN) from my DEM and attempted to use the Tin Edge tool to find hard breaklines, but it did not give me any values.
        • I created an accumulation array, positing that the areas of least accumulation would be ridgelines.
          • I created a flow direction grid, as well as an output drop raster, which shows the ratio of max. change in elev. from each cell along the direction of flow to the path length between centers of cells, expressed in percentages.
          • I computed a flow accumulation grid, using the flow direction grid as my input raster and the drop grid as my input weight raster (because doing so applies a weight to each cell depending on the max change of elevation in it; otherwise there is a default weight of one for each cell).
          • This helped me find a reasonably accurate representation of hydrology, but was inaccurate with respect to ridges. If I trusted this info, it would tell me that Whistler is a ridge.
        • I attempted to produce a Topographic Position Index (TPI)
          • To produce this model, I downloaded a toolbox called CorridorDesigner that replicates the topographic position index model from ArcView 3.x.
          • It separates cells into 1) canyon bottoms; 2) gentle slopes; 3) steep slopes; 4) ridgetops
          • Unfortunately, it was no more accurate than the curvature model.
        • Ultimately, I was forced to concede that your input resolution constrains the precision of your analysis, which of course makes perfect sense.
    • When people refer to terrain traps, they are rarely talking about large gullies; rather, terrain traps are nearly always smaller than 30m. I included them in my analysis merely to make the point that safe travel involves avoiding terrain traps, but my least-cost path likely fails to identify a single one. 

Wind Direction

     To calculate hazard due to windloading, I used the prevailing wind direction. This is problematic for two reasons. First of all, it doesn't account for the fact that the wind direction often deviates from the prevailing wind direction. In addition, it doesn't compensate for the effect that alpine environments have on the wind. For example, valleys channel wind in different directions than it was originally blowing; large areas of exposed rock might be considerably warmer than the surrounding snow or glaciers, creating localized pressure gradients.

     Additionally, there is often a lot of variation of aspect on slopes. There could well be pockets that windload even on 'safer' slopes. This is a problem, because as mentioned earlier, it only takes one 'sweet spot' to release an entire slope.


Subjectivity of Weights

     By nature, multicriteria evaluations are subjective. The weights I assigned my criteria are based on my 22 years of experience in the mountains. It should be mentioned that I do not consider myself an avalanche expert by any stretch of the imagination. I believe that I have a decent working knowledge of snow stability, and a sensible level of acceptable hazard. Furthermore, I believe that I know my limits in avalanche terrain and understand the risks involved in travelling through avalanche terrain. That said, Bruce Tremper makes the point that avalanche victims are often described as 'experienced' backcountry users, and that people usually vastly overestimate their avalanche knowledge. Though I like to think that I do not fit that category, I can't state with any confidence that I have the requisite knowledge or experience to make a truly educated guess regarding the appropriate weights each of my criteria deserves.


Runout Paths

     My least-cost route path may remain  mostly in terrain that isn't likely to slide itself, but it also appears to traverse terrain that is exposed to considerable hazard from above. I was simply unable to figure out how to create a buffer around these slopes. It would be a complicated procedure, because the runout distance of an avalanche depends largely upon the vertical distance that the avalanche runs, as well as the degree of the slope below.

     Even if I were able to figure out how to model that for each potential avalanche path, it may not be possible to construct a route that avoids them successfully. This is why the 'extreme' rating comes with the recommendation to forego backcountry travel entirely. If natural releases are coming down all around you, there might not exist a route that safely bypasses all of them. Avalanches often run surprisingly far, sometimes even going uphill on the other side of a valley.



Potential For Improvement

Finer Resolution DEMs

      Producing high resolution DEMs involves considerable expense. The 30m DEM that I used for my analysis is, as far as I have been able to tell, the highest resolution available for the Whistler area. Happily, it would only be necessary to produce high resolution DEMs for areas that see a lot of backcountry recreation, which are few and far between. Of course, people certainly do venture into more remote places, but they expect to be in the position of evaluating hazard for themselves anyway. This model is only intended to make the casual backcountry user a little safer. Therefore, it might well be feasible to produce such a hazard model for high-traffic areas despite the expense. This would likely require investment from resorts, or perhaps state or provincial governments, however, as Avalanche Centers rarely have a surplus of available funds.


Weather Events
     
     As mentioned in the introduction, avalanche hazard is largely dependent upon the interaction between constantly fluctuating weather conditions and terrain characteristics such as the ones I considered in my analysis. To be effective and practically applicable, avalanche forecast models such as the one I have produced must take into account current weather information such as snowfall, temperature, and wind direction. However, just as there are traffic maps that constantly update to account for current traffic conditions, so could avalanche forecast maps update to account for changing weather conditions. Popular backcountry recreation areas often have weather stations nearby in representative terrain. Whistler, for example, keeps track of the temperature and snowfall at a variety of elevations. This information could be used to constantly update avalanche forecast models.

     The biggest foreseeable impediment to such a system would be accurately and effectively modeling how the wind is affected by alpine environments. It would likely be very difficult to arrive at an effective predictor of the effects of windloading. It would accordingly be very difficult to monitor the snowpack depths and dynamics over a large area.


Subjective Weighting

     In the section above entitled 'Problems and Inaccuracies,' I called attention to the fact that one could question the accuracy of the weights I assigned to each of my criteria due to the inherent subjectivity and my status as an avalanche amateur. Multicriteria evaluations are inherently subjective, so this is not something that could be entirely mitigated. However, by calling upon the knowledge of experts such as UBC's Dave McClung, Bruce Tremper, and others - and there are a substantial number of competent avalanche professionals - one could probably produce a fairly accurate weighting scheme via an average of their responses. The weights would likely have to be adjusted to account for variations in climate (i.e. continental vs. maritime) in addition to other considerations.


Trial and Error

     Avalanche prediction is a very complex and challenging practice, as much art as science. Any computer-based prediction model would likely be fairly inaccurate at first. However, it might be possible to improve upon the model based upon trial and error. Once the platform is implemented, it would likely be fairly straightforward to include additional considerations.



Additional Caveats

Other Objective Hazards

     The model that I have produced only accounts for avalanche hazard; it does not account for other objective hazards such as crevasses. It is possible, and even likely, that my 'least hazardous path' actually traverses glaciers riddled with crevasses. Crevasses can be very dangerous, especially given the fact that they are often covered by a thin bridge of snow, rendering them invisible. Someone blindly following the 'least hazardous path' might well avoid becoming the victim of an avalanche only to die at the bottom of a crevasse.

     The picture at the top of this page accredited to Lee Lau displays the Platform and Curtain Glaciers riddled with crevasses. This picture was taken in a particularly low snow year, but crevasses are nearly always present in some locations.


Gizmo Madness

     In recent years, there have been significant advances in avalanche 'safety' equipment. Avalanche beacons have become more user-friendly, and backpacks with airbags intended to float you to the surface of the snowpack in the event of an avalanche have come to market. Such 'gizmos' contribute to a false sense of safety that Bruce Tremper calls 'gizmo madness' (Tremper 206). Many people act as though their beacon is a magical amulet that will keep them safe from avalanches, but this is far from the case. Tremper has observed that advances in technology merely push the 'stupid line' higher; that is, wearing a beacon might well push your personal level of acceptable risk higher, though this attitude could get you into trouble.

     Any advance in technology has the potential to have the same effect. It is a reasonable concern that my prediction model could do the same. Armed with the 'least hazardous path' downloaded on their trusty GPSs, people might march out into the backcountry with their blinders on, nose to their LCD screen. But recall, the path still crosses terrain that is on the higher end of the hazard spectrum. Even if it didn't, it traverses hazardous terrain that can still potentially slide. Additionally, weather conditions change very rapidly in alpine environments, and with them avalanche conditions. A slope that may be safe in the morning may slide after lunch. Thus, following the least hazardous path could potentially get someone into trouble.

     Prediction models, like avalanche beacons, are not substitutes for a brain and awareness of one's surroundings. That means that it is necessary to have at least a basic understanding of what terrain might be dangerous. While an avalanche hazard map such as the one I have produced is better than nothing at all, safety in the backcountry requires, and likely will always require, a certain amount of knowledge and experience.



 Utility and Immediate Applications   

Utility

     Despite the considerations I have listed above, a model such as the one produced in this analysis still has the potential to be reasonably accurate. It would be necessary to make several of the improvements I have made above, but were it not possible to accurately model windloading, for example, this would not mean that the model was useless. Even if the resolution remained at 30m, this avalanche forecast model would still be an improvement upon the current standard. It would demonstrate that certain aspects, elevations, and slopes require more caution than others, rather than asserting a blanket statement about the hazard in the entire backcountry.

     No model will ever predict the safety of a slope with 100% accuracy. Even an avalanche professional cannot predict the safety of a slope they are standing on, no matter how many tests he or she does, with 100% accuracy - though of course they can make reliable educated guesses. Avalanches do not lend themselves to prediction. One point on an otherwise stable slope could be unstable; if the slab above it is sufficiently elastic, a release could propagate from this point, triggering 'sympathetic releases,' wherein adjacent slopes slide as well, though they may be around a corner. These are the aforementioned 'sweet spots', and they are impossible to model.

     However, that doesn't mean that the model I have produced is useless. Uncertainty is inherent to any forecasting model: how often is the weatherman correct? Though an avalanche forecast generally has a higher premium on success than a weather forecast, uncertainty is nonetheless acceptable. Furthermore, I am not making any claims that a slope is 'safe'. Note that the scale on my hazard map identifies areas of greater and less hazard rather than going from 'safe' to 'hazardous'. This, I hope, effectively communicates that avalanche terrain is always potentially dangerous. Someone relying entirely on a map to tell them how to safely traverse a given area has no business being there in the first place.   


Education Tool
 
     I have acknowledged that the model I have produced has some deficiencies; it is not sufficiently accurate to rely on as an exact representation of actual avalanche hazard. Some may criticize it as entirely useless even as a hazard model, asserting that it might provide backcountry users with a false sense of confidence about the areas that appear safer. Even if that is true, the model I have produced would be useful as an educational tool. Such models could be used in avalanche classes to demonstrate the variability of avalanche hazard across an area that one might expect to travel. Even if it is not implemented as an actual hazard model, it would effectively demonstrate the inadequacy of the current 'low to extreme,' blanket-statement hazard model, as well as dispelling the notion that there are any completely 'safe' areas in the backcountry.



Acknowledgements

     First and foremost, I would like to thank Professor Brian Klinkenberg, Alejandro Cervantes, and Jose Aparicio, all of the University of British Columbia, for their invaluable direction and assistance. Without their patient tutelage and help, this project would not have been possible.

    I would also like to thank Lee Lau for giving me permission to use his photographs of the Spearhead Range. Lee is an avid backcountry skier who exemplifies the appropriate margin of safety one should observe in the backcountry. While he most often skis low-angle 'meadow-skipping' descents that many would find boring, when conditions are right for it he skis steep, gripping, fatally-exposed lines that would make most people shake in their boots. Lee is also an intellectual property rights lawyer, so I thought it especially prudent to ascertain that I had his permission before using his photographs. He has even more of them on his Spearhead Traverse trip report on his personal website.



Further Reading

     In the introduction, I mentioned that anyone interested in learning more about avalanche dynamics or snow safety should consult this section. I have provided two books that are both worth considering.

     In the section above entitled 'Problems and Inaccuracies', I mentioned several impediments to the completion of a practically applicable avalanche hazard map. The first was that a 30m DEM resolution appears to be too coarse. The solution to requires only the appropriate funds. While this may prove to be an insurmountable hurdle to the continuation of such a project, it is not conceptually challenging in the least. The other three problems I mentioned however demand further consideration. These were the difficulty of modeling the windloading of snow in GIS; the subjectivity inherent to weighting criteria; and the difficulty of modeling runout paths over a wide area. I have provided below some studies that would serve as good starting points towards resolving these issues.


Avalanche Safety; Dynamics

     Bruce Tremper's book 'Staying Alive in Avalanche Terrain', referenced throughout this study, is an invaluable guide to safe backcountry skiing. It should be considered required reading before venturing into avalanche-prone terrain. It is accessible and relatively easy to understand, and it deals almost exclusively safety in the backcountry, as opposed to snow science.

 Image courtesy of www.mountaineersbooks.org
     
   
 Those interested in the science behind avalanches should read 'The Avalanche Handbook' by Dave McClung and Peter Schaerer.

 Image courtesy of www.mountaineersbooks.org




Windloading

A broad overview of alpine climatology
  • This explanation provides a basic understanding of the way in which alpine climatology can affect wind patterns.
  • A brief excerpt: "Mountains have predictable wind patterns, which interfere with the thermal stratification in a typical way. Similar to the sea and land breeze in coastal areas, winds tends to blow from the highland into the valley at night (contributing to the above-mentioned inversions), and from valley to mountain during the day, when high elevations undergo a relatively faster heating than the lowlands."

Several studies have been published specifically on the spatial modeling of snow in alpine terrain:
  • Gauer, P. 2001. 'Numerical modeling of blowing and drifting snow in alpine terrain,' Journal of Glaciology. (47)156, 97-110.
  • Winstral et al. 2002. 'Spatial Snow Modeling of Wind-Redistributed Snow Using Terrain-Based Parameters,' Journal of Hydrometeorology, V. 3, No.5.
 
 There is also an Italian study that may be of some relevance:
  • Orlandini, S. and Alberto Lamberti, 2000.  'Effect of Wind on Precipitation Intercepted by Steep Mountain Slopes,' J. Hydrologic Engrg. V. 5, No. 4. pp. 346-354
  • This is a study on the effects of wind on rain specifically, and does not appear to address snow. Its purpose is to better understand the processes behind 'debris flow phenomena' in Cortina d'Ampezzo, a town in the Northern Italian Dolomites. As wind distributes snow far more effectively than it does water, their results are unlikely to be applicable to avalanche hazard mapping, their methodology might be. It would be interesting to see how they have quantified the distributive effects of wind in alpine terrain.


Subjectivity of Weights

     While I was unable to find any studies of the appropriate weighting of criteria for avalanche hazard mapping specifically, I was able to find approaches to the same methodology for a related study: landslide hazard mapping. The following three studies all take slightly different approaches to removing the subjectivity of weighting from their hazard models.
  • Balteanu, Dan, 2009. 'GIS Landslide Hazard Map of Romania', GIM International [online], Vol. 23, Issue 4.
    • Excerpt: A number of qualitative and quantitative models and methods are available for computing landslide-hazard and susceptibility maps, e.g. likelihood ratio, neutral network, logistic regression, and fuzzy-logic models. The choice of one or other is based on several elements, including the scale of assessment, degree of detail in a study, data availability and the purpose of the study. For our landslide-hazard risk map we use a Landslide Hazard Index (LHI) method based on quantitatively defined weighted values. Expert analysis, combined with a long history of landslide mapping and assessment and field experiments, play an important role in this method. The expert judgement involved a large number of studies and assessments undertaken at different scales, and geomorphological mapping of Romanian territory at the scale of 1:200,000. 
  • Gritzner, Mandy Lineback, W. Andrew Marcus, Richard Aspinali, and Stephan G. Custer, 2001. 'Assessing landslide potential using GIS, soil wetness modeling and topographic attributes, Payette River, Idaho,' Geomorphology, Vol. 37 Issues 1-2, pp.149-165
    • This study conducted an analysis of 559 landslides, and then applied a Bayesian probability model based on combinations of elevation, slope, aspect, and wetness to determine which criteria exhibited the closest relation to landsliding (it was elevation, followed by slope).
    • A similar approach could likely be used for avalanches, especially given that there are areas such as Rogers Pass wherein they keep meticulous records of avalanche characteristics.
    • Interestingly, they had the same problems that I did with raster resolution using a 30m DEM, and concluded that this was the only remaining impediment to theability of researchers to effectively model landscape scale landsliding
  • Sivakumar Babu, G.L., and M.D. Mukesh, 'Landslide Analysis in Geographic Information Systems,' http://www.gisdevelopment.net
    • Excerpt from abstract: GIS is being used to conduct multi-criteria evaluations for landslide hazard mapping. This paper demonstrates the ability of the GIS to incorporate the spatially varying data of ground elevation, soil properties, etc. in the engineering analysis of the slope stability. The key factor in landslide hazard mapping is the assessment of risk associated with the failures.


Runout Distances

     There are several studies regarding runout distances, but both appear to use historical data about the slide paths. One is avalanche related, and the other is related to pyroclastic flows. It would be worthwhile to look at both to see if some of their methodology might be more broadly applicable.



Additional Reading

     University of British Columbia professor Dave McClung has authored or co-authored a large number of studies of avalanche dynamics, some of which are particularly relevant to avalanche forecasting. Though I confess I have not had time to read through them all, here is a selection that appear to be relevant:

    • Borstad, C.P and D.M. McClung (2009). Sensitivity analyses in snow avalanche dynamics modeling and implications when modeling extreme events. Canadian Geotechnical Journal 46 (9), 1011-1023.
    • Grimsdottir, H. and D. McClung (2006). Avalanche risk during backcountry skiing - An analysis of risk factors. Natural Hazards 39 (1), 127-153.
    • McClung, D.M. (2005). Risk-based definition of zones for land-use planning in snow avalanche terrain, Canadian Geotechnical Journal 42 (4), 1030-1038.
    • Roeger, C., McClung, D. and R. Stull (2004) Verified combination of numerical weather and avalanche prediction models at Kootenay Pass, British Coloumbia, Canada, Annals of Glaciology 38, 215-222.
    • Floyer, J. and D.M. McClung (2003). Numerical avalanche prediction: Bear Pass, British Columbia, Canada, Cold Regions Science and Technology 37 (3), 333-342.
    • Haegeli, P. and D.M. McClung (2003). Avalanche characteristics of a transitional snow climate - Columbia Mountains, British Columbia, Canada, Cold Regions Science and Technology 37 (3), 255-276.
    • McClung, D.M. (2002). The elements of applied avalanche forecasting - Part I: The human issues. Natural Hazards 26 (2) 111-129.
    • McClung, D.M. (2002). The elements of applied avalanche forecasting - Part II: The physical issues and the rules of applied avalanche forecasting. Natural Hazards 26 (2) 131-146.
    • McClung, D.M. (2001). Extreme avalanche runout: a comparison of empirical models. Canadian Geotechnical Journal 38: 1254-1265.
    • Roeger, C., McClung, D., Stull, R., Hacker, J. and H. Modzelewski (2001). A verification of numerical weather forecasts for avalanche prediction. Cold Regions Science and Technology 33 (2-3), 189-205.
    • Keylock, C., D.M. McClung and M. Magnússon (1999). Avalanche risk mapping by simulation. Journal of Glaciology 45 (150): 303-314.
    • McClung, D.M. and J.Schweizer (1999). Skier triggering, snow temperatures and the stability index for dry slab avalanche initiation. Journal of Glaciology 45 (150): 190-200.
    • (thesis he refereed): Haegeli, P. (2004). Scale analysis of avalanche activity on persistent snow pack weaknesses with respect to large scale backcountry avalanche forecasting, Ph.D. Thesis, Atmospheric Science - Earth and Ocean Science, UBC, 250 pp.
    • Mingo, L. and D.M. McClung (1998). Crocus test results for snowpack modelling in two snow climates with respect to avalanche forecasting. Annals of Glaciology 26, 347-356.


Back To Results - Continue to References

Project Content and Site Design © Sam Wright - University of British Columbia - 2009