SYSTEMS:THE WORLD, THE FLESH, THE METAL
Sue A. Ferguson
In the past it was not uncommon to close access to large areas of the mountains for long periods of time because of avalanche dangers. Now, modern management of mountainous areas is requiring roads to be kept open longer, larger areas for skiing be available, and safe environment for recreation and development. These management needs are placing a growing amount of pressure on those responsible for avalanche mitigation.
Unfortunately, many are finding that newly developed techniques for analyzing and controlling avalanche hazard simply are not adequate to handle the requirements of today's avalanche safety programs. Because of these inadequacies avalanche control workers are finding themselves more frequently in positions of risk. Not surprisingly, a recent study has shown that avalanche control work is much more hazardous than other professions of risk (Wilbour, 1991). There are many reasons why avalanche technology may be falling behind the needs of mountain management. For example, many avalanche forecasters now must look at so much data before making a decision that it is often overwhelming; or the data is summarized to such an extent that it no longer describes a true physical process. Another problem is that there has been little organization in transferring technology from the laboratory to the field. Many scientific studies that have been performed on a certain, finite set of circumstances are unilaterally (and incorrectly) applied to a broad range of conditions. Another related problem is that there are still many aspects of snow, mountain weather, and avalanches that remain unknown. Field workers seem to be asking the same unanswered questions each season.
Instead of dwelling on the unknown, this paper proposes ways to visualize the complex system of avalanches to more effectively manage avalanche hazards. One goal of this approach is to utilize all available data in a form that represents clear physical features and in a way that will not create confusion. Another goal of this approach is to utilize all available data in a form that represents clear physical features and in a way that will not create confusion. Another goal is to use those scientific tools, which are available, in more appropriate ways. Perhaps by visualizing avalanche phenomena in such a way, new insight into features previously not understood can be revealed.
Because avalanche activity depends upon the interaction of several big systems like patterns of snow stability, human exposure, and tools for control, the proposed techniques for visualizing avalanche problems must include ways to observe one or more simultaneous interactions. Unfortunately, traditional analytic methods that concentrate on analyzing individual components of systems may not be adequate to comprehend and manipulate this large, complex system.
Several years ago Stafford Beer, and English economist, suggested a more reasonable approach to understanding big systems that borrow from the mathematics of set theory (Beer, 1965). He suggests carving up the not-so-understood big system into interacting sub-systems. Comprehension occurs by watching the sub-systems interact, not smoothing and averaging the variance of each component, but allowing each to behave with its own, sometimes erratic, natural entropy.
Beer illustrates his ideas by describing problems encountered in popular big systems; the world, the flesh and the metal (themes loosely borrowed from Catholicism). Because the avalanche system is so large in itself, the following uses "Beer-like" categories, not as complete systems, but as subsystems of the entire avalanche phenomena. As a summary, some ways to help visualize the interacting components of these sub-systems are suggested.
The world of avalanches can be thought of as a subsystem of the earth's ecosystem. Fluctuating weather conditions build a variety of snow layers on top of undulating topography. This creates a complex system where the balance of strength to stress varies significantly over time and space. Components of this world-system include snow crystal shapes, layer densities, wind speed, air temperature, slope angle, ground cover, etc.
An example of an analytical approach that studies one component of this large, complex system is the relationship of depth hoar to avalanche formation. Observations of the "sugar-like" characteristics of depth hoar and rammsonde profiles (e.g., Armstrong and Ives, 1976) have led many to believe that depth hoar is least stable when it is fully developed. However, if depth hoar is observed as it interacts with other layers in the snowpack, closer understanding of the strength of the whole snowpack system that contains a depth hoar layer is obtained. Field observations of avalanche fractures have shown that less developed faceting of grains often plays a more important role in the formation of avalanches than fully developed depth hoar (e.g., Stethem and Perla, 1980; Ferguson, 1983; David McClung, personal communication). Indeed, more careful observations of depth hoar's mechanical properties have shown it to be significantly stronger in a variety of stress regimes than overlying layers of less well developed faceted crystals (e.g., Bradley and others, 1977; Armstrong,1986).
Observing how snow layers interact with each other in terms of stability has been done statistically (Ferguson, 1984). Using multi-variate cluster analysis techniques, it was found that unstable snow stratigraphy can be distinguished from stable stratigraphy if combinations of parameters are viewed simultaneously.
Statistical models are useful for observing several properties of interacting system components, even when little is known about their physical relationships. In addition, statistical models perform well for some forecast situations. However, when the components of such systems have a high degree of entropy, as shown by their variance about a calculated mean, these multi-variate statistical models reflect more nuances of the mathematics than physical relationships. They are unable to predict outlying events that occur beyond what has been defined as normal.
Avalanches fall under the category of high entropy systems. Components vary significantly both temporally and spatially. For example, densities measured in avalanching snow have a strong mean around 350 kg/m3. However, they range from less than 30 kg/m3 in very soft slab avalanches to nearly 1000 kg/m in slush flows (Ferguson, 1984). This causes statistically-based predictive models to fail when required to forecast "unusual" events, like those that so often plague avalanche control programs. Other statistically-based models that use weather inputs for avalanche prediction (e.g., nearest-neighbors approach by Buser and others, 1987; discriminate analysis by Bakkehoi, 1987) have similar difficulties.
Another way to understand the variety of snow layer combinations that produce avalanches is to observe components of the subsystem of snow interacting with components of other subsystems, like terrain and weather. In this way, it should be possible to begin predicting outlier events.
Most field personnel observe the interaction of parameters qualitatively and many local-area forecasters have developed graphical ways to help them follow the development of predefined critical factors. These "storm plots" typically graph meteorological data with avalanche occurrence as a time series. Some adjust the scale of their graphs to better observe patterns of instability (David Hamre, personal communication). Others have developed computer programs that will allow them to use the forecaster's experience to change defined critical factors on a daily basis. The computer searches a data base to locate days with similar conditions so that the forecaster can compare related past avalanche activity to help evaluate expected current activity (Gary Poulsen, personal communication).
Recently, some attempts have been made to combine physical models of snow metamorphism and related mechanics to develop quantitative forecast guidance using methods of expert systems (Lafeuille and others, 1987). As knowledge of snow and avalanche characteristics grows and computing power increases, these types of models could by a practical solution to the problems encountered when trying to understand and manipulate the avalanche world-system.
Human interaction with the avalanche world is what creates the hazards many are hired to mitigate. Components of the flesh-system are simply people who work, play or travel in and around avalanche-prone areas. Some characteristics of these components include an inexperienced, frequently falling skier in a group of more skilled back-country adventurers; a foreign visitor who does not respect closure signs; unsuspecting road travellers, etc.
Those who make a practice of studying human behavior are comfortable with Beer's suggestion to watch components of the flesh-system interact. However, many snow and avalanche scientists fall back on traditional analytic techniques when trying to comprehend or manipulate human behavior around avalanche terrain. These traditional observations have led avalanche specialists to attempt controlling human exposures to avalanche hazards by warning of dangers and regulating access. Using posted signs to control behavior is often ineffective, especially when there is nothing obvious that would indicate a hazard. Controlling usage of avalanche terrain is certainly difficult.
This newspaper headline, "Moderate avalanche danger causes only moderate death,' (2/10/85 News-Tribune, Tacoma, Washington) illustrates a common, misguided perception of the avalanche warning terminology, which has been adopted as a U.S. standard (Williams, 1980; Moore and Marriot, 1981). Studies on the use of avalanche hazard scales is still in its infancy in this country (e.g., Ferguson and others, 1986). More quantitative means of evaluating the content and interpretation of hazard ratings have been suggested by work in France (Lafeuille and Pahaut, 1988).
Difficult problems also have arisen in many areas where the usage of avalanche terrain must be regulated. Perhaps the most dramatic example was seen in Colorado, where the entire community became involved in solving the problems of "yo-yo" skiers (Woodrow, 1987). Yo-yo skiers have been defined as those repeatedly using privately controlled ski areas to gain access to National Forest public lands. Several fatalities among this group prompted a review board to establish guidelines for managing ski areas boundaries.
Another way to understand and manipulate the systems of flesh is to observe how decisions are made. Qualitative reviews of avalanche accidents (e.g., Stethem and Shearer, 1980; williams and Armstrong, 1984) are common ways to observe the decision-making process. Surveys of people who frequently travel in avalanche terrain is another (e.g., Sutherland and McPherson, 1986).
Other observations are able to group the "flesh-system" into separate classes. For example, statistics maintained on avalanche-related fatalities in the United States show that mountain climbers and ski tourers are more often victims of avalanches than other groups like snowmobilers, motorists, persons at work, or residents (e.g., Williams and Armstrong, 1984).
More casual observations can classify the flesh-system according to climate. For example, back-country users in areas with frequent storms often allow the weather to make decisions for them. That is, a stormy cold day with poor visibility will be reason enough not to descend a steep snow slope. This type of decision-making does not allow the back-country traveller to develop skills at interpreting snow stability, and their knowledge level can remain dangerously low for long periods of time. On the other hand, Back-country users in climates with frequent clear periods will explore vast areas of steep terrain. Their skills at interpreting snow stability are constantly tested. They reach high levels of knowledge very quickly.
Observing decision-making processes allow for better understanding of the flesh-system. This, in turn, should provide new insights for potential programs that are able to model such things as dispersed recreation, land use, and risk management. In the future, it may be possible to better manipulate and control the safety of those who work, live, or play in avalanche-prone areas.
The metal-system combines those machines and tools that we build yo help investigate, control, or create avalanches. Components of the metal-system include explosives, explosive delivery machines, thermometers, magnifying glasses, shear-frames, precipitation gauges, rescue transceivers, probe poles, computers, data management programs, prognostic models, etc. Since we have a hand in building these systems, we ought to be able to understand how they work and manipulate their outcome. This is not necessarily true.
Tools used to describe a complex system that do not allow for variability are often misinterpreted or of little use. For example, as 1970 the ram penetrometer was promoted as the key instrument for determining snow structure without pit excavation (Alta Avalanche Study Center, 1970). This metal rod is pounded vertically through the snowpack. It provides a complicated measure of compaction and disaggregation. The instrument works well for detecting thick layers of crusts. However, when thin layers or layers of more subtle strength properties are suspected, a ramsonde provides false or conflicting information. Recently it has lost its popularity as a tool for avalanche stability analysis.
Tools that account for variability provide more useful information. For example, the most common component of wind that avalanche forecasters usually are interested in is that which transports snow into the avalanche starting zones. This typically means wind speeds between 5 and 10 m/s as measured approximately 10 meters above a snow surface (Kind, 1981). However, today's heated anemometers can measure wind speeds from less than 1 m/s to well over 50 m/s. It is now possible for forecasters to relate low wind speeds with subtle metamorphic process on the snow surface, or high wind speeds with significant sculpting of avalanche terrain.
Another trait of a good machine is that it compensate for missing data in a reasonable fashion. For example, when an anemometer becomes rimed, the measured variation in wind direction decreases and measured wind speed is reduced. An observant forecaster would know that conditions exist to cause the instrument to become rimed. He/she will seek information from other sensors and with his/her knowledge of local terrain, estimate realistic values for wind speed and direction at the defective anemometer site.
Many instruments and physical models currently being developed can accommodate a wide range of variability. In addition, many process-oriented prognostic models now are built to adjust for missing data. Promising efforts have been made recently to use reliable data sets as inputs for models that recreate events for different times or locations. For example, CROCUS is a model that reconstructs snow layering at varying locations and times from available snow stratigraphy and meteorological data (Brun and others, 1989). Tools like these are giving the metal-system new dimension and applicability to avalanche mitigation efforts.
VISUALIZING COMPLEX SYSTEMS
Understanding the entire big-system of avalanches requires a way of watching components of each sub-system (the world, the flesh, and the metal) interact. Experienced observers have learned a great deal about small portions of the avalanche phenomena by visualizing the interactive data in a qualitative way. However, to use these observations to manipulate and control the avalanche phenomena, a more quantitative method of visualization is suggested that can include components of the world-system, flesh-system, and metal-system, simultaneously.
A recent step forward in understanding and manipulating big systems is the use of computerized geographic information systems (GIS). These allow large volumes of spatial data to be displayed simultaneously, usually using terrain as a base and other data as overlays. This method of visualizing information has been used to display information on historic avalanche activity (Cartwright and others, in press). However, because snow weather, and terrain configurations on a snow-covered slope change frequently over time, the full potential of GIS as a tool for avalanche forecasting requires prognostic models as the basis for some inputs.
Imagine a GIS built for a local ski area. Base data from a digital elevation model (DEM) shows all those undulations in terrain that affect mechanical properties of an overlying snow cover. These data are digitized from maps made by an experienced avalanche safety manager or consultant. The first overlay on this DEM ground surface would include fall vegetation.
An affordable array of sensors at the base, mid-way, and ridge-top collects information on wind, temperature, precipitation, perhaps relative humidity and snow depth, and maybe solar radiation. These data provide inputs into models that reconstruct the snow stratigraphy at each observation site. Meso-scale wind flow and precipitation prognostic models are then used to extrapolate snow stratigraphy to each grid point on the DEM. Additional programs that model solar radiation, relative humidity, and temperature are used to model heat and vapor fluxes to determine metamorphic processes in the snow cover. These would provide GIS overlays of snow stratigraphy over space and time.
Models that predict the mechanical strength of the snow layers are used to help determine areas of potential instability. These are updated and/or modified with observations from avalanche control activities. Placement of explosive shots can be pin-pointed on the DEM and subsequent effects over the entire area of the avalanche path can be calculated. Overlays of snow stratigraphy can be changed to accommodate the calculated change in mechanical structure.
Data from avalanche occurrence is continually added, either by personal observation, remote visible or infrared photography, radar, of pressure transducers. Time of the avalanche event, location, areal extent, fracture surface boundaries, and mass build-up in the run-out zone are all inputs into this imaginary GIS. These data may alter the base DEM or snow stratigraphy overlays. Additional overlays of frequent avalanche paths and structure locations can be added to help visualize potential problem areas. Also dispersed population data can be added from models and/or observation to aid in hazard assessments.
A forecaster with all this information available on a graphical computer can quickly review the entire system ar a glance, or watch interacting components as they unfold in time or space. Much like the captain of a space ship exploring an unknown planet, the avalanche forecaster can "fly" over the terrain, viewing coarse samples of information; or he/she can zoom in on a potential problem area and investigate small-scale components of the systems.
Many of the tools to build such a GIS currently exist. Most spatial data and that acquired historically on the scale of a medium-sized ski area can fit into common personal computers, like an IBM-compatible 386 PC with about 3MB of memory. Time-lapse displays of the information require larger computing power, like that found in work stations or mini-computers. With careful preparation and organization of each overlay, an GIS of this sort could be updated and improved with each new understanding of the system. Such a GIS may become a valuable tool for the avalanche safety manager who must deal with the pressures of a modern society.
Alta Avalanche Study Center, 1970, "Instrumentation for snow, weather, and avalanche observations," USDA - Forest Service, Wasatch National Forest, Snow Safety Guide Number 2, 80 pp.