4. TERRAIN STABILITY MAPPING METHODS

[Back to TOC] [Previous] [Next]

The previous chapter summarized the uses, types, terrain attributes, map scales and units of terrain stability maps. Because of the many terrain attributes and parameters involved, there are many possible methods to produce a terrain stability map. Nine of the methods are applied to landslide initiation zones and discussed in Section 4.1. Four of the methods are applied to landslide runout zones and discussed in Section 4.2.

Useful reviews of terrain stability/landslide hazard and risk mapping methods have been prepared by Hansen (1984), Brabb (1984), Varnes (1984), Brand (1988), Hartlen and Viberg (1988), Hutchinson (1992), Gee (1992), Van Westen (1993) and Wu et al (1996).

4.1 In the Initiation Zone

[Back to TOC] [Previous] [Next]

In the initiation zone, landslide hazard mapping is most common type, although landslide risk mapping is sometimes carried out. Landslide hazard mapping usually involves, among other things, predicting and expressing the probability of a landslide occurring. The approaches vary from qualitative to quantitative. The following Methods A through I, modified from a classification proposed by Van Westen (1993), review the main aspects of these methods. A summary of the methods is provided in Table 4.1.

Method A -- Landslide distribution analysis requires the preparation of a process inventory map for individual landslides such as debris slides or debris flows, or for a group of landslides. The process inventory map is a simple, objective, but qualitative, form of a landslide hazard map. It shows the distribution and magnitude of recent landslide events by the number and size of landslides. It can then be used for more elaborate landslide hazard analysis. If used to calibrate other types of terrain stability maps, or by themselves, process inventory maps must be used cafefully, as they record only past landslide activity within a specific time interval. They provide no information on the landslide potential of areas other than those that experienced landslides during the time interval used for the study.

Some types of landslides, such as those involving the failure of thin colluvial material, are cyclical in nature. When such landslides occur at a given site, a period of stability follows while the forest regenerates and new surficial material develops. A landslide distribution analysis usually records those sites which failed recently, but ignores those which are mature and 'primed' for a landslide. Thus, a thorough landslide hazard assessment should also consider factors other than landslide distribution.

Landslide distribution analyses are particularly unreliable if a prediction of landslide hazards is required for changed conditions, such as following road construction, clear cut logging or reservoir flooding. In such cases, it is necessary to use statistical or judgmental extrapolation from areas that have already undergone such change, described below as a probabilistic univariate analysis.

The bibliography lists approximately 20 publications concerned specifically with landslide distribution analyses. These range from maps of large rock avalanche sites, for example Abele (1974) and Cruden et al (1988), through maps of debris slides, for example Rood (1984), to snow avalanches, for example Scheiss (1989).

Method B -- Landslide activity analysis is a refinement of the landslide distribution analysis, by which information is included on a process inventory map from several different time periods. Landslide activity analysis maps show changes in landslide sites with time. The objective, qualitative data are usually obtained from the interpretation of air photos from several different years. Landslide activity analysis still may not recognize areas which have not been active, but are potentially unstable.

The most useful landslide activity analyses are carried out for areas of slow movement where it is possible to distinguish sequential activity. An example of this method of mapping applied to land use planning in Switzerland is Bonnard and Noverraz (1984). A special application of landslide activity analysis is the comparison of landslide occurrence before and after a certain activity, such as timber harvesting, for example Swanson et al (1982).

Method C -- Landslide density analysis is a second possible phase in the processing of landslide distribution or landslide activity, and is used to calculate some form of landslide density, and therefore, although objective like the former two methods, is somewhat quantitative. This calculation may be done in three ways:

average the number of landslides per unit area in a map unit, for example Howes (1987);

calculate the percentage of unstable area in a map unit, for example O'Loughlin (1972); or

draw contours of equal landslide density (isopleths), for example Wright et al (1974), DeGraff (1985) and DeGraff and Canuti (1988).

The first method of calculation is suitable for medium or small scale mapping. The second is more appropriate for larger map scales, especially where the size of the unstable areas varies. The isopleth method is more suited to areas of weak rocks or fine grained soils, characterized by abundant and relatively deep seated landslides.

Landslide densities are sometimes subjectively grouped into 'susceptibility classes', for example Hicks and Smith (1981).

The limitations of landslide distribution and landslide activity analyses apply equally to landslide density analysis. The most important use of all three methods is to document past events and to provide calibration for predictive techniques using other terrain stability mapping methods.

Method D - Subjective geomorphic analysis involves the delineation of polygons based on several terrain attributes from air photo interpretation and fieldwork. The mapper then subjectively assigns a qualitative terrain stability/landslide hazard class to each map polygons, based on the air photo interpretation, field observations and his/her experience. The rules of assignment of the classes are not specified and may vary from one polygon to another. Geomorphic recognition of potentially unstable terrain is often strongly guided by observations of existing landslides. Computer-generated slope class maps may be used to focus subjective mapping. Subjective geomorphic analysis is highly flexible and can be very effective at a variety of scales and degrees of effort. Its drawbacks are a lack of repeatability and total reliance on the skills and experience of the mapper.

Kienholz (1978) provides an example of this method. The mapping is at a large scale and complemented by an elaborate system of illustrative geomorphological symbols. The 1:25,000 scale French ZERMOS maps also use a subjective geomorphic method (Champetier de Ribes 1987). The results are translated into a four colour traffic light zoning system: two shades of red distinguish two zones of landslide hazard; orange represents zones of potential, uncertain or minor hazard; and green delineates zones of no perceived hazards. The zone definitions differ from one region to other. Similar mapping was carried out in the 1970s in the United States by the US Forest Service (Bailey 1972 and 1974), and in British Columbia by the BC Ministry of Transportation and Highways (Haughton 1978).

Subjective geomorphic analyses were incorporated into the guidelines for the recognition of environmentally sensitive areas (ESA) by the BC Ministry of Forests (1992). The ESA mapping method provided general guidelines to assist the subjective decisions of the individual mapper, based on geomorphology and recognition of past landslide activity. The ESA mapping method has recently been revised and is now referred to as the 'reconnaissance mapping terrain stability' (BC Ministry of Forests 1995a).

Subjective geomorphic analyses are predominate in snow avalanche hazard mapping, for example Ives and Bovis (1978) and Freer and Schaerer (1980). This method of mapping is supplemented by snow avalanche probability of occurrence analyses on defined paths.

Method E -- Subjective rating analysis is an outgrowth of the subjective geomorphic method where an algorithm for the assignment of terrain stability/landslide hazard classes is established for the entire study area, as opposed to each polygon as in the latter method. The classes are assigned subjectively by judgmental weighting of various relevant terrain attributes, referred to as 'blind' weighting by Gee (1992). Although consistent in a study area, landslide hazard classes may vary between different study areas.

Relevant terrain attributes are usually assigned on the basis of map polygons. The terrain attributes most often used include slope gradient, surficial materials and geomorphic processes. Additional factors such as soil drainage, soil depth, and vegetation cover may also be used. Many subjective rating analyses include the presence of existing landslides as an important factor.

The complexity of the algorithm can vary considerably, from simple qualitative combinations of terrain attributes to complicated quantitative tables of weighting factors. Gee (1992) found that increasing the complexity of the algorithm, however, does not often improve the reliability of the results. Defining a subjective rating algorithm requires a high degree of specific local knowledge and experience. In theory, once an algorithm is defined by an experienced mapper, a less qualified person can be charged with collecting the terrain attributes and determining the landslide hazard classes. The algorithm, however, is not rigid and thus requires continual use of skilled judgement. There is some danger of oversimplification of the analytical results.

An advantage of a subjective rating analysis method is that a record of the procedure exists and the assignment of classes can be independently reviewed. Algorithms should not be exported outside the area in which they were developed and tested.

Many different subjective rating analysis methods have been used in different parts of the world. A few address only one type of landslide, for example Aulitzky (1980) which deals with the relative activity of debris flow fans, while most address the whole spectrum of landslides, for example Gee (1992). This method has been applied in a simple form at the 1:50,000 scale to the BC Terrain Classification System (Ryder and MacLean, 1980; Resources Inventory Committee 1996a).

The method of pre-harvest, 1:20,000 scale terrain stability mapping developed in the 1970s by MacMillan Bloedel Limited (Bourgeois, 1978) is a subjective rating analysis method. Over the years this method has been adapted and adopted by other forest companies and the BC Ministry of Forests (BC Ministry of Forests, 1995a). The terrain stability/landslide hazard classes are defined in terms of expected performance or probability of occurrence of landslides following timber harvesting or road building.

Method F -- Relative univariate analysis uses relative statistical methods to produce an objective, qualitative link between terrain stability/landslide hazard classes and actual observed performance of slopes. The relative correlation is based on the assumption that terrain units that have critical terrain attributes similar to terrain units that have failed in the past are most likely to fail in the future.

Most of the statistical methods reported in the literature are relative methods. This reflects the fact that landslides occur infrequently and therefore it is difficult to assign a numerical probability of occurrence based upon rigorous statistics (refer to Section 2.3). Most relative methods are therefore based on a spatial distribution of occurrence, and are useful in establishing a relative probability of occurrence for areas most likely to generate landslides.

The data of observed slope performance can take two forms:

landslide density expressed in number of events per unit area (this is suitable for relatively large numbers of relatively small events, especially at medium to small map scales); and

areal density expressed as the percentage of unstable area to total area (this is suitable for larger events and larger map scales).

In relative univariate analyses, the relationship between the performance data and the terrain attributes is examined separately for each attribute. This relationship represents a set of weighting factors which are added, or otherwise combined, to produce a relative terrain stability/landslide hazard class similar in form to the subjective rating analysis.

The relative univariate analysis is a simple and logical method which can be extended to consider more terrain attributes if necessary, however, the amount of work required in overlaying the various parameter maps and combining the weights is considerable. In addition, the analyses require a detailed landslide inventory with a large number of landslide events, otherwise, any calibration may not be statistically significant.

The advantage of the relative univariate analysis is that it allows the influence of individual terrain attributes to be studied. It is therefore useful in studies concerned with the selection of terrain attributes. Van Westen (1993) describes a method of constructing an algorithm by trial and error. The weights of individual terrain attributes were added to the landslide hazard class one by one and the result was examined by statistical comparison to the landslide density maps. Those terrain attributes which did not improve the correlation were rejected. The disadvantage of the relative univariate analysis is that it can only estimate relative probabilities of occurrence, not absolute probabilities of occurrence.

A pioneering example of the relative univariate analysis approach is the landslide mapping of San Mateo County, California by Brabb et al (1972). A detailed geological map and a slope map of the county were prepared. The per cent area within 35 geologic map units covered by landslide deposits was estimated by use of a grid overlay. The geologic map units were then grouped into six classes, from Class I with 0% area covered by landslide deposits to Class VI with 54%-70% covered by landslide deposits, representing the relative probability of landsliding from very low (Class I) to very high (Class VI). The landslide deposits themselves were shown as a separate class, Class L. Each geologic map unit was then further evaluated to determine which terrain attributes were critical for the occurrence of landslides. For example, if few or no landslide deposits formed on low slopes, the class number for the geologic map unit was reduced.

Rollerson (1992) applied a relative univariate analysis to a large sample of landslides in clear cut terrain, mapped at 1:20,000 scale, on the Queen Charlotte Islands. Nine terrain attributes were used in addition to natural and logging-induced landslide densities. About 50% to 80% of the terrain attributes examined were found to have a statistically significant influence on post-logging landslides. Different terrain attributes were significant for clear cut landslides as opposed to road related failures. Rollerson did not combine the individual factor weights into landslide hazard classes. Instead he produced a probabilistic multi-parameter classification, discussed below.

Method G -- Probabilistic univariate analysis uses objective probabilistic statistical methods to produce a quantitative link between terrain stability/landslide hazard classes and actual observed performance of slopes. A quantitative correlation extends the relative univariate analysis by assuming that the probability of future landslides can be predicted from the frequency of landslides in similar failed terrain units over a given time period.

In this method, a statistical correlation is sought between the probability of occurrence and a single terrain attribute or a prescribed combination of several terrain attributes (multi-parameter classification). The probability of occurrence is usually a spatial distribution, although in some cases where the landslide density map can be correlated with a time period, it is also expressed as a temporal probability of occurrence. The probabilistic univariate analysis is usually applied on the basis of terrain polygons.

This method is practical because it is simple to implement and test. Selection of relevant terrain attributes and definition of classes, however, requires careful and thorough work. A potential source of error, which is common to all statistical methods, is the quality and detail of the landslide frequency data on which the correlations are based. A further potential source of error is the delineation and classification of polygons by the mapper during the data collection phase. Because mapping variability can influence the resultant landslide frequencies correlated with a particular multi-parameter terrain class, combining individual terrain types into generalized classes reduces this problem somewhat, and tends to smooth over differences between mappers.

The probabilistic univariate analysis method has been used in a number of forestry related studies in British Columbia, for example Rollerson and Sondheim (1985), Howes (1987) and Rollerson (1992). In the forest industry background data consists of landslide occurrence during the critical 5-15 year time period following logging. The predicted spatial probability of occurrence relates to the same time period and can therefore be converted into a temporal probability of occurrence.

The choice of relevant terrain attributes and their use in establishing the multi-parameter classification is done by judgement or by trial and error by testing different combinations of parameters. The selection of terrain attributes can be guided by a parallel relative univariate analysis of each separate terrain attribute. Rollerson and Sondheim (1985) tried different classifications based on slope, slope morphology, surface material, aspect and the occurrence of natural landslides, and found that different classifications were needed for clear cut and road related landslides. Howes (1987) defined 15 multi-parameter classes based on landform, drainage, soil depth, slope angle and morphology and the presence of gully erosion.

Method H -- Probabilistic multivariate analysis uses objective multiple regression methods to establish a correlation between probability of occurrence and a group of terrain attributes. The method can be applied on a site specific basis, for example Pack (1985), or on an overlay polygon basis, for example Carrara (1983).

A simple version of probabilistic multivariate analysis is the matrix approach suggested by DeGraff and Romesburg (1984). Using overlays of maps delineated by terrain attribute polygons, they defined a separate class for each combination of independent terrain attributes. For example, using three terrain attributes, such as bedrock, slope and drainage, with four classes in each, the resulting matrix had 4 x 4 x 4 = 43 = 64 possible classes. While conceptually simple, the large number of combination classes, which can result even with a few terrain attributes, requires a detailed data base of landslide occurrences, to achieve statistically significant correlation.

More formal multiple regression and discriminant statistical analyses, using as many as 25 terrain attributes, have been conducted by Carrara (1983, 1991) with the help of a GIS. Van Westen (1993) tested similar procedures on a carefully mapped study area and found that no significant correlations resulted due to insufficient quality of the input data. He found that both relative and probabilistic univariate analyses produced satisfactory results with the same data.

The main disadvantage of the probabilistic multivariate analysis is that it excludes the experience and judgement of the mapper in producing correlations. Thus, the results are totally dependent on the quality of the data.

Method I -- Slope stability analysis methods usually use the infinite slope stability equation to assist with mapping. This equation, of which there are a number of variations, determines the factor of safety of a relatively long shallow slope segment with uniform or assumed characteristics. The factor of safety is used in the engineering sense as the ratio by which the shear strength of the slope material exceeds the shear stresses in the material. A factor of safety of 1 or less indicates that failure is imminent. Some authors suggest using constant material strength properties for a study area. This permits mapping of the variation of the landslide hazard primarily as a function of slope. This technique, however, is only meaningful in very small areas.

Slope stability analyses show either the distribution of the factor of safety (deterministic method) or the distribution of the probability of the factor of safety being less than one (probabilistic method). The deterministic method must be applied at grid points, while the probabilistic method can be applied to polygons (Hall et al, 1994).

The results of such analyses must be interpreted with care. Geotechnical engineers and geoscientists should be aware of the difficulty of obtaining a realistic factor of safety even at a single, thoroughly sampled and instrumented site because of the difficulty in determining the soil strength and groundwater parameters and the failure mechanism. Probabilistic analysis of the parameters is even more difficult and the probabilistic calculation of a factor of safety variation over a large area is as much based on judgement and informed guessing as any of the techniques discussed earlier. The apparent definiteness of the numerical output should not obscure this fact.

In addition, slope stability analyses neglect a number of items, including anisotropy of slope properties, seepage pressures, presence and strength of thin weak layers, effects of non-planar sliding surface, and three dimensional effects. Despite these cautionary remarks, slope stability analysis is a useful tool to improve judgmental assignment of landslide hazard classes, but should be checked against experience and actual performance.

The 'Level I Stability Analysis' (LISA), developed by the US Forest Service (Hammond et al 1992) is a computer-based probabilistic mapping method applied to polygons and is the best developed of the stability techniques. Level I mapping is intended for general resource allocation purposes and normally includes only limited field checking. Level II mapping, carried out in the project planning stage, is intended to predict the response of the terrain to specific treatment. It requires fairly extensive field work. Level III mapping is used for critical site stabilization before and during construction and requires detailed site specific fieldwork (Hall et al 1994).

4.2 In the Runout Zone

[Back to TOC] [Previous] [Next]

The methods reviewed in the previous section are primarily used to determine the spatial probability of occurrence of landslides in the initiation zone. That is, they show the distribution of probability of occurrence, magnitude and/or intensity where landslides are most likely to occur. The effects of the landslides, however, extend downslope, and therefore there is often a need to map the extent and nature of landslide runout. Landslide risk maps can be derived from the landslide hazard maps by including elements at risk, vulnerability and consequence.

Methods of terrain stability mapping in the runout zone are relatively few at present, but can be grouped into four, Methods J to M. The following reviews the main aspects of these methods. The methods are summarized in Table 4.2. It should be noted that landslide hazard mapping in the initiation zone is a pre-requisite for both landslide hazard and risk mapping in the runout zone.

Method J -- Hazard consequence analysis is a subjective extension of landslide hazard mapping applied to the initiation zone by which those zones are annotated to indicate the potential downslope damage should a landslide occur. Therefore, no separate mapping is required beyond the addition of a consequence rating to the landslide hazard polygon or point.

An example is the consequence classification used by Howes (1987) that provides a relative estimate of the potential of debris from a landslide entering a stream or a body of water, should a landslide occur within a particular polygon. In assigning a consequence, knowledge of the runout characteristics of the potential slide is implied, but not directly mapped.

Method K -- Runout zone analysis is used for many landslide and snow avalanche risk mapping projects, where the greatest potential for damage is in the runout zone. The initiation zones are identified and the landslide hazards are determined, but the initiation zones may not appear on the map.

The landslide runout zone may be shown simply as having or not having potential for being affected by the landslide hazard, or may be shown by zones based on landslide hazard intensity parameters such as velocity, depth of flow and/or deposits, range of impact pressures, or a combination of several parameters, and associated probabilities. For example, the Swiss national standard for snow avalanche zonation uses a combination of probability of occurrence and maximum impact pressure to define a 3-class "traffic light" system. A similar system was used for debris flow mapping in Colorado by Mears (1977).

In reconnaissance work, runout zones may be defined by a probability of occurrence. All elements within this zone are considered at risk until more detailed, site specific investigations are carried out. This type of runout mapping has been carried out in British Columbia for a number of Official Community Plans. A recent detailed study of debris flow hazards on the Cheekeye Fan near Squamish delineated four runout zones (Hungr and Rawlings 1995, Sobkowicz et al 1995). Each zone was associated with the probability of occurrence of three different size ranges of debris flows, each characterized by qualitative and quantitative descriptions of flow behaviour and likely consequences. The probabilities of occurrence were used to predict the probability of death to individuals (PDI) of the area and to estimate the probability of material damage to development on the fan.

There are a variety of methods for predicting the probability of occurrence, distance and character of landslide and snow avalanche runout. These include:

analyses of observational data for actual repeated events;

analyses of historical and dendrochronological evidence;

analyses of geomorphological and stratigraphic evidence;

empirical relationships concerning the flow path geometry;

empirical relationships concerning other parameters such as source area and magnitude; and

dynamic models of runout.

The methods of runout prediction are presently evolving. A review, oriented towards large rock slides, has recently been carried out by Hungr and Evans (1993). The international geotechnical societies (International Association of Engineering Geology, International Society of Soil Mechanics and Foundation Engineering and International Society of Rock Mechanics) have recently created a working group to prepare a 'suggested method' for the prediction of rapid landslide movement in the runout zone (Sassa 1993).

Method L - Linear path movement analysis includes mapping linear features such as debris flows paths, which generally follow recognizable lines or paths, and the resulting effects of that movement. The Gully Assessment Procedure (BC Ministry of Forests 1995b) begins by defining streams that are susceptible to debris flows. The stream channels are assessed for:

sideslope and headwall stability which indicates the potential of the sideslopes and headwalls to produce debris slides or flows;

channel stability determined qualitatively as a combination of available debris and transport capability of the channel; and

effects of debris flows considering the potential to deliver debris into a stream.

The procedure can be used to produce a qualitative map showing locations where delivery of debris into the stream system is most likely, where along the channel debris flows are likely to initiate, and potential downstream impacts. An extension of this system, which considers movement of erosion products and sediment transport in flowing streams, has been developed by Hogan and Wilford (1989) and is applied to an entire drainage.

Ellen et al (1993) carried out a 1:30,000 scale, GIS supported analysis of debris flows in Hawaii. A landslide hazard map for the initiation zone was prepared by a probabilistic univariate analysis, supplemented by data on average erosion rates of young volcanic terrain. Next a magnitude-probability of occurrence relationship for the initiation of a debris flow was established. The landslide hazard map was sampled at random points to generate random debris flows that followed existing drainage paths. An empirical relationship between erosion and deposition rates in cubic metres per metre of travel, and based on slope angle and degree of confinement, was used to determine the length of runout. The resulting movement lines were then transferred to the map to show the probability of debris flow damage at various points.

Method M - Landslide movement analysis is more general than linear path movement analysis in that it can be used for all types of landslide movement, linear or otherwise. It is a combination of thorough landslide hazard mapping in the initiation zone and thorough landslide hazard or risk mapping in the runout zone. Landslide movement analysis is more frequently carried out retroactively to document specific landslide events that have already occurred, rather than proactively to predict such events and the associated impacts.

There is, in principle, no great difficulty in combining terrain stability maps of the initiation zone, prepared by any of the methods listed in Section 4.1, with terrain stability maps of the runout zones, prepared by described in the previous paragraphs. Such combined maps are rarely prepared, however, perhaps because of the extensive work required in both the initiation and the runout zones.

Table 4.1 Summary of Terrain Stability Mapping Methods in the Initiation Zone

Table 4.2 Summary of Terrain Stability Mapping Methods in the Runout Zone

[Back to TOC] [Previous] [Next]

HTML Created:Mar 98

Copyright © 1997 Province of British Columbia

Published by the Resources Inventory Committee