The following DRAFT manuscript has been submitted for consideration in GeoManitoba, the 65th Canadian Geotechnical Conference:
Landslide Risk Affecting Linear Infrastructure, and Risk Inferred from Linear Infrastructure
BGC Engineering Inc., Victoria, British Columbia, Canada
Geospatial data can be used to estimate the potential for development of landslides or other hazards. This can be illustrated in the form of landslide susceptibility maps, which show the anticipated future spatial distribution of landslides, or as landslide hazard maps, which include consideration of the anticipated temporal distribution, or expected frequency. Such hazard maps may be compared with the location and vulnerability of potentially affected elements at risk, including linear infrastructure such as roads, railways, pipelines or power transmission lines, to develop an a spatial understanding of future landslide risk, which is a measure of the expected loss over time. It may be less obvious that the spatial distribution of specific linear infrastructure, namely roads, may be used to feed back into the analysis of risk to other spatially distributed elements at risk, including human habitation and inhabitants. This paper will explore various spatial relationships to develop an understanding of qualitative risk to linear infrastructure, and then show how linear infrastructure (i.e. road distribution) can be used to further assess qualitative risk to human habitation, using large landslides in sensitive clay and seismic hazard in eastern Canada as working examples.
Les données géospatiales peuvent être utilisés pour estimer le potentiel de développement de glissements de terrain ou autres dangers. Ceci peut être illustré sous la forme de cartes de susceptibilité aux glissements de terrain, qui montrent la répartition future anticipée spatiale des glissements de terrain, ou que des cartes de risques de glissements de terrain, qui comprennent l’examen de la répartition prévue temporelle, ou de la fréquence attendue. Ces cartes de risques peut être comparé à l’emplacement et la vulnérabilité des éléments potentiellement touchés à risque, y compris les infrastructures linéaires comme les routes, voies ferrées, pipelines ou de lignes de transport d’électricité, de développer une compréhension d’un spatiale du risque de glissements de terrain avenir, qui est une mesure de la perte attendue au cours du temps. Il peut être moins évident que la répartition spatiale des infrastructures linéaires spécifiques, à savoir les routes, peut être utilisé pour alimenter de nouveau dans l’analyse des risques à d’autres éléments répartis dans l’espace à risque, y compris l’habitat humain et les habitants. Cet article explore différentes relations spatiales pour développer une compréhension du risque qualitative à une infrastructure linéaire, puis montrer comment l’infrastructure linéaire (c.-à-la distribution de route) peuvent être utilisés pour évaluer plus qualitative des risques à l’habitation humaine, en utilisant grands glissements de terrain dans l’argile sensible et des risques sismiques dans l’Est du Canada à titre d’exemples de travail.
Linear infrastructure can be affected by numerous natural hazards, including landslide, flooding or earthquake hazards, as a few examples. Where these hazards can be mapped with some degree of confidence, they can be overlaid in geographical information systems (GIS) by linear infrastructure to identify areas attracting higher levels of risk. Such analyses can serve to inform disaster response planning, which could involve either the development of redundant branches in the linear network, or provision of resources to respond for rapid restoration.
Linear infrastructure can be examined as a potential element at risk in risk analysis of natural hazards potentially threatening vital transportation or communication arteries. It may be less obvious that specific linear infrastructure, namely road networks, can also be used as a proxy for population density, thus serving as a foundation for risk analysis of natural hazard threatening human habitation.
This paper examines these dual roles of linear infrastructure of road networks in risk analysis: first as a direct input in evaluating risk to the road network itself; and second as an indirect input in evaluating risk to human habitation and related infrastructure. The paper uses eastern Canada as a working example, examining two different natural hazards to demonstrate the general concepts: large landslides in sensitive clay; and, earthquake shaking.
2 LANDSLIDE SUSCEPTIBILITY IN EASTERN CANADA
Quinn et al. (2010) presented a landslide susceptibility map for the Saint Lawrence lowlands of eastern Ontario and southern Quebec. That map was developed using the weights of evidence approach, a bivariate statistical method, using five different geospatial themes: ground elevation; soil type; overburden thickness; land use; and, flow accumulation in the drainage network. In recent work sponsored by CN, Canadian Pacific and Transport Canada, BGC Engineering was engaged to upgrade the model. Key improvements included:
– expansion of the model to cover the entire area of glaciomarine deposition;
– use of updated topographic data and improved methods in GIS for inferring drainage networks; and
– inclusion of additional geospatial themes in the model.
The upgraded model, shown in Figure 1, includes consideration of twelve different geospatial themes, of which five were excluded from the final model (i.e. LANDSAT satellite imagery; stream gradient; seismic hazard; bedrock type; and, land use), and seven were included: soil type; overburden thickness; stream flow accumulation; stream sinuosity; local relief; and, slope angle. The resulting model has been divided into four qualitative susceptibility categories: low; low to moderate; moderate to high; and, high. These qualitative categories can be assigned a quantitative meaning through further analysis of the available landslide inventory data, as illustrated in Figure 2. The “low” category, for example, corresponds to an expected spatial frequency of approximately 0.2 times the average spatial frequency within the overall study area, whereas the “high” category corresponds to an expected frequency roughly 40 times the average. With an understanding of the typical characteristics of large landslides in sensitive clay in eastern Canada, as described by, for example, Quinn et al. (2011), the map shown in Figure 2 may be taken as showing generalized landslide hazard within the study area. New qualitative descriptors have been assigned to the quantitative categories, so “high” susceptibility is now described as “very high” landslide frequency, and “low” susceptibility is now described as “rare.” These new terms are carried forward for the subsequent analysis of landslide hazard affecting linear infrastructure.
3 LANDSLIDE HAZARD AND RISK AFFECTING LINEAR INFRASTRUCTURE
The maps illustrated in Figures 1 and 2 present a spatial interpretation of the potential distribution of future large landslides in sensitive clay in eastern Canada. In order to estimate risk to potentially affected elements at risk, it is also necessary to determine whether such elements at risk are likely to be present given the occurrence of a landslide, to understand the vulnerability of such elements at risk given the occurrence of the hazard, and to understand the temporal distribution of the hazard. The following discussion examines presence of potentially affected elements at risk; temporal distribution and vulnerability are topics for later, more detailed study.
Linear infrastructure may be exposed to landslide hazards in sensitive clay. Quinn et al. (2011) document a number of case studies from the literature where specific infrastructure was damaged or destroyed by these hazards, including roads, power lines, gas pipelines, and a highway bridge. An examination of risk to specific linear infrastructure may commence by overlaying the landslide hazard model with a linear network. Figure 3 shows major roads in eastern Ontario and southern Quebec as one possible example. Greater detail of this interpretation is provided for the area around Montreal, Quebec, in Figure 4.
Examination of the map in Figure 4 suggests that landslide hazards are widely present, but localized, and overlap with the primary road network in selected areas. Figure 5 shows the landslide hazard model within a 1 km wide corridor along the primary road network. This interpretation serves to illustrate where landslides are more or less likely to affect the primary roads.
The map in Figure 5 shows that landslide hazards affect a relatively small proportion of the overall road network. Landslides can be expected to be rare, relative to average conditions, along 46.9 % of the primary road network. The moderate, elevated and very high frequency categories affect approximately 35.8, 15.7 and 1.6 % of the transportation network.
This mapping leads to a preliminary assessment of landslide hazard and qualitative risk on a regional scale, thus serving to establish priorities for more detailed study in selected areas. Such further study might include detailed air photo analysis to confirm or revise the understanding of the hazard, review of road construction details to assess vulnerability to different effects of large landslides, and potentially site reconnaissance and subsurface investigation.
The preceding analysis of hazard and risk affecting the primary road network can be repeated for other linear networks, including railroads, pipelines and power and communications transmission lines.
4 ROAD DENSITY AS A PROXY FOR POPULATION DENSITY
The previous section discusses hazard and risk associated with large landslides in sensitive clay affecting linear infrastructure. Other potentially affected elements at risk may be better represented as being either spatially distributed or point features. The following section discusses the use of road network data to make inferences about the spatial distribution of human habitation. Road density can be used as a proxy for population density, thus serving as the basis for modelling the spatial distribution of risk to human habitation associated with spatially distributed geological hazards. Here, “risk to human habitation” is intended to include, broadly, risk of casualties including injury and loss of life, or risk of financial loss associated with damage to human settlements.
One way to assess hazard and risk to human habitation associated with natural hazards across a large area is to compare, in GIS, the spatial locations of expected hazards with spatial locations of development. This may be modelled as a first estimate according to population density. Population data are available for Canadian municipalities from Statistics Canada (2007). Figure 6 shows municipal boundaries in southeastern Ontario and southern Quebec, and Figure 7 shows the inferred population density, by municipality, within the area of the former Champlain Sea.
Figure 8 shows a larger scale view of inferred municipal population density, focussing on the area around Ottawa, Ontario and Hull, Québec. The Regional Municipality of Ottawa Carleton is a single municipality, covering a substantial area. This municipality has a wide range of population density, as it includes: an urban core; suburbs in, for example, Orleans and Kanata; a number of small rural communities; and, extensive farmland and other green space in a broad green belt around the urban centre, and between the rural villages. Figure 9 shows the distribution of streets within the area, and one can see a high concentration of roads in the urban core and major suburbs, and much lower road density in the outlying rural areas.
Population data are available from Statistics Canada (2007) at a variety of scales of aggregation, including Census Subdivisions, which correspond approximately to municipal subdivisions (as shown in Figure 6), and Census Tracts, which are smaller subdivisions within larger municipalities. Figure 10 shows a plot of all road density (as km/km2) and population density (as km-2) data for census subdivisions in Ontario and Census Tracts in Quebec. This provides data covering several orders of magnitude in both population density and road density. A best fit relationship can be obtained as Pp ~ 28 PR2, where PP and PR are population density and road density, respectively. A probable lower bound relationship, capturing 85 % of the data (thus mean minus one standard deviation) is Pp ~ 13 PR2. Figure 11 shows the same data plotted on a log-log graph, demonstrating that the best fit relationship remains valid across a wide range of meaningful values, spanning population densities between 1 and > 10,000 km-2.
The relationship illustrated in Figures 10 and 11 can be used to generate a map of inferred population density, as shown in Figure 12. This map has concentrations of population density in the major urban centres of Ottawa-Hull, Montreal and Québec, as expected, and similar to the previous population density model shown in Figure 7. However, this model has significantly more resolution within municipal boundaries, as illustrated in Figure 13, which shows inferred population density in the Ottawa-Hull area compared with road distribution.
It may be noted that similar approaches to modelling population density have been used in health studies (e.g. Owens et al. 2010), but the author is not aware of previous work in natural hazard research.
5 QUALITATIVE RISK INFERRED FROM ROAD DENSITY
The preceding section presented a method for inferring population density from road density, thus providing a better resolution population density model for risk assessment than otherwise available from municipal population data. This model can then be compared with the spatial distribution of natural hazards to develop inferences about risk to human habitation associated with those hazards. This section uses two very different natural hazards – large landslides in sensitive clay, and earthquake shaking – to illustrate potential uses of this population density model in natural hazard and risk analysis at the regional scale.
Figure 14 presents a proposed risk matrix for large landslides affecting human habitation. The values within the matrix represent the product of the relative landslide frequency and inferred population density. These have been subdivided into arbitrary bins using a logarithmic scale, with “low,” “elevated” and “high” risk categories associated with products less than 1000, less than 10,000, or greater than 10,000, respectively. While the selection of these limits is arbitrary, yielding qualitative categories, they can be assigned to distinct quantitative meanings through further analysis. However, even without quantitative meaning, the qualitative categories are useful in support of landslide risk management, as they serve to identify areas of relatively higher or lower landslide risk. The “elevated” and “high” risk categories occupy only 7.1 and 0.3 % of the study area, respectively. Therefore further efforts to better understand and manage landslide risk can focus on spatial priorities involving relatively small areas within the overall study area.
Landslide hazards are only one of many natural hazards affecting human habitation in eastern Canada. Earthquakes are also relatively common in eastern Canada, resulting in higher seismic hazard than most of Canada apart from the west coast and areas in the Arctic. Figure 16 shows the distribution of seismic hazard in eastern Ontario and southern Quebec, represented as peak ground acceleration associated with the 1 in 475 year earthquake, as obtained from Halchuk and Adams (2008). Figure 16 also shows recorded earthquake activity as obtained from Natural Resources Canada (1012).
Seismic hazard in Canada is modelled as uniform hazard spectra, which allows design motions to be evaluated in consideration of the dynamic response of both the ground and built infrastructure. Buildings of different heights, for example, will respond differently to the higher or lower frequency components of an earthquake record. Similarly, the geological materials comprising the foundation may amplify or de-amplify ground motions, again depending on the frequency content of the earthquake record, but also depending on the stiffness and thickness of overburden materials overlying competent rock.
Site-specific and facility-specific considerations must be examined when evaluating site specific seismic hazard and risk. However, when examining the potential for impact to human habitation on a regional scale, it may be more convenient to defer to a more simplistic analysis, relying only on PGA. The Modified Mercalli Intensity (MMI) scale (US Geological Survey, 2012) provides a qualitative measure of earthquake damage from earthquakes, and has been correlated with PGA by various authors, including Wald et al. (1999). Table 1 provides a description of MMI values of interest in the study area, along with a proposed PGA scale.
Figure 16. Seismic hazard in eastern Canada, shown as peak ground acceleration for the 1 in 475 year event. Inferred PGA from Halchuk and Adams (2008) and recorded earthquakes from Natural Resources Canada (2012).
Figure 17 shows the MMI values in eastern Ontario and southern Quebec expected to be associated with the 1 in 475 year earthquake PGA values. This provides a qualitative indication of the anticipated level of earthquake damage in different parts of the study area associated with the 1 in 475 year event. Two things should be noted when examining this map. First, the expected MMI values are inferred from literature sources that relied on damage data from past earthquakes that often occurred in areas with lower standards of design and construction. The MMI values may therefore be somewhat conservative; however, the map is presented as a relative scale only. The absolute MMI values are less important than the range of values, and their distribution, which serves to focus interest to areas of more intense shaking. Second, damage will only occur where built infrastructure exists. Some areas of elevated MMI may suffer little or no future damage, due to the absence of huma habitation.
Table 2 presents a preliminary qualitative risk matrix for earthquakes in eastern Canada affecting human habitation. The qualitative risk values (i.e. L, M and H) have been obtained by multiplying population density by values ranging from 1 (for MMM = IV) to 100,000 (for MMI = IX), and subdividing for values less than 105, less than 107, and greater than 107.
Figure 18 shows the distribution of qualitative seismic risk to human habitation within the area of the former Champlain Sea in eastern Ontario and southern Quebec. The major urban centres of Ottawa-Hull, Montreal and Québec attract moderate risk, due to their concentrations of population and moderate seismic hazard. The highest seismic risk is shown to occur east of Quebec City on either side of the Gulf of Saint Lawrence, where population density is lower, but earthquake motions are expected to be higher.
The preceding analyses of landslide and earthquake risk to human population are both intended to be interpreted as preliminary, qualitative assessments, intended to help establish priorities for further study or allocation of resources for disaster response. These analyses are not suitable for interpretation at a local, site-specific scale, where local ground conditions and the nature of actual development will play very important roles in determining risk.
Landslide risk and earthquake risk tend to concentrate in different areas, depending on the distributions of human settlement and the applicable natural hazard. However, there is some significant overlap, as the three major urban centres of Ottawa-Hull, Montreal and Québec each attract elevated landslide and earthquake risk, relative to average conditions in the study area. Thus, these areas deserve heightened attention for further study or emergency preparedness planning. Areas of further interest for landslide risk include:
– Chicoutimi, Jonquière and La Baie area;
– Trois-Rivières and Bécancour area;
– Pembroke area;
– Bellefeuile, Sant-Jerôme, Prévost and Sainte-Sophie area;
– Drummondville area;
– Sorel-Tracey area;
– Victoriaville area;
– Shawinigan area;
– Batiscan area;
– Pont-Rouge and Neuville area;
– Matane area; and
While each of these areas attracts greater than average relative landslide risk, it may be noted that absolute risk can be expected to be quite low, as historical landslides in the study area have caused less than one fatality per year, on average, which is much lower than many voluntary risks accepted by residents in the study area, and probably lower than risk associated with many other natural hazards. However, this provides some initial guidance in establishing priorities for further work to mitigate landslide risk.
Earthquake risk, which is expected to be similarly low in an absolute sense, concentrates outside the major urban centres along both sides of the Gulf of Saint Lawrence relatively close to Québec. The areas with significantly elevated relative risk extend along the north side of the Saint Lawrence from Petite-Rivière-Saint-Paul and Baie-Saint-Paul to Saint-Siméon, and along the south side from L’Islet and Saint-Jean-Port-Joli to Rivière-du-Loup and Saint-Antonin.
A variety of natural hazards may affect linear infrastructure like roads, railroads, pipelines and transmission lines. Where reasonable estimates of hazard occurrence are available, these can be compared in GIS to the spatial distribution of linear infrastructure to develop an understanding of risk to infrastructure. This paper has presented a qualitative assessment of landslide risk affecting the primary road network in eastern Ontario and southern Quebec as a working example of this approach.
Road distribution can also be used to model population density, thus serving as a proxy basis for evaluating risk to human habitation. This paper has provided a population density model for the Saint Lawrence Lowlands, and has also provided preliminary qualitative risk models for large landslides and earthquake shaking affecting human habitation. The spatial distribution of roads is therefore a fundamental component in evaluating regional scale risk associated with natural hazards to both the road themselves and to the human settlement that led to development of the roads.
The writer would like to acknowledge Chris Bunce and Eddie Choi at Canadian Pacific, Mario Ruel at CN and Merrina Zhang, Lon Nadler and Geoff Anderson of Transport Canada for sponsorship of the work that led to refinement of the landslide susceptibility model presented in this paper.
Halchuk, S., and Adams, J. 2008. Fourth generation seismic hazard maps of Canada: maps and grid values to be used with the 2005 National Building Code of Canada. Geol. Survey of Can. Open File 5813, 32 pgs.
Natural Resource Canada. 2012. National Earthquake Database. Available: http://www.earthquakescanada.nrcan.gc.ca/stndon/NEDB-BNDS/bull-eng.php. [Accessed March 2012].
Owens, P.M., Titus-Ernstoff, L., Gibson, L., Beach, M.L., Beauregard, S., and Dalton, M.A. 2010. Smart density: a more accurate method of measuring rural residential density for health-related research. International Journal of Health Geographics. 9: 8.
Quinn, P.E., Hutchinson, D.J. Diederichs, M.S., and Rowe, R.K. 2010. Regional scale landslide susceptibility mapping using the weights of evidence approach: an example applied to linear infrastructure. Canadian Geotechnical Journal. 47: 905-927.
Quinn, P.E., Hutchinson, D.J. Diederichs, M.S., and Rowe, R.K. 2011. Characteristics of large landslides in sensitive clay in relation to susceptibility, hazard and risk. Canadian Geotechnical Journal. 48: 1212-1232.
Statistics Canada. 2007. Population and dwelling counts: A portrait of the Canadian population. Available at: http://www12.statcan.ca/census-recensement/2006/rt-td/pd-pl-eng.cfm [Accessed February, 2012]
United States Geological Service. 2012. The modified Mercalli intensity scale. Available: http://earthquake.usgs.gov/learn/topics/mercalli.php. [Accessed March 2012].
Wald, D.J., Quitoriano, V., Heaton, T.H., and Kanamori, H. 1999. Relationships between peak ground acceleration, peak ground velocity, and modified Mercalli intensity in California. Earthquake Sectra. 15(3): 557-564.