This paper provides a qualitative risk model for large landslides in sensitive clay in eastern Canada. The model is based on prior work on landslide susceptibility, which is extended to regional landslide hazard, and inferred population density, which is taken to represent the spatial distribution of potentially vulnerable elements at risk. The resulting risk model’s validity is explored by comparing its predictions with the location of several historic damaging landslides. The model is found to be useful for supporting prioritization of further study or risk mitigation measures, and includes specific limitations: it is intended for the regional scale and cannot be interpreted in relation to individual specific facilities; it over-predicts risk in some areas; and, it does not completely address landslide risk associated with triggering by extreme earthquakes. The use of a 10 km by 10 km square grid provides a uniform indication of qualitative risk that may be used for regional prioritization and planning purposes. The three major urban centres of Ottawa, Montreal and Québec attract the highest risk due to their concentrations of human settlement and physical infrastructure. A number of other isolated areas contain smaller, but still significant, concentrations of landslide risk.
Sensitive clay, hazard, risk, Champlain Sea, large landslides
This paper examines hazard and risk associated with large landslides in sensitive clay in the Saint Lawrence Lowlands of eastern Ontario and southern Quebec in eastern Canada. The study area is illustrated in Figure 1, which shows the approximate extent of marine invasion of the Champlain Sea, represented approximately by the 200 m elevation contour, and the spatial extent of drainage basins draining into this area. This physiographic region is characterized by gentle sloping plains, carved by numerous minor rivers and streams draining into the Ottawa and Saint Lawrence Rivers, which bisect the area, and confined between the rugged uplands of the Appalachain mountains to the south, and Canadian Shield to the north. The area contains significant deposits of fine grained sediments that formed in a marine environment in the post-glacial Champlain Sea, which existed over 10,000 years ago following the retreat of the Laurentian ice sheet. Isostatic rebound has raised the region some 200 m, exposing the sediments to the atmosphere and susbequent runoff and erosion. Numerous landslides have developed as a natural stage of river valley development. Many of these landslides are very large, and they can occur suddenly with little warning. Large landslides in sensitive clay have caused fatalities, injuries and property damage, with the most notorious historic events being the 1908 landslide at Notre-dame-de-la-Salette, and the 1971 landslide at Saint-Jean-Vianney. Previous investigation by the author and collaborators has examined landslide suscptibility, or the relative tendency for large landslides to occur in different parts of the study area. This paper extends that work to examine risk, or expected loss associated with this hazard. This necessarily requires an understanding of the spatial distribution of potentially vulnerable elements at risk within the study area. An inferred population density model is compared with landslide susceptibility to obtain a qualitative risk model. The spatial representation of risk is examined in an effort to obtain priorities for further investigation or mitigation, and limitations of the risk model are discussed.
2 PREVIOUS WORK
2.1 Landslide Susceptibility
Landslide susceptibility may be defined as the tendency for conditions to exist that are favourable to the development of new landslides (Quinn et al. 2011a). This is an indication of the potential for future landslides to occur, but with no information about the expected temporal frequency or return period. A landslide susceptibility map will show where future landslides are more or less likely to occur, but will generally include no reference to annual probability of occurrence.
Quinn et al. (2010) presented a landslide susceptibility model for the western part of the study area. That work was based on a bivariate statistical method comparing landslide occurrence within a part of the study area (i.e. landslide inventory area in NTS 31H) with that of five other spatial themes: soil type; land use; overburden thickness; elevation; and, flow accumulation in the stream network. That model was extended outside the landslide inventory area to cover the western half of the Saint Lawrence Lowlands, and was validated for use at the regional scale through air photo study outside NTS 31H.
The original landslide susceptibility model presented by Quinn et al. (2010) was limited by several factors, including its spatial extent, and its inability to consider riverbank height, earthquake triggering and other conditioning or triggering factors. Additional work was done to expand the model’s spatial coverage and to improve its predictive power through incorporation of additional spatial themes. An updated and expanded model, with improved predictive power, has been presented by Quinn (in press 2012a). This newer model was also developed using a bivariate statistical method comparing landslide occurrence to other geospatial data, including: soil type; overburden thickness; flow accumulation; elevation; stream sinuosity; local relief; and, slope angle. The new model was developed with consideration of other factors, including land use, bedrock type, LANDSAT satellite imagery, stream gradient and seismic hazard. Each of these five factors was excluded from the final model for various reasons. Figure 2 shows the landslide susceptibility model, which is subdivided into four broad categories: low; low to moderate; moderate to high; and high. These qualitative categories can be described in terms of the average spatial density of landslides, as illustrated in Figure 3. The density of landslides within the “low” susceptibility areas is 0.22 times that of the average within NTS 31H. By contrast, density of landslides within the “high” susceptibility areas is 40 times the average, or about 200 times that of the “low” susceptibility areas. The spatial density of landslides ranges from 0 to 3 x 10-2/ha, with a mean value of 7 x 10-4/ha.
2.2 Landslide Hazard and Risk
The terms hazard and risk are used in this paper generally as proposed by Varnes and the International Association of Engineering Geology Commission on Landslides and Other Mass Movements (1984), and as used by Soeters and van Westen (1996). The following definitions have been modified slightly from those presented by Quinn et al. (2011a).
Landslide hazard may be defined as the probability of occurrence of a potentially damaging landslide phenomenon within a specified area and period of time. Landslide hazard maps delineate areas with respect to the annual probability of occurrence (or return period) of a specific landslide hazard.
Landslide risk may be defined as a measure of expected loss, given the potential occurrence of the landslide hazard. Expected loss can be examined in different ways. For example, one may compute the annual expected loss of life, or capital value of expected infrastructure damage, associated with the occurrence of a given hazard. Risk may be expressed as the product, or some other mathematical combination, of probability of occurrence of a hazard and magnitude or severity of potential consequences. Such consequences may be expressed as the product of the likelihood of exposure (i.e. probability of spatial correlation of the specific element at risk and the occurring hazard) times the vulnerability of the element at risk. Vulnerability is the tendency for a given element at risk to be harmed, given exposure to the hazard.
Risk can be examined at different scales, and may consider either a single hazard or multiple hazards. At the largest (i.e. closest) scale, one can examine risk to a specific individual or a single fixed facility. Examination of risk to an individual or single facility may serve to identify appropriate insurance coverage, or to determine specific mitigative measures necessary to reduce risk of specific damage or loss to some acceptable level. Examination of risk at a much smaller (i.e. regional) scale may be more appropriate for identifying spatial priorities, to support the development of a rational prioritization scheme for allocation of limited resources in efforts to reduce the expected loss over a very large area. This paper examines risk at the regional scale, covering the entire extent of the Saint Lawrence Lowlands, and extending to the boundaries of the major watersheds draining into these lowlands. This work is intended to identify specific areas that are expected to suffer greater losses due to future large landslides in sensitive clay, and thus may serve to assist in the prioritization of future more detailed study or planning or implementation of mitigative measures.
Quinn et al. (2011a) presented a preliminary discussion of landslide hazard and risk in eastern Canada. That paper discusses: the general nature of the specific landslide hazard; the spatial distribution of large landslides within the study area; the temporal distribution of large landslides in the study area; and, the interaction between landslides and receptors, or elements at risk. This latter component gives an initial indication of landslide risk. The following four paragraphs provide a summary of key findings from that paper.
Large landslides in sensitive clay are often described as retrogressive earth slides. A careful examination of the literature and review of available air photos suggests these landslides can be broadly categorized in three primary types: flake slides; lateral spreads; and, earth flows. Quinn et al. (2011b) have suggested that these three broad types may represent different points or extremes along a spectrum of behavior of a single landslide type; however, for the purpose of discussion of hazard and risk, it is convenient to talk of these three broad categories. Flake slides are translational earth slides where the displaced material moves laterally as a single intact block. These have been reported in the literature in Sweden and Norway (e.g. Gregerson and Loken, 1979; Eide and Bjerrum, 1955; Drury, 1968), but are believed to be uncommon or absent in eastern Canada, with a very small number of possible cases recorded in the literature (Grondin and Demers, 1996; and Conlon, 1966). Lateral spreads involve translational displacement accompanied by disruption and subsidence of the displaced material, resulting in a distinct “rib and ridge” or horst and graben morphology. These landslides typically involve little or no evident flow of liquefied clay. Canadian examples have been reported by Eden (1956), Ells (1908), Carson (1979), Wilson and Mackay (1919), and Bégin et al. (1996). Earth flows involve translation, flow and complete disruption of the displaced material, often resulting in a landslide crater with little or no remaining debris. Canadian examples have been reported by Tavenas et al. (1971), Schwab et al. (2004), Geertsema et al. (2006), and Hurtubise and Rochette (1956).
Lateral spreads and earth flows in sensitive clay in eastern Canada tend to have different geometric characteristics. At least 90 % of all large landslides in sensitive clay in eastern Canada are lateral spreads, which tend to be either short and wide, or roughly equi-dimensional. The remaining 10 % (or less) occur as earth flows, which tend to be either long and narrow, or bottleneck shaped. Earth flows tend to be much larger than lateral spreads, with median areas of 16,000 and 8,000 m2, respectively. The debris from earth flows also tends to travel longer distances than that from lateral spreads, with mean travel distances from cases reported in the literature being nearly three times as long for flows as for spreads, at roughly 1700 m and 640 m, respectively.
Large landslides in sensitive clay in eastern Canada, defined as events with least dimension larger than 100 m, have been estimated by Quinn et al. (2011a) to occur with a total average frequency of about 2.5 x 10-7 to 3 x 10-6 per hectare per annum, with a best estimate obtained from the geometric mean of roughly 1 x 10-6. The expected temporal frequency of large landslides in the low, low to moderate, moderate to high and high susceptibility areas is therefore 2 x 10-7, 4 x 10-6, 8 x 10-6, and 4 x 10-5 per hectare per annum. This is illustrated in Figure 4, which may be considered to represent a regional scale landslide hazard map.
The consequences of large landslides in sensitive clay have been documented in a number of case studies in the literature, as summarized by Quinn et al. (2011a). The tendency for damage may be overestimated from these case studies as one might expect these would have focused on damaging events, and that less attention would have been paid in the literature to notable landslides that did not cause significant damage. Nevertheless, the summary findings provide an indication of the potential for damaging effects to elements at risk, including human occupants and physical infrastructure. Most of the documented landslides caused damage to physical infrastructure such as buildings, transportation facilities, power lines or pipelines. Buildings were damaged or destroyed in 50 % of the case studies, and linear infrastructure was damaged in 38 %. Miscellaneous damage to other point facilities occurred in 22 % of the case studies. The reported damage was caused by loss of ground in the landslide crater in 50 %, and by flowing debris or flooding in 16 % of the case studies. All buildings, roads and other facilities located within the footprint of landslide craters were destroyed, suggesting 100 % vulnerability of any facilities within the landslide footprint. Vulnerability of facilities outside the landslide crater footprint to effects from flooding or debris impact is less well understood but is less than 100 %. Human occupants were present on or immediately adjacent to 50 % of the documented landslides, and fatalities occurred in just over half of these cases. In case studies where the exact numbers of both casualties and survivors are documented, there were a total of two fatalities and 78 survivors. This suggests that human vulnerability (i.e. the probability of casualties given the presence of humans in the event of a landslide) is less than 100 % and potentially much lower. It may be noted in support of this claim that most landslide casualties in eastern Canada were caused by two specific events: the Notre-dame-de-la-Salette landslide of 1908, with 33 fatalities, and the 1971 Saint-Jean-Vianney landslide, with 31 fatalities. In the latter landslide, a large but unknown number of residents and bus passengers escaped the landslide alive. In the former case, many inhabitants also escaped without injury.
Development of a landslide risk map from the landslide hazard map presented in Figure 4 requires knowledge of the location and vulnerability of elements at risk. When considering risk at the regional scale, it is proposed that one can estimate the likely presence of elements of risk – which may include human occupants or physical infrastructure – in a general way from an estimate of population density. Population density data provide a measure of the spatial probability of presence of elements at risk, but does not address vulnerability, which can be expected to vary based on many factors, including standard of construction as one example. While vulnerability will vary, perhaps considerably, for different facilities or occupants, it should be reasonable to expect that the typical, or average, vulnerability will be similar in different areas, assuming similar types of construction and patterns of settlement. Therefore it is proposed that population density, combined directly with hazard, yields a qualitative measure of risk. Detailed population density models are not directly available, and the resolution of population data varies across the study area, depending on the size of census space. Quinn (in press, 2012b) has presented a methodology for estimating regional scale population density from road network data, which is used in the following section to develop a qualitative estimate of landslide risk.
3 QUALITATIVE LANDSLIDE RISK
3.1 Calculation of Risk
Population density in eastern Canada may be inferred from road density according to DP ~ 28 DR2, where DP is population density in units of persons per square kilometer, and DR is road density in kilometers per square kilometer, as reported by Quinn (in press 2012b). An inferred population density model for the study area is presented in Figure 5, which shows ranges of population density on a logarithmic scale with decimal divisions.
Landslide risk has been estimated by multiplying relative landslide density, which is expected to relate linearly with landslide frequency, by population density. The mean population density is on the order of 150-200 persons/km2. Therefore, where relative landslide density times population density is less than 150 to 200, risk is less than average within the study area, and where it exceeds that value, it is higher than average. Table 1 presents a proposed risk categorization scheme for the study area, with three broad categories of risk: low; elevated; and, high. The categories have been derived from quantitative input parameters, but are considered to be qualitative outputs, useful for relative comparisons of risk, but less applicable for obtaining precise quantitative risk estimates. However, one can obtain a preliminary quantitative estimate of landslide risk to human life from basic available data. There have been nearly 80 confirmed fatalities in the study area due to large landslides in sensitive clay in eastern Canada in recorded history, over a period of approximately 100-150 years, in an affected area of roughly 85,000 km2. The average rate of landslide fatalities is therefore 6 to 9 x 10-6 fatalities per km2 per annum, and this average rate would be applicable where landslide density times population density is roughly 150 to 200, which corresponds approximately to the “low” risk category. This estimated risk level increases by about an order of magnitude for the “elevated” risk category, and by another order of magnitude for the “high” risk category. Similar analogies may be drawn for the potential for financial loss associated with damage to, or destruction of, physical infrastructure. While the numeric outputs discussed above should be treated with some skepticism, it should be evident that the proposed categories do provide a rational basis for making relative distinctions of risk. Therefore, the highest levels of landslide risk in the study area are concentrated in the “elevated” and “high” risk categories, which each comprise approximately 7.1 % and 0.8 % of the study area, respectively.
It is important to note that while the landslide risk map includes “elevated” and “high” risk categories of relative risk, even where risk is shown as “high” in a relative sense, it is still quite low in an absolute sense. The expected number of landslide-related fatalities, for example, is less than one per year within a population of approximately 7.6 million within the whole study area. This is much lower than many other important sources of risk. For example, traffic fatalities are far more abundant. However, since risk is concentrated in a relatively small proportion of the study area, the absolute risk in “elevated” and “high” risk areas is higher than average, and the qualitative risk map serves to focus attention for more detailed study or mitigation efforts.
3.2 Validation of Risk Model
The susceptibility model discussed previously is expected to isolate future landslide occurrence to a relatively small proportion of the overall study area. However, much of the study area is either unpopulated or infrequently used/occupied, including areas that have been mapped as moderate to high or high susceptibility. Therefore, many landslides can occur with minimal chance of impact to people, buildings or other infrastructure. The risk map, by contrast, is intended to isolate areas that are more or less likely to have future damage to existing facilities, or other impact, such as loss of life.
One way to test the validity of the model is to look at historic damaging landslides to see whether they occurred within higher risk areas of the map. Good location data are available for a small number of past damaging landslides, including the following:
• 2010 Saint-Jude, Quebec – house destroyed, 4 fatalities;
• 1972 Saint-Jean-Vianney, Quebec – majority of subdivision destroyed, 31 fatalities (Tavenas et al. 1971);
• 1924 Kenogami, Quebec – mill facilities damaged, complete mill shut down (Brzezinski, 1971);
• 1955 Nicolet, Quebec – school destroyed, church property damaged, 4 fatalities (Crawford and Eden, 1964);
• 1918 Saint-Thuribe, Quebec – 1 fatality (Morin, 1947); and
• 1908 Notre-dame-de-la-Salette, Quebec – numerous homes destroyed, 33 fatalities (Ells, 1908).
Each of these listed landslides are shown in Figures 7 to 11, which also show the inferred risk categories. Each of these landslides is located in an area of the map delineated as “elevated” or “high” risk. Additionally, the 1993 Lemieux landslide (Evans and Brooks, 1994) occurred in an area of “high” risk. While one person was injured by this landslide from driving into its crater, more significant disaster was averted by the prior evacuation and relocation of the rural town of Lemieux. Of note, the future occurrence of that landslide had been predicted from air photo interpretation by Gagnon (1972). Therefore, the damaging landslides that have been documented in the literature, and one more recent incident widely publicized in Canadian news (i.e. 2010 Saint-Jude), have each occurred in areas of higher than average risk, lending credence to the risk model.
3.3 Spatial Representations of Risk for Prioritization
The primary purpose in developing the risk model is to enable the identification of higher risk areas, to support the prioritization of plans for either detailed study or development and implementation of risk mitigation measures. A first attempt at prioritization may be made by considering risk within political boundaries. Figure 12 shows municipal boundaries within the Saint Lawrence Lowlands superimposed over the risk model. This map shows concentrations of risk near the three major urban centres of Ottawa, Montreal and Quebec City, plus sporadic areas of elevated risk elsewhere across the study area.
A first effort at spatial risk prioritization is shown in Figure 13, which shows the total, or aggregate, risk by municipality. This has been computed as a weighted sum of risk category for each pixel of the risk map within a given municipal area, with “elevated” and “high” risk pixels assigned a value of 10, and 100 respectively, consistent with their relative risk being greater than that for “low” risk by one or two orders of magnitude. The resulting map yields a number of high risk municipalities in eastern Ontario, much of which is not significantly affected by landslide hazard. This result is intuitively unsatisfying, as it places an overemphasis on eastern Ontario, due to the large average municipality area.
An alternative prioritization approach ranks the municipalities based on risk density, where this density is obtained by dividing the total (aggregate) risk by area. The result is illustrated in Figure 14. This approach yields results that are closer to expectations, with concentrations of risk in areas known to be affected by large landslides. However, the variable municipal size still influences the outcome. For example, Ottawa, which has the largest area of any municipality in the study area, is shown with a uniform moderate risk density across its entire extent. By contrast, the metropolitan areas around Montreal and Québec City have finer discretization of risk density, due to the finer division of municipalities. It may be expected that the Ottawa area has variable risk similar to that shown for the Quebec municipalities immediately north, across the Ottawa River from Ottawa, yet this variability is not evident in the map.
An alternative to municipality-based prioritization is shown in Figure 15, which shows total (aggregate) risk, calculated as previously described, for 1026 equal area 10 km by 10 km grid squares across the study area. In this case, risk density is directly proportional to total risk since the areas are equal. This approach therefore eliminates any bias associated with varying municipal area. In Figure 16, this grid-based risk interpretation has been modified to show the top ~ 5 %, 10 % and 20 % (i.e. top 50, 100 and 200) of all grid squares in terms of total risk. Examination of the higher risk grid squares in Figure 16 in comparison with municipal boundaries allows the development of an initial ranking of elevated risk areas, and these are summarized in Table 2. It can be seen that the three metropolitan areas dominate the risk assessment, with the greater Montreal area, including rural areas to the east along Richelieu River, containing 20 of the 50 highest risk grid squares. The Ottawa and Quebec City areas each contain six of the 50 highest risk grid squares. A section of Rivière l’Assomption contains four of these highest risk grid squares, and the Chicoutimi and Trois-Rivières areas each contain two. Ten other areas contain isolated high risk grid squares. These areas might be considered to rate as the highest priorities for further study; note, however, that the risk map in Figure 16 also contains several other clusters of grid squares in the top 10 or 20 %, so this priority list is not complete.
3.4 Limitations of the Risk Model
3.4.1 Resolution of the Model
Quinn et al. (2010) and Quinn (in press 2012a) both describe limitations on the applicability of the regional scale landslide susceptibility model, which represents half of the qualitative risk calculation presented herein. The landslide susceptibility model is meant to be interpreted at the regional scale, and should not be examined at larger scale, in relation to specific facilities. Such evaluations require a different, site-specific approach. The risk map is therefore also intended for use only at the regional scale, and cannot be used to guide decisions for specific localities or individual facilities.
3.4.2 Overprediction of Risk in Selected Areas
The landslide susceptibility model is expected to over-predict landslide incidence in certain parts of the study area. In particular, several areas within the core of metropolitan Montreal are mapped as having elevated susceptibility, and many of these have been confirmed by the author through air photo interpretation to have little or no potential for landslide occurrence. It has not been practical to complete air photo survey of the entire study area, so some parts of the map retain some degree of conservatism. Focussed attention on the mapped high risk areas, beginning with air photo analysis, should serve to eliminate areas of over-prediction of risk, and would therefore be a useful first step in any further work.
3.4.3 Risk Associated with Earthquake Triggering of Large Landslides
The previous discussion of qualitative risk, and its relation to quantitative risk, relies to some degree on data derived from historic damaging landslides. European colonization of the study area is relatively recent, extending no more than about 400 years in the past, a small fraction of the more than 10,000 years since retreat of the Laurentian ice sheet, during which time large landslides have occurred throughout the study area. Many of the largest landslides in the study area occurred prior to significant population growth. This includes more than a dozen large landslides east of Ottawa, dated by Aylsworth and Lawrence (2003) to major earthquakes about 4500 and 7000 years ago. This also includes a number of large landslides in Quebec associated with the magnitude M7 Charlevoix earthquake of 1663, including in particular one landslide occupying several square kilometers surrounding the crater of the 1971 Saint-Jean-Vianney landslide (Leggett and LaSalle, 1978), and more than one very large landslide at Shawinigan (Desjardins, 1980).
Figure 17 shows the large landslides east of Ottawa mapped by Aylsworth and Lawrence (2003), and illustrates that the susceptibility model does not do a perfect job of isolating extremely large landslides associated with very long return period major earthquakes. This important limitation in the spatial prediction of landslide incidence associated with major earthquakes extends to the risk model. Therefore the risk model may not capture all of the risk associated with large landslides triggered by extreme earthquake events.
Figure 18 shows the location of large landslides in and near Shawinigan mapped by Desjardins (1980), believed to be associated with the 1663 Charlevoix earthquake, as observed by Jesuit missionaries and confirmed through carbon dating. These landslides occurred prior to significant settlement, and the current population in this built up area is approximately 40,000. If an earthquake similar to the 1663 event had occurred and triggered similar landslides in this area subsequent to urbanization, one would anticipate a very large number of casualties and significant property damage. Such extreme events are not considered directly in the risk model, and might dominate risk calculations if it were possible to make a confident assessment of return periods and magnitudes, and associated loss.
The spatial comparison of landslide hazard and population density has resulted in development of a qualitative landslide risk model for eastern Ontario and southern Quebec of eastern Canada. The validity of the model has been investigated qualitatively by comparison with the locations of known historic damaging landslides. The main purpose of the qualitative risk model is to support the development of priorities for further study, or for planning and implementation of risk reduction measures. It should be noted that the risk model has specific limitations: it is intended to be interpreted at the regional scale, and not for specific individual facilities; it is known to overpredict risk in some areas, and so its findings should be checked through air photo study where warranted; and, it does not fully address landslide risk due to large landslides triggered by extreme earthquake events. The spatial representation of the risk model has been examined in different ways, and a uniform grid of 10 km by 10 km squares provides a useful basis for preliminary risk ranking. The major urban centres of Ottawa, Montreal and Quebec City have the greatest risk exposure due to the concentration of population, and a large number of other smaller areas have been identified as having elevated landslide risk.
The development of the landslide susceptibility model was sponsored by Transport Canada, CN Rail and CP Rail. The geospatial analysis conducted in development of the road density proxy for population density was sponsored by BGC Engineering Inc. as internal research and development.
Aylsworth, J.M., and Lawrence, D.E. 2003. Earthquake-induced landsliding east of Ottawa: A contribution to the Ottawa Valley Landslide Project, In: Proceedings of Geohazards 2003, 57-64.
Bégin, C., Evans, S.G., Parent, M., Demers, D., Grondin, G., Lawrence, D.E., Aylsworth, J.M., Michaud, Y., Brooks, G.R., and Couture, R. 1996. Le glissement de terrain d’avril 1996 à Saint-Boniface-de-Shawinigan, Québec: observations et donnés preliminaries. Current Research 1996-E, Natural Resources Canada, 215-223.
Brzezinski, L. 1971. A review of the 1924 Kenogami landslide. Canadian Geotechnical Journal, 8(1): 1-6.
Carson, M.A. 1979. On the retrogression of landslides in sensitive muddy sediments: Reply. Canadian Geotechnical Journal, 16: 431-444.
Conlon, R.J. 1966. Landslide on the Toulnoustuc River, Quebec. Canadian Geotechnical Journal, 3(3): 113-144.
Crawford, C.B. and Eden, W.J. 1964. Nicolet landslide of November 1955, Quebec, Canada. In: Engineering Geology Case Histories, 4: 45-50.
Desjardins, R. 1980. Tremblements de terre et glissements de terrain: Corrélation entre des datations au 14C et des données historiques à Shawinigan, Québec. Géographie physique Quaternaire, 34(3): 359-362.
Drury, P. 1968. The Hekseberg landslide, March 1967: An illustration of the Romerike clay landslide problem. Norwegian Geotechnical Institute, Publication No. 75, 27-31.
Eden, W.J. 1956. The Hawkesbury land slide. Division of Building Research, National Research Council of Canada
Eide, O., and Bjerrum, L. 1955. The slide at Bekkelaget. Geotechnique, 5: 88-100.
Ells, R.W. 1908. Report on the landslide at Notre-Dame de la Salette, Lièvre River, Quebec. Geological Survey Branch, Department of Mines, Ottawa, 1-15.
Evans, S.G. and Brooks, G.R. 1994. An earthflow in sensitive Champlain Sea sediments at Lemieux, Ontario, June 20, 1993, and its impact on the South Nation River. Canadian Geotechnical Journal, 31: 384-394.
Gagnon, H. 1972. La photo aérienne dans les etudes de glissement de terrain. Revue de Géographie de Montréal, XXVI (4): 381-406.
Geertsema, M., Cruden, D.M., and Schwab, J.W. 2006. A large rapid landslide in sensitive glaciomarine sediments at Mink Creek, northwestern British Columbia, Canada. Engineering Geology, 83: 36-63.
Gregerson, O., and Loken, T. 1979. The quick-clay slide at Baastad, Norway, 1974. Engineering Geology, 14: 183-196.
Grondin, G., and Demers, D. 1996. The 1989 Saint-Ligouri flakeslide: characterization and remedial works. Proceedings 7th International Symposium on Landslides, 743-748.
Hurtubise, J.E., and Rochette, P.A. 1956. The Nicolet Slide. In: Proceedings of the Canadian Good Roads Association, 143-155.
Leggett, R.F., and LaSalle, P. 1978. Soil studies at Shipshaw, Quebec: 1941 and 1969. Canadian Geotechnical Journal, 15: 556-564.
Morin, L.-G. 1947. La coulee d’argile de Saint-Louis (Comté de Richelieu). Le naturaliste Canadien, 74(5-6): 125-143.
Quinn, P.E. 2012a in press. Landslide susceptibility in sensitive clay in eastern Canada: some practical considerations and results in development of an improved model. Submitted to: Canadian Geotechnical Journal, April 2012.
Quinn, P.E. 2012b in press. Road density as a proxy for population density in regional scale risk modelling. Submitted to: Canadian Geotechnical Journal, April 2012.
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. 2011a. Characteristics of large landslides in sensitive clay in relation to susceptibility, hazard and risk. Canadian Geotechnical Journal. 48: 1212-1232.
Quinn, P.E., Diederichs, M.S., Hutchinson, D.J., and Rowe, R.K. 2011b. A new model for large landslides in sensitive clay using a fracture mechanics approach. Canadian Geotechnical Journal. 48: 1151-1162.
Schwab, J.W., Geertsema, M., and Blais-Stevens, A. 2004. The Khyex River landslide of November 28, 2003, Prince Rupert British Columbia Canada. Landslides, 1: 243-246.
Soeters, R., and van Westen, C.J. 1996. Slope instability recognition, analysis and zonation. Chapter 8 from Landslides, Investigation and Mitigation, Special Report 247, Transportation Research Board, National Research Council, National Agency Press, Washington, DC, 129-177.
Tavenas, F., Chagnon, J.-Y., and La Rochelle, P. 1971. The Saint-Jean-Vianney landslide: Observations and eyewitness accounts. Canadian Geotechnical Journal, 8: 463-478.
Varnes,D.J., and the International Association of Engineering Geology Commission on Landslides and Other Mass Movements on Slopes. 1984. Landslide hazard zonation: a review of principles and practice. UNESCO. 1-63.
Wilson, M.E., and MacKay, B.R. 1919. Landslide adjacent to Rivière Blanche, St. Thuribe, Parish of St. Casimir, Portneuf County, P.Q. In: Report on Mining Operations in the Province of Quebec during 1918, Quebec Bureau of Mining Annual Report 1918, 152-156.