A Review of Methods for Determining Landslide Hazard Models and the Applicability of these Models to Urban Environments John Garvey

Transcription

1 A Review of Methods for Determining Landslide Hazard Models and the Applicability of these Models to Urban Environments John Garvey Introduction Landslides are natural occurrences in our environment. By natural, I mean that the earth s surface regularly experiences the down slope movement of material. For human settlements located upon this surface, landslides pose a significant danger. It is the intersection of human development and natural processes which constitute a natural hazard (Guzzetti et al 1999). It is imperative to assess the degree of these hazards to life and property and to inform our decisions of these assessments. To this end, landslide hazard modeling has played a pertinent role and provides a critical application of geomorphic concepts. Many studies relating to landslide mapping introduce their research by giving figures about the number of lives lost or the damage caused by earthquakes annually. They continue by saying that urban areas are often subject to landslides and that the increasing percentage of population moving into cities and the development of cities into more hazardous areas leads to an increased need for accurate landslide hazard modeling. These introductory narratives give significance or provide the so what factor. Perhaps these perfunctory introductions are required to secure funding? But given that the significance of their research is often attributed to mitigating risk in urban areas, it is curious that few studies are conducted in cities and that their models do not include variables that are specific to the urban environment. This is not to say that landslide studies have not included anthropogenic influence. There is a wealth of research focused on the impact of human activity into natural spaces. Some of these activities that have been researched include: logging (Swanson and Dyrness

2 1975), agriculture (Clerici et al 2002), and road building (Guthrie 2002). These studies, although providing valuable insight into the dynamics of landslides, may not be applicable models for urban environments. Landslide hazard models generally have three steps. The first step is to construct an inventory of landslides in a given area. This inventory is either depicted as point locations on maps or turned into some sort of mapping unit of landslide density. The second step is to gather a list of instability factors that are thought to correlate with landslides occurrence. Clerici et al (2002) has grouped these factors into 5 categories: geomorphologic, geologic, hydrologic, land use, and vegetative. The final step is to determine the exact relationship between instability factors and the landslide inventory. This is done in two ways. Instability factors could be run through a model to generate a landslide distribution. This created distribution would then be compared to the landslide inventory which is considered the real distribution. Factors are tweaked until the created and real distribution fall within an acceptable error range. The second way is to correlate instability factors with the landslide inventory using some form of multivariate statistics. The aim is to determine the relative importance of each factor or set of factors in explaining landslide distribution. The assumption behind these methods is that conditions that generated landslides in the past will explain future conditions (uniformitarianism) (Guzzetti et al 1999). This paper will explore the techniques used to model landslides. To do this, I will conduct a literature review that specifically focuses on methods. The first section will discuss and critique how landslide inventories are created. Next, I will list and review some of the different instability factors employed by researchers and how these factors are constructed. Finally, I will examine the applicability of these methods in urban areas and make

3 suggestions regarding possible urban-specific inputs that could be incorporated into these models. Landslide Inventory The first step in constructing a landslide hazard model is to create an inventory of past landslides. These inventories have been generated in three ways: 1) landslides are identified using aerial or satellite imagery (Guthrie 2002); 2) qualitative accounts using media sources or interviews (Lee et al 2004) and 3) field observations. These methods are not exclusive and often field observations are used in tandem with remote sensing in order to ground truth results (Guzzetti et al 1999). Guthrie (2002) in his study of the effects of logging roads on landslide occurrence used stereo-pairs of aerial photographs to create inventories of landslide prior and post road construction. The aerial photographs ranged in resolution, the most resolute having a nominal scale of 1:15,000 and the least being 1:30,000. One problem he identified was the undercounting of landslides due to the inability to see slides underneath forest canopy and the difficulty in identifying smaller slides, < 500 m 2. In a subsequent article, Guthrie stated that the minimum identifiable slide of 500 m 2 translated on a 1:20,000 scaled photograph as a ~1 mm 2 area (Guthrie and Evans 2004). Satellite imagery of insufficient resolution also makes the identification of small landslides difficult. Lee, Chwae, and Min (2002) used satellite imagery with a resolution of 5.8 meters. Problems with resolution can be reduced with the use of more advanced technology such as Light Detection and Ranging (LIDAR) and the availability of better satellite imagery like IKONOS (1 meter resolution). The lack of resolution in imagery-based inventories results in the undercounting of small-scale landslides. Guthrie and Evans (2004) acknowledged that small landslide could account for 85 percent of the total number of landslides. The undercounting of landslides

4 has created unexplainable data points in graphs depicting size and frequency of landslides; it is generally accepted that power function curves best represent this relationship. These disconformities, estimated to be 75 percent of the data, have been attributed to undercounting due to map scale and minimum resolvable units (Stark and Hovius 2001). Another problem that I have not seen mentioned in the literature is errors generated in correlating environmental factors with an inaccurate landslide inventory. The second method used to gather a landslide inventory incorporates a mix of textual sources and/or interviews. Guzzetti et al (1999) in their study analyzed four newspaper, interviewed 24 eyewitnesses, and reviewed 180 technical and scientific reports. Chau, Chwae, and Min (2004) in his study of landslides in Hong Kong used annual reports published by the Geotechnical Engineering Department. In both cases, record keeping of the event was inconsistent across different time periods. Often, the magnitude and location of the events were often impossible to accurately determine. Chau et al (2003), in a study of rock fall occurrences, discusses in a long but relevant quote, the difficulty of using these sources. However, some information of these rockfall records, such as the time of occurrence, the size of boulder, and even the rate of occurrence, are missing in some of these reports. This makes the compilation and analysis of the data more difficult. In addition, the geology of the slope, the slope angle, the vertical and horizontal travel distances of the boulder were not reported in these reports. To overcome these inaccuracies, Guzzetti et al (1999) applied a confidence buffer around point locations to account for locational error. Like with aerial photographs, the results would undercount smaller landslides. Often the difficulties in constructing a complete inventory or in assessing the accuracy of reconnaissance inventories are addressed using field observations. Field

5 observations can be done in test sites like in Guzzetti et al (1999) or with windshield surveys, in some cases, using helicopters (Guthrie 2002). This method, although being the most accurate, is too costly and time consuming to be used over large areas (Chau, Chwae, and Min 2004). Instability Factors The second step in constructing landslide models is to gather data relating to the terrain. These characteristics are called instability factors. Instability factors are environmental conditions that are correlated with landslides. These factors include variables representing geomorphology, hydrology, geology, land use, and vegetation (Clerici et al 2002). These characteristics of the terrain are then classified and stacked, usually within a Geographic Information System (GIS). Guzzetti et al (1999) gives an interesting account of how these themes, when combined, create unique conditions mapping units. In their study of the Umbria and Marche regions, 522 unique conditions were identified. To fit into the scope of this paper, I will limit the review to instability factors that I feel are most challenging to capture in urban environments. These are slope characteristics, hydrology, vegetation, and land use. Geomorphology is usually represented by slope. There are several characteristics of slope that have been examined. These characteristics are generally derived from Digital Elevation Models (DEM). Slope angle is the most common. Clerici et al (2002) generated slope angle using a DEM that was created from a contour map with intervals ranging from 5 to 25 m. The contour lines were digitized on-the-fly and then interpolated in a raster grid. This article does not say what cell sizes were created. Clerici et al then classified slope angle frequencies into five categories using natural breaks, other studies have used equal interval (Ermini, Catani, and Casagli 2005). Another slope characteristics that is derived from a DEM

6 is slope curvature the concavity or profile of slope (Lee, chwae, and Min 2004, Ermini, Catani, and Casagli 2005). A final slope characteristic is aspect direction of slope. Slope aspect is related to rainfall, with different aspects receiving different amounts of rain (Smyth and Royle 2000). In order to assess these slope characteristics it is important to understand Digital Elevation Models and, in particular, the effect that poor resolution has on calculating slope angle, curvature, and collection area. According to the National Geospatial Development Center, low resolution DEM have several affects: 1)underestimate steepness on steep slopes and overestimate steepness on flatter slopes; 2) slope curvature decreases in magnitude; and 3) upslope contributing areas are overestimated in upper landscape positions but are underestimated in lower landscape positions 1. Hydrology, in this paper, will refer to any factor that directly refers to water. The most obvious is rainfall which is generally gathered at rainfall gauges. Up-Slope Catchment area, although derived from DEMs and denoting a physical area, is used to quantify the total input of water to the slope area (Vanacker et al 2003). Interestingly, few articles discuss methods used to collect rainfall data or the relationship between gauging stations and their study sites. Jakob and Weatherly (2003) give the most comprehensive review. They mention the effects of elevation in their site as well as discuss differences in rain gauges-- whether they record hourly precipitation and rainfall intensity. Smyth and Royle (2000) indicate that rainfall is not evenly distributed across the landscape with south facing slopes in his site experiencing greater rainfall than other aspects. The last instability factor category is land use. Land use is the one theme that incorporates human activity. The studies that I review use only a few land use types. Ermini, 1 This list was taken from an article, Multi-scale Terrain Analysis to Improve Soil Survey, that had no identified author. The link is: ( Last Viewed 3/17/07.

7 Catani, and Casagli (2005) uses rangeland, cultivated land, grassland, and woods. Ermini, Catani, and Casagli states that these land uses were used as general groups because they have been shown to have differential controls on landslide occurrence. Although no article details how land use layers were constructed, I believe that a combination of remote sensing and field observation were used. In a study of land use patterns and sedimentation transport to a river, Vanacker et al (2005) construct a land use map using photo-analysis. Urban Speculations & Factors One difficulty presented by urban environments is constructing a landslide inventory. Aerial photography is not effective for identifying slides where the terrain is under forest canopy. In cities, the majority of the land is covered under layers of concrete and landscaping. Chau, Chwae, and Min (2004), in his study of the hillslopes surrounding Hong Kong, addresses this issue by relying on more textual sources. These sources though tend not to be precise in locating slides and in undercounting total number of slides. Additionally, as in the Guzzetti et al study (1999), a variety of sources like newspapers and interviews provide a non-consistent form of data collection where records from different time periods collected different sets of information. A second problem with the methods employed relates to the complex landscape of cities. Land uses do not fit into regional areas like grassland or forest. Rather, the city is a mélange of buildings, roads, parks, landscaping, etc. This complex of land uses, when superimposed upon layers of instability factors, would create millions of unique combinations. Additionally, these different urban structures, if located on slopes, require engineering such as grading and retention walls. Smyth and Royle (2000) looked at one example of this when studying favella communities. Cut and fill methods used in these developments significantly altered the morphology of the slope. These parcel by parcel

8 alterations to slope characteristics challenge the use of DEMs whose resolution is not capable of detecting nuances on the parcel scale. These micro-patterns of slope use could lead to significant changes in slope hydrology and sediment transport (Vanacker et al 2005). The final problem relates to the accurate quantification of instability factors. For example, research reviewed for this study relied upon total rainfall to quantify the amount of water in the system. In cities, rainfall would only be one input, leaky pipes and irrigation would be other sources. Yang et al (1998) attempted to quantify the amount of groundwater recharge associated with leaking water mains and sewer lines. In many of the study sites, groundwater recharge from these artificial sources was greater than from precipitation. Conclusion Landslides as a natural hazard have been widely studied. The primary purpose of these studies is to understand landslide potential in order to assess levels of risk to human settlements. Cities, due to their dense populations and sprawling forms are considered to have more potential for harm and, therefore, landslide modeling could play a significant role in urban development. To mitigate this danger, landslide hazard models have been constructed. These models are constructed by combining an inventory of landslides with a variety of environmental factors relating to slope stability. The end product for this combination is the classification of lands based upon susceptibility to landslides or degree of hazard. One problem with these studies is that they are not conducted in urban areas. Additionally, the models that they employ do not include urban-specific variables such as building weight, slope complexity from terracing and retaining walls, and artificial sources of water such as irrigation and leaking pipes. This gap is more surprising considering the stated purpose of these studies to identify risks in urban environments. Perhaps these models of

9 landslides are not applicable in urban environments. Lee et al (2004) concludes, the methods used in this study are valid for generalized planning and assessment purposes, although they may be less useful at the site-specific scale, here local geological and geographic heterogeneities may prevail. This quote hints at the complexity that urban environments present and the difficulty of capturing a heterogeneous environment with models intended to be generalizable.

10 Source List Chau, K.T., R.H.C.,Wong, J.Liu and C.F.,Lee Rockfall Hazard Analysis for Hong Kong Based on Rockfall Inventory. Rock Mechanics and Rock Engineering, 36(5): Chau, K.T, Y.L., Sze, M.K., Fung, W.Y. Wong, E.L., Fong and L.C.P, Chan Landslide hazard analysis for Hong Kong using landslide inventory and GIS. Computer Geosciences, 30: Clerici, A., S., Perego, C. Tellini and P.,Vescovi A procedure for landslide susceptibility zonation by the conditional analysis method. Geomorphology, 48(1): Ermini,, L. F.,Catani, and N.,Casagli Artificial Neural Networks applied to landslide susceptibility assessment. Geomorphology, 66: Guthrie, R.H The effects of logging on frequency and distribution of landslides in three watersheds on Vancouver Island, British Columbia. Geomorphology, 43: Guthrie, R.H. and S.G. Evans Analysis of Landslide Frequencies and Characteristics in a Natural System,, Coastal British Columbia. Earth Surfaces, Processes, and Landforms, 29: Guzzetti, F., A.,Carrara, M.,Cardinali and P.,Reichenbach Landslide hazard evaluation: a review of current techniques in a multi-scale study, Central Italy. Geomorphology, 31: Jakob, M. and H., Weatherly A hydroclimatic threshold for landslide initiation on the North Shore Mountains of Vancouver, British Columbia. Geomorphology, 54: Lee, S.,U.,Chwae and K.,Min Landslide susceptibility mapping by correlation between topography and geological structure: the Janghung area, Korea. Geomorphology, 46: Ermini, L, F., Catani and N.,Casagli Artificial Neural Networks applied to landslide susceptibility assessment. Geomorphology, 66: Lee, S., J., Ryu, J.,Won and H., Park Determination and application of the weights for landslide susceptibility mapping using an artificial neural network. Engineering Geology, 71: Smyth, C. and S., Royle Urban landslide hazards: Incidence and causative factors in Niterio, Rio de Janeiro. Applied Geography, 20: Stark, C. and N., Hovius The characterization of landslide size distributions. Geophysical Research Letters, 28(6): Swanson F.J. and C.T. Dyrness Impact of clear-cutting and road construction on soil erosion by landslides in the western Cascade Range, Oregon. Geology Journal, 3(7): Vanacker, V., M., Vanderschaeghe, G., Govers, E., Willems, J., Poesen, J., Deckers and B. DeBievre Linking hydrological, infinite slope stability and land-use change models through GIS for assessing the impact of deforestation on slope stability in high Andean watersheds. Geomorphology, 52:

11 Vanacker, V., A.,Molina, G., Govers, J.,Poesen, G.,Dercon, and S.Deckers River channel response to short-term human-induced change in landscape connectivity in Andean ecosystems. Geomorphology, 72: Yang.Y., D.N. Lerner, M.H. Barrett, and J.H. Tellam Quantification of groundwater recharge in the city of Nottingham, UK. Environmental Geology, 38(3):