LANDSLIDE HAZARD MAPPING BY USING GIS IN THE LILLA EDET PROVINCE OF SWEDEN

Transcription

1 LANDSLIDE HAZARD MAPPING BY USING GIS IN THE LILLA EDET PROVINCE OF SWEDEN Arzu ERENER 1, Suzanne LACASSE 2, Amir M. KAYNIA 3 1 Geodetic and Geographic Information Technologies, Middle East Technical University, 06531, Ankara/TURKEY, Phone: , Fax: , erener@metu.edu.tr 2 International Centre for Geohazards, NGI, Oslo, Norway, phone: , Fax: , suzanne.lacasse@ngi.no 3 International Centre for Geohazards, NGI, Oslo, Norway, phone: , fax: , amir.kaynia@ngi.no KEY WORDS: Hazard Mapping, FOSM, Probability Likelyhood ABSTRACT Quick clay slides occur frequently in Sweden and cause serious damages and costs. Reliable hazard and risk assessment approaches would help to mitigate the consequences of such landslides. This paper evaluates the hazard of clay slides in the Lilla Edet munucipality in Sweden using Geographical Information Systems (GIS). This tool provides a powerful tool to model the landslide susceptibility accounting for the spatial analysis and prediction. Landslide hazard areas are analysed and mapped using the landslide-occurrence factors, by the probability likelihood ratio method as well as by the first order second moment (FOSM) method. The governing factors such as slope angle, aspect, curvature, plan, and elevation are recognised for the probability likelihood ratio method. These factors are extracted from the light detection and ranging (LIDAR) data. The historical landslide locations are digitized from the hardcopy map of the region. The LIDAR data and aerial photos are used to confirm the historical landslide locations. In the FOSM the application of method a probabilistic slope stability model is integrated into GIS environment. The performance of the landslide mapping is assessed using the actual landslide locations and the relative operating curve (ROC) method. An ROC curve evaluates how well the methods predict landslides. The analyses show satisfactory agreement between the hazard map and existing data on landslide locations and validate the methodology used in this investigation. 1. INTRODUCTION Landslides in quick clay occur frequently in Scandinavian countries and cause damages and cost. After the Tuve landslide, the Swedish government decided to carry out a survey mapping of landslide hazard in built and planned areas within certain landslide susceptible sites (SGI Raport, 2006). The spatial probability of occurrence of clayslide in the Lilla Edet municipality in Sweden is analysed to help to reduce the consequences of such slides. For assessing landslide susceptibility, different methodologies are available in the literature. In this study two different approaches were used. The first approach was based on a univariate probability analysis. In this analysis, often called the frequency ratio model (FRM), the spatial relationship between the landslide locations and each landslide-related factor is analysed (Lee and Tu Dan, 2005; Lee and Pradhan, 2006). Frequency ratio values for each category of the influencing factors are computed using Geographical Information System (GIS) tools. Also the relationship between landslide related factors and the landslide locations are analysed by GIS tools. Then these relationships are used for each factor's rating in the overlay analysis in the FRM method. The ratings of the factors are summed up to give a landslide susceptibility index and susceptibility map. A second method applied in this study for susceptibility mapping is the First Order Second Moment (FOSM) method. A FOSM method is one of the widely used methods due to its relative simplicity in formulation and straightforward mathematics (Chen et. al., 2007). This method was applied recently by Duzgun & Ozdemir (2006); Duzgun & Karpuz, (2006), and Chen et. al., (2007). In this study the shear strength was combined with the slope

2 map of the clay slide area, based on a probabilistic slope stability model, in order to assess the hazard and generate a susceptibility map by using GIS for the Lilla Edet municipality. The predictive ability of the models generated by FRM and FOSM models are compared by using the Relative Operating Characteristics (ROC) curve. The ROC curves signify how well the models predict landslides. As a result of these comparisons, the FRM model appeared to be a better tool than the FOSM Study Area The study region, with an area km 2 (Figure 1), is located on the southwest part of the municipality of Lille Edet in the Västra Götaland province and it is situated in the southern part of the Göta Älv River valley in southwest Sweden. The Göta River Valley is the area with the highest frequency of landslides in Sweden. This is due to its special geological history with deep deposits of soft clay surrounded by outcropping bedrocks. Serious landslides covering large areas have occurred although most of the landslides are small and have relatively shallow slip surfaces. Figure 1. Study region in Lilla Edet For the hazard assessment of landslide in LilleEdet municipality the LIDAR data were used to extract the factors that cause instability of the slopes. A detailed digital terrain model (DTM) was accomplished by using the Airborne Laser scanning data and it was used as a basis for stability calculation purposes. Laser scanning of the terrain from an aircraft can deliver a detailed DTM with much better accuracy compared to existing topographic maps. The TopEye airborne topographic survey system was used to scan the topography covering, in the surroundings of the Göta Älv River in the area. The LIDAR system delivers 5-10 measurements per square meter, and the accuracy in the x, y, and z measurements are better than 10 cm.(sgi, 2006).In this study, DTM created with 5m 5m resolution was used to extract factors that cause instability of the slopes. From the generated DTM the slope, aspect, curvature, and curvature plan were calculated. The historical clay slide boundaries were drawn from the hardcopy map of the study region and their locations were verified by the contour maps and aerial photos. After analysis of landslide related factors and landslide locations, the relationship of each factor and landslide locations was used as each factors rating in the overlay analysis. So the ratings of the factors were summed to give a landslide susceptibility index and susceptibility map. For the calculation of FOSM method the shear strength, which is the basic variable in the formula of reliability index, was calculated for three different values of slope failure depths, namely 5, 10 and 15 m. The computations are described in more detail in the following. 3. APPLICATION OF METHODS Frequency Ratio Method (FRM) A FRM provides a correlation between the historical slide locations and the various influencing factors under consideration (Lee and Pradhan, 2007). To apply the probabilistic model, a spatial

3 database of landslide related factors were designed and constructed. Five factors were considered in calculating the probabilities. For calculation of the factors a Triangular Irregular Network (TIN) was constructed from the LIDAR of the region in 3D Analyst. The generated TIN is converted to raster format through interpolation and a DTM is constructed (Figure 2). Figure 2. DTM of the study region Figure 3. Curvature map of the study region By using the 3D Analyst, Raster surface extension and DTM, the curvature map (Figure 3), and plan curvature are created. Using the DTM one can also produce the slope and the aspect maps of the region. Figure 4 shows the slope map of the region. According to this map, the inclination angle of the region changes between 0 to 72 degrees. The slope angle affects the overall rate of movement downslope.. After acquisition of the landslide related factors, the historical landslide locations were digitised from the report by SGI (Viberg, 1982). The scarps of the landslides were confirmed by using the contour map of the region by expert opinion. From these analyses, 8 clay slides were mapped for the site (Figure 5). All slide locations were converted to 5 m grids and a total of 1783 cells for the landslide locations were generated for the analysis. Figure 4. Slope map of the study region Figure 5. Slide locations in the study region Having established the data, the FRM was applied to provide a correlation between the historical slide locations and the various influencing factors under consideration (Lee, 2007). In this approach the percentage of the cells in a given parameter range are computed for both the landslide area and for the whole region. Then the FR value for each parameter range is computed as the ratio between these percentages for that range (Lee and TuDan., 2005).

4 FR values greater than 1 indicate higher correlation with the landslide occurences. FR values much less than 1 (eg. below 0.85) show low correlation. The relationship between landslide locations and slope are given in Table 1 (in the table, the highly correlated variables to landslide locations, i.e. FR>1, are highlighted in bold). The values in the table indicate that slope values between 10 0 ~ 15 0 have the highest probability of 1.88 and the slope values between 5 0 and 25 0 have higher correlation to landslide occurrence whereas the slope angle lower than 5 0 and larger than 25 0 do not show a distinct relationship to landslide occurrence. Tabel 1. Frequency ratio of steepness map (factor = SLOPE, slope inclination) Class Landslide Occurrence * Pixels in domain ** Ratio A, Number PLO=(A/B)*100 (%) C, Number PIF=(C/D)*100(%) PLO/PIF 0 0 ~ ~ ~ ~ ~ > *B=1783 (total number of landslides containing pixels) **D=53196 (total number of pixels in the study area) The analysis of aspect correlation to historical slide locations showed that the frequency of slides was most abundant on Northeast, North and Northwest facing slopes and the frequency of landslides was lowest on Southeast, South, and East facing slopes. Similar FR analyses (not shown here) indicated that the elevations at landslide locations lie mostly between 9m to 27m. Therefore, these ranges were considered in determining the intervals of elevation parameters. The elevation class between 15 m to 20 m has the highest probability of 2.82 occurrences of slide Application of First Order Second Moment (FOSM) Second approach to generate susceptibility map of the Lille Edet municipality was applied by integrating a probabilistic slope stability model into the GIS environment. The FOSM reliability approach was used to compute the probability of slope failure in the cells; these probabilities were then used to assess the susceptibility, (see, for example, Duzgun & Ozdemir, 2006). In this study, the soil s shear strength was used together with the slope map of the area in a FOSM approach. The limit state function, G, was defined as the difference between the resistance stress R f and the driving traction D f, as follows: G = R f - D f (1) In this study, the limit state function was formulated by using R f = s (2) D f = γ m H sin(α) (3) Where γ m = Unit weight of soil, α = Slope angle, s = Shear strength of soil, and H = Depth to sliding surface μ The reliability index (β), as defined by Cornell (1969) is: β = z Var(z) (4) The larger values of β indicate smaller occurrence probabilities of failure P f (Chen et. al., 2007). Applying FOSM to Eqns (1) - (3), one gets: μ z = s γ m H Sin (α) (5) Var(z) = σ s (6) The basic variable in this case is the shear strength because its uncertainty is much higher compared to other variables. The amount of uncertainty in shear strength is represented by its

5 variance ( σ ). As a result, the reliability index for this application takes the following form: β s γ 2 s * H * Sin( α) m = (7) σ s The shear strength profile for the site is plotted in Figure 6 and is discussed in detail in LessLoss Deliverable 24 (2007). The estimated undrained shear strength is described in the area close to the river with the following formula: c u (z) = 23+z z > 1 m (8) where c u = undrained shear strength [kpa], and z = depth under the ground surface [m] However, in the area further away from the river, the undrained shear strength is estimated to be described by: c u (z) = 27 1 m < z < 8 m (9) c u (z) = * z z> 8 m (10) Figure 6. Shear strength From the profiles in Figure 6 the standard deviation of the shear strength was calculated for 3 different z values, namely 5, 10, and 15m. The γ values were taken equal to 20 kn/m 3 in the calculations. Then for each cell of the study region, the values of β were computed by using Equation 7 in spatial the Analyst extension of ArcGIS. 4. SUSCEPTIBILITY MAPPING OF FRM & FORM For generating the susceptibility map of the FRM the calculated frequency ratios for each factor is considered. Each layer in the GIS is assigned the ratio that was calculated for that layer for each parameter range. Then after calculating the frequency ratio of each layer by the GIS software, the grids were overlain and the index for susceptibility mapping was calculated by summing up the indexes for each cell in each layer. The summed values were between 0.34 to 7.85 and the standard deviation was equal to After calculation of the total frequency ratios for each cell, the index was classified into 3 classes as low, medium, and high susceptibility Figure 8. In the Figure 9 the historical locations of landslide is also added and it demonstrates that the resulting map matches very well the landslide locations.

6 Figure 8. FRM Susceptibility Map Figure 9. High susceptible zones overlaid with slide locations and air photo For generating the susceptibility map of the FORM the reliability index computed for each cell is considered. For the creation of susceptibility map a β value of 1.88 was considered to be a threshold for the safety assessment of the slope (Düzgün, 2000). Hence regions with β values less than 1.88 are classified as unsafe regions, while safe regions were those with β values greater than or equal to Figure 10 illustrates the susceptibility map for Lille Edet based on a slide surface of depth 15 m. As seen in Figure 11, the FOSM shows clearly the boundaries of actual landslides. Figure 10. of FOSM Susceptibility map with 15m height Figure 11. Susceptible zone of Lilla Edet with 15m height and landslide locations overlayed 5. EVALUATION OF RESULTS To evaluate the performance of the susceptibility mapping methods, the method of the Relative Operating Curve (ROC) was used (Santiago, 2006) and the results are shown in Figure 13 and Tables 2. According to this method, the further the curve lies above the reference line, the more accurate is the method. The area under the curve for the FR method in Table 6 is 0.88 which is very satisfactory (it is quite close to the ideal value of 1.0). Also as seen from the table the asymptotic significance is less than 0.05, which means that using the probability model is much better than guessing. For the FOSM reliability method, the area under the curve in Table 7 is only 0.54 which is considerably lower than that of the FR method. Therefore, the FR method can be considered to be a more reliable mapping method for this site.

7 Figure 14. ROC Curve of FRM and FOSM Table 6. Area under the Curve Metod Area Std. Error Asymptotic Sig. Asymptotic 95% Confidence Interval Lower Bound FRM FOSM Upper Bound 6. SUMMARY AND CONCLUSIONS In this study in the LillaEdet municipality of Sweden, quick clay slides were analysed and mapped, using the landslide-occurrence factors, by the probability likelihood ratio method as well as by the first order second moment (FOSM) method. By the application of frequency ratio method (probability likelihood ratio) the relationship of landslide related factors and the landslide locations were analysed by GIS tools. In the FOSM the application of method a probabilistic slope stability model is integrated into GIS environment. The predictive ability of the models generated by FRM and FOSM models are compared by using relative operating characteristics (ROC) curve. As a result of these comparisons, the FRM model appears to be a better tool than the FOSM. 7. REFERENCE Chen J.C., Jan C.D., Lee M.H., Probabilistic analysis of landslide potential of an inclined uniform soil layer of infinite length: theorem. Environ Geol, Vol. 51, pp Cornell, C.A A probability-based structural code. Journal of the American Concrete Institute, 66: Düzgün H.S.B., Karpuz C.2006, GIS-Based Landslide Risk Assessment for Bandirma Harbor Evaluacion de Riesgo Deslizamiento Basado en SIG para el Puerto de Bandirma Düzgün H.S.B., A. Ozdemir, Landslide risk assessment and management by decision analytical procedure for Dereko y, Konya, Turkey. Nat Hazards, Vol 39, pp Lee S, Tu Dan N. 2005, Probabilistic landslide susceptibility mapping in the Lai Chau province of Vietnam:Focus on the relationship between tectonic fractures and landslides, Environ Geol. 48: Lee. S. and Pradhan B., Landslide hazard mapping at Selangor, Malaysia using frequency ratio and logistic regression models, Landslides, Vol. 4, pp Santiago B (2006) Validation and Evaluation of Predictive Models in Hazard Assessment and Risk Management, Natural Hazards 37: Swedish Geotechnical Institute (2006). Risk Mitigation for earthquakes and landslides integrated project-lidar data for slope stability analysis, LessLoss Deliverable 6. Viberg L., Kartering och klassificering av lerområdens stabilitetsforutsaattningar, Swedish Geotechnical Institute, Raport no:15.