ANALYSIS OF LANDSLIDE HAZARD MAPPING METHODS: REGRESSION MODELS VERSUS WEIGHT RATING

摘要

Hazards due to landslides cause huge sufferings to human and substantial economic and environmental losses throughout the world. For the assessment of the landslide hazard, different methodological approaches are proposed. They are mainly grouped into two: qualitative and quantitative methods. The quantitative methods eliminate the problem of subjectivity of the qualitative methods. On the other hand, qualitative methods have simpler methodological procedures. Among the quantitative methods, statistical techniques are one of the most popular ones. This study aims at providing a comprehensive comparison of quantitative and qualitative landslide hazard assessment maps, as well as proposing a new approach to enhance the performance of qualitative methods. The conventional statistical analysis can only produce, average and global parameter estimates for the whole study region. However, there are local variability in the factors causing landslides due to spatial correlation; i.e. geographically nearby units will have similar characteristics, while the ones, which are farer apart will have less similar properties. Therefore, as regression methods based on ordinary least square estimation, they are lack in modelling the spatial correlation structure of the parameters. For this reason, in this study ordinary logistic regression (OLR) is extended by spatial regression (SR) techniques for landslide hazard assessment. Morover, a heuristic approach is developed for obtaining the susceptibility map based on assigning weights to landslide causing factors. In this approach a weight rating system is adapted, where weights were assigned to different causitive factors. In both of the quantitative and qualitative analyses, slope, aspect, curvature, which are derived from Digital Elevation Model (DEM), lineaments and vegetation index, which are extracted from Landsat Thematic Mapper (TM) satellite images, elevation which is obtained from the contour map are considered to be the landslide influencing factors. The proposed models are implemented to a case study area from More and Romsdal region of Norway. Hazard maps from SR.OLR and susceptibility map from heurtictic approach are prepared and compared. It is concluded that SR model have better explantion capability for landslide occurrences than OLR model and heuristic model.