Weights-of-evidence model applied to landslide susceptibility mapping in a tropical hilly area

Abstract

A study demonstrating the application of the weights-of-evidence model (a Bayesian probability model) to landslide susceptibility mapping using geographical remote sensing (GIS) in a tropical hilly area of Malaysia is presented. In the first stage, a landslide related spatial database was created. Seven landslide conditioning factors were considered for the susceptibility analysis. Using landslide location and a spatial database containing information such as topography, soil, lithology, land cover and lineament, the weights-ofevidence model was applied to calculate each relevant factor's rating for the Cameron Highlands area in Malaysia. The topographic database including information on slope angle, slope aspect, plan curvature and distance from drainage was developed from a digital elevation model (DEM); the lithology and the distance from the lineament were derived from the geological database; soil texture was derived from the soil database; land cover and normalized difference vegetation index (NDVI) information were extracted from Landsat Thematic Mapper (TM) satellite imagery. Tests of conditional independence were performed for the selection of landslide conditioning factors, allowing nine combinations in total. Finally, landslide susceptibility maps were constructed using the ratings of each landslide conditioning factor. The resultant susceptibility maps were validated using the receiver operating characteristics (ROCs) based area under curve (AUC) method. Landslide locations were used to validate the results of the landslide susceptibility map and the verification results showed 97% accuracy for model 5, which employed a combination of parameters. Plan curvature, distance from drainage, distance from lineament, lithology and land cover performed better than other combinations of landslide conditioning factors. © 2010 Taylor & Francis.