Debris flow susceptibility mapping for initiation areas at medium scale: a case study in Western Norway

In recent years, rapid mass movements such as debris flow and debris avalanches resulted in a significant impact on Norwegian society and economy. The need for dispelling the uncertainty inherent in landslide risk assessment has encouraged the development of hazard and susceptibility maps. Different statistically-based modelling methods, in combination with geographic information systems (GIS), have been extensively used to ascertain landslide susceptibility in quantitative terms. This thesis proposes a bivariate statistical method (Weights of Evidence) for assessing the spatial proneness of debris flows within Førde and Jølster municipalities (Western Norway), where emphasis is put on the critical conditions of initiation. Since no feasible landslide database could be exploited for susceptibility mapping at medium scale, this thesis addressed the realisation of a new inventory. By coupling pre-existing data from remote sensing and field observations, circa 1100 debris flow initiation areas were outlined and differentiated in four categories with geomorphological repeatable features. Simple topography-based parameters such as slope, upslope contributing area, curvature and roughness were used to find significant statistical differences between the initiation areatypes. Moreover, they were employed together with other thematic maps as informative layers for landslide modelling. In order to test the model fitting performance, the ROC curves method is used in this thesis. The evaluation of different discretization schemes and combinations of the above-mentioned variables led to individuate models with different performances in terms of success rates. The best model is obtained by using only a combination of slope, flow accumulation and elevation (82% true positive rate), while the manual adjustment of the classification scheme did not lead to significant improvements.