Landslide susceptibility assessment by bivariate methods at large scales: Application to a complex mountainous environment

Statistical assessment of landslide susceptibility has become a major topic of research in the last decade. Most progress has been accomplished on producing susceptibility maps at meso-scales (1:50,000–1:25,000). At 1:10,000 scale, which is the scale of production of most regulatory landslide hazard and risk maps in Europe, few tests on the performance of these methods have been performed. This paper presents a procedure to identify the best variables for landslide susceptibility assessment through a bivariate technique (weights of evidence, WOE) and discusses the best way to minimize conditional independence (CI) between the predictive variables. Indeed, violating CI can severely bias the simulated maps by over- or under-estimating landslide probabilities. The proposed strategy includes four steps: (i) identification of the best response variable (RV) to represent landslide events, (ii) identification of the best combination of predictive variables (PVs) and neo-predictive variables (nPVs) to increase the performance of the statistical model, (iii) evaluation of the performance of the simulations by appropriate tests, and (iv) evaluation of the statistical model by expert judgment. The study site is the north-facing hillslope of the Barcelonnette Basin (France), affected by several types of landslides and characterized by a complex morphology. Results indicate that bivariate methods are powerful to assess landslide susceptibility at 1:10,000 scale. However, the method is limited from a geomorphological viewpoint when RVs and PVs are complex or poorly informative. It is demonstrated that expert knowledge has still to be introduced in statistical models to produce reliable landslide susceptibility maps.