Spatial validation of statistical models of landslide susceptibility.
Résumé
Evaluation of the quality of landslide susceptibility assessments with statistical models is very often performed using either time or spatial partitioning procedures of the dataset. Less attention has been focused on the application of the weights or probabilities of landslide susceptibility from one independent site to another independent site with the same environmental characteristics (predictive variables) and the same landslide types (observed variable). As the calibration and validation steps are essential points in susceptibility analyses, the aim of this work is to estimate the robustness of a Bayesian inference statistical model by applying the best set of predictive variables identified on a first area to two distinct areas without any additional calibration procedure, and to quantify the degree of model fit. The model is applied to three alpine catchments of the South French Alps developed in black marls.