Data requirements for the assessment of shallow landslide susceptibility using logistic regression
Shallow landslides are an abundant phenomenon in mountain regions. Since these processes often endanger settlements and infrastructure it is important to estimate their spatial occurrence. Hence, various modelling techniques for the area-wide assessment of shallow landslide susceptibility are applied (i.e. heuristic, statistically- and physically-based approaches). Amongst these, statistically-based approaches are based on the assumption that factors promoting landslides in the past will also facilitate landsliding in future. Therefore a shallow landslide inventory for the area of interest including sufficient observations for training and validation of the model as well as a high-quality digital terrain model are prerequisites. With the help of a multi-annual shallow landslide inventory and derivatives of two airborne laser scanning campaigns (i) the optimal spatial resolution of the digital terrain model, (ii) the ideal training-to-validation split and (iii) the minimal number of observed landslides required for the assessment of shallow landslide susceptibility using logistic regression are investigated. Predictors are based on the digital terrain models and comprise slope angle, aspect, minimum and maximum curvature, slope length and topographic wetness index. The objectives are discussed for three study areas in Vorarlberg, Austria. Results of the modelling experiments show best performances using a digital terrain model with a spatial resolution of 5 m and a training-to-validation split of 3:7. Regarding the inventory size at least 150 mapped landslides were necessary to achieve acceptable results. However, it is recommended that at least 400 observed landslide locations at a minimum landslide density of 3 landslides/km2 are considered for the statistically-based assessment of shallow landslide susceptibility.