Abstract
The ultimate goal of most statistically-based landslide susceptibility studies is to portray areas that a more or less prone to future slope instability. A number of strategies have been presented to sample the position of the underlying training observations (i.e. landslide locations) within raster-based landslide susceptibility models. Previous research emphasized that the location of past landsliding should preferably be represented by variables that describe the conditions before landslide occurrence (pre-landslide conditions). The assumption behind is that an in-depth description of post-landslide conditions might hamper the identification of susceptible terrain that did not fail yet, but is likely to be affected by future landsliding. In practice, however, data on pre-landslide conditions is rarely available. This contribution outlines the main outcomes of recently published research [cf. 1]. The aim was to elaborate differences between landslide susceptibility models based on post-landslide digital terrain models (DTMs) and their counterparts calibrated with approximated pre-landslide DTM derivatives. In this context, also the associated effects of raster resolution and landslide size were considered. The pairwise model confrontations (post- vs. pre-landslide) showed that the DTM raster resolution and the size of the geomorphic phenomena of interest (i.e. smaller vs. larger landslides) controlled whether and how the modelling results differed. The experiments indicate that commonly available post-landslide DTMs can still reasonably be utilized to derive landslide susceptibility models in case they are resampled to a comparably low resolution (i.e. with respect to landslide size).