Landslides can be directly associated with the topographic characteristics of an area. Topography is
encoded in gridded Digital Elevation Models (DEMs) and derived Land‐Surface Parameters (LSPs)
such as curvature and slope. So far, mainly continuous terrain data were integrated in landslide
mapping processes. Recently, the use of terrain objects, i.e. discrete surface patches, has been
proposed, especially due to the steadily increasing spatial resolutions of DEMs.
The main aim of this study is to create terrain objects that are relevant for digital landslide mapping
in an object‐based image analysis (OBIA) framework. In OBIA terrain objects are produced by
applying image segmentation algorithms to LSPs. In addition to curvatures and slope, more complex
LSPs such as topographic openness and sky‐view factor are explored. The algorithm of choice is
multiresolution segmentation, a region‐growing technique that merges pixels to objects based on
the similarity in values. Various LSP stacks are segmented to produce homogenous terrain objects
for the study area in Taiwan. By changing the algorithmic parameters different scales of terrain
objects – ranging from hillslope to sub‐landslide size – are created for the same LSP layer stack.
Multiresolution segmentation is optimised statistically through multi‐scale analysis of local variance.
The landslide relevance of the terrain object scales is visually checked against orthophotos and DEM
visualisations, and is quantified by correlation with manually delineated landslide polygons.
The study gives clues about which parameters and LSPs are optimal for the multiresolution
segmentation algorithm in order to produce terrain objects that are most relevant to landslides. The
experiences made in this study are important for the development of an operational landslide
mapping system in OBIA, which integrates different types of remote sensing data (optical, SAR,
2013 GRSG AGM - Status and Developments in Geological Remote Sensing