Landslides occur every year in many areas of the world, causing casualties, economic and environmental losses.
Landslide inventory maps are important to document the extent of the landslide phenomena in a region, for risk
estimation and management, and to study landscape evolution. We present a method to facilitate the semi-automatic
recognition and mapping of event induced shallow landslides. The method is based on the combination in a Bayesian
framework of information extracted from High Resolution optical multispectral satellite images and Digital Elevation
Models (DEM). The landslide membership probability is estimated from post-event satellite images using a supervised
image classification method. The likelihood of landslide occurrence is obtained adopting a “data-driven” approach,
intersecting existing landslide inventories with maps of morphometric parameters (slope and curvature) calculated from
the DEM. We tested the method in the Huaguoshan basin, Taiwan, where it proved capable of detecting and mapping
landslides triggered by Typhoon Morakot in August 2009. Compared to other pixel-based approaches, the method
reduces significantly the typical “salt-and-pepper” effect of landslide classifications, and allows the internal classification
of landslide areas in landslide source areas and landslide travel and depositional (“run out”) areas.