Slope unit-based landslide susceptibility zonation

Landslide susceptibility zonation is essential for disaster management and control in mountainous regions. Most landslide susceptibility zonations up to now are pixel-based and somehow are impracticable in landslide hazard management. In this paper, we propose a procedure for slope unit-based landslide susceptibility zonation with a case study of Shenzhen, China. First, the flat terrain is removed by slope classification, and then slope units are derived through watershed segmentation of mean curvature. The impact factors of a slope unit are assigned the mean or majority values of the factors of all pixels within that unit, respectively. The slope units containing the existing landslides are picked as positive training examples. Applying the one-class Support Vector Machine (SVM) under a 20% holdout cross validation strategy, we successfully predict slope units as safe or landslide-prone. Compared with a pixel-based method, the slope unit-based method significantly decreases the computing costs and predicts reasonable landslide-prone areas without obvious growth of omission error.