Abstract. The morphology of landslides is influenced by the slide/flow of the material downslope. Usually, the distance of the movement of the material is greater than the width of the displaced material (especially for flows, but also the majority of slides); the resulting landslides have a greater length than width. In some specific geomorphologic environments (monoclinic regions, with cuesta landforms type) or as is the case for some types of landslides (translational slides, bank failures, complex landslides), for the majority of landslides, the distance of the movement of the displaced material can be smaller than its width; thus the landslides have a smaller length than width. When working with landslide inventories containing both types of landslides presented above, the analysis of the length and width of the landslides computed using usual geographic information system techniques (like bounding boxes) can be flawed. To overcome this flaw, I present an algorithm which uses both the geometry of the landslide polygon minimum oriented bounding box and a digital elevation model of the landslide topography for identifying the long vs. wide landslides. I tested the proposed algorithm for a landslide inventory which covers 131.1 km2 of the Moldavian Plateau, eastern Romania. This inventory contains 1327 landslides, of which 518 were manually classified as long and 809 as wide. In a first step, the difference in elevation of the length and width of the minimum oriented bounding box is used to separate long landslides from wide landslides (long landslides having the greatest elevation difference along the length of the bounding box). In a second step, the long landslides are checked as to whether their length is greater than the length of flow downslope (estimated with a flow-routing algorithm), in which case the landslide is classified as wide. By using this approach, the area under the Receiver Operating Characteristic curve value for the classification of the long vs. wide landslides is 87.8 %. An intensive review of the misclassified cases and the challenges of the proposed algorithm is made, and discussions are included about the prospects of improving the approach with further steps, to reduce the number of misclassifications.
Read full abstract