Object-Based Point Cloud Analysis for Landslide and Erosion Monitoring

Abstract

Today point clouds from close-range sensing are used for operational erosion and landslide monitoring. Distances between points from multi-temporal acquisitions can indicate surface deformation, while a designation of the underlying geomorphological processes is often handicapped by complex terrain structures and vegetation. We present an approach to landslide monitoring that integrates semantic information and three-dimensional deformation detection automatically. Surface changes are assigned to (i) semantic object classes (landslide scarp, eroded area, deposit) and (ii) to spatially contiguous, individual objects (like parts of the landslide scarp and moving clods of turf and soil). We demonstrate this object-based approach with a time series of 13 topographic Light Detection and Ranging point clouds, covering a site affected by shallow landsliding. The results of this case study illustrate how the presented methods translate the unstructured point clouds into information on geomorphological process dynamics to support erosion and landslide assessment.