Glacier Surface Segmentation Using Airborne Laser Scanning Point Cloud and Intensity Data

As glaciers are good indicators for the regional climate, most of them presently undergo dramatic changes due to climate change. Remote sensing techniques have been widely used to identify glacier surfaces and quantify their change in time. This paper introduces a new method for glacier surface segmentation using solely Airborne Laser Scanning data and outlines an object-based surface classification approach. The segmentation algorithm utilizes both, spatial (x,y,z) and brightness information (signal intensity) of the unstructured point cloud. The observation intensity is used to compute a value proportional to the surface property reflectance – the corrected intensity – by applying the laser range equation. The target classes ice, firn, snow and surface irregularities (mainly crevasses) show a good separability in terms of geometry and reflectance. Region growing is used to divide the point cloud into homogeneous areas. Seed points are selected by variation of corrected intensity in a local neighborhood, i.e. growing starts in regions with lowest variation. Most important features for growing are (i) the local predominant corrected intensity (i.e. the mode) and (ii) the local surface normal. Homogeneity is defined by a maximum deviation of ±5% to the reflectance feature of the segment starting seed point and by a maximum angle of 20° between surface normals of current seed and candidate point. Two-dimensional alpha shapes are used to derive the boundary of each segment. Building and cleaning of segment polygons is performed in the Geographic Information System GRASS. To force spatially near polygons to become neighbors in sense of GIS topology, i.e. share a common boundary, small gaps (<2 m) between polygons are closed. An object-based classification approach is applied to the segments using a rule-based, supervised classification. With the application of the obtained intensity class limits, for ice <49% (of maximum observed reflectance), firn 49-74% and snow ≥74%, the glacier surface classification reaches an overall accuracy of 91%.