Efficiency of local minima and GLM techniques in sinkhole extraction from a LiDAR-based terrain model

Abstract

The aim of this paper was to study reliable automated delineation possibilities of karst sinkholes using a LiDAR-based digital terrain model (DTM) with pixel-based classifications. We applied two approaches to extract sinkholes: (1) general linear modeling (GLM) with morphometric indices derived from DTM; (2) and a local minima-based delineation using only LiDAR DTM as the input layer. The outcome of the local minima was significantly different from the reference ones but found all the sinkholes without previous knowledge of the area. The GLM-based outcome did not differ statistically from the reference. Results showed that these approaches were efficient in detecting sinkholes based on LIDAR derivatives, and can be used for risk assessment and hazard preparedness in karst areas: GLM had an overall accuracy of 89.5% and local minima had an accuracy of 92.3%; both methods identified sinkholes but also had commission errors, identifying depressions as sinkholes.

Keywords

• karst mapping
• sinkhole identification
• general linear model
• statistical evaluation
• sink fill