Automated delineation of karst sinkholes from LiDAR-derived digital elevation models

Abstract Sinkhole mapping is critical for understanding hydrological processes and mitigating geological hazards in karst landscapes. Current methods for identifying sinkholes are primarily based on visual interpretation of low-resolution topographic maps and aerial photographs with subsequent field verification, which is labor-intensive and time-consuming. The increasing availability of high-resolution LiDAR-derived digital elevation data allows for an entirely new level of detailed delineation and analyses of small-scale geomorphologic features and landscape structures at fine scales. In this paper, we present a localized contour tree method for automated extraction of sinkholes in karst landscapes. One significant advantage of our automated approach for sinkhole extraction is that it may reduce inconsistencies and alleviate repeatability concerns associated with visual interpretation methods. In addition, the proposed method has contributed to improving the sinkhole inventory in several ways: (1) detection of non-inventoried sinkholes; (2) identification of previously inventoried sinkholes that have been filled; (3) delineation of sinkhole boundaries; and (4) characterization of sinkhole morphometric properties. We applied the method to Fillmore County in southeastern Minnesota, USA, and identified three times as many sinkholes as the existing database for the same area. The results suggest that previous visual interpretation method might significantly underestimate the number of potential sinkholes in the region. Our method holds great potential for creating and updating sinkhole inventory databases at a regional scale in a timely manner.