Abstract:

LiDAR (Light Detection and Ranging) surveys of karst terrains provide high-resolution digital elevation models (DEMs) that are particularly useful for mapping sinkholes. In this study, we used automated processing tools within ArcGIS (v. 10.0) operating on a 1.0 m resolution LiDAR DEM in order to delineate sinkholes and closed depressions in the Boyce 7.5 minute quadrangle located in the northern Shenandoah Valley of Virginia. The results derived from the use of the automated tools were then compared with depressions manually delineated by a geologist. Manual delineation of closed depressions was conducted using a combination of 1.0 m DEM hillshade, slopeshade, aerial imagery, and Topographic Position Index (TPI) rasters. The most effective means of visualizing depressions in the GIS was using an overlay of the partially transparent TPI raster atop the slopeshade raster at 1.0 m resolution. Manually identified depressions were subsequently checked using aerial imagery to screen for false positives, and targeted ground-truthing was undertaken in the field. The automated tools that were utilized include the routines in ArcHydro Tools (v. 2.0) for prescreening, evaluating, and selecting sinks and depressions as well as thresholding, grouping, and assessing depressions from the TPI raster. Results showed that the automated delineation of sinks and depressions within the ArcHydro tools was highly dependent upon pre-conditioning of the DEM to produce “hydrologically correct” surface flow routes. Using stream vectors obtained from the National Hydrologic Dataset alone to condition the flow routing was not sufficient to produce a suitable drainage network, and numerous artificial depressions were generated where roads, railways, or other manmade structures acted as flow barriers in the elevation model. Additional conditioning of the DEM with drainage paths across these barriers was required prior to automated 2delineation of sinks and depressions. In regions where the DEM had been properly conditioned, the tools for automated delineation performed reasonably well as compared to the manually delineated depressions, but generally overestimated the number of depressions thus necessitating manual filtering of the final results. Results from the TPI thresholding analysis were not dependent on DEM pre-conditioning, but the ability to extract meaningful depressions depended on careful assessment of analysis scale and TPI thresholding.