Morphometric comparisons between automated and manual karst depression inventories in Apalachicola National Forest, Florida, and Mammoth Cave National Park, Kentucky, USA

Karst depression catalogs are critical to assessing the hydrology and geohazards of an area; yet, the delineation of these features within a landscape can be a difficult, time-consuming and subjective task. This study evaluates the efficacy of karst depression inventorying using an automated fill-difference method operating on high-resolution lidar-derived digital elevation models (DEMs). The resulting catalog is compared with existing karst depression inventories for two low-development areas of the USA, Mammoth Cave National Park (MACA) and Apalachicola National Forest (ANF), where karst depressions have been mapped previously using a manual closed-contour approach. The automated fill method captures 93 and 85 % of these previously mapped karst depressions at MACA and ANF, respectively. Field observations and topographic analysis suggest that the omitted features were likely misclassified within the existing catalogs. The automated routine returns 797 and 3377 additional topographic depressions, at MACA and ANF, respectively, which are not included in the existing catalogs. While the geology of ANF is mostly homogenous Quaternary deposits, the newly identified, typically smaller-scale depressions found within MACA tend to be disproportionally located in non-carbonate-dominated formations, where the development of karst may be restricted by geologic heterogeneity. Within both areas, the size distributions of the two inventories are statistically identical for features larger than ~103 m2 in area or ~3 m in depth. For individual depressions captured by both methods at MACA, the automated fill-difference routine tends to return a slightly larger estimate of depression size and aggregate small depressions into larger ones. Conversely, at ANF, some low-relief depressions may be disaggregated by the fill-difference technique, with a trend toward smaller estimated depression areas when the automated method is employed. The automated fill-difference method, operating on high-resolution lidar-derived DEMs, can reproduc