Novel parameters for sinkhole mapping in the anthropocene: Example from a karst-dominated landscape of the Missouri Ozarks, USA

Understanding the spatial distribution of sinkholes in karst terrains is essential for evaluating groundwater resources, hydrological processes, and mitigating hazards such as land subsidence. Due to the rapidly evolving dynamic nature of sinkholes, automating sinkhole mapping is essential for maintaining accurate inventories. Additionally, human activities modify natural depressions and landforms, which complicates the task of distinguishing sinkholes from other types of depressions. This study introduces four novel morphometric parameters—berm-drop, width-depth ratio, maximum flow accumulation, and flow accumulation-to-area ratio—to improve automated sinkhole mapping, particularly in the Anthropocene, where human impacts on the landscape are pervasive. These four new parameters are integrated with eleven literature-based morphometric parameters into a Random Forest Model (RFM) to automate sinkhole mapping in the karst-dominated landscape of the Mark Twain National Forest, located in the Ozark Plateau of southeast Missouri, USA. The results demonstrate that the RFM offers great ability—about 95% overall accuracy—to distinguish sinkholes and non-sinkholes particularly human-modified depressions, such ponds. The berm-drop parameter plays a notable role in this distinction. This capability is particularly valuable in the study area, which is dominated by agricultural activities featuring a large number of ponds. The RFM identified over 4000 sinkholes across the entire study area, far surpassing the existing inventory of just 271 sinkholes maintained by the Missouri Department of Natural Resources. The model was particularly effective in detecting sinkholes of moderate to larger depths (depth ≥0.6 m), about 99% accurate in identifying sinkholes. Further, the RFM identified sinkholes of all sizes (area) accurately. This is particularly important, as larger and deeper sinkholes present heightened risks, including potential for substantial property damage and greater subsidence hazards. Thus, this research underscores the value of integrating these novel parameters for more accurate, large-scale sinkhole mapping, which is vital for informed land-use planning, hazard mitigation, and groundwater protection in karst regions.