Improved Automated Mapping of Sinkholes Using High-Resolution DEMs

Abstract

Sinkholes are common surface manifestations of the presence of networks of subsurface caverns in areas where the bedrock geology is dominated by soluble rocks such as limestones. Accurate mapping of sinkholes is crucial as they are hazardous to transportation infrastructure and may serve as conduits of contaminants to the groundwater. The use of high-resolution digital elevation models extracted from LiDAR and tools in ArcGIS have made it a simple task to automate the process of identification of closed depressions. However, these automated methods do not differentiate between sinkholes and other man-made depressions. Multivariate statistical methods such as linear discriminant analysis, quadratic discriminant analysis, and logistic regression were used to produce predictive models based on selected shape factor values such as circularity, sphericity, and curvature. Curvature values, especially when combined with circularity, were found to be the most powerful variables in separating closed depressions into sinkholes and other artificial depressions.