APPLYING DISCRIMINANT ANALYSIS TOWARDS AUTOMATED SINKHOLE MAPPING METHODS

Sinkholes are common geomorphologic features in karst topography resulting from dissolution of soluble rocks such as limestones. High resolution LiDAR-derived DEMs allow simple visual recognition of sinkholes, but due to their large number, manual mapping can be extremely time consuming. The purpose of this research is to develop an automated extraction method for sinkholes from an airborne LiDAR-derived digital elevation model (DEM) for the Shenandoah Valley of Virginia. The DEM has to initially be pre-processed by lowering the elevation of transit routes. The core methodology to map depressions from DEMs is to use the Fill tool in ArcGIS to artificially fill depressions and subtract the filled DEM from the original. The resulting DEM will show all depressions, converted into polygons, that do not necessarily represent sinkholes. Three steps are followed to identify sinkholes based on the criteria of being circular depressions. Step one separates conjoined sinkholes, which initially may be represented as one noncircular polygon, into several distinctive polygons that retain their primary circular geometry. Step two calculates the eccentricity, circularity, and sphericity of each polygon to be used as criteria to identify circular depressions. Step three uses the Fisher discriminant function (F), based on discriminant analysis, to combine the three variables. The dividing point (C) was found to be ~1.99 such that values F < C represent sinkholes. The results indicate that sinkholes can be identified with 82.4% accuracy. To expedite the workflow in ArcGIS, Python script tools were created to extract depressions and identify sinkholes.