Using ALSM to map sinkholes in the urbanized covered karst of Pinellas County, Florida—2. Accuracy statistics

Abstract

We apply the logic of clinical epidemiological studies to quantify the accuracy of mapping sinkholes by ALSM in the 750 km² Pinellas County. By such studies, a new diagnostic procedure is tested by comparing the diagnoses in a clinical trial to diagnoses on the same patients from a more reliable, but more elaborate and expensive procedure (“gold standard” in epidemiological context). A relatively undeveloped, 65 km² focus area where we have aerial photographs that are effectively contemporaneous with the ALSM flights serves as the “clinical trial”. The xy-locations in the focus area are the “patients” in the trial. The “diagnostic test” for having “sinkhole disease” is inclusion in a database of sinkhole polygons delimited by ALSM contours (“ALSM-alone”), as detailed in Part 1. The standard of comparison (“gold standard” would be an overstatement in the absence of geophysical testing) is inclusion in a database of sinkhole outlines derived by best judgment of conjunctive interpretation of ALSM and aerial photography. GIS intersections that indicate the sensitivity and specificity of the test (ALSM-alone) are 43 and 98.3%, respectively, and, in the focus area where the prevalence of “sinkhole disease” is 4.7%, the positive and negative predictive values are 55.5 and 97.2%, respectively. Over much of the rest of the county, where only the test can be applied, the prevalence of sinkholes is sufficiently small that it cannot be determined to be any different from zero given the paucity of interpreted sinkholes (positive diagnoses) and the low specificity of the test method. The conclusion, therefore, is that contemporaneous aerial photography is essential to compile an ALSM-derived database that aims to state that the given xy-points lay inside or outside of topographic depressions in the covered karst of west-central Florida.