Abstract
Sinkholes are the major cause of concern in Florida for their direct role on aquifer vulnerability and potential loss of lives and property. Mapping sinkhole susceptibility is critical to mitigating these consequences by adopting strategic changes to land use practices. We compared the analytical hierarchy process (AHP) based and logistic regression (LR) based approaches to map the areas prone to sinkhole activity in Marion County, Florida by using long-term sinkhole incident report dataset. For this study, the LR based model was more accurate with an area under the receiver operating characteristic (ROC) curve of 0.8 compared to 0.73 with the AHP based model. Both models performed better when an independent future sinkhole dataset was used for validation. The LR based approach showed a low presence of sinkholes in the very low susceptibility class and low absence of sinkholes in the very high susceptibility class. However, the AHP based model detected sinkhole presence by allocating more area to the high and very high susceptibility classes. For instance, areas susceptible to very high and high sinkhole incidents covered almost 43.4% of the total area under the AHP based approach, whereas the LR based approach allocated 20.7% of the total area to high and very high susceptibility classes. Of the predisposing factors studied, the LR method revealed that closeness to topographic depression was the most important factor for sinkhole susceptibility. Both models classified Ocala city, a populous city of the study area, as being very vulnerable to sinkhole hazard. Using a common test case scenario, this study discusses the applicability and potential limitations of these sinkhole susceptibility mapping approaches in central Florida.