PREDICTION OF POTENTIAL AREAS OF SINKHOLE DEVELOPMENT IN SOUTHWESTERN INDIANA USING MULTIPLE REGRESSION ANALYSIS

Results of a multiple regression analysis, including prediction intervals, were used to develop a probability-based map of potential areas of sinkhole development in southwestern Indiana based on landscape-level variables as predictors. The dependent variable used in the analysis was the log of the sinkhole density in number of sinkholes/km², derived from an inventory of 154,925 sinkholes. The analysis was conducted for five large (8-digit scale Hydrologic Unit Code) watersheds in southwestern Indiana and northern Kentucky.

The study area was divided into regular units of analysis at a scale of a quarter section in the public land survey system (160 acres or 0.65 km²). For this analysis, 28,832 analysis cells were used. The conceptual model views karst development to be more probable near specific geologic materials or features, especially in humid-temperate climates. The final best-fit equation accounted for 67% of the observed variability in the log sinkhole density and contained 14 statistically significant (99% confidence level) independent variables; five were related to lithology (including one interaction term, which combined geology and soils), two were land-use classes, and seven were related to terrain. Influential variables in the conceptual model were limestone, soil parent materials derived from limestone dissolution, and soil moisture and topographic parameters that might describe preferential flow to or from sinkhole features. The spatial distribution of potential sinkhole-development risk was mapped using 99% confidence intervals for the predicted log sinkhole density values.

The resultant map of sinkhole-development risk shows a well-defined karst zone trending from northwest to southeast along the Mitchell Plateau. The “extremely high” risk category has a high density of existing karst features, covering 2,617 km² (14%) of the study area. The “high risk” category borders the extremely high category, covering 2,471 km² (13%). The “moderate risk” category (5,027 km², 27%) includes sinkholes developed in discontinuous limestone units as well as near faults where offsets have brought limestone closer to the surface. The “low risk” category covers 46% of the study area (8,554 km², 46%), and is dominated by noncarbonate bedrock geology, thick soil, or loess cover.