Predicting sinkholes by means of probabilistic models

A quantitative method for assessing sinkhole susceptibility and hazard has been developed and independently tested in an evaporite karst of NE Spain. Three genetic types of sinkholes have been mapped: 947 small cover-collapse sinkholes, 23 large collapse sinkholes and 24 large subsidence depressions. By analysing the statistical relationships between the sinkholes and the potential conditioning factors, susceptibility models have been developed. These were evaluated using several strategies that have allowed us to assess quantitatively the quality of the models. The best susceptibility model for the small collapse sinkholes was transformed into a hazard map by considering the frequency and average size of an independent population of new occurrences. The susceptibility and hazard models obtained indicate that it is possible to produce reasonably good predictions on the distribution and frequency of sinkholes in the study area.