Probabilistic sinkhole modelling for hazard assessment

A method for quantitatively assessing sinkhole susceptibility (spatial probability) and hazard (spatio‐temporal probability) has been developed and independently tested in a 50 km² sector of the Ebro Valley evaporite karst. Three genetic types of sinkholes have been mapped in the floodplain and a terrace surface: 947 small cover‐collapse sinkholes (type 1, terrace), large collapse sinkholes (type 2, floodplain) and large subsidence depressions (type 3, floodplain). The type 1 sinkhole inventory includes two temporal populations: 447 sinkholes formed before 24 November 2005, and 500 between that date and 2 November 2006. Sinkhole susceptibility models have been elaborated analysing the statistical relationships between the sinkholes of the 2005 inventory and a set of potential conditioning factors. The independent evaluation (validation) of the susceptibility models by means of several strategies (random, sequentially excluded, and temporal) has allowed us to select the most significant variables for each sinkhole type and assess quantitatively the quality of models; which are reasonable for the three sinkhole types. Validation has also provided information on the contribution of specific variables and the effect of changing their accuracy to the prediction capability of models. Susceptibility models for type 3 sinkholes have been validated satisfactorily with the 2006 sinkhole inventory (temporal validation). The best susceptibility model has been transformed into a hazard map considering the frequency of sinkholes that occurred in each susceptibility class between 2005 and 2006, as well as their average size. The susceptibility and hazard models obtained could be used as an objective basis for the application of mitigation measures, either of preventive or corrective nature. Copyright © 2008 John Wiley & Sons, Ltd.