Development and validation of sinkhole susceptibility models in mantled karst settings. A case study from the Ebro valley evaporite karst (NE Spain)

• Published in: Engineering geology Vol. 99; no. 3; pp. 185 - 197
• Main Authors: Galve, J.P., Bonachea, J., Remondo, J., Gutiérrez, F., Guerrero, J., Lucha, P., Cendrero, A., Gutiérrez, M., Sánchez, J.A.
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Amsterdam Elsevier B.V 23.06.2008; Elsevier
• Subjects: Applied sciences; Buildings. Public works; Earth sciences; Earth, ocean, space; Ebro valley; Engineering and environment geology. Geothermics; Evaporite karst; Exact sciences and technology; Geotechnics; Natural hazards: prediction, damages, etc; Prediction modelling and validation; Sinkhole susceptibility; Soil mechanics. Rocks mechanics; Europe; Spain; Sinkhole susceptibility; Evaporite karst; Prediction modelling and validation; Ebro valley; Collapse; Cartography; Geology; Vulnerability; Risk analysis; Forecast model; Modeling; Case study; Karst; Database; Evaporites
• ISSN: 0013-7952; 1872-6917
• DOI: 10.1016/j.enggeo.2007.11.011

A preliminary sinkhole susceptibility analysis has been carried out in a stretch 50 km 2 in area of the Ebro valley alluvial evaporite karst (NE Spain). A spatial database consisting of a sinkhole layer and 27 thematic layers related to causal factors was constructed and implemented in a GIS. Three types of sinkholes were differentiated on the basis of their markedly different morphometry and geomorphic distribution: large subsidence depressions (24), large collapse sinkholes (23), and small cover-collapse sinkholes (447). The susceptibility models were produced analysing the statistical relationships between the mapped sinkholes and a set of conditioning factors using the Favourability Functions approach. The statistical analyses indicate that the best models are obtained with 6 conditioning factors out of the 27 available ones and that different factors and processes are involved in the generation of each type of sinkhole. The validation of two models by means of a random-split strategy shows that reasonably good predictions on the spatial distribution of future dolines may be produced with this approach; around 75% of the sinkholes of the validation sample occur on the 10% of the pixels with the highest susceptibility and about 45% of the area can be considered as safe.