Assessment of land subsidence susceptibility in Semnan plain (Iran): a comparison of support vector machine and weights of evidence data mining algorithms

Abstract

Land subsidence is a geo-hazard that leads to slow or rapid decrease in ground level. This can result in geological, environmental, hydrogeological, and economic impacts. Land subsidence has already occurred in more than 300 plains in Iran. Semnan plain is one of the most important areas undergoing this phenomenon. In general, miscellaneous methods have been employed around the world to assess land subsidence susceptibility. In this study, support vector machine and weights of evidence Bayesian theory were applied to assess land subsidence susceptibility. In the first step, the required information on the history of subsidence in the study area was provided. Locations of the land subsidence were specified by Landsat 8 satellite images and field surveys. Twelve conditioning factors from different basic layers including topography, geology, land use, and groundwater table were considered for modeling. Spatial correlation between land subsidence locations and effective factors was calculated using weights of evidence Bayesian theory. Land subsidence susceptibility maps were created using support vector machine and weights of evidence models. ROC curve, sensitivity, specificity, Cohen’s Kappa index, and fourfold cross-validation were employed to validate the obtained land subsidence susceptibility maps. In Semnan plain, AUC for the support vector machine and weights of evidence models was 0.748 and 0.726, respectively, demonstrating that the given models hold an acceptable accuracy for land subsidence susceptibility mapping; however, the accuracy of the support vector machine is higher than that of weights of evidence model. Results of this research can help policy makers as well as environmental and urban planners.