Prioritization of effective factors in the occurrence of land subsidence and its susceptibility mapping using an SVM model and their different kernel functions

This study attempted to map land subsidence susceptibility using a support vector machine (SVM) model and their different kernel functions in Kerman province, Iran. Initially, land subsidence locations were recognized using extensive field surveys and Google Earth images and, subsequently, a land subsidence distribution map was created in a GIS environment. Then, different effective factors in the occurrence of land subsidence in the study area including percentage slope, slope aspect, altitude, profile curvature, plan curvature, topographic wetness index (TWI), distance from river, lithological units, piezometric changes, land use and normalized difference vegetation index (NDVI) were selected as independent variables for the modeling process. Land subsidence susceptibility maps in the study area were produced using an SVM model and different kernel functions related to it such as linear, polynomial, sigmoid and radial basis functions. The results of model validation using 30% of the unused locations in the modeling process and receiver operating characteristic (ROC) showed that the maps of land subsidence susceptibility obtained from the SVM technique and kernel functions had the highest accuracy with AUC values of 0.894 to 0.857. According to the results of prioritization of effective factors, piezometric data (utilization of groundwater), NDVI and altitude were the most significant factors affecting the occurrence of land subsidence in Kerman province. Therefore, the results of spatial modeling of land subsidence and their susceptibility maps have a key role in the planning of land allocation and water resource management in the study area.