Gully erosion susceptibility mapping using artificial intelligence and statistical models

Abstract

In this article, the gully erosion susceptibility map (GESM) was developed for the Abdanan region, Ilam province, Iran using frequency ratio (FR), logistic regression (LR), an ensemble of radial basis function (RBF) and imperialist competitive algorithm (ICA) models. At first, 74 gully erosion locations have been recognized and then 52 (70%) cases had been regarded for training, while 22 (30%) cases validate the models. To model the GESM, 12 affecting factors on gully erosion were identified and prepared in a GIS environment, including altitude, slope angle, slope aspect, plan curvature, topographical wetness index (TWI), normalized differential vegetation index (NDVI), soil type, rainfall, distance to river, distance to road, lithology and land use in the study area. According to the results of LR model, slope aspect factor (2.439) is the most important and NDVI factor (–1.462) has the least importance on the occurrence of gully erosion. Also, the results of FR model show that MuPlaj class (FR = 2.5) of lithology factor is the most important in gully erosion. The ROC–AUC has been applied to confirm the accuracy of the models. The ROC–AUC results show efficiencies of 89.9%, 88.3% and 85.1% for RBF–ICA, LR and FR models, respectively.