30 - Spatial Modeling of Gully Erosion: A New Ensemble of CART and GLM Data-Mining Algorithms

Abstract

Gully erosion susceptibility mapping is a vital tool for natural resources management and planning development activities in semiarid environments. The purpose of the present study is to model gully erosion susceptibility, using an ensemble of CART (classification and regression tree)–GLM (general linear model) models within a geographical information system (GIS) and R statistical environment. Gully erosion locations were identified and a gully inventory map was developed using intense field surveys. In total, 174 gully locations were identified, and locations were divided into two categories; training (70%) and validating (30%), respectively. Twelve layers of gully erosion conditioning factors including, slope gradient, slope aspect, slope length, altitude, plan curvature, topographical wetness index, soil types, distance from river, distance from road, distance from lineament, drainage density, and land use and land cover were selected for modeling aims. Finally, the receiver operating characteristic curves for CART–GLM ensemble and alone models such as CART and GLM were constructed and the area under the curve (AUC) was computed for verification and accuracy purposes. The results revealed that AUC values are 0.789 (78.9%), 0.798 (79.8%), and 0.723 (72.3%) for CART–GLM, CART, and GLM models, respectively. Therefore, CART is almost identical to CART–GLM ensemble and both are better than GLM. The final outcome maps can be useful for planning and policy makers in respect to agricultural activities and environmental protection.