Comparison of Ensemble Logistic Model Tree with Logistic Regression for Landslide Susceptibility Mapping

Abstract

Landslide susceptibility mapping is considered as the first important step in landslide risk assessments. This maps is a useful tool for land use planning. The main purpose of this study is to compare performance of ensemble data mining technique of logistic model tree (LMT) with logistic regression (LR) for landslide susceptibility modeling in the Sarkhoon watershed, Chaharmahal and Bakhtiari province. For this purpose, at first, a landslide inventory map including 98 landslides was constructed using history records, and extensive field surveys. In addition, a total of 100 non-landslide points were also identified to obtain the highest classification accuracy. The landslide and non-landslide points were randomly divided into two groups with a 70/30 ratio for modelling and validation sets. Twenty conditioning factor were selected based on literature review, geo-environmental properties and landslide occurrence mechanism in the study area. Subsequently, the LMT and LR data mining techniques were applied to identify the influence of conditioning factors on landslide occurrence of the training dataset and assess landslide susceptibility. Finally, the performance of applied models in landslide susceptibility mapping were investigated with the validation set using receiver operating characteristics (ROC) curve. The results showed that the LR model, with a validation AUC of 0.797, is the better technique for landslide susceptibility mapping than the LMT (AUC = 0.740). Therefore, both models are reliable tools for spatial prediction of landslide susceptibility. Also, the results of LR model illustrated that among the conditioning factors, the distance to road, a weight of -0.808 has the highest importance.