Spatial Prediction of Landslide Susceptibility Using Random Forest Algorithm

Abstract

Intelligent data analytics approaches are popular in landslide susceptibility mapping. This chapter develops a random forest (RF) approach for spatial modeling of landslide susceptibility. A total number of 78 landslide locations are identified using field survey, 55 of which are randomly selected to model landslide susceptibility and remaining 23 locations considered for model validation. Twelve predictor variables are selected: elevation, slope percentage, slope aspect, plan curvature, profile curvature, distance from roads, distance from streams, distance from faults, lithological formations, land use, soil type, and topographic wetness index (TWI) to create an RF model for landslide susceptibility mapping. The results of RF model are evaluated using efficiency (E), true positive rate (TPR), false positive rate (FPR), true skill statistic (TSS), and area under receiver operating characteristic curve (AUC) in training and validation steps. RF model registered excellent goodness-of-fit with AUC = 93.6%, E = 0.887, TSS = 0.776, TPR = 0.905, FPR = 0.129, and predictive performance with AUC = 90.7%, E = 0.777, TSS = 0.559, TPR = 0.809, FPR = 0.25. Intelligent data analytic method, therefore, has a significant promise in tackling challenges of landslide susceptibility mapping in large regions, which may not have sufficient geotechnical data to employ a physically based method.