Application of GIS-based bivariate statistic for prediction of landslide susceptibility mapping in Lindu District, Indonesia

Abstract

This study utilized Geographic Information Systems (GIS) and bivariate statistical models to delineate landslide susceptibility in Lindu District, Indonesia. The results of Google Earth image interpretation and expert validation identified around 391 landslide locations and randomly classified into training (70%) and validation (30%) datasets. Fifteen landslide conditioning factors: elevation, slope, aspect, curvature, plan curvature, profile curvature, stream power index, topographic wetness index, road, river, fault, land use, normalized difference vegetation index, lithology and precipitation are combined with landslide training to obtain each factor weight and factor class from the Weight of Evidence (WoE) and Informative Value (IV) models. Both models were then validated using the area under curve (AUC). The AUC model accuracy results show that the success rates of the WoE and IV models are 81.93% and 80.36%, and the prediction rates are 80.83% and 77.35%, respectively. This study aids local governments in landslide risk mitigation planning.