Landslide susceptibility modeling based on remote sensing data and data mining techniques

The main purpose of this study is to apply three advanced landslide susceptibility models (Fisher’s linear discriminant analysis (FLDA), forest by penalizing attributes (FPA), and functional tree (FT)) to evaluate the landslide susceptibility in Muchuan County, China. Firstly, 12 landslide conditioning factors were prepared based on the local geo-environmental characteristics and relevant studies, including slope angle, slope aspect, elevation, profile curvature, plan curvature, topographic wetness index (TWI), land use, normalized difference vegetation index (NDVI), lithology, soil, distance to roads, and distance to rivers. Secondly, a landslide inventory map consisting of 279 landslides was randomly divided into training and validation datasets with the ratio of 70/30. Then, a certainty factors (CF) model was used to analyze the relationship between landslides and the classes of landslide conditioning factors, and a Chi-Squared statistic was used for contribution analysis. Finally, the receiver operating characteristic (ROC) curve (AUC-value) and some statistical parameters are used to compare and verify the three landslide susceptibility models. After comparison and validation, the FPA model should be used as the best model for landslide susceptibility evaluation in Muchuan County in this study. This study can provide valuable information for local governments or organizations to study slope stability. At the same time, the landslide susceptibility map obtained can be a tool for reference for infrastructure planning, engineering design, and disaster mitigation design in the study area.