Landslide Susceptibility Analysis and Mapping Using Statistical Multivariate Techniques: Pahuatlan, Puebla, Mexico

Susceptibility analyses are frequently based on the idea that landslides occur in the same areas where they have taken place previously, and also in areas under similar conditions. Based on that assumption, four different statistical techniques-Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Logistic Regression (LRA), and Neural Networks (NN)-have been applied for the municipality of Pahuatlan, Puebla, Mexico. The base for the analysis was a geomorphological landslide inventory derived from the stereo-interpretation of Very High Resolution (VHR) satellite images. The quality of each model was controlled by using ROC curves and Cohen's Kappa coefficient. Also, a temporal validation with a data set of landslides occurred on 2012 was carried out for each model. The resulting analysis showed that the aspect, the slope angle and the lithological unit were the variables with the highest weight associated with the occurrence of landslides in the study area.