Landslide Susceptibility Assessment Using Binary Logistic Regression in Northern Philippines

Abstract

In a landslide prone area in mountainous northern Philippines, landslide susceptibility using binary logistic regression was investigated. Landslide data were randomly divided into training and validation sets using 80% and 20% proportions, respectively. A detailed logistic regression procedure was applied and presented herein. Nine landslide conditioning factors were used. Based on the coefficients obtained, the most influential factors were NDVI followed by land use/land cover, slope aspect, lithology and slope angle. Distance to lineament, distance to road, plan and profile curvature showed no influence in the model generated. Training and validation accuracy were good, amounting to 91% and 86%, respectively. Using training data, 80%, 11%, 5%, 2% and 2% of the landslides were associated with the very high, high, moderate, low and very low susceptibility classes, respectively. Using validation data, the proportions were 82%, 10%, 5%, 3% and 0%, respectively. The strong influence of NDVI affirms its major role in modelling landslide susceptibility. It supports the strong potential of revegetation of precarious slopes in complementing ongoing structural slope stabilization and rehabilitation measures.