Earth-flow susceptibility assessment in the Marvello River basin (Sicily, Italy)

In this study, statistical models of earth-flow susceptibility were prepared using logistic regression. The analyses were carried out in a small (51 km2) basin of western Sicily, where 1,376 earth-flows were identified. To predict the spatial distribution of the mapped landslides, outcropping lithology and seven topographic attributes were exploited as explanatory variables. Before calculating these variables, a reconstruction of the pre-failure topography was performed. To evaluate the predictive skill and the robustness of the models, two groups made of five random subsets of earth-flows and stable cells were prepared. Absences of the first group were selected as individual cells whereas those of the second group where picked up using circular areas randomly distributed within the stable portions of the basin. The largely satisfactory accuracy and stability of the models demonstrated the reliability of the reconstructed pre-landslide topography. Furthermore, models prepared with the circular groups of absences showed better predictive skills.