Model-based assessment of shallow landslides susceptibility at a petrochemical site in Brazil

Abstract

The Serra do Mar mountain range is the main site of shallow-landslides occurrence in Brazil. The application of physically-based models is an effective method to predict landslide susceptibility, which is of great importance in hazard assessments and urban-planning studies. Thus, the objective of this study is to compare landslide susceptibility scenarios created with SHALSTAB and SINMAP at two large watersheds (Mogi and Perequê) in Cubatão, Latin America’s largest petrochemical site. Model calibration is based on the landslide scars inventory of the 1985 event, the geotechnical parameters derived from soil samples and the topography sourced from a 5 m resolution DEM. Using the Receiver Operating Characteristics (ROC) analysis to assess model performance, SHALSTAB emerges as the best-fit model for both watersheds due to higher concentration of landslide scars in unstable areas (true positive results) and higher global accuracy. Even though SINMAP had similar degree of success, it was slightly less accurate and failed more often in the identification of potentially unstable areas. Comparative performance studies of physically-based models are fundamental to support effective and reliable hazard assessments in mountain regions, providing an outlook in how to proceed with more detailed studies.