MAPPING LANDSLIDES AND ASSESSING LANDSLIDE SUSCEPTIBILITY IN THE RíO LA CARBONERA WATERSHED, ON THE SE FLANK OF PICO DE ORIZABA VOLCANO, MEXICO

We prepare a multi-temporal landslide inventory and use Geographic Information Systems (GIS) to create and compare landslide susceptibility maps. GIS has been used to map and assess landslide susceptibility heuristically, statistically, or deterministically at local or regional scale. In Mexico, a paucity of detailed landslide inventory maps and landslide susceptibility maps limits systematic comparison of landslide susceptibility models. The lack of systematic comparison compromises the reliability of the models and can lead to their abuse. We address this deficiency by evaluating four landslide models: 1) a landslide susceptibility map created by adaptation of the Landslide Hazard Zonation protocol of the Washington State Department of Natural Resources, Forest Practices Division; 2) the Stability Index MAPping (SINMAP) model; 3) and 4) landslide susceptibility maps created using, respectively, Multiple Logistic Regression (MLR) and Multicriteria Evaluation (MCE) models. The susceptibility maps are validated by using a contingency table and the area under the receiver operating characteristic (ROC) curve. We evaluate how well these models predict landslide location, and we identify the advantages and limitations of each model. We use the Río La Carbonera watershed as a case-study area on the southeastern flank of Pico de Orizaba (5,675 m a.s.l.). The watershed has seasonally high rainfall, rock types susceptible to landslides, a high degree of weathering, and steep slopes. Small landslides and debris flows of at least 10² m³ occur continually and frequently impact and damage human settlements and disrupt economic activity. Landslide distribution is ascertained through a landslide inventory map (235 landslides) and a related geo-dataset created from multi-temporal aerial photographs and field work. The results suggest that in the study area the MLR model has a stronger predictive capability than the MCE and SINMAP models. The landform model is a generalized method that cannot be assessed by ROC or contingency table techniques. The technique and its implementation are presented and discussed. This research is funded by Grant PAPIIT # IN102115 from the National Autonomous University of Mexico (UNAM).