Landslide susceptibility mapping using a logistic regression approach: a case study in Colorado Springs, El Paso County, Colorado

Abstract

A landslide is defined as a mass-wasting process of soil, rock or debris that moves downslope (U.S. Geological Survey, 2004). Landslides can occur on any terrain if the geomorphic, geological, and hydrological conditions are right. Landslides can cause significant property damage and even fatalities. Located in the foothills of the Rocky Mountains, Colorado Springs has proved to be highly susceptible to landslides. In summer 2015, approximately 200 homes were destroyed due to natural hazards including landslides in the area (FEMA, 2015). Efforts to quantify accuracy of landslide susceptibility maps by failure mechanism in Colorado Springs have been limited, thus warranting this study. The goal of this study is to evaluate the effectiveness of a geology-slope logistic regression (logit) model used to create landslide susceptibility maps. The inputs of the geology-slope logit model include only more readily available geology and topography data. To assess the effectiveness of this geology-slope logit model, a borehole-slope logit model that includes cohesion, friction angle, unit weight, groundwater depths, and topography was also developed. The two models were applied to study landslide susceptibility in a six square mile area of southwestern Colorado Springs. The results of both models were compared and evaluated for their accuracy by using a Receiver Operating Characteristics (ROC) curve method. The landslide susceptibility maps are specific to failure mechanisms present in the study area including circular failures within the colluvium deposits, circular failures within the weathered shale and planar failures within the weathered shale. The landslide susceptibility maps were produced for each failure mechanism present in the study area. The results show that the geology-slope model yields an ROC ranging from 53% to 62%, and the borehole-slope model yields an ROC ranging from 52% to 79%. The borehole-slope model outperformed the geology-slope model for circular failures in the colluvium deposits and planar failures within the weathered shale. The geology-slope model had a more successful ROC curve accuracy than the borehole-slope model for circular failures within the weathered shale. In addition, the results indicate that geology was the significant input in the geology-slope model, while soil strength parameters were the significant input for the borehole-slope model. In summary, the logit models developed in this study were used to create failure-mechanism-specific landslide susceptibility maps, which have not been done in previous studies for the Colorado Springs area. The landslide susceptibility maps produced in this study provide insights for prioritizing areas of additional field investigation. The geology-slope model can be easily and rapidly applied to other study areas since both geologic and topographic information are readily available.