Loading...
Using logistic regression to generate and compare six landslide inventory-based susceptibility maps in El Paso County, Colorado
Killen, Ashton A.
Killen, Ashton A.
Citations
Altmetric:
Editor
Date
Date Issued
2023
Date Submitted
Keywords
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
Landslides in Colorado cause millions of dollars of damage, destroy homes and infrastructure, and cause loss of life. Within the state, the town of Colorado Springs and surrounding El Paso County has been identified as an area that needs ongoing attention due to the severity of landslide risk. In Colorado Springs, some of the highest risk areas are expensive houses in the foothills and on mesas. Most of these areas are either already mapped as landslide deposits or are located on top of the Pierre Shale, a unit that is highly susceptible to landslides. Most research into the failure mechanisms of landslides in Colorado Springs has been focused directly on the geology, slope angle, and geotechnical data from boreholes. This study uses a landslide database of 561 events in the area to improve on previous predictive methods, and also expands the list of potential influencing factors by including geology, slope angle, aspect, terrain roughness, curvature, plan curvature, elevation, and topographic wetness index. The parameters were placed into a binary logistic regression model to determine their significance and generate models. The most significant parameters were curvature, elevation, slope, topographic wetness index, and geology, with an area under curve (AUC) of 0.9535 (Model 1). Additionally, five more models were built from subsets of these significant parameters: slope and geology (AUC 0.9249, Model 2), curvature, elevation, and slope (AUC 0.9321, Model 3), curvature, elevation, and topographic wetness index (TWI) (0.9468, Model 4), slope and elevation (AUC 0.9293, Model 5), and TWI and geology (AUC 0.9095, Model 6). Landslide susceptibility maps were generated from each of these models, with five classifications, low, low-moderate, moderate, moderate-high, and high. The models that generated the best map were Model 2 and Model 5. Model 5 is recommended for use because it uses parameters never used before and only uses two simple parameters to generate the susceptibility map.
Associated Publications
Rights
Copyright of the original work is retained by the author.