Spatial prediction of landslide hazard using logistic regression and GIS

Gorsevski, P.V.; Gessler, P.; Foltz, R.B. 2000. Spatial Prediction of Landslide Hazard Using Logistic Regression and GIS. Presented at the 4^th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4): Problems, Prospects and Research Needs, September 2 - 8, 2000, Banff, Alberta, Canada. 9 p.

Keywords: Landslides, Landslide hazard, Slope stability, Geographic Information System, Spatial Prediction, Multivariate models, Logistic Regression

Links: HTML [colorado.edu]

Abstract: The Inland Northwest U.S. has experienced significant and widespread landslide events in recent years. This research focuses on the Clearwater National Forest (CWNF) in central Idaho, and outlines the development of datasets, methods, and maps for the spatial prediction of landslide hazard. Terrain attributes are directly calculated from digital elevation models (DEMs) and include variables such as elevation, slope, plan and profile curvature, flow path length, and specific catchment area. The DEM-derived datasets are combined with a diversity of other digital datasets including climate surfaces, bedrock geology, roads, landslide events, and landcover. The datasets are combined in a geographic information system (GIS) for landslide exploration of correlation and statistical model development to predict the probability of occurrence or hazard. Logistic regression is used to predict the probability of occurrence in tandem with the receiver operator characteristic (ROC) curve as a measure of performance of a predictive rule. The ROC curve is a plot of the probability of having true positive identified landslides versus the probability of false positive identified landslides as the cut-off probability varies. The resulting product may be used for forest planning and decision support for maintenance, obliteration or development of forest roads. Predictions of increased local climate variability and extreme events portend that hazard maps are critical for minimizing negative impacts on forest infrastructure and environmental degradation due to sediment pulses and stream course obliteration.

Moscow FSL publication no. 2000i