GIS-Based Logistic Regression for Landslide Susceptibility Analysis in Western Washington State

Introduction

Natural disasters are usually not randomly distributed and occur in different regions of the world due to climatic patterns, topography, or tectonic activity. Droughts, floods, and hurricanes are prominent examples of natural disasters but there are also other, lesser reported disaster occurrences, that effect most nations: landslide activities (Sidle and Ochiai, 2006). In the United States, landslides occur or have occurred in all fifty states (Highland, 2004). The primary regions of landslide occurrences are the Intermountain West, mountainous and hilly regions of the eastern U.S., and the mountainous and coastal areas of the West Coast (Highland, 2004). The State of Washington experienced hundreds of landslides during the winters of 1995, 1996, and 1997 due to record rainfall, and these slides caused significant property damage and disrupted transportation networks (Baum et al., 2005). Devastating to the local economy, these severe slides stimulated renewed interest in identifying areas prone to slope failure (Baum et al., 2005). Recently, the town of Oso in western Washington experienced a tragic landslide resulting in significant home destruction, the death of 43 people, and major damage to state Highway 530 (Stromberg, 2014). Identifying potential landslide areas has once again been revived in western Washington (Stromberg, 2014).

In order to minimize potential adverse impacts of landslides, Geographic Information Systems (GIS) have been used to map landslide susceptibility (Dahal et al., 2007). Many techniques have been developed and applied in the literature to produce susceptibility maps (Kavzoglu et al., 2014). Landslide susceptibility is assessed through qualitative and quantitative approaches (Dahal et al., 2007). Qualitative approaches rely entirely on expert opinion, and susceptible areas are often categorized by such terms as “very high” or “very low” (Castellanos Abella and Van Westen, 2008). The two types of quantitative approaches utilized for landslide susceptibility mapping are deterministic and statistical (Yalcin et al., 2011). Deterministic methods are primarily based on slope stability studies and are only applicable when ground conditions are uniform across the entire study area (Dahal et al., 2007; Yalcin et al., 2011). The statistical approach is an indirect susceptibility mapping methodology, which involves analyzing the historical link between landslide-controlling factors (i.e. independent variables) and the distribution of landslides (dependent variable) (Atkinson and Massari, 1997; Dahal et al. 2007; Yalcin et al., 2011). The most widely used statistical approaches in landslide susceptibility mapping include bivariate and multivariate statistical techniques such as logistic regression (Devkota et al., 2013). Other quantitative approaches include data mining models such as support vector machine, decision tree methods, artificial neural networks, etc. (Devkota et al., 2013).

Most published literature on landslide susceptibility mapping involves comparing known landslide susceptibility approaches, or adjusting universally accepted methodologies in order to produce more accurate maps. Many studies have also used repeated methodologies from prior literature but at different geographic locations or scales. The purpose of this research is to quantitatively identify high risk areas to landslide occurrence within western Washington by developing a GIS-based landslide susceptibility model. Multivariate logistic regression was utilized as the statistical methodology. In addition, this paper addresses the level of risk associated with the Oso landslide. This research aims to aid western Washington land-use managers in identifying areas prone to landslides in order to save lives and mitigate environmental and economic costs.