Assessing susceptibility to landslides: Using models to understand observed changes in slopes Netra R. Regmi ⁎, John R. Giardino, John D. Vitek Department of Geology and Geophysics and High Alpine and Arctic Research Program (HAARP), Texas A&M University, College Station, TX 77843, USA abstract article info Article history: Received 9 January 2010 Received in revised form 14 May 2010 Accepted 15 May 2010 Available online 7 July 2010 Keywords: Landslides Weight of evidence Frequency ratio Fuzzy operators Susceptibility map West-central Colorado A map of landslide susceptibility is a necessary tool for proper planning and selection of sites for agriculture, infrastructure and other human developments. The Paonia–McClure Pass area of Colorado, USA, is well known for active mass movements. Large losses of property and risks to people highlight the need to accurately map susceptibility to shallow landslides and to identify safe locations for infrastructure and residential development. We mapped 735 active mass movements and 17 factors about each one. The weights of evidence, frequency ratio of landslides, and fuzzy-logic method were used to create an optimum map of landslide susceptibility. Weights of the evidence were used to categorize continuous factor data, frequency ratios of shallow landslides were used to assign the membership values for the categories of the factors, and the fuzzy-logic method was used to integrate the membership values. Four models from the fuzzy-inference network of mapping susceptibility to shallow landslides were developed based on the combination of factors using five types of fuzzy operators. The first inference network model was comprised of the combination of factors, which are independent of each other. The second, third and the fourth inference network models were developed such that factors are not necessarily independent of each other. These models combine all dependent and independent factors, based on the expert's knowledge. Intermediate steps in the second, third and fourth models were developed by combining the fuzzy factors in the first step by fuzzy-OR, fuzzy- AND, and fuzzy-OR plus fuzzy-AND operations, respectively. All models predicted similar percentages of observed shallow landslides with the fuzzy-gamma operation. Although the prediction capabilities of all the models are not significantly different, the fourth model is the best because it is the only model that accommodates the under-sampled and missed landslide data and the effect of increasing and decreasing gamma values. The first and third models create a problem if a category of a factor has a 0 membership value because of the absence or under-sampling of shallow landslides. The second model incurs the highest increasing effect of gamma values, and the third model incurs the highest decreasing effect of gamma values. The approaches described in this paper reduce the uncertainties associated with the categorization of continuous data, determination of fuzzy-membership values, and the combination of factors that causes shallow landslides. © 2010 Elsevier B.V. All rights reserved. 1. Introduction Maps of landslide hazards/susceptibilities and risks are necessary tools for engineers, earth scientists, planners and decision makers to select appropriate sites for development of agriculture, construction and other human activity. Numerous articles have been published on mapping hazards and susceptibilities to landslides. The methods of mapping landslide susceptibility can be categorized into qualitative or knowledge-based (Carrara and Merenda, 1976; Kienholz, 1978; Fenti et al., 1979; Ives and Messerli, 1981; Rupke et al., 1988, Regmi et al., 2010a), quantitative or statistical (Carrara, 1983; Carrara et al., 1991, 1999; Anbalagan, 1992; Juang et al., 1992; Maharaj, 1993; Gokceoglu and Aksoy, 1996; Van Westen et al., 1997; Atkinson and Massari, 1998; Pachauri et al., 1998; Guzzetti et al., 1999; Rautela and Lakhera, 2000; Gritzner et al., 2001; Sakellariou and Ferentinou, 2001; Cevik and Topal, 2003; Gorsevski et al., 2003; Lee, 2004; Tangestani, 2004; Ayalew and Yamagishi, 2005; Regmi et al., 2010a,b) and deterministic methods (Chowdhury, 1976; Chowdhury and Bertoldi, 1977; Wu and Sidle, 1995; Gokceoglu and Aksoy, 1996). In summary, qualitative or knowledge-based methods are based on field observations and a priori knowledge of the expert, in which the expert identifies landslides and makes a priori assumptions about those sites where movement has occurred and is likely to occur again. The expert then develops decision rules or assigns weighted values for the classes of index maps and overlays them to develop a map of landslide susceptibility. Geomorphology 122 (2010) 25–38 ⁎ Corresponding author. Tel.: + 1 517 803 5578. E-mail address: netraregmi@neo.tamu.edu (N.R. Regmi). 0169-555X/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.geomorph.2010.05.009 Contents lists available at ScienceDirect Geomorphology journal homepage: www.elsevier.com/locate/geomorph