Logistic regression versus artificial neural networks: landslide susceptibility evaluation in a sample area of the Serchio River valley, Italy

Abstract

This article presents a multidisciplinary approach to landslide susceptibility mapping by means of logistic regression, artificial neural network, and geographic information system (GIS) techniques. The methodology applied in ranking slope instability developed through statistical models (conditional analysis and logistic regression), and neural network application, in order to better understand the relationship between the geological/geomorphological landforms and processes and landslide occurrence, and to increase the performance of landslide susceptibility models. The proposed experimental study concerns with a wide research project, promoted by the Tuscany Region Administration and APAT-Italian Geological Survey, aimed at defining the landslide hazard in the area of the Sheet 250 ‘‘Castelnuovo di Garfagnana’’ (1:50,000 scale). The study area is located in the middle part of the Serchio River basin and is characterized by high landslide susceptibility due to its geological, geomorphological, and climatic features, among the most severe in Italy. Terrain susceptibility to slope failure has been approached by means of indirect-quantitative statistical methods and neural network software application. Experimental results from different methods and the potentials and pitfalls of this methodological approach have been presented and discussed. Applying multivariate statistical analyses made it possible a better understanding of the phenomena and quantification of the relationship between the instability factors and landslide occurrence. In particular, the application of a multilayer neural network, equipped for supervised learning and error control, has improved the performance of the model. Finally, a first attempt to evaluate the classification efficiency of the multivariate models has been performed by means of the receiver operating characteristic (ROC) curves analysis approach

Landslide susceptibility assessment in the Upper Orcia Valley (Southern Tuscany, Italy) through conditional analysis: a contribution to the unbiased selection of causal factors

“…The statistical approach, instead, is founded on the multivariate relationships between causal factors and past and present landslide occurrence. The multivariate relationships are often identified through conditional analysis (BonhamCarter et al, 1989;Carrara et al, 1995), discriminant analysis (Agterberg, 1974;Carrara, 1983;Carrara et al, 1995Carrara et al, , 2003Baeza and Corominas, 2001), linear or logistic regression (Atkinson and Massari, 1998;Guzzetti et al, 1999, and references therein; Gorsevski et al, 2000;Dai and Lee, 2003;Ohlmacher and Davis, 2003;Ayalew and Yamagishi, 2005), and artificial neural networks (Aleotti et al, 1996;Lee et al, 2001;Wang and Sassa, 2006;Falaschi et al, 2009;Pradhan and Lee, 2010).…”

Influence of management of variables, sampling zones and land units on LR analysis for landslide spatial prevision

“…Among this group, the majority of authors employs a sample population randomly extracted from the whole study area (Dai and Lee, 2002;Dai et al, 2004;Chau and Chan, 2005;Yesilnacar and Topal, 2005;Garcia-Rodriguez et al, 2008;Mathew et al, 2009;Choi et al, 2012), other researchers select only a portion of the surface to be used (Nefeslioglu et al, 2008;Nandi and Shakoor, 2009). Further approaches consist in using a different number of stable and landslide prone cells (0/1 = 1), considering as a sample population either the whole study area (Bernknopf et al, 1988;Ohlmacher and Davis, 2003;Ayalew and Yamagishi, 2005;Chen and Wang, 2007;Falaschi et al, 2009;Chauhan et al, 2010;Erener et al, 2010;Rossi et al, 2010;Yalcin et al, 2011) or a part of the same (Gorsevski et al, 2000;Can et al, 2005;Van Den Eeckhaut et al, 2006;Greco et al, 2007;Sorriso-Valvo et al, 2009). …”

“…Coding of independent variables has been dealt with by several authors according to two different approaches; the first approach, which is used by the vast majority of authors (Bernknopf et al, 1988;Dai and Lee, 2002;Ohlmacher and Davis, 2003;Dai et al, 2004;Ayalew and Yamagishi, 2005;Can et al, 2005;Chau and Chan, 2005;Van Den Eeckhaut et al, 2006;Chen and Wang, 2007;Garcia-Rodriguez et al, 2008;Nefeslioglu et al, 2008, Mathew et al, 2009Falaschi et al, 2009;Chauhan et al, 2010;Rossi et al, 2010), creates layers having binary variables (dummy variables) for all the categories or classes of each dependent variable (a variable having ten categories or classes, generates the same number of binary variables). Such approach is sensible when a limited number of independent variables is available and when such variables are -in turn -articulated in a few categories or classes.…”

“…Among these units, except for a less number of authors who make use of slope units (Carrara et al, 1991(Carrara et al, , 1995Komac, 2006;Rossi et al, 2010) and homogeneous land units (Can et al, 2005;Falaschi et al, 2009), cells matrix is the most widely used land unit by the authors considered (Bernknopf et al, 1988;Gorsevski et al, 2000;Dai and Lee, 2002;Ohlmacher and Davis, 2003;Dai et al, 2004;Ayalew and Yamagishi, 2005;Chau and Chan, 2005;Yesilnacar and Topal, 2005;Van Den Eeckhaut et al, 2006;Greco et al, 2007;Chen and Wang, 2007;Garcia-Rodriguez et al, 2008;Nefeslioglu et al, 2008;Mathew et al, 2009;Sorriso-Valvo et al, 2009;Nandi and Shakoor, 2009;Chauhan et al, 2010;Erener et al, 2010;Yalcin et al, 2011;Choi et al, 2012). Authors almost homogeneously perform the fitting and weight assessment steps for independent variables and for the variables of analysis application (assessment of P (y)).…”

“…Among the several statistical procedures adopted to assess the susceptibility to mass-movements, Logistic Regression (LR), has increasingly been adopted (Bernknopf et al, 1988;Gorsevski et al, 2000;Dai and Lee, 2002;Ohlmacher and Davis, 2003;Dai et al, 2004;Ayalew and Yamagishi, 2005;Can et al, 2005;Chau and Chan, 2005;Yesilnacar and Topal, 2005; Van Den Eeckhaut et al, 2006;Greco et al, 2007;Chen and Wang, 2007;Garcia-Rodriguez et al, 2008;Nefeslioglu et al, 2008;Mathew et al, 2009;Sorriso-Valvo et al, 2009;Falaschi et al, 2009;Nandi and Shakoor, 2009;Chauhan et al, 2010;Erener et al, 2010;Rossi et al, 2010;Yalcin et al, 2011;Choi et al, 2012). Different procedures of the LR application have been employed, and they mainly differ in terms of sampling modalities, variables transformation, and selection of reference land units.…”