Optimal landslide susceptibility zonation based on multiple forecasts

Abstract

Environmental and multi-temporal landslide information for an area in Umbria, Italy, was exploited to produce four single and two combined landslide susceptibility zonations. The 78.9 km² study area was partitioned in 894 slope units, and the single susceptibility zonations were obtained through linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), logistic regression (LR), and by training a neural network (NN). The presence or absence of landslides in the slope units in the period from pre-1941 to 1996 (training set) was used as the dependent variable for the terrain classification. Next, adopting a regression approach, two “optimal” combinations of the four single zonations were prepared. The single and the combined zonations were tested against landslides in the 9-year period from 1997 to 2005 (validation set). Different metrics were used to evaluate the quality of the susceptibility zonations, including degree of model fit, uncertainty in the probability estimates, and model prediction skills. These metrics showed that the degree of model fit was not a good indicator of the model forecasting skills. Zonations obtained through classical multivariate classification techniques (LDA, QDA and LR) produced superior predictions when compared to the NN model, that over fitted the landslide information in the training set. LDA and LR produced less uncertain zonations than QDA and NN. The combined models resulted in a reduced number of errors and in less uncertain predictions; an important result that suggests that the combination of landslide susceptibility zonations can provide “optimal” susceptibility assessments.

Introduction

Landslide susceptibility (LS) is the likelihood of a landslide occurring in an area on the basis of local terrain conditions (Brabb, 1984). It is the degree to which an area can be affected by future slope movements, i.e. an estimate of “where” landslides are likely to occur (Guzzetti et al., 1999, Guzzetti et al., 2005, Guzzetti et al., 2006a, Guzzetti et al., 2006b). In mathematical language, LS is the probability of spatial (geographical) occurrence of slope failures, given a set of geo-environmental conditions (Chung and Fabbri, 1999, Guzzetti et al., 2005, Guzzetti et al., 2006a). Susceptibility does not consider the temporal probability of failure (i.e., when or how frequently landslides occur), nor the magnitude of the expected landslide (i.e., how large or destructive the failure will be) (Committee on the Review of the National Landslide Hazards Mitigation Strategy, 2004). For this reason, landslide susceptibility is different from landslide hazard (Guzzetti et al., 2005, Guzzetti et al., 2006a, Guzzetti, 2006).

The concepts, principles, techniques and methods for LS evaluation are known, and can be found, among others, in Carrara (1983), Brabb (1984), Hansen (1984), Varnes and IAEG Commission on Landslides and Other Mass-Movements (1984), van Westen (1994), Soeters and van Westen (1996), van Westen et al. (1997), Aleotti and Chowdhury (1999), Chung and Fabbri (1999), Guzzetti et al. (1999), Vandine et al. (2004), Crozier and Glade (2005), and Guzzetti (2006).

In the last two decades, the availability of (i) fast and efficient personal computers, (ii) low cost, commercial and open source GIS and statistical software, and (iii) thematic and environmental information readily available in digital format, have facilitated the preparation of LS zonings. Authors have started to compare LS models, and the associated terrain zonations, prepared exploiting different classification methods, and to evaluate their performances (e.g., Carrara et al., 1992, Carrara et al., 1995, Chung and Fabbri, 1999, Barredo et al., 2000, Lee, 2004, Süzen and Doyuran, 2004, Ayalew et al., 2005, Yesilnacar and Topal, 2005, Davis et al., 2006, Kanungo et al., 2006, Lee and Sambath, 2006, Wang and Sassa, 2006, Irigaray et al., 2007, Carrara et al., 2008, Song et al., 2008, Van Den Eeckhaut et al., 2009). However, inspection of the literature reveals that little work has been done to determine strategies, and to evaluate operational methods for the optimal assessment of LS in an area (e.g., Carrara et al., 1995). This work attempts to bridge this gap, exploiting multivariate classification techniques. Conversely, the work is not intended to evaluate the quality and role of the landslide and environmental information used to obtain the LS zonations.

We start by recognizing that an LS zonation is a form of quantitative forecast of the spatial (geographical) distribution of landslides (Chung and Fabbri, 2003, Fabbri et al., 2003, Guzzetti et al., 2005, Guzzetti et al., 2006a, Guzzetti et al., 2006b), and that multiple zonations can be prepared for an area exploiting the same thematic (landslide and environmental) information. Next, for a study area in central Umbria, Italy (Fig. 1), we exploit environmental and landslide information to calibrate (train) and validate (test) four independent LS forecasting models. To obtain the four geographical forecasts, we adopt three multivariate statistical classification techniques, and we train a neural network. Lastly, adopting a regression approach, we obtain two optimal combinations of the four individual zonations, and we test their predictive performance against independent landslide information. We conclude discussing the results obtained, and presenting a script for the R free software environment for statistical computing (http://www.r-project.org/) for the production and quantitative assessment of LS models, and the associated terrain zonations.