Spatial prediction of landslide susceptibility using an adaptive neuro-fuzzy inference system combined with frequency ratio, generalized additive model, and support vector machine techniques
Introduction
The landslide is known as a natural hazard phenomenon that plays a critical role in the evolution of landscapes and represents a serious vulnerability in many areas of the world (Moosavi and Niazi, 2016). According to the latest statistical analyses of geological disasters from the Ministry of Land and Resources of the People's Republic of China, landslide is still the most commonly encountered geological disaster (http://www.cigem.gov.cn). In addition, landslides also cause thousands of casualties and fatalities, hundreds of billions of dollars in damage, and environmental problems each year (Aleotti and Chowdhury, 1999, Gutiérrez et al., 2015). In China, too many regions have suffered from landslide occurrences, and they have seriously threatened the safety of residents, transportation, electrical transmission lines, communication facilities, and gas and oil pipelines in the last several years (Lin et al., 2012, Ma et al., 2015, Wang et al., 2015, Xu et al., 2014, Yuan et al., 2015, Yuan et al., 2016, Yuan et al., 2013, Zhou et al., 2013). Specifically, people living in Hanyuan County (the study area) have been seriously affected by landslide activities and are under a high-risk threat of landslides due to the widely distributed mountainous and hilly areas in the county. Therefore, an assessment of the landslide susceptibility in this region is both valuable and important.
Generally, landslide damages could be largely decreased via spatial prediction of landslide susceptibility (Pradhan, 2010). Over the years, many researchers have studied landslide susceptibility assessment using GIS (Geographic Information System) and RS (Remote Sensing), and some of their studies have employed statistical models such as bivariate statistical analysis, logistic regression, and discriminant analysis (Alkhasawneh et al., 2014, Regmi et al., 2014, Chen et al., 2017c, Tsangaratos and Ilia, 2016, Wang et al., 2015). Multi-criteria analysis methods have also widely been used in landslide susceptibility evaluation (Erener et al., 2016, Feizizadeh et al., 2013a, Feizizadeh et al., 2013b, Feizizadeh et al., 2014b, Pourghasemi et al., 2013). Many researchers have considered uncertainty and probabilistic approaches to landslide susceptibility (Chen et al., 2016d, Feizizadeh and Blaschke, 2014, Feizizadeh et al., 2014a, Pourghasemi et al., 2014, Youssef et al., 2016a). In addition, some new data mining approaches also have been applied in the landslide literature in recent years, such as fuzzy logic, neuro-fuzzy, and artificial neural network models (Chen et al., 2017e, Lee et al., 2015, Vakhshoori and Zare, 2016), support vector machines (Chen et al., 2016b, Hong et al., 2015, Tien Bui et al., 2016b), maximum entropy (MaxEnt) (Pourghasemi et al., 2012, Chen et al., 2017b, Felicísimo et al., 2013), multivariate adaptive regression splines (Chen et al., 2017d, Conoscenti et al., 2015), kernel logistic regression (Chen et al., 2017f, Hong et al., 2015, Tien Bui et al., 2016b), alternating decision tree (Hong et al., 2015, Pham et al., 2016a), naive Bayes (Pham et al., 2015, Tien Bui et al., 2012), classification and regression tree (Chen et al., 2017g, Youssef et al., 2016b), and random forest (Chen et al., 2017g, Youssef et al., 2016b). In summary, various approaches have been carried out for landslide susceptibility assessment worldwide.
Nevertheless, some state-of-the-art models, such as the adaptive neuro-fuzzy inference system (Chen et al., 2017a, Nasiri Aghdam et al., 2016, Tien Bui et al., 2016a) combined with frequency ratio (ANFIS-FR) and the generalized additive model (GAM) (Pourghasemi and Rossi, 2016), have rarely been applied in the research field of landslide susceptibility assessment. The ANFIS performs best in conditions where there is no applicable method to select membership and parameters (Dehnavi et al., 2015), and this method has exhibited good performance for landslide susceptibility evaluation in some case studies (Dehnavi et al., 2015, Nasiri Aghdam et al., 2016, Pradhan, 2013). ANFIS-FR is the combination of machine learning and bivariate statistical approaches, both of which can evaluate the relationship between landslide conditioning factors and that of individual factor class and landslides. GAM is an extended version of the generalized linear model; this model replaces the linear predictor with an additive one, which can utilize categorical and continuous data as well as model the continuous data as a nonlinear smoothing function, and was used in landslide susceptibility analysis by several investigators (Park and Chi, 2008).
Overall, all the above methods have been applied successfully and efficiently in many individual studies. Out of these methods, SVM has been applied widely for landslide susceptibility assessment (Pham et al., 2016b). However, the ANFIS-FR and GAM methods have rarely been applied for landslide prediction. Moreover, their performance has not been compared in the literature. Thus, in this study, these three data mining approaches, namely, the ANFIS-FR, GAM, and SVM models, were applied, and their results were compared for prediction of landslide susceptibility for Hanyuan County, China.
Section snippets
Study area
The study area (Hanyuan County) is located near Yaan City, Sichuan Province, in the southwest part of China between longitudes 102°16′–103°00′E and latitudes 29°05′–29°43′N (Fig. 1). It covers an area of approximately 2388 km2. The study area is characterized as highly mountainous, with elevation values varying from 632 to 3940 m above sea level, decreasing from northwest to southeast. The mean altitude is 1956 m and the standard deviation is approximately 603 m. The climate of the study area is
Methodology
The methodology of this study is shown in Fig. 2. There are mainly four steps in the current study: (1) data preparation including preparation of a landslide inventory map and landslide conditioning factors, (2) multi-collinearity analysis and consideration of the correlation between landslide locations and conditioning factors, (3) landslide susceptibility modeling using three data mining models (ANFIS-FR, GAM, and SVM), and (4) validation of landslide susceptibility maps produced by the ROC
Correlation between landslides and conditioning factors using the FR model
The spatial relationship between landslides and the conditioning factors using the frequency ratio model is shown in Table 3. In this table, in the case of slope aspect, landslides are most abundant on south-facing (1.59), southwest-facing (1.28), and west-facing (1.10) slopes. The frequency ratios are lowest on flat, northeast-facing, and north-facing slopes. For the altitude factor, the frequency ratios decrease with increasing altitude. The altitudes between 632–1284 m and 1284–1773 m have
Discussion
Spatial prediction of landslides is considered one of the most difficult tasks in landslide hazard and risk assessment (Tien Bui et al., 2016b). However, the prediction accuracy of a landslide model depends on the method used. Therefore, the investigation of new methods is highly necessary and it will help reach reasonable conclusions (Chen et al., 2017f). In the current study, we evaluated and compared three machine learning methods such as ANFIS-FR, GAM, and SVM for landslide susceptibility
Conclusions
Landslide susceptibility assessment and its mapping constitute one of the primary steps in research on landslide hazard and risk. The aim of the present study was spatial modeling of landslide susceptibility in Hanyuan County, China using ANFIS-FR, GAM, and SVM data mining techniques. For this aim, twelve landslide conditioning factors were selected in the study area. The layers were slope aspect, altitude, slope angle, TWI, plan curvature, profile curvature, distance to rivers, distance to
Acknowledgments
This research was supported by China Postdoctoral Science Foundation funded project (Grant No. 2017M613168), Scientific Research Program Funded by Shaanxi Provincial Education Department (Program No. 17JK0511), and Open Fund of Shandong Provincial Key Laboratory of Depositional Mineralization & Sedimentary Minerals (Grant No. DMSM2017029).
References (120)
- et al.
The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan
Geomorphology
(2005) - et al.
GIS-based logistic regression for landslide susceptibility mapping of the Zhongxian segment in the Three Gorges area, China
Geomorphology
(2010) - et al.
Landslide susceptibility mapping based on GIS and support vector machine models for the Qianyang County, China
Environ. Earth Sci.
(2016) - et al.
Application of frequency ratio and weights of evidence models in landslide susceptibility mapping for the Shangzhou District of Shangluo City, China
Environ. Earth Sci.
(2016) - et al.
Performance evaluation of GIS-based new ensemble data mining techniques of adaptive neuro-fuzzy inference system (ANFIS) with genetic algorithm (GA), differential evolution (DE), and particle swarm optimization (PSO) for landslide spatial modelling
Catena
(2017) - et al.
Landslide spatial modeling: introducing new ensembles of ANN, MaxEnt, and SVM machine learning techniques
Geoderma
(2017) - et al.
A comparative study of logistic model tree, random forest, and classification and regression tree models for spatial prediction of landslide susceptibility
Catena
(2017) - et al.
Susceptibility mapping of shallow landslides using kernel-based Gaussian process, support vector machines and logistic regression
J. Afr. Earth Sci.
(2016) - et al.
Assessment of susceptibility to earth-flow landslide using logistic regression and multivariate adaptive regression splines: a case of the Bence River basin (western Sicily, Italy)
Geomorphology
(2015) - et al.
A new hybrid model using step-wise weight assessment ratio analysis (SWARA) technique and adaptive neuro-fuzzy inference system (ANFIS) for regional landslide hazard assessment in Iran
Catena
(2015)