Abstract
In this study, landslide susceptibility assessments were achieved using logistic regression, in a 523 km2area around the Eastern Mediterranean region of Southern Turkey. In reliable landslide susceptibility modeling, among others, an appropriate landslide sampling technique is always essential. In susceptibility assessments, two different random selection methods, ranging 78–83% for the train and 17–22% validation set in landslide affected areas, were applied. For the first, the landslides were selected based on their identity numbers considering the whole polygon while in the second, random grid cells of equal size of the former one was selected in any part of the landslides. Three random selections for the landslide free grid cells of equal proportion were also applied for each of the landslide affected data set. Among the landslide preparatory factors; geology, landform classification, land use, elevation, slope, plan curvature, profile curvature, slope length factor, solar radiation, stream power index, slope second derivate, topographic wetness index, heat load index, mean slope, slope position, roughness, dissection, surface relief ratio, linear aspect, slope/aspect ratio have been considered. The results showed that the susceptibility maps produced using the random selections considering the entire landslide polygons have higher performances by means of success and prediction rates.
Keywords Landslide sampling strategy Logistic regression Landslide susceptibility Southern Turkey
Introduction
Landslide susceptibility refers to the subdivision of the land into homogeneous areas based on different likelihoods that certain types of landslide may occur in the future and is assessed using, either qualitative or quantitative meth-ods. The rapid development in Geoinformation Sciences for the past few decades has led to the assessment of numerous approaches to determine landslide susceptibility. The landslide susceptibility methods are generally based on the notion that the future landslides are likely to occur under similar geo-environmental conditions to the ones of past landslides. Overviews of the methods, available for landslide susceptibility assessment, can be found in Dai et al. (2002), Wang et al. (2005), Chacon et al. (2006), Fell et al. (2008a,b), van Westen et al. (2008), Corominas et al.
(2014), Hungr (2016) and Rossi and Reichenbach (2016). The choice of the most appropriate method depends on the type of landslide, the extent of the study area and the availability of the spatial data. Statistical and machine learning methods are the most common quantitative data-driven methods used for regional scale landslide suscepti-bility assessments (e.g., Corominas et al.2014; Goetz et al.
2015; Chen et al. 2017).
In data-driven methods, the selection of the terrain mapping units (Guzzetti et al.1999) either vector-based or grid-based, is required to determine how landslides will be sampled, to prepare the training and validation datasets for the landslide susceptibility modeling. In grid-based land-slide susceptibility assessments several strategies are used to sample landslide affected pixels in accordance with the landslide inventory maps (e.g., Suzen and Doyuran 2004; Can et al. 2005; Clerici et al. 2006; Van Den Eeckhaut et al. 2006; Gorum et al. 2008; Nefeslioglu et al. 2008; Yilmaz 2010; Nefeslioglu et al. 2012; Regmi et al.2014; Hussin et al. 2016). Regmi et al. (2014) summarized the sampling strategies from landslide affected pixels into four & Tolga C¸an
tolgacan@cukurova.edu.tr
1 Department of Geological Engineering, C¸ ukurova University,
classes: as a single pixel from landslide scarp, the entire landslide area, the center of the landslide mass and the center of the landslide scarp. In some studies, landslide affected pixels are taken from the outside of the landslide mass using the main scarp upper age approach (e.g., Dai and Lee2003; Clerici et al.2006) and seed–cell approach (e.g., Suzen and Doyuran 2004; Nefeslioglu et al. 2008). Pre-landslide terrain conditions in term of slope angles were considered as a parameter in the production of land-slide susceptibility maps by Van Den Eeckhaut et al. (2006) and Gorum et al. (2008).
Once the selection of landslide affected pixels are con-sidered another selection procedure is also required to separate the landslides into train and validation sets. In general, 70–80% of the landslides are selected for train data set and the remaining 20–30% of the landslides are selected for the validation dataset.
In this study, two different landslide sampling strategies were practiced during the selection of the train and vali-dation datasets. In the first selection method, landslide pixels were randomly selected based on their identity numbers considering the whole landslide polygons corre-sponding 78–83% for the train and 17–22% validation sets. In the second one, an equal number of landslides affected pixels were randomly selected in any part of the landslide mass. The test area selected for this study is located in the Eastern Mediterranean region of Turkey that is known as one of the slightly landslide-prone areas in Turkey (Duman et al. 2011; Can et al. 2013). However, landslides are abundant in several regions due to the local geologic, geomorphologic and physical factors and caused intermit-tently significant losses to the settlements and infrastruc-tures (Duman et al.2009; Tekin2014; Tekin et al.2015). Landslides adversely affect the residential areas, agricul-tural lands, engineering structures such as transportation systems and crude oil pipeline (Tekin et al.2015). For that reason, a reliable landslide susceptibility map is essential for regional land use planning strategies. In this study, considering two different landslide sampling strategies, landslide susceptibility models using logistic regression method were carried out and prediction-success skills of the maps were compared in Kadirli town of Eastern Mediterranean region of Turkey (Fig.1).
Study Area and Spatial Data Preparation
The study area is located between 36402000 and 36170000latitude, 372903000and 371403000longitude in the Eastern
Mediterranean region of Turkey and covers 523 km2. The study area represents rugged and semi-mountainous topo-graphical features (see Fig.1). The climate is of Mediter-ranean type, with a typical hot and relatively dry summer
between June and August, and a wet season during autumn, winter, and spring. According to the closest meteorological station at Kadirli town, the annual precipitation ranges between 464 and 1133 mm with the mean value of 790.0 mm. The 85% of the total rainfall occurs between December and February. Heavy rainfall is the main land-slide triggering factor in the study area.
Landslide Inventory
A reliable landslide inventory data is a basic requirement for quantitative zoning of landslide susceptibility, hazard, and risk. Preparing landslide inventory maps is important to document the spatial distribution of landslides in a region together with their types, patterns, recurrence and statistics of slope failures (Fell et al. 2008a; van Westen et al.2008; Van Den Eeckhaut et al.2012; Guzzetti et al.
2012). The Turkish landslide inventory map database prepared by Duman et al. (2009) was used in this study after updating the recent landslides by field survey and analyzing Google Earth images. 185 landslides with mean landslide area of 0.07 km2 were delimited by their full extent as polygons including run-out zones (see Fig.1). All the landslides are classified as deep-seated and active according to the national landslide inventory mapping standards by Duman et al. (2005). Landslides damage the roads, agricultural lands and the crude oil pipeline in places (Fig.2).
Landslide Conditioning Factors
The study area is mainly represented by low-lying hillside slopes with elevation ranging between 48 and 695 masl (see Fig. 1), where 94% of the landslides are located from 200 to 600 masl. The gentle slopes lower than 10 and higher than 20 correspond to 50 and 20% of the study area, respectively. Landslides are mostly concentrated on the slopes between 10 and 20 (Fig.3a). Geologically, the study is in the Misis-Andirin Tectonic High (Kozlu1997). The pre-Pliocene basements rocks in this tectonic zone are overlain discordantly by the Kadirli Formation which hosts 95% of the landslides in the study area. The Kadirli For-mation is mainly composed of weakly cemented coarse conglomerate, conglomerate-sandstone, sandstone and loosely cemented clay-sand-pebble levels of Pliocene age crops out in the 67% of the study area (Fig.3b). The geological map of the study area was converted to raster format with grid cell size of 25 9 25 m, which is the same for the other raster data, and each geological formation was assigned by binary value (1 and 0). The land use map is classified into bare, bare and agriculture, plantation and shrub, lake, uncultivated farrow and forest (Fig.3c) using Landsat 7 satellite image. Bare lands with agricultural
areas correspond to 59% of the study area comprising 81% of the landslides. The study area was classified into 10 landform classes according to the Weiss (2001) using topographic position index (Fig.3d). It has seen that the 45% of the study area is classified as Open slopes, followed by plains (25%) and the other landform classes in smaller percentages. The open slopes located between ridges and valleys have 69% of the landslides in the study area. Curvature, plan and profile curvatures (Moore et al.1991), slope length factor (Moore and Burch 1986a, b), solar radiation (Fu and Rich 1999), stream power index, topo-graphic wetness index (Moore et al. 1993), Slope second derivate, heat load index, linear aspect (McCune and Keon
2002), the mean-slope, roughness (Riley et al.1999), dis-section (Evans1972) surface relief ratio (Pike and Wilson
1971), slope/aspect ratio Stage’s (1976) COS, SIN and Roberts and Cooper (1989) topographic radiations aspect index parameters were calculated using a 3 9 3 cell neighborhood around the center grid cell. The results were summarized in Table1.
Methods
Logistic Regression
Logistic regression (LR) is an efficient multivariate statis-tical method to investigate the relationship between a binary dependent variable with a set of continuous and/or categorical independent variables. LR is one of the most widely used methods for landslide susceptibility mapping (Atkinson and Massari 2011). The probability of the observed dependent variable as a function of the indepen-dent variables are expressed by the maximum likelihood method. In LR model, the linear combination of the inde-pendent variables X1…Xn can be expressed by the fol-lowing equation (Eq.1),
Z¼ b0þ b1X1þ b2X2þ þ bnXn ð1Þ
where b0 is the constant and b1…bn are the maximum likelihood estimates.
And the estimated logistic probability of landslide occurrences in each grid cell is calculated from the Eq.2,
P¼ 1=1 þ e Zð Þ ð2Þ
where P is the probability of landslide occurrence and Z is the weighted linear combination of the independent vari-ables. The estimated logistic probabilities vary between 0 and 1 (Dai et al.2004). The spatial relationship between the location of each landslide and the 20 conditioning param-eters were calculated using the logistic regression model following Eqs.1and2.
In LR, the significance of the coefficients bi can be tested using the Wald test, which obtained by comparing the maximum likelihood estimate of every bi with the standard error. The Wald statistical values increase with the significance of bi coefficients (Hosmer et al. 2013). Forward stepwise selection procedure was utilized to integrate the independent variables with a significant con-tribution to the presence of landslides.
Construction of Training and Validation Models
The grid cell was selected as the terrain mapping unit which determined how landslides will be sampled to pre-pare the training and validation datasets for the suscepti-bility modeling. In landslide susceptisuscepti-bility assessments several strategies are used to sample landslide affected
pixels. Regmi et al. (2014) summarized the sampling strategies from landslide affected pixels into four classes as a single pixel from landslide scarp, the center of the landslide mass, the center of the landslide scarp and the entire area of a landslide. The main scarp upper age approach (e.g., Donati and Turrini 2002) and seed–cell approach (e.g., Suzen and Doyuran2004) select the pixels in undisturbed morphological conditions. In some studies, landslide affected pixel are selected within the partial landslide mass (e.g., Duman et al.2006; Nefeslioglu et al.
2012). According to the grid cell size of 25 m, the landslide data set is represented by 21,749 grid cells (13.59 km2) where the total study area has 830,489 pixels. Landslide affected pixels were randomly divided into two groups for analysis and validation set using two different sampling strategies. The first group random samples were generated based on the identity numbers of the landslides considering the entire polygon, varying between 78 and 83% for the analysis and 17 and 22% for the validation set. The second group random samples were selected from any part of the landslide affected pixels keeping the same proportions of the first one. Three random selections for the landslide free Fig. 2 Active deep-seated landslides in the study area; rotational landslides on cut slopes of a road (a), in an orchard (b) and complex landslides on hillside slopes (c, d)
grid cells of equal proportion were also applied for each of the landslide affected data set.
Model validation is a fundamental step in landslide susceptibility study. Validation refers to comparing the model predictions for assessing its accuracy or predictive power. The performance of the susceptibility models was evaluated using the area under the receiver operating characteristic curves (AUC) (e.g. Begueria2006; Van Den Eeckhaut et al.2006) and success rate and prediction rate curves (Chung and Fabbri 2008) for the entire area and validation datasets.
Success and prediction rate curves plot the percentage of the study area in each susceptibility class against the
percentage of landslide area in the same susceptibility class. Success rate curves are constructed from the sus-ceptibility model of the entire area obtained by the training dataset. Prediction rate curves are built considering vali-dation data set to measure the prediction skill of suscepti-bility model. This method works by creating specific rate curve which explains the percentage of known landslides that fall into each defined level of susceptibility classes and displays as the cumulative frequency diagram.
ROC curves plot ‘‘sensitivity’’ versus 1-‘‘specificity’’, where sensitivity is the proportion of landslide affected mapping units that are correctly classified as susceptible, and ‘‘specificity’’ is the proportion of landslide free Table 1 Descriptive statistics of
the continuous independent variables for the study area and the landslides
Variables Minimum Maximum Mean SD
Study area (n = 830,489 count)
Digital elevation model 48.07 695.00 244.45 112.31
Slope 0.00 63.39 10.84 8.01
Curvature - 14.08 11.75 0.00 0.89
Plan curvature - 5.78 6.98 0.01 0.49
Profile curvature - 8.35 8.30 0.01 0.56
Roughness 0.00 7.37 2.41 1.03
Topographic wetness index - 0.49 24.03 4.47 4.96
Stream power index 0.00 10.90 0.60 0.99
Solar radiation 0.26 1.06 0.85 0.11
Heat load index 1968.00 10,000.00 7555.21 1152.75
Second derivative slope 0.00 45.19 6.06 5.20
Dissection 0.00 1.00 0.46 0.25
Slope length factor 0.00 244.87 1.84 3.10
Mean slope 0.00 52.18 10.84 6.78
Linear aspect 0.00 359.00 180.94 101.87
Slope/aspect ratio 625.00 1395.30 643.38 24.16
Surface relief ratio 0.00 0.90 0.46 0.17
Landslide area (n = 21,749 count)
Digital elevation model 146.00 621.33 309.33 84.08
Slope 0.00 39.94 12.93 5.87
Curvature - 5.51 6.89 0.10 0.84
Plan curvature - 3.87 3.67 0.05 0.47
Profile curvature - 3.40 4.28 0.05 0.50
Roughness 0.00 4.87 2.80 0.56
Topographic wetness index 0.27 24.03 3.22 1.76
Stream power index 0.00 9.48 0.81 0.97
Solar radiation 0.36 1.06 0.84 0.12
Heat load index 2930.00 9998.00 7207.74 1251.16
Second derivative slope 0.00 36.95 6.09 4.23
Dissection 0.00 1.00 0.48 0.17
Slope length factor 0.00 103.14 2.46 2.83
Mean slope 0.54 31.43 12.89 4.39
Linear aspect 0.00 359.00 146.54 97.09
Slope/aspect ratio 625.00 815.20 645.17 19.82
mapping units that are correctly classified as landslide free. The AUC is a quantitative measure of the model perfor-mance. As the distributions of the probabilities estimated by the model become more distinct. The plot of the ROC curve rises more rapidly, and the AUC increases from 0.5 to its theoretical maximum of 1.0. The general guidelines for the degree of discrimination suggested for the AUC value between 0.8 and 0.9 is considered as excellent dis-crimination where AUC C 0.9 is considered outstanding discrimination (Hosmer et al.2013).
Landslide Susceptibility Maps
The logistic regression analyses using a forward stepwise procedure were applied to each different sampling training data set. The landslide susceptibility maps for the whole area were then obtained from the regression coefficients of
each model. As a first step to compare the performances of the two different sampling strategies and to choose the foremost model, the success and prediction rate curves together with the area under ROC curves were calculated. The area under ROC curves can be summarized quantita-tively by the accuracy or prediction ability of the devel-oped landslide susceptibility model. The area under ROC curve that ranges from 0.5 to 1.0 describes the discrimi-nation as follows: 0.9–1, outstanding discrimidiscrimi-nation; 0.8–0.9, excellent discrimination; 0.7–0.8, acceptable dis-crimination; 0.6–0.7, poor disdis-crimination; and 0.5–0.6, no discrimination (Hosmer et al.2013).
Also, the resulted landslide susceptibility maps have been verified using the two rules proposed by Can et al. (2005) that first, the majority of the landslides should locate in the high and very high susceptibility classes, and second, the high and very high susceptibility classes should cover the small percentages of the area (Bai et al. 2010; Fig. 4 Success and prediction rate curves of dataset 1 (a), dataset 2 (b) and dataset 3 (c) with AUC values of ROC curves for the landslide susceptibility maps prepared using the random selection of the entire landslide polygons
Fig. 5 Success and prediction rate curves of dataset-a (a), dataset-b (b) and dataset-c (c) with AUC values of ROC curves for the landslide susceptibility maps prepared using random selection within the landslide polygons
Pradhan and Lee 2010; Pourghasemi et al. 2012; Boure-nane et al.2016).
The selection of the probability intervals in suscepti-bility classes also varies widely such as natural Jenks, equal interval etc. which has a great influence on the partition and the proportion of the susceptibility classes in the study area. In this study, an equal interval of probability classes was chosen which resulted in a small percentage of the study area containing a large percentage of the existing landslides in the high and very high susceptibility classes. Results for the first group random selection approach showed that the high–very high susceptibility classes cor-respond 17.31, 16.95 and 23.98% of the area comprising 77.34, 80.43 and 71.07% of all landslides, respectively (Fig.4). The high–very high susceptibility classes of the second random selection datasets are 24.41, 32.38, 33.64% comprising 71.28, 67.43, 71.26% of all landslides (Fig.5). It was found that the success and prediction rates for the random sampling datasets of entire polygons have higher scores (Fig.4) than the second model (Fig.5). The sus-ceptibility models with highest performances from each random selection were given in Fig.6. According to the best landslide susceptibility model shown in Fig.6a, the 16.95% of the study area is located within the high–very
high susceptibility classes comprising the 84.84% of the validation and 80.43% of all landslides.
In susceptibility models, the common variables, that positively related to the presence of landslides entered into the regression models, were found as Kadirli Formation, elevation, roughness, bare and upland drainages whereas alluvium, deep canyons, topographic wetness index and slope variables were entered into the model with negatively related to landslides. The variables entered in the best logistic regression model were summarized in Table2. Landform factors together with elevation and Kadirli For-mation appeared to be the most important variables con-trolling the landslides. Consequently, the spatial distribution of the most susceptible areas within the Kadirli Formation were found in upland drainage classes with high elevation and high roughness values.
Conclusions
In the present study, two different landslide sampling techniques on the performance of the landslide suscepti-bility maps, were evaluated. In landslide susceptisuscepti-bility models, it has been seen that the selection of the landslide Fig. 6 Landslide susceptibility models produced by selection of the entire landslide polygon (a) and partial landslide polygon (b)
affected pixels from the entire landslide polygon has higher prediction and success rates than the selection of the same proportion of pixel within any part of the landslide masses. Morphometric features within a landslide mass have great diversity considering the depletion zone and the accumu-lation zone, the age and the activity of a landslide. In general, slopes are higher towards the main scarp and gentle along the foot of the landslide. The type of landslide, the first-time failure and the secondary movements, the strength of the material, vegetation and the climatic factors have great importance on the present morphology of landslides. The entire landslide polygon selection elimi-nates some of these inconsistences that were mentioned above in landslide susceptibility modelling. The results suggest that the substantial accuracy and the reliability of the landslide susceptibility zonation is sufficient to be used in land use planning for public safety and early phases of site selection studies for engineering structures.
hazard assessment and risk management. Natural Hazards, 37(3), 315–329.https://doi.org/10.1007/s11069-005-5182-6. Bourenane, H., Guettouche, M. S., Bouhadad, Y., & Braham, M.
(2016). Landslide hazard mapping in the Constantine city, Northeast Algeria using frequency ratio, weighting factor, logistic regression, weights of evidence, and analytical hierarchy process methods. Arabian Journal of Geosciences, 9(2), 154.
https://doi.org/10.1007/S12517-015-2222-8.
Can, T., Duman, T. Y., Olgun, S., Corekcioglu, S., Karakaya-Gulmez, F., Elmacı, H., Hamzacebi, S., & Emre, O. (2013). Landslide database of Turkey. In Proceedings of the TMMOB GIS congress, November 2013, Ankara. [In Turkish].
Can, T., Nefeslioglu, H. A., Gokceoglu, C., Sonmez, H., & Duman, T. Y. (2005). Susceptibility assessments of shallow earthflows triggered by heavy rainfall at three catchments by logistic regression analyses. Geomorphology, 72(1–4), 250–271.https:// doi.org/10.1016/j.geomorph.2005.05.011.
Chacon, J., Irigaray, C., Fernandez, T., & El Hamdouni, R. (2006). Engineering geology maps: Landslides and geographical infor-mation systems. Bulletin of Engineering Geology and the Environment, 65(4), 341–411. https://doi.org/10.1007/s10064-006-0064-z.
Chen, W., Pourghasemi, H. R., Kornejady, A., & Zhang, N. (2017). Landslide spatial modeling: Introducing new ensembles of ANN, MaxEnt, and SVM machine learning techniques. Geoderma, 305, 314–327.https://doi.org/10.1016/j.geoderma.2017.06.020. Chung, C. J., & Fabbri, A. G. (2008). Predicting landslides for risk
analysis—spatial models tested by a cross-validation technique. Geomorphology, 94(3–4), 438–452. https://doi.org/10.1016/j. geomorph.2006.12.036.
Clerici, A., Perego, S., Tellini, C., & Vescovi, P. (2006). A GIS-based automated procedure for landslide susceptibility mapping by the conditional analysis method: The Baganza valley case study (Italian Northern Apennines). Environmental Geology, 50(7), 941–961.https://doi.org/10.1007/s00254-006-0264-7.
Corominas, J., van Westen, C., Frattini, P., Cascini, L., Malet, J. P., Fotopoulou, S., et al. (2014). Recommendations for the quan-titative analysis of landslide risk. Bulletin of Engineering Geology and the Environment, 73(2), 209–263.https://doi.org/ 10.1007/s10064-013-0538-8.
Dai, F. C., Deng, J. H., Tham, L. G., Law, K. T., & Lee, C. F. (2004). A large landslide in Zigui County, Three Gorges area. Canadian Geotechnical Journal, 41(6), 1233–1240. https://doi.org/10. 1139/T04-049.
Dai, F. C., & Lee, C. F. (2003). A spatiotemporal probabilistic modelling of storm-induced shallow landsliding using aerial photographs and logistic regression. Earth Surface Processes and Landforms, 28(5), 527–545.https://doi.org/10.1002/esp.456. Upper slopes - .755 0.077 96.622 0.470 Midslope ridges - .547 0.083 43.292 0.578 Deep canyons - 1.509 0.096 246.367 0.221 Mean slope - .147 0.012 157.121 0.863 Slope/aspect ratio - .015 0.002 58.865 0.985 Roughness 1.958 0.121 259.918 7.087 Curvature - .172 0.029 36.114 0.842 Second derivative slope - .036 0.003 115.644 0.964 Linear aspect - .003 0.000 300.991 0.997 Bare - .592 0.036 267.798 0.553 Bare-agriculture - .378 0.040 89.456 0.685 Uncultivated farrow - .128 0.037 11.992 0.880 Forest - .544 0.043 158.552 0.580 Kadirli Formation 1.852 0.044 1794.722 6.376 Alluvium - 3.267 0.582 31.525 0.038 Constant 4.446 1.263 12.394 85.275
Dai, F. C., Lee, C. F., & Ngai, Y. Y. (2002). Landslide risk assessment and management: An overview. Engineering Geol-ogy, 64(1), 65–87. https://doi.org/10.1016/S0013-7952(01)00093-X.
Donati, L., & Turrini, M. C. (2002). An objective method to rank, the importance of the factors predisposing to landslides with the GIS methodology: Application to an area of the Apennines, (Val-nerina; Perugia, Italy). Engineering Geology, 63(3–4), 277–289.
https://doi.org/10.1016/S0013-7952(01)00087-4.
Duman, T. Y., Can, T., & Emre, O. (2011). 1:1.500.000 Scaled Landslide inventory map of Turkey. General Directorate of Mineral Research and Exploration Special Publication Series-27, Ankara.
Duman, T. Y., Can, T., Emre, O., Kecer, M., Dogan, A., Ates, S., et al. (2005). Landslide inventory of northwestern Anatolia, Turkey. Engineering Geology, 77(1–2), 99–114.https://doi.org/ 10.1016/j.engeo.2004.08.005.
Duman, T. Y., Can, T., Gokceoglu, C., Nefeslioglu, H. A., & Sonmez, H. (2006). Application of logistic regression for landslide susceptibility zoning of Cekmece Area, Istanbul, Turkey. Environmental Geology, 51(2), 241–256. https://doi.org/10. 1007/s00254-006-0322-1.
Duman, T. Y., Can, T., Olgun, S., Nefeslioglu, H. A., Durmaz, S., Hamzacebi, S., Elmaci, H., & Corekcioglu, S. (2009). 1: 500.000 Scaled Turkish landslide inventory map. Hatay Quadrangle, General Directorate of Mineral Research and Exploration Special Publication Series-24, 23 S. Ankara. [In Turkish]. Evans, I. S. (1972). General geomorphometry, derivatives of altitude,
and descriptive statistics. In R. J. Chorley (Ed.), Spatial analysis in geomorphology (pp. 17–90). New York, NY: Harper & Row. Fell, R., Corominas, J., Bonnard, C., Cascini, L., Leroi, E., Savage, W. Z., et al. (2008a). Guidelines for landslide susceptibility, hazard and risk zoning for land-use planning commentary. Engineering Geology, 102(3–4), 99–111. https://doi.org/10. 1016/j.enggeo.2008.03.014.
Fell, R., Cororninas, J., Bonnard, C., Cascini, L., Leroi, E., Savage, W. Z., et al. (2008b). Guidelines for landslide susceptibility, hazard and risk-zoning for land use planning. Engineering Geology, 102(3–4), 85–98. https://doi.org/10.1016/j.enggeo. 2008.03.022.
Fu, P., & Rich, P. M. (1999). Design and implementation of the Solar Analyst: an ArcView extension for modeling solar radiation at landscape scales. Proceedings of the 19th Annual ESRI User Conference, San Diego, USA. Available fromhttp://www.esri. com/library/userconf/proc99/proceed/papers/pap867/p867.htm. Goetz, J. N., Brenning, A., Petschko, H., & Leopold, P. (2015).
Evaluating machine learning and statistical prediction techniques for landslide susceptibility modeling. Computers & Geosciences, 81, 1–11.https://doi.org/10.1016/j.cageo.2015.04.007.
Gorum, T., Gonencgil, B., Gokceoglu, C., & Nefeslioglu, H. A. (2008). Implementation of reconstructed geomorphologic units in landslide susceptibility mapping: The Melen Gorge (NW Turkey). Natural Hazards, 46(3), 323–351. https://doi.org/10. 1007/s11069-007-9190-6.
Guzzetti, F., Carrara, A., Cardinali, M., & Reichenbach, P. (1999). Landslide hazard evaluation: A review of current techniques and their application in a multi-scale study, Central Italy. Geomor-phology, 31(1–4), 181–216. https://doi.org/10.1016/S0169-555x(99)00078-1.
Guzzetti, F., Mondini, A. C., Cardinali, M., Fiorucci, F., Santangelo, M., & Chang, K. T. (2012). Landslide inventory maps: New tools for an old problem. Earth-Science Reviews, 112(1–2), 42–66.https://doi.org/10.1016/j.earscirev.2012.02.001. Hosmer, D. W., Lemeshow, S., & Sturdivant, R. X. (2013). Applied
logistic regression (Vol. 3)., Wiley series in probability and statistics Hoboken, NJ: Wiley.
Hungr, O. (2016). A review of landslide hazard and risk assessment methodology. In S. Aversa et al. (Eds.), Landslides and engineered slopes. Experience, theory and practice(pp. 3–27). Hussin, H. Y., Zumpano, V., Reichenbach, P., Sterlacchini, S., Micu,
M., van Westen, C., et al. (2016). Different landslide sampling strategies in a grid-based bi-variate statistical susceptibility model. Geomorphology, 253, 508–523.https://doi.org/10.1016/j. geomorph.2015.10.030.
Kozlu, H. (1997). Tectono-stratigraphic units of the Neogene basins Iskenderun, Misis–Andırın and their tectonic evolution in the Eastern Mediterranean region. Unpublished Ph.D. thesis, Natural Science Institute, Cukurova University, Adana, Turkey, p. 189 [in Turkish].
McCune, B., & Keon, D. (2002). Equations for potential annual direct incident radiation and heat load. Journal of Vegetation Science, 13(4), 603–606. https://doi.org/10.1111/j.1654-1103.2002. tb02087.x.
Moore, I. D., & Burch, G. J. (1986a). Physical basis of the length-slope factor in the universal soil loss equation. Soil Sciences Society America Journal, 50, 1294–1298.
Moore, I. D., & Burch, G. J. (1986b). Modeling erosion and deposition: Topographic efects. Transactions ASAE, 29, 1624–1640.
Moore, I. D., Gessler, P. E., Nielsen, G. A., & Peterson, G. A. (1993). Soil attribute prediction using terrain analysis. Soil Science Society of America Journal, 57, 443–452.
Moore, I. D., Grayson, R. B., & Landson, A. R. (1991). Digital terrain modelling: A review of hydrological, geomorphological, and biological applications. Hydrological Processes, 5, 3–30. Nefeslioglu, H. A., Gokceoglu, C., & Sonmez, H. (2008). An
assessment on the use of logistic regression and artificial neural networks with different sampling strategies for the preparation of landslide susceptibility maps. Engineering Geology, 97(3–4), 171–191.https://doi.org/10.1016/j.enggeo.2008.01.004. Nefeslioglu, H. A., San, B. T., Gokceoglu, C., & Duman, T. Y.
(2012). An assessment on the use of Terra ASTER L3A data in landslide susceptibility mapping. International Journal of Applied Earth Observation and Geoinformation, 14(1), 40–60.
https://doi.org/10.1016/j.jag.2011.08.005.
Pike, R. J., & Wilson, S. E. (1971). Elevation relief ratio, hypsometric integral, and geomorphic area altitude analysis. Geological Society of America Bulletin, 82, 1079–1084.
Pourghasemi, H. R., Mohammady, M., & Pradhan, B. (2012). Landslide susceptibility mapping using index of entropy and conditional probability models in GIS: Safarood Basin, Iran. Catena, 97, 71–84.https://doi.org/10.1016/j.catena.2012.05.005. Pradhan, B., & Lee, S. (2010). Delineation of landslide hazard areas on Penang Island, Malaysia, by using frequency ratio, logistic regression, and artificial neural network models. Environmental Earth Sciences, 60(5), 1037–1054. https://doi.org/10.1007/ s12665-009-0245-8.
Regmi, N. R., Giardino, J. R., McDonald, E. V., & Vitek, J. D. (2014). A comparison of logistic regression-based models of suscepti-bility to landslides in western Colorado, USA. Landslides, 11(2), 247–262.https://doi.org/10.1007/s10346-012-0380-2.
Riley, S. J., Degloria, S. D., & Elliot, R. (1999). A terrain ruggedness index that quantifies topographic heterogeneity. Intermountain Journal of Sciences, 5(1–4).
Roberts, D. W., & Cooper, S. V. (1989). Concepts and techniques of vegetation mapping. In D. Ferguson, P. Morgan, & F. D. Johnson (Eds.), Land classifications based on vegetation: Applications for resource management (pp. 90–96). Ogden, UT: USDA Forest Service GTR INT-257.
Rossi, M., & Reichenbach, P. (2016). LAND-SE: A software for statistically based landslide susceptibility zonation, version 1.0. Geoscientific Model Development, 9, 3533–3543.
Montanarella, L., & Nadim, F. (2012). Statistical modelling of Europe-wide landslide susceptibility using limited landslide inventory data. Landslides, 9(3), 357–369. https://doi.org/10. 1007/s10346-011-0299-z.
landslide susceptibility mapping by conditional probability and artificial neural networks. Environmental Earth Sciences, 60(3), 505–519.https://doi.org/10.1007/s12665-009-0191-5.
