1 Introduction

A natural disaster can be defined as some rapid, instantaneous, or profound impact of the natural environment upon the socio-economic system (Alexander 1993). In the last decades, there has been a dramatic increase in loss of lives and properties, as well as injuries and damages to the environment, as a result of natural disasters. In addition to increase in population, rapid and unconscious urbanization has also played an important role in these losses. Turkey, a country ranked among the most affected by natural disasters worldwide, has frequently witnessed the dreadful consequences of natural disasters such as earthquakes, landslides, floods, avalanches, and so on. Of these, landslides are the second destructive natural disaster in the country and constitute approximately 30 % of the whole damage. Therefore, it becomes necessary to assess landslides in order to mitigate landslide consequences by preparing landslide inventory, susceptibility, and hazard or risk maps. Landslide inventory and susceptibility maps, in particular, are of great importance since they provide important information for decision makers and planners and constitute the basic information for hazard and risk maps. In addition, effective utilization of these maps can considerably reduce damages and losses (Ercanoglu 2005). Although the hazard and risk concepts are very important, it is not usually possible to obtain required data for landslide hazard or risk assessment parameters. Thus, in this study, landslide susceptibility assessment is preferred in a landslide-prone area in Turkey.

Landslide susceptibility is defined as a quantitative or qualitative assessment of the classification, volume (or area), and spatial distribution of landslides that exist or potentially may occur in an area. Although it is expected that landsliding will occur more frequently in the most susceptible areas, in the susceptibility analysis, time frame is explicitly not taken into account (Fell et al. 2008). In other words, landslide susceptibility analysis involves developing an inventory of landslides that have occurred in the past together with an assessment of the frequency (annual probability) of the occurrence of landslides (Cascini 2008). In order to assess landslide susceptibility, there are qualitative and quantitative methods such as heuristic analysis (geomorphic analysis and qualitative map combination), knowledge-based analysis (based on a learning process), statistical analysis (discriminant analysis, factor analysis, logistic regression, and so on), and deterministic analysis (classical slope stability theory and factor of safety approach) (Soeters and Van Westen 1996; Fell et al. 2008). In a different study, Ayalew et al. (2005) categorized these methods into three groups such as semi-qualitative (simple ranking and rating and analytical hierarchy process), quantitative (bivariate, multivariate statistical analyses, neural networks and fuzzy approaches), and hybrid methods (index based and training based). To add another viewpoint, recent developments in geographic information system (GIS), remote sensing, and computer technologies allowed users to perform landslide susceptibility assessments in more feasible ways. Nowadays, practically all research on landslide susceptibility and hazard mapping makes use of digital tools for handling spatial data such as GIS, global positioning systems (GPS), and remote sensing (Van Westen et al. 2008). Thus, utilization of the quantitative methods has enormously increased and has become much more common than the qualitative ones. As for the landslide susceptibility assessment parameters (i.e., the causal or conditioning factors), users have a multitude of options to produce different input data layers based on geological, topographical, land use, and geomorphological features of the area concerned. Regardless of the assessment method, the input data for landslide susceptibility assessment should be selected on a basis of causes of actual and past instabilities. However, analysis of cause–effect relationship is not always simple, as a landslide is seldom linked to a single cause (Aleotti and Chowdhury 1999). Therefore, it becomes important to make a selection of the significant input parameters in landslide susceptibility assessments and modeling to obtain more reliable results. This selection depends usually upon experience, size of the area, time, scale, landslide type, methodology to be applied, project budget, data availability, and reliability (Glade and Crozier 2005). The crucial stage of this process is the selection of reliable data from what may be available (Aleotti and Chowdhury 1999).

In this study, taking into consideration the above-mentioned issues, the major objectives for selecting parameters and assessing landslide susceptibility were twofold: firstly, selecting or ordering the effects of input parameters on landslide occurrences and secondly, evaluating the landslide susceptibility by Analytical Hierarchy Process (AHP) based on the first stage. In order to achieve the first objective, the max–min (MM) method, one of the similarity methods in fuzzy logic (Ross 1995), was considered to represent the input parameter effects on landslide occurrences. A new index, namely landslide relation indicator (LRI), was proposed for the landslide susceptibility assessments. It depicts the parameter effect on landslide occurrences and allows users to order the parametric weights. In addition, the proposed index was adapted to AHP method. By doing so, it was thought that the subjectivity concept in AHP was alloyed by the proposed methodology in the present study.

2 Characteristics of the study area

The study area is located in the Western Black Sea region of Turkey (Fig. 1) and is known as one of the most landslide-prone areas in the country. It encompasses four 1/25,000 quadrangles and covers approximately 567 km2, located in the west of Karabuk City. The digital elevation model (DEM) was obtained from the 1/25,000 scale topographical maps provided from the General Command of Mapping of Turkey. The DEM of the study area, containing 1418975 pixels with 20 m × 20 m resolution, represents that the topographical elevations range from 95 m to 1,736 m. Slope angle values vary between 0° and 59°. In the region, typical Black Sea climate prevails, and annual average rainfall is 518.2 mm (http://tumas.meteoroloji.gov.tr). The biggest residential location in the study area is the Yenice District of Karabuk City. There are also small scattered villages such as Döngeller, Kadıköy, Akmanlar, Keyfaller, and Güney. The main stream is the Yenice River, which passes through the study area approximately in its east–west direction. There are also Kelemen, Simsir, Pazarlı, and Güney Rivers, which form a dentritic drainage system in the study area.

Fig. 1
figure 1

Location map of the study area

Full size image

There are eight different lithological units in the study area, ranging from Precambrian to Quaternary ages (Fig. 2). The oldest lithological units are metagranitoid, marble, and granite of Precambrian age. These units are covered by massive Jurassic limestones outcropped at highly steep slopes (Tuysuz et al. 2004). Jurassic limestones are overlaid by Lower Cretaceous conglomerates and sandstone–mudstone–limestone alternations. Upper Cretaceous age unit, known as Ulus formation, comprises of sandstone, claystone, and siltstone and represents turbiditic flysch character. This unit is very dominant in the Black Sea region, and its areal extent starts from south of Yenice and goes through in NW direction toward the Eastern Black Sea region. It is highly susceptible to weathering and has a weak structure. There is no landslide evidence in the resistant part of this unit (i.e., rock materials), while the non-resistant soil parts (i.e., weathering products of rocks) are highly prone to landslide occurrences, particularly in gentle slopes (Ercanoglu 2005). Quaternary alluvial deposits (Qal and Qy) are the youngest units and are generally observed along the beds of the rivers and talus of slopes, respectively. Structural elements in the study area were obtained from the digital map prepared by MTA (2002) in vector format. However, no data exist regarding the type of structural elements in this map whether they are faults or folds.

Fig. 2
figure 2

Geological map of the study area and mapped landslides

Full size image

To prepare landslide inventory map, at the first stage, landslide locations were determined by air photograph interpretations. Then, during the field studies, these locations were verified and mapped. In addition, landslide inventories of the previous works carried out in some parts of the study area (e.g., Ercanoglu and Gökçeoglu 2002; Ercanoglu et al. 2004; Can et al. 2005) were also confirmed and updated during the field studies to fulfill the actual landslide inventory. Totally, 101 landslide locations covering approximately 8.6 % of the study area were mapped based on these studies. Of these, 86 landslides were classified as rotational earth slides, while 15 landslides were classified as earth flow based on the landslide classification system proposed by Cruden and Varnes (1996) (Fig. 3). The largest landslide in the study area covers an area of 1.94 km2, and the smallest one encompasses 7,655 m2. Based on the information gathered from the local people, most of the landslides occurred after heavy rainfall and snowmelt events in different periods in the region. However, it was not possible to obtain much reliable data for each landslide from the records or people. But, it could be concluded that the main triggering factor of the landslide events was completely a result of the meteorological peaks.

Fig. 3
figure 3

Some views from the landslides: a rotational slide; b earth flow; and c typical hummocky topography for rotational slides

Full size image

3 Methodology

The first step in every landslide assessment includes the collection of available information and data related to the area concerned. Based on the purpose of mapping (i.e., susceptibility, hazard, or risk), these data can be subdivided into four groups such as landslide inventory data, environmental (causal) factors, triggering factors, and elements at risk (Van Westen et al. 2008). Of these, landslide inventory and environmental factors such as DEM, slope angle, aspect, lithology, land use, faults, and hydrological features are essential to landslide susceptibility mapping. However, there are no universal guidelines for selecting the parameters that influence landslides in susceptibility mapping (Ayalew et al. 2005). Thus, establishment of a link between causal parameters and landslides becomes a very important and a problematic issue.

The users have a lot of options to model landslide susceptibility, taking into consideration different data layers. The number of data layers in landslide susceptibility mapping may range from only a few parameters (e.g., Corominas et al. 2003; Moreiras 2005) to several parameters (Yesilnacar and Topal 2005 (17 parameters); Meusburger and Alewell 2008 (15 parameters)). A detailed study regarding the considered causal parameters in landslide susceptibility modeling was carried out by Hasekiogullari (2010). It was based on 114 scientific studies that were indexed in science citation index (SCI) and published in the period between 2000 and 2010 in different scientific journals. One of the major goals of that study was to evaluate the parametric preferences of the researchers when preparing landslide susceptibility maps. Based on the results of the study carried out by Hasekiogullari (2010), slope angle was the most commonly preferred causal parameter among the scientists. The other considered parameters and their utilization frequencies used in these studies are shown in Fig. 4. The question herein comes to mind why slope angle, lithology, aspect, topographical elevation, curvature, land use, and so on are much more preferred than the other parameters among the scientists. Of course, there is a theoretical background when selecting these parameters. For example, slope angle is the most important parameter in gravitational movements, while slope aspect might reflect differences in soil moisture and vegetation. Similarly, lithology controls the type of movement, and land-use/land-cover characteristics are one of the main components in stability analyses (Van Westen et al. 2008). In addition to the theoretical relationships with experience and data availability, there are also technical reasons to use some other parameters. For example, if there is a module, let us say drainage density, in the GIS code used in a study, the user may want to use this feature as an input parameter in the landslide susceptibility analyses since the user may think a logical relationship between the drainage density and landslides. The development of a clear hierarchical methodology in landslide assessments is necessary to obtain an acceptable cost-benefit ratio and to ensure the practical applicability of the assessment (Soeters and Van Westen 1996). The scale parameter is also important for these studies and selecting landslide-related parameters. The working scale for a slope instability analysis is determined by the requirements of the user for whom the survey is executed. Although the selection of the scale of analysis is usually determined by the intended application of mapping results, the choice of a mapping technique remains open. This choice depends upon type of problem and availability of data, financial resources, and time for the investigation, as well as the professional experience of those involved in the survey (Soeters and Van Westen 1996). In this study, 1/25,000 scale maps such as topographical and land-use maps and their derivatives were used, based on the data at hand and data availability. Another aspect is that there are no fixed or standard value of weight and rating for landslide-related features. Furthermore, the determination of the weight and rating values are very important to landslide susceptibility analyses (Lee et al. 2004). This could be done by several methods such as statistical analyses (e.g., Ercanoglu et al. 2004; Ayalew and Yamagishi 2005; Guzzetti et al. 2008; Thiery et al. 2007; Nefeslioglu et al. 2008; Bai et al. 2009; Nandi and Shakoor 2009; Poudyal et al. 2010; Pradhan and Lee 2010), artificial intelligence techniques (e.g., Ercanoglu and Gökçeoglu 2002, 2004; Ercanoglu 2005; Ermini et al. 2005; Gomez and Kavzoglu 2005; Yesilnacar and Topal 2005; Kanungo et al. 2006; Pradhan et al. 2009; Yılmaz 2010), and expert opinion approaches (e.g., Barredo et al. 2000; Abella and Van Westen 2008; Ercanoglu et al. 2008; Ruff and Czurda 2008). The users have also a lot of options to produce landslide susceptibility maps using different methodologies such as artificial intelligence techniques (fuzzy logic, artificial neural networks, neurofuzzy), statistical methods (logistic regression, multivariate statistics, etc.), and combination and/or comparison of these methods. For example, some recent applications for artificial intelligence techniques can be found in Pradhan (2010a, 2011), Pradhan and Buchroithner (2010), Bui et al. (2011), Oh and Pradhan (2011), Sezer et al. (2011), and Akgun et al. (2012). For statistical applications and comparative studies, the details can be found in Yılmaz (2009, 2010), Pradhan (2010b, c), Pradhan and Pirasteh (2010), Pradhan et al. (2010a, b), Ercanoglu and Temiz (2011), and Akgun (2012).

Fig. 4
figure 4

Distribution of the causal parameters considered in landslide susceptibility studies

Full size image

This study attempted to evaluate landslide susceptibility by AHP, which can be considered in the expert opinion approaches. Generally, following steps are essential to perform AHP: (1) to break a complex unstructured problem down into its components; (2) to arrange the factors in a hierarchic order; (3) to assign numerical values to subjective judgment on the relative importance of each factor; and (4) to analyze the judgments to determine priorities to be assigned to these factors (Saaty and Vargas 2001). Perhaps, the most discussed step in this approach is the third step since it contains subjectivity with its general form. Although subjectivity is not necessarily bad when it is based on an expert opinion (Van Westen 2000), a new index, namely LRI, was adapted to overcome this problematic issue with this study. It is based on the “max–min” (MM) method, one of the similarity methods in fuzzy logic (Ross 1995). Similar to AHP, the MM method works on establishing relationships between elements of two or more data sets and operates on pairs of the data points. Although the name sounds similar to the fuzzy “max–min composition” method (Ercanoglu and Gökçeoglu 2004), it is different from the fuzzy composition operation. The MM method is expressed in the following equation:

$$ r_{\text{ij}} = \frac{{\sum\nolimits_{k = 1}^{\text{m}} {{\text{min(}}x_{\text{ik}} ,x_{\text{jk}} )} }}{{\sum\nolimits_{k = 1}^{\text{m}} {{\text{max(}}x_{\text{ik}} ,x_{\text{jk}} )} }}\quad i, \, j = 1,2, \ldots ,n $$
(1)

where r ij is the similarity relation value, x ik and x jk are the members of the different data sets, and min and max are the minimum and the maximum values of the two data sets, respectively. Close values of r ij to 1 indicate the similarity of the two data sets, while close values of r ij to 0 show dissimilarity. The proposed LRI is based on the calculation of two different r ij values, depicting the parameter effect on landslide occurrences by considering the parameters and landslide locations. It also allows users to order parameters based on the calculated LRI values, particularly for selecting the effective parameter combinations in the landslide susceptibility analyses. It is expressed in the following equation:

$$ {\text{LRI}} = \sum {r_{\text{ij(PR)}} \times r_{\text{ij(LS)}} } $$
(2)

where r ij(PR) is the total calculated r ij values of all the considered parameters with each other, r ij(LS) is the r ij value of any parameter with landslide locations, and LRI is the landslide relation indicator.

In order to calculate the LRI values for each parameter, 13 input parameter maps were prepared from different data sources (Table 1). As it is clear from Table 1, lithology parameter was not considered in the analyses. The main reason was that the mapped landslides were only located in the Upper Cretaceous flysch unit in the study area (see Fig. 2). Those parameters were almost compatible with the “top parameters” given in Fig. 4, frequently used in landslide susceptibility analyses. However, it should be noted that these parameters were gathered by the data availability and possibilities at hand in this study. Then, all vector-type parametric data were converted to raster files including the landslide inventory. Frequency ratio (FR) model proposed by Lee and Talib (2005) was used to associate these parameter maps with landslide locations. FR is the ratio of where landslides occurred in the total study area and also is the ratio of the probability of a landslide occurrence to a non-occurrence for a given attribute (Lee and Talib 2005). FR for each parameter was calculated in the GIS environment, considering randomly selected 70 % of overall data for modeling and 30 % for validation stages, respectively.

Table 1 Considered parameters in the analyses and their sources
Full size table

The calculated FR values were tabulated in Table 2. Also, these values were normalized (NFR) to express landslide susceptibility in [0, 1] interval, and the NFR values were assigned to the input parameter maps (Fig. 5). The next stage was to evaluate the LRI values of the input parameters. Thus, all input parameter and landslide inventory maps (including 0 for non-landslide and 1 for landslide locations) were converted to the data files (“*.dat”). To perform these analyses and to calculate LRI values for each parameter, a computer code, “LRI. bas”, was written using the Q-Basic programming language. The work-flow diagram was given in Fig. 6. In order to calculate LRI value, two r ij values such as r ij(PR) and r ij(LS) are needed, as stated in Eq. 2. The first one (i.e., r ij(PR)) defines the landslide susceptibility relationship between the parameter pairs, while the second one (i.e., r ij(LS)) reflects the relationship between the considered parameter and landslide locations. This was needed because of the fact that two parameters could be compatible with each other but could not be consistent with the landslide locations, or vice versa. Thus, r ij(PR) and r ij(LS) values were calculated separately using “LRI.bas” and are given in Table 3. As can be seen from Table 3, the highest LRI value was obtained from slope angle (SLP) parameter as 2.716, and its order was the first. This means that the slope angle (SLP) and distance to streams (DTS) are the first two effective parameters on landslide occurrences in the study area, while the distance to ridges (DTRI) parameter has the lowest effect based on the LRI values. Thus, evaluation of the relative importance of the considered parameters with respect to the LRI values allowed selecting in which order the considered parameters contributed to landslide occurrences objectively.

Table 2 Distribution of parameter subgroups in landslide locations and in the study area
Full size table
Fig. 5
figure 5

Considered input parameters: a slope; b aspect; c plan curvature; d topographical elevation; e NDVI; f land use; g distance to streams; h distance to roads; i distance to structural elements; j distance to ridges; k stream power index; l sediment transport capacity index; and m wetness index

Full size image
Fig. 6
figure 6

Flowchart of the “LRI.bas” computer code

Full size image
Table 3 LRI, r ij(PR), and r ij(LS) values of the considered parameters
Full size table

In order to produce landslide susceptibility maps, AHP method was chosen since it considers parameter pairs similar to the LRI approach explained above. It is a multicriteria decision-making process using the relative importance of the parameters contributing to the event to produce parameter weights and evaluates the consistency of pairwise comparison parameter (Barredo et al. 2000). To date, AHP was successfully applied in many studies (e.g., Ayalew et al. 2005; Komac 2006; Yalçin and Bulut 2007; Akgün et al. 2008; Ercanoglu et al. 2008; Yalçin 2008; Wang et al. 2009; Akgün and Türk 2010; Huang et al. 2010). Generally, an expert rates the considered parameters based on a scale proposed by Saaty (1977) (Table 4). Pairwise importance of the parameters for an event is assessed by an expert, based on the values given in Table 4. These values are placed in an “n × n” matrix (n: the number of parameters) in the form of rows-to-columns order. To perform AHP analyses, WEIGHT and Multi-Criteria/Multi-Objective Decision Wizard (MCE) modules of Idrisi Taiga were used. The first module develops a set of relative weights for a group of parameters in a multicriteria evaluation. The weights generated by this module are produced by means of the principal eigenvector of the pairwise comparison matrix (Eastman 2009). This module also calculates the consistency ratio (CR), which shows the considered weights consistent or not. CR values less than 0.1 indicate a good consistency, while CR value exceeding 0.1 shows an inconsistent matrix and should be reevaluated. The MCE module provides a process in which multiple layers are aggregated to yield a single output map. It has three different procedures such as Boolean intersection, order weighted average (OWA), and weighted linear combination (WLC). Of these, the WLC was chosen to produce landslide susceptibility maps. It is a procedure of multiplying each factor by its factor weight and adding the results (Eastman 2009).

Table 4 Scale of relative importance between parameter pairs suggested by Saaty (1977)
Full size table

In this study, instead of using an expert opinion to fulfill the parametric pairwise matrices, we made use of the calculated r ij(PR) and LRI values. This procedure allowed using a data-driven approach when assigning the parametric pairwise values. In order to achieve this, r ij(PR) values ranging in [0, 1] interval were adapted to the scale provided by Saaty (1977) (Table 5). The main idea herein is to build a logical connection between r ij(PR) and relative importance values. This is realized by considering the relationship between the two parameter pairs with low r ij(PR) values (e.g., 0 ≤ r ij(PR) ≤ 0.2) should have the highest intensity of importance (e.g., 9 or 1/9 in the AHP scale) since one of the parameters dominates the landslide process and has absolute effect on landslide occurrence than the other one. Similarly, if the two parameters have almost the same effect (i.e., together contributing or not) on landslide occurrence or non-occurrence, r ij(PR) value between the two parameters must be high and has almost equal importance on the event (e.g., 1 in the AHP scale).

Table 5 r ij(PR) values corresponding the relative importance values in AHP
Full size table

Given the above-mentioned issues, intensity of importance values among the parameter pairs were filled out in the matrices with the corresponding r ij(PR) values in Table 5. An example of a completed matrix with nine input parameters (selected by the orders given in Table 3) is given in Table 6. It should be noted that the intensity of importance values may be fractional or integer (ranging from 1/9 to 9) based on the location of parameter either in row or in column. For example, if the r ij(PR) value was calculated as 0.725 between the SLP and ASP parameters, the intensity of importance value would be 3 or 1/3 based on the values tabulated in Table 5. In this condition, the user could make a decision according to the LRI values of the considered parameters (LRISLP: 2.716; LRIASP: 1.663). If the SLP is in the row location and the ASP is in the column location (as in the case in Table 6), the importance value will be 3 since the SLP has higher effect on landslide occurrence due to its higher LRI value. In contrast, if the ASP had been located in the row and the SLP had been located in the column of the matrix, the importance value would have been 1/3 (reciprocal value in the matrix) because of LRISLP: 2.716 > LRIASP: 1.663. Eventually, taking into consideration these assignments, a total of 11 matrices starting from first 3 parameters (SLP, DTS, and NDVI) with the highest LRI values to 13 parameter combinations (i.e., all parameters) were prepared. In other words, there were 11 different landslide susceptibility maps to be produced. Of these, three landslide susceptibility maps were shown in Fig. 7 in order to save space. Based on the CR values, except for 4 parameter combinations of 10, 11, 12, and 13, the CR values of 3–9 parameter combinations were obtained below 0.1. In other words, the parameter combinations ranging from 3 to 9 indicated the compatibility, while the others were not acceptable because of their CR values (Fig. 8). Thus, the maps produced by 10–13 parameters were not taken into consideration for the validation process. In addition, calculated weights for each parameter combinations were also shown in Fig. 8.

Table 6 Comparison matrix for the nine parameters
Full size table
Fig. 7
figure 7

Landslide susceptibility maps produced by: a 3 parameters; b 6 parameters; and c 9 parameters

Full size image
Fig. 8
figure 8

CR values’ parametric weights of the AHP models

Full size image

4 Results and validation

In the present study, landslide susceptibility was assessed in a landslide-prone area covering approximately 567 km2 in the Western Black Sea region of Turkey. A total of 101 landslides were mapped in the study area. In order to assess landslide susceptibility, basically, AHP model was considered. As mentioned before, 13 different parameter maps were used to produce landslide susceptibility maps. A total of 11 landslide susceptibility maps were produced for the study area using different parameter combinations, based on LRI concept and AHP. Of these, four landslide susceptibility maps were omitted because of the limitation related to CR values explained in the previous chapter. For these susceptibility maps, different susceptibility levels were obtained. In order to express spatial distribution or variability of landslide susceptibility, there are several ways such as quantiles, natural breaks, equal intervals, and standard deviations (Ayalew and Yamagishi 2005; Pradhan and Lee 2010). In this study, equal area division approach was followed to represent spatial variations in landslide susceptibility values varying from “very low” to “very high”, dividing the [0, 1] interval by five classes with a unit of 0.2 increment. For all considered landslide susceptibility maps, spatial distributions of susceptibility levels are given in Table 7.

Table 7 Areal coverages of landslide susceptibility for the maps and AUC values of the AHP models
Full size table

In order to validate the produced landslide susceptibility maps, Relative Operating Characteristics (ROC) module of Idrisi Taiga was applied using the validation data. This module evaluates the validity of a model that predicts the location of the occurrence of a class by comparing a suitability image depicting the likelihood of that class (i.e., the input image) and a Boolean image showing where that class actually exists (i.e., the reference image) (Eastman 2009). The module calculates the area under curve (AUC) value with the defined threshold (selected as 100 in the analyses). The value of 1 indicates that there is a perfect spatial agreement between the reference image and the input image, while the value of 0.5 is the agreement that would be expected due to chance. The ROC module also produces the true-positive (TP) % and false-positive (FP) % values for each threshold that constitute the curve from which the AUC is calculated. Based on the module results (Table 7), the best performance was gathered from the 9-parameter-combination landslide susceptibility map with the AUC value of 79.7 % (Fig. 9) using the validation data.

Fig. 9
figure 9

Performance of the landslide susceptibility map produced by 9 parameters

Full size image

5 Conclusions and discussion

AHP model is conventionally based on a rating system provided by expert opinion. In fact, expert opinion is very useful in solving complex problems like landslides. However, to some extent, opinions may change for every individual expert and thus may be subjected to cognitive limitations with uncertainty and subjectivity. Another aspect is that data-driven methods are also powerful in landslide susceptibility mapping and contain less subjectivity. In order to minimize these drawbacks both in selecting parameters and in applying methodology, LRI, a new index, was proposed in this study. The LRI is applicable in any landslide susceptibility assessment method and allows users to order parametric importance before the landslide susceptibility analyses application. It is based on two similarity relation values depicting parametric relationships (by parametric pairwises) on landslide occurrences and landslide locations individually (by each parameter and landslide locations). The first one defines the landslide susceptibility relationships between the parameter pairs, while the second one reflects the relationship between the parameters and landslide locations. The main idea in the proposed approach is to reflect the parameter effects on landslide occurrences, considering the landslide system as a whole. In other words, two considered parameters could be compatible with each other on landslide locations but could not be consistent with landslide locations, or vice versa. For example, LUS (land use) parameter has the ninth order with respect to LRI values in the whole system. However, it is in the third order if only the landslide locations were considered. This means that LUS parameter is compatible with the mapped landslide locations, while it is not in accordance with the other parameters contributing the landslide system since most of the landslides occur only in agricultural lands (83.78 % of the landslide locations). Based on the LRI calculations, the parameter effects on landslide occurrences in order were the slope angle, distance to streams, NDVI, topographical elevation, plan curvature, aspect, stream power index, wetness index, land use, distance to structural elements, sediment transport capacity index, distance to roads, and distance to ridges, respectively. Taking into consideration this parametric order, different parameter combinations ranging from 3 to 13, a total of 11 AHP assessments were performed. Based on the CR values, 7 landslide susceptibility maps were produced since CR values of 10-11-12-13 parameter combinations were greater than 0.1. Of these, landslide susceptibility map produced by 9 parameter combinations showed the best performance with AUC value of 0.797 and was considered as satisfactory.

The difficulty in obtaining data and the issue of selecting independent data to analyze, that is, the parameters that are thought to be causally related, remain challenges in implementing more data-based approaches (Akgun 2012). Although the AHP method is fundamentally based on expert opinion, it is thought that the selection of parameters on landslide occurrences by LRI approach alloys the subjectivity concept in this method. Furthermore, LRI approach can be used in any landslide susceptibility assessment whatever the methodology is conducted. However, it should be noted that the reliability of the results is directly affected by the landslide location data, that is, the landslide inventory map. Research results will be helpful in developing new regulations on territory protection and facilities as well as land management purposes for local administrations, decision makers, and planners.