1 Introduction

Landslides are controlled by mechanical laws that can be determined empirically, statistically or in deterministic methods. Conditions that cause landslides (instability factors), or are directly or indirectly linked to slope failures, can be collected and used to build the predictive models of landslide occurrence (Dietrich et al. 1995). Landslide susceptibility is the likelihood of a landslide occurring (Dewitte et al. 2010) in an area on the basis of the local terrain conditions. It is the degree to which a terrain can be affected by slope movements, i.e. an estimate of ‘where’ landslides are likely to occur in the future. In mathematical language, landslide susceptibility is the probability of spatial occurrence of slope failures, given a set of geo-environmental conditions. Landslide susceptibility assessment has become a major subject for authorities responsible for regional land use planning and environmental protection. More generally, landslide susceptibility consists in the assessment of what has happened in the past, and landslide hazard evaluation consists of the prediction of what will happen in the future (Guzzetti 2005). The landslide susceptibility map is based on the distribution of past landslides, slope steepness, type of soil/bedrock, structure, hydrology and other pertinent data (Nandi and Shakoor 2009). Many different methods and techniques for evaluating landslide susceptibility have been proposed and tested (Akgun et al. 2008; Atkinson and Massari 2011; Brenning 2005; Clerici et al. 2006; Dahal et al. 2008; Dewitte et al. 2010; Ercanoglu et al. 2008; Ghosh and Carranza 2010; Gorsevski and Jankowski 2010; Jaiswal et al. 2010; Lee 2005; Lee and Dan 2005; Lee et al. 2004, 2007; Lee and Pradhan 2006, 2007; Nefeslioglu et al. 2008; Oh et al. 2009; Oh and Pradhan 2011; Ozdemir 2009; Pirasteh et al. 2009; Pistocchi et al. 2002; Pradhan et al. 2006, 2010a, b; Pradhan and Youssef 2010; Rossi et al. 2010; Saha et al. 2005a, b; Sezer et al. 2011; Sterlacchini et al. 2011; van Den Eeckhaut et al. 2009, 2010; van Westen 2004; Yeon et al. 2010; Yilmaz 2009, 2010; Youssef et al. 2009; Zhou et al. 2002). General overviews of research into landslide susceptibility can be found in the work of Leroi (1996), Aleotti and Chowdhury (1999), Guzzetti et al. (1999), Dai et al. (2001), Van Westen (2004) and Van Westen et al. (2008).

The reliability of landslide susceptibility maps depends mostly on the amount and quality of available data, the working scale and the selection of the appropriate methodology of analysis and modelling (Ayalew and Yamagishi 2005). Van Westen (2004) gives an overview of recent developments in the use of Geographical Information Systems and Earth Observation, which have been applied for improved landslide inventory mapping, landslide susceptibility and hazard assessment, elements at risk mapping and finally landslide vulnerability and risk assessment (Van Westen 2004).

Earthquakes and landslides are the most serious geological hazards in Turkey, when comparing losses resulting from all natural hazards. During the period between 1950 and 2005, landslides constitute 33% of all evacuations of settlements due to natural disasters (Duman et al. 2009). The main objective of this paper was to produce a landslide susceptibility map for the Sultan Mountains in Turkey, using a Weights of Evidence (WOE) method with landslide and landslide causative factor databases developed with Geographical Information System (GIS). In the study area, landslides occur frequently due to climatologic, geomorphologic and geologic conditions. In this region, especially the south-western part of Turkey, frequent landslides often result in significant damage to people and property. On 13 July 1995, in Senirkent, which is located 50 km away from the study area in the west, 74 lives were lost, numerous houses were destroyed and property damage resulted from a debris flows caused by prolonged and heavy rainfall. Therefore, to mitigate any damage arising from landslides, it is necessary to assess scientifically susceptible areas.

This manuscript investigates landslide susceptibility and the effect of landslide-related factors at Sultan Mountains, using the WOE and the GIS. Nineteen important causative factors for landslides were selected, and corresponding thematic data layers were prepared in GIS. Input data were collected from the topographic maps and field study investigations or observations. Numerical weight for different categories of these factors was determined based on a statistical approach and then integrated into the GIS environment to arrive at a landslide susceptibility map of the area. The landslide susceptibility map classifies the area into four classes of landslide susceptible zones, i.e. high, moderate, low and very low. Assessing landslide susceptibility with limited background information and data is a constant challenge for engineers, geologists, planners, landowners, developers, insurance companies and government entities (Nandi and Shakoor 2009; Aleotti and Chowdhury 1999).

2 The study area

The study area is located in the Central Anatolia Region of Turkey, in the north-western portion of the province of Konya (Fig. 1). The study area encompasses the Sultan Mountains and its surroundings. It is located in the west of Akşehir in Central Turkey. The investigated region has a total area of about 373.112 km2 (932780 grid) and extends from latitude 4,229,000 to 4,258,000 m North and from longitude 349,000 to 369,000 m East. The location map of the study area is given in Fig. 1. Study area boundaries were determined so as to encompass settlement centres and the major roads.

Fig. 1
figure 1

Location map of the study area

Full size image

The region exhibits uneven mountainous topographical features. The geological, geomorphological and hydrogeological environment of the study area is favourable to landslide activity. Several landslides have occurred in the past; there are also numerous recorded instances of active landslides in the study area. Landslides in the study area terrain have varying controlling factors, such as lithology and slope, that seem to be the main controlling factors in causing slope instability.

3 Materials and methods

Numerous landslide risk factors have been used to define susceptibility maps. According to the Wu and Sidle (1995), these factors can be grouped into two types: (1) the intrinsic factors that contribute to landslides, such as topography, geology and hydrogeology, and (2) the extrinsic variables that tend to trigger landslides, such as intense rainfall, earthquakes and landscape modifications (e.g. urbanization, road construction and mining). In this study, the WOE analysis method (Dahal et al. 2008; Deng 2010; Mathew et al. 2007; Neuhäuser and Terhorst 2007; Oh and Lee 2010; Regmi et al. 2010; Zhu and Wang 2009) and GIS techniques were used to extract an evaluation of landslide susceptibility in the study area. In this study, 19 main categories of landslide affecting factors are selected and defined. These categories are geology, relative permeability, land use/land cover, precipitation, elevation, slope, aspect, total curvature, plan curvature, profile curvature, wetness index, stream power index, sediment transport capacity index, attitude, distance to drainage, distance to fault, drainage density, fault density and spring density map. Each category is subdivided into different classes by its value or features. The maps of geology and tectonic data, as well as relative permeability, land use/land cover and landslide inventory data, are vector maps, and these maps are converted to raster maps with grid size of 20 × 20 m for compilation in the database. However, other maps used in this study are derived from the DEM in raster format with a grid size of 20 × 20 m. For all data layers used in the model, a spatial resolution of 20 m by 20 m was chosen (Atkinson and Massari 2011). It is often inadvisable to use a spatial resolution finer than 20 m because published maps and field data are commonly more generalized, especially when carrying out studies on a regional scale. Using generalized data on a fine grid may induce false relations into the data modelling. The landslide characteristics and parameters that are influential for landslide susceptibility (slope stability), and considered in the present study, are described below.

3.1 The landslide characteristics

Landslide inventory mapping is the systematic mapping of existing landslides in a region using different techniques such as field survey, air photo interpretation and literature search for historical landslide records (Jaiswal et al. 2010). A landslide inventory map provides the spatial distribution of the locations of existing landslides (Nandi and Shakoor 2009). In the study area, the slopes are subjected to small spread instability and small landslides developed in previous years. The shallow landslides in the study area occur in the soil’s overlying bedrock. A landslide inventory map of the study area was constructed/completed in the year 2009 by the field investigation consisted of a geologic and geomorphologic surveys at a 1:25,000 scale map. The typical landslide morphology of visible scarps and hummocky topography can be recognized in the field (Fig. 2).

Fig. 2
figure 2

Examples of landslides on slopes in the study area; a, b, c complex shallow and deep-seated landslides, d a rotational slide along the Akşehir-Cankurtaran road, e soil flow, f debris flow (the white dashed line marks the scarp)

Full size image

In the landslide inventory map, 90 landslides (covering 7537 grid) were identified in the study area and are shown as green- and red-coloured polygons in Fig. 3. They covered an area of 3.004 km2, accounting for 0.808% of the study area. The minimum, mean and the maximum landslide areas are 0.0027, 0.0334 and 0.1692 km2, respectively (Table 1). Slides and flows are the two main types of mass movements that occur most often in the study area. Rotational or complex landslides are the most common types of slides. The second common type of failure is earth and debris flow in highly weathered alteration zones. Most of the slides that occur in the study area are deep-seated (depth of failure surface >5 m). In the landslide inventory map, 80 of the landslides (out of 90 landslides) were identified as within the training landslide area and were shown as green-coloured polygons in Fig. 3. They covered an area of 2.7612 km2, accounting for 0.74% of the study area. The minimum, mean and the maximum training landslide areas are 0.0028, 0.034 and 0.169 km2, respectively. In the remaining test area (in control area), 10 landslides were identified as within the test landslide area (Pradhan et al. 2010a, b) and are represented by red-coloured polygons in Fig. 3. The minimum, mean and the maximum test landslide areas are 0.0028, 0.02644 and 0.0828 km2, respectively. They covered an area of 0.2536 km2, accounting for 0.068% of the total study area. A digitized map of landslide boundaries was produced using ArcGIS 9.3. A vector-to-raster conversion was performed to provide a raster layer of the landslide areas. The training landslide area was used in analyses, and the results of the analyses were then implemented over the entire study area, including the test area. The landslides in the test area were used solely for verification of the analysis results.

Fig. 3
figure 3

Digital elevation model (DEM) and location map of observed training (green polygons) and test landslides (red polygons) in the study area

Full size image
Table 1 Type and scarp area of landslides (number with asterisks symbol as 12* displays test landslide)
Full size table

3.2 Landslide affecting factors

3.2.1 Geological and tectonic factors

A field survey of exposed rocks within the study area was undertaken to establish the different lithological, geological and structural features of the study area. The study area geologically consists of sedimentary and metamorphic rock units. The stratigraphic sequence of the study area is ranging in age from Cambrian to Holocene (Fig. 4). The geology and tectonic of the Sultan Mountain and its vicinity have been studied by many researchers, including Atalay (1975), Demirkol (1982) and Boray et al. (1985). Based on the studies by Demirkol (1982), the geological units are divided into fourteen classes. These units/formations, from bottom to top, are Çaltepe, Seydişehir, Engilli, Harlak, Kocakızıl, Deresenek, Hacialabaz, Bağkonak and old and young alluvium. The Bağkonak formation and alluvium units cover all the old units discordantly. These units are described below and presented in Fig. 4. In this study, the geology of the Akşehir vicinity has been described by Demirkol (1982). The Çaltepe formation (Cac) and its member, the Çaltepe limestone member (Cacl), are the oldest units in the study area. The Early–Middle Cambrian-aged Çaltepe formation consists of sedimentary units, such as whitish grey and black layers containing crushed limestone and dolomite. These units are observed in the southern parts of the study area. Karstic forms, such as karens and dolines, were observed at many places in this formation. The Çaltepe limestone member is composed of nodular limestone. This member is located in the upper part of Çaltepe formation. The Çaltepe formation is thrust into the Seydişehir formation. The Seydişehir Formation (CaOs), dating back to the late Cambrian or early Ordovician era, overlies and conforms to the Çaltepe formation. The formation consists of an alternating sequence of metaconglomerate, metasandstone and chlorite/serisite/quartz schist. Thin layers of light green and yellow schist overlie metaconglomerate and metasandstone bed. This formation is conformably overlain by the Engilli formation. The Seydişehir formation is the most widespread unit of the study area. The Engilli formation (De), from the Middle Ordovician period, unconformably overlies the Seydişehir formation. This formation consists of medium to thick, jointed, hard, purple, yellow and white quartz beddings. It is intercalated by thin greyish limestone. There are more than two sets of discontinuity in the rock mass of this formation. The thickness of this formation is approximately 220 m. These units, changing from thick bedding to massive formations, are compact and hard. Limestone member of Engilli formation (Del) is a Middle Ordovician-aged unit consisting of greyish, medium to thick bedding and jointed limestones. There are many coral fossils in these rocks. The thickness of the member is approximately 40 m. The Engilli formation gradually transitions into the overlying Harlak formation. Dating from the carboniferous-aged Harlak formation (Ch) consists of purple-green phyllite with thin schistosity, metasandstone and metaconglomerate. Chlorite and serisite is observed condensative. This formation gradually passes with Engilli quartzite. Kocakızıl formation (Cko) dates back to the Early Carboniferous period and is overlain unconformably by the Harlak formation. This formation consists of a brown and red base conglomerate, crystallized limestones, dolomites and dolomitise limestones. Lithologies of this formation are very hard, shared and have thick bedding. The Kartalkaya formation (Cka) is Early Carboniferous-aged and conformably overlies the Kocakızıl formation. It consists of grey-whitish coloured, very hard, shared and middle to thick bedding clayey limestone, marls and quartz. This formation is overlain unconformably by the Çay Group. The Çay Group (C) dates back to the Permian age and is represented by the Deresenek formation, the Deresenek limestone member and the Toprakkale formation in the vicinity of Akşehir. The outcrops of the Toprakkale formation are not observed in the study area. This group unconformably overlies the Kartalkaya formation. The Deresenek formation (CPd) is Permian-aged and consists of metamorphic units such as marble, crystallized limestone, dolomite and chalk schist. Rocks are generally schistose, yellow and brown coloured, very hard and crushed and small folding. The Limestone member of the Deresenek formation (CPdl) consists of marl, limestone and recrystallized limestone. This member is transitional to the underlying Deresenek formation and is overlain by the Hacialabaz limestone. The Hacilabaz formation (Jh) unconformably overlies the Palaeozoic base. This formation begins in conglomerate and continues up to limestone and dolomite alterations. Rocks in this formation are dark greyish-black coloured, ranging from thin to thick bedding. The thickness of this unit ranges from 150 to 300 m. Karstification is widespread in the Early Jurassic-aged Hacialabaz formation. The Bağkonak formation (Tb), dating back to the Tertiary age, consists of loosely cemented sandstone, siltstone and greyish-yellowish coloured, thick bedding mudstone. Old alluvial (Qalo) and Young alluvial (Qaly) deposits, dating to the Quaternary age, overlay the old formations, is observed in the eastern region of the investigation site. These deposits are mainly composed of clay, silt and sand with local gravel and boulder. The Bağkonak formation and alluviums cover all the old units discordantly. The landslide area and outcropping area of these units, observed in the study region, are given in Fig. 4 and Table 2.

Fig. 4
figure 4

Geological map of the study area (modified from Demirkol 1982)

Full size image
Table 2 Class and landslide area, and computed weights and contrast values for landslide affecting factors
Full size table

Important folding is formed in the study area as a result of Alpine and Caledonian orogeny. Strike slips and dip slip faults are related to hard and breakable limestones, in which joint systems are generally well developed. The folding axes trends, in harmony with the direction of the Sultan Mountains, have NW–SE strikes. The Cankurtaran nape is located at the south part of the study area, and due to this nape, the Late Cambrian–Ordovician-aged Seydişehir formation is covered by the Early–Middle Cambrian-aged Çaltepe formation. The Aksehir Active Fault in the Sultan Mountains is the main border fault on the SW side of the Akşehir-Afyon Graben, with a dip slip rate of 0.3 mm/year (Kocyigit and Ozacar 2003). The length of the Aksehir fault is about 80 km (Demirkol 1982). The main fault in the study area is the Akşehir Fault, which extends in a NW–SE direction. Other faults are observed at the south and south-eastern parts of Engilli.

3.2.2 Relative permeability

In this study, a relative permeability map is also considered a conditioning factor of the landslide. A relative permeability map is prepared for the permeability of lithologies of formations which are exposed in the study area (Fig. 5a). In the study area, all units constitute differentiated hydrogeological complexes. The Çaltepe limestone member, Engilli, Deresenek limestone member and Hacialabaz formations, which consist of crushed and karstic limestone, dolomite and crushed quartzite lithologies, are classified as ‘very permeable’ unit. A minimum of three joint sets is characteristic for most outcroppings in these units. Most joint systems are closely spaced and crosscut each other, suggesting that they are well connected in a three-dimensional network. Quaternary-aged old and new alluvial deposits (mainly composed of gravels, sands and clay-silts, silty sand and silty gravel), Çaltepe formations (consisting of limestone and dolomite), Kocakizil formations (consisting of conglomerates, crystallized limestones, dolomites and dolomitise limestones) and Deresenek formations (consisting of metamorphic units such as marble, crystallize limestone, dolomite and chalk schist) are classified as ‘permeable’ unit. The Kartalkaya and Bağkonak formations consist of marl, clayey limestone, crystallize limestones, sandstone, siltstone and mudstones and are classified as ‘low permeability’ unit. Finally, the Seydişehir and Harlak formations consist of foliated metamorphic rocks such as metaconglomerate, metasandstone, chlorite–serisite schist and phyllite and are classified as ‘impervious’ unit (Fig. 5a). Mainly, the central parts of the study area have very permeable units (13.376 km2), while permeable (148.035 km2) and low permeable units (48.378 km2) are dominant in the north-eastern and central parts of the study area. The southern and centre part of the study area have impermeable units (164.08 km2). Most of landslide areas are observed in the permeable and impermeable relative permeability classes (Table 2).

Fig. 5
figure 5

a Relative permeability, b land use/land cover, c precipitation and d attitude map of the study area

Full size image

3.2.3 Land use/land cover

Land use and land cover influence slope behaviour, and they play important roles in the stability of slopes. The land use/land cover map of the studied area is taken from the District Directorate of the Konya-Turkey Forest. According to this map, there are more than 30 land cover/land use types in this area. For the purpose of this study, the land use map is classified as ‘soil and earthen land’, ‘forest’, ‘agriculture’, ‘residential area’ and ‘pasture land’. Most of the study area is covered by forest and agriculture land (Fig. 5b). The residential zone, which occupies the smallest area, is predominant at lower elevations in the north-eastern part of the study area. It is clear that the landslide area is very low (0.0045 km2) in pasture land areas, whereas landslide area reaches 1.985 km2 in forest areas. Many of landslides are in the forests and earthen land area (Table 2).

3.2.4 Precipitation

Precipitate data from the meteorological stations (Çay, Sultan Mountains, Kumdanlı, Yalvaç, Bağkonak, Gelendost and Şarkikaraağaç) located in the vicinity of the study area were used to develop a precipitation map for the study area. The average annual precipitations recorded at the Çay, Sultan Mountains, Akşehir, Kumdanlı, Yalvaç, Bağkonak, Gelendost and Şarkikaraağaç Meteorology Stations were 531, 492, 591, 442, 508, 485, 577 and 468 mm, respectively. The typical annual precipitation in the region is about 442–591 mm, although a higher amount of localized precipitation is common and is known as ‘lake effect precipitation’. The presence of lakes in the region affects the weather conditions. An annual rainfall map of the study area was prepared by the inverse distance-weighted (IDW) method using the annual precipitation values measured at the meteorology stations and surrounding area, which are given above. Then, it is converted to raster format and classified into the five classes, to obtain a classified annual precipitation map (Fig. 5c and Table 2). The resolution of the precipitation map is 20 × 20 m. Precipitation and landslides are directly related; as precipitation increases, the number of landslides also increases. Landslides are abundant in areas in which there is high precipitation.

3.2.5 DEM and derivatives

The 1/25,000 scaled topographic map is used to generate the Digital Elevation Model (DEM). The contour interval of this map is 10 m. A DEM with a 20 × 20 m grid size has been created using the elevation contours. The DEM is used as input to derive a thematic map of the elevation, slope, aspect, curvature, plan curvature, profile curvature, topographic wetness index, stream power index and sediment transport capacity index (Pradhan et al. 2010a, b).

The elevation map is derived from the DEM. The altitude in the area ranges from 960 to 2,221 m above sea level. The dominant altitude ranges are 960–1,100 m and 1,600–2,000 m. Most landslides occur in the elevations between 1,100 and 1,600 m and especially elevations between 1,100 and 1,300 m. There are no or very few landslides in the elevations between 1,600 and 1,700 m (Table 2).

The slope angle in the study area is characterized by low and high angles. The original slope angle values range from 0° to 48°. The slopes with angles lower than 5° cover most of the investigation area. Slope values were reclassified into the ten categories with interval of 5°. The slope analyses showed that landslide area increases gradually along with an increase in slope angle until the range of 25–30° and then decreases beyond that range (Table 2).

Mainly, the slope aspect is related to the general physiographic trend of the area. The general physiographic trend of the area is NW–SE, and an important number of the landslides observed in the studied area have NE and NW failure directions. The slope aspect is classified into eight categories, with each category being 45°. North-east, east and south-east-facing slopes contain the largest number of landslides, with 83% (2.234 km2) of the training landslide area. Westerly facing slopes (SW, W and NW) make up the smallest proportion of the research area with 26.6% (99.178 km2) of the total area and 14.5% (0.387 km2) of training landslides. Slope aspect analyses showed that landslides were most abundant in the range 292.5–157.5° facing slopes (Table 2).

Curvature attributes are based on second derivatives: the rate of change of a first derivative such as slope or aspect, usually in a particular direction. The three curvatures most frequently computed are total, plan and profile curvature (Wilson and Gallant 2000). In this study, the total curvature values, ranging −5.589 to 7.51, were classified into five zones. The dominant total curvature value range is −2.5 to 2.5. Curvature values between −2.5 and 2.5 comprise 98.5% (367.637 km2) of the research area and contain 98.4% (2.7189 km2) of the training landslide area (Table 2). The plan curvature values, ranging from −3.77 to 3.01, were classified into seven classes to obtain a classified plan curvature map. Plan curvature values between −1 and 2 make up 98.7% (368.381 km2) of the research area and contain 99.5% (2.7486 km2) of the training landslide area (Table 2). Finally, the profile curvature values, ranging from −5.34 to 4.133, were classified into six classes to obtain a classified profile curvature map. Profile curvature values between −1 and 0 comprise 98.9% (369.112 km2) of the research area and contain 98.7% (2.727 km2) of all training landslide area (Table 2).

The topographical wetness index (TWI) is the ratio between the specific catchment area and the slope (Wilson and Gallant 2000). The Wetness index values, in the wetness map, range from 2 to 22. The values of the wetness index are divided into five classes. Relationships between the wetness index and landslide occurrence indicated that landslides generally occur at the wetness index range 2–8. As the wetness index increases, the landslides decrease (Table 2).

The erosion processes can be considered to be one of the main conditioning factors of landslide occurrence (Duman et al. 2006). The Stream Power Index (SPI) of the secondary topographical attributes was used in this study to reflect the erosive power of the terrain (Wilson and Gallant 2000), and it is a measure of the erosive power of the flowing water based on the assumption that the discharge (q) is proportional to the specific catchment area (A s ). It predicts the net erosion in the areas of profile convexity and tangential concavity (flow acceleration and convergence zones) and the net deposition in the areas of profile concavity (zones of decreasing flow velocity; Wilson and Gallant 2000). The specific catchment area (A s ) was calculated from DEM using TAUDEM software (Tarboton 2002). It is determined that the index values range from 0 to 831,070 in the prepared SPI map of the study area. In this study, due to the increase in visual quality, log (SPI) is calculated using SPI values. Then, the log (SPI) values were classified into five zones. The stream power index analyses showed that the index value in a range of 1–4 indicates the abundance of landslide occurrence, whereas low log numbers (SPI) (in a range 0–1) and high log (SPI) (in a range 4–5) indicate few landslide occurrences (Table 2).

The sediment transport capacity index (LS) was derived from unit stream power theory and is equivalent to the length–slope factor in the Revised Universal Soil Loss Equation in certain circumstances (Wilson and Gallant 2000). This equation is sometimes used to predict the locations of net erosion and net deposition areas. Erosion indices can be computed directly from the slope and drainage area attributes calculated in TAS software (Lindsay 2005). A prepared Sediment Transport Capacity Index (LS) map of study area is shown that the values of LS range from 0 to 932. The sediment transport capacity map of the study area was divided into five categories. Relationships between LS and landslide occurrence indicated that landslides generally occur at the LS range 31–66 (Table 2).

3.2.6 Attitude map

Geometric relationship between rock bed and slope can influence the condition of groundwater flow and hence the process of landslides (Cruden 1989). The structural setting is also usually considered strongly affecting the slope stability, with a different degree of land sliding strictly connected to the different relationship between bedding attitude and slope aspect and angle (Clerici et al. 2002). Planar land sliding may occur easily if the topographic aspect (βt) and dip direction of the bedding (βb) lie in approximately the same direction (i.e. βt−βb ≤ [(20°)2]1/2 and the dip angle of bedding (αb) is less than the topographic slope angle (αt). In landslide susceptibility assessments, attitude in terms of landslide activity is taken into consideration a conditioning factor. In the whole study area, 250 dip direction values (from 0° to 360°) and dip angle values (from 0° to 90°) for bedding measured in the field were collected. Dip direction and dip angle maps of discontinuities (bedding and schistosity) are formed using the values of dip direction and dip angle, respectively. Then, an area in which the difference between the discontinuity dip direction angle and topographic aspect angle is less than ±20° and the dipping angle of discontinuity is less than the topographic slope angle is determined. This situation is named Y (class 1) and another situation is named N (class 2). Hence, the attitude map is constituted from two values Y and N (Fig. 5d). An area in which planar sliding may easily occur is shown as Y (1). The sum of the area in which the discontinuity dip angle is less than the slope angle and in which the difference in angle between the discontinuity dip direction and slope aspect is less than 20° is 1.2% (4.597 km2) of the total studied area (Table 2). Even so, in the situation mentioned above, it is determined that the landslide area is only about %10 of the training landslide area. It shows that land sliding depends on the discontinuous position, and this factor may cause about 10% of total training landslides.

3.2.7 Proximity maps

Proximity of local lineament features (i.e. faults and streams) was used to investigate any cause–effect relationships between lineament and landslide occurrence. In this study, drainage lines are produced from DEM using TAUDEM software (Tarboton 2002). Distance to drainage lines (streams or rivers) are also main causes of landslides (Sezer et al. 2011). Our analysis buffer map of all drainage lines was prepared in GIS showing a buffer width of 250 m from these drainage lines. The values of distance on the drainage map are divided into five classes (Table 2). The landslide distribution in these buffer zones is shown in Table 2. It is evident that the majority of the landslides are located in the drainage buffer zone of 50–150 m (Table 2).

Distance from faults and the thrust faults are also main causative reasons for landslides (Melchiorre et al. 2010) in that the presence of these tectonic structures breaks the rock mass, reducing its strength (Donati and Turrini 2002; Pradhan et al. 2010a, b). The Active Akşehir Fault is located in the north-eastern part of the investigated area. Besides this, many small faults trending NW–SE and NE–SW are also present in the study area. Fault lines were derived from 1:25,000 scale geology map. For our analysis, a buffer map of all faults was prepared in GIS showing the buffer width to be 500 m from these faults. The values of distance to the fault are divided into ten classes with the 50-m intervals. The landslide distribution in these buffer zones is shown in Table 2. It is evident that the majority of the landslides are located in the 100- to 200-m zone. The analysis of landslides’ distance from fault shows that as the distance from the fault decreases and increases, landslides decrease (Table 2).

3.2.8 Density maps

A density map displays the mass, number, length and/or area within a given area or volume. Drainage, fault and spring density maps have been produced for the studied area. Drainage lines from DEM, length of faults from geological map and springs from topographical map of the study area are extracted. The spatial data of faults, drainage lines and springs are then integrated in GIS to determine the susceptibility of landslides for each density class. Drainage density (Melchiorre et al. 2010; Suzen and Doyuran 2004) was considered to be an influential parameter in the present study. The drainage lines are also used to calculate the kilometre square density of drainage lines in the whole study area. The values obtained a range from 0 to 4 m/km2 and are divided into four classes. It is evident that the majority of the landslides are located in the 0–1 m/km2 class of drainage density (Table 2).

The procedure to obtain the fault density map of the study area is as follows: (1) The faults in the study region are digitized from the geological map using ArcGIS 9.3 (ESRI 2008). (2) A polygon map is created in ArcGIS with circles with 560 m in radius. (3) The total length of faults in each circle is determined. Then, the value of total fault length per surface area in each circle is calculated and assigned to the attribute of each circle. (4) The map is converted into a raster map. Finally, the fault length per unit area is determined in the fault density map. The faults are extracted from the geological map, 1/25,000 scaled, of study area. The fault density values are found to vary between zero and 203 cm/km2. These values are divided into five classes with the 50 cm/km2 intervals. It is determined that all of the landslides are located in the 0–150 cm/km2 fault density zone (Table 2).

A spring density map displays the count of springs within a given area (in 1 km2). In this study, 440 springs are determined on the topographic map, 1/25,000 scaled, of the study area. Using a 20 × 20 m grid, the spring density values have been calculated and are found to vary between zero and 25 spring per km2. These values are divided into five classes with the five intervals. It is determined that the majority of landslides locate in the 0–1 (spring/km2) class of spring density (Table 2).

3.3 Weights of evidence method

Various approach methods have been used in assessing an area’s susceptibility to landslides. These methods include the logistic regression technique, weighting factors, weighted linear combination of instability factors, multivariate regression, information value, weights of evidence, frequency ratio approach, landside nominal risk factor, genetic algorithm, likelihood ratio, certainty factors and fuzzy sets to landslide susceptibility zonation.

The theory’s background and the algorithm of weights of evidence (WOE), as well as its application in landslide susceptibility assessment, have been presented in Lee et al. (2002), Van Westen et al. (2003), Lee and Choi (2004), Ermini et al. (2005), Wang and Sassa (2005), Chang and Chao (2006), Liu et al. (2006), Chang and Chien (2007), He and Beighley (2007), Neuhäuser and Terhorst (2007), Dahal et al. (2008), Barbieri and Cambuli (2009), Nandi and Shakoor (2009), Regmi et al. (2010), and Hapke and Plant (2010). The idea behind WOE is that several binary patterns can be combined to predict another binary pattern (Bonham-Carter 1994). The WOE is a statistical method based on Bayes’ theorem. WOE is a quantitative data-driven method used to combine data sets (Bonham-Carter 1994). It uses a log-linear form of the Bayesian probability model to estimate the relative importance of evidences by statistical means. This method was developed at the Canadian Geological Survey (Bonham-Carter et al. 1989) and applied to the mapping of mineral potential. In the Bayesian approach, prior and posterior probabilities are among the most important concepts. An advantage of Bayesian probabilistic modelling is the possibility of incorporating uncertainty into the susceptibility model and to explicitly consider expert knowledge, which often exists for the investigated area (Chung and Fabbri 1999).

Weights of evidence method are denoted mathematically by the Barbieri and Cambuli (2009) as below: ‘If a part A f of the studied area is affected by landslide phenomena, the prior probability of finding a landslide within A t (total studied area) is:

$$ P_{f} = \frac{{A_{f} }}{{A_{t} }} $$
(1)

This initial estimation can be then increased or decreased depending on the relationships between analysed factors and landslides. The probability of finding one of the factors examined in the study area is given by:

$$ P_{v} = \frac{{T_{v} }}{{A_{t} }} $$
(2)

where T v  = total area occupied by a certain class of a certain factor (e.g. slope class 20–35%).

For the whole territory surveyed, the probability of finding a landslide in the areas occupied by the n class of the j parameter is the ratio between the probability of finding a landslide inside the territory occupied by the n class of the j parameter and the probability of finding an area occupied by the n class of the j parameter in the whole territory:

$$ P\left\{ {A_{f} /T_{v} } \right\} = \frac{{P(T_{v} \cap A_{f} )}}{{P_{v} }} = \frac{{P\left\{ {T_{v} /A_{f} } \right\}}}{{P_{v} }} \times P_{f} $$
(3)

Similarly, the posterior probability of finding a landslide in the areas not occupied by the n class of the j parameter is:

$$ P\left\{ {A_{f} /T_{v} } \right\} = \frac{{P(T^{\prime}_{v} \cap A_{f} )}}{{P^{\prime}_{v} }} = \frac{{P\left\{ {T^{\prime}_{v} /A_{f} } \right\}}}{{P^{\prime}_{v} }} \times P_{f} $$
(4)

For mathematical reasons, probabilities can be expressed more conveniently as odds:

$$ O = \frac{P}{1 - P} $$
(5)

Then, Eqs. 3 and 4 can be expressed as follows, using the weights of evidence statistical method as noted by Barbieri and Cambuli (2009):

$$ O\left\{ {A_{f} /T_{v} } \right\} = \frac{{O\left\{ {T_{v} /A_{f} } \right\}}}{{O_{v} }} \times O_{f} $$
(6)
$$ O\left\{ {A_{f} /T^{\prime}_{v} } \right\} = \frac{{O\left\{ {T_{v} /A_{f} } \right\}}}{{O^{\prime}_{v} }} \times O_{f} $$
(7)

In the weights of evidence method, the natural logarithms are applied to both Eqs. 6 and 7 to obtain:

$$ \ln O_{T} \left\{ {A_{f} /T_{v} } \right\} = W_{v}^{ + } + \ln O_{f} $$
(8)
$$ \ln O_{T} \left\{ {A_{f} /T^{\prime}_{v} } \right\} = W_{v}^{ - } + \ln O_{f} $$
(9)

where W + is the positive weight to be assigned when the class n of the j parameter is present and W − is the negative weight to be assigned when the class n of the j parameter is absent.The weights are calculated by the following equations:

$$ W^{ + } = \ln \frac{{P\left\{ {T_{v} /A_{f} } \right\}}}{{P\left\{ {T_{v} /A^{\prime}_{f} } \right\}}}\; = \ln \left[ {\frac{{\frac{{T_{v} \cap A_{f} }}{{A_{f} }}}}{{\frac{{T_{v} \cap A^{\prime}_{f} }}{{A^{\prime}_{f} }}}}} \right] = \ln \left[ {\frac{{\frac{{{\text{Landslide}}\;{\text{area}}\;{\text{in}}\;{\text{the}}\;{\text{considered}}\;{\text{class}}}}{{{\text{Total}}\;{\text{landslide}}\;{\text{area}}}}}}{{\frac{{{\text{Stabil}}\;{\text{area}}\;{\text{in}}\;{\text{the}}\;{\text{considered}}\;{\text{class}}}}{{{\text{Total}}\;{\text{stable}}\;{\text{area}}}}}}} \right] $$
(10)
$$ W^{ - } = \ln \frac{{P\left\{ {T^{\prime}_{v} /A_{f} } \right\}}}{{P\left\{ {T^{\prime}_{v} /A^{\prime}_{f} } \right\}}}\; = \ln \left[ {\frac{{\frac{{T^{\prime}_{v} \cap A_{f} }}{{A_{f} }}}}{{\frac{{T^{\prime}_{v} \cap A^{\prime}_{f} }}{{A^{\prime}_{f} }}}}} \right] = \ln \left[ {\frac{{\frac{{{\text{Total}}\;{\text{landslide}}\;{\text{area}}\;{\text{in}}\;{\text{the}}\;{\text{other}}\;{\text{classes}}}}{{{\text{Total}}\;{\text{landslide}}\;{\text{area}}}}}}{{\frac{{{\text{Total}}\;{\text{stabil}}\;{\text{area}}\;{\text{in}}\;{\text{the}}\;{\text{other}}\;{\text{classes}}}}{{{\text{Total}}\;{\text{stable}}\;{\text{area}}}}}}} \right] $$
(11)

The W + positive weight is directly proportional to the influence that the n class of the j parameter has on landslide development. To analyse the influence of several parameters on the distribution of landslides in the area, the weights of each parameter are summarized, as these parameters are mutually statistically independent:

$$ \ln O_{T} \left\{ {A_{f} /T_{v1}^{k} \cap T_{v2}^{k} \cap T_{v3}^{k} \ldots \ldots \ldots \cap T_{vn}^{k} } \right\} = \sum\limits_{i = 1}^{n} {W_{vi}^{k} } + \ln O_{f} $$
(12)
$$ \ln P_{\text{finale}} = \sum {W + \ln O_{f} } $$
(13)
$$ P_{\text{finale}} = {\text{Exp}}\left( {\sum {W + \ln O_{f} } } \right) $$
(14)

where k can take the sign + or − depending on the presence or absence of this parameter.

The difference between the positive and negative weights, as computed for each class of each parameter analysed, is a good indicator of its relationship with landslides:

$$ C = W^{ + } - W^{ - } $$
(15)

The value of C is typically between 0 and 2; when the value of C tends to zero, the presence of the considered parameter does not affect the distribution of landslides in the area; conversely, when C is approximately two or more, the correlation is significant (Barbieri and Cambuli 2009). T is the sum of the product between posterior probability and the area, extended to all elementary cells in which the territory was divided: ideally, T should then be approximately equal to the total landslide area. In practical applications, it frequently happens that T is greater than the landslide area, because there are generally some conditional dependencies between the variables; if T does not exceed landslide area for more than 15%, conditional independence between the considered variables exists’. One of the key assumptions in the weights of evidence (WE) modelling is that the predictor patterns have to be conditionally independent. When this assumption is violated, WE posterior probability estimates are likely to be biased upwards (Deng 2009; Agterberg and Cheng 2002; Sterlacchini et al. 2011).

4 Application of weights of evidence method

In the analysis, scarp areas of landslides were taken in consideration (Dewitte et al. 2010; Gorsevski and Jankowski 2010). In this study, landslide scarp areas were first identified as polygons. Then, the polygons were converted to a raster map with 30 m * 30 m grid spacing. A total of 3,016 landslide points, derived from this raster map, were used in the analysis. For the purpose of this research, all of the landslides in the study area were divided into two groups: training landslides and test landslides (Ghosh and Carranza 2010; Pradhan et al. 2010a, b). Eighty landslides out of all landslides, located in the study area, were used as the ‘training landslides’, and the remaining landslides (10) situated in the same area were designated as the ‘test landslides.’ The term ‘all landslides’, as used in this paper, includes both the training landslides and the test landslides. In the landslide inventory map, 80 landslides (covering 6903 grid) were identified in the training area and are shown as green polygons in Fig. 3. In the remaining landslides, 10 landslides (covering 634 grids) were identified and are represented by red polygons in Fig. 3. The training landslides were used in analyses, and the results of the analyses were then implemented over the entire study area, including the test area. The landslides in the test area were used solely for the verification of the analysis results. The landslide susceptibility analysis was performed on the training area of the study area using the WOE method. In the analysis, landslide instability factors (geology, relative permeability, land use/land cover, elevation, slope gradient, slope aspect, curvature, plan curvature, profile curvature, topographic wetness index, stream power index, sediment transport capacity index, attitude, distance to drainage, distance to fault, drainage density, fault density and spring density) were considered to be variables responsible for landslides (Van Den Eeckhaut et al. 2010). First, the various thematic maps, which show factors that affect landslides, were overlapped with the landslide map. On the basis of these intersections, for each of the landslide-related factors, the weights, contrasts and weights of evidence probability values were calculated using Eqs. 10, 15 and 14 (Table 2, Fig. 6).

Fig. 6
figure 6

The procedure of the susceptibility assessment

Full size image

Then, the conditional independence was tested before the integration of the predictor patterns to map the landslide susceptibility. The chi-square values for testing the conditional independence between all pairs of binary patterns for each factor were calculated at the 95% significance level and 1 degree of freedom. Calculated chi-square values greater than the table values (7.9) suggest that the pairs are not significantly different (Table 3). Among 324 pairwise comparisons, 81 of the pairwise comparison couples were found independent of each other (Table 3). For example, the pairs of geology and elevation show a conditional dependence (χ2 = 1,459), but the pairs of geology and slope show conditional independence (χ2 = 1.41). This implies that geology and elevation could not be used together to map the landslide susceptibility, but geology and slope could be used together to map the landslide susceptibility. In the study, several models were derived from the independent factors examined. Generally, the slope angle is among the most important input factors, together with the lithology, which influence slope stability (Bednarik et al. 2010). Giving preference to geology and slope angle, which play an important role in the formation of landslides, and considering these together with other independent factors, four different models were generated. The independent factors of the designed models are given in Table 4. Landslide susceptibility map was prepared from each model. Then, the accuracy of each map was tested using the relative operating characteristics.

Table 3 Calculated chi-square values for testing the conditional independence between all pairs of binary patterns with each factor in the 95% significance level (the bold values are dependent)
Full size table
Table 4 Four possible model combinations of the independent factors based on the chi-square statistics
Full size table

In this study, landslide susceptibility map was prepared from each model. Accuracy of each model was evaluated by calculating relative operating characteristics (ROC) (Mathew et al. 2007; Oh and Pradhan 2011; Oh and Lee 2010; Rossi et al. 2010; Van Den Eeckhaut et al. 2010; Yilmaz 2009; Yeon et al. 2010). In the ROC method, the area under the ROC curve (AUC) values ranging from 0.5 to 1.0 are used to evaluate the accuracy of the model (Nandi and Shakoor 2009). In order to apply the ROC method to the study area, a concise and representative data set was prepared using randomly selected pixels from landslides (1,000 pixels) and non-landslides (1,000 pixels) locations in the investigated area. The prediction capability of each model is determined by the area under the curve (Fig. 7 and Table 4). The area value of under the ROC curve for Model 1 was found to be 0.920, with an estimated standard error of 0.006 (Fig. 7 and Table 5). The area under the ROC curve values for model 2, model 3 and model 4 were calculated as 0.916, 0.884 and 0.898, with estimated standard errors of 0.07, 0.008 and 0.008, respectively (Fig. 7). These results indicate that model 2, model 3 and model 4 are a poor estimator, whereas model 1 is a good estimator of landslide susceptibility in the study area. The WOE model showed 92% accuracy, indicating that it can be used for landslide susceptibility mapping. The model shows quite a reasonable performance with respect to similar studies carried out by other researchers (Oh and Lee 2010; Vahidnia et al. 2010; Lee 2007; Pradhan 2010; Pradhan and Lee 2010; Yilmaz 2010; Hapke and Plant 2010). Model 1 has four factors: geology, slope, profile curvature and spring density. Prepared landslide susceptibility map for model 1 is given in Fig. 8.

Fig. 7
figure 7

ROC curve evaluation for models

Full size image
Table 5 Accuracy assessment of the four models of susceptibility to landslides based on the area under the ROC curve
Full size table
Fig. 8
figure 8

Landslide susceptibility map of study area obtained from WOE analyses method

Full size image

5 Results and discussion

The relationship between landslide and landslide-related factors is as follows (Table 2, Fig. 9). The contrast is positive for a positive spatial association and negative for a negative spatial association (Lee and Choi 2004). The contrast is the rating of each class of each factor that influences landslide occurrence. The results of contrast value (Fig. 9 and Table 2) analyses revealed that the factors with the highest landslide probabilities are Deresenek formation limestone member and Harlak formations (Ch) of the lithological units, the class of permeable unit of relative permeability, the classes of forest and soiled with earthen land use or land cover, the classes of 560–600 mm of precipitation, the classes of 1,100–1,600 m of altitude, the classes of 20–50° of slope, the classes of 22.5–157.5° of slope aspect, the classes of 0–2.5 total curvature, the classes of 0–1 of plan curvature, the classes of −3 to −1 and 0–1 of profile curvature, the classes of 2–8 of wetness index, the classes of 2–3 of the logarithmic stream power index, the classes of 31–167 of sediment transport capacity index, the classes of 50–100 m of distance to drainage pattern, the classes of 100–200 m of distance to fault, the classes of 1–2 of drainage density, the class of 2–3 of fault density and the class of 0–1 spring density (Fig. 9).

Fig. 9
figure 9

Graph showing the variation in weight contrasts with classes of landslide affecting factors

Full size image

The results of contrast value analyses are shown that the factors with the highest slope stability probabilities are:Seydişehir formation (CaOS) and Engilli formation limestone member (Del) of the lithological units, the classes of high and low permeable unit of relative permeability, the class of pasture land cover, the class of 520–560 mm of precipitation, the classes of 1,000–1,100, and 1,600–1,700 m of altitude, the classes of 0–10° of slope, the classes of flat and 22.5–337.5° of slope aspect, the classes of −1 to 0 and 1–2 of plan curvature, the class of −1 to 0 of profile curvature, the classes of 10–22 of wetness index, the classes of 1–2 of the logarithmic stream power index, the classes of 0–12 and 167–392 of sediment transport capacity index, the classes of 0–50 m of distance to drainage pattern, the classes of 400–500 m of distance to fault, the classes of 2–3 of drainage density, the class of 1 of fault density and the class of 1–5 spring density (Fig. 9).

The four landslide effecting parameter layers (geology, slope, profile curvature and spring density map) of the entire study area are used to generate landslide susceptibility map. With the WOE method applied to the training data set, a landslide susceptibility zoning map (LSZ) is obtained as shown in Fig. 8. The value of each cell in the map represents landslide susceptibility values. In this map, maximum, minimum, mean and standard deviation of landslide susceptibility values are 0.499, 0, 0.038 and 0.0611, respectively. The range of landslide susceptibility is classified into four categories for general purposes. These four categories are very low (<= mean value (0.038) or 0 ~ 7.6%), low (mean value +1 standard deviation or 7.6~19.8%), moderate (mean value + 2 standard deviation or 19.8~64.10%) and high (>mean value + 2 standard deviation or 64.1~100%). The landslide susceptibility map shows that the areas of the very low, low, moderate and high landslide susceptibility classes are 223.979 km2 (60%), 85.858 km2 (23%), 29.864 km2 (8%) and 33.597 km2 (9%), respectively (Table 6, Fig. 10).

Table 6 The area and WOE probability values of landslide susceptibility classes obtained LSZ
Full size table
Fig. 10
figure 10

The landslide susceptibility values of landslide susceptibility classes and the percentage area in these classes

Full size image

The accuracy of the final LSZ map is evaluated based on the observed landslides (Mathew et al. 2007; Regmi et al. 2010). First, the final LSZ map is checked by overlaying with the observed landslide map. We consider a landslide as a ‘correct’ prediction when at least part of it situated in a high susceptibility zone. Otherwise, the prediction is considered ‘wrong.’ Based on the above criteria, the 80 landslides in the study area are predicted by the model as shown in Table 7 and Fig. 11. By overlaying the predicted LSZ map with the training data, the result indicates that 63 of the 80 observed training landslides are correctly predicted and 17 of the 80 observed training landslides are wrongly predicted, as shown in Table 7. On the other hand, Fig. 11 shows the LSZ map with the observed test landslides indicating the different levels of prediction. For the modelling validation, the constructed LSZ map is overlaid with the 10 observed test landslides. It follows that the nine test landslides are correctly predicted (90%), as shown in Fig. 11.

Table 7 Summary of the prediction accuracy of the landslide susceptibility map compared with the training and test data sets
Full size table
Fig. 11
figure 11

The prediction accuracy of the landslide susceptibility zone map compared with the training (black coloured bars) and test (red coloured bars) landslides

Full size image

Comparing the observed training landslides map with the landslide susceptibility map, 78.75% of training landslides fall into high susceptibility area and 18.75% of training landslides fall into medium landslide susceptibility zone (Table 8, Fig. 11). On the other hand, comparing observed test landslides map with the landslide susceptibility map, 90% of test landslides fall into a high susceptibility zone (Fig. 11). Also, 96% of all observed landslides fall into medium and high susceptibility zones. Based on the evaluation of the prediction accuracy of the observed training and test landslides, it can be remarked that the LSZ map, because most of the observed landslides are correctly predicted by its medium and high landslide susceptibility classes, can be considered highly believable.

Table 8 The distribution of the landslides with respect to the landslide susceptibility zones at the study area
Full size table

Landslide susceptibility is high in the Harlak and Deresenek formations, but it is low in the young alluvial, old alluvial, Bağkonak and Hacialabaz formations. The Harlak formation is composed mainly of argillaceous components, phyllites, shales and schists, but the Deresenek formation limestone member is mainly composed of marls and limestones. It was observed that phyllites and schists were very weathered and altered. In addition to weakening rocks, the presence of weak planes in phyllites and schists causes landslides. (Damm and Terhorst 2010). The phyllite, schist and marls have a low strength, and upon contact with water, they soften and their shear strengths become lower. This lower shear strength causes increased landslide susceptibility. On the other hand, the Harlak formation, stratigraphically, is located under the Deresenek formation. The infiltration by rainfall water of permeable limestones of the Deresenek formation decreases the shear strength of the unpermeable lithologies of the Harlak formation, such as phyllites and schists, and also forms and increases pore water pressure on these lithologies. In these areas, low shear strength increases landslide susceptibility. Field observations indicated that many failures occurred within the weathering zone or contact zone of the Harlak formation. The areas most prone to instability are located on slopes on the contact zone, between the Harlak and Deresenek formations. The infiltration of rainfall, formation of the saturated zone and rise in groundwater tables in the contact zone of the less-permeable bedrock (Harlak formation) increase the pore water pressure and reduce the shear strength of the lithologies (Harlak formation). Landslide susceptibility is low in the old alluvial, young alluvial and Hacialabaz formations (Figs. 8 and 9). This is the reason that the slope angle of the alluvial area is low, and the shear strength of slope material, which consists of the limestone in the Hacialabaz formation, is very high. In areas where the profile curvature values range from 0 to 3, the landslide susceptibility is higher. It is known that the negative profile curvature values define the concave areas where slope degrees are increasing and erosion is probable, but the positive values define the convex areas where slope degree is decreasing and depositing is probable. In the areas where the profile curvature values range between 0 and 1(in deposition areas), landslide susceptibility is higher. It is determined that landslide susceptibility is very low in the western parts of the investigation area (Fig. 8), where slope angles and annual precipitation are also very low. Landslide susceptibility is high in the zone where topographical elevation is between 1,100 and 1,500 m and slope angle is very high. Landslide susceptibility is lower in areas where topographical elevations are between 960–1,000 m and 2,000–2,200 m. The low topographical elevation areas have low slope angles and low shear strength lithologies, but lithologies in the high topographical elevation areas have high strength and high slope angles. Landslide susceptibility is quite low in the area with a low slope. Landslide occurrences have been observed in almost every aspect class, but the slopes facing the north-east—south-east (between 22.5° and 157.5° from north) seem to be relatively more susceptible to landslide occurrences.

6 Conclusions

In this study, a landslide susceptibility map has been constructed using the relationship between landslide locations and causative factors. In the Sultan Mountain area (Akşehir), the relationships between landslides and various factors were investigated using the weights of evidence method. Then, 19 combinations of the factors were derived. For this purpose, landslide inventory, geology, relative permeability, land use/land cover, elevation, slope gradient, slope aspect, curvature, plan curvature, profile curvature, topographic wetness index, stream power index, sediment transport capacity index, attitude, distance to drainage, distance to fault, drainage density, fault density and spring density layers were used. For the analysis, calculated contrasts were overlaid to calculate the weights of evidence probability values. Then, these values were mapped, and the results (Mathew et al. 2007; Regmi et al. 2010) were validated by the correlation between the observed training-control landslide location and the results. The resulting landslide susceptibility map matches the actual landslides well, with about 92.2% of the identified landslides in the investigation area occurring in high susceptibility areas. When compared with the prepared landslide susceptibility map, all of the landslides identified in the study area are found to be located in the high (92.2%) and moderate (7.8%) susceptible zones. This assessment is also supported by the performance analysis. As a consequence, the landslide susceptibility map can be accepted as valuable performance for the planning purposes.

Based on the findings of this study, the classes of rocks most sensitive to landslide in the study area are the Harlak formation, Deresenek limestone member and Deresenek formations. The described landslides are typically located in the lithologies, including the high permeable layers and impermeable layers, such as claystone, siltstone and mudstone layers. The areas of the high permeable, permeable and impermeable classes are the most sensitive classes to landslide occurrence, but the areas of low permeability are not susceptible to landslides. Considering this finding, it is possible to say that one of the main conditioning factors of the landslides in the study area is lithology. The classes of land use/land cover most sensitive to landslide are earthen and forest areas. Areas of high precipitation are very conducive to landslide occurrence. Landslides do not occur in the very low (<1,000 m) and high (>1,700 m) altitude classes. Many of landslides occurred between 1,100–1,600 m in altitude. The presence of lineaments greatly reduces the strength of rock mass units, and they contribute positively to making the slopes unstable. It is observed widely in the Sultan Mountain that, close to regional geological lineaments, the land sliding phenomenon is very severe.

The final map indicates that most of the observed landslide areas that are well predicted belong to high or very high landslide susceptibility classes. Hence, the high and very high landslide susceptibility classes in the final map can be considered highly believable. This map could help citizens, planners and engineers reduce losses caused by existing and future landslides by means of prevention, mitigation and avoidance.