Introduction

Habitat fragmentation is mostly mentioned within the concept of climate change; however, roads have also great potential to cause degradation and fragmentation effects on flora and fauna (Trombulak and Frissell 2000). These potential effects begin with the pre-process stage of road constructions and continue with the post-processing stages of road network (Forman and Alexander 1998; Seiler 2001; Forman 2006). Due to barrier effects of road networks, wild animals might face long-term population decline as well as increased wildlife collision (Wc) rates, which result in deaths and financial losses on roads (Theobald et al. 1997; Seiler 2005; Morelle et al. 2013).

There have been an increasing interest in road ecology and related research areas around the world, because of habitat fragmentation effects that result in Wc and conflict with international conservation policies. Road ecology is one of the main issues regarding conflicts between wildlife and human. The road networks are usually the major source of this conflict in and around forested areas. In order to overcome negative effects of road networks on wild animals, the wildlife passages have become the most common solution in many European and American countries. Based on the type of engineering structures that are preferred to link, the habitats are named under a number of terms such as ecological passages (eco-passages), ecosystem bridges, wildlife bridges, wildlife tunnels, wildlife culverts, eco-ducts, and wildlife underpasses and overpasses (Iuell et al. 2003; Clevenger and Waltho 2005; Coffin 2007; Glista et al. 2009). Eco-passage is the most common term used for wildlife passages that somehow penetrate the road network and connect fragmented habitats of wild animals (Clevenger et al. 2001). Besides, it is not only wild animals but also domestic animals pass through these road engineering structures located over or under the road network (Gülci and Akay 2015). In addition, these engineering structures are designed for hydrological and geological purposes and it can also be considered for ecological purposes under the road ecology disciplines (Spellerberg 1998; Coffin 2007).

Determining optimum locations for eco-passages is a very important stage to prevent wildlife and human as well as to ensure biological conservation (Iuell et al. 2003). There are several approaches to find optimum locations for eco-passages at road networks (Clevenger 2005). Wildlife movement corridor-based approach is well known and mostly the preferred method in which optimized movement or migration routes assist decision makers to find optimum solutions for eco-passage locations (Beier and Loe 1992; Beier and Noss 1998; Conrad et al. 2012; Cushman et al. 2013; Gülci 2014). The distribution of wild animals should be evaluated with generalization concept considering eco-geographic variables that affect habitat requirements of the animals (Donovan et al. 1987; Guisan et al. 2002). However, the researchers have been often faced with difficulties to generate prediction models about habitats of wild animals, since generally, many input data may not be expressed in numeric values in environmental researches. Yet, indirect decision variables such as land use, stand type, topography, etc. ought to circumvent with generalized model usage for estimating distribution of species according to their habitat preferences (Guisan et al. 2002; Doswald et al. 2007).

By using advanced geographical information system (GIS) tools and improved mathematical optimization techniques such as multi criteria evaluation (MCE) methods makes it possible to assess the dynamic and complex relations between live organisms (i.e., animals, plants, others) and non-living environments (i.e., water, rocks, others) based on specified decision variables (Gontier 2008; Gülci 2014). There are many wildlife conservation and management-related studies where habitat suitability model (HSM) has been commonly used to classify the habitats based on ecological requirements of wild animals (Hirzel et al. 2001; Store and Kangas 2001; Gontier et al. 2010; Carvalho et al. 2012). The literature data and expert opinions are also very useful data filling the gaps on field data absences to develop HSM (Clevenger et al. 2002). Surely, expert opinion or literature data can be converted into the numbers; however, the relative importance of decision variables should be investigated with MCE techniques by decision makers in ecological studies (Guisan and Zimmermann 2000). For instance, relative importance of variables have been usually found out by using three methods. (1) Rank sum, (2) reciprocal (ratio matrix), and (3) exponent procedures are well-known methods in MCE procedures (Malczewski 2000; Öztürk and Batuk 2011).

Recent studies indicated that GIS-based MCE methods can provide optimum locations for ecological passages (eco-passages) based on HSM by decision makers (Gülci 2014). The most common MCE methods include weighting, analytical hierarchy process (AHP) and technique for order preference by similarity to ideal solution (TOPSIS) (Akay et al. 2011; Öztürk and Batuk 2011). The advantages of MCE approach are that it easily expresses the relations between species and environment with different habitat factors, integrates with GIS, and utilizes empirical models and expert knowledge (Store and Jokimäki 2003; Seiler 2005). Furthermore, GIS-based MCE methods are powerful tools for habitat suitability analysis and there are many successful studies for Wc prevention and ecological conservation studies (Craver 1991; Eastman et al. 1995; Alexander et al. 2004; Malczewski 2006; Gülci 2014).

Briefly, generated habitat suitability models help to design movement corridors and intersecting with the motorways should be determined to locate eco-passages. In the solution process of locating eco-passages at the mitigation points on roadways by using movement corridor, there are many alternative methods for solving this ecological matrix. Least cost path (LCP) is one of the well-known methods which is originated from a graph theory for estimating the best (least cost) paths of wildlife corridors (LaRue and Nielsen 2008; Conrad et al. 2012). LCP logic was explained with the first algorithm in 1959 by Dijkstra, and intensively, it has been used not only in computer science but also in many other science branches since then (Yu et al. 2003; Siniedovich 1980). LCP has also been evaluated in wildlife studies to estimate crossing points of roads and wildlife crossing places (Huck et al. 2011; Cushman et al. 2013).

This study aims to evaluate ecological passages along a sample motorway within a Mediterranean city of Osmaniye in Turkey by using GIS-based MCE method. In the solution processes, HSM was developed to estimate logical movement corridor, and then the best route was determined by using LCP method for roe deer (Capreolus capreolus) as target species in the study. Finally, mitigation points where movement corridor intersected with the motorway were evaluated for eco-passages.

Materials and methods

Study area

The study area is located at the city of Osmaniye in the Mediterranean Region of Turkey. This region is one of the most important hotspots in terms of richness in flora and fauna (Şekercioğlu et al. 2011). The study was carried out at some part of Motorway 52 which connects Adana and Şanlıurfa motorways. Motorway 52 passes through within the border of Osmaniye Forest Enterprises Directorate (FED) (Fig. 1). The length of the motorway within the study area is approximately 60 km and consists of two tunnels. The motorway has three lanes at each side (total of six lanes), and both sides of the road were wired with fences along the entire roadway for security purposes. Motorway 52 was opened for traffic in 2006, and it carries averagely 18,934 vehicles annually in research area according to General Directorate of Highways (GDH 2014). The maximum speed limits for articulated trucks and automobiles are 60 and 120 km/h, respectively (GDH 2014).

Fig. 1
figure 1

Location of the study area and route of Motorway 52 in Osmaniye FED

Full size image

The Osmaniye FED has a total land of 160,790 ha in which 91,563 ha was covered with forests. The land use along the Motorway 52 road corridor is mostly covered by agricultural lands, pastures, forest lands, and openings. Agricultural vegetation types around roadway are cultivated with wheat, corn, peanuts, and non-irrigated farm lands with olive tree and few other fruit trees. Forest lands are mostly covered by degraded, mixed and native forests with Brutian pine (Pinus brutia), Oak (Quercuss spp.), and forest openings (GDNCN 2013). Settlements around roadway and its surroundings altered into agricultural facilities that are unsuitable habitats for wildlife species. However, rural countryside of Osmaniye FED, which is especially located in the southern part of the study area, has very suitable habitats for fauna. The common wild medium and small size terrestrial mammalian species are wild boar (Sus scrofa), roe deer (Capreolus capreolus), jackal (Canis aureus), and fox (Vulpes vulpes). The domestic animals are cat, dog, cattle, goat, and sheep (GDNCN 2013).

Target species and decision variables

Roe deer (Capreolus capreolus), which are regularly surveyed by the General Directorate of Nature Conservation and National Parks, have been chosen as target species. The population of roe deer was reported as 192 in the region, and this number will increase since they are strictly under protection with law and listed as very important wild animals by FED in terms of ecotourism purposes (GDNCN 2013). Moreover, roe deer’s potentially high Wc rate in general effected it to be determined as the target species (Morelle et al. 2013).

The information about roe deer habitat requirements were mainly obtained from previous studies (Çanakçioglu and Mol 1996; Sempere et al. 1996; Rossi et al. 2001; Madsen et al. 2002; Ratikainen et al. 2007; Li-wei et al. 2006; Pagon 2010). Because of local differences in roe deer preferences, specific experiments and feedbacks from local hunters and experts were also carefully recorded (Clevenger et al. 2002; Doswald et al. 2007).

In this research, indirect eco-geographic variables were used in order to construct HSM (Guisan and Zimmermann 2000; Zimmermann 2004). There are nine indirect decision variables which are mostly preferred to estimate the most suitable and the least suitable habitat places for roe deer considering species behaviors such as sheltering, feeding, hiding, and mating. These variables are (1) land use (LU) types assessed with the use of CORINE data, (2) vegetation cover (VC), (3) stand age and crown closer (SaC) assessed with the use of Turkish forest operation plans, (4) elevation (E), (5) slope (S) assessed with the use of ASTER DEM (30 m × 30 m), (6) road density (RD), (7) road distance (RDist) variables calculated and assessed with the use of road network and ASTER DEM, (8) water distance (WDist) assessed with use of water lines and ASTER SYM, and (9) topographic position (TP) assessed with the use of ASTER DEM considering Weiss (2001) classification.

Constructing the HSM

The model involves large-scale geographic area with specified decision variables for roe deer. GIS is utilized, since it is the best tool for making a relation between decision variables representing ecological constraints of roe deer. ArcGIS 9.3, a common GIS software, was chosen for the pre-processing and post-processing stages based on raster surface models (Gontier 2008). The objective was focused on ecological passages and to determine the best crossing points (mitigation points = hotspots) in a reasonable computer time. First, all main and sub-factors were first calculated in MS Excel and then combined in GIS environment to integrate with MCE methods (Store and Jokimäki 2003; Malczewski 2006).

Suitability degree estimate

The rating scale table was designed as habitat suitability index (HSI) to describe linguistic and non-standardized data changed into mathematical expression or otherwise standardizing (ESM103 1981; Wakely 1988; Clevenger et al. 2002; Alexander et al. 2004). The rating criteria were divided into five linguistic functions considering literature and expert-based knowledge about ecological preferences of roe deer (Clevenger et al. 2002; Gontier et al. 2010). Rating score range from 1 to 10, where “1” represents excellent and “10” represents very poor habitats (Table 1). There is also another rating score of “0” for extreme habitats that means absence of roe deer. Table 1 has been also used for rating each of sub-criteria (p) of decision variables LU, VC, SaC, S, RD, RDist, WDist, E, and TP. As a result of the rating process issued with 0 or among 1 to 10 scores (r), suitability degree (SD) was computed for each decision variable by using Eq. (1).

$$ \left\{\begin{array}{l}r\in \left[x,\left\{0\right\},{Z}^{+}\right]\Rightarrow 0\le x\le 10\\ {}p\in \left[x,{Z}^{+}\right]\Rightarrow x\ge 1\end{array}\right\}\Rightarrow \left\{\begin{array}{l} LU={p}_1r+{p}_2r+\dots +{p}_nr\\ {}VC={p}_1r+{p}_2r+\dots +{p}_nr\\ {}SaC={p}_1r+{p}_2r+\dots +{p}_nr\\ {}S={p}_1r+{p}_2r+\dots +{p}_nr\\ {}RD={p}_1r+{p}_2r+\dots +{p}_nr\\ {} WDist={p}_1r+{p}_2r+\dots +{p}_nr\\ {} RDist={p}_1r+{p}_2r+\dots +{p}_nr\\ {}E={p}_1r+{p}_2r+\dots +{p}_nr\\ {}TP={p}_1r+{p}_2r+\dots +{p}_nr\end{array}\right. $$
(1)
Table 1 Rating criteria for the decision variables
Full size table

Data preparation and combination with GIS

Geoprocessing was driven by GPS data and 1/25,000 scaled topographic maps were used for decision variables which were generated in both vector and raster format. Once vector data were corrected, all vector-based eco-geographic data layers were transformed into raster data (30 m × 30 m) with the help of ArcGIS 9.3 extensions (Spatial Analyst) (Esri 1996). Calculated SD values for decision variables were saved in text (.txt) format with MS Excel software. Then, SD values were used to generate raster grids (30 m × 30 m) in order to evaluate layers in GIS environment. And, “Join” tool which is available in ArcGIS 9.3 was used to combine SD values for sub-criteria evaluation. Finally, possible literature data and expressions in decision variables-ranked scores were converted into percentage (%) expression with simple linear regression (R 2 = 0.9958) which has the highest correlation (Eq. 2). Table 2 indicates the scale of SD classes.

$$ y=-10,636X+108 $$
(2)
Table 2 Classes for constraint assessment values
Full size table

Generalization process

SD generalized from 0 to 1 and relative importance (W i ) values calculated with reciprocal (ratio matrix) belong to spatial arrangement for object functions of decision variables (ESM103 1981; Wakely 1988). The mean of SD (\( {\overline{p}}_i \)) was determined for each of the sub-criteria spatial distribution of the decision variables in weighted linear overlay process (Eq. 3). SD value of “0”, which represents the absolute absence, was not added to the arithmetic mean calculation. Relative importance (W i ) was calculated using Eq. (4) with the help of the function of variables, based on the proportional relationship between the numbers of pairs of intensity distributions (ESM103 1981; Malczewski 2000; Öztürk and Batuk 2011).

$$ \frac{{\displaystyle \sum_{i=1}^n\left({p}_1w+{p}_2w+\dots +{p}_nw\right)}}{p_n}={\overline{p}}_i $$
(3)
$$ {W}_i = \frac{\left(1/{\overline{p}}_i\right)}{\sum \left(1/{\overline{p}}_i\right)} $$
(4)

Overlay process

Relative importance between decision variables were used to develop HSM which helps to identify active weights (β) to determine severity (w i ) between the decision variables which are obtained from the scoring table set (Eq. 5) (Eastman et al. 1995; Malczewski 2000). Hence, the last permeable surface was investigated in the identification of roe deer’s mostly preferred habitats and suitability arrangements in study area.

$$ Z=\frac{{\displaystyle \sum_{i=1}^nwi\times xi}}{{\displaystyle \sum_{i=1}^nwi}} $$
(5)

Path scenario and mitigation points

Considering the wildlife corridor approach, LCP method was implemented to identify the most and worst suitable areas for roe deer in order to estimate movement corridor (Adriaensen et al. 2003; Zimmermann 2004). The method was implemented by using specified decision variables and roe deer habitat preferences information which were obtained from literature data and expert knowledge. The method was used for assessing the potential roe deer mitigation point after intersected passages with motorway (i.e., expressed as mitigation points or hotspots). And, spatial analysis process was performed with ArcGIS 9.3 tools with the extension of Queens’ pattern (Yu et al. 2003).

Raster surface was examined considering eight neighbor cells (3 × 3 cells) for cumulative cost or effective movement corridor analysis (Fig. 2). Two accumulated cost (C) algorithms were performed to move between core habitats for roe deer on raster cells. Excellent places where roe deer has ability to mate, feed, and hide areas were decided as the starting point (S 0) and forward destination (d). Horizontal and vertical movements from the start point to the forward destination (S 0 to d 2, S 0 to d 4, S 0 to d 6, S 0 to d 8) were calculated with Eq. (6), and other movement types were crosswise movements from the start point to the forward destination (S 0 to d 1, S 0 to d 3, S 0 to d 5, S 0 to d 7) on raster cells (Eq. 7).

$$ \begin{array}{l}{C_{S_0}}_{,{d}_i}=\frac{\left({C}_{S_0}+{d}_i\right)}{2}+{C}_{S_0}\\ {}i=\left[2,4,6,8\right]\end{array} $$
(6)
$$ \begin{array}{l}{C_{S_0}}_{,{d}_i}=\frac{\left({C}_{S_0}+{d}_i\right)}{2}*\sqrt{2}+{C}_{S_0}\\ {}i=\left[1,3,5,7\right]\end{array} $$
(7)
Fig. 2
figure 2

Directional movement pattern for the start point to the forward destination (3 × 3 pixels)

Full size image

Considering HSM, one mostly preferred S 0 point and nine d points were determined for roe deer in Osmaniye FED. The movement directions were calculated from weighted surface by using the given equation. The accumulative costs between points were spatially evaluated by LCP with the help of ArcGIS 9.3 (Esri 1996). Then, the best path or possible movement corridors for roe deer were determined between S 0 and d points. Logically determined movement corridors were intersected with Motorway 52, and the mitigation points were evaluated for road engineering structures if it could be assessed as ecological passage for roe deer.

Field works

During field works, observation activities were performed for whole destination points (d) in order to control the area and to compare the actual condition with predicted habitat suitability. Photo-trap images were also used for providing proof evidence of roe deer existence in the area. In addition, to control the predicted movement corridor if roe deer have used these directions, mitigation points were evaluated for availability and suitability of road engineering structures (i.e., wildlife culverts, wildlife bridges, eco-ducts) at Motorway 52. Besides, mitigation points and the circular area with a radius of 250 m from mitigation points were surveyed for road engineering structures and footprints of roe deer.

Results and discussion

Expert knowledge and literature data-based ecological preferences for Roe deer were summarized in given matrix table (Table 3). All decision variables ranking from 1 to 10 scores including “0” were described for preparing standardization and generalization process in GIS. Decision variables and functions were crossed with each other for finding constraints. Thus, each function could have been ranked for sub-criteria in decision variables considering ecological constraints for roe deer.

Table 3 Roe deer ecological preferences for nine decision variables with five functions
Full size table

Prediction of roe deer’s habitat preference was determined with nine decision variables and their importance was estimated from 0 to 1. HSM overlay process was calculated with Eq. (8) which expresses roe deer’s spatial preferences weighting between nine decision variables with simple linear model. In other words, ranked subdivision classes of decision criteria were evaluated with reciprocal (ratio matrix) in MS Excel table. Reciprocal assessment is an easy and accurate method to understand relation and weights (relative importance) between decision variables (Öztürk and Batuk 2011). The results showed that RDist was the most effective function, while RD was the least effective function in this research.

$$ \mathrm{H}\mathrm{S}\mathrm{M}=\left(\begin{array}{l}\left(0.09\times RD\right)+\left(0.10\times VC\right)+\left(0.10\times E\right)+\\ {}\left(0.10\times WDist\right)+\left(0.11\times LU\right)+\left(0.11\times TP\right)+\\ {}\left(0.12\times SaC\right)+\left(0.12\times S\right)+\left(0.15\times RDist\right)\end{array}\right) $$
(8)

It was observed that human activities have been raised day by day in the region and this process has caused territorial reduction in Osmaniye FED. Road networks are one of the most important factors for habitat fragmentation due to their barrier effects on habitats of roe deer and other animals (Trombulak and Frissell 2000; Seiler 2001; Forman 2006). Forest lands have been sharply divided into three habitat blocks by main roads (Motorway 52) and minor roads. Especially, Motorway 52 has made it almost impossible for roe deer to move between habitats. Figure 3a, b shows predicted HSM for roe deer based on suitability degrees and HS classes, respectively. It was found that main roads and settlements have relatively poor habitat suitability, and the degree of suitability increases as getting away from these structures. Figure 3b was generated by using Eq. 2 for better understanding of suitability range in study area.

Fig. 3
figure 3

Predicted HSM for roe deer based on suitability degrees (a) and HS classes (b)

Full size image

WLC is one of the very useful, easy, and quick methods in GIS-based MCE methods. Yet, WLC method puts on very rigid SD because of the additive process on multiple variables. As a result of the additive manner, predicted HSM could not explain “absent” and “very poor” places for Roe deer (Guisan et al. 2002). Surely, WLC method represents useful tool for prediction and quick assessments when the baseline information and expert feedbacks are lacking and time limitation do not allow for data collection (Clevenger et al. 2002; Doswald et al. 2007). WLM played rigid classification and evaluation manner released as like Boolean operators and HSM which could give three main functions of excellent, average, and poor habitats between suitability value of 84.97 and 27.93 %.

Accuracy of HSM results was controlled by field observations, wildlife management plans, and photo-trap studies. Wildlife protection and management area provides important opportunities to evaluate species absence/evidence situations in research area for target species (GDNCN 2013). After HSM was completed based on literature data and expert feedbacks, most preferred areas were detected through field observations for roe deer evident in the research area (Fig. 4). It was found that roe deer was actively available at nights and in day times. Thus, literature data and expert-based HSM provided important answers to the question about distribution of roe deer in Osmaniye FED (Doswald et al. 2007; Clevenger et al. 2002).

Fig. 4
figure 4

Detected female (a) and male (b) roe deer (Capreolus capreolus)

Full size image

The results also indicated that LCP scenario provided accumulative best routes where estimated movement paths or corridors could be used easily by roe deer. Totally, ten movement corridors were generated for roe deer in the study area. All possible best habitats were examined to represent whole area with ten target points for best movement corridor. Figure 5 shows all LC movement corridors with the mitigation point A and B on Motorway 52. Both A and B mitigation points with its surrounding area (500 m2) was surveyed in the field. Point A had no engineering structure for wildlife passage, while an infrastructure built for hydrologic function was located about 70 m far from the point at Motorway 52. Point B had no engineering structure for wildlife passage either, and an infrastructure built for hydrologic function was located about 145 m away from the point at Motorway 52. On the other hand, there was a motorway tunnel which was located 145 m away from point B and it was suitable for natural wildlife overpass.

Fig. 5
figure 5

LC roe deer movement paths and mitigation points on Motorway 52

Full size image

It was found that both mitigation points were not suitable for roe deer crossings between the estimated suitable habitats in the study area. For instance, within motorway corridors, wire security fences (about 2 m high), steel road barriers, and steep cut-fill slopes prevent roe deer crossings (Fig. 6). New constructions and reorganizations were obviously required to make these mitigation points useful and actual ecological passages for roe deer. Around the mitigation point B, there was a canal barrier which was constructed for hydroelectric power plant. Even though urbanization, power plants, mining, forest operations, and other factors affect ecological passages, all kinds of earth work and movements during these constructions start with the location of road network; therefore, essential issue must be implementing principles of road ecology in order to protect wildlife and ensure biodiversity (Şekercioğlu et al. 2011).

Fig. 6
figure 6

General overview for mitigation point B (a) and wire fence along Motorway 52 (b)

Full size image

Turkey’s biodiversity in general have faced many threats especially due to improper constructions, overgrazing, excessive irrigation, and draining of wetlands that require immediate conservation challenges for sustainability (Şekercioğlu et al. 2011). It is clear that solution to these problems needs multidisciplinary studies with contribution of research institutions, practitioners, and non-governmental organizations. Besides, previous studies indicated that GIS-based MCE using WLC method has great potential to solve the problems. On the other hand, environmental and structural issues should be evaluated efficiently and effectively considering economic, environmental, and social factors (Store and Jokimäki 2003; Malczewski 2006; Gontier et al. 2010; Conrad et al. 2012).

Conclusion

This research was to investigate the location of ecological road structures considering wildlife habitat conservation. GIS-based MCE method using WLC and LCP modelling provided optimum solutions for positioning strategic points (mitigation points = crossing points = hotspots) for roe deer in study area. Complex ecological relations between road network and wildlife were standardized and generalized by evaluating specified decision variables in the solution process. Decision variables representing ecological preferences of roe deer were expressed in numerical values with ranking based on literature data and expert feedbacks to produce understandable results for wildlife corridors. The fragmented habitats and wildlife corridors of roe deer were investigated and clearly defined along the roadway of Motorway 52. The visual evidence of roe deer was captured with photo-trap in corridors defined with the help of estimated HSM and LCP method. The results indicated that computer-based models provide accurate and quick solutions for positioning ecological passages to ensure sustainability of wildlife habitat and reduce environmental effects of road networks on wild animals. In this study, the methodology was performed for single species; thus, future studies should consider available species in a study area.