1. Introduction
Roads affect wildlife ecosystems in various ways, including changing the land use patterns of animals, altering their behavioral patterns [
1,
2], and affecting dynamic community structures and components of ecosystems [
3,
4]. These factors influence the sustainability of animal species [
5,
6]. New roads affect wildlife in many ways, for example, they fragment communities and form barriers to movement [
7,
8,
9,
10]. Among these, fatal animal-vehicle collisions have direct impact on wildlife [
7,
8]. This phenomenon, called wildlife-vehicle collisions (WVCs), poses a serious threat to wild animals by endangering numerous animal species and also to vehicle drivers by causing loss of human life or property [
11]. The global data on wildlife killed in WVCs shows that the species (Mammals) that crossroads have a higher probability of being killed while crossing than other species (birds, reptiles, amphibians) [
12,
13,
14]. It is estimated that more than a million vertebrates are involved in fatal WVCs per day in the U.S. [
15]. On expressways, where vehicle speed and vehicular traffic are greater than on other roads, the occurrence of roadkill and isolation of animals are especially frequent [
16]. Expressways are wider than other roads, and their proximity to forests significantly increases the risk of WVCs on them compared to that on other roads. As described above, expressways pose threats to animals. However, industrial development is stimulating the construction of new expressways, which in turn, would expectedly increase the occurrence of WVCs.
The rising social and economic costs of WVCs have prompted researchers to conduct cost-benefit analyses on mitigation strategies [
17,
18]. The construction of wildlife crossings, designed to connect wild animal habitats divided by expressways, is one of the most well-known mitigation measures [
19,
20]. Wildlife crossings for ungulates and faunae are located below or above expressways, and they aim to provide safe passage to these types of animals. Although WVC mitigation measures such as wildlife crossings and fences are highly effective at reducing WVCs, their construction warrants considerable investment [
11]. Additionally, these mitigation measures tend to rely on past experiences rather than empirical analysis. Therefore, it is necessary to rank WVC reduction measures in terms of their effectiveness to establish the priority of WVC reduction measures by performing a quantitative analysis of the causes of WVC occurrence.
Compared to past studies on WVC that have mainly addressed the status of WVCs, many recent studies have focused on analyzing the causes of WVCs and developing approaches to mitigate them. These studies include analyses of WVC occurrence depending on landscape variables [
21,
22,
23], studies on the social costs of WVCs [
11,
17], and research verifying the effects of WVC mitigation strategies [
24]. In the extensive literature on WVC occurrence, many studies on the correlation between WVC occurrence and landscape variables have been conducted worldwide [
21,
22], but such studies have been rarely conducted in Korea. Therefore, it is necessary to analyze the correlation between WVC occurrence and landscape by using Korea’s WVC occurrence data to formulate a WVC reduction plan.
Animals are mainly found near their shelters and near locations where food is readily available. Expressways are built through wildlife habitats. Therefore, wildlife will cross any expressway built within their home range, which is the primary cause of WVCs. For this reason, it is extremely important to study the relationship between the landscape around WVC hotspots and the home ranges of wildlife.
In this study, we obtain geographical information regarding animal migration based on WVC occurrences on the expressway and designated WVC hotspots. A “hotspot” is an area in which the occurrence risk of WVCs is statistically higher than that in other regions [
25]. It is vital to identify these hotspots for reducing collision-related threats to wildlife [
26,
27]. Research on WVC hotspots involving wild animals has been widely used to prioritize locations for WVC reduction [
22]. It is essential to identify spatial and temporal hotspots and analyze their effects on WVC occurrence to establish mitigation measures and thus avoid high WVC.
Insufficient data regarding wildlife distribution in a given area can limit one’s ability to predict the occurrence of WVCs in the area. Therefore, uniformly collected WVC data are extremely important for WVC prediction. In addition, analysis conducted using a WVC occurrence map may not precisely indicate the accident concentration, especially at spots where more than two WVCs have occurred. Therefore, it is necessary to analyze the density of WVC occurrence at the same spot. This will help to predict the occurrence of WVCs more accurately and to establish an effective WVC reduction plan.
To this end, in the present study, we aim to understand the landscape characteristics of the locations frequented by water deer and the locations of WVC occurrence. Then, we identify WVC hotspots and analyze their spatial characteristics. Finally, we analyze the WVC cluster characteristics of the identified WVC hotspots. In doing so, we analyze the spatial characteristics of WVC occurrence to gather the basic data necessary for implementing WVC reduction plans on expressways.
3. Results and Discussion
3.1. Spatial Analysis Using Land Cover
To determine the land cover in the area where water deer exhibit active behavioral patterns, the points of appearance of water deer, as recorded in the national ecosystem survey, were used (
Table 2). The results of our analysis of the land cover around the locations at which water deer were found in Cheongju, Boeun, and Sangju based on the data from the National ecology survey indicated that Forested Areas showed the highest ratio (79.17%), followed by Agricultural Land (12.44%), Grassland (2.84%), and Water (2.02%).
The land cover of the areas with WVC occurrence was analyzed and compared with the land cover in the areas where water deer appeared. The type of land cover in areas close to the points of occurrence of WVCs indicated that Forested Areas comprised the majority of the WVC areas (69.72%), followed by Agricultural Land (17.92%), Used Area (6.19%), and Grassland (4.09%).
A comparison of the land cover in the areas where water deer appeared and the land cover in the areas where WVCs occurred revealed that the types of land cover in the areas where water deer frequently appeared were Forested Area and Agricultural Land. Moreover, the types of land cover in the areas where WVCs occurred were Forested Area, Agricultural Lands, and Used area, which were similar to the land cover results of the areas in which water deer appeared in the natural ecosystem survey.
The home range of water deer is Forested Areas, Agricultural Land, and Wetland [
33]. In addition, the results of previous studies indicated that Forested areas and Agricultural Land tended to increase the occurrence of WVCs [
38]. Given that the land cover in the areas of WVC occurrence is similar to the habitat range of water deer, the appearance probability of water deer is high. This increases the occurrence probability of WVCs.
The density of WVCs per unit area (km
2) for the types of land cover around the CSE was analyzed (
Table 3). The results indicate that the density of WVCs in Used Areas was the highest (27.41 cases per km
2), followed by Wetland (12.57 cases per km
2), Forested Areas (10.90 cases per km
2), and Agricultural Lands (10.58 cases per km
2). These results suggest that there is a difference between the appearance of water deer and the type of land cover where WVCs occur frequently.
Therefore, the differences in spatial characteristics between the areas where water deer appeared (National Ecology Survey) and the areas where WVCs occurred were compared. Based on the land cover in the study area (Cheongju, Boeun, and Sangju), the relative ratios of land cover in areas with water deer appearance (National Ecosystem Survey) and land cover in areas with WVC occurrence were analyzed (
Table 3).
As of 2010, the types of land cover in all Korean territories were as follows: Forested Areas (67.8%), Agricultural Lands (21.1%), Used Areas (4.1%), and Grassland (2.9%) [
39]. The land cover across Korea and the land cover in the area where water deer appeared were compared. The results indicated that the land cover ratios of Forested Area, Grassland, Wetland, and Bare Land in the areas where water deer appeared were higher than the corresponding ratios for all of Korea. The land cover ratios of Forested Area, Grassland, and Wetland in the areas where water deer appear are higher than those of Korea because CSE is located near Songnisan National Park. For this reason, there is a high probability that a larger number of water deer will inhabit the study area compared to other areas in Korea. In previous studies, the higher the habitat quality, the more vulnerable the area was to WVCs occurring near expressways [
22]. Therefore, the study area has a favorable environment for water deer to inhabit, and it has a high occurrence probability of WVCs.
The area in which water deer appeared had a higher ratio of Forested Area and Wetland compared to the study area, but the ratio of Used Area was lower. According to the present study, water deer frequent Forested Areas and Wetland, and this finding is consistent with the results of previous studies [
33] on the habitats of water deer. However, in the areas with WVC occurrence, the land cover ratios of Used Areas, Agricultural Land, and Grassland were higher than those in the study area. Used Area, Agricultural Land, and Grassland are areas characterized by low elevation and high openness. In previous studies, water deer had a high habitat density at elevations lower than 300 m [
40]. Therefore, the density of water deer is expected to be high in Used Area, Agricultural Land, and Grassland. Used Area and Grassland are wide and open terrains, so water deer can access them easily.
Based on the results, correlations with land cover in the areas where water deer appeared around the CSE and the areas in which WVCs occurred were statistically analyzed by conducting Spearman’s correlation test. As a result, the correlation of land cover between the areas where water deer appeared and the areas where WVCs occurred was −0.535, which is a negative correlation.
3.2. WVC Hotspot Modelling
Based on an analysis of the location data of WVC occurrence with KDE, WVC occurrence density was divided into five sections (
Table 4,
Figure 2). In Section I, which had the highest WVC occurrence density, 35.49 cases of WVC occurred per km, while 23.98 cases occurred per km in Section II. By contrast, the number of WVC occurrences per km in section V, which had the lowest WVC density, was 1.75.
The sections that exhibited higher WVC densities (cases per km) than that of the CSE overall (12.36 cases per km) were defined as WVC hotspots, and these hotspots were located in Sections I and II. The WVC density in the hotspots was 26.96 cases per km. The occurrence of WVCs in Section I was 20.28 times higher than that in Section V. Section I accounted for 5.30% of the CSE and 15.23% of the WVCs occurring along the entirety of the CSE. Section II accounted for 15.23% of the CSE and 29.54% of WVCs along the CSE. Based on these values, it can be concluded that even though the WVC Hotspots merely accounted for 20.53% of the CSE study route, the WVC occurrence ratio in the corresponding sections was 44.77%. These statistics indicate that most of the WVCs on the CSE occurred in Sections I and II. The WVC occurrence densities in Sections I and II were 26.31 per km2 and 21.24 per km2, respectively. A high percentage of the WVCs occurred in Sections I and II, and the WVC density in Section I was 12.96 times higher than that in Section V.
Based on the probability of WVC density on the CSE, each section was overlaid with a land cover map to analyze the land cover ratio of each KDE-derived section. The land cover ratios in the areas with WVC Hotspots were as follows: Forested Area = 65.33%, Agricultural Land = 19.58%, and Used Area = 7.06%. Compared to the WVC occurrence points (
Table 2), WVC hotspots had low ratios of Forested Area. Compared to the land cover in areas with water deer appearance (National Ecosystem Survey), the ratios of Used Area, Agricultural Land, Grassland, and Barren Land in the WVC Hotspot areas were high.
The density of WVC occurrence by land cover was analyzed (
Table 5). As a result, it was found that along the entire CSE and in the areas with WVC Hotspots, the highest WVC occurrence density was in the Used Area. The WVC density in the WVC hotspots generated in the Used Area was three times higher than that in the Forested Area. The density of WVC by land cover was highest in the Used Area, except in Section I.
Since WVCs are easily affected by land cover [
41], we found it necessary to compare the land cover maps of WVC Hotspots and other sections on the CSE. The land around Sections I and II had lower percentages of Forested Areas than the land in Sections III, IV, and V, and higher percentages of Used Area and Grassland. This result is consistent with that of a previous study on the suitability of water deer habitats with MaxEnt, which suggested that water deer prefer Used Areas among the different land cover types [
42]. The WVC Hotspots were characterized by higher percentages of open terrains such as Used Area, Grassland, and Barren Land. The open terrains, where roads are located, have fewer cut slopes than Forested Areas, which allows animals to easily access the expressway and increases the occurrence rate of WVCs [
43]. Water deer, which especially prefer to live on the edges of forests, tend to appear more frequently in Grasslands than in Forested Areas. The WVC occurrence density results provide evidence that WVC cases occur more frequently in open terrains [
43,
44]. Moreover, according to the field survey results, the closer a facility is to a Used Area, the higher is the occurrence probability of WVCs [
45] because in these areas, food is easy to find, and the occurrence probability of WVC is high due to the flat terrain.
Meanwhile, Sections IV and V, where WVC densities involving water deer were relatively low, had higher ratios of Forested Areas. A forested land cover and proximity to roads are spatial factors that can predict the WVCs of different ungulates (roe deer in Austria and France and white-tailed deer
Odocoileus virginianus in Illinois and Pennsylvania) [
46,
47,
48,
49,
50]. For instance, as the distance between a forest and a road increased by 100 m, the collision risk with moose decreased by 15% [
49]. In a previous study, it was demonstrated that forested habitats, which provide food sources to wildlife, and proximity to roads significantly influence the occurrence risk of WVCs [
49]. Thus, Sections IV and V, which had higher ratios of forests with trees and herbaceous plants, provided abundant food sources and resting places for water deer, such as lairs. Hence, in the areas with forested habitats, fewer WVCs involving water deer occurred compared to those in Sections I and II.
Two independent sample t-tests were conducted to analyze the statistical differences between the land covers in the WVC hotspots (Sections I and II) and non-WVC hotspots (Sections III, IV and V). The results showed a statistically significant difference (t = 5.655, significance level of 0.000, p < 0.05) in the land cover of the WVC Hotspot.
3.3. WVC Occurrence Cluster Analysis
The WVC distribution pattern in each of the five sections classified using the KDE function based on the WVC occurrence density probability was analyzed using the NNM (
Table 4). The distribution pattern in Section II was the most clustered with an NNR of 0.046 while that in Section V was the most dispersed with an NNR of 1.950 (
Table 6).
The WVC distribution patterns in the WVC Hotspots, entire CSE, and five sections analyzed using the KDE technique were examined using the NNM. While the CSE as a whole had an NNR of 0.097, the WVC hotspots had an NNR value of 0.043. These NNR values indicate that the WVC occurrence pattern in the hotspots followed a highly clustered distribution compared to that along the entire expressway. The distribution in Section I was relatively dispersed compared to those in Sections II and III. The WVC cases occurring in Section I were distributed over the entirety of Section I. The WVC Hotspots had high distributions and densities of wildlife because they were the transit regions for wildlife [
51,
52].
Section 1, as classified using KDE, accounted for 5.30% of the CSE, but the density of WVC occurrence in this section was high. Since the occurrence probability of WVCs was high throughout Section I, a distributed distribution was observed in Section I.
The NNR values of Sections II and III were 0.046 and 0.050, respectively, indicating clustered distributions. In Section II, the density of WVC occurrence in the Used Area was the highest at 81.53 cases per km
2 (
Table 5), but the land coverage ratio of Used Area was only 6.98% (
Table 4). The density of WVC occurrence in the Used Area in section II was 3.83 times higher than that in the entirety of section II. Therefore, the occurrence of WVCs was high in a small patch of Used Area, which implies that WVC occurrence in the section followed a clustered distribution. As for the land cover in Section III, the proportions of Forested Area and Agricultural Land were extremely high, accounting for 86.69% of the total land in Section III (
Table 4). However, the types of land cover with high WVC densities were Wetland and Used Area. Wet Land and Used Area had small distributions in Section III. Therefore, WVC occurrence in Section III tended to follow a clustered distribution compared to the other sections.
4. Conclusions
Water deer are exposed to many risks due to expressway construction, and for this reason, they are involved in WVCs. Most of the WVCs that occurred on the CSE involved water deer. Therefore, we identified WVC hotspots on the CSE and investigated the effect of spatial distribution on WVC occurrence based on land cover in the areas where water deer appeared and land cover in the areas where WVCs occurred.
Most of the lands in the areas where water deer appears were Forested Areas and Agricultural Lands. The land cover ratios of Forested Area and Wetland in the areas where water deer appeared were higher than the corresponding values in the study area (Cheongju, Boeun, and Sangju), whereas the land cover ratio of Used Area was lower than that in the study area. In the areas with WVC occurrence, the land cover ratios of Used Area and Agricultural Land were higher than those in the study area.
An analysis of the occurrence density of WVCs indicated that many WVCs occur in Used Areas. Moreover, the density of WVC hotspots was the highest in Used Areas. Through cluster analysis, it was confirmed that rate of occurrence of WVCs was high in Used Areas. Used Areas have a smaller cut slope than Forested Areas, meaning that wildlife can easily access the expressway, and these results are consistent with those of a previous study, where it was reported that WVCs occur frequently on open terrains.
Furthermore, the occurrence of WVCs was found to be influenced by the spatial distribution around the CSE, and the WVC occurrence pattern tended to be spatially clustered. WVC occurrence followed a clustered distribution rather than a random pattern [
49]. Moreover, WVC occurrence points were affected by the local circumstances and local landscapes [
50]. Similar results were reported in other studies conducted by road ecologists, indicating that WVC occurrence patterns tend to be spatially clustered instead of random [
26,
51,
52,
53]. Therefore, to establish an effective policy for reducing WVCs on a road, the WVC characteristics and the spatial distribution of the road should be considered together.
WVCs have a significant influence on the safety of humans and animals. Therefore, it is essential to analyze the exact location, period, and causes of WVCs in high-risk areas and establish effective mitigation measures [
54]. To reduce the probability of WVC occurrence on expressways, the locations of WVC mitigation facilities should be selected based on analyses of each animal’s behavioral and ecological characteristics. New mitigation measures should address the factors affecting the spatial distribution of WVCs.
Herein, we quantitatively analyzed the occurrence of WVCs and studied the spatial characteristics that affect WVC occurrence. Therefore, the results of this study provide a quantitative priority of various landscapes for establishing a WVC reduction plan in the future. A limitation of this study is that we did not consider the characteristics of various ecosystems because we conducted the study on water deer without focusing on the underreporting of WVC data, and land cover was used among various spatial characteristics. Therefore, if the characteristics of WVC hotspots are analyzed considering the characteristics of various ecosystems in the future, an appropriate WVC reduction plan can be established.