Introduction

Habitat fragmentation, associated with human population expansion and the following landscape modification, is of major concern to agencies of wildlife conservation [1,2]. Disconnected landscape impedes dispersal, reduces gene flow and aggravates inbreeding in endangered species [3]. Well-connected landscape is considered necessary for long-term viability of extinction-prone metapopulation [2,4]. Since Merriam [5] firstly introduced the concept and Taylor et al. [6] defined it as “the degree to which the landscape impedes or facilitates movement among resource patches”, people are putting increasing efforts in maintaining and restoring landscape connectivity.

Numerous connectivity models have been put forward but there has been no consensus over their definitions and algorithms [7,8]. Despite disagreement over these connectivity models, researchers agree that involving species-specific perceptions of landscape determines, to a large extent, the success of a model in predicting potential connectivity between patches [7,9,10,11]. This consensus has motivated the concept of functional connectivity [7], which has rapidly expanded beyond stepping stones, migratory stopovers and habitat mosaics [2,10]. With the help of Graph Theory and Geographic Information System (GIS), a growing number of studies model least-cost functional connectivity between patches [12,13,14]. The functional model uses species-specific habitat preference, often based on expert opinion and literature [15,16], to depict the least-cost movement corridors or paths that provide better opportunities for successful dispersals. Advantages of this methodology include modest data requirements, a great benefit-to-effort ratio, visualized outputs and a convenient update process [7]. There exists, however, a common complaint about the expert-driven resistance estimates of landscape features, known as the “subjective translation problem” [17]. To improve the quality of and confidence in the least-cost algorithm, researchers suggest testing and refining the resultant map with empirical data, among others, results from landscape genetics are increasingly used [17,18,19].

Przewalski’s gazelle, endemic to the Qinghai-Tibetan Plateau, is one of the most endangered antelope species in the world [20] and listed as Endangered on the IUCN Red List [21] and as a Category I Key National Protected Wild Animal Species in China [22]. The population underwent significant decline from the 1950s onwards and is now constrained to several isolated habitat remnants around the Qinghai Lake (Figure 1) [22,23,24]. Since the 1990s, local populations of Przewalski’s gazelle have experienced different trends [23,25]. Gazelle populations in patches experiencing either rebounds in population size or declines in habitat suitability may disperse to search for potential habitats [25,26]. On the other hand, Yang et al. [27] found significant genetic differentiation in the species, which call for gazelle dispersals between patches. Although habitats of the gazelle are relatively flat moraine covered with steppe, sand dunes and meadows, the gazelle may not move along the shortest Euclidean distance due to habitat degradation. In this context, it is necessary to assess functional connectivity among patches, including extant distributed areas and new potential habitats.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Habitat suitability map with extant and potential habitat patches of Przewalski’s gazelle.

Przewalski’s gazelle are currently constrained in twelve isolated habitat remnants: Yuanzhe (YZ), Hudong (HD), Ketu (KT), Sand Island (SI), Ganzihe (GZ), Hargai (HG), Taliexuanguo (TL), Bird Island (BI), Kuairma (KM), Shengge (SG), Wayu (WY) and Ranquhu (RQ). The suitability scores range from one to ten, representing increasing habitat quality or movement resistance. Lakes in light blue and high mountains (> 4 000 m in elevation) in grey are avoided by the gazelle.

https://doi.org/10.1371/journal.pone.0080065.g001

Thus far, except for a few inter-patch movements reported by local pastoralists and inferred by researchers [25,27,28], no studies have been conducted with regard to dispersal connectivity for Przewalski’s gazelle. Here, we present a spatially explicit model of habitat suitability and least-cost functional connectivity between the populations. We further evaluate and test the model using the latest evidence from landscape genetics of the gazelle [27]. This study presented thorough and visualized information for making conservation plans. In a general sense, we provided a heuristic framework to model landscape connectivity for species with similar biological characteristics and habitat patchiness to Przewalski’s gazelle.

Methods

Results

Experts ranked slope as the strongest and elevation as the weakest in influencing both habitat selection and movement of Przewalski’s gazelle (Table 2 and 3). In the habitat suitability model, human density and NDVI followed the strongest factor. In the cost surface model, distance to settlement was the second strongest factor while distance to individual house was the fourth one. There were no disparities in ranking with the other landscape features, which had similar relative influencing weights between habitat suitability and cost surface.

Among the extant distributed patches, average habitat suitability score was highest in SG (9.31) followed by GZ (9.25), HG (9.01), TL (9.01), YZ (8.80), KM (8.71), SI (8.60), HD (8.19), RQ (7.92), KT (7.76), WY (7.68) and BI (7.57). Patches with suitability higher than 9.50 covered a total area of 1 753 km² (2.7% of the analytical window). Among these patches, five potential habitats (P1-P5) were identified and their area amounted to 638 km² (1.1% of the analytical window). Of the five potential habitats, the shortest distance to current ranges was 2.5 km from TL to P1 and three existing patches (GZ, HG and TL) were in the vicinity of P1 (within 7.6 km). P2 was relatively further away but within 20 km distance from four populations (KT, SI, GZ and TL). P3, P4 and P5 were over 30 km away from proximate extant patches (33.1 km P3-KT, 49.0 km P4-TL and 31.6 km P5-BI) and isolated by high mountains or towns (Figure 1).

Fifteen pairs of least-cost functional corridors and paths were depicted between neighboring pairwise patches including eleven extant and two potential patches (P1 and P2) within 25 km of one another (Figures 2, 3 and 4). Least-cost paths were all longer than the corresponding Euclidean distances except nearly equality between GZ and HG. Least-cost paths ranged from 0.2 km to 26.8 km (averaging 12.4 km) while Euclidean distances from 0.2 km to 22.8 km (averaging 10.2 km). Of the 15 least-cost paths, four held the cumulative movement cost lower than 10 000, four higher than 20 000 and seven in between. The least movement cost was found between GZ and HG (291) while the highest was between KT and P2 (42 268). The maximum cost of 1-km-corridor (corridor slice with minimum width of one km at the narrowest point along their length) ranged from 600 to 45 000, averaging 17 607. The 1-km-corridors of HD-KT, HG-TL, KT-P2, SI P2 and TL-P2 split into branches with several small areas of unsuitable landscape in the middle and near the edges. If we set the cost value of the least-cost path HD-KT (between which gazelle movements were considered present) as a threshold for inter-patch dispersal, there would be six traversable corridors, i.e., GZ-HG, TL-P1, HG-P1, SG-KM, KT-SI and HD-KT in descending ranking of permeability (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. More traversable corridors and least-cost paths among extant populations of Przewalski’s gazelle.

Four pairs of corridors and least-cost paths, GZ-HG (A), SG-KM (B), KT-SI (C) and HD-KT (D), were considered traversable if we set the cost value of the least-cost path HD-KT as a threshold for inter-patch dispersal.

https://doi.org/10.1371/journal.pone.0080065.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Less traversable corridors and least-cost paths among extant populations of Przewalski’s gazelle.

Five pairs of corridors and least-cost paths, HG-TL (A), HD-YZ (B), GZ-TL (C), WY-RQ (D) and SI-GZ (E), were considered less traversable if we set the cost value of the least-cost path HD-KT as a threshold for inter-patch dispersal. A photo of the newborn Przewalski’s gazlle found in SG was displayed in panel F.

https://doi.org/10.1371/journal.pone.0080065.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Corridors and least-cost paths between extant populations and potential patches of Przewalski’s gazelle.

Six pairs of corridors and least-cost paths, TL-P1 (A), HG-P1 (B), TL-P2 (C), GZ-P2 (D), SI P2 (E) and KT-P2 (F), were mapped for Przewalski’s gazelle between potential patches and neighboring extant populations.

https://doi.org/10.1371/journal.pone.0080065.g004

Discussion

We found that pastures in the study area were of low quality with limited suitable habitats for Przewalski’s gazelle in patchy landscape (Figure 1). Human expansion and excessive livestock grazing has caused deterioration of nearly all grasslands in the region [37]. Four populations (YZ, HD, KT and SI) at the eastern and BI population at the western shore of the Qinghai Lake are struggling to survive in restricted space neighboring to large sand dunes [25]. Two small populations (WY and RQ in 2008 and 2009, respectively) were newly found in degraded rangelands of the Gonghe Basin to the south of the Qinghainanshan Mountain. They might immigrate from the Qieji (QJ) population 20 km away driven down the Shazhuyu River by local degraded habitat [25]. In these areas, conservation efforts should be targeted at restoration of degraded habitats and human-gazelle conflicts [25,26]. Pastures of relatively higher quality for Przewalski’s gazelle are mainly identified in areas along the Buha River and those from the northern shore of the Qinghai Lake to the southern foot of Datong Mountain. This is consistent with the pattern predicted using the Maximum Entropy Approach [38,39], enhancing our confidence in the model output.

Our modeling indicated that regional and patch-to-patch connectivity for Przewalski’s gazelle is weak. Various barriers drive gazelles to face local extirpations due to stochastic events and inbreeding depressions [19,24]. On a regional scale, the barrier effects of natural features, including high mountains, lakes and long geographical distances, have existed for hundreds of thousands of years since the species formed [27]. According to these effects, Przewalski’s gazelle could be subdivided into four meta-groups: G1 (YZ, HD, KT, SI, GZ, HG and TL), G2 (BI), G3 (SG and KM) and G4 (WY and RQ). Movements between these meta-groups are few. In particular, the Qinghainanshan Mountain and a new highway along the southern foot of the mountain almost completely impede gazelle movement between Gonghe populations (G4) and those in the Qinghai Lake Basin (G1-G3). Genetic distances between Gonghe populations and the others (G1-G3) are the farthest among all pairwise populations [27]. In another study, evidence from skull morphology suggests that the genetic differentiation has reached the subspecies level in Gonghe populations [40].

Within each meta-group (G1-G4), anthropogenic barriers disable gazelle dispersals along Euclidean distances between patches. The settlement (Hudong) and surrounding farmlands bend the functional connectivity YZ-HD to the foot of a hill where human activities is less dense with scattered pastoralists’ houses. Similarly, connectivity SI-GZ is disrupted by a highway and a tourist attraction (Shaodao). Yang et al. [27] found significant genetic differentiation between these pairwise patches and local pastoralists reported no inter-patch migrants. National highway No. 315 with heavy car traffic runs between TL and the neighboring HG and GZ. The highway, together with the town Ganzihe, presents a barrier reducing gazelle movement. There is also a highway between KT and SI but with less traffic. The movement cost was relatively low and one first-generation migrant from KT to SI was detected [27]. This indicates a traversable connectivity between the two populations. However, the future of matrix permeability between the two remains uncertain due to increasing traffic commensurate with prosperous eco-tourism. The movement cost is the lowest between GZ and HG, which are located close to and separated by one fenced railway (Qinghai-Tibet). Local pastoralists reported gazelles passing through the eight underpasses (<10 m in width) and Yang et al. [27] suggest that the railway does not significantly impact the gene flow. The desert between HD and KT and streamside along Buha River between SG and KM are comparably traversable in that there are few barriers in between. Of the five identified potential patches, P1 has the shortest cost distance to the nearest extant populations (e.g., HG and TL) and P2 follows (in the vicinity of TL and GZ). The connectivity WY-RQ is relatively straight due to the homogeneity of the landscape across the area. The degraded grassland in the basin may unintentionally facilitate movement between the two patches.

The reduced landscape connectivity for Przewalski’s gazelle could be enhanced in case of implementing appropriate restoration programs. Habitat restoration inside and between patches should be a priority [13,25,38]. Overgrazing should be controlled by reducing the size of domestic livestock. Human activities inside or near the 1-km-corridors, including settlements, tourism attractions and livestock herding, should be controlled to a lesser degree. Following habitat restoration, inter-patch corridors could be restored wide enough that they could be more resilient to edge effects, climatic changes and anthropogenic influences [2,17]. Barrier effect of transportation routes could be mitigated by construction of highway crossing structures, which have been applied successfully in many other species [41,42,43,44]. However, there are currently no under- or over-crossing structures on the highways in the Qinghai Lake region. The existing railway underpasses are either not specially designed for wild ungulates. An immediate recommendation is to construct wider crossing structures where least-cost corridors intersect with transportation routes. Additionally, the conservation of northern populations (HG, GZ and TL) should be prioritized in the sense of preserving a viable source of the gazelle metapopulation. These populations have nearly half of the gazelles (736/1544=47.7%), and the populations are increasing [25]. If the connectivity from these populations is preserved or regenerated, they will be beneficial in maintaining genetic diversity, sustaining extant metapopulation and occupying new potential habitats to the north.

There exists a common controversy over the ‘subjective translation’ of the expert-opinion and literature method [17]. However, this method has been used in many connectivity studies for a wide range of vertebrates and invertebrates [17,45] and found to be a good fit to independent empirical data [13,15,16]. The knowledge about Przewalski’s gazelle of the experts we invited in this study help improves the model. Another limitation is that we did not include the effect of grassland fencing which does harms to gazelles [25,46,47]. It was impractical to locate or model the spatial configuration of the fencing at a regional scale. However, we argue that the exclusion has little effect on the model outputs because grassland fencing is distributed homogeneously inside and between patches.

The functional connectivity modeled here is really a starting point rather than a final solution. It is a putative model based on expert opinions and literature reviews rather than depicted using large frequently sampled dataset of locations obtained by GPS or radio telemetry [1,12]. We concur that Przewalski’s gazelle does not necessarily use the least-cost corridors given the complex and stochastic attributes of animal movement [2,4]. The modeling, however, identified a swath of lands with higher permeability, providing the most potential opportunities for successful dispersals between patches. Findings in landscape genetics [27], our previous field work and allegations by local pastoralists have provided first-step validation of the connectivity pattern. Before corridors are finalized, model outputs should be further assessed by field surveys of actual use of corridors by Przewalski’s gazelle. Following this, we encourage studies on gazelle movement with the help of GPS telemetry.

As Mallon and Jiang [48] pointed out: conservation of wild grazing ungulates in Central Asia in the long run presents a challenge, especially when considering migratory behavior of the species in nomadic landscape. However, conservation biology addresses crises in biodiversity preservation [49] and we should take immediate actions when there remain populations of Przewalski’s gazelle and some suitable habitat remnants. Such conservation impetus may also be provided by the fact that the gazelle still has an evolutionary potential and moderate nuclear genetic diversity in spite of a genetic bottleneck [27,28]. The habitat suitability and functional connectivity modeled in this study provides conservation practitioners with an initial, but thorough and visualized, look at the connectivity network for Przewalski’s gazelle. In conjunction with previous research findings on the gazelle [25,26,27,38,39,50,51], this study has direct applications to decision making process in which limited funding could be targeted at those patches and corridors identified as conservation priorities. In a general sense, we suggest that decision makers should involve habitat connectivity in conservation planning for species with similar biological and ecological characteristics living in patchy landscape.

Acknowledgments

We thank the supports from the Administration of Qinghai Lake National Nature Reserve and Wildlife and Nature Reserve Management Bureau of Qinghai Province. We are grateful to the following gazelle experts for their opinions on the parameter assignments: Liu Zhensheng, Liu Bingwan, Li Zhongqiu, Yang Ji, Mardan Turghan, Ping Xiaoge, Luo Zhenhua and Chen Jing. We also thank Tang Songhua, Li Feng, Liu Zhao, Zhu Jiawei and Li Linlin for kind suggestions during the modeling procedure. We appreciate the help of Jennifer Chin in Duke University in USA on earlier drafts of the manuscript.