Habitat suitability models are useful to understand species distribution and to guide management and conservation strategies. The grey wolf (Canis lupus) has been extirpated from most of its historic range in Pakistan primarily due to its impact on livestock and livelihoods. We used non-invasive survey data from camera traps and genetic sampling to develop a habitat suitability model for C. lupus in northern Pakistan and to explore the extent of connectivity among populations. We detected suitable habitat of grey wolf using a maximum entropy approach (Maxent ver. 3.4.0) and identified suitable movement corridors using the Circuitscape 4.0 tool. Our model showed high levels of predictive performances, as seen from the values of area under curve (0.971±0.002) and true skill statistics (0.886±0.021). The main predictors for habitat suitability for C. lupus were distances to road, mean temperature of the wettest quarter and distance to river. The model predicted ca. 23,129 km2 of suitable areas for wolf in Pakistan, with much of suitable habitat in remote and inaccessible areas that appeared to be well connected through vulnerable movement corridors. These movement corridors suggest that potentially the wolf range can expand in Pakistan’s Northern Areas. However, managing protected areas with stringent restrictions is challenging in northern Pakistan, in part due to heavy dependence of people on natural resources. The habitat suitability map provided by this study can inform future management strategies by helping authorities to identify key conservation areas.
Citation: Kabir M, Hameed S, Ali H, Bosso L, Din JU, Bischof R, et al. (2017) Habitat suitability and movement corridors of grey wolf (Canis lupus) in Northern Pakistan. PLoS ONE 12(11): e0187027. https://doi.org/10.1371/journal.pone.0187027
Editor: Bi-Song Yue, Sichuan University, CHINA
Received: June 11, 2017; Accepted: October 12, 2017; Published: November 9, 2017
Copyright: © 2017 Kabir et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper (Please, see Table 1 and figures).
Funding: Main funding for this study came from the Research Council of Norway (Grant 204202/F20) and Snow Leopard Trust. The principal investigator were Dr. Richard Bischof and Dr. Muhammad Ali Nawaz. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
The distribution of species in space and time is a central topic in ecology. Species distribution models (SDMs) are increasingly important for investigating the requirements of species and for conservation planning [1–5]. Such models provide valuable quantitative information on the threats, such as areas of where there is high risk from humans, or where are the required resources  and they help identify conservation priorities [7–11].
The conservation of large carnivores remains challenging, in part due to a poor understanding of the complex spatial dynamics that facilitate population persistence . The habitat requirements of such species deserve particularly close attention because they generally require large home ranges, are negatively impacted by changes in land use and are killed because of the threats they pose to livelihoods [13–16]. The grey wolf (Canis lupus) is a prime example of the drastic reduction in former ranges as a result of intensive persecution. Wolves were once widely distributed throughout the Palearctic and Nearctic biogeographic areas [17, 18]. However, global wolf range has decreased by 33% over the last century . In many areas, agricultural expansion into marginal areas of wolf habitat has increased depredation of livestock and subsequently increased poaching, resulting in a numerical and spatial contraction of grey wolf populations .
A major concern of modern conservation efforts is identifying remaining habitat that is suitable for a species to occupy [19–23]. SDMs have proven to be effective at predicting habitat suitability for large carnivores, including wolves [24–26]. Habitat and conflict management can be implemented using the results from monitored wolf populations once potential areas are identified. In addition, knowledge of potentially suitable wolf habitat can be integrated into landscape planning . Indeed, evidence from elsewhere suggests that map-based conservation planning can help facilitate human-wolf coexistence by identifying areas where the potential conflict caused by livestock depredation is high [27, 28].
Several studies have shown that the long-term survival of large vertebrates is achieved by both protecting source populations and providing dispersal opportunities between suitable patches [29,30]. Ecological corridors can help them connect local populations, allowing individuals free dispersal between populations . Wolf dispersal patterns across the landscape can better predict where new wolf populations may appear . Animals use a wide variety of mechanisms to select suitable habitat and being aware of habitat use details is important for corridor design . Connectivity analysis is particularly important for wolves because allow them to know this animal can move through the existing habitat .
Wolf populations in Pakistan have suffered population declines and range contraction [34, 35]. They are now confined to remote, barren, mountainous regions and extensive deserts . Numerous factors are thought to be responsible for their decline. The expansion of agricultural practices and land conversion has caused habitat loss. The movement of herders up the altitudinal gradient because of climate warming has further reduced available habitat and increased the impact of retaliatory killings of wolves. These predators move to lower altitudes during heavy snowfall, further increasing the chances of being killed due to livestock depredation .
Although a several studies have addressed GIS and modelling analyses of C. lupus in different area in the world [24–26], no research has been conducted on grey wolf in northern Pakistan. Our goal was to model, through the use of non-invasive survey data from camera traps and genetic sampling, a habitat suitability model for grey wolves in northern Pakistan and to explore the extent of connectivity between populations. We identified: a) the first geographical distribution analysis for C. lupus in northern Pakistan and which ecological factors may be limiting the species distribution in the study area; b) the corridors in northern Pakistan where the landscape would facilitate dispersal of C. lupus to provide an understanding of landscape permeability for large carnivores in a largely unsuitable matrix and to present conservation agencies with useful information should grey wolves continue to disperse into the region.
Materials and methods
The Northern Areas of Pakistan (35-37o N and 72-75o E) are dominated by glaciated mountains crests, narrow valleys, ravines, cliffs and rough ridges . They fall in the watersheds of the Karakorum, Himalaya and Hindu Kush mountain ranges (Fig 1). The Karakoram-Pamir landscape is a combination of several unique agro-ecological units that merge into one another . Climatic conditions vary widely, ranging from the monsoon-influenced moist temperate zone in the western Himalayas to the semi-arid cold deserts of the northern Karakorum and Hindu Kush . Four vegetation zones can be distinguished along the altitudinal gradients, namely alpine dry steppes, subalpine scrub zones, alpine meadows and permanent snowfields . These varied climatic conditions and ecosystems support rare and endangered animals such as the snow leopard (Uncia uncia), brown bear (Ursus arctos), grey wolf (Canis lupus), Himalayan lynx (Lynx lynx), Marco Polo sheep (Ovis ammon polii), musk deer (Moschus chrysogaster), blue sheep (Pseudois nayaur), Himalayan Ibex (Capra ibex sibirica), flare-horned markhor (C. f. cashmirensis), Ladakh urial (Ovis orientalis vignei) and woolly flying squirrel (Eupetaurus cinereus).
Genetic sample collection
The field surveys were conducted during the period 2010–2012 within the following protected areas (PAs): Machiara National Park, Musk Deer National Park, Khunjerab National Park, Broghil Valley National Park, Qurumber National Park, Shandoor-Handrab National Park, and Deosai National Park, as well as outside the PAs in Laspur, Yarkhun, Misgar, and Chapursan, covering suspected wolf range. We covered seven national parks and eight non-protected study sites. Survey areas were divided into grids cells of 5 × 5 km (except in Khunjerab National Park and Shimshal where the grid size was taken as 10 × 10 km) on GIS maps. Surveys points were randomly selected within each grid cell and a 50 m radius around each point was searched for wolf scats. We searched 1,736 points within the study area. We also searched for scats at camera trap locations and whilst hiking along livestock trails and manmade tracks (which were also used by wildlife). About 1,000 scat samples of carnivores were collected and preserved in 20 ml bottles filled with 95% alcohol.
We confirm that: 1) the institution leader of this research (Carnivore Conservation Lab, Department of Animal Sciences, Quaid-i-Azam University, Islamabad) has all field permit to work in all protected areas; 2) the owner of the land in non-protected study sites gave us the authorization to conduct the study on this site; and 3) no wolfs were harmed and sacrificed during this research. We only have collected wolf scats and image by camera traps.
DNA analysis and species identification
Deoxyribonucleic acid (DNA) was extracted from 15 mg of feces using the DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany) recovered in a total volume of 200 μL. Blank extractions were systematically performed to monitor possible contaminations. Species identification was performed through next generation sequencings (NGS) by amplifying DNA extract using primer pair 12SV5F (5’-TAGAACAGGCTCCTCTAG-3’) and 12SV5R (5’- TTAGATACCCCACTATGC-3’)  targeting about 100-bp of the V5 loop of the mitochondrial 12S gene . The sequencing was carried out on the Illumina Genome Analyzer IIx (Illumina Inc.), using the Paired-End Cluster Generation Kit V4 and the Sequencing Kit V4 (Illumina Inc.), following the manufacturer’s instructions. The sequence reads were analyzed using OBI Tools (http://www.prabi.grenoble.fr/trac/OBITools). Taxon assignation was achieved using the ecoTag program  in comparison with a reference database for vertebrates. This reference database was built by extracting the relevant part of the mitochondrial 12S gene from the European Molecular Biology Laboratory’s (EMBL) nucleotide library using the ecoPCR program . Genetic results revealed identification of 80 samples belonging to wolves.
Camera traps were installed in 798 locations during the period 2009–2017 and were separated by a horizontal buffer of at least 1 km (Table 1). Camera trap locations were identified based on landscapes characteristics—ridges, cliff bases, draw—preferred by carnivores and the presence of carnivore signs . A single motion-triggered digital camera with infrared flash (HC500/PC900, Reconyx, Holmen, WI, USA) was deployed at each location on a steel pole (50–60 cm) driven into the ground. Camera traps were set to take consecutive images (1-s picture interval) when triggered and were typically kept active at a given location for 10–40 days . Camera trap sites were baited with fish oil. Commercial trapping scent lures were deployed in some randomly selected sites .
Model preparation: Selection of presence data and environmental variables
Records obtained by scats collection and camera trapping of C. lupus were screened in ArcGis (version 9.2) for spatial autocorrelation using average nearest neighbour analysis to remove spatially correlated data points and guarantee independence [46–48]. After this selection, from an initial dataset of ca. 131 presence records, only 25 unrelated locations were used to generate current SDMs of C. lupus.
To produce SDMs for C. lupus in northern Pakistan, we considered initially a set of 28 environmental variables. We included altitude, 19 bioclimatic variables, land cover, slope, soil type, distance to roads, distance to rivers, distance to settlements, vector ruggedness measure and normalized difference vegetation index. Bioclimatic variables and altitude were obtained from WorldClim database (www.worldclim.org/current) . Land cover was obtained from the Global Land Cover 2000 (available from https://lta.cr.usgs.gov/glcc/globdoc2_0). Distance to roads, distance to rivers, distance to settlements were calculated by using Euclidean distance tool in Arc GIS 10.0. Soil (FAO, 2003, digital soil map of the world), vector ruggedness measure (SRTM 90m DEM by Center for Nature and Society, Peking University.) and normalized difference vegetation index were obtained from NASA website (http://modis-land.gsfc.nasa.gov/vi.html). The MODIS normalized difference vegetation index product is computed from atmospherically corrected bi-directional surface reflectance that have been masked for water, clouds, heavy aerosols and cloud shadows. Global MOD13A2 data are provided every 16 days at 1-kilometer spatial resolution as a gridded level-3 product in the Sinusoidal projection.
In order to remove any variables that were highly correlated before generating the models, we calculated a correlation matrix using Pearson’s technique and selected only the variables for which r < 0.70 . From this first set of predictors, we selected only the variables that were most representative of the species’ ecological requirements [12, 16, 17, 18, 20, 24, 25, 26, 33, 34]. After this analysis, eight environmental variables were selected considering their applicability to the scale of our study area, relevant predictive power, and their suspected biological importance [51, 52]. All the variables were prepared—conforming cell size [30-arc second resolution (0.93 × 0.93 km = 0.86 km2 at the equator)], geographic extent, projection, and ASCII—using the ‘resample’, ‘clip’, ‘mask’, and ‘conversion’ tools in Arc GIS 10.0. Finally, the following eight environmental variables were used for model training: distance to roads (m), distance to rivers (m), mean temperature of wettest quarter (°C), mean diurnal range (°C), soil, annual precipitation (mm), altitude (m) and global land cover 2000.
SDMs rely on presence-absence data or presence-only data [7, 9, 53–55]. The use of presence-only is recommended when absence data has a high degree of uncertainty relative to presence data, which is especially true when detection rates are poor [25,26]. We modeled wolf distribution using Maxent (ver. 3.4.0) as it is recognized as a better performer with presence-only data, especially with small numbers of occurrence points [56–58]. To build the models, we used the presence records (defined “sample” in Maxent) of C. lupus selected as described above and the environmental variables (defined “environmental layers” in Maxent). In the setting panel, we selected the following options: auto features; random seed; write plot data; remove duplicate presence records; give visual warming; show tooltips; regularization multiplier (fixed at 1); 10,000 maximum number of background points; 1,000 maximum iterations; and, finally, we achieved a 20 replicates effect with cross-validation run type as suggested by Pearson et al.  for testing small samples, this run type makes it possible to replicate n sample sets removing a locations at each step [47, 60, 61]. All other parameters were left by default. These settings are conservative enough to allow the algorithm to get close to convergence and optimize performance .
The final logistic output gave suitability values from 0 (unsuitable habitat) to 1 (suitable habitat). The 10 percentile training presence (i.e. the value above which the model classifies correctly 90% of the training locations) was selected as the threshold value for defining the species’ presence. This is a conservative value commonly adopted in species distribution modelling studies, particularly those relying on datasets collected over a long time by different observers and methods [11, 47]. This threshold was used to reclassify our model into binary presence/absence map.
We used Jackknife sensitivity analysis to estimate the actual contribution that each variable provided to the geographic distribution models. During this process, Maxent generated three models: first, each environmental variable was excluded in turn and a model was created with the remaining variables to check which one of the latter was the most informative. Second, a model was created by individually by adding each environmental variable to detect which variable had the most information not featuring in the other variables. Third, a final model was generated based on all variables. Response curves derived from univariate models were plotted to know how each environmental variable influence presence probability.
AUC assesses the discrimination ability of the models and its value ranges from 0 (equaling random distribution) to 1 (perfect prediction). AUC values > 0.75 correspond to high discrimination performances . TSS compares the number of correct forecasts, minus those attributable to random guessing, to that of a hypothetical set of perfect forecasts. It considers both omission and commission errors and success as a result of random guessing; its values range from -1 to +1, where +1 corresponds to perfect agreement and zero or less to a performance no better than random .
Modelling potential movement corridors
A spatial corridor model was developed using the distribution map of wolves in Circuitscape 4.0 software (http://www.circuitscape.org/downloads) . We used Circuitscape 4.0 to model connectivity and movement corridors of grey wolf in Pakistan across the landscape. Circuitscape treats the landscape as a conductance surface, where each pixel represents a resistor with an assigned resistance value. Pairwise electrical resistances between locations are calculated by running a theoretical electrical current between each population pair, with one population being set as the current source and the other as the ground . Contrary to least cost resistance methods, Circuitscape does not assume that animals disperse according to previous knowledge of the surroundings, but is based on random walks . It thus links populations through multiple pathways , such that connectivity between habitat patches increases according to the number of connected pathways, and the effective resistance between two populations is derived from the overall resistance across all pathways. We used SDM output as conductance layer and 24 nodes to run movement corridors of grey wolf in Circuitscape 4.0. The nodes were used to represent different areas where we have confirmed wolf presence in northern Pakistan. Clearly, we have not used all the nodes to run Circuitscape because otherwise this procedure would become too complex. We have used a very low number of nodes and have chosen them as the most important areas of wolf movement in northern Pakistan. We converted the nodes into a grid file in Arc GIS 10.0. Both the habitat suitability map (created by Maxent) and the nodes file were converted into ASCII format for a Circuitscape model run. We used the option of conductance instead of resistance because the landscape is represented as a conductive surface with low resistances assigned to landscape feature types [9, 66, 67]. Finally, the final map of movement corridors was reclassified into three categories, respectively, low, moderate and high to better represent the most important areas for C. lupus movements.
Camera traps and genetic analysis
We obtained 51 wolf presence records from camera traps and 80 from the genetic analysis of scat samples collected from the distribution range of wolves (Fig 2). Most presence records were obtained from national parks. Presence records along the altitudinal gradients ranged from 3,000 m (Musk Deer National Park) to 4,700 m (Khunjerab National Park). PAs with higher frequency of presence records were Deosai National Park (Himalayan range), Chitral Gol National Park (Hindu Kush range), Khunjerab (Karakorum range) and Broghil National Park (Pamir range). Outside the PAs, the highest wolf encounter was recorded from the Khanbari Valley in Gilgit-Baltistan. There was no presence record from Terich, Astor, Misgar, Chipurson, Shimshal and Hisper Hooper Valley. Overall, wolf detection was low, suggesting thin and patchy population.
Habitat suitability and model validation
The Maxent model suggested that there was suitable wolf habitat within the areas chosen as suspected wolf habitat range (Fig 3). The binary map discriminated between areas typically used by wolves and those considered unsuitable (Fig 4). The most suitable areas identified from the models were located predominantly within PAs and most inaccessible areas with minimum human disturbance, and overall, mainly along the narrow valley and around summer livestock pastures. The model suggested that there was less suitable habitat in lower altitude areas with more human access.
Scales show the probability of presence ranging from 0 to 1.
White = Absence; Red = Presence.
In relation to the distribution range, suitable areas were quantified based on habitat suitability modelling (Fig 4). The model predicted ca. 23,129 km2 of potential distribution of wolf in Northern Pakistan.
The jackknife-cross-evaluation test yielded the relative contribution and permutation of each environmental variable using Maxent. Distances to road, mean temperature of the wettest quarter and distance to river contributed most to the model. Soil, altitude, annual precipitation and land cover contributed relatively little. Response curves showed how the logistic prediction changed as environmental variables varied while keeping all other environmental variables at their average sample values. The probability of wolf presence was 0.7 at 0 km2 of distance from the road and stream, but it drastically to 0.05 at a distance of 0.5 km2. Mean temperature of the wettest quarter ranged from -10°C to 10°C with a max value (0.6) of probability of presence around 0°C.
The jackknife test also revealed the importance of different variables and their impact on model efficacy. Distance to river was the most important variable in determining model prediction in training, test and AUC evaluation. The distance to river increased the gain more than any other variables when added in isolation. On the other hand, the mean temperature of the wettest quarter variable decreased the gain most when omitted i.e. it contains the most information not present in other variables (Fig 5).
It shows how important each variable is in explaining wolf presence when used separately (cobalt blue), and how the model is affected when each variable is left out (aqua). Dark blue bars = Importance of single variable, light blue bars = loss in model gain, when variable is omitted. Red bar = total model gain. Alt = Altitude; bio_12 = Annual Precipitation; bio_2 = Mean Diurnal Range (Mean of monthly); bio_8 = Mean Temperature of Wettest Quarter; glc2000 = Global land cover 2000; road_dist = Distance to roads; river_dist = Distance to rivers; Soil = Soil.
Finally, our model showed high levels of predictive performances as can be seen from the values of area under curve (0.971±0.002) and true skill statistics (0.886±0.021).
Potential movement corridors
The corridor modelling generated estimates of habitat connectivity among scattered wolf populations in northern Pakistan (Fig 6). Four chunks of suitable habitat were identified within the Himalayas, Pamirs, Hindukush, and Karakorum mountains ranges of northern Pakistan. Sub-populations have strong, but unprotected connections and corridors movement existed between all major areas of wolf habitat. The model identified weak linkages between populations found at lower altitudes with high disturbance rates. Among the PAs, Chitral Gol, Broghil and Qurumber National Park had wide potential corridors comprised of suitable habitat connecting core areas. Furthermore, the analysis revealed appropriate dispersal habitat between Musk Deer and Deosai National Park, and similarly between Qurumber, Broghil, and Khunjerab National Park to allow for wolf dispersal.
Maxent model and movement corridors of grey wolf in Pakistan
This study represents the first large-scale assessment of wolf distribution, habitat suitability and movement corridors in Pakistan. Our model identified areas with suitable habitat and corridors through which wolves may travel to reach new territories across northern Pakistan. The model showed considerable predictive performance, showing AUC value > 0.9 that may be placed among the highest in published models [5, 55, 68] and excellent values of the True Skill Statistic, corresponding to a very high predictive capacity [69–71].
Overall, our model found that the most suitable areas for wolves are located in mountainous regions where human disturbance is limited. Yet, it was also clear that wolves are relatively flexible in their use of habitat at the landscape scale. Along the altitudinal gradients, wolf presence was recorded ranging from a moist temperate zone in Musk Deer National Park up to the alpine zone in Khunjerab National Park. In general, wolves could potentially live in any area where human tolerance and prey populations are adequate to support viable numbers . Wolves show different patterns of habitat selection based on time (year, season, time of day) and areas in which they were observed [73, 74]. Our data supported previous observations that wolves occur in various types of habitat and shows low habitat specificity and high levels of ecological resilience compared with other large carnivores [34, 72, 75].
Wolf habitat selection patterns at a fine scale appear to be influenced by complex interactions between habitat attributes and human disturbances [73, 76]. We found that wolf presence depended on the type of anthropogenic disturbance in the area. Most roads in the study area are unpaved with minimum disturbance and traffic pressure. We observed that wolves avoided main roads and tracks, but followed livestock tracks and small, unpaved roads with low disturbance. We suspect that wolves use roads for traveling, scent-marking, and territorial patrolling, but have also developed cryptic behavioral responses to roads, likely driven by the increased risks associated with human presence [77, 78]. Wolves are likely to select secondary gravel or unpaved roads for hunting due to the greater visibility and mobility . The presence of road networks may drive wolves toward suitable habitat types. Wolf’s tolerance to human disturbances increased in suitable habitat types .
The results of our model suggested that the distance to road was an important predictor of wolf presence [76, 80, 81]. This finding suggested that wolves may seek to minimize the probability of encountering humans this [80, 81].The proximity to rivers was the second most important predictor. Riparian habitats provide wolves with increased opportunities to hunt wild prey and are also important in den selection [82, 83]. In our landscape shepherds prefer moving along streams and established temporary stay which may also attracts wolves to feed on livestock. Wolves were concentrated in lower areas due to snow-caused aggregation of prey during the winter season. Frozen rivers and lakes are often used by wolves to travel faster  . Others environmental variables such as altitude, annual precipitation and land cover were among the variables that contributed least to the SDMs for C. lupus in northern Pakistan.
We found that summer huts, temporary settlements, and grazing pastures limited wolf distribution. Wolves exhibited some tolerance towards humans, enabling them to persist within a mosaic of human-altered and naturally occurring habitat. Wherever primary habitat is rare, wolves tend to be dispersed in meadows and rangelands, or in less-natural landscapes such as mixed-use agro ecosystems [86,87]. A similar relationship between the number of inhabitants of settlements and avoidance of close surroundings by wolves was observed in Poland . Wolf populations in closer contact with human-active areas indicated tolerance to human activities .
Resistance modelling indicated the presence of habitat corridors for wolves in northern Pakistan. These corridors could link potential habitats and movement corridor between PAs. Habitat connectivity is not uniform in the Himalayas and population connectivity between the Pamirs and Himalayas range is very weak i.e. based on our corridor modelling analysis. Wolf populations in the Hindu Kush appear to be well connected with the population of the Pamirs and Karakorum’s, which is also enhanced by the establishment of PAs, including the Broghil, Qurumber and Khunjerab National Parks. PAs have become islands of habitat within a mosaic of agriculture and development, and although at a slower rate than non-PAs, anthropogenic activities persist even within the boundaries of PAs . Chitral, Broghil and Qurumber were identified as areas of likely wolf activity based on habitat quality and connectivity to other patches of high-quality habitat. The wolf population in Deosai appears to be connected with the population of Central Karakorum National Park, but only weak connected with other potential habitat. The wolf populations at the Musk Deer and Khanbari study sites appear to be isolated.
Highly suitable habitat was also detected outside of PAs with minimum levels of anthropogenic activities. For large carnivores, sub-optimal habitats might serve as corridors linking habitats necessary for survival and reproduction and also to prevent inbreeding depression . Grey wolves are able to travel through habitats considered poor in the search for an area to form a new pack . “Pioneering” wolves have been known to disperse over large distances and settle in new habitats far from the nearest source population [34, 93]. The populations of the Hindu Kush, Pamirs and Karakorum appear to be connected through movement corridors, but these needs to be protected to facilitate safe use by dispersing wolves.
There are two main limitation to our model. First, no prey availability estimates were available. Second, data was only collected during winter and wolf habitat selection patterns may vary between seasons. The dataset was influenced by the species’ patchy distribution, its rarity throughout the landscape, its seasonal surface occurrence, and its location (when active) on steep and rocky (often impassable) terrain . Previous studies showed that wolf distribution at the landscape scale was influenced primarily by prey availability and human infrastructure . Assuming prey biomass varies with habitat type, studies on carnivores demonstrate the potential for deriving accurate habitat and connectivity models [96, 97].
Application of habitat modelling
With respect to species presence, the model predicted habitat suitability reasonably well. Among the four major chunks of high-quality habitat identified, one is protected, one is partially protected, and the rest are weakly protected. The most suitable habitat in the Hindu Kush range (Khanbari) lacks PAs and has poor connections with other populations. The eastern part of the Pamir range and southern Himalayas are partially connected with the western Karakorum and northern Himalayan populations, respectively. Our study provides a better idea of where wolves may disperse to in case numbers increase in the future, and help to identify priority areas for community engagement, management zones and proactive planning .
Habitat models developed in the current study will support wolf conservation in three ways. First, habitat maps provide a tool to identify suitable habitat and movement corridors and provided a guide map for investing limited conservation resources. Second, wildlife managers prioritize the establishment of more PAs covering suitable habitat and movement corridors to extend the PA network for the long-term survival of wolf populations in Pakistan. Third, it is a challenging issue in northern Pakistan to manage protected areas in stringent categories, due to heavy dependence of people on natural resources.
We acknowledge the logistics support of provincial wildlife departments of Gilgit-Baltistan, Khyber-Pakhtunkhwa, and Azad Jammu and Kashmir, and local communities. Field work would not have been possible without the support of the Snow Leopard Foundation in terms of equipment and trained field staff (particularly Muhammad Younas, Muhammad Ayoub, Khurshid Ali Shah, and Siraj Khan). Mr Doost Ali provided much needed GIS support and data. We are grateful to the anonymous reviewers for the very constructive comments that allowed us to improve a first manuscript version.
- 1. Bosso L, Mucedda M, Fichera G, Kiefer A, Russo D. A gap analysis for threatened bat populations on Sardinia. Hystrix. 2016;27(2).
- 2. Hirzel AH, Helfer V, Metral F. Assessing habitat-suitability models with a virtual species. Ecol Model. 2001;145(2–3): 111–121.
- 3. Jarvis AM, Robertson A. Predicting population sizes and priority conservation areas for 10 endemic Namibian bird species. Biol Cons. 1999;88(1): 121–131.
- 4. Sheehan KL, Esswein ST, Dorr BS, Yarrow GK, Johnson RJ. Using species distribution models to define nesting habitat of the eastern metapopulation of double‐crested cormorants. Ecol Evol. 2017;7(1): 409–418. pmid:28070303
- 5. Smeraldo S, Di Febbraro M, Ćirović D, Bosso L, Trbojević I, Russo D. Species distribution models as a tool to predict range expansion after reintroduction: A case study on Eurasian beavers (Castor fiber). J Nat Conserv. 2017;37: 12–20.
- 6. Guisan A, Thuiller W. Predicting species distribution: Offering more than simple habitat models. Ecol Lett. 2005;8(9): 993–1009.
- 7. Oberhauser K, Wiederholt R, Diffendorfer JE, Semmens D, Ries L, Thogmartin WE, et al. A trans‐national monarch butterfly population model and implications for regional conservation priorities. Ecol Entomol. 2017;42(1): 51–60. 0.1111/een.12351.
- 8. Rhodes JR, Wiegand T, Mcalpine CA, Callaghan J, Lunney D, Bowen M, et al. Modeling species' distributions to improve conservation in semiurban landscapes: Koala case study. Conserv Biol. 2006;20(2): 449–459. pmid:16903106
- 9. Roscioni F, Rebelo H, Russo D, Carranza ML, Di Febbraro M, Loy A. A modelling approach to infer the effects of wind farms on landscape connectivity for bats. Landsc Ecol. 2014;29(5): 891–903.
- 10. Roscioni F, Russo D, Di Febbraro M, Frate L, Carranza M, Loy A. Regional-scale modelling of the cumulative impact of wind farms on bats. Biodivers Conserv. 2013;22(8): 1821–1835.
- 11. Russo D, Di Febbraro M, Cistrone L, Jones G, Smeraldo S, Garonna AP, et al. Protecting one, protecting both? Scale‐dependent ecological differences in two species using dead trees, the rosalia longicorn beetle and the barbastelle bat. J Zool. 2015;297(3): 165–175.
- 12. Hinton JW, Proctor C, Kelly MJ, van Manen FT, Vaughan MR, Chamberlain MJ. Space use and habitat selection by resident and transient red wolves (Canis rufus). PLoS ONE. 2016;11(12): e0167603. pmid:28002495
- 13. Enserink MV, Vogel G. The carnivore comeback. Science. 2006;314: 746–749. pmid:17082433
- 14. Gordon IJ. What is the future for wild, large herbivores in human-modified agricultural landscapes? Wildl Biol. 2009;15(1): 1–9.
- 15. Doherty TS, Dickman CR, Johnson CN, Legge SM, Ritchie EG, Woinarski JCZ. Impacts and management of feral cats Felis catus in Australia. Mammal Rev. 2017;47(2): 83–97.
- 16. Newsome TM, Boitani L, Chapron G, Ciucci P, Dickman CR, Dellinger JA, et al. Food habits of the world's grey wolves. Mammal Rev. 2016;46(4): 255–269.
- 17. Vila C, Amorim IR, Leonard JA, Posada D, Castroviejo J, Petruccia-Fonseca F, et al. Mitochondrial DNA phylogeography and population history of the grey wolf Canis lupus. Mol Ecol. 1999;8(12): 2089–2103. pmid:10632860
- 18. Mech LD, Boitani L. Wolves: Behavior, ecology, and conservation. 1st ed. Chicago: University of Chicago Press; 2003.
- 19. Mech LD. The wolf: Ecology and behavior of an endangered species. 1st ed. Minneapolis: University of Minnesota Press; 1981.
- 20. Mesler JI. Modeling habitat suitability and connectivity of gray wolf (Canis lupus) populations in the Pacific Northwest. M.Sc. Thesis, Humboldt State University. 2015. Available from http://humboldt-dspace.calstate.edu/bitstream/handle/10211.3/150029/Mesler_Jacob_I_SU2015.pdf;sequence=1.
- 21. Hilty JA, Lidicker WZ Jr, Merenlender A. Corridor ecology: The science and practice of linking landscapes for biodiversity conservation. Washington D.C.: Island Press; 2012.
- 22. Lindenmayer DB, Fischer J. Habitat fragmentation and landscape change: An ecological and conservation synthesis. 2nd ed. Washington D.C.: Island Press; 2013.
- 23. Saura S, Bodin O, Fortin M-J. Stepping stones are crucial for species' long‐distance dispersal and range expansion through habitat networks. J Appl Ecol. 2014;51(1): 171–182.
- 24. Hendricks SA, Clee PRS, Harrigan RJ, Pollinger JP, Freedman AH, Callas R, et al. Re-defining historical geographic range in species with sparse records: Implications for the Mexican wolf reintroduction program. Biol Conserv. 2016;194: 48–57.
- 25. Otis J-A, Thornton D, Rutledge L, Murray DL. Ecological niche differentiation across a wolf‐coyote hybrid zone in eastern North America. Divers Distrib. 2017;23(5): 529–539.
- 26. Subba SA, Shrestha AK, Thapa K, Malla S, Thapa GJ, Shrestha S, et al. Distribution of grey wolves Canis lupus lupus in the Nepalese Himalaya: Implications for conservation management. Oryx. 2017;51(3): 403–406.
- 27. Carroll C, Zielinski WJ, Noss RF. Using presence/absence data to build and test spatial habitat models for the fisher in the Klamath Region, U.S.A. Conserv Biol. 1999;13(6): 1344–1359.
- 28. Mech LD, Harper EK, Meier TJ, Paul WJ. Assessing factors that may predispose Minnesota farms to wolf depredations on cattle. Wildl Soc Bull. (1973–2006) 2000;28(3): 623–629.
- 29. Margules CR, Pressey RL. Systematic conservation planning. Nature. 2000;405: 243–253. pmid:10821285
- 30. Mech SG, Hallett JG. Evaluating the effectiveness of corridors: A genetic approach. Conserv Biol. 2001;15(2): 467–474.
- 31. Beier P, Noss RF. Do habitat corridors provide connectivity? Conserv Biol. 1998;12(6): 1241–1252.
- 32. Danchin E, Heg D, Doligez B. Public information and breeding habitat selection. In: Clobert J, editor. Dispersal. Oxford: Oxford University Press; 2001. pp. 243–258.
- 33. Ciucci P, Reggioni W, Maiorano L, Boitani L. Long-distance dispersal of a rescued wolf from the Northern Apennines to the Western Alps. J Wildl Manage. 2009;73(8): 1300–1306.
- 34. Mech LD, Boitani L. Wolf social ecology. In: Mech LD, Boitani L, editors. Wolves: Behavior, ecology, and conservation. Chicago: University of Chicago Press; 2003. pp. 1–34.
- 35. Sheikh KM, Molur S. Status and Red List of Pakistan's mammals: Based on conservation assessment and management plan workshop, 18–22 August 2003. Islamabad: IUCN Pakistan. 2004.
- 36. Roberts TJ. The mammals of Pakistan. 2nd ed. Karachi: Oxford University Press. 1997.
- 37. Lovari S, Sforzi A, Scala C, Fico R. Mortality parameters of the wolf in Italy: does the wolf keep himself from the door? J Zool. 2007;272(2): 117–124.
- 38. Bischof R, Ali H, Kabir M, Hameed S, Nawaz MA. Being the underdog: An elusive small carnivore uses space with prey and time without enemies. J Zool. 2014;293(1): 40–48.
- 39. Xu J, Grumbine RE, Shrestha A, Eriksson M, Yang X, Wang Y, et al. The melting Himalayas: Cascading effects of climate change on water, biodiversity, and livelihoods. Conserv Biol. 2009;23(3): 520–530. pmid:22748090
- 40. Jacobose JP. Climatic records from Northern Areas of Pakistan. Culture Area Karakorum Newsletter. 1993;3: 13–17.
- 41. Riaz T, Shehzad W, Viari A, Pompanon F, Taberlet P, Coissac E. ecoPrimers: Inference of new DNA barcode markers from whole genome sequence analysis. Nucleic Acids Res. 2011;39(21): 145–145. pmid:21930509
- 42. Shehzad W, Riaz T, Nawaz MA, Miquel C, Poillot C, Shah SA, et al. Carnivore diet analysis based on next generation sequencing: Application to the leopard cat (Prionailurus bengalensis) in Pakistan. Mol Ecol. 2012;21(8): 1951–1965. pmid:22250784
- 43. Pegard A, Miquel C, Valentini A, Coissac E, Bouvier F, François D, et al. Universal DNA-based methods for assessing the diet of grazing livestock and wildlife from feces. J Agric Food Chem. 2009;57(13): 5700–5706. pmid:19566081
- 44. Ficetola GF, Coissac E, Zundel S, Riaz T, Shehzad W, Bessière J, et al. An in silico approach for the evaluation of DNA barcodes. BMC genomics. 2010;16(11): 434.
- 45. Bischof R, Hameed S, Ali H, Kabir M, Younas M, Shah KA, et al. Using time-to-event analysis to complement hierarchical methods when assessing determinants of photographic detectability during camera trapping. Methods Ecol Evol. 2014;5(1):44–53.
- 46. Bosso L, di Febbraro M, Cristinzio G, Zoina A, Russo D. Shedding light on the effects of climate change on the potential distribution of Xylella fastidiosa in the Mediterranean Basin. Biol Invasions. 2016;18(6): 1759–1768.
- 47. Bosso L, Luchi N, Maresi G, Cristinzio G, Smeraldo S, Russo D. Predicting current and future disease outbreaks of Diplodia sapinea shoot blight in Italy: Species distribution models as a tool for forest management planning. For Ecol Manage. 2017;400: 655–664.
- 48. Kwon H-S, Kim B-J, Jang G-S. Modelling the spatial distribution of wildlife animals using presence and absence data. Contemp Probl Ecol. 2016;9(5): 515–528.
- 49. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A. Very high resolution interpolated climate surfaces for global land areas. Int J Climatol. 2005;25(15): 1965–1978.
- 50. Booth GD, Niccolucci MJ, Schuster EG. Identifying proxy sets in multiple linear regression: An aid to better coefficient interpretation. Ogden: US Dept. of Agriculture; Forest Service, Intermountain Research Station; 1994.
- 51. Guisan A, Zimmermann NE. Predictive habitat distribution models in ecology. Ecol Model. 2000;135(2–3): 147–186.
- 52. Post E, Forchhammer MC. Synchronization of animal population dynamics by large-scale climate. Nature. 2002;420: 168–171. pmid:12432390
- 53. Bosso L, De Conno C, Russo D. Modelling the risk posed by the zebra mussel Dreissena polymorpha: Italy as a case study. Environ Manage. 2017:60(2): 304–313. pmid:28493016
- 54. Ficetola GF, Thuiller W, Miaud C. Prediction and validation of the potential global distribution of a problematic alien invasive species—the American bullfrog. Divers Distribut. 2007;13(4): 476–485.
- 55. Rebelo H, Jones G. Ground validation of presence‐only modelling with rare species: A case study on barbastelles Barbastella barbastellus (Chiroptera: Vespertilionidae). J Appl Ecol. 2010;47(2): 410–420.
- 56. Engler R, Guisan A, Rechsteiner L. An improved approach for predicting the distribution of rare and endangered species from occurrence and pseudo-absence data. J Appl Ecol. 2004;41(2): 263–274.
- 57. Elith J, Graham CH, Anderson RP, Dudík M, Ferrier S, Guisan A, et al. Novel methods improve prediction of species’ distributions from occurrence data. Ecography. 2006;29(2): 129–151.
- 58. Wisz MS, Hijmans RJ, Li J, Peterson AT, Graham CH, Guisan A. Effects of sample size on the performance of species distribution models. Divers Distrib. 2008;14(5): 763–773.
- 59. Pearson RG, Raxworthy CJ, Nakamura M, Peterson AT. Predicting species distributions from small numbers of occurrence records: A test case using cryptic geckos in Madagascar. J Biogeogr. 2007;34(1): 102–117.
- 60. Guo Y, Li X, Zhao Z, Wei H, Gao B, Gu W. Prediction of the potential geographic distribution of the ectomycorrhizal mushroom Tricholoma matsutake under multiple climate change scenarios. Sci Rep. 2017;7(46221). pmid:28393865
- 61. Tang CQ, Dong Y-F, Herrando-Moraira S, Matsui T, Ohashi H, He L-Y, et al. Potential effects of climate change on geographic distribution of the tertiary relict tree species Davidia involucrata in China. Sci Rep. 2017;7(43822). pmid:28272437
- 62. Phillips SJ, Anderson RP, Dudík M, Schapire RE, Blair ME. Opening the black box: an open‐source release of Maxent. Ecography. 2017;40(7): 887–893.
- 63. Fielding AH, Bell JF. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv. 1997;24(1): 38–49.
- 64. Allouche O, Tsoar A, Kadmon R. Assessing the accuracy of species distribution models: Prevalence, kappa and the true skill statistic (TSS). J Appl Ecol. 2006;43(6): 1223–1232.
- 65. McRae BH, Shah VB. Circuitscape user guide. Santa Barbara: The University of California; 2009. Available from: http://circuitscape.org.
- 66. Cushman SA, Landguth EL, Flather CH. Evaluating population connectivity for species of conservation concern in the American Great Plains. Biodivers Conserv. 2013;22(11): 2583–2605.
- 67. Saura S, Estreguil C, Mouton C, Rodríguez-Freire M. Network analysis to assess landscape connectivity trends: Application to European forests (1990–2000). Ecol Indic. 2011;11(2): 407–416.
- 68. Domiguez-Vega H, Monroy-Vilchis O, Balderas-Valdivia CJ, Gienger CM, Ariano-Sánchez D. Predicting the potential distribution of the beaded lizard and identification of priority areas for conservation. J Nat Conserv. 2012;20(4): 247–253.
- 69. Hanspach J, Kühn I, Pompe S, Klotz S. Predictive performance of plant species distribution models depends on species traits. Perspect Plant Ecol Evol Syst. 2010;12(3): 219–225.
- 70. Taylor C, Cadenhead N, Lindenmayer DB, Wintle BA. Improving the design of a conservation reserve for a critically endangered species. PLoS ONE. 2017;12(1): e0169629. pmid:28121984
- 71. Wogan GOU. Life history traits and niche instability impact accuracy and temporal transferability for historically calibrated distribution models of North American birds. PLoS ONE. 2016;11(3): e0151024. pmid:26959979
- 72. Fritts SH, Stephenson RO, Hayes RD, Boitani L. Wolves and humans. In: Mech LD, Boitani L, editors. Wolves: Behavior, ecology, and conservation. Chicago: University of Chicago Press; 2003. pp. 289–316.
- 73. Ciucci P, Masi M, Boitani L. Winter habitat and travel route selection by wolves in the northern Apennines, Italy. Ecography. 2003;26(2): 223–235.
- 74. Milakovic B, Parker KL, Gustine DD, Lay RJ, Walker ABD, Gillingham MP. Habitat selection by a focal predator (Canis lupus) in a multiprey ecosystem of the northern Rockies. J Mammal. 2011;92(3): 568–582.
- 75. Carroll C, Noss RF, Schumaker NH, Paquet PC. Is the return of the wolf, wolverine, and grizzly bear to Oregon and California biologically feasible? In: Maehr DS, Noss RF, Larkin JL, editors. Large Mammal Restoration: Ecological and sociological challenges in the 21st century. Washington D.C.: Island Press; 2001. pp. 25–46.
- 76. Houle M, Fortin D, Dussault C, Courtois R, Ouellet J-P. Cumulative effects of forestry on habitat use by gray wolf (Canis lupus) in the boreal forest. Landsc Ecol. 2010;25(3): 419–433.
- 77. Barja I, de Miguel FJ, Bárcena F. The importance of crossroads in faecal marking behaviour of the wolves (Canis lupus). Naturwissenschaften. 2004;91(10): 489–492. pmid:15349691
- 78. Zimmermann B, Nelson L, Wabakken P, Sand H, Liberg O. Behavioral responses of wolves to roads: Scale-dependent ambivalence. Behav Ecol. 2014;25(6): 1353–1364. pmid:25419085
- 79. Lesmerises F, Dussault C, St-Laurent M-H. Major roadwork impacts the space use behaviour of gray wolf. Landsc Urban Plan. 2013;112: 18–25.
- 80. Whittington J, St Clair CC, Mercer G. Spatial responses of wolves to roads and trails in mountain valleys. Ecol Appl. 2005;15(2): 543–553.
- 81. Gurarie E, Suutarinen J, Kojola I, Ovaskainen O. Summer movements, predation and habitat use of wolves in human modified boreal forests. Oecologia. 2011;165(4): 891–903. pmid:21207232
- 82. Packard JM. Wolf behavior: Reproductive, social, and intelligent. In: Mech LD, Boitani L, editors. Wolves: Behavior, ecology, and conservation. Chicago: University of Chicago Press. 2003; pp. 35–65.
- 83. Latham ADM. Wolf ecology and caribou-primary prey-wolf spatial relationships in low productivity peatland complexes in northeastern Alberta. Ph.D. Thesis, University of Alberta. 2009.
- 84. Musiani M, Okarma H, Jędrzejewski W. Speed and actual distances travelled by radiocollared wolves in Białowieża Primeval Forest (Poland). Acta Theriol. 1998;43(4): 409–416.
- 85. Ansorge H, Kluth, G, Hahne, S. Feeding ecology of free-living wolves Canis lupus in Saxony. In: Macholán M, Bryja J, Zima J, editors. 4th European congress of mammology, Brno, Czech Republic, July 27–August 1, 2003: Program and abstracts and list of participants. Brno: European Mammalogy; 2003. pp. 70.
- 86. Jędrzejewski W, Jędrzejewska B, Zawadzka B, Borowik T, Nowak S, Mysłajek RW. Habitat suitability model for Polish wolves based on long-term national census. Anim Conserv. 2008;11(5): 377–390.
- 87. Treves A, Martin KA, Wydeven AP, Wiedenhoeft JE. Forecasting environmental hazards and the application of risk maps to predator attacks on livestock. BioScience. 2011;61(6): 451–458.
- 88. Theuerkauf J, Jędrzejewski W, Schmidt K, Gula R. Spatiotemporal segregation of wolves from humans in the Białowieża Forest (Poland). J Wildl Manage. 2003;67(4): 706–716.
- 89. Ahmadi M, Kaboli M, Nourani E, Shabani AA, Ashrafi S. A predictive spatial model for gray wolf (Canis lupus) denning sites in a human-dominated landscape in western Iran. Ecol Res. 2013;28(3): 513–521.
- 90. Gaveau DLA, Epting J, Lyne O, Linkie M, Kumara I, Kanninen M, et al. Evaluating whether protected areas reduce tropical deforestation in Sumatra. J Biogeogr. 2009;36(11): 2165–2175.
- 91. Mech LD. Prediction failure of a wolf landscape model. Wildl Soc Bull. 2006;34(3): 874–877.
- 92. Merrill SB, Mech LD. Details of extensive movements by Minnesota wolves (Canis lupus). Am Midl Nat. 2000;144(2): 428–433.
- 93. Wabakken P, Sand H, Liberg O, Bjärvall A. The recovery, distribution, and population dynamics of wolves on the Scandinavian peninsula, 1978–1998. Can J Zool. 2001;79: 710–725.
- 94. Mooney TL. Predicting hydromantes shastae occurrences in Shasta County, California. M.Sc. Thesis, University of Southern California. 2010. Available from http://digitallibrary.usc.edu/cdm/ref/collection/p15799coll127/id/408488.
- 95. Potvin F, Breton L, Courtois R. Response of beaver, moose, and snowshoe hare to clear-cutting in a Quebec boreal forest: A reassessment 10 years after cut. Can J For Res. 2005;35(1): 151–160.
- 96. Linkie M, Chapron G, Martyr DJ, Holden J, Leader-Williams N. Assessing the viability of tiger subpopulations in a fragmented landscape. J Appl Ecol. 2006;43(3): 576–586.
- 97. Wikramanayake E, McKnight M, Dinerstein E, Joshi A, Gurung B, Smith D. Designing a conservation landscape for tigers in human-dominated environments. Conserv Biol. 2004;18(3): 839–844.
- 98. Houts ME. Using logistic regression to model wolf habitat suitability in the northern Rocky Mountains. Banff: World Wolf Congress; 2003.