Anchoring and adjusting amidst humans: Ranging behavior of Persian leopards along the Iran-Turkmenistan borderland

Understanding the space use and movement ecology of apex predators, particularly in mosaic landscapes encompassing different land-uses, is fundamental for formulating effective conservation policy. The top extant big cat in the Middle East and the Caucasus, the Persian leopard Panthera pardus saxicolor, has disappeared from most of its historic range. Its spatial ecology in the areas where it remains is almost unknown. Between September 2014 and May 2017, we collared and monitored six adult leopards (5 males and 1 female) using GPS-satellite Iridium transmitters in Tandoureh National Park (355 km2) along the Iran-Turkmenistan borderland. Using auto-correlated Kernel density estimation based on a continuous-time stochastic process for relocation data, we estimated a mean home range of 103.4 ± SE 51.8 km2 for resident males which is larger than has been observed in other studies of Asian leopards. Most predation events occurred in core areas, averaging 32.4 ± SE 12.7 km2. Although neighboring leopards showed high spatiotemporal overlap, their hunting areas were largely exclusive. Five out of six of leopards spent some time outside the national park, among human communities. Our study suggests that a national park can play an ‘anchoring’ role for individuals of an apex predator that spend some time in the surrounding human-dominated landscapes. Therefore, we envisage that instead of emphasizing either land sharing or land sparing, a combined approach can secure the viability of resilient large carnivores that are able to coexist with humans in the rugged montane landscapes of west and central Asia.


Introduction
Wide-ranging apex predators have spatial needs that may push them to wander beyond the boundaries of protected areas [1,2]. Prey availability and environmental productivity are major factors driving predator space use [3,4]. Predator movement patterns are also regulated by their population density [1,5,6] and climatic disturbance in resource availability [7]. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 findings are relevant for better management of many montane areas, where islands of small reserves are surrounded by densely populated human areas.

Ethics statement
The study was conducted in Tandroueh National Park, Iran. The Iranian Department of Environment reviewed all sampling, trapping and handling procedures and approved permits for the work conducted (93/16270). The trapping and handling protocol was also approved by the University of Oxford's Ethical Review Committee (BMS-ERC-160614).

Study area
We studied leopards in Tandoureh NP, north-eastern Iran (ca. 20 km from the Turkmenistan border) from September 2014. The park has been protected since 1968 and covers 355 km 2 . It is characterized by mountains covered with wormwood Artemisia sp. and scattered juniper trees Juniperus sp. Elevation and annual precipitation range from 1,000 to 2,600 m and 250 to 300 mm, respectively.
There is no human settlement inside the park. Besides the two main cities, i.e. Dargaz (Iran) and Ashgabat (the capital of Turkmenistan) near our study area (Fig 1), local communities live in villages with population ranging between 30 to 400 households. They are mainly sheep and goat herders.
The main prey species for leopards include urial Ovis orientalis, bezoar goat Capra aegagrus, and wild pig Sus scrofa. The availability of leopard prey in Tandoureh is affected by the national park boundaries. Wild medium-sized prey are available only inside the park, whereas domestic animals are found exclusively outside the park. The only exceptions are wild pigs, which are occasionally found in multi-use areas, outside the national park.
For programming the collars' fix rates, we followed Knopff et al. [35] who recommended recording fixes every 3 hours to enable the identification of spatially aggregated GPS points, or clusters. However, to increase fix success rates [36] fixes were taken hourly during the last week of each month. Also, a 'virtual fence' option enabled us to upload the area's boundary, so that when leopards left the defined area fix rate could be increased to hourly. Bjørneraas et al.   [37] recommended that to analyze animal movement and behavior, fixes obtained immediately after collaring should be excluded because the animal is likely to behave abnormally. Therefore, we omitted the first 4 days for all collar data, associated with the earliest known kill made by the leopards after collaring (M1/Borzou).
We also investigated the potential kill sites of collared leopards. Kills were defined by clusters of GPS fixes, i.e. locations where leopards remained overnight (6 PM to 6 AM) within a radius of 200 meters. Candidate GPS clusters were investigated for possible kill remains. Prey species were categorized as "small" as < 15 kg, including red fox Vulpes vulpes, Indian crested porcupine Hystrix indica and birds or "medium" as ! 15 kg, such as urial, bezoar goat, wild pig, domestic sheep Ovis aries and domestic dog Canis familiaris. Young wild ungulates and domestic animals (< 1 year) were also included in medium-sized prey.

Statistical analysis
We screened the data for two types of errors which are typical in GPS locations: missing location fixes (i.e. unsuccessful attempts of a GPS fix) and location errors of successfully acquired fixes (i.e. the difference between the recorded location and the animal's true location) [37]. After removing missing fixes, erroneous locations and outliers were screened based on identification of locations arising from unrealistic movement patterns with minimal loss of data, using a script developed by Bjørneraas et al. [37] implemented in the R environment for statistical computing [38]. We defined conservative movement values for leopards as Δ = 30,000 m; μ = 15,000 m; α = 5000 m/h; θ = −0.97 corresponding to turning angles between 166˚and d 194˚; Δ is a distance threshold over which an individual could not possibly travel between consecutive intervals, μ is a distance that leopard can move between two fixes and α is speed.
Multiple home range estimators are suggested to facilitate comparison with other studies that use just one method. We used three estimators for quantifying home ranges of the leopards: minimum convex polygon (MCP), kernel density estimator (KDE) and auto-correlated KDE (AKDE). Both MCP and KDE are popular for estimating animals' home ranges, but they suffer from fundamental flaws that could degrade data quality. MCP lacks an underlying probabilistic model whereas the kernel is a nonparametric, probabilistic method, which calculates home range area based on the complete utilization distribution (UD, i.e., the probability distribution defining the animal's use of space [39]). However, KDE assumes that the data are independent and identically distributed whereas relocation data that are ordered in time are inherently auto-correlated (i.e. an individual's position, velocity, or acceleration measured at one point in time are statistically correlated with the same measurements in the past and future). Therefore, we also used the recently developed AKDE method, a continuous-time approach which is a fully generalized KDE to account for auto-correlated bivariate Gaussian density estimation for relocation data [40].
For each animal, we plotted an empirical variogram, which is the estimated semi-variance in positions as a function of the time lag separating observations to visually inspect the autocorrelation structure of the relocation data. Upward curvature at zero to short time lags indicates velocity autocorrelation while the long-lag behavior of the variogram illustrates space use. Thus, range residents are expected to reach an asymptote on a timescale that roughly corresponds to the home-range crossing time data [41]. In the absence of proof of range residency, we excluded them from estimating population-level movement metrics.
We used package 'ctmm' version 0.4.0 [41] to perform three movement models. The Independent Identically Distributed (IID) process assumes uncorrelated positions and velocities which is equal to the conventional KDE [40]. The Ornstein-Uhlenbeck (OU) process combines a random search model without space use constraint (Brownian motion) with a tendency to remain in a particular home range. Finally, the Ornstein-Uhlenbeck Foraging (OUF) process features both velocity autocorrelation time scale (a measure of path sinuosity) and restricted space use [41,42]. Both the OU and OUF model processes accommodate autocorrelated data to estimate home range size and crossing time (day).
Starting values derived from semi-variograms were used for maximum likelihood model fitting. Suitable models were fitted to the data using maximum likelihood estimation and best models were selected based on their AICc weight. The best model for each individual leopard was used to calculate movement parameters and home range; the latter defined as area within 95% % UD isopleths of AKDE estimates.
Core areas of space use, defined as the area within which an animal spends a maximum amount of time, was estimated using an individual-based quantitative approach, following Vander Wal and Rodgers [43]. Thus, the AKDE utilization distribution area was plotted against isopleths to determine the point at which the proportional home range area begins to increase at a greater rate than the probability of use (slope = 1). The value of the corresponding isopleth determines the boundary of the core area [43]. We then assessed the position of kills made by collared leopards in relation to the core area of their home ranges. We also calculated seasonal AKDE home ranges to explore variation in space use. Seasons were defined as spring = March to May, summer = June to August, fall = September to November and winter = December to February.
We used AKDE estimates to quantify an animal's utilization distribution (UD), i.e., the probability distribution defining the animal's use of space. Then we used AKDE's 'UD' option to illustrate the 'static interaction', i.e. the spatial overlap of 2 home-ranges and congruence in their utilization distributions [27], ignoring the temporal sequence of movement paths [44]. Our pairwise static interaction analysis was based only on those fixes obtained from the period of time when both animals were collared, partitioned by seasons. Therefore M4 was excluded from this analysis, because he was collared after other individuals' collars dropped off.
We calculated range overlap using function overlap in package 'ctmm' version 0.4.0 [41] which uses the Bhattacharya coefficient as an approximate measurement of the amount of overlap between two statistical samples. The overlap function incorporates movement models and calculates the overlap of their auto-correlated kernel density. For each pair of neighbors, we calculated the proportion of home range overlap of individual A on B and vice versa. A value of 1 implies that the two distributions are identical, while a value of 0 implies that the two distributions share no area in common.
In addition to home range and crossing time, two other movement parameters, the velocity autocorrelation time scale (a measure of path sinuosity) and mean distance travelled per day were also calculated [41,42] by the OUF model. All statistical analysis were implemented in R environment for statistical computing [38].

Results
Between September 2014 and May 2017, we collared and monitored six leopards (5 males and 1 female) using GPS-satellite Iridium collars, comprising 4 adults and 2 young individuals in Tandoureh National Park. GPS collars collected between 54 and 368 days data per individual, representing a total of 56.7 monthly leopard study periods (Table 1). Our overall fix rate was high (mean 85.0% ± SE 7.6) and we obtained a total of 22226 GPS locations for 1702 leoparddays (283.7 ± SE 50.8 days/leopard). No erroneous fixes or spikes in movement were detected in our data, despite using very conservative movement parameters to screen location errors.
Overall, 17.9% of GPS fixes were located outside the park ( Table 1). The five collared leopards which were observed outside the park varied substantially in the amount of time spent on multi-use lands (villages, farmlands and pastures), ranging between 2.2 to 43.8% (Fig 1). Only the leopard M2/Bardia did not leave the park limits.

Home range size and overlap
Based on objective assessment of variograms (Fig 2), a clear asymptote was reached for three adult males M2, M3 and M4, showing their constrained space use as resident individuals. In contrast, both young leopards (F5 and M6) lacked an asymptote, evidence for lack of range residency. F5 was tracked for only 54 days which was probably not long enough to show range residency. M1 (old male) showed a mixed ranging pattern. He showed resident behavior until almost 5.5 months after collaring when his semi-variance increased and he started his excursions outside the park along the borderland's communities with regular returns to the national park.
We excluded non-resident individuals which did not constrain their space use (F5 and M6) and the old male (M1) that appeared to become a non-resident wandering animal from the estimates of home range size. Accordingly, mean AKDE home range was calculated to be 103.4 ± SE 51.8 km 2 for resident males which was slightly larger than their non-correlated KDE home range size estimates (1.0 to 1.1 times; Table 1). M6, possibly a dispersing young male, showed the largest range use in one year, expanding from Iran into Turkmenistan, resulting in an elongated range with 81.6 km between farthest fixes (Table 1). His AKDE analysis revealed that he finally settled in Turkmenistan, according to his core area which was placed primarily within the Turkmen territory (Fig 1). This male had the largest difference between AKDE and KDE (AKDE > 3KDE). Mean estimated core area size for resident males was 32.4 ± 12.7 km 2 , which were represented by the 62% to 67% isopleths of the utility distribution (Table 1). There was no consistent seasonal difference in AKDE home ranges for resident males (F 5, 6 = 1.72, P = 0.26). Although our sample size was small, individual variations in seasonal home range size can be seen (Fig 3). The two resident males (M2 and M3) tended to have their smallest AKDE estimates during winter when snow covered higher elevations confine their ranging to lower areas (Fig 3).
In total, we found 139 kill remains where leopards preyed (n = 130) or scavenged (n = 9), belonging to 10 species, mostly medium sized prey (93.1%). On average, only 10.7 ± 3.8% of kills made by resident males were found outside the core areas of AKDE home ranges. No sign of predation or scavenging outside of the core area was found for the only collared female (F5/ Iran) during her short tracking period while the largest proportion of outside core area's kills belonged to the M6, the young non-resident male (22.2%, Table 1).
All leopards with neighboring ranges showed moderate home range overlap, varying from 0.29 to 0.64 (mean = 0.43 SE 0.06; Table 2). Home range overlap was also similar between resident males (0.44 SE 0.10) and resident-transient individuals (0.39 SE 0.06).

Discussion
Our home range estimates for Persian leopards were larger than those reported in previous Asian leopard ranging studies. Range overlap between conspecifics was relatively high and the majority of predation events occurred within home range core areas. Excursions to areas occupied by people occurred on 17.9% of all leopard collaring days with very wide variation among individuals, pointing to the importance of a combining land sparing and land sharing approaches for leopard conservation.

Home range size
Although our data showed remarkable individual variation, leopards in Tandoureh occupied the largest home ranges recorded so far for Asian landscapes [45][46][47], with the exception of an adult male tracked in an arid montane habitat in central Iran (670 km 2 [48]). The home ranges of predators scale with body mass [49] and habitat productivity, which affects prey biomass  [4]. The large body masses of Persian leopards [50] and the low primary productivity of the landscapes (e.g. annual rainfall 250-300 mm in northeastern Iran) are likely to be two key determinants of their larger home range sizes. Seasonality may partly explain variation in leopard home-range sizes at the population scale [4]. We found no evidence of consistent seasonal variation in home range size, in accordance with previous leopard studies [30,31,46,47]. Nonetheless, our data is consistent with previous observations [31,46] in suggesting that seasonal variation in home range size is an individual behavior rather than a population level trait. Both adult males (M2 and M3) restricted their ranging to lower elevations during winter when higher elevations are covered with snow and are extremely cold, both being factors known to be constraints for leopard habitat selection [23,51].

Home range overlap
Leopards showed substantial home range overlaps in Tandoureh. The estimates of home range overlaps in the current study were considerably larger than reported in many previous studies on leopards (Table 3). There are two possible explanations. The majority of previous studies were based on VHF telemetry, which may miss significant animal movements and consequently result in smaller home range overlap estimates [52]. Likewise, the conventional KDE and MCP estimation generally provide a lower bound for the estimate of home range area [40], and consequently result in overlap reduction. Alternatively, the higher home range overlap observed in Tandoureh is attributable to the topographic features of this rugged landscape that can facilitate co-existence of multiple individuals. Landscape heterogeneity and topographic features can provide restricted detectability for leopards and promote their spatiotemporal overlap.
Predation occurred mostly in parts of the home range used exclusively by each leopard, i.e. home range core areas. We know of only two other studies evaluating the spatial configuration of hunting grounds in regard to felids' core areas. Predation events were reported to be more frequent outside core areas for both other case studies, i.e. jaguar Panthera onca [54] and puma Puma concolor [55]. Amongst home ranges with high degrees of spatial overlap, exclusive hunting areas can facilitate coexistence of multiple individuals. Competition over resources, including kills, is a cause of intraspecific agonistic behavior in leopards [56] and resulted in the death of the only collared female leopard in Tandoureh.
An obvious limitation of our study is the small sample size of GPS collared leopards, of which most were male (five out of six). Clearly the findings on a single female risk influence by individual idiosyncrasy [57]. Nevertheless, besides Simcharoen et al. [46] pioneering work (with eight collared leopards), our study is the most intensive study ever conducted on Asian leopards in terms of sample size and collaring period, reflecting the difficulty of working in the harsh landscapes in which Asian leopards persist.

Conclusion and synthesis
Home range, as described by Powell and Mitchell [58], is "that part of an animal's cognitive map of its environment that it chooses to keep updated". We speculate that our findings support an 'anchoring' and 'adjustment' paradigm in the use of space. Anchoring and adjustment, are cognitive biases in the assessment of risk first described in humans by Tversky and Kahneman [59]. According to this psychological heuristic, when people assess the magnitude of a risk, they start with an implicitly suggested reference point (the "anchor") and make adjustments to it to reach their estimate. We do not imply the same mechanisms underlying this heuristic in humans apply to leopards, not least as the concept of 'rational choice' has a different meaning in non-humans [60]. If only in the form of a helpful analogy, the national park may be functioning as an 'anchor' for leopards (and probably many other animals) while they adjust their cognitive space use beyond the park boundaries (where, incidentally, they are not often associated with stock raiding). None of our collared leopards was killed by humans, whereas in the absence of properly managed protected areas, leopards can experience high rates of human-induced mortality in multiuse lands [8,61].
Such anchoring and adjustment behavior supports the proposition that, in Asia's rugged landscapes, a combination of land sparing and land sharing strategies at multiple spatiotemporal scales has the potential to ensure viability of leopards and other big cats. Properly-managed conservation areas (spared lands) are of paramount importance for securing high densities of large carnivores, insofar as they control poaching of carnivores and their prey species. Nonetheless, their space use outside-conservation areas must be managed through promoting the existence of carnivores in human-dominated landscapes ("land sharing"), with minimized levels of conflict with stock breeders.
With 25% of the global land surface area, mountain ecosystems support a wide range of ecosystem services and biodiversity [62]. Climate change is expected to have a radical effect on biodiversity in mountainous areas [63], forcing northward and upward range shifts in many mammalian species [64][65][66], including humans [67]. Asian mountains can serve as climate refugia for big cats [68,69], despite the fact that only one third of their current extant range remains as suitable habitat in the next half century [69,70]. Land use change is the main driving factor for range losses in threatened mammalian carnivores [71]. Conservation policy should clearly be proactive wherever possible for sparing montane refugia, preferably larger and better-connected areas, to anchor a high density of breeding nuclei of large cats in Asia's rugged landscapes. Nonetheless, many montane protected areas are not large enough to meet extensive spatial requirements, high energy needs and hierarchical social interaction of big cats [2]. Therefore, bolstering the coexistence model (i.e. land sharing) is inevitable in order to support viability of both big cats and human communities, which are strongly dependent on reduced water resources in high altitudes. Future research might usefully explore the interaction between the land sharing and sparing, and how it can support both larger carnivore viability and human livelihoods, particularly in the context of montane landscapes.