lable at ScienceDirect

Animal Behaviour 100 (2015) 22e37Contents lists avaiAnimal Behaviour

journal homepage: www.elsevier .com/locate/anbehavImpact of risk on animal behaviour and habitat transition probabilities

Byron du Preez a, *, Tom Hart b, Andrew J. Loveridge a, David W. Macdonald a

a Wildlife Conservation Research Unit, Department of Zoology, University of Oxford, Oxford, U.K.b Department of Zoology, University of Oxford, Oxford, U.K.a r t i c l e i n f o

Article history:Received 7 July 2014Initial acceptance 26 August 2014Final acceptance 14 October 2014Published onlineMS. number: 14-00552R

Keywords:fearinteractionintraguild competitionleopardlionMarkov chainniche shift* Correspondence: B. du Preez, Wildlife Conservatioof Zoology, University of Oxford, Recanati-Kaplan CenRoad, Tubney, Oxford OX13 5QL, U.K.

E-mail address: bydupreez@gmail.com (B. du Pre

http://dx.doi.org/10.1016/j.anbehav.2014.10.0250003-3472/ 2014 The Association for the Study of ALions, Panthera leo, and leopards, Panthera pardus, coexist in space and compete for resources. Althoughdirect killing of leopards by lions has been recorded, avoidance behaviour is an important part of leopardecology that is difficult to measure through direct observation. Using tracking data from simultaneouslycollared lions and leopards, we investigated the effect of lion proximity on the behavioural ecology ofleopards. We show that proximity to lions influenced leopard habitat use, transition probability andbehaviour. Within enclosed habitats, lions were allowed to get closer to leopards before leopardsengaged in a flight response. Visual observation data suggest that lions and leopards infrequently comeinto direct contact. However, tracking data indicate that avoidance was based on relative habitat coverand detectability, and as a result the two species were often located within close proximity. Finding newsignals of interaction and avoidance within two well-studied predators with relatively small sample sizessuggests that this approach may have value to other systems, such as predator/prey interactions, orrelationships between sympatric species, and at a scale hitherto not possible. This could be used toinvestigate the costs and benefits of animal foraging where competitive exclusion may occur, and isrelevant for the large number of animals that are difficult to observe. 2014 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.Predators can influence the ecology and population dynamics ofother species not only through direct killing, but also indirectly byeliciting behavioural changes in response to risk (Creel, Winnie, &Christianson, 2013; Laundre, Hernandez, & Altendorf, 2001;Valeix, Loveridge, et al., 2009). The impacts of predators are typi-cally measured in terms of total individuals killed (Lima, 1998) orbiomass consumed (e.g. Bodendorfer, Hoppe-Dominik, Fischer, &Linsenmair, 2006; Hayward, O'Brien, Hofmeyr, & Kerley, 2007;Karanth & Sunquist, 2000). However, the indirect effects of pre-dation, including behavioural modifications that result in ecologicalniche shifts, can affect the fitness, demographics and populationdensity of a species (Creel et al., 2013; Valeix et al., 2010). Ungulatedistribution and habitat selection are influenced by the risk ofpredation, and ungulates avoid areas of high predation risk andhabitats that reduce their ability to escape (Laundre et al., 2001;Valeix et al., 2010), even though this behaviour negatively affectstheir rate of food ingestion and nutrient quality (e.g. Creel et al.,2013; Hernandez & Laundre, 2005). If predation risk is high, itcan lead to ecological niche shifts that negatively affect then Research Unit, Departmenttre, Tubney House, Abingdon

ez).

nimal Behaviour. Published by Elspopulation dynamics of the subordinate species (Brown, Laundre,& Gurung, 1999; Spitz, Rousseau, & Ridoux, 2006).

Aggressive intraguild interactions between terrestrial carni-vores have been well documented (e.g. Creel et al., 2013; Harihar,Pandav, & Goyal, 2011; Kamler, Davies-Mostert, Hunter, & Mac-donald, 2007; Kamler, Stenkewitz, & Macdonald, 2013; Karanth &Sunquist, 2000; McDougal, 1988; Palomares & Caro, 1999), and aswith herbivores, dominant carnivore species may also indirectlyaffect subordinate carnivore species by inducing behavioural ad-justments associated with risk (e.g. Brown et al., 1999; Laundreet al., 2001). It is therefore conceivable that subordinate carni-vores incur significant costs as a result of increased vigilance,avoidance behaviour, kleptoparasitism and restricted use of sharedhabitats (e.g. Creel et al., 2013; Lima,1998; Valeix et al., 2010). Thesecosts have been less well researched, but understanding the effectsthat predation and competition for resources have on the behav-iour and distribution of key species is fundamental to informedmanagement (e.g. Funston, Groom, & Lindsey, 2013), especially asprotected areas, which are limited in size, become increasinglyisolated (Lindsey, Romanach, Tambling, Chartier, & Groom, 2011).

Interpreting the indirect impacts of predation on an animal'sbehavioural ecology requires insight into the circumstantial de-cisions that an individual faces, which may be reflected in its ac-tivity and habitat use (Basille, Calenge, Marboutin, Andersen, &evier Ltd. All rights reserved.

Delta:1_given nameDelta:1_surnameDelta:1_given nameDelta:1_surnamemailto:bydupreez@gmail.comhttp://crossmark.crossref.org/dialog/?doi=10.1016/j.anbehav.2014.10.025&domain=pdfwww.sciencedirect.com/science/journal/00033472http://www.elsevier.com/locate/anbehavhttp://dx.doi.org/10.1016/j.anbehav.2014.10.025http://dx.doi.org/10.1016/j.anbehav.2014.10.025http://dx.doi.org/10.1016/j.anbehav.2014.10.025

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 23Gaillard, 2008; Jessopp, Cronin, & Hart, 2013), because where ananimal is and what it is doing are influenced by its need for bothshort- and long-term risk avoidance (Lima,1998; Valeix, Fritz, et al.,2009; Whitehead & Jonsen, 2013). Variation in risk between hab-itats, and the effect this has on behaviour, therefore requires seriousconsideration (Lind and Cresswell, 2005; Lone et al., 2014). Spatialheterogeneity of habitat and vegetation density influences thedistribution of risk within the landscape (Laundre et al., 2001;Valeix, Loveridge, et al., 2009), and because predation risk in-fluences habitat selection (Creel and Winnie, 2005; Valeix, Fritz,et al., 2009; Valeix et al., 2010) it is thus important to take intoaccount different habitat types when investigating interspeciesinteractions. Determining habitat use and transition rates betweenhabitat types with regard to predator proximity is therefore infor-mative regarding the impact of risk on behavioural ecology.

However, unbiased observation of wild animals, and in partic-ular nocturnal predators, is difficult (e.g. Broekhuis, Cozzi, Valeix,McNutt, & Macdonald, 2013; Funston, Mills, & Biggs, 2001; Hart,Mann, Coulson, Pettorelli, & Trathan, 2010; Jessopp, Cronin, &Hart et al., 2013), and attempts to do so may affect their behav-iour or expose their position to potential predators or prey. At thelandscape level, quantification of the behavioural response topredator proximity via direct observations is virtually impossible(Valeix et al., 2010); however, biologgers are commonly andincreasingly used in ecological studies. These biologgers are able torecord large time-referenced data sets on location, body orientationand temperature (among other parameters), fromwhich utilizationdistributions, habitat selection, speed of movement and distancesbetween individuals can be determined.

Inferring patterns of animal behaviour indirectly is limited in-sofar as it precludes highly detailed description of activity. How-ever, even broad behavioural classification is insightful whenrelated to the environment and distance from predators (e.g.Broekhuis et al., 2013; Milinski & Heller, 1978), and investigatingthe transitions between behavioural states, rather than focusing onthem individually, may allow identification of the environmentalfactors that induce changes in behaviour (Bagniewska, Hart,Harrington, & Macdonald, 2013; Hart et al., 2010; Krebs & Davies,1991). For example, consider the transition of Ade

lie penguins,Pygoscelis adeliae, between water, where they forage but are also atrisk of predation by leopard seals, Hydrurga leptonyx, and ice, onwhich they are generally safe from predation (Ainley, Ballard, Karl,& Dugger, 2005); although movements between environmentsoccur frequently, the probability of transition from water to landmay be directly related to the presence and proximity of seals, i.e.the behavioural change points are coincident with a predator.

Using biotelemetry track data from both species, we aimed toinvestigate the effect of lion, Panthera leo, predation risk on thebehavioural ecology of leopards, Panthera pardus, specifically, howthe presence and proximity of lions influenced leopard behaviourand habitat transition probability. Lions and leopards aremorphologically similar intraguild competitors that have extensiverange and habitat overlap in Africa (Bauer, Nowell, & Packer, 2013;Henschel et al., 2008), as well as overlap in prey species utilization(Bodendorfer et al., 2006; Ogara et al., 2010). Leopards are solitaryin nature, and smaller than lions (leopards: females 35e50 kg,males 50e75 kg; lions: females 140e160 kg, males 180e240 kg;Kock, Meltzer, & Burroughs, 2006), and leopards are consequentlyat risk of aggressive persecution by the physically and numericallydominant carnivore (e.g. Bonesi & Macdonald, 2004; Kamler et al.,2007, 2013; McDougal, 1988; see Fig. A1 in Appendix 1).

Studies of diving animals have used hidden Markov models(HMMs) to calculate the probability of transiting between as few astwo dive states to accurately determine environmental predictorsof behaviour (Bagniewska et al., 2013; Hart et al., 2010; Jessoppet al., 2013). Specifically, Hart et al. (2010) linked dive depth andduration to hidden behavioural states that were classified as eitherforaging or not foraging, and investigated the transitions betweenthese states and the environmental conditions that triggered them.HMMs assume that the system is Markovian, with distinct stateswhere the transition from time step t to t 1 is conditional on thestate at t rather than previous states, and estimate the unknownstates and the transition probabilities between them (Bagniewskaet al., 2013). As Markov chain analyses consider state persistenceand transitions between states, this makes them ideal for analysisof nonindependent temporal data such as movement trajectories(Hart et al., 2010; Patterson, Basson, Bravington, & Gunn, 2009).Markov chain analyses control for variables such as differentamounts of habitat and density of predators, prey and competitors,and allow direct comparison of behaviour between habitat typesand different areas.

Here we employed a Markov chain analysis of relatively high-resolution track (i.e. the same GPS fix rate used for detecting dy-namic interactions between big cats; e.g. Benhamou, Valeix,Chamaille-Jammes, Macdonald, & Loveridge, 2014) and environ-mental data to investigate behavioural changes in two terrestrialpredators. This studywas inspired by the pioneeringwork of Brownet al. (1999) and Laundre et al. (2001) regarding the ecologicalconcept of fear, but is novel in that we had access to leopard pop-ulations both with and without lions present, as well as before andafter the introduction of lions. A Markov chain analysis of thissystem is appropriate as it measures and removes the confoundingeffects of spatial autocorrelation, habitat variability and prey den-sity, and focuses on the behavioural changes that occur whenleopards come into close proximity to lions.

In the field of large-mammal ecology, properly controlled, large-scale in situ experimental manipulations are rare, and notoriouslydifficult to carry out (Bonesi &Macdonald, 2004; Creel et al., 2013).We took advantage of a unique opportunity to experimentally testthe impact of lions on leopard behavioural ecology when an elec-trified fence that had historically excluded lions from a section ofour study site (hereafter Kwalusi) was removed (see Fig. 1). Lionswere present and abundant in the area adjacent to Kwalusi (here-after Mazunga), and the fenced exclosure effectively created asharp contrast in competition and predation by lions on an other-wise contiguous population of leopards. The fence was removed 18months after the study began, allowing lions to move into andcolonize the Kwalusi area, rapidly balancing their density betweenthe two sites (du Preez, Loveridge, & Macdonald, 2014a). We thuscompared, at the individual level, the behaviour of the leopardsubpopulations with and without the risk of predation by lions, aswell as the effect of the removal of the fence and rapid introductionof lions on the leopard population. Where lions were present, wealso assessed the variation in predator proximity and resultant levelof leopard response, so as not to simply dichotomize risk into onlypredator present/absent classes (e.g. Creel, Christianson, Liley, &Winnie, 2007; Creel et al., 2013).

Lions represent a significant source of mortality for leopards(Bailey, 1993; and see Fig. A1), and leopards are likely to activelyavoid lions so as to reduce risk. However, avoidance behaviour maybe a trade-off between running away, thereby alerting lions to theirpresence, and hiding, thereby risking close-quarter detection.Leopard behaviour and habitat use are thus likely to reflect both thereal and perceived risk associated with lions (e.g. Brown et al., 1999;Laundre et al., 2001). We therefore expected to find changes in thetransition rate fromopen to closed habitat anda change in behaviourwithin eachhabitatwhen lionswere present compared towhen theywere absent. Moreover, as this behaviour is lion-moderated andbased on detection of lions, we expected leopard behaviour tochange based on the proximity of the nearest tracked lion.

Figure 1. Map of the 3743 km2 Bubye Valley Conservancy where the research was conducted, and its location and relative size within Zimbabwe (inset top right). The Mazunga(control site) and Kwalusi (experiment site) areas are indicated on the map. The black-and-white line indicates the electrified fence that excluded lions from the Kwalusi area east ofthe fence (the exclosure). Imagery Copyright 2014 TerraMetrics, Inc., www.terrametrics.com.

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3724Because of the potentially lethal consequences of interactionwith lions, and the differing amounts of available cover betweenhabitat types (and therefore differential levels of detection risk andpotential for escape), we hypothesized first that (H1) leopardhabitat transition probability would increase from more openhabitat types to more closed habitat types where lions were pre-sent. Second, we hypothesized that (H2) leopard behaviour be-tween each habitat type would differ in areas where lions werepresent. Finally, as detection probability and the relative level ofrisk are an inverse function of distance (related to habitat density),we hypothesized that (H3) leopard behaviour would be dependenton the proximity of lions.

METHODS

Study Site

The Bubye Valley Conservancy (BVC) is a privately ownedwildlife conservancy of 3743 km2 located in the MatabelelandSouth province of Zimbabwe, centred at 3070E, 21300S, and about550 m above sea level. The mean annual rainfall between 2007 and2012 was 351 76 mm, which fell mostly between November andMarch. May to August are the driest months of the year. Thevegetation consists largely of mopane, Colophospermum mopane,and acacia, Acacia spp., scrub and woodland, transected by seasonalriver-lines and associated riparian vegetation, with scattered opengrassland areas in between. BVC contains the full complement ofendemic megafauna, including elephant, Loxodonta africana, bothspecies of rhinoceros (Ceratotherium simum and Diceros bicornis),lion, leopard and buffalo, Syncerus caffer. Zimbabwean veterinarylegislation pertaining to management of buffalo results in thecomplete enclosure of BVC by a double fence. Lion and leopard arethe principal large carnivores, and, although present, the abun-dance of spotted hyaena, Crocuta crocuta, wild dog, Lycaon pictus,and cheetah, Acinonyx jubatus, remain low (du Preez, n.d.). Thereare abundant and diverse herbivore populations, especially zebra,Equus quagga, wildebeest, Connochaetes taurinus, and impala,Aepyceros melampus. All permanent water sources are maintainedwith boreholes, and culling is used as a tool for ungulate populationcontrol (K. Leathem, personal communication). On 25 August 2011the fence separating the Kwalusi management section from themain body of the conservancy was removed, and as a result lionswere able to enter and colonize this area (see Fig. 1).

Ethical Note

Captured leopards and lions were handled by project staffqualified by attendance at Zimbabwe's Physical and ChemicalCapture ofWild Animals Course and held valid drugs licences (B. duPreez, Dangerous Drugs Licence No. 600131). The animals werecaptured under permit from the Zimbabwe Parks and WildlifeManagement Authority (23(1) (C) (ii) 13/2008) and with permis-sion from the landowner and conservancy management, followingthe ASAB/ABS recommendations for the Use of Animals in Research.

Adult lions were darted remotely using a Dan-Inject (DAN-INJECT ApS, Sellerup Skovvej 116, Brkop, Denmark, DK 7080) CO2propelled 1.5 ml dart, which delivered drugs intramuscularly toeither the shoulder or the rump of the animal. Lions were dartedstrictly during the coolest part of the day, either at dawn or dusk.Leopards were trapped in cages measuring 1.50 0.75 m and

http://www.terrametrics.com

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 250.60 m high, with a 2 2 cm wire mesh that prevented body partsfrom getting caught, and a door-stopper, combined with a lock, toprevent complete closure causing injury to the tail. Traps werepositioned at midday when leopards are least likely to be active,and were set in the late afternoon, before the evening peak inleopard activity. Leopardswere trapped at night, and the traps werechecked at first light each morning. The traps were covered beforethe leopard was darted through the mesh using the Dan-Inject at apressure of 4 bars, and the animal was left in the trap untilcompletely immobilized. Subadult leopards that were caught weredarted in the trap so that they could be safely removed withoutcausing injury to themselves or research staff, but were notcollared. Captured leopards that were not collared were visuallymonitored after capture until they had moved off on their owninitiative and were no longer visible, and approximately 8 h latertheywere then physically tracked on foot until it was clear that theyhad left the area and were moving with consistent stride anddirectionality. Small cubs were not able to trigger the trap, andleopard trapping was strictly done in conjunctionwith camera trapsurveys so that specific animals could be targeted, which preventedheavily pregnant females and those with dependent cubs frombeing captured. Only adult animals of both species were collared.Lions in late stage pregnancy were not captured, and although it isimpossible to detect early pregnancy, the drugs are extensivelyused by veterinarians, and have no known effects on fetuses.

The capture drugs used to immobilize the animals were a com-bination of a dissociative anaesthetic (Zoletil; lion dosage:1.445 0.157 (range 1.042e1.786) mg/kg; leopard dosage:2.752 0.181 (range 2.400e3.000)mg/kg;manufactured by VirbackRSA, Halfway House, South Africa) and a sedative (Xylazine; liondosage: 0.578 0.063 (range 0.417e0.714) mg/kg; leopard dosage:1.310 0.156 (range 1.000e1.500) mg/kg; manufactured by CPPharma, Burgdorf, Germany), which was reversed with an alpha-2antagonist (Atipamezole; lion dosage: 0.058 0.006 (range0.042e0.071) mg/kg; leopard dosage: 0.131 0.016 (range0.100e0.150) mg/kg; manufactured by Novartis, Isando, South Af-rica) administered intramuscularlywithin 60 min of immobilization.Doses were calculated for each species and sex separately, and ani-malsweremonitored until they hadmade a full recovery andhad leftthe area (up to 5 h, but usually less than 3 h). Collared animals weremonitored via radiotracking for the next 48 h after capture to ensurethat there were no lingering effects. No adverse effects from thedrugs were observed, and none have been reported in the literature.

The lion collars weighed 1.10 kg, and the leopard collars 0.65 kg,which represents 0.69% and 1.30% of the body weight of thesmallest individual collared of each species, respectively. GPS col-lars were replaced approximately 18e24months after deployment,when the batteries began to fail (which was indicated by noticeablyweaker VHF signal detection). Recollaring of known individualsallowed long-term population monitoring.

Leopard and Lion at BVC

Although the abundance of each carnivore is irrelevant to theMarkov chain analysis, both lion and leopard density at each studysite, and before and after the removal of the exclosure fence, areprovided simply to indicate the potential for competition betweenthem. Lion density was measured both inside and outside theexclosure, before and after the fence was removed, using spoortransects (Funston et al., 2010; Stander, 1998). In 2011 no lions weredetected inside the Kwalusi exclosure (460 km2), while at the sametime the lion density in the adjacent Mazunga area (2293 km2) wascalculated at 0.202 0.018 lions/km2. In 2012, after removal of thefence, the lion density had increased to 0.202 0.020 lions/km2 atthe exclosure site, and had remained at a similar density of0.187 0.012 lions/km2 in the surrounding area (du Preez et al.,2014a). Leopard density was calculated using spatially explicitcaptureerecapture modelling of camera data (du Preez et al.,2014a). The adult leopard density at Kwalusi was0.054 0.011 leopards/km2 in 2011 before lions were present, and0.061 0.011 leopards/km2 in 2012 when lions were present. AtMazunga the adult leopard density was 0.046 0.010 leopards/km2 in both 2011 and 2012.

GPS Radiotelemetry Tagging

Between 2010 and 2013, 21 lions (nine female and 12 male) and15 leopards (seven female and eightmale) were fittedwith custom-built GPS radiotelemetry collars (Africa Wildlife Tracking, Pretoria,South Africa). Six leopards were collared at the Kwalusi site whilethe area was lion free, and an additional leopard was collared afterthe fence was removed and lions had entered. Eight leopards werecollared at the Mazunga site. The number of collared leopards ateach site represents approximately 26% of the standing Kwalusileopard population and 38% of the standing Mazunga leopardpopulation (du Preez, Loveridge, & Macdonald, 2014b).

All collars simultaneously recorded a GPS positional fix hourlyfrom 1600 to 0800 hours when the animals were most active, andthen every 2 h at 1000, 1200 and 1400 hours during the heat of theday when both lions and leopards are least active (e.g. Bothma, Nel,& Macdonald, 1984; Cozzi et al., 2012; Valeix et al., 2010). This fixschedule was a compromise between the resolution of the data andthe life span of the collar batteries, which in our case lasted forapproximately 18e24 months before they needed replacement.GPS collar data were analysed to determine habitat use, transitionbetween habitats and behaviour within each habitat type. All sta-tistical analyses were done in R 3.0.1 (The R Foundation for Sta-tistical Computing, Vienna, Austria, http://www.r-project.org).

We conducted a static test of horizontal GPS collar location ac-curacy in riparian habitat to ensure that the circular error probable(CEP) was not adversely affected by the density of the vegetation(e.g. Rempel & Rodgers, 1997). Five GPS collars were placed in afixed location within riparian habitat and left for 1 week, afterwhich the 50% CEP for each GPS collar was calculated. This resultedin a mean SE 50% CEP of 9.94 0.89 m, which we considered tobe acceptable for our study (e.g. Agouridis et al., 2004; D'eon,Serrouya, Smith, & Kochanny, 2002; Rempel & Rodgers, 1997).

The maximum precision of the GPS collars we used is approxi-mately 2.5 m (M. Haupt, Africa Wildlife Tracking, personalcommunication), and to help assess the accuracy of each locationalfix, the horizontal dilution of precision (HDOP) value is coinci-dentally recorded. The HDOP values may be multiplied by 2.5 m tocalculate the potential error of the fix, and for our data themean SE of the HDOP values recorded were in grassland2.10 0.09, in scrub 2.17 0.08 and in riparian habitat 2.19 0.09.We compared the mean HDOP values for each individual animal ineach habitat type, and tested these for a significant difference usinga general linear mixed-effects model, with individual included as arandom effect. Although the mean HDOP value increased slightlywith habitat density, there was no significant difference in HDOPbetween each habitat type (F2,98 0.287, P 0.752, R2 0.01,power 0.096).

Habitat Use

Habitat classification was performed in ArcMap 10.1 (ESRI,www.esri.com), using a Bing (Microsoft) satellite map at a scale of1:40 000. Description of cover regarding risk of detection needs tobe objective (Jones, 1968), and therefore the area within the studysite was classified into one of three broad habitat types derived

http://www.r-project.orghttp://www.esri.com

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3726from the satellite imagery, based on relative vegetative density: (1)grassland, with no ground or vertical cover, (2) scrub, with groundcover but no vertical cover, and (3) riparian habitat, with bothground and vertical cover (in the form of trees, up which leopardscan find refuge from lions; e.g. Calenge, Dufour, & Maillard, 2005;Godvik et al., 2009; de Knegt et al., 2011). Habitat classificationinto broad categories was relatively straightforward as all threehabitat types differ obviously from one another in terms of speciescomposition and vegetative density, and are thus visuallydistinguishable.

The points of greatest change in vegetation cover and compo-sition defined the ecotones between habitat classes (Fortin &Drapeau, 1995), and geospatial vector shapefiles for each werecreated in ArcMap. To ensure classification consistency in theresulting habitat map, we inspected random GPS points in situ andcompared these to the digital map. GPS tracking data for each in-dividual were then imported into ArcMap, merged with the habitatmap (using the Merge function in the Geoprocessing menu), andthe habitat class attribute for each GPS fix was extracted by thehabitat shapefile within which it was located (using Extract underSpatial Analyst Tools in the ArcMap Toolbox).

Habitat-to-habitat Transition Probability

A habitat transition matrix of five consecutive GPS fixes wascreated for each individual GPS fix location, with each matrixconsisting of the fix datum at t along with the two previous(t 1, t 2) and two subsequent (t 1, t 2) fixes for each indi-vidual leopard (see R script in Appendix 2). Five was the lowestnumber of consecutive fixes required to create a transition matrix.The reason that we used the minimum number of fixes possiblewas because the aim was to show very short-term behaviouralencounters, so we wished to avoid any averaging effect of a longermoving window.

The proportional transition probability of the animal eitherstaying in the same habitat or moving to a different habitat typewas thus calculated (i.e. a dependency metric; see Jessopp et al.,2013). With three different habitat types there are nine possibletransitions: one from each habitat to either the same habitat orone of the other two habitat types. The mean likelihood of eachpossible transition to and from each habitat class for each indi-vidual could then be related to environmental predictors, such asvarying proximity to lions.

The distance of each individual to every other simultaneouslycollared animal was determined for every fix datum (see R script inAppendix 3), and the habitat transition probabilities to and fromeach habitat class were compared at varying distances (100 m in-tervals) between leopards and lions. Thus the relationship betweenthe proximity to lions and leopard habitat use, behaviour andtransition probabilities could be determined. A general linearmixed-effects model, with a single fixed effect of lion presence, wasused to determine difference in habitat transition probabilitiesbetween the areas with and without the risk of lion predation. Toremove any pseudoreplication, individual was included as arandom effect.

A large proportion of direct interactions between species couldbe missed as a result of the hourly resolution of the collar data, anddetections of interaction between individuals would be reducedbecause of avoidance behaviour occurring between scheduled fixes(Creel et al., 2013). Investigating predatoreprey dynamics with GPScollar data therefore probably underestimates the predator effectbecause encounters both between fixes and with uncollaredpredators go undetected. If the frequency of interactions betweenpredators and prey is underestimated, then the level of antipred-ator response in the population will be underrated (Creel et al.,2013). To reduce the observed impact of unknown lions on theanalysis of leopard behaviour, only leopard data collected within1000 m of collared lions were considered for comparison betweenthe areas where lions were either present or absent. Lions areaggressively territorial and neighbours largely avoid each otherbecause direct interactions can result in injury or death (e.g.Benhamou et al., 2014). A threshold distance of 1000 m fromknown lions was therefore chosen within which to measure leop-ard behaviour, because this distance is great enough that it alloweda large sample of leopard data to be investigated, while also smallenough to prevent the effect of unknown lions, whichwere unlikelyto be frequently within 1000 m of territorially dominant in-dividuals (see Mosser & Packer, 2009), from influencing leopardbehaviour.

Generalized Habitat Transitions where Lions were Present

Transitions between habitats could be directionally generalizedin terms of movement between open and closed types; forexample, a transition from grassland to scrub would be consideredmoving from open to closed habitat, while riparian to scrub wouldbe a closed to open direction. The open-closed/closed-open direc-tionality of habitat transition at varying distances to lions wastherefore determined at varying lion proximity to investigatefurther the relative importance of habitat density and transitionsbetween habitats by leopards for reducing risk.

Leopard habitat-to-habitat transition probabilities were calcu-lated at increasing distance from the nearest known lion within100 m intervals. The mean likelihood of a leopard remaining in anyparticular habitat type versus moving to a different habitat type att 1 was determined for each interval and individual leopard. Themean probabilities of leopards moving from closed to open(scrub/ grass, riparian/ scrub, riparian/ grass) and open toclosed (grass/ scrub, grass/ riparian, scrub/ riparian) habitattypes at varying proximity to lions were thus calculated, and theeffect of lion proximity on leopard habitat transition direction wastested using a polynomial analysis with the lm function in R.

Habitat-specific Leopard Behaviour

As a proxy for leopard behaviour within each habitat class, thestraight-line distance between consecutive hourly fixes wascalculated, andwas defined as step length (Fieberg& Borger, 2012;Valeix et al., 2010). This is effectively a proxy for the speed (v d/t)at which a leopardmoveswithin each habitat type, and is indicativeof a latent behaviour (e.g. Bagniewska et al., 2013; Hart et al., 2010;Whitehead& Jonsen, 2013). Lion step lengths are known to be shortwhen foraging, but long and highly directional when avoidingdanger such as people or other lions (Valeix, Hemson, Loveridge,Mills, & Macdonald, 2012). Similarly, as lions represent a sourceof risk to leopards, leopard habitat-specific step lengths weredetermined at varying distances from lions, and compared to thoseat the Kwalusi site when lions were absent.

RESULTS

Habitat Transition Probabilities

The mean probability of every possible habitat-to-habitat tran-sition for each collared leopard was calculated both in the absence(Fig. 2a) and in the presence of lions (Fig. 2b), betweenwhich therewas a difference in risk of predation. A general linear mixed-effectsmodel, with individual included as a random effect, was used to testthe effect of lion presence on each leopard habitat transitionprobability, which showed a significant increase in the probability

0

20

40

60

100

GS GR SG SS SR RG RS RR

(a)

*

i

ii

iiiiv

v

0

20

40

60

100

GG GS GR SG SS SR RG RS RR

(b)

*

i

ii iii

iv

v80H

abit

at t

ran

siti

on p

roba

bili

ty (

%)

GG

Habitathabitat transition

80

Figure 2. Leopard transition probabilities to and from each habitat type, where lions were (a) absent and (b) present. The solid bars indicate the mean probability averaged acrossindividuals for each habitat type, and the error bars represent the standard error of the mean. G grassland; S scrub; R riparian; GG grassland-to-grassland transition;GS grassland-to-scrub transition, etc. An asterisk (*) above the corresponding bars indicates where a particular habitat transition is significantly different between the areas wherelions were present and absent. Roman numerals ieiii above the corresponding bars indicate where lions had a negative effect on habitat transition probability, and Roman numeralsiv and v indicate where lions had a positive effect on habitat transition probability, as predicted based on relative habitat density.

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 27of leopards moving from grassland to scrub where lions werepresent (see Table A1 in Appendix 4 for statistical results). Althoughnonsignificant, there was a lower mean probability of leopardsmoving from all other habitats (riparian, scrub and grassland) tograssland habitat, and a greater mean probability of leopardsmoving from scrub to riparian habitat, and remaining in riparianhabitat, where lions were present.

Transitions where Lions were Present

There was a relatively low likelihood of leopards moving fromclosed to open habitat types, except when at close proximity tolions (Fig. 3), which had a significant influence on the probability ofleopards moving out of the particular habitat type in which theywere located (Table A2). There was no significant influence of lionproximity on the probability of leopards moving in the direction ofopen to closed habitat (Table A2), which remained relatively high atall distances from lions.

Habitat-specific Behaviour

There was no difference in mean leopard step lengths betweenhabitat types where lions were absent (Table A3) or where lionswere present but when their proximity to the leopards was notaccounted for (F2,19 2.286, P 0.079, R2 0.191, power 0.553).However, when leopard behaviour was examined within 1000 mfrom lions, there was an increase in leopard stepelength in grass-land habitat, which was significantly greater than the step lengthsin both scrub and riparian habitat types (Table A3, Fig. 4).

The relationship between leopard step length in grasslandhabitat and lion proximity was therefore investigated further;leopard data were divided into 100 m intervals from the nearestsimilarly collared lion, and the mean step length for each intervalwas determined. A general linear mixed-effects model, with indi-vidual included as a random effect, showed that leopard step lengthwas significantly related to their distance from lions (F1,13 12.686,P 0.003, R2 0.494, power 0.942; Fig. 5).

DISCUSSION

Here we used Markov chain analyses to investigate the trackdata of sympatric carnivores, which allowed us to detect newsignals of interaction between two relatively well-studied species,and reveal the impact of one species on the habitat transitions andbehaviour of the other. Markov chain analyses account for theserial dependence of track data, and directly link behaviour toenvironmental data in an integrated manner (Patterson et al.,2009). The Markov chain analysis facilitated the identification oftransition points between leopard behavioural states in relation tothe proximal cause of lion risk, and thus removed the effects ofconfounding variables such as prey density and the presence ofother carnivores.

Investigating the impact of one species on the behaviour ofanother is important, because the cost of habitat transitionswhen avoiding predators could lead to reduced foraging effi-ciency, fitness and ultimately population density, while failure toavoid predators altogether may result in mortality. Predators, andthe risk and fear of encountering them, therefore have a signif-icant impact on the behavioural ecology of a species (e.g. Brownet al., 1999; Laundre et al., 2001). In this study, the existence ofthe lion exclosure and its subsequent removal provided anunprecedented opportunity to determine the impact of lionson leopard behaviour by simultaneously comparing neighbour-ing leopard subpopulations under differing levels of risk, wherepart of this variation was experimentally manipulated, partnatural.

1000 1500 20000

20

40

60

100

80

Tran

siti

on p

roba

bili

ty (

%)

500Distance to lions (m)

Figure 3. The probability of leopards moving in the direction of closed to openhabitat at varying proximity to lions. Points indicate the mean transition probability,averaged across each individual leopard, and the error bars indicate the standard errorof the mean. The curve represents the quadratic polynomial of the general linearmodel.

0

200

400

600

800

1200

Scrub Riparian

(a)

0

200

400

600

800

1200

Grassland Scrub Riparian

(b)

Grassland

Habitat type

Leop

ard

ste

p l

engt

h (

m/h

)

1000 1000

Figure 4. The average hourly distance (step length) moved by leopards in each habitat type, compared between where lions were (a) absent and (b) present. The solid bars indicatethe mean probability averaged across individuals for each habitat type, and the error bars represent the standard error of the mean.

0

500

1000

2000

1500

Leop

ard

ste

p l

engt

h (

m/h

)

500 1000 1500 2000

Distance to lions (m)

Figure 5. Leopard step length in grassland habitat at varying proximity to lions. Pointsindicate the mean step length, averaged across each individual leopard, and the errorbars indicate the standard error of the mean. The black line represents the linearmodel.

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3728

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 29Animal movement and behaviour are influenced by environ-mental circumstances, and may be impacted by both habitatcharacteristics and predation risk (e.g. Valeix et al., 2010). A typicalantipredator response may be indicated by retreat to safe habitat(Sih, 1997), and our results show that leopard behaviour andtransition probabilities were related to the proximity of lions;leopards moved to denser habitat types to reduce risk. Reactionsto lions were greatest in grassland habitat, where leopardbehaviour changed when at closer proximity to lions, as indicatedby the increase in their speed of movement. This response occursbecause within a landscape where there is the threat of predation,faster, straight-line movement in open areas reduces exposuretime and risk of detection (e.g. Valeix et al., 2010; Zollner & Lima,2005).

Where lions were present, leopards were significantly morelikely to move from open grassland habitat to the denser scrubhabitat, which offers more ground cover where they are able tohide. Similarly, leopards also showed a reduction in transitionprobability from all habitat types to grassland, and an increase intransition probability to riparian habitat. Open habitat types,particularly grassland, are likely to be dangerous areas where lionsare a concern for leopards, as such habitats provide neither sub-stantial ground cover nor trees up which to escape, and lions canoutrun leopards (L. Hunter, unpublished data, cited in Balme,Hunter, & Slotow, 2007). Where there is predation risk, leopardsare therefore likely to have an increased probability of movingfrom grassland habitat and into denser habitats that providerelatively more safety from detection and avenues of escape, andthis was confirmed by the comparison of the leopard habitattransition probabilities between the sites of lion presence/absence, as well as increased step length in open areas so as toreduce the time exposed to risk. With nowhere to hide in grass-land habitat, this response is also likely to be stronger at closerproximity to lions, which was also shown by the analysis of steplength versus distance to lions. Similarly, even though predationrisk does not ultimately affect long-term habitat selection byherbivores, it has been shown that they may still exhibit short-term risk-sensitive behaviour in the vicinity of lions (Valeixet al., 2010). For animals at risk of predation, the nonlethal ef-fects of predators on their behavioural ecology may have a greaterimpact on their population dynamics than mortality alone (Brownet al., 1999).

Scrub and riparian habitat appeared to be equally adequateregarding their use by leopards in the presence of lions, as therewas little difference in leopard behaviour and transition proba-bilities between them. This finding confirms a previous study ofhabitat use by leopards for hunting, in which areas of interme-diate ground cover were found to be optimal (Balme et al., 2007),and it is likely that where the available ground cover is adequatefor leopards to escape being detected by their prey, this covermay also aid in their concealment from predators. However, ifleopards are detected by lions at close proximity in scrub habitatwhere there are no trees up which to climb, then leopards maystill have little chance of safely escaping. If a leopard were hidingfrom a lion that was on a convergent trajectory, it would there-fore need to weigh the costs of remaining concealed and therebyrisking detection at close proximity, or breaking cover andalerting the predator to their presence (e.g. Broom & Ruxton,2005).

Cryptic animals at risk of predation, including subordinate car-nivores such as the leopard in this context, are likely to be wary ofand detect larger and more numerous predators before theythemselves are detected (Broom & Ruxton, 2005; Brown et al.,1999). One strategy to minimize risk, upon detection of a pred-ator, is to immediately leave the general area before theythemselves are detected; even if this means crossing exposedhabitat to reach safety. Leopards may also employ a multistage, orsophisticated, antipredator response where possible (e.g. Hemmi& Pfeil, 2010), relying on camouflage and cryptic behaviour (Caro,2014) until the risk of detection is imminent, and then flushing,rather than breaking cover early and risk alerting an otherwiseunaware predator (Broom & Ruxton, 2005). This would explain thepattern observed in the closed/ open habitat transition proba-bility, which showed a low probability of transition from closed toopen habitat types, except when at close proximity to lions wherethere was a significant increase in the transition probability out ofthe immediate area and habitat.

Leopards displayed significantly different (H1) habitat transitionprobabilities and (H2) behaviour between the areas of lion presenceand absence, as well as (H3) increased step length at closer prox-imity to lions. We have therefore shown that lions do affect thebehavioural ecology of leopards, and we accept our hypotheses.However, we also note that the true scale of the observed effectswas possibly masked by the presence of uncollared lions, as well asthe unrecorded interactions between GPS fixes, and may thereforebe even greater than was actually shown.

Sexually size-dimorphic male and female leopards are likely tobehave differently (e.g. Loarie, Tambling, & Asner, 2013; Valeixet al., 2010) as they have different energy requirements, access toprey types and home range sizes (Bailey, 1993), and thereforesubsequently differing levels of predation risk. Generalist pop-ulations may also contain individual specialists, and behaviouraldifferences may result in variation in response measurements (e.g.Bagniewska et al., 2013). However, despite these influences, thegeneral pattern of leopard habitat transitions and behaviour relatedto lion presence remained clear in this study. Fifteen tagged leop-ards and 21 lions provided sufficient data to investigate predationrisk at a range of distances between the species, despite the inter-species encounter frequency being low.

In addition to considering leopards at the population level, aninteresting example of one GPS-collared male leopard allowed thesimultaneous investigation of behavioural differences at the in-dividual level between areas with and without lions, as this in-dividual utilized both sides of the electrified fence while it was inplace. This leopardwas able to freely cross the fence, and displayeddifferent behaviour between areas where lions were present andabsent. The behavioural differences shown by this individual be-tween areas with and without lions reflect similar differencesdetected at the population level, and suggest an individual abilityto adapt to the immediate level of risk (see Appendix 5 for acomprehensive description and analysis of this individual'shabitat transitions and behaviour at both sites of lion presence/absence).

While reducing risk can lead to a lower probability of mortalityin the short-term, this may still not prevent significant levels ofinterspecific killing in the long-term (Lima, 1998), as competitorswith a large niche overlap will have correspondingly large fre-quencies of encounter. This is particularly relevant in protectedareas, which are of limited size and increasingly isolated (Lindseyet al., 2011), and where the potential for dispersal away from high-risk areas is reduced. The BVC lion population has increased fromjust 17 individuals (of which only five were female), reintroducedto the area in 1999, to an estimated 523 37 in 2012 (du Preezet al., 2014a). This equates to a current lion density of18.950 1.340 lions per 100 km2, which is one of the highest re-ported in Africa (Creel & Creel, 1997), and is still increasing (duPreez et al., 2014a). Similarly, additional interactions with otherintraguild competitor species may also affect the behaviouralecology of leopards (e.g. Vanak et al., 2013). However, here wewere specifically interested in the potential impact of the

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3730burgeoning lion population; the presence and influence of othercarnivore species would not affect the Markov chain analysis ofbehavioural switches related to lion predation risk that we presenthere.

Conclusion

Detecting new signals of interaction and avoidance betweentwo relatively well-studied species indicates the wide applicabilityof this method to the ever-expanding field of biotelemetry. Inves-tigation of behavioural transitions, with regard to the environmentand varying proximity to other species, provides information notonly on the location of an animal or its mean habitat use, but alsoon why that animal is where it is, and doing what it is doing. TheMarkov chain technique that we have presented here could befurther developed and applied to management of sensitive species,or highlight areas of potential conservation value, as well as allowaccurate predictions of interspecies interaction and habitat use inunstudied areas.

Acknowledgments

We thank the BVC shareholders and management for allowingus to conduct the research, Blondie Leathem for his constant sup-port, Paul Trethowan who helped in the field, Mark Jessopp forearly discussions on the concepts of behavioural transitions, andtwo anonymous referees who provided useful and critical com-ments on the manuscript. This project was funded by The DarwinInitiative for Biodiversity Grant 17-031. D.W.M. gratefully ac-knowledges the support of the Robertson and Recanati-KaplanFoundations. B.dP. thanks The Beit Trust for scholarship fundingand support.

References

Agouridis, C. T., Stombaugh, T. S., Workman, S. R., Koostra, B. K., Edwards, D. R., &Vanzant, E. S. (2004). Suitability of a GPS collar for grazing studies. Transactionsof the Asae, 47(4), 1321e1329.

Ainley, D. G., Ballard, G., Karl, B. J., & Dugger, K. M. (2005). Leopard seal predationrates at penguin colonies of different size. Antarctic Science, 17(3), 335e340.http://dx.doi.org/10.1017/S0954102005002750.

Bagniewska, J. M., Hart, T., Harrington, L. A., & Macdonald, D. W. (2013). HiddenMarkov analysis describes dive patterns in semiaquatic animals. BehavioralEcology, 24(3), 659e667. http://dx.doi.org/10.1093/beheco/ars217.

Bailey, T. N. (1993). The African leopard: Ecology and behavior of a solitary felid. NewYork, NY: Columbia University Press.

Balme, G., Hunter, L., & Slotow, R. (2007). Feeding habitat selection by huntingleopards Panthera pardus in a woodland savanna: prey catchability versusabundance. Animal Behaviour, 74, 589e598. http://dx.doi.org/10.1016/J.Anbehav.2006.12.014.

Balme, G. A., Slotow, R., & Hunter, L. T. B. (2010). Edge effects and the impact of non-protected areas in carnivore conservation: leopards in the Phinda-MkhuzeComplex, South Africa. Animal Conservation, 13(3), 315e323. http://dx.doi.org/10.1111/J.1469-1795.2009.00342.X.

Basille, M., Calenge, C., Marboutin, E., Andersen, R., & Gaillard, J.-M. (2008).Assessing habitat selection using multivariate statistics: some refinements ofthe ecological-niche factor analysis. Ecological Modelling, 211(1e2), 233e240.http://dx.doi.org/10.1016/j.ecolmodel.2007.09.006.

Bauer, H., Nowell, K., & Packer, C. (2013). Panthera leo. IUCN Red List of ThreatenedSpecies. Version 2013-2 Retrieved 10 January, 2014, from: http://www.iucnredlist.org/details/15951/0.

Benhamou, S., Valeix, M., Chamaille-Jammes, S., Macdonald, D. W., & Loveridge, A. J.(2014). Movement-based analysis of interactions in African lions. AnimalBehaviour, 90, 171e180. http://dx.doi.org/10.1016/j.anbehav.2014.01.030.

Bodendorfer, T., Hoppe-Dominik, B., Fischer, F., & Linsenmair, K. E. (2006). Prey ofthe leopard (Panthera pardus) and the lion (Panthera leo) in the Comoe andMarahoue National Parks, Cote d'lvoire, West Africa. Mammalia, 70(3e4),231e246. http://dx.doi.org/10.1515/Mamm.2006.037.

Bonesi, L., & Macdonald, D. W. (2004). Impact of released Eurasian otters on apopulation of American mink: a test using an experimental approach. Oikos,106(1), 9e18. http://dx.doi.org/10.1111/J.0030-1299.2004.13138.X.

Bothma, J. D. P., Nel, J. A. J., & Macdonald, A. (1984). Food niche separation be-tween 4 sympatric Namib Desert carnivores. Journal of Zoology, 202(Mar),327e340.Broekhuis, F., Cozzi, G., Valeix, M., McNutt, J. W., & Macdonald, D. W. (2013). Riskavoidance in sympatric large carnivores: reactive or predictive? Journal of An-imal Ecology, 82(5), 1098e1105. http://dx.doi.org/10.1111/1365-2656.12077.

Broom, M., & Ruxton, G. D. (2005). You can runeor you can hide: optimal strategiesfor cryptic prey against pursuit predators. Behavioral Ecology, 16(3), 534e540.http://dx.doi.org/10.1093/Beheco/Ari024.

Brown, J. S., Laundre, J. W., & Gurung, M. (1999). The ecology of fear: optimalforaging, game theory, and trophic interactions. Journal of Mammalogy, 80(2),385e399. http://dx.doi.org/10.2307/1383287.

Calenge, C., Dufour, A. B., & Maillard, D. (2005). K-select analysis: a new method toanalyse habitat selection in radio-tracking studies. Ecological Modelling, 186(2),143e153. http://dx.doi.org/10.1016/j.ecolmodel.2004.12.005.

Caro, T. (2014). Antipredator deception in terrestrial vertebrates. Current Zoology,60(1), 16e25.

Cozzi, G., Broekhuis, F., McNutt, J. W., & Schmid, B. (2013). Comparison of the effectsof artificial and natural barriers on large African carnivores: implications forinterspecific relationships and connectivity. Journal of Animal Ecology, 82(3),707e715. http://dx.doi.org/10.1111/1365-2656.12039.

Cozzi, G., Broekhuis, F., McNutt, J. W., Turnbull, L. A., Macdonald, D. W., & Schmid, B.(2012). Fear of the dark or dinner by moonlight? Reduced temporal partitioningamong Africa's large carnivores. Ecology, 93(12), 2590e2599.

Creel, S., Christianson, D., Liley, S., & Winnie, J. A. (2007). Predation risk affectsreproductive physiology and demography of elk. Science, 315(5814). http://dx.doi.org/10.1126/Science.1135918, 960e960.

Creel, S., & Creel, N. M. (1997). Lion density and population structure in the SelousGame Reserve: evaluation of hunting quotas and offtake. African Journal ofEcology, 35(2), 83e93. http://dx.doi.org/10.1111/J.1365-2028.1997.062-89062.X.

Creel, S., & Winnie, J. A. (2005). Responses of elk herd size to fine-scale spatial andtemporal variation in the risk of predation by wolves. Animal Behaviour, 69(5),1181e1189. http://dx.doi.org/10.1016/j.anbehav.2004.07.022.

Creel, S., Winnie, J. A., Jr., & Christianson, D. (2013). Underestimating the frequency,strength and cost of antipredator responses with data from GPS collars: anexample with wolves and elk. Ecology and Evolution, 3(16), 5189e5200. http://dx.doi.org/10.1002/ece3.896.

D'eon, R. G., Serrouya, R., Smith, G., & Kochanny, C. O. (2002). GPS radiotelemetryerror and bias in mountainous terrain. Wildlife Society Bulletin, 30(2), 430e439.

Fieberg, J., & Borger, L. (2012). Could you please phrase home range as a question?Journal of Mammalogy, 93(4), 890e902. http://dx.doi.org/10.1644/11-mamm-s-172.1.

Fortin, M. J., & Drapeau, P. (1995). Delineation of ecological boundariesecomparisonof approaches and significance tests. Oikos, 72(3), 323e332. http://dx.doi.org/10.2307/3546117.

Funston, P. J., Frank, L., Stephens, T., Davidson, Z., Loveridge, A., Macdonald, D. M.,et al. (2010). Substrate and species constraints on the use of track incidences toestimate African large carnivore abundance. Journal of Zoology, 281(1), 56e65.http://dx.doi.org/10.1111/J.1469-7998.2009.00682.X.

Funston, P. J., Groom, R. J., & Lindsey, P. A. (2013). Insights into the management oflarge carnivores for profitable wildlife-based land uses in African Savannas.PLoS One, 8(3), e59044. http://dx.doi.org/10.1371/journal.pone.0059044.

Funston, P. J., Mills, M. G. L., & Biggs, H. C. (2001). Factors affecting the huntingsuccess of male and female lions in the Kruger National Park. Journal of Zoology,253, 419e431. http://dx.doi.org/10.1017/S0952836901000395.

Godvik, I. M. R., Loe, L. E., Vik, J. O., Veiberg, V., Langvatn, R., & Mysterud, A. (2009).Temporal scales, trade-offs, and functional responses in red deer habitat se-lection. Ecology, 90(3), 699e710. http://dx.doi.org/10.1890/08-0576.1.

Harihar, A., Pandav, B., & Goyal, S. P. (2011). Responses of leopard Panthera pardus tothe recovery of a tiger Panthera tigris population. Journal of Applied Ecology,48(3), 806e814. http://dx.doi.org/10.1111/J.1365-2664.2011.01981.X.

Hart, T., Mann, R., Coulson, T., Pettorelli, N., & Trathan, P. (2010). Behaviouralswitching in a central place forager: patterns of diving behaviour in the mac-aroni penguin (Eudyptes chrysolophus). Marine Biology, 157(7), 1543e1553.http://dx.doi.org/10.1007/s00227-010-1428-2.

Hayward, M. W., Adendorff, J., Moolman, L., Hayward, G. J., & Kerley, G. I. H. (2007).The successful reintroduction of leopard Panthera pardus to the Addo ElephantNational Park. African Journal of Ecology, 45(1), 103e104. http://dx.doi.org/10.1111/J.1365-2028.2006.00673.X.

Hayward, M. W., O'Brien, J., Hofmeyr, M., & Kerley, G. I. H. (2007). Testing pre-dictions of the prey of lion derived from modeled prey preferences. Journal ofWildlife Management, 71(5), 1567e1575. http://dx.doi.org/10.2193/2006-264.

Hemmi, J. M., & Pfeil, A. (2010). A multi-stage anti-predator response increasesinformation on predation risk. Journal of Experimental Biology, 213(Pt 9),1484e1489. http://dx.doi.org/10.1242/jeb.039925.

Henschel, P., Hunter, L. T. B., Breitenmoser, U., Purchase, N., Packer, C., Khorozyan, I.,et al. (2008). Panthera pardus. IUCN Red List of Threatened Species. Version2013-2 Retrieved 10 January, 2014, from: http://www.iucnredlist.org/details/15954/0.

Hernandez, L., & Laundre, J. W. (2005). Foraging in the landscape of fear and itsimplications for habitat use and diet quality of elk Cervus elaphus and bisonBison bison. Wildlife Biology, 11(3), 215e220. http://dx.doi.org/10.2981/0909-6396(2005)11[215:Fitlof]2.0.Co;2.

Jessopp, M., Cronin, M., & Hart, T. (2013). Habitat-mediated dive behavior in free-ranging grey seals. PLoS One, 8(5), e63720. http://dx.doi.org/10.1371/journal.pone.0063720.

Jones, R. E. (1968). A board to measure cover used by Prairie Grouse. Journal ofWildlife Management, 32(1), 28e31. http://dx.doi.org/10.2307/3798233.

http://refhub.elsevier.com/S0003-3472(14)00417-5/sref1http://refhub.elsevier.com/S0003-3472(14)00417-5/sref1http://refhub.elsevier.com/S0003-3472(14)00417-5/sref1http://refhub.elsevier.com/S0003-3472(14)00417-5/sref1http://dx.doi.org/10.1017/S0954102005002750http://dx.doi.org/10.1093/beheco/ars217http://refhub.elsevier.com/S0003-3472(14)00417-5/sref4http://refhub.elsevier.com/S0003-3472(14)00417-5/sref4http://dx.doi.org/10.1016/J.Anbehav.2006.12.014http://dx.doi.org/10.1016/J.Anbehav.2006.12.014http://dx.doi.org/10.1111/J.1469-1795.2009.00342.Xhttp://dx.doi.org/10.1111/J.1469-1795.2009.00342.Xhttp://dx.doi.org/10.1016/j.ecolmodel.2007.09.006http://www.iucnredlist.org/details/15951/0http://www.iucnredlist.org/details/15951/0http://dx.doi.org/10.1016/j.anbehav.2014.01.030http://dx.doi.org/10.1515/Mamm.2006.037http://dx.doi.org/10.1111/J.0030-1299.2004.13138.Xhttp://refhub.elsevier.com/S0003-3472(14)00417-5/sref12http://refhub.elsevier.com/S0003-3472(14)00417-5/sref12http://refhub.elsevier.com/S0003-3472(14)00417-5/sref12http://refhub.elsevier.com/S0003-3472(14)00417-5/sref12http://dx.doi.org/10.1111/1365-2656.12077http://dx.doi.org/10.1093/Beheco/Ari024http://dx.doi.org/10.2307/1383287http://dx.doi.org/10.1016/j.ecolmodel.2004.12.005http://refhub.elsevier.com/S0003-3472(14)00417-5/sref17http://refhub.elsevier.com/S0003-3472(14)00417-5/sref17http://refhub.elsevier.com/S0003-3472(14)00417-5/sref17http://dx.doi.org/10.1111/1365-2656.12039http://refhub.elsevier.com/S0003-3472(14)00417-5/sref19http://refhub.elsevier.com/S0003-3472(14)00417-5/sref19http://refhub.elsevier.com/S0003-3472(14)00417-5/sref19http://refhub.elsevier.com/S0003-3472(14)00417-5/sref19http://dx.doi.org/10.1126/Science.1135918http://dx.doi.org/10.1126/Science.1135918http://dx.doi.org/10.1111/J.1365-2028.1997.062-89062.Xhttp://dx.doi.org/10.1016/j.anbehav.2004.07.022http://dx.doi.org/10.1002/ece3.896http://dx.doi.org/10.1002/ece3.896http://refhub.elsevier.com/S0003-3472(14)00417-5/sref23http://refhub.elsevier.com/S0003-3472(14)00417-5/sref23http://refhub.elsevier.com/S0003-3472(14)00417-5/sref23http://dx.doi.org/10.1644/11-mamm-s-172.1http://dx.doi.org/10.1644/11-mamm-s-172.1http://dx.doi.org/10.2307/3546117http://dx.doi.org/10.2307/3546117http://dx.doi.org/10.1111/J.1469-7998.2009.00682.Xhttp://dx.doi.org/10.1371/journal.pone.0059044http://dx.doi.org/10.1017/S0952836901000395http://dx.doi.org/10.1890/08-0576.1http://dx.doi.org/10.1111/J.1365-2664.2011.01981.Xhttp://dx.doi.org/10.1007/s00227-010-1428-2http://dx.doi.org/10.1111/J.1365-2028.2006.00673.Xhttp://dx.doi.org/10.1111/J.1365-2028.2006.00673.Xhttp://dx.doi.org/10.2193/2006-264http://dx.doi.org/10.1242/jeb.039925http://www.iucnredlist.org/details/15954/0http://www.iucnredlist.org/details/15954/0http://dx.doi.org/10.2981/0909-6396(2005)11[215:Fitlof]2.0.Co;2http://dx.doi.org/10.2981/0909-6396(2005)11[215:Fitlof]2.0.Co;2http://dx.doi.org/10.1371/journal.pone.0063720http://dx.doi.org/10.1371/journal.pone.0063720http://dx.doi.org/10.2307/3798233

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 31Kamler, J. F., Davies-Mostert, H. T., Hunter, L., & Macdonald, D. W. (2007). Predationon black-backed jackals (Canis mesomelas) by African wild dogs (Lycaon pictus).African Journal of Ecology, 45(4), 667e668. http://dx.doi.org/10.1111/J.1365-2028.2007.00768.X.

Kamler, J. F., Stenkewitz, U., & Macdonald, D. W. (2013). Lethal and sublethal effectsof black-backed jackals on cape foxes and bat-eared foxes. Journal ofMammalogy, 94(2), 295e306. http://dx.doi.org/10.1644/12-Mamm-a-122.1.

Karanth, K. U., & Sunquist, M. E. (2000). Behavioural correlates of predation by tiger(Panthera tigris), leopard (Panthera pardus) and dhole (Cuon alpinus) inNagarahole, India. Journal of Zoology, 250, 255e265.

de Knegt, H. J., van Langevelde, F., Skidmore, A. K., Delsink, A., Slotow, R., Henley, S.,et al. (2011). The spatial scaling of habitat selection by African elephants.Journal of Animal Ecology, 80(1), 270e281. http://dx.doi.org/10.1111/j.1365-2656.2010.01764.x.

Kock, M. D., Meltzer, D., & Burroughs, R. (2006). Chemical and physical restraint ofwild animals. In A training and field manual for African species. Greyton, SouthAfrica: IWVS Africa.

Krebs, J. R., & Davies, N. B. (1991). Economic decisions and the individual. InJ. R. Krebs, N. B. Davies, & J. Parr (Eds.), An introduction to behavioural ecology(3rd ed.). (pp. 48e76). Cambridge, MA: Blackwell Scientific Publications.

Laundre, J. W., Hernandez, L., & Altendorf, K. B. (2001). Wolves, elk, and bison:reestablishing the landscape of fear in Yellowstone National Park, USA.Canadian Journal of Zoology, 79(8), 1401e1409. http://dx.doi.org/10.1139/Cjz-79-8-1401.

Lima, S. L. (1998). Nonlethal effects in the ecology of predator-prey inter-actionseWhat are the ecological effects of anti-predator decision-making?Bioscience, 48(1), 25e34. http://dx.doi.org/10.2307/1313225.

Lindsey, P. A., Romanach, S. S., Tambling, C. J., Chartier, K., & Groom, R. (2011).Ecological and financial impacts of illegal bushmeat trade in Zimbabwe. Oryx,45(1), 96e111. http://dx.doi.org/10.1017/50030605310000153.

Lind, J., & Cresswell, W. (2005). Determining the fitness consequences of anti-predation behavior. Behavioral Ecology, 16(5), 945e956. http://dx.doi.org/10.1093/Beheco/Ari075.

Lone, K., Loe, L. E., Gobakken, T., Linnell, J. D. C., Odden, J., Remmen, J., & Mysterud, A.(2014). Living and dying in a multi-predator landscape of fear: roe deer aresqueezed by contrasting pattern of predation risk imposed by lynx and humans.Oikos, (123), 641e651. http://dx.doi.org/10.1111/j.1600-0706.2013.00938.x.

Loarie, S. R., Tambling, C. J., & Asner, G. P. (2013). Lion hunting behaviour andvegetation structure in an African savanna. Animal Behaviour, 85(5), 899e906.http://dx.doi.org/10.1016/j.anbehav.2013.01.018.

McDougal, C. (1988). Leopard and tiger interactions at Royal Chitwan National Park,Nepal. Journal of the Bombay Natural History Society, 85, 609e610.

Milinski, M., & Heller, R. (1978). Influence of a predator on optimal foragingbehavior of sticklebacks (Gasterosteus aculeatus L). Nature, 275(5681), 642e644.http://dx.doi.org/10.1038/275642a0.

Mosser, A., & Packer, C. (2009). Group territoriality and the benefits of sociality inthe African lion, Panthera leo. Animal Behaviour, 78(2), 359e370. http://dx.doi.org/10.1016/J.Anbehav.2009.04.024.

Ogara, W. O., Githai, N. J., Andanje, S. A., Oguge, N., Nduati, D. W., & Mainga, A. O.(2010). Determination of carnivores prey base by scat analysis in Samburucommunity group ranches in Kenya. African Journal of Environmental Science andTechnology, 4(8), 540e546.

Palomares, F., & Caro, T. M. (1999). Interspecific killing among mammaliancarnivores. American Naturalist, 153(5), 492e508. http://dx.doi.org/10.1086/303189.

Patterson, T. A., Basson, M., Bravington, M. V., & Gunn, J. S. (2009). Classifyingmovement behaviour in relation to environmental conditions using hiddenFigure A1. Image series of lions chasing and killing a leopard in grasslaMarkov models. Journal of Animal Ecology, 78(6), 1113e1123. http://dx.doi.org/10.1111/J.1365-2656.2009.01583.X.

du Preez, B. D. (n.d.). [Annual carnivore density surveys of the Bubye ValleyConservancy, Zimbabwe]. Unpublished raw data.

du Preez, B. D., Loveridge, A. J., & Macdonald, D. W. (2014a). Experimentally testingan intraguild hostility hypothesis: the impact of lions Panthera leo on leopardsPanthera pardus. Submitted manuscript.

du Preez, B. D., Loveridge, A. J., & Macdonald, D. W. (2014b). To bait or not to bait: acomparison of camera-trapping methods for estimating leopard Panthera par-dus density. Biological Conservation, (176), 153e161.

Rempel, R. S., & Rodgers, A. R. (1997). Effects of differential correction on accuracy ofa GPS animal location system. Journal of Wildlife Management, 61(2), 525e530.http://dx.doi.org/10.2307/3802611.

Sih, A. (1997). To hide or not to hide? Refuge use in a fluctuating environment.Trends in Ecology & Evolution, 12(10), 375e376. http://dx.doi.org/10.1016/S0169-5347(97)87376-4.

Spitz, J., Rousseau, Y., & Ridoux, V. (2006). Diet overlap between harbour porpoiseand bottlenose dolphin: an argument in favour of interference competition forfood? Estuarine, Coastal and Shelf Science, 70(1e2), 259e270. http://dx.doi.org/10.1016/j.ecss.2006.04.020.

Stander, P. E. (1998). Spoor counts as indices of large carnivore populations: therelationship between spoor frequency, sampling effort and true density. Journalof Applied Ecology, 35(3), 378e385. http://dx.doi.org/10.1046/J.1365-2664.1998.00313.X.

Valeix, M., Fritz, H., Loveridge, A. J., Davidson, Z., Hunt, J. E., Murindagomo, F., et al.(2009). Does the risk of encountering lions influence African herbivorebehaviour at waterholes? Behavioral Ecology and Sociobiology, 63(10),1483e1494. http://dx.doi.org/10.1007/s00265-009-0760-3.

Valeix, M., Hemson, G., Loveridge, A. J., Mills, G., & Macdonald, D. W. (2012).Behavioural adjustments of a large carnivore to access secondary prey in ahuman-dominated landscape. Journal of Applied Ecology, 49(1), 73e81. http://dx.doi.org/10.1111/J.1365-2664.2011.02099.X.

Valeix, M., Loveridge, A. J., Chamaille-Jammes, S., Davidson, Z., Murindagomo, F.,Fritz, H., et al. (2009). Behavioral adjustments of African herbivores to predationrisk by lions: spatiotemporal variations influence habitat use. Ecology, 90(1),23e30. http://dx.doi.org/10.1890/08-0606.1.

Valeix, M., Loveridge, A. J., Davidson, Z., Madzikanda, H., Fritz, H., &Macdonald, D. W. (2010). How key habitat features influence large terrestrialcarnivore movements: waterholes and African lions in a semi-arid savanna ofnorth-western Zimbabwe. Landscape Ecology, 25(3), 337e351. http://dx.doi.org/10.1007/S10980-009-9425-X.

Vanak, A. T., Fortin, D., Thaker, M., Ogden, M., Owen, C., Greatwood, S., et al. (2013).Moving to stay in place: behavioral mechanisms for coexistence of African largecarnivores. Ecology, 94(11), 2619e2631. http://dx.doi.org/10.1890/13-0217.1.

Whitehead, H., & Jonsen, I. D. (2013). Inferring animal densities from tracking datausing Markov chains. PLoS One, 8(4), e60901. http://dx.doi.org/10.1371/journal.pone.0060901.

Zollner, P. A., & Lima, S. L. (2005). Behavioral tradeoffs when dispersing across apatchy landscape. Oikos, 108(2), 219e230. http://dx.doi.org/10.1111/J.0030-1299.2005.13711.X.APPENDIX 1. LIONeLEOPARD INTRAGUILD PREDATIONnd habitat. Copyright Stanslaus Mwampeta, Serengeti Lion Project.

http://dx.doi.org/10.1111/J.1365-2028.2007.00768.Xhttp://dx.doi.org/10.1111/J.1365-2028.2007.00768.Xhttp://dx.doi.org/10.1644/12-Mamm-a-122.1http://refhub.elsevier.com/S0003-3472(14)00417-5/sref41http://refhub.elsevier.com/S0003-3472(14)00417-5/sref41http://refhub.elsevier.com/S0003-3472(14)00417-5/sref41http://refhub.elsevier.com/S0003-3472(14)00417-5/sref41http://dx.doi.org/10.1111/j.1365-2656.2010.01764.xhttp://dx.doi.org/10.1111/j.1365-2656.2010.01764.xhttp://refhub.elsevier.com/S0003-3472(14)00417-5/sref43http://refhub.elsevier.com/S0003-3472(14)00417-5/sref43http://refhub.elsevier.com/S0003-3472(14)00417-5/sref43http://refhub.elsevier.com/S0003-3472(14)00417-5/sref44http://refhub.elsevier.com/S0003-3472(14)00417-5/sref44http://refhub.elsevier.com/S0003-3472(14)00417-5/sref44http://refhub.elsevier.com/S0003-3472(14)00417-5/sref44http://dx.doi.org/10.1139/Cjz-79-8-1401http://dx.doi.org/10.1139/Cjz-79-8-1401http://dx.doi.org/10.2307/1313225http://dx.doi.org/10.1017/50030605310000153http://dx.doi.org/10.1093/Beheco/Ari075http://dx.doi.org/10.1093/Beheco/Ari075http://dx.doi.org/10.1111/j.1600-0706.2013.00938.xhttp://dx.doi.org/10.1016/j.anbehav.2013.01.018http://refhub.elsevier.com/S0003-3472(14)00417-5/sref49http://refhub.elsevier.com/S0003-3472(14)00417-5/sref49http://refhub.elsevier.com/S0003-3472(14)00417-5/sref49http://dx.doi.org/10.1038/275642a0http://dx.doi.org/10.1016/J.Anbehav.2009.04.024http://dx.doi.org/10.1016/J.Anbehav.2009.04.024http://refhub.elsevier.com/S0003-3472(14)00417-5/sref52http://refhub.elsevier.com/S0003-3472(14)00417-5/sref52http://refhub.elsevier.com/S0003-3472(14)00417-5/sref52http://refhub.elsevier.com/S0003-3472(14)00417-5/sref52http://refhub.elsevier.com/S0003-3472(14)00417-5/sref52http://dx.doi.org/10.1086/303189http://dx.doi.org/10.1086/303189http://dx.doi.org/10.1111/J.1365-2656.2009.01583.Xhttp://dx.doi.org/10.1111/J.1365-2656.2009.01583.Xhttp://refhub.elsevier.com/S0003-3472(14)00417-5/sref55http://refhub.elsevier.com/S0003-3472(14)00417-5/sref55http://refhub.elsevier.com/S0003-3472(14)00417-5/sref55http://refhub.elsevier.com/S0003-3472(14)00417-5/sref55http://dx.doi.org/10.2307/3802611http://dx.doi.org/10.1016/S0169-5347(97)87376-4http://dx.doi.org/10.1016/S0169-5347(97)87376-4http://dx.doi.org/10.1016/j.ecss.2006.04.020http://dx.doi.org/10.1016/j.ecss.2006.04.020http://dx.doi.org/10.1046/J.1365-2664.1998.00313.Xhttp://dx.doi.org/10.1046/J.1365-2664.1998.00313.Xhttp://dx.doi.org/10.1007/s00265-009-0760-3http://dx.doi.org/10.1111/J.1365-2664.2011.02099.Xhttp://dx.doi.org/10.1111/J.1365-2664.2011.02099.Xhttp://dx.doi.org/10.1890/08-0606.1http://dx.doi.org/10.1007/S10980-009-9425-Xhttp://dx.doi.org/10.1007/S10980-009-9425-Xhttp://dx.doi.org/10.1890/13-0217.1http://dx.doi.org/10.1371/journal.pone.0060901http://dx.doi.org/10.1371/journal.pone.0060901http://dx.doi.org/10.1111/J.0030-1299.2005.13711.Xhttp://dx.doi.org/10.1111/J.0030-1299.2005.13711.X

Grassland (6%)Scrub (76%)

Riparian (18%)

(a)

Grassland (3%)

Scrub (49%)

Riparian (48%)

(b)

Figure A3. Comparison of habitat use by male leopard LM4 between (a) the exclosure, without lions, and (b) the other side of the fence, with lions.

Figure A2. Male leopard LM4's GPS data while the electrified fence (black line) was in place. Lions were present on the western side of the fence, but excluded from the eastern side.Imagery Copyright 2014 TerraMetrics, Inc., www.terrametrics.com.

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3732

http://www.terrametrics.com

0

200

400

600

800

1000

1400

Scrub Riparian

(a)

0

200

400

600

800

1200

1400

Grassland Scrub Riparian

(b)

1200M

ale

leop

ard

LM

4 st

ep l

engt

h (

m)

Grassland

Habitat type

1000

Figure A4. LM4's average hourly distance moved (step length) in each habitat type where lions were absent compared with where they were present. The error bars represent thestandard error of the mean.

0

20

40

60

100

GS GR SG SS SR RG RS RR

(a)

0

20

40

60

100

GG GS GR SG SS SR RG RS RR

(b)

80

Mal

e le

opar

d L

M4

hab

itat

tra

nst

ion

pro

babi

lity

(%

)

80

GG

Habitathabitat transition

Figure A5. LM4's habitat-to-habitat transition probabilities on either side of the electrified fence between areas where lions were (a) absent and (b) present. The error barsrepresent the standard error of the mean.

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 33

Figure A6. (a) GPS data for a lioness, FL3, up to the day before removal of the electrified fence (dark grey); (b) FL3's data after removal of the electrified fence (original positionindicated in light grey), showing her movement into the area within the original exclosure in which LM4 once inhabited before his disappearance after the removal of the fence andentry of the lions. Imagery Copyright 2014 TerraMetrics, Inc., www.terrametrics.com.

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3734

http://www.terrametrics.com

B. du Preez et al. / Animal Behaviour 100 (2015) 22e37 35APPENDIX 2. R SCRIPT: HABITAT TRANSITION MATRIX

B. du Preez et al. / Animal Behaviour 100 (2015) 22e3736APPENDIX 3. R SCRIPT: SIMULTANEOUS DISTANCES BETWEENSIMULTANEOUSLY COLLARED ANIMALSTable A2APPENDIX 4. RESULTS OF ALL STATISTICAL TESTSTable A1Comparison of leopard habitat transition probabilities between where lions arepresent versus where lions are absent

From To P F Power Effect size (R2)

Grassland Grassland 0.247 1.598 0.237 0.186Grassland Scrub 0.016a 10.019 0.866 0.589Grassland Riparian 0.698 0.164 0.068 0.023Scrub Grassland 0.196 1.924 0.280 0.161Scrub Scrub 0.404 0.759 0.139 0.071Scrub Riparian 0.216 1.746 0.259 0.149Riparian Grassland 0.388 0.825 0.147 0.084Riparian Scrub 0.831 0.048 0.055 0.005Riparian Riparian 0.260 1.446 0.222 0.138

a Denotes a significant result.

Comparison of leopard habitat transition probabilities versus lion proximity be-tween open habitat and closed habitat types

From To P F Power Effect size (R2)

Closed Open 0.001a 9.528 0.997 0.641Open Closed 0.621 0.254 0.079 0.014

a Denotes a significant result.

Table A3Comparison of leopard step length between habitats where lions are present versuswhere lions are absent

Lions P F Power Effect size (R2)

Absent 0.942 0.060 0.059 0.008Present 0.046a 4.022 0.704 0.401

a Denotes a significant result.

Table A6Comparison of leopard LM4's habitat transition probabilities between where lionsare present versus where lions are absent

From To P F Power Effect size (R2)

Grassland Grassland