The potential of occupancy modelling as a tool for monitoring wild primate populations

The potential of occupancy modelling as a tool formonitoring wild primate populationsA. Keane1,2,3, T. Hobinjatovo4, H. J. Razafimanahaka4, R. K. B. Jenkins1,4,5 & J. P. G. Jones1

1 School of the Environment, Natural Resources and Geography, Bangor University, Bangor, Gwynedd, UK2 Department of Anthropology, University College London, London, UK3 Institute of Zoology, Zoological Society of London, Regent’s Park, London, UK4 Madagasikara Voakajy, Antananarivo, Madagascar5 Durrell Institute of Conservation and Ecology, School of Anthropology and Conservation, University of Kent, Canterbury, Kent, UK

Keywords

power; distance sampling; transect;mammal; survey; lemur; Madagascar;occupancy.

Correspondence

Aidan Keane, Department of Anthropology,University College London, 14 TavitonStreet, London WC1H 0BW, UK.Email: [email protected]

Editor: Res AltweggAssociate Editor: Michael Schaub

Received 8 July 2011; accepted 21February 2012

doi:10.1111/j.1469-1795.2012.00575.x

AbstractPrimates are a global conservation priority, with half of known species consideredthreatened with extinction. Monitoring trends in primate populations is importantfor identifying species in particular need of conservation action, and evaluatingthe effectiveness of interventions. Most existing primate survey methods aim tomeasure abundance. However, obtaining estimates of abundance with acceptableprecision to detect changes in population is often expensive and time consuming.Evidence from other taxa suggests that estimating occupancy (the proportion ofthe area used by the species) may be less resource-intensive, yet still provide usefulinformation for monitoring population trends. We investigate the potential ofoccupancy modelling for monitoring forest primates using a case study of threespecies of diurnal lemurs in the eastern rainforest of Madagascar. We estimateddetectability and occupancy from a survey with three visits to 30 sites. Ourestimates suggest that precision in occupancy estimates would be maximized byvisiting a larger number of sites (therefore with limited repeat visits) for Indri indri,whereas the optimal monitoring design for Eulemur fulvus and Propithecusdiadema, which showed very low detectability in our surveys, involves more fre-quent visits to fewer sites. Power analyses suggested that a meaningful reductionin occupancy could be detected with reasonable effort for easily detected species,but the method may prove impractical for more cryptic species. Primates pose anumber of practical challenges for occupancy modelling, including choosingappropriate survey designs to satisfy closure assumptions. We suggest that if theseissues can be overcome, occupancy modelling has the potential to become avaluable addition to the monitoring toolbox for the study of forest primates.

Introduction

Primates are a global conservation priority (Chapman,Lawes & Eeley, 2006; Schipper et al., 2008). Although theyare ecologically (Chapman, 1995), economically (Wilkie &Carpenter, 1999) and culturally (Baker et al., 2009) impor-tant, they are also highly threatened by habitat loss (Col-ishaw & Dunbar, 2000), hunting (Fa & Brown, 2009) anddisease (Köndgen et al., 2008), with 49% of species classifiedas threatened according to the International Union for Con-servation of Nature (IUCN)’s Red List criteria (IUCN,2010). Monitoring is important to inform management deci-sions (e.g. to identify priority species or sites for conservationinterventions) and to audit the effect of management actionsor policies (Jones et al., 2011). Unfortunately, although pri-mates are the focus of a great deal of research, the methodsused to monitor populations remain poorly developed.

It has long been recognized that monitoring mammals intropical forests is difficult (Plumptre, 2000) and primatespresent particular challenges (Buckland et al., 2010a). Manyspecies are cryptic, occur at low densities and occupy largehome ranges (Nowak, 1999). In addition, primates are oftenalert, highly mobile and likely to respond to the presence ofresearchers (e.g. Savage et al., 2010), thereby violating theassumptions of many survey techniques (Keane, Jones &Milner-Gulland, 2011). Buckland et al. (2010a,b) recentlyhighlighted several methodological flaws that are commonin primate surveys, including poor sampling designs andfailures to properly account for imperfect detection. Conse-quently, there is a pressing need to continue to develop thesuite of techniques available for monitoring primate speciesof conservation concern.

A key issue in most biological surveys is the problem ofimperfect detection: individuals may not be detected even if

bs_bs_banner

Animal Conservation. Print ISSN 1367-9430

Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London 1

they are present within the sampled area. If not properlyaccounted for, this can introduce bias into estimates ofpopulation abundance, distribution or species richness(MacKenzie et al., 2006). Several approaches, includingdistance sampling, multiple-observer surveys and capture–recapture techniques, can be used to derive rigorous, unbi-ased measures of the abundance of a species by explicitlymodelling the probability of detection (e.g. Buckland et al.,2010a,b). In some cases, however, obtaining estimates ofabundance with an acceptable level of precision can be timeconsuming and expensive (e.g. Walsh & White, 1999; Ogutuet al., 2006). In the context of conservation, where budgetrestrictions are the norm (James, Gaston & Balmford,1999), surveys using occupancy as a state variable have beenadvocated as a less resource-intensive alternative to distancesampling for the monitoring of wild populations (MacKen-zie et al., 2006).

Occupancy may be defined as the proportion of an areathat is occupied or used by a species (MacKenzie et al.,2002). In many situations, a precise measure of occupancycan be obtained with less effort and cost than would berequired to obtain a precise estimate of abundance (Josephet al., 2006). As with abundance-based methods, occupancysurveys can be designed to allow probability of detection tobe explicitly modelled (MacKenzie et al., 2006). If it is meas-ured at an appropriate scale, occupancy is positively corre-lated with the population size for many species and canserve as an adequate surrogate for abundance; in othercases, occupancy may be of interest in its own right (Mac-Kenzie & Nichols, 2004). Occupancy modelling is thereforeincreasingly used for monitoring threatened populations ina wide variety of contexts [e.g. bat species in the US (Weller,2008); forest plant communities in Australia (Penman,Binns & Kavanagh, 2009); elephants in Zimbabwe (Martinet al., 2010); and frogs and toads in Britain (Sewell, Beebee& Griffiths, 2010)]. To date, however, there has been littleeffort to examine the strengths and weakness of occupancy-based approaches in the context of primate conservation.

To our knowledge, formal occupancy surveys have onlybeen evaluated as a means of monitoring primate popula-tions in one previous study. Guillera-Arroita et al. (2010)carried out surveys of the Alaotran gentle lemur (Hapalemuralaotrensis), a small, critically endangered primate that isendemic to the Lac Alaotra wetlands in north-easternMadagascar. However, because of the unique nature of thissystem (the lemur is restricted to reed beds only accessible byboats along established routes) and the prohibitive difficul-ties of implementing a probabilistic sampling scheme in thisspecies’ habitat, the applicability of this study for informingthe design of occupancy surveys for primate populations inother habitats is limited. Approximately 90% of the world’sprimates live in tropical forests (Mittermeier, 1988), and inthis paper, we provide the first examination of occupancymodels as a tool for monitoring primate populations inforest habitats.

For single-species occupancy models, the choice of anoptimal sampling design depends on the proportion of sam-pling units that are occupied by a species and the probability

that it is detected when it is present in a sampling unit(MacKenzie & Royle, 2005). We present data from a casestudy, in which pilot occupancy surveys were carried out forthree lemur species with differing ecologies in an area offorest in eastern Madagascar. These species are mobile andresponsive, exemplifying many of the challenges that mustbe overcome for effective primate monitoring. Using ourestimates of occupancy and probability of detection derivedfrom these data, we estimate the survey effort that would berequired to detect changes in occupancy over time andinvestigate whether these model parameters are affected bythe presence of artisanal gold mining, an important poten-tial threat to lemur populations in the study area. Based onour findings, we go on to assess the suitability of thisapproach for monitoring multiple species of lemur in thesame surveys and discuss the broader implications of thesefindings for the design of monitoring protocols for primatepopulations in the wild.

Methods

Study area

The study was conducted in a forest known locally as‘Mangabe’, which is currently being established as a newprotected area with provisions for sustainable use. TheMangabe Forest is located in the Moramanga District ineastern Madagascar (Fig. 1). Even though the forest hasprovisional protected area status, which, in association withother laws, prohibits the hunting of lemurs, extraction ofminerals and conversion of natural forest into agriculture,there are chronic and growing threats to the wildlife. Lemursare hunted for food (Jenkins et al., 2011) and to protect cropsusing traps, snares and guns. Illegal gold mining in Mangabehas resulted in increased immigration and the establishmentof temporary settlements within the forest; both of theseappear to be linked with lemur hunting. For the purposes ofmanagement, there is therefore a need to establish a baselineof the status and distribution of lemur species in the newprotected area, and to investigate whether mining poses animportant threat to the species’ populations.

A preliminary survey in Mangabe conducted in 2009revealed the presence of seven lemur species. The diurnaland cathemeral lemurs are of higher conservation concern(Table 1) and are generally preferred over nocturnal speciesfor consumption. In this study, we therefore chose to focuson three day-active species, all of which are easily identifiedand are hunted for food (Table 1).

Survey design

We divided the survey area into a grid of square samplingunits, 200 ¥ 200 m in size. Cells within the area were disre-garded if our land cover map indicated that they were notcompletely forested. The remaining cells were individuallynumbered and a subset of 30 was chosen at random. Becauseof the difficult nature of the terrain and the presence of a large

Monitoring wild primate populations using occupancy modelling A. Keane et al.

2 Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London

river in the area, we drew a further sample of 20 cells, whichserved as replacements if cells within our original sampleproved to be inaccessible (e.g. due to flooding).

Each cell was visited three times by a team of oneresearcher and two local guides during the morning or after-noon when the target species are usually active. In order tominimize differences in the effort expended, the team fol-lowed a regular route through each cell: initially walking in

a straight line along one side of the square, then walking inthe opposite direction along a parallel line through thecentre of the cell, before finally walking back along the otheropposite side, finishing at the opposite corner of the squarefrom their starting point. All sightings of the target specieswere recorded. To further minimize potential bias, onlydirect visual observations of lemurs were recorded as pres-ences (so as to avoid the inclusion of individuals heardcalling from a location outside the sampling unit). The pres-ence or absence of gold mining (which can easily be inferredfrom the presence of small, distinctive pits dug in theground) was also noted. Each grid square took approxi-mately 2.5 h to complete.

The repeat visits to each cell were carried out at 3- to5-day intervals in order to minimize possible effects of dis-turbance on the behaviour of the lemurs. However, the threespecies are mobile and have home ranges that are likely to beconsiderably larger than the 200 ¥ 200-m sampling units.For example, Wrangham et al. (1993) estimated that the dayranges of Eulemur fulvus and Indri indri are 140 and 250 m,respectively. Consequently, it is likely that the survey maynot have been closed to changes in occupancy. Assumingthat movement throughout the home range is random, weinterpret our estimates of ‘occupancy’ as reflecting the use of

Figure 1 Map indicating the location of the study area within Madagascar, and details of the Mangabe forest, including the sampling grid andthe cells visited in this study.

Table 1 Lemur species identified from Mangabe forest duringsurveys in 2009 and 2010. Bold indicates species that are the focusof this study

Species IUCN red list Activity Body mass (kg)

Indri indri EN Diurnal 6.0–9.5Propithecus diadema EN Diurnal 6.0–8.5Eulemur fulvus NT Cathemeral 1.7–2.1Hapalemur griseus VU Cathemeral 0.70–0.85Cheirogaleus sp. LC Nocturnal 0.35–0.40Microcebus sp. DD Nocturnal 0.045–0.048Avahi laniger LC Nocturnal 1.0–1.4

LC, least concern; NT, near threatened; VU, vulnerable; EN, endan-gered; DD, data deficient; IUCN, International Union for Conservationof Nature.

A. Keane et al. Monitoring wild primate populations using occupancy modelling

Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London 3

the area rather than true occupancy and ‘probability ofdetection’ as a compound measure of the probability thatthe species is currently occupying the site and the probabil-ity that it is detected (see MacKenzie et al., 2006 for adiscussion).

Analyses

We fitted a series of single-season occupancy models to thedata for each species separately using the ‘occu’ functionfrom the package ‘unmarked’ (Fiske & Chandler, 2011) in Rversion 2.13.0 (R Development Core Team, 2011). Single-season occupancy models involve the estimation of twoparameters: the probability that the species in question ispresent in a cell, or occupancy (y), and the probability thatthe species is detected given that it is present (P). The basicstructure of these models is described by MacKenzie et al.(2002).

We considered models in which occupancy and probabil-ity of detection were either fixed or modelled as a function ofthe presence of mining in the survey area, giving a candidateset of four models for each species. Following MacKenzie &Bailey (2004), we assessed the goodness-of-fit of the modelsby comparing their X2 statistics to reference distributionscalculated from a parametric bootstrap. We then assessedthe relative fit of the models within each candidate setaccording to their Akaike information criterion (AIC)values (Akaike 1974). Models were ranked, with the lowestAIC value indicating the best fit to the data, and AICweights were calculated. These weights were used to identifythe best-fitting model and to calculate model-averaged esti-mates and confidence intervals (CIs) for the probability ofdetection and occupancy of the three species in areas withand without mining (Burnham & Anderson, 2002).

Based on our estimates of probability of detection andoccupancy, we calculated the precision with which occu-pancy could be estimated for the three species under severalalternative sampling schemes. Following MacKenzie &Royle (2005), these calculations were based on asymptotic(i.e. large sample) assumptions for the variance of the occu-pancy estimator. We also performed a simulation-basedpower analysis to examine the effect of total survey effortand the distribution of effort between the number of sites (s)and the number of repeat visits to each site (K) on the powerto detect changes in the area occupied by each species. Inthis case, we consider adequate power to be greater than orequal to 0.8 (i.e. probability of a false-negative result, ortype II error, less than or equal to 0.2). For each scenariowithin the power analysis, two encounter histories were gen-erated at random from our estimates of occupancy andprobability of detection: the first using the original esti-mates, and the second having occupancy 0.3 lower than theoriginal estimates (e.g. occupancy of 0.6 would be reducedto 0.3). This represents a substantial decline which a typicalconservation monitoring scheme might wish to detect. Prob-ability of detection was assumed to be identical in each ofthe encounter histories. Two models were fitted to the datausing the ‘occu’ function from the package ‘unmarked’: one

in which occupancy parameters were estimated separatelyfor the two sites, and a null model in which a single occu-pancy parameter is estimated for both sites. The two modelswere compared using a likelihood ratio test and we recordedwhether the test was significant at a = 0.05. This procedurewas repeated 1000 times for each scenario, and the resultswere interpreted as the power to say that any decline hadoccurred, given that the true decline was 0.3.

ResultsDuring our surveys, we observed E. fulvus 17 times at 16sites, Propithecus diadema 16 times at 15 sites and I. indri 17times at 12 sites. Tests of goodness-of-fit performed on thesaturated models did not detect a lack of fit for any ofthe three species (P. diadema: X2 = 5.30, P = 0.42; I. indri:X2 = 9.70, P = 0.11; E. fulvus: X2 = 5.12, P = 0.44). For eachspecies, comparison of AIC values suggested that the bestsupported model was the simplest one in which occupancyand probability of detection were the same in areas withevidence of mining as those without it (Supporting Infor-mation, Tables S1–S3).

Our model-averaged estimates of probability of detectionwere generally low, estimated to be P = 0.17 (95% CI: 0.08–0.27) for P. diadema, P = 0.18 (95% CI: 0.09–0.28) forE. fulvus and P = 0.36 (95% CI: 0.10–0.62) for I. indri(Fig. 2a). Both P. diadema and E. fulvus were estimated touse the entire survey area (y = 1.0, 95% CIs: 0.98–1.00 inboth cases), and I. indri estimated to use y = 0.57 of the area(95% CI: 0.20–0.94; Fig. 2b). Generally, the AIC values andmodel-averaged parameter estimates provide no evidencefor differences in occupancy or probability of detectionbetween cells where mining was observed and areas where itwas not. The model-averaged CIs for occupancy and prob-ability of detection were derived using asymptotic assump-tions, which are known to be unreliable for estimates on theboundary of parameter space (i.e. when a parameter is esti-mated as 0 or 1). We therefore calculated a complementaryset of CIs for the best supported models for each speciesusing likelihood profiling which may perform better underthese conditions (Fiske & Chandler, 2011). The profile CIsfor probability of detection were as follows: P. diadema,95% CI: 0.11–0.30; E. fulvus, 95% CI: 0.12–0.30; I. indri,95% CI: 0.14–0.54. The profile CIs for use of the area wereas follows: P. diadema, 95% CI: 0.65–1.0; E. fulvus, 95% CI:0.69–1.0; and I. indri, 95% CI: 0.31–1.0 (Fig. 2).

The observed values of occupancy and probability anddetection were used to predict the precision of estimates ofoccupancy that could be obtained for the three species,assuming three levels of total effort (Fig. 3). For P. diademaand E. fulvus, which have high occupancy but low detecta-bility, the minimal standard errors for estimates of occu-pancy are obtained if each site is visited 20 times, althoughprecision is similar if sites are visited between 14 and 25times. High numbers of visits per site imply that fewer dif-ferent sites can be sampled. By contrast, the most preciseestimates of occupancy of I. indri are obtained when agreater number of sites are each visited five times.

Monitoring wild primate populations using occupancy modelling A. Keane et al.

4 Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London

Based on our initial findings, we examined the power todetect changes in occupancy for P. diadema/E. fulvus and forI. indri (Fig. 4), and carried out a sensitivity analysis toexamine the robustness of the power analysis to uncertaintyin our parameter estimated (see Supporting Information,Fig. S1). The optimal survey designs involved 5 repeat visitsper site for I. indri and 14 visits per site for P. diadema/E. fulvus. Under these designs, a much smaller level of totalsurvey effort is needed to obtain adequate power for I. indrithan for P. diadema/E. fulvus. Power greater than or equal to0.8 could be obtained for I. indri for a total survey effort of275 site-visits (i.e. 55 sites, each visited 5 times). ForP. diadema/E. fulvus, achieving this level of power wouldrequire a total of 448 site-visits (32 sites, each visited 14times). To obtain power greater than or equal to 0.8 for allthree species in the same survey, the minimum total effortrequired would be 495 site-visits, visiting 45 sites 11 timeseach.

DiscussionThe vast majority of primate surveys seek to estimate abun-dance (Morgan et al., 2006) or relative abundance (Rovero

et al., 2006) at a site to compare between sites (Kümpel et al.,2008) or investigate trends over time (Teelen, 2007). Abun-dance is usually the most informative state variable for thepurposes of population monitoring, but obtaining precisemeasures of abundance may not always be cost-effective(Joseph et al., 2006) and attempts to estimate primate abun-dance have often suffered from design limitations, whichmake interpretation of the results problematic (Bucklandet al., 2010a). In this study, we have used a case study ofoccupancy modelling for three lemur species to explore thepotential of occupancy surveys as an alternative approach tomonitoring wild primate populations. We obtained prelimi-nary estimates of the occupancy and detectability of threespecies of lemur, but their precision was limited by the smallsamples collected during our surveys and may have beenaffected by violations of the closure assumption. As a result,there is undoubtedly considerable scope for refining oursurvey design to improve its efficiency.

Our methodology was chosen in order to balance anumber of conflicting design requirements. We attempted tosatisfy the assumption of ‘closure’ (i.e. the assumption thatthe occupancy status of cells within the sampling area doesnot change within the survey) by minimizing the length of

Figure 2 Estimates of probability of detec-tion and occupancy for three species oflemur. Points indicate the maximum likeli-hood estimate and lines indicate 95%confidence intervals (CIs) around those esti-mates. Black points/lines correspond tomodel-averaged estimates, with CIs calcu-lated using asymptotic assumptions. Theseassumptions may be unreliable for param-eter estimates at the boundary (e.g. theestimates of occupancy for Propithecusdiadema and Eulemur fulvus). Grey points/lines correspond to estimates from thesingle best-fitting model for each species,with CIs calculated from likelihood profiling.

Figure 3 Predictions of the precision ofestimates of occupancy that can beobtained for three lemur species, based onasymptotic assumptions. Three levels oftotal effort are shown, each representingthe total number of site-visits that are made(i.e. number of sites ¥ number of visits persite). SE, standard error.

A. Keane et al. Monitoring wild primate populations using occupancy modelling

Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London 5

time between repeat visits to each cell we surveyed. However,we also set out to minimize the risk that the presence oflemurs in an area could be affected by the history of previousvisits to a cell (e.g. lemurs that had previously encounteredthe researcher in the area might alter their behaviour byhiding or leaving the area, or conversely might approach theresearcher). Achieving this required longer periods to be leftbetween repeat visits to each area and researchers to be quietand careful during the surveys. An alternative approach tomitigating the problems of behavioural responses wouldhave been to model the effects of observers on the behaviourof the monitored species, thereby allowing surveys to becarried out over a shorter period of time.

In practice, the relatively small size of our sampling unitsand the time between repeat visits means that it is perhapsunlikely that the survey was closed to changes in occupancybetween repeat visits. Assuming that movement in and outof the sites is random, MacKenzie et al. (2006) suggestedthat the interpretation of the parameters y and P should bemodified so that they represent the proportion of the areaused and the probability that the species is using a siteduring the survey and detected by it. In this study, the lowestimates of detectability (P) we obtained for E. fulvus andP. diadema might reflect the effects of temporary emigrationrather than low probabilities of detection per se. Interpretedappropriately, the parameter estimates remain unbiased(MacKenzie et al., 2006), but may present practical prob-lems for monitoring because low apparent probabilities ofdetection necessitate a greater level of effort to produce

precise estimates of the population’s state. There has alsobeen little research to investigate how the parameters inter-preted in this way relate to underlying population processes.For example, we might expect a population decline to resultin lower estimates of both y (reflecting a reduction in theproportion of the study area that is now used by the species)and P (reflecting a reduction in the probability that a sitewill be occupied at any given time). However, it is likely thatthe extent to which changes in ‘use’ are positively correlatedwith changes in population size will depend on the corre-spondence between the scale of sampling units and the dis-tance over which monitored species can be expected to moveduring the sampling period: the greater the mismatchbetween these scales, the less likely it is that monitoring usewill provide useful information about critical populationprocesses (cf. MacKenzie & Royle, 2005). Ultimately, theform of these relationships and the extent to which useconstitutes a practical state variable for monitoring remainunclear and require further study.

The second trade-off we were forced to address concernedthe consistency of the survey method and the ease with whichit could be followed. In the absence of natural discrete sam-pling units, we adopted a relatively formal sampling design inwhich the survey teams followed a pre-defined route aroundselected cells on a square grid. Although it was more difficultto implement, this design helped to ensure that a similar levelof effort was devoted to searching for lemurs in each cell,reducing the potential for unmodelled heterogeneity in prob-ability of detection, which can lead to biased estimates ofoccupancy (Royle, 2006). A simple improvement in the effi-ciency of the design might have been achieved by stopping thesearch as soon as the target species had been detected (a‘removal design’; MacKenzie & Royle, 2005). In our study, ateam of three people could survey 1–3 grid squares in a day.The fieldwork presented here (90 site-visits) was completedby three teams in the field over 23 days (this incorporated 3full days of training and testing and 11 days, during whichsites were located and prepared, overlapping with 15 days inwhich surveys were conducted). By changing the researchprotocol to allow the teams to leave a square as soon as allthree species had been detected, the number of squares sur-veyed per day might have increased substantially. Care mustbe taken, however. ‘Removal designs’ of this sort can be lessrobust to violations of modelling assumptions than standardoccupancy designs (MacKenzie & Royle, 2005). Anotherpotential modification to the design would be to recordseparately sightings that occurred on each of the three ‘sec-tions’ walked within each sampled unit. Records from thesesecondary sampling periods are sufficiently close togetherthat closure may be reasonably assumed even for mobilespecies and these data would then allow movement in and outof the sampling units between repeat visits to be explicitlymodelled (see Rota et al., 2009). This modification wouldrequire very little additional effort, but care would be neededto ensure that records from the three sections are independ-ent of one another (i.e. that observing the species on onesection does not affect its probability of being observed onthe other sections).

Figure 4 Total amount and distribution of survey effort required toachieve an acceptable level of power (� 0.80) to detect a decline of0.3 in the occupancy of lemur populations (a = 0.05). The areashaded light grey indicates survey designs that produce acceptablepower for Indri indri (y = 0.6, P = 0.4), while the hatched area indi-cates survey designs that produce acceptable power for Propithecusdiadema/Eulemur fulvus (y = 0.95, P = 0.2).

Monitoring wild primate populations using occupancy modelling A. Keane et al.

6 Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London

For the three species included in this study, power analy-ses based on our estimates of occupancy and detectabilityindicated the level of effort that might be required for suc-cessful population monitoring and highlighted the chal-lenges of designing surveys for multiple species. Here, wecompared three species of lemur which differ in terms oftheir size and behaviour. For easily detectable species, suchas I. indri, our power analyses suggest that occupancy-basedmonitoring might be successfully implemented with a rea-sonable level of effort (~275 site-visits in total). A great dealof effort would be required for a monitoring scheme to beable to detect the same changes in the occupancy of E. fulvusand P. diadema. In general, it is likely that any primatemonitoring scheme will require a large investment of effortand careful design. The power analyses also suggest that theoptimal design for the two groups of species is very different(i.e. many sites each visited 5 times for I. indri, and far fewersites visited 14 times for E. fulvus and P. diadema). Moni-toring species with very different characteristics togetherinevitably requires more effort than a single survey opti-mized for a single species. In this case, however, a surveydesigned to strike an optimal balance for monitoring thethree species together would only require 11% more effortthan a survey designed specifically for P. diadema/E. fulvus,and 32% less effort than conducting two surveys separatelyfor P. diadema/E. fulvus and I. indri.

Our findings and experience suggest a number of avenuesfor future research. Further investigation is required to deter-mine whether the design of sampling units we adopted herewas the most appropriate. For example, surveys of specieswith large home ranges might require larger sampling units toproduce useful information. Despite the large volume ofliterature on sampling, there appears to be relatively littlepractical guidance relating the performance of differentchoices of sampling unit to the ecological characteristics ofthe target species. Similarly, the ideal choice of sampling unitwill also depend on the study’s aims. In this case, although wefound no detectable effect of mining on occupancy ordetectability for any of the three species, we do not interpretthis as strong evidence that mining does not affect lemurs.Rather, it seems likely that the effects of mining manifestthemselves at a larger scale in the landscape so that thepresence or absence of mining at the scales measured heremay bear little relation to its status as a threat. Furthermore,the small sample sizes used in this study may have limited thepower to detect the effects of mining. For certain species, fargreater efficiency might be achieved by monitoring using callsor other signs of presence as an alternative to visual cues.Indri indri, for example, produces loud vocalizations, whichcan be heard over a very wide area. In this case, occupancymodelling approaches based on point counts might allowmonitoring to be conducted relatively cheaply over muchlarger scales, although any such method would need to becarefully designed to account for possible variations in pat-terns of vocalization (e.g. in response to environmental con-ditions or disturbance; Geissmann & Mutschler, 2006).

In the face of continuing threats to wild primates world-wide (Mittermeier et al., 2009), the need for effective and

efficient monitoring of primate populations remains strong.Occupancy modelling has increasingly been recognized asa valuable tool for the study of many species in a varietyof contexts, but has not previously been applied to the studyof forest-dwelling primates. By exploring its potential,and highlighting key challenges that must be overcome, wehope that this study will encourage the development of apractical, targeted programme of research to establishthe strengths and weaknesses of alternative monitoringmethods for primate conservation.

AcknowledgementsWe are very grateful to two anonymous reviewers whosecomments greatly improved the manuscript. We thank theMinistry of the Environment and Forests in Antananarivoand Moramanga for the permission and support and ourcollaborators at the Department of Animal Biology, Uni-versity of Antananarivo. We are also grateful to Ra Razafi-manantsoa Germain, Randriarimalala Jimmy Rochelle,Rakotoarimanana Jean Doré, Randriamialisoa Céléstin,Randriambololona Doré, Rasamy Joseph Etienne, EmileRazanakoto and Luciennot Nirinaseheno Raharinjanaharyfor the assistance in the field. This work was funded by theUK-Government’s Darwin Initiative (grant 17-1127) andthe Rufford Foundation. We thank the Cambridge StudentConference of Conservation Science for funding T.H.’sinternship in the UK to work on this paper.

References

Akaike, H. (1974). A new look at the statistical model iden-tification. IEEE Trans. Autom. Control 19, 716–723.

Baker, L.R., Tanimola, A.A., Olubode, O.S. & Garshelis,D.L. (2009). Distribution and abundance of sacredmonkeys in Igboland, southern Nigeria. Am. J. Primatol.71, 574–586.

Buckland, S., Plumptre, A., Thomas, L. & Rexstad, E.(2010a). Design and analysis of line transect surveys forprimates. Int. J. Primatol. 31, 833–847.

Buckland, S., Plumptre, A., Thomas, L. & Rexstad, E.(2010b). Line transect sampling of primates: can animal-to-observer distance methods work? Int. J. Primatol. 31,485–499.

Burnham, K.P. & Anderson, D.R. (2002). Model selectionand multi-model inference. A practical information-theoretic approach. 2nd edn. New York: Springer.

Chapman, C.A. (1995). Primate seed dispersal: coevolutionand conservation implications. Evol. Anthropol. 4, 74–82.

Chapman, C.A., Lawes, M.J. & Eeley, H.A.C. (2006). Whathope for African primate diversity? Afr. J. Ecol. 44, 116–133.

Colishaw, G. & Dunbar, R. (2000). Primate conservationbiology. Chicago: University of Chicago Press.

Fa, J.E. & Brown, D. (2009). Impacts of hunting onmammals in African tropical moist forests: a review andsynthesis. Mammal Rev. 39, 231–264.

A. Keane et al. Monitoring wild primate populations using occupancy modelling

Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London 7

Fiske, I. & Chandler, R. (2011). Unmarked: an R packagefor fitting hierarchical models of wildlife occurrence andabundance. J. Stat. Soft. 43, 1–23.

Geissmann, T. & Mutschler, T. (2006). Diurnal distributionof loud calls in sympatric wild indris (Indri indri) andruffed lemurs (Varecia variegata): implications for callfunctions. Primates 47, 393–396.

Guillera-Arroita, G., Lahoz-Monfort, J.J., Milner-Gulland,E.J., Young, R.P. & Nicholson, E. (2010). Using occu-pancy as a state variable for monitoring the criticallyendangered Alaotran gentle lemur Hapalemur alaotrensis.Endanger. Species Res. 11, 157–166.

IUCN (2010). IUCN red list of threatened species. Version2010.4. URL: http://www.iucnredlist.org (downloaded on27 October 2010).

James, A.N., Gaston, K.J. & Balmford, A. (1999). Balanc-ing the Earth’s accounts. Nature 401, 323–324.

Jenkins, R.K.B., Keane, A., Rakotoarivelo, A.A.,Rakotomboavonjy, V., Randrianandrianina, F.H.,Razafimanahaka, H.J., Ralaiarimalala, S.R. & Jones,J.P.G. (2011). Analysis of patterns of bushmeat con-sumption reveals extensive exploitation of protectedspecies in eastern Madagascar. PLoS ONE 6, e27570.

Jones, J.P.G., Collen, B., Atkinson, G., Baxter, P.W.J.,Bubb, P., Illian, J.B., Katzner, T.E., Keane, A., Loh, J.,Mcdonald-Madden, E., Nicholson, E., Pereira, H.M.,Possingham, H.P., Pullin, A.S., Rodrigues, A.S.L., Ruiz-Gutierrez, V., Sommerville, M. & Milner-Gulland, E.J.(2011). The why, what, and how of global biodiversityindicators beyond the 2010. Conserv. Biol. 25,450–457.

Joseph, L.N., Field, S.A., Wilcox, C. & Possingham, H.P.(2006). Presence–absence versus abundance data formonitoring threatened species. Conserv. Biol. 20, 1679–1687.

Keane, A., Jones, J.P.G. & Milner-Gulland, E.J. (2011).Encounter data in resource management and ecology:pitfalls and possibilities. J. Appl. Ecol. 48, 1164–1173.

Köndgen, S., Kühl, H., N’Goran, P.K., Walsh, P.D.,Schenk, S., Ernst, N., Biek, R., Formenty, P., Mätz-Rensing, K., Schweiger, B., Junglen, S., Ellerbrok, H.,Nitsche, A., Briese, T., Lipkin, W.I., Pauli, G., Boesch,C. & Leendertz, F.H. (2008). Pandemic human virusescause decline of endangered great apes. Curr. Biol. 18,260–264.

Kümpel, N., Milner-Gulland, E., Rowcliffe, J. &Cowlishaw, G. (2008). Impact of gun-hunting on diurnalprimates in continental Equatorial Guinea. Int. J. Prima-tol. 29, 1065–1082.

MacKenzie, D. & Bailey, L. (2004). Assessing the fit of site-occupancy models. J. Agric. Biol. Environ. Stat. 9, 300–318.

MacKenzie, D.I. & Nichols, J.D. (2004). Occupancy as asurrogate for abundance estimation. Anim. Biodivers.Conserv. 27, 461–467.

MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege,S., Andrew Royle, J. & Langtimm, C.A. (2002). Estimat-ing site occupancy rates when detection probabilities areless than one. Ecology 83, 2248–2255.

MacKenzie, D.I. & Royle, J.A. (2005). Designing occu-pancy studies: general advice and allocating survey effort.J. Appl. Ecol. 42, 1105–1114.

MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H.,Bailey, L.L. & Hines, J.E. (2006). Occupancy estimationand modeling. Inferring patterns and dynamics of speciesoccurrence. London: Academic Press.

Martin, J., Chamaillé-Jammes, S., Nichols, J.D., Fritz, H.,Hines, J.E., Fonnesbeck, C.J., MacKenzie, D.I. & Bailey,L.L. (2010). Simultaneous modeling of habitat suitability,occupancy, and relative abundance: African elephants inZimbabwe. Ecol. Appl. 20, 1173–1182.

Mittermeier, R.A. (1988). Primate diversity and the tropicalforest. In Biodiversity: Chapter 16, 145–154. Wilson, E.O.& Peters, F.M. (Eds). Washington: National Academy ofSciences.

Mittermeier, R.A., Wallis, J., Rylands, A.B., Ganzhorn,J.U., Oates, J.F., Williamson, E.A., Palacios, E.,Heymann, E.W., Kierulff, M.C.M.,Yongcheng, L., Supri-atna, J., Roos, C., Walker, S., Cortés-Ortiz, L. & Schwit-zer, C. (2009). Primates in peril: the world’s 25 mostendangered primates 2008–2010. Primate Conserv. 24,1–57.

Morgan, D., Sanz, C., Onononga, J. & Strindberg, S.(2006). Ape abundance and habitat use in the GoualougoTriangle, Republic of Congo. Int. J. Primatol. 27, 147–179.

Nowak, R.M. (1999). Walker’s primates of the world. Balti-more: Johns Hopkins University Press.

Ogutu, J.O., Bhola, N., Piepho, H.P. & Reid, R. (2006).Efficiency of strip- and line-transect surveys ofAfrican savanna mammals. J. Zool. (Lond.) 269, 149–160.

Penman, T.D., Binns, D.L. & Kavanagh, R.P. (2009).Patch-occupancy modeling as a method for monitoringchanges in forest floristics: a case study in southeasternAustralia. Conserv. Biol. 23, 740–749.

Plumptre, A.J. (2000). Monitoring mammal populationswith line transect techniques in African forests. J. Appl.Ecol. 37, 356–368.

R Development Core Team. (2011). R: a language and envi-ronment for statistical computing. Vienna: R Foundationfor Statistical Computing.

Rota, C.T., Fletcher, R.J. Jr, Dorazio, R.M. & Betts, M.G.(2009). Occupancy estimation and the closure assump-tion. J. Appl. Ecol. 46, 1173–1181.

Rovero, F., Struhsaker, T., Marshall, A., Rinne, T.,Pedersen, U., Butynski, T., Ehardt, C. & Mtui, A. (2006).Abundance of diurnal primates in Mwanihana Forest,Udzungwa Mountains, Tanzania. Int. J. Primatol. 27,675–697.

Monitoring wild primate populations using occupancy modelling A. Keane et al.

8 Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London

Royle, J.A. (2006). Site occupancy models withheterogeneous detection probabilities. Biometrics 62,97–102.

Savage, A., Thomas, L., Leighty, K.A., Soto, L.H. &Medina, F.S. (2010). Novel survey method finds dramaticdecline of wild cotton-top tamarin population. Nat.Commun. 1, 30.

Schipper, J., Chanson, J.S., Chiozza, F., Cox, N.A., Hoff-mann, M., Katariya, V., Lamoreux, J., Rodrigues,A.S.L., Stuart, S.N., Temple, H.J., Baillie, J., Boitani, L.,Lacher, T.E., Mittermeier, R.A., Smith, A.T., Absolon,D., Aguiar, J.M., Amori, G., Bakkour, N., Baldi, R.,Berridge, R.J., Bielby, J., Black, P.A., Blanc, J.J.,Brooks, T.M., Burton, J.A., Butynski, T.M., Catullo, G.,Chapman, R., Cokeliss, Z., Collen, B., Conroy, J.,Cooke, J.G., da Fonseca, G.A.B., Derocher, A.E.,Dublin, H.T., Duckworth, J.W., Emmons, L., Emslie,R.H., Festa-Bianchet, M., Foster, M., Foster, S., Garshe-lis, D.L., Gates, C., Gimenez-Dixon, M., Gonzalez, S.,Gonzalez-Maya, J.F., Good, T.C., Hammerson, G.,Hammond, P.S., Happold, D., Happold, M., Hare, J.,Harris, R.B., Hawkins, C.E., Haywood, M., Heaney,L.R., Hedges, S., Helgen, K.M., Hilton-Taylor, C.,Hussain, S.A., Ishii, N., Jefferson, T.A., Jenkins, R.K.B.,Johnston, C.H., Keith, M., Kingdon, J., Knox, D.H.,Kovacs, K.M., Langhammer, P., Leus, K., Lewison, R.,Lichtenstein, G., Lowry, L.F., Macavoy, Z., Mace,G.M., Mallon, D.P., Masi, M., McKnight, M.W.,Medellín, R.A., Medici, P., Mills, G., Moehlman, P.D.,Molur, S., Mora, A., Nowell, K., Oates, J.F., Olech, W.,Oliver, W.R.L., Oprea, M., Patterson, B.D., Perrin,W.F., Polidoro, B.A., Pollock, C., Powel, A., Protas, Y.,Racey, P., Ragle, J., Ramani, P., Rathbun, G., Reeves,R.R., Reilly, S.B., Reynolds, J.E., Rondinini, C., Rosell-Ambal, R.G., Rulli, M., Rylands, A.B., Savini, S.,Schank, C.J., Sechrest, W., Self-Sullivan, C., Shoemaker,A., Sillero-Zubiri, C., De Silva, N., Smith, D.E., Srini-vasulu, C., Stephenson, P.J., van Strien, N., Talukdar,B.K., Taylor, B.L., Timmins, R., Tirira, D.G., Tognelli,M.F., Tsytsulina, K., Veiga, L.M., Vié, J.-C., William-son, E.A., Wyatt, S.A., Xie, Y. & Young, B.E. (2008).The status of the world’s land and marine mammals:diversity, threat, and knowledge. Science 322,225–230.

Sewell, D., Beebee, T.J.C. & Griffiths, R.A. (2010). Opti-mising biodiversity assessments by volunteers: the appli-cation of occupancy modelling to large-scale amphibiansurveys. Biol. Conserv. 143, 2102–2110.

Teelen, S. (2007). Primate abundance along five transectlines at Ngogo, Kibale National Park, Uganda. Am. J.Primatol. 69, 1030–1044.

Walsh, P.D. & White, L.J.T. (1999). What it will take tomonitor forest elephant populations. Conserv. Biol. 13,1194–1202.

Weller, T. (2008). Using occupancy estimation to assess theeffectiveness of a regional multiple-species conservationplan: bats in the Pacific Northwest. Biol. Conserv. 141,2279–2289.

Wilkie, D.S. & Carpenter, J. (1999). Can nature tourismhelp finance protected areas in the Congo Basin? Oryx33, 333–339.

Wrangham, R.W., Gittleman, J.L. & Chapman, C.A.(1993). Constraints on group size in primates and carni-vores: population density and day-range as assays ofexploitation competition. Behav. Ecol. Sociobiol. 32, 199–209.

Supporting informationAdditional Supporting Information may be found in theonline version of this paper:

Figure S1. Sensitivity analysis of the power of surveys todetect changes in occupancy of Indri indri (left column) orPropithecus diadema/Eulemur fulvus (right column) touncertainty in probability of detection (P) or occupancy (yor psi). The panels in the top row relate to a survey designedfor I. indri (number of visits per site, K = 5 and number ofsites, s = 55), the middle row relates to a survey designed forP. diadema/E. fulvus (K = 14, s = 32), and the bottom rowrelates to a survey which adopts the optimal compromise formonitoring the three species together (K = 11, s = 45).Power to detect a decline of 0.3 in occupancy is indicated onthe y axes, values of P are shown on the x axes, whiledifferent values of y are indicated by different colouredlines. Vertical dashed grey lines indicate the actual value ofP estimated from our study.Table S1. Summary of model selection for Indri indri. In themodel specifications, P is the probability of detection and yis occupancy. The notation (.) indicates that the componentwas modelled as a single parameter for all areas, while(mining) indicates that different parameter estimates wereobtained for areas with and without indications of miningactivity.Table S2. Summary of model selection for Propithecusdiadema. The notation (.) indicates that the component wasmodelled as a single parameter for all areas, while (mining)indicates that different parameter estimates were obtainedfor areas with and without indications of mining activity.Table S3. Summary of model selection for Eulemur rufus.The notation (.) indicates that the component was modelledas a single parameter for all areas, while (mining) indicatesthat different parameter estimates were obtained for areaswith and without indications of mining activity.

Please note: Wiley-Blackwell is not responsible for thecontent or functionality of any supporting materials suppliedby the authors. Any queries (other than missing material)should be directed to the corresponding author for the paper.

A. Keane et al. Monitoring wild primate populations using occupancy modelling

Animal Conservation •• (2012) ••–•• © 2012 The Authors. Animal Conservation © 2012 The Zoological Society of London 9