Using Occupancy-Based Camera-Trap Surveys to Assess the Critically Endangered Primate Macaca Nigra Across Its Range in North Sulawesi, Indonesia

Using occupancy-based camera-trap surveys to assess the Critically Endangered primate Macaca nigra across its range in North Sulawesi, Indonesia

C ASPIAN L. JOHNSON,HARRY H ILSER,MATTHEW L INKIE,RIVO R AHASIA F RANCESCO R OVERO,WULAN P USPARINI,IWAN H UNOWU A LFONS P ATANDUNG,NOVIAR A NDAYANI,JOHN T ASIRIN L UKITA A. NISTYANTARA and A NDREW E. BOWKETT

Abstract Primates are one of the most threatened groups of camera locations surveyed for  months (, camera days) mammals. Understanding their patterns of population oc- across three consecutive seasons is the effort required to de- currence and abundance, especially in response to threats, tect such change with % certainty. Our study underscores is critical for informing conservation action. The crested the importance of well-managed protected areas and intact black macaque Macaca nigra is the only Critically En- forests for the long-term survival of the crested black ma- dangered species of Sulawesi’s seven endemic macaques. caque, and tests the effectiveness of camera traps to monitor Little is known about its distribution or its response to primates at the landscape scale. deforestation and hunting. We conducted a camera-trap Keywords Camera trap, presence–absence, Macaca nigra, survey across the entire species range using an occupancy- occupancy, power analysis, primates, detectability, cost– based analytical approach to () establish the first range- benefit, modelling wide baseline of occurrence, () investigate how environmental and anthropogenic factors influence occurrence, Supplementary material for this article is available at () identify priority conservation subpopulations, and doi.org/./S () test the efficacy of the sampling and analytical protocol for temporal monitoring of M. nigra using occupancy as the state variable. From , camera-trap days, M. nigra Introduction was detected on  days at  of the  camera locations. Species occupancy was . and highest inside protected lobally, an estimated % of primate species are at areas and closed canopy forest. We identified eight distinct Grisk of extinction (i.e. categorized as threatened on the subpopulations, based on distribution and forest fragment IUCN Red List), with % of all primate species being in a size. To inform future monitoring, we used a power analysis state of decline (Estrada et al., ). The threats to primate to determine if our effort would allow us to detect inter- species are broad and wide-ranging, with many acting in syn- annual occupancy declines of %, and found that  ergy to hasten population declines, but hunting for trade and habitat loss from logging and agricultural expansion are the primary drivers (Cowlishaw & Dunbar, ; Estrada et al., CASPIAN L. JOHNSON* (Corresponding author, orcid.org/0000-0001-6990- ). To understand the effects of threats on primate popu- 7952) Selamatkan Yaki Foundation, Manado, Indonesia E-mail [email protected] lations, bespoke monitoring programmes are critical.

HARRY HILSER† Selamatkan Yaki Foundation, Manado, Indonesia Monitoring of primate populations provides forewarn- ing of declines and, by recognizing causal factors, a good MATTHEW LINKIE,WULAN PUSPARINI,IWAN HUNOWU,ALFONS PATANDUNG and NOVIAR ANDAYANI Wildlife Conservation Society, Bogor, Indonesia monitoring programme can identify mitigating solutions.

RIVO RAHASIA‡ Natural Resource Conservation Agency, North Sulawesi, However, monitoring primates using abundance metrics Indonesia can be resource intensive and time consuming at large   FRANCESCO ROVERO§ Tropical Biodiversity Section, Museo delle Scienze, spatial scales (Walsh & White, ; Cavada et al., ), Trento, Italy especially if the focal species are hunted and therefore wary JOHN TASIRIN University of Sam Ratulangi, Manado, Indonesia of humans, or live in remote and inaccessible locations

LUKITA A. NISTYANTARA BOGANI¶ Nani Wartabone National Park (Fashing & Cords, ). To address this, several recent

ANDREW E. BOWKETT Whitley Wildlife Conservation Trust, Paignton, UK studies have recommended using occupancy as the state *Also at: Bristol Zoological Society, Bristol, UK variable, rather than abundance, as a more cost-effective †Also at: Department of Geography, Exeter University, Exeter, UK option (Sanje mangaby Cercocebus sanjei,Roveroetal., ‡Also at: Selamatkan Yaki Foundation, Manado, Indonesia ;indriIndri indri,Keaneetal.,; Alaotra reed lemur §Also at: Department of Biology, University of Florence, Florence, Italy  ¶Also at: Natural Resource Conservation Agency, North Sulawesi, Indonesia Hapalemur alaotrensis,Guillera-Arroitaetal., ). Received  February . Revision requested  April . Occupancy modelling corrects for detectability issues by Accepted  July . First published online  September . using presence/absence (detection/non-detection) data to

Oryx, 2020, 54(6), 784–793 © The Author(s), 2020. Published by Cambridge University Press on behalf of Fauna & Flora International doi:10.1017/S0030605319000851 Downloaded from https://www.cambridge.org/core. IP address: 170.106.35.93, on 01 Oct 2021 at 02:11:50, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0030605319000851 Macaca nigra 785

estimate the proportion of sites occupied by the species of M. nigra (O’Brien & Kinnaird, ; Rosenbaum et al., interest (Mackenzie et al., ), and as such is considerably ; Palacios et al., ) and recent landcover maps less labour intensive and less costly to implement than (KLHK, ) to determine the extent of potential habitat count-based metrics (Royle & Nichols, ; Joseph et al., that would be ecologically capable of supporting a macaque ). To this end, camera traps produce data that are suited population and use this to define our study area (Fig. ). to occupancy analysis and therefore present a viable survey These included sites characterized by forest cover and/or method, especially for elusive primates living in landscapes scrub habitats. North Sulawesi has altitudes of –, m with difficult terrain. Here, we use an occupancy based and has a wet season during October–May and a dry season camera-trap approach to conduct a baseline assessment during June–September. for the crested black macaque Macaca nigra, a hard-to- detect and mostly terrestrial species of forest-dwelling primate endemic to the northernmost peninsula of the Methods Indonesian island of Sulawesi. The crested black macaque exemplifies the plight of Data collection: camera trapping many forest-dwelling primates. Resource extraction and We first divided the potential habitat into sampling units land requirements for agriculture in Sulawesi have resulted (termed sites) of  ×  km (n =  sites). This size was cho- in the loss of most of its lowland forest habitat (Margono sen as it is close to the largest recorded M. nigra home range et al., ), creating a fragmented habitat mosaic that is  of . km (O’Brien Kinnaird, ), but larger than the probably restricting gene flow between subpopulations  purported average of . km (Riley, ). A sample of  (Evans et al., ). Added to this are retaliatory killings of these sites was then selected to be surveyed using camera in response to foraging in crops, and hunting for bush- traps ( Reconyx HC, Holmen, USA;  Cuddeback meat consumption, which in turn makes the species shy Black Flash, DePere, USA; one Bushnell Natureview HD and secretive in large parts of its range. Live View, Overland Park, USA). This was done in two Numerous presence-only studies have explored the dis- phases. Firstly, during January –October ,wesur- tribution and status of M. nigra (Mackinnon & Mackinnon, veyed  sites. This included all  sites across the Tang- ; Sugardjito et al., ;Lee,; Rosenbaum et al., koko Nature Reserve, and  sites selected with random ; Melfi et al., ;Palaciosetal.,; Kyes et al., ) systematic sampling within the larger region of the southern and have estimated the population size to be ,–, Bogani Nani Wartabone National Park landscape (Fig. ). individuals (Riley, ). Sharp declines of up to %over Secondly, during March–July , we randomly selected   years in M. nigra populations have indicated the serious- sites from across patches of potentially suitable habitat be- ness of the threats facing the species (Supriatna & Andayani, tween these protected areas. Although implemented in dif- ). Macaca nigra is now categorized as Critically En- ferent years because of the number of cameras available, we dangered on the IUCN Red List (Supriatna & Andayani, conducted the actual survey phases within a relatively short ) and is one of the top  most threatened primates period of time, minimizing the risk of violating the assump- (Schwitzer et al., ). Although previous work has high- tion that the population was closed. lighted the vulnerability of M. nigra, the limited survey efforts Because of logistical constraints, camera traps were not yielded incomplete knowledge on distribution, status and re- uniformly deployed within each site. Rather, we deployed sponses to anthropogenic threats at a regional scale. In this cameras in the part of the site most accessible, provided it study, we use camera-trap data to () investigate the influence was .  m from the site’s border. Our grid-based approach of environmental and anthropogenic factors on M. nigra ensured sufficient spacing between camera traps. We posi- occupancy across its native range () provide an empirical tioned cameras along wildlife trails, attaching them to tree baseline of its status, using occupancy as the state variable trunks at a height of c.  cm, with the sensor aimed parallel of interest, and () evaluate the efficacy of our protocol to to the ground facing the monitoring area. On average each monitor primates at the landscape scale. camera was deployed for a minimum of  months. Four locations, however, produced no data as the cameras were either stolen or malfunctioned; we therefore present data Study area collected from  camera traps. Macaca nigra is native to North Sulawesi province, where its range extends from the northern tip and some of the small Data collection: covariates surrounding islands (Lembeh, Manadotua and Talise) to the Onggak-Dumoga River and the south-east landscape Occupancy of sites by M. nigra was expected to vary across of Bogani Nani Wartabone (Johnson et al., ). Within North Sulawesi according to habitat and anthropogenic this area, we used knowledge of the habitat preferences of factors (Rosenbaum et al., ). We therefore derived a

FIG. 1 Camera-trap placements across North Sulawesi province in relation to protected and unprotected areas and the main habitat types. The line represents the boundary between Macaca nigra and Macaca nigrescens (Johnson et al., ).

spatial dataset of covariates that could potentially explain Chandler, )inR .. (R Development Core Team, any heterogeneity in M. nigra occupancy. These were pres- ). Single-season occupancy models estimate both the ence of forest cover (Yes/No), elevation, slope, normalized probability that a site is occupied (ψ) and the probability difference vegetation index (NDVI), distance to roads and that the species is detected if it is present (p) using a max- settlements, protected area (Yes/No), level of anthropogenic imum likelihood approach (Mackenzie et al., ). All nu- disturbance in the landscape (measured by the Human merical covariates were first standardized into z-scores and Footprint Index; Venter et al., ), and distance from assessed for collinearity using Pearson’s rank correlation. the edge of continuous forest (Table ). All covariates were As no covariates were found to exhibit collinearity . ., extracted for the exact camera location, except for NDVI, no covariates were excluded in any of the candidate models slope and elevation. As a measure of landscape rough- (Supplementary Table ). ness, and therefore accessibility, we calculated the mean of We next examined the effect of covariates (Table )on slope and elevation across each site. We calculated NDVI our parameter of interest, ψ. However, we expected that a using red and near infrared spectral bands in Landsat  number of covariates that influence occupancy of M. nigra imagery (Hansen et al., ), averaged across each site. may also affect the abundance of M. nigra and therefore All spatial datasets for covariates were calculated and de- detectability (Royle & Nichols, ). Additionally, there rived in ArcGIS . (Esri, Redlands, USA). are other factors that could also influence the detectability. We therefore began our model selection process by ex- plicitly accounting for p. Following recommendations of  Data analyses: occupancy modelling MacKenzie et al. ( ), we identified a suitable covariate structure for p whilst holding the covariate structure for We discretized our sampling data, following Rovero & ψ constant. Covariate structure for p was assessed using Spitale (), into  consecutive -day sampling occasions the Akaike information criterion corrected for small sample (of ,  and  days, a -day duration best facilitated model sizes (AICc). We then fixed the covariate structure for p with convergence). For each site and each occasion, a  indicated the covariate structure that had the lowest AICc (Burnham detection (a photograph) and  indicated non-detection (no & Anderson, ) and proceeded to assess the role of our photograph) of M. nigra. covariates on ψ, ranking the competing models according to To assess the influence of ecological and anthropogenic their AICc values. If multiple models were within  ΔAICc factors on occupancy, we fitted single-season occupancy points, they were considered equally supported (Burnham models to the data using the package unmarked (Fiske & & Anderson, ) and estimates were instead derived by

TABLE 1 Definition and predicted effect of covariates used to model variation in detectability and occupancy of Macaca nigra across sites.

Covariate Definition Type Expected effect (rationale) Forest Whether camera was located within closed canopy Categorical Positive (as M. nigra is predominantly forest- forest or bush/scrub habitat (source: KLHK, 2015) dwelling, occupancy/detectability will be higher in forest than in bush/scrub) NDVI Normalized Difference Vegetation Index, averaged Continuous Positive (high NDVI values should be associated across grid cell (Hansen et al., 2013) with greater food availability) Protected Whether camera was located within a protected Categorical Positive (protected areas should have lower levels of Area area boundary (KLHK, 2015) disturbance, with less hunting & forests more intact) Roads Euclidean distance (km) from camera location to Continuous Positive (as distance from the nearest road increases, nearest paved road a site typically becomes less accessible & therefore less likely to be subject to disturbance) Village Euclidean distance (km) from camera location to Continuous Positive (sites in close proximity to a village may nearest settlement/village be more frequently visited by people & hunting pressure may be higher) Edge Euclidean distance (km) from camera location to Continuous Positive (as distance from the forest edge increases, the forest edge (negative for cameras in non- the site becomes more difficult to access, & ex- forested habitat) ploitation & disturbance may therefore be lower) Elevation Mean elevation (m) across entire site (calculated Continuous Positive (increasing elevation & steeper slope angles from digital elevation model; NASA/METI/AIST/ contribute to landscape roughness; rougher land- Japan Spacesystems, 2009) scapes will be harder for people to access & disturbance will be lower) Slope Mean slope angle (degrees) across entire site Continuous Positive (see elevation) Human Expressed as a standardized index for (1) built Continuous Negative (higher values indicate higher intensity of Footprint environments, (2) crop land, (3) pasture land, human influence) Index (4) population density, (5) night-time lights, (6) railways, (7) roads (Venter et al., 2016) Camera Make of camera trap Categorical Unknown p (as different cameras could potentially Model vary in how well they detect animals, this is a potential source of heterogeneity in detectability)

averaging coefficients from across these multiple models populations, it enables easy demarcation of distinct forest using the MuMIn package (Barton, )inR. We used a blocks that may be exposed to different threats. parametric bootstrap to check the adequacy of our model  fit using the χ approach on the most saturated model (Burnham & Anderson, ). Data analyses: optimal study design and power analysis Once the best (or averaged) occupancy model was identified, we used the corresponding occupancy probability to es- We evaluated the optimal effort for an occupancy-based timate the proportion of the  sites likely to be occupied by camera-trap monitoring protocol for M. nigra by first  M. nigra. We then mapped predicted occupancy probability of estimating the number of sites (s) and occasions (K, -day M. nigra across its native range using the approach suggested sampling periods) required to achieve a desired level of pre- ψ by Rovero & Spitale (). This approach involved fitting our cision for the and p estimators. These were calculated best model to a data frame containing a grid of covariate using estimates from the best model and the corresponding  values at the scale of  ×  km across the range of M. nigra. asymptotic variance of the occupancy indicator (Equation ;  Maps of estimated occupancy were created in ArcMap . Mackenzie & Royle, ): (Esri, Redlands, USA). TS = s × K Finally, we used predicted occupancy to identify regions K × c (1 − p∗) that may be important to the species based on the presence = (1 − c) + (1) cˆ ∗− × × − K−1 of characteristics associated with a high occupancy probabil- var( ) p K p (1 p)  ity. We classified these as spatially distinct areas with .  km where TS is total effort, s is the number of sites, K is the of continuous forest cover and predicted occupancy . .. number of occasions, ψ the probability of occupancy, p We defined spatially distinct as being separated by .  m the probability of detection provided the site is occupied, of habitat not capable of supporting the species. Although and p* the probability that the species is detected at least separation of forest patches as we define it here does not ne- once after K occasions. As the number of occasions can be cessarily preclude movement of M. nigra individuals between increased free of additional cost in a camera-trap survey

TABLE 2 Model selection results for covariate effects in determining occupancy probability of M. nigra across its native range in North Sulawesi. The top ranked models are shown as those with ΔAIC c ,  followed by the null and average model. All models depicted (ΔAICc , ) were included in the averaged model. 1 2 3 Δ 4 5 ψ6 ± 7 ψ ± 8 ± Model NPar AICc AICc Wi / SE Cumulative Wi SE/p SE ψ(Forest, PA) p(PA, Forest, HFI) 7 1666.5 0.00 0.28 0.28 0.663 ± 0.076 ψ(Forest) p(PA, Forest, HFI) 6 1666.6 0.09 0.27 0.54 0.662 ± 0.062 ψ(Forest + NDVI) p(PA + Forest + HFI) 7 1668.0 1.54 0.13 0.67 0.661 ± 0.073 ψ(Forest + Edge) p(PA + Forest + HFI) 7 1668.2 1.70 0.12 0.79 0.662 ± 0.072 ψ(Forest, PA, NDVI) p(PA, Forest, HFI) 8 1668.3 1.75 0.12 0.91 0.661 ± 0.085 Null model p(.)ψ(.) 1695.8 30.4 0.655 ± 0.046 0.296 ± 0.013 Average model 0.662 ± 0.081 0.280 ± 0.025

 PA, protected area; HFI, Human Footprint Index.  Number of parameters.  Akaike information criterion corrected for small sample size.  Relative difference in AICc values compared to top-ranked model.  AICc model weight.  Estimate of occupancy ± SE.  Cumulative AICc model weight.  Estimate of detection probability ± SE.

(at least until a camera needs to be serviced, in this case after TABLE 3 Summary of conditional model averaged parameters based c.  months, or K = ) we defined the optimum monitoring on the best-supported models identified in Table . Estimates of effort for M. nigra as one that achieves the target standard the β coefficient are reported for standardized covariates (scaled    error of . in our estimates whilst minimizing s.Wedid to mean = and unit variance of ). See Table for the models with covariates for both ψ and p. this by resolving the Equation for s and assuming a duration of  months for the surveys. Parameter1 Estimate ± SE z value P(.|z|) As smaller absolute declines in occupancy become more p(PA: inside) 0.518 ± 0.130 3.99 , 0.001* detectable as the number of seasons of monitoring increases p(Forest: inside) 0.357 ± 0.188 1.90 0.057 (Beaudrot et al., ), we calculated the number of seasons p(HFI) −0.177 ± 0.068 2.60 0.009* needed to detect a % annual occupancy decline if the op- ψ(Forest: inside) 1.149 ± 0.528 2.18 0.030* ψ ± timal monitoring effort (calculated by Equation ) is chosen (PA: inside) 0.677 0.452 1.50 0.134 ψ ± as the long-term monitoring protocol (survey effort = s × K). (NDVI) 0.200 0.232 0.86 0.390 ψ(Edge) 0.192 ± 0.295 0.65 0.516 To do this, we simulated data for an annual % decline  across a time series. We used the estimates from our best PA, protected area; HFI, Human Footprint Index. model to determine the initial input parameters for occu- *Significant. pancy and detection and set colonization to zero, so as to simulate a decline. We then took the generated data and fragments, distributed across North Sulawesi, indicating a fitted it to a dynamic occupancy model without covariates greater distribution than previously known. (Mackenzie et al., ). These were then assessed to deter- Investigating covariate influence on detection probabil- mine the number of seasons required to detect an annual ity, whilst holding occupancy constant, revealed p as a posi- decline of % with % confidence. Our methods follow tive function of being inside both a protected area and for- those detailed by Beaudrot et al. () and all simulations est, and a negative function of the Human Footprint Index and analyses were done in R, using code available on Github (Supplementary Table , Supplementary Fig. ). This model (Ahumada, ). gave a detection probability of . ± SE . and was sub- sequently used to explore which combination of covariates best explained M. nigra occupancy. Results From the different combinations of covariates, the most parsimonious occupancy model also included protected  Detection and occupancy area and forest (Table ). M. nigra occupancy was higher in forests (β coefficient . ± SE .) and inside protected From , camera-trap days, M. nigra was detected on  areas (. ± SE .). However, five models were ranked separate days in  of  sites, yielding a naïve occupancy ,  ΔAICc units and thus considered equally supported estimate of .. These  camera locations confirmed (Table , Fig. ). We therefore estimated detectability and the presence of the species in  spatially distinct forest occupancy by averaging across these models, resulting in

FIG. 2 The functional relationships of predicted occupancy, with standard error, of M. nigra, with covariates included, in the best-fit models (Table ): (a) protective area status, (b) forest cover (bush/scrub or closed canopy forest), (c) Euclidean distance from forest edge (negative values for cameras in non-forested habitat), and (d) Normalized Difference Vegetation Index (NDVI).

p = . and ψ = . (Table ). By taking this estimate needed by increasing the number of repeats, without addition- for occupancy and extrapolating it to all sites across the al cost, and thereby reducing overall survey costs. Under this  landscape (n =  or  km ), we estimated the area option, if the number of repeats is increased to  ( days,  of occupancy of M. nigra to be , km . Finally, our or c.  months), the number of camera-trap sites could be  goodness-of-fit test (χ = ,P=.) indicated no reduced to  (Fig. b). As the cheapest means of surveying evidence for a lack of fit for the most saturated model, M. nigra with suitable precision, this is the protocol that we suggesting that the more parsimonious models provide an advocate. Simulating the power provided by this protocol, if adequate description of the data. According to our criteria applied across multiple seasons,  years would be sufficient of predicted occupancy . . and continuous forest cover to detect a % decline in occupancy with % certainty.  .  km , we identified eight distinct regions that probably contain important subpopulations (Fig. , Supplementary Discussion Table , Supplementary Fig. ). Our study demonstrates an extension in the known range Optimal study design and power analysis of M. nigra compared to previous data. We show that camera-trap data coupled with occupancy analysis can Considering the estimates from our averaged model and provide a robust baseline assessment of a predominantly assuming constant probabilities, an optimal survey design ground-dwelling forest primate of conservation concern.  would have  camera-trap sites, each with  repeats (. We estimated an occupancy of . (, km ) for M. months when each repeat constitutes a duration of  days) nigra, which represents the first baseline for future re- to achieve a target standard error of . (Fig. a). However, gional monitoring of the species. Estimated occupancy for camera traps further repeats are easily gained, without was only slightly higher than the naïve occupancy (.), incurring further cost, by simply delaying the retrieval of the which is probably an artefact of a long survey duration, cameras. Therefore, it is possible to reduce the number of sites which resulted in a high cumulative detection probability,

FIG. 3 Predicted occupancy across the mainland native range of M. nigra. Eight potentially important landscapes, defined as spatially distinct continuous forest  blocks .  km with predicted occupancy . ., are identified by numbers (Supplementary Fig. , Supplementary Table ).

thereby reducing its influence on occupancy. Detection loss within protected area; Hansen et al., ), a factor probability, however, varied across sites, and was higher in that probably implies reduced hunting and disturbance. closed canopy forests and in areas with a lower Human However, protected areas currently cover only %ofthe Footprint Index. We attribute this to a preference by M. area estimated to be occupied by the species. Considering nigra for forests and a sensitivity to human disturbance, that M. nigra, like many other primates (Estrada et al., which cause a reduced abundance of the species in non- ), is vulnerable to anthropogenic activities, the species forested and disturbed habitats, with less abundant species remains at risk across % of its range. typically being less detectable (Royle & Nichols, ). Our modelling approach predicted eight spatially dis- However, this could also be a consequence of M. nigra tinct regions that are likely to support important sub- being more elusive and cautious in more distur-bed areas. populations of M. nigra (Fig. , Supplementary Table , Although occupancy was significantly higher in closed Supplementary Fig. ). Five of these landscapes repre- canopy forest than in bush/scrub habitats, we were not sent the first scientific record of the species’ presence able to test for effects of forest type (pristine vs degraded/ (Supplementary Table ). In combination with evidence logged) because of the limited resolution of available satel- of range expansion (Johnson et al., ), this finding is of lite imagery. We used NDVI as a remotely sensed proxy for conservation relevance and contradicts speculations that vegetation structure and green biomass (Myneni et al., ) Tangkoko contains the last viable population (Supriatna & but, although retained in the best models, it was not signifi- Andayani, ; Palacios et al., ;Kyesetal.,; cant. This is probably the result of the overriding import- Engelhardt et al., ). Although Tangkoko does appear to ance of forest presence over the quality of that forest in hold an important and viable population (Engelhardt et al., determining occupancy. The dependence of M. nigra on ) it comprises just % of the total area occupied by the forest is nevertheless apparent, highlighting the species’ species (Supplementary Table ). Therefore, although we susceptibility to forest conversion to other land uses. acknowledge that population density and viability of sub- Along with a preference for forest cover, estimated occu- populations will vary across the species’ range, the signifi- pancy was higher within protected areas, which is probably cance of these additional areas for the species cannot be a result of both the continuous extent of optimal habitat ignored. and lower anthropogenic disturbance. Deforestation ob- Our discovery of previously undocumented popula-  served within protected areas during – was much tions,inadditiontotheincreasedrangesizeof km lower than outside (.% outside protected area vs .% recently reported (Johnson et al., ), results in an

In addition to contributing valuable information on the distribution and status of the species, our study facilitates the design of a targeted monitoring protocol, which we define as  sites surveyed with camera traps for  days ( × -day occasions) per season. Testing the power of this design under simulation, we found it would be possible to detect an annual change in occupancy of % (a decline we consider sufficiently severe to initiate a conservation intervention), with % certainty if implemented over  seasons. By demonstrat- ing the robustness of our suggested protocol in meeting the objectives of such a monitoring programme, researchers can be confident in investing resources in its long-term imple- mentation (Yoccoz et al., ;Legg&Nagy,). Meaningful monitoring studies, especially at large spatial scales, are considerable investments of limited conservation resources. Procurement of the cameras required for a study such as this constitutes a high cost, which must be justified given the availability of other presence/absence survey methods such as line transect surveys. We were able to use data collected during this study to conduct a basic feasibility comparison between the two methods and found that the cost efficiency of camera traps was far greater than that of a more labour intensive transect method, even after considering the cost of cameras (Sup- plementary Table ). As such, although initial costs are high, camera traps may provide a cheaper alternative for monitoring certain species. FIG. 4 Total survey effort required (sites × repeats) (a), and total In conclusion, our study provides a robust baseline for number of unique camera sites required (b) to achieve a given precision in the occupancy estimate as a function of the number M. nigra occupancy and, through simulation, indicates of repeats required per camera. Repeats are represented by how the monitoring of a generally elusive and Critically months; a single repetition has a duration of  days and  month Endangered primate through camera trapping could be is therefore six repeats. Curves were created based on the feasible at a landscape level. We emphasize the necessity averaged model (p = . and ψ = .) and on the asymptotic of such a monitoring approach for understanding popula- properties of the maximum likelihood estimates. The curves tion responses to ecological and anthropogenic factors and correspond to different standard errors in the occupancy hence informing conservation efforts. The baseline we estimate. report forms the basis on which temporal trends in M. nigra occupancy could be detected, and the study  area of occupancy of , km and an extent of occur- design we suggest provides a robust means of detecting  rence of , km . The Critically Endangered status of potential declines within a suitable timeframe. In demon- M. nigra was a result of suspected severe population de- strating the potential of camera-trap data analysed in an clines. As we only provide a baseline, using new survey occupancy framework to monitor M. nigra,weaddtoa methodology, we can neither support nor refute this. growing body of literature that promotes occupancy as a fea- However, with a greater range than previously thought, sible alternative to abundance-based metrics for monitoring our findings may influence the threatened status of the wild primates (Guillera-Arroita et al., ; Keane et al., species and therefore have implications for any future ; Gerber et al., ; Rovero et al., ). review of the species’ Red List status. With this in mind, we caution that the species remains highly threatened, Acknowledgements We thank the Ministry of Science and particularly outside protected areas, and emphasize the Technology for permission to conduct this research in Indonesia; need for standardized monitoring that follows the proto- Bogani Nani Wartabone National Park management authority for col we recommend. This will identify trends in the popu- research permission; BKSDA North Sulawesi, in particular Pak lation and lead to a clearer picture of the species’ status. Yakub Ambagau, for support of our research; our field assistants Nicodemus Malir, Hanisa Silayar, Fandy Pongantung, Adha Yakseb, In the meantime, we stress the need for increased conser- Omega Judistira, Idham Pohi, Muhamad Ihwan, Sukrisno and vation management in the landscapes identified as im- Luciano Gawina; Graeme Gillespie for providing advice during the portant (Supplementary Table ). conception of this project; and two anonymous reviewers for their

useful comments. Funding was provided by the Global Environ- GUILLERA-ARROITA, G., LAHOZ-MONFORT, J.J., MILNER-GULLAND, mental Facility, as administered through the project Enhancing the E.J., YOUNG, R.P. & NICHOLSON,E.() Using occupancy as a Protected Area System of Sulawesi, Fondation Segré and Wildlife state variable for monitoring the Critically Endangered Alaotran Reserves Singapore Conservation Fund. gentle lemur Hapalemur alaotrensis. Endangered Species Research, , –. Author contributions Study design and fieldwork: CJ, RR, LN, HH, HANSEN, M.C., POTAPOV, P.V., MOORE, R., HANCHER, M., AP, WP, IH; data analysis: CJ, FR; writing: CJ, AB, ML, HH, FR. TURUBANOVA, S.A., TYUKAVINA,A.&TOWNSHEND, J.R.G. () High-resolution global maps of st-century forest cover change.  – Conflicts of interest None. Science, , . JOHNSON, C.L., HILSER, H., ANDAYANI, N., HUNOWU, I., LINKIE, M.,  Ethical standards This research abided by the Oryx guidelines on PATANDUNG,A.&BOWKETT, A.E. ( ) Camera traps clarify ethical standards. the distribution boundary between the crested black Macaque (Macaca nigra) and Gorontalo macaque (Macaca nigrescens) in North Sulawesi. International Journal of Primatology, , References –. JOSEPH, L.N., FIELD, S.A., WILCOX,C.&POSSINGHAM, H.P. () AHUMADA, J.A. () PAOccPaper. github.com/jaahumadap/ Presence–absence versus abundance data for monitoring threatened PAOccPaper/tree/master/ShinyApps/powerPaper [accessed species. Conservation Biology, , –.  November ]. KEANE, A., HOBINJATOVO, T., RAZAFIMANAHAKA, H.J., JENKINS, BARTON,K.() MuMIn: multi-model inference. R package version R.K.B. & JONES, J.P.G. () The potential of occupancy modelling . cran.r-project.org/web/packages/MuMIn/index.html [accessed as a tool for monitoring wild primate populations. Animal January ]. Conservation, , –. BEAUDROT, L., AHUMADA, J., O’BRIEN, T.G. & JANSEN, P.A. () KLHK () Kementrian Lingkunan Hidup dan Kehutanan. Detecting tropical wildlife declines through camera-trap geoportal.menlhk.go.id/arcgis/home [accessed  January ]. monitoring: an evaluation of the Tropical Ecology Assessment and KYES, R.C., ISKANDAR, E., ONIBALA, J., PAPUTUNGAN, U., LAATUNG, Monitoring protocol. Oryx, , –. S. & HUETTMANN,F.() Long-term population survey of the BURNHAM, K.P. & ANDERSON, D.R. () Model Selection and Sulawesi black macaques (Macaca nigra) at Tangkoko Nature Multimodel Inference: A Practical Information-Theoretic Approach. Reserve, North Sulawesi, Indonesia. American Journal of Springer-Verlag, New York, USA. Primatology, , –. CAVADA, N., CIOLLI, M., BARELLI,C.&ROVERO,F.() Optimizing LEE,R.() The impact of hunting and habitat disturbance on the field and analytical procedures for estimating densities of arboreal population dynamics and behavioural ecology of the crested black and threatened primates in tropical rainforest. American Journal of macaque (Macaca nigra). PhD thesis. University of Oregon, Eugene, Primatology, , . USA. CLAYTON,L.&MILNER-GULLAND, E.J. () The trade in wildlife in LEGG, C.J. & NAGY,L.() Why most conservation monitoring is, North Sulawesi, Indonesia. In Hunting for Sustainability in Tropical but need not be, a waste of time. Journal of Environmental Forests (eds J. Robinson & E. Bennett), pp. –. Columbia Management, , –. University Press, New York, USA. MACKENZIE, D.I. & ROYLE, J.A. () Designing occupancy studies: COWLISHAW,G.&DUNBAR,R.() Primate Conservation Biology. general advice and allocating survey effort. Journal of Applied University of Chicago Press, Chicago, USA. Ecology, , –. ENGELHARDT, A., MUNIZ, L., PERWITASARI-FARAJALLAH,D.& MACKENZIE, D.I., NICHOLS, J.D., LACHMAN, G., DROEGE, S., ROYLE, WIDDIG,A.() Highly polymorphic microsatellite markers for A. & LANGTIMM,C.() Estimating site occupancy rates the assessment of male reproductive skew and genetic variation in when detection probabilities are less than one. Ecology, , Critically Endangered crested macaques (Macaca nigra). –. International Journal of Primatology, , –. MACKENZIE, D.I., NICHOLS, J.D., HINES, J.E., KNUTSON, M.G. & ESTRADA, A., GARBER, P.A., RYLANDS, A.B., ROOS, C., FRANKLIN, A.B. () Estimating site occupancy, colonization, and FERNANDEZ-DUQUE, E., DI FIORE,A.&LI,B.() Impending local extinction when a species is detected imperfectly. Ecological extinction crisis of the world’s primates: why primates matter. Society of America, , –. Science Advances, ,e. MACKENZIE, D.I., NICHOLS, J.D., ROYLE, J.A., POLLOCK, K.H., EVANS, B.J., SUPRIATNA, J., ANDAYANI,N.&MELNICK, D.J. () BAILEY, L.L. & HINES, J.E. () Occupancy Estimation and Diversification of Sulawesi macaque monkeys: decoupled evolution of Modeling. Inferring Patterns and Dynamics of Species Occurrence. mitochondrial and autosomal DNA. Evolution, , –. Academic Press, Cambridge, USA. FASHING, P.J. & CORDS,M.() Diurnal primate densities and MACKINNON,J.&MACKINNON,K.() Cagar Alam Gn. biomass in the Kakamega forest: an evaluation of census methods Tangkoko-Dua Saudara, Sulawesi Utara Management Plan, and a comparison with other forests. American Journal of –. Unpublished report. WWF, Bogor, Indonesia. Primatology, , –. MARGONO, B., POTAPOV, P., TURUBANOVA, S., STOLLE,F.&HANSEN, FISKE,I.&CHANDLER,R.() Unmarked:anR package for fitting M. () Primary forest cover loss in Indonesia over –. hierarchical models of wildlife occurrence and abundance. Journal Nature Climate Change, , –. of Statistical Software, , . MELFI, V., TASIRIN, J., JONAS, A., YOSEP, A., SABINTOE, B., GERBER, B.D., WILLIAMS, P.J. & BAILEY, L.L. () Primates and HOUSSAYE,F.&KAMBEY,R.() Impact of Disturbance of cameras: noninvasive sampling to make population-level inferences Wild Macaca nigra Populations and Habitats in Sulawesi. while accounting for imperfect detection. International Journal of Unpublished report to LIPI, Whitley Wildlife Conservation Fund, Primatology, , –. Paignton, UK.

MYNENI, R.B., HALL, F.G., SELLERS, P.J. & MARSHAK, A.L. () The ROVERO, F., MARTIN, E., ROSA, M., AHUMADA, J.A. & SPITALE,D. interpretation of spectral vegetation indexes. IEEE Transactions on () Estimating species richness and modelling habitat Geoscience and Remote Sensing, , –. preferences of tropical forest mammals from camera trap data. NASA/METI/AIST/JAPAN SPACESYSTEMS () ASTER Global PLOS ONE, ,e. Digital Elevation Model, NASA EOSDIS Land Processes DAAC. doi. ROVERO, F., MTUI, A., KITEGILE, A., JACOB, P., ARALDI,A.&TENAN, org/./ASTER/ASTGTM. [accessed  February ]. S. () Primates decline rapidly in unprotected forests: evidence O’BRIEN,T.&KINNAIRD, M.F. () Behavior, diet, and movements from a monitoring program with data constraints. PLOS ONE, of the Sulawesi crested black macaque (Macaca nigra). International ,e. Journal of Primatology, , –. ROYLE, A.J. & NICHOLS,J.D.() Estimating abundance from O’BRIEN, T.G. & KINNAIRD, M.F. () Differential vulnerability of repeated presence–absence data or point counts. Ecology, , –. large birds and mammals to hunting in North Sulawesi, Indonesia, SCHWITZER, C., MITTERMEIER, R., RYLANDS, A., CHIOZZ,F., and the outlook for the future. In Hunting for Sustainability in WILLIAMSON, E., MACFIE,E.&COTTON,A.() Primates in Tropical Forests (eds J. Robinson & E. Bennett), pp. –. Peril: The World’s  Most Endangered Primates –. Columbia University Press, New York, USA. Conservation International, Arlington, USA. PALACIOS, J.F.G., ENGELHARDT, A., AGIL, M., HODGES, K., BOGIA, SUGARDJITO, J., SOUTHWICK, C.H., SUPRIATNA, J., KOHLHAAS, A., R. & WALTERT,M.() Status of, and conservation BAKER, S.C., ERWIN,J.&LERCHE,N.() Population survey of recommendations for, the Critically Endangered crested black macaques in northern Sulawesi. American Journal of Primatology, macaque Macaca nigra in Tangkoko, Indonesia. Oryx, , –. , –. RDEVELOPMENT CORE TEAM () R: a language and environment SUPRIATNA,J.&ANDAYANI,N.() Macaca nigra. In IUCN Red for statistical computing. Version ... R Foundation for Statistical List of Threatened Species : e.TA. dx.doi.org/. Computing, Vienna, Austria. /IUCN.UK..RLTS.TA.en [accessed  January RILEY, E.P. () The endemic seven: four decades of research on the ]. Sulawesi macaques. Evolutionary Anthropology, , –. VENTER, O., SANDERSON, E.W., MAGRACH, A., ALLAN, J.R., BEHER, J., ROSENBAUM, B., O’BRIEN, T.G., KINNAIRD,M.&SUPRIATNA,J. JONES, K.R. & WATSON, J.E.M. () Sixteen years of change in the () Population densities of Sulawesi crested black macaques global terrestrial human footprint and implications for biodiversity (Macaca nigra) on Bacan and Sulawesi, Indonesia: effects of conservation. Nature Communications, , . habitat disturbance and hunting. American Journal of Primatology, WALSH, P.D. & WHITE, L.J.T. () What it will take to monitor , –. forest elephant populations. Conservation Biology, , –. ROVERO,F.&SPITALE,D.() Species-level occupancy analysis. YOCCOZ, N.G., NICHOLS, J.D. & BOULINIER,T.() Monitoring In Camera Trapping for Wildlife Research (eds F. Rovero & of biological diversity in space and time. Trends in Ecology and F. Zimmermann), pp. –. Pelagic Publishing, Exeter, UK. Evolution, , –.