Habitat determinants of golden‐headed lion tamarin (Leontopithecus chrysomelas) occupancy of cacao agroforests: Gloomy conservation prospects for management intensification

Organismal distributions in human‐modified landscapes largely depend on the capacity of any given species to adapt to changes in habitat structure and quality. The golden‐headed lion tamarin (GHLT; Leontopithecus chrysomelas) is an Endangered primate from the Brazilian Atlantic Forest whose remaining populations occupy heterogeneous landscapes consisting primarily of shade cacao (Theobroma cacao) agroforestry, locally known as cabrucas. This cash crop can coexist with high densities of native tree species and holds a significant proportion of the native fauna, but its widely extolled wildlife‐friendly status is increasingly threatened by management intensification. Although this potentially threatens to reduce the distribution of GHLTs, the main determinants of tamarin's occupancy of cabrucas remain unknown, thereby limiting our ability to design and implement appropriate conservation practices. We surveyed 16 cabruca patches in southern Bahia, Brazil, and used occupancy modeling to identify the best predictors of GHLT patch occupancy. Key explanatory variables included vegetation structure, critical resources, landscape context, human disturbance, and predation pressure. We found a negative relationship between GHLT occupancy and the prevalence of jackfruit trees (Artocarpus heterophylus), which is likely associated with the low representation of other key food species for GHLTs. Conversely, cabrucas retaining large‐diameter canopy trees have a higher probability of GHLT occupancy, likely because these trees provide preferred sleeping sites. Thus, key large tree resources (food and shelter) are currently the main drivers of GHLT occupancy within cabruca agroecosystems. Since both factors can be directly affected by crop management practices, intensification of cabrucas may induce significant habitat impacts on GHLT populations over much of their remaining range‐wide distribution.

trees have a higher probability of GHLT occupancy, likely because these trees provide preferred sleeping sites. Thus, key large tree resources (food and shelter) are currently the main drivers of GHLT occupancy within cabruca agroecosystems.
Since both factors can be directly affected by crop management practices, intensification of cabrucas may induce significant habitat impacts on GHLT populations over much of their remaining range-wide distribution.

| INTRODUCTION
As a result of burgeoning human demands on Earth's natural resources, human-modified landscapes have expanded relentlessly, particularly in species-rich biomes (Watson et al., 2016). The longterm persistence of many taxa is therefore strongly contingent on their capacity to survive in such novel habitat mosaics (Tabarelli, Peres, & Melo, 2012). Species capacity to deal flexibly with habitat change-such as those related to microclimate, vertical stratification, and availability and quality of food resources-will determine their ability to occupy and persist in many anthropogenic habitats (Purvis, Gittleman, Cowlishaw, & Mace, 2000).
A pantropical meta-analysis has shown that primate assemblages in human-modified habitats, such as those resultant of logging and agriculture practices, can present declines of 17-43% in biodiversity metrics (e.g., abundance, density, and species richness), with more detrimental effects when the forest is converted to agricultural land (Almeida-Rocha, Peres, & Oliveira, 2017). However, agro-mosaics and agroforests can support or subsidize populations of many primate species due to the more heterogeneous nature of vegetation at these sites (Almeida-Rocha et al., 2017;Estrada, Raboy, & Oliveira, 2012), representing a viable "win-win" solution in reconciling human economic demands with biodiversity conservation (Perfecto & Vandermeer, 2008). The capacity of these agro-systems to retain native forest biodiversity depends on the amount of residual forest cover in the landscape and the type and intensity of management practices (Cassano, Barlow, & Pardini, 2014;Steffan-Dewenter et al., 2007).

Deforestation of the Atlantic Forest of Bahia, northeastern
Brazil, has been so severe that only~11% of the original forest remains (SOS Mata Atlântica & Instituto Nacional de Pesquisas Espaciais, 2018). Much of forest conversion occurred due to cacao (Theobroma cacao) cultivation that began in 1,746 and expanded rapidly in the late 19th century, particularly in southern Bahia (Piasentin & Saito, 2014). Since the cacao expansion period, much of the original forest of southern Bahia has been converted to shadecacao agroforestry systems, locally known as cabrucas (Piasentin & Saito, 2014). Traditional cabrucas are established by replacing the native forest understory with cacao trees that grow underneath the canopy of predominantly native tree species that are retained for their shade, in addition to trees that were subsequently either planted or regenerated (Alves, 1990). Consequently, the vertical structure of cabrucas is very simplified compared with intact forests, but much more complex than monoculture systems, such as "suncacao" and other annual or perennial crops (Alves, 1990). According to the Executive Committee of the Cacao Cropland Plan (CEPLAC), cabrucas should retain between 25 and 30 shade trees/ha (Mandarino, 1981), but traditional cabrucas of this region usually retain a much higher tree density: an average of 197 (70-480) shade trees/ha of which~63% (18-100%) are native species (Schroth et al., 2015).
These cabrucas maintain a vegetation complexity that enables a significant proportion of native fauna to use them as habitat, supplementary resources, and/or dispersal corridors between forest patches (Faria, Paciencia, Dixo, Laps, & Baumgarten, 2007).
Due to their compatibility with both biodiversity conservation and forest carbon storage (Schroth et al., 2011(Schroth et al., , 2015, cabrucas are considered a wildlife-friendly production system. Unfortunately, this status is now threatened by land-use intensification (Schroth et al., 2011).
Former Brazilian environmental legislation banned native tree felling within cabrucas (Federal Decree no 6.660 of 21 November 2008, chapter VIII, article 28), but the State Decree no. 15.180 (Chapter 2, Section IV) published by the Bahia Government in 2014 (hereafter referred to as the management decree) sanctioned the legal removal of shade trees in high-density cabrucas to increase cacao yields, allowing landowners to retain a minimum native tree density of 40 stems/ha. This tree density threshold is far below that observed in traditional cabrucas of southern Bahia (Schroth et al., 2015), and will undoubtedly make this agroecosystem far more structurally simplified than it has been until now (Figure 1), potentially diminishing their wildlife-friendly status (Cassano et al., 2014;Schroth et al., 2015) and negatively affecting many endangered species.
The golden-headed lion tamarin (GHLT, Leontopithecus chrysomelas) is an endangered small-bodied (~620 g) callitrichid primate endemic to the Brazilian Atlantic Forest whose geographic range has been severely reduced by deforestation and is currently dominated by cabrucas . The most recent assessment of vegetation cover within the GHLT range indicates that approximately 60% is currently covered by cabrucas, with greater dominance in the eastern range-the region containing the most viable populations (Zeigler, Fagan, DeFries, & Raboy, 2010)-where cabrucas represent about 47% (± 33% SD) of the landscape . The GHLT diet consists mainly of ripe fruits, arthropods and small vertebrates (Rylands, 1989), and in cabrucas is largely comprised of the exotic jackfruit (Artocarpus heterophylus), which is widely available almost all year-round (Oliveira, Neves, Raboy, & Dietz, 2011). Other key resources for this species are bromeliads, the main microhabitat used for arthropod foraging (Rylands, 1989). In cabrucas, GHLTs typically occur at a mean density of 0.12 (0.04-0.21) ind./ha  live in groups of 2-15 individuals, usually with one breeding female (Baker, Bales, & Dietz, 2002). All group members sleep together preferentially in large tree cavities (Rylands, 1989), which may be a constraint on group size. GHLTs usually repeat the use of individual trees in cabrucas more than in forests (Oliveira, Hankerson, Dietz, & Raboy, 2010), which may increase predation risk since some predators can learn the location of the sleeping sites (Franklin, Hankerson, Baker, & Dietz, 2007). Also, GHLTs usually prefer the lower levels of the vertical strata in forests, but they increase the use of the upper levels in cabrucas-probably due to the distribution of food resources and travel routes-being even more exposed to aerial predators (Almeida-Rocha, De Vleeschouwer, Reis, Grelle, & Oliveira, 2015).
Despite their ability to use cabrucas , GHLTs do not occupy all cabruca patches within its range . Therefore, identifying cabruca features that favor GHLT occupancy is critical to effectively advocate for management practices that will best protect this species and maintain the wildlifefriendly status of cabrucas, particularly given the current policy context that encourages widespread management intensification.
Here, we investigate which habitat and landscape characteristics facilitate the occupation of cabrucas by GHLTs, as well as the role of natural and domesticated predators in this process. GHLTs experience a higher predation risk in cabrucas-mainly from raptorscompared to relatively undisturbed forests .
Domestic dogs (Canis familiaris) are highly abundant at cabrucas, and attacks on GHLTs have been reported . Since habitat alteration can lead to unbalanced trophic interactions (Irwin, Raharison, & Wright, 2009), predation pressure may exert a strong influence on GHLT cabruca occupancy.
We, therefore, expected to find that patch occupancy is positively related to (a) vegetation structural complexity (e.g., density and height of shade trees, vertical stratification, canopy connectivity and abundance of lianas); (b) availability of key trophic resources (e.g., key tree species for feeding, sleeping, and foraging); and (c) total amount of vegetation cover at the local and landscape scales. In contrast, patch occupancy should be negatively related to (d) predation risk (i.e., an abundance of potential predators); and (e) management intensification of shade-cacao plantations (i.e. high frequency of weeding, high density of cacao trees, and low shade cover).

Our research adhered to the American Society of Primatologists'
Principles for Ethical Treatment of Primates. Since we did not use any invasive field technique, it was not necessary to obtain approval from any Brazilian committee for this study.

| Study area
From May 2014 to May 2015, we surveyed 16 cabruca sites located within the GHLT geographic range, covering 12 municipal counties (encompassing an area of~4,000 km 2 ) of southern Bahia, Brazil ( Figure 2). The cabruca sites were at least 11 km apart (mean F I G U R E 1 Profiles of the vertical structure and diversity of trees in (a) mature (old-growth) forests, (b) secondary forests, (c) traditional cabrucas, and (d) intensified cabrucas. Bars on the right of the panels indicate mean canopy height. Illustration by Gastón Giné ALMEIDA-ROCHA ET AL. | 3 of 15 distance: 52 km; range: 11-114 km), ensuring an appropriate level of spatial independence. The study region is characterized by a high level of deforestation and fragmentation, especially in the western portion of species range where the dominant vegetation is tropical seasonal semi-deciduous forest . The eastern portion retains the largest and most intact forest fragments, with the coastal evergreen tropical rainforest as the dominant vegetation type . The mean annual temperature and rainfall are 24°C and 2,500 mm, respectively, with no marked seasonality (Mori, Boom, de Carvalho, & dos Santos, 1983).

| Golden-headed lion tamarin survey
All sampling was performed by Almeida-Rocha JM with the help of a field assistant. Playback was used to systematically search for GHLTs in each study area over three nonconsecutive days. The number of visits was defined a priori based on the GHLT detection history of a previous study developed within cabrucas using the same techniques (L. G. Neves [personal communication, November, 2013]). Visits to the same cabruca site were separated by at least 1 week to avoid animals habituating to playbacks (Dong & Clayton, 2009), but all surveys within the same site were completed within 30 days.
Using Landsat images from Google Earth (Google, 2016), we deployed a pre-selected sampling grid within each cabruca site, so playbacks could be performed at the intersection points of this grid ( Figure 3). The methodology consisted of playing an adult male GHLT long-call-which in this genus typically attracts neighboring groups before territorial encounters (Peres, 1989)-to stimulate intraspecific responses by attracting counter-calls. To do this, we used a Sony ICD-PX470 digital voice recorder and a portable Anchor Audio AN-MINI Speaker (frequency response: 100 Hz-15 kHz ± 3 dB).
F I G U R E 2 Natural vegetation cover within the geographic range (the dark gray region on the map of Brazil) of the golden-headed lion tamarin (Leontopithecus chrysomelas), and the location of the 16 cabruca sites surveyed in this study (red circles). Classification of vegetation cover is based on Landau, Hirsch, and Musinsky (2008) At each playback point, a complete long-call was directed towards the four cardinal points, holding the speaker~2 m above ground, followed by a 5-min on-site wait-and-listen interval. When responses were detected, we recorded their location, time, direction, and the number of vocalizations. These parameters helped us to assess whether more than one group responded to the playbacks. To reduce the chance of detecting the same group more than once on the same day, playback points were spaced 200 m apart to prevent overlap in their auditory range (~100 m), which had been previously tested experimentally using a radio-collared GHLT group from another study. Previous studies used the same distance (Kierulff & Rylands, 2003;Raboy et al., 2010).
Each sampling grid had at most 15 playback points (equivalent to a sampling area of 60 ha, considering the playback range) to enable sampling of all points in the morning (06:00 hr-11:00 hr) when GHLTs are most active in cabrucas (Reis, 2012). Total sampling effort amounted to 612 playbacks (24-48 per site). At each visit, we started from a different playback point to increase detection probability by considering any possible variation in the use of space by the groups throughout the day. We used a thermo-hygrometer to record mean air temperature and humidity levels during each visit so we could model the effect of these parameters on species detectability (Waser & Waser, 1977). Playback surveys were never performed during rainy weather or strongly windy conditions.

| Predator surveys
Based on points of occurrence and distribution maps, we identified 15 mammalian carnivores and 46 diurnal raptor species that can occur in the study region (Tables S1 and S2). We classified them as potential predators of GHLTs based on (a) records of predation on primates, (b) records of attacks on primates, (c) body mass, (d) typical prey size, (e) records of mammals in the diet, (f) degree of dietary specialization, and (g) foraging strategy. To make this classification more systematic, each criterion received a categorical value, with high values attributed to characters that favor GHLT predation (Table S3). These values were then summed to create a Potential Predation Index (PPI) which was used to rank all species according to their probability of preying on GHLTs, attributing greater weights to categories (a), (b), and (g), which were considered most important.
Details on this classification are presented in Tables S1-S3.
F I G U R E 3 Schematic representation of the study design showing the sampling grid (delimited by the transects represented here as dashed gray lines), the survey area (delimited by a 100-m radius of playback hearing range), and the surrounding area (delimited by a 1-km radius from the boundaries of the survey area) of a cabruca site. The sampling grid shows the location of the playback points for both golden-headed lion tamarins (black circles) and diurnal raptors (blue circles), as well as the cameratrapping stations (red circles), and the vegetation plots (green circles). Active searches for diurnal raptors were conducted in the external transects (spaced by 400 m), and landscape metrics were calculated within the surrounding area ALMEIDA-ROCHA ET AL.

| Nonaerial predators
We sampled nonaerial predators simultaneously with GHLT surveys using four to six digital camera-traps (Tigrinus® 6.0D) per site, depending on the size of the sampling grid. Camera-trap stations were spaced at least 300 m apart and positioned near playback points ( Figure 3). At each station, one camera was fixed to a tree at~50 cm above the ground and baited with a banana lure (10 ml), carnivore essence (Bobcat urine; 10 ml) and sardine oil (10 ml), specifically selected to attract potential GHLT predators such as felids and mustelids (Schlexer, 2008). Baits and lures were placed separately into perforated pots protected from rain and animal consumption that were attached to wooden sticks at~50 cm above the ground and 2 m perpendicular to the camera. In all cabruca sites, the cameras were operated simultaneously for 24 hr during consecutive days for an overall total sampling effort of 128 ± 28 camera-trap/days per site (which is following the recommendations of Espartosa, Pinotti, & Pardini, 2011).
Camera-trap stations were checked weekly to replace baits, lures, batteries, memory cards, or the cameras themselves in case of occasional malfunction. Malfunctioning cameras were replaced and kept operating longer to compensate for any losses in sampling effort. Photographs of conspecifics recorded within a 24-hr period were considered as a single record (i.e., the same individual), unless individual recognition was possible through natural marks, as in the case of domestic dogs. In the case of social species, such as coatis (Nasua nasua), we used "group" rather than "individual" records.
Finally, we used camera-trap records to estimate the total abundance of nonaerial predators at each cabruca site.

| Aerial predators
A combination of active search, playback, and point count was used to survey for diurnal raptors, which were sampled during two nonconsecutive days at each site. These surveys were carried out after GHLT surveys were completed to avoid interference in the behavior and detectability of the tamarins. Sampling was carried out between 06:00 and 12:00 hr, the peak period of activity for most diurnal raptors (Mañosa, Mateos, & Pedrocchi, 2003;Thiollay, 1989), avoiding rainy and windy days (Granzinolli & Motta-junior, 2008).
Between 06:00 and 09:00 hr, sampling was carried out within cabruca sites, focusing on forest species that only occasionally fly above the canopy (Thiollay, 1989), but also searching for soaring species that commence flight activity later. An active search was carried out throughout the sampling grid using Yukon Futurus Pro We performed targeted playbacks to detect the presence of a set of diurnal raptors known to respond to calls (Zorzin, 2011;JABM [pers. obs., November, 2012]) using a modified version of the methods proposed by Granzinolli and Motta-junior (2008). These species included gray-headed kite (Leptodon cayanensis), barred forest-falcon (Micrastur ruficollis), collared forest-falcon (Micrastur semitorquatus), bicolored Hawk (Accipiter bicolor), and black hawk-eagle (Spizaetus tyrannus). We used recordings from Wiki Aves (http://www.wikiaves. com/), preferentially selecting those from the study region, and avoiding aggressive vocalizations and duets. At each playback point, recordings of all focal species were played in a pre-established order considering both body size and aggressive behavior, since largerbodied species could repel smaller raptors. Thus, we played the vocalizations of the smallest and least aggressive raptor first. Each vocalization was played continuously for 3 min, holding the speaker at~2 m above the ground and rotating it 360°at a constant rate, followed by a 3-min on-site wait.
Most raptors start soaring when thermals are well-formed, so the best period to perform point count techniques is between 09:00 and 12:00 hr (Mañosa et al., 2003;Thiollay, 1989). During this period, we recorded all individuals using visual or vocal cues from a fixed point located on hilltops adjacent to the study area (Mañosa et al., 2003). At six of the 16 sites where the relief was very flat, we performed two complementary point counts located at~100 m from the edge of the cabruca, separated by mean distances of 690 ± 170 m.
We split our efforts between these two points so that we remained at each point for 1 hr 30 min.
Except for single point counts, we changed the location of the initial sampling point in the second visit to ensure the detection of species with different activity peaks at all points (Jones, 2000).
Given that even small raptors occupy home ranges of up to 100 ha (Thiollay, 1989), repeated detections of the same species in the same site were attributed to the same individual, unless more than one individual was observed simultaneously. Based on these records, we estimated the total abundance of aerial predators at each cabruca site. Overall sampling effort amounted to 64 playback points (4 per site), 91 hr 11 min of active searches (4 hr 48 min-6 hr 45 min per site), and 96 hr of point counts (6 hr per site). and frequency of all shade tree species recorded at each site, according to Curtis and McIntosh (1951). Then, based on checklists of tree species used by GHLTs for food and shelter (Cardoso, 2008;Catenacci, De Vleeschouwer, & Nogueira-Filho, 2009;Catenacci, Pessoa, Nogueira-Filho, & De Vleeschouwer, 2016;Oliveira et al., 2010, we identified the key tree species recorded in our study sites (Table S5). Finally, we summed the IVI values of these key tree species per site. We also used the jackfruit trees IVI separately for the analysis due to its particular importance in the diet of GHLTs at cabruca sites (Oliveira et al., 2010). Further details on this index and all other habitat variables are presented in the Supplementary Material (Table S4).

| Landscape context
Based on Landsat 8 images from 2016 (30 m resolution) provided by Google Earth (Google, 2016) and using the raster package (Hijmans et al., 2016) in R 3.3.1 (R Core Team, 2019), we measured the minimum linear distance between each playback point to the nearest household and fragment edge. We calculated the percentage of vegetation cover in the surroundings of each cabruca site by defining a 1-km radius buffer from the survey area' boundaries ( Figure 3) and extracting all visually identified clear-cut areas from this region using the Quantum GIS 2.18.2 (http://www.qgis.org/). The spectral difficulty of accurately distinguishing cabrucas from forests using satellite images did not pose a problem for our analysis because we were primarily interested in quantifying the total amount of available habitat for GHLTs, and cabruca is habitat for this species .

| Occupancy modeling
Occupancy estimate (Ψ) represents the proportion of an area that is occupied by a given species (Mackenzie et al., 2002). Using Mark 8.x software (White & Burnham, 1999), we fitted single-season occupancy models-which assume that the population is closed to changes in occupancy inside each sampling unit during the survey season-to estimate GHLTs occupancy in cabrucas, and modeled the detection probability (p) as imperfect, considering that GHLTs may be present in an area but may not always be detected. Before analyses, we assessed the pairwise correlations between all variables collected in the field and extracted from satellite images through a Spearman correlation test, using the Stats package in R (R Core Team, 2019). For each pair, we excluded one highly correlated variable (r ≥ .6), always keeping the variables that enabled us to test all hypotheses. In this way, we removed shade tree species richness, bromeliads, canopy height, and connectivity from the analyses. We Since we were primarily interested in determining the most important covariates that influenced Ψ and p, we built a model set based on all possible additive covariate combinations (Doherty, White, & Burnham, 2012), which resulted in 2,517 competing models.
We then calculated the cumulative AIC c weight (w + ) for each covariate to interpret their relative importance on the estimates (Burnham & Anderson, 2002). The final estimates of Ψ and p were model-averaged considering the weighted mean among all competing models (Burnham & Anderson, 2002). We assessed the fit of the most general model (i.e. the model with the greatest number of parameters) by estimating the overdispersion parameter c-hat through 10,000 bootstrap samples (Mackenzie & Bailey, 2004) in the PRESENCE 11.7 software (http://www.mbr-pwrc.usgs.gov/software/ presence.shtml).

| GHLT and predator surveys
We obtained 31 GHLT responses at 29 playback points within eight sites (50%). At two additional sites, we did not obtain playback responses, but we occasionally detected GHLTs while walking between playback points. The playbacks also elicited responses from diurnal raptors, including some potential predators such as the Southern Caracara (Caracara plancus), the Crane Hawk (Geranospiza caerulescens), the Mantled Hawk (Pseudastur polionotus), the Roadside Hawk (Rupornis magnirostris), and the Black Hawk-eagle.
A total of 10 native mammal species plus domestic dogs, domesticated livestock, and humans were recorded at cabruca sites using camera traps. Such records included three potential predators assigned to low to moderate GHLT predation probabilities (Table 1): domestic dogs (PPI = 11), recorded in 15 cabruca sites (94%); coatis (PPI = 9), recorded in three cabruca sites (19%); and tayras (Eira barbara; PPI = 11), recorded in two cabruca sites (13%). Except for domestic dogs, all potential predators were recorded within low activity cabruca plots near regenerating forest patches. Capuchin monkeys (PPI = 11) were occasionally recorded at one site while moving between playback points (Table 1).
We recorded at least 18 species of diurnal raptors, including 15 potential predators with varying GHLT predation probabilities ( Table 2). The most common of these were assigned to medium to high potential to prey on GHLTs (

| Habitat structure and quality, and landscape context
Cabruca sites have a mean density of 623 ± 182 cacao trees/ha and 182 ± 60 shade trees/ha, with a mean shade tree diameter of 37.2 ± 30.7 cm, median canopy height of 15.6 ± 2.6 m, and mean shade cover of 73% ± 10% (Table S6). A total of 79 shade tree species were identified (15 ± 5 species/site), 46 of which are used by GHLTs for food and/or shelter (Table S5).
The IVI of those key resource tree species, and particularly of jackfruit trees, ranged between 15-78% and 0-19%, respectively (Table S6). Direct or indirect signs of hunting (traps, hunters, and/or shotgun blows) and logging (chainsaw noise and stumps) were recorded at 10 and 9 of the 16 sites, respectively (Table S6). Vegetation cover in the surroundings ranged from 73% to 96% between cabruca sites (Table S6).

| GHLT occupancy
Our goodness-of-fit test revealed no evidence of overdispersion χ 2 = 15.17; P = 0.29; c-hat = 1.09). The most parsimonious model explaining GHLT occupancy had a low AIC c weight (w + = 0.15), suggesting a high degree of uncertainty as to which is the best model (Table 3)

| DISCUSSION
We investigated the determinants of GHLT occupancy within cabrucas of southern Bahia, Brazil, focusing on the specific influence of vegetation structure, habitat quality, agroforestry management intensity, landscape context, and predation pressure. The two features that most affected GHLT occupancy-the preponderance of jackfruit trees and the diameter of shade trees-are related to the availability of key resources (food and shelter) and both may be directly affected by the intensification of management practices.
We found a negative relationship between the IVI of jackfruit trees and GHLT occupancy. At first, this may seem counterintuitive since jackfruits represent one of the most important food resources for GHLTs within cabrucas (Oliveira et al., 2010). However, as the IVI index was derived from key resource trees, high values of jackfruit trees IVI imply a prevalence of this species over other key food species-such as Myrtaceae trees, which are largely used as food resources by GHLTs in cabrucas (Oliveira et al., 2010)-resulting in lower availability of complementary resources. This suggests that, when sites become highly dominated by jackfruit trees, they may fail to provide enough complementary resources to satisfy the metabolic and nutritional requirements of GHLTs. This situation may occur whenever farmers favor cacao shading by fast-growing tree species with dense crownsas is the case of jackfruit trees and Erythrina spp.-rather than maintaining a diversified native tree composition in which old-growth species are more common (Rolim & Chiarello, 2004).
We found that cabrucas retaining wide-diameter shade trees were more likely to contain GHLTs, which is probably related to the availability of suitable sleeping sites . In addition to boosting GHLT occupancy, retaining larger trees also contributes to climate change mitigation, since trees larger than 35 cm in diameter account for a disproportionate fraction of the carbon storage within cabruca systems (Schroth et al., 2015).
The tree diameter profile of cabrucas will vary depending on which species are used to shade the cacao understory and on the frequency with which the understory is weeded (Sambuichi & Haridasan, 2007), which can therefore largely determine the extent to which cabrucas are wildlife-friendly. Since natural regeneration of most shade trees is suppressed through weeding (Rolim & Chiarello, 2004), cabrucas are composed of an unstable land-sharing system in which the longterm persistence of GHLTs, as well as many other vertebrate species, depends heavily on replanting key resource species. A possible way to improve the conservation value of cabrucas under production intensification is to prioritize the retention/replanting of larger diameter tree species that have already been identified as important for the regional fauna (Oliveira et al., 2010), ensuring a diversified and balanced tree species composition.
The probable reason why we failed to detect significant effects of presumably important habitat features, such as shade tree density and canopy connectivity, is that our study sites did not span critical thresholds for such features. This is not a failure of our study design but the reality of traditional cabrucas. We can reasonably expect these features to become more important if legally sanctioned management intensification is implemented. If intensification is unavoidable, we strongly recommend the monitoring of cabruca plots both before and after intensification, so that we can understand how intensification will impact GHLT populations (and other species from the regional fauna) and design mitigation strategies accordingly.
Similarly, we failed to detect a significant influence of the amount of vegetation cover within surrounding cabruca landscapes probably because the range of values estimated for our study sites was high.
Our approach considered cabrucas in estimates of vegetation cover because this agroecosystem is a key habitat for GHLTs , so these high values reflect the regional context at the T A B L E 1 Camera trapping records (number of individuals/groups) of mammalian species at 16 cabruca sites, and species classification according to their capacity to prey on GHLTs (y = yes, n =no) based on their Potential Predation Index (PPI)
time of this study. We, therefore, highlight that our results apply to landscapes containing high amounts of habitat availability for GHLTs.
However, landscape-scale habitat amount would likely become more important for GHLTs should this study be repeated in highly deforested landscapes.
Contrary to our expectations, predators apparently do not play a decisive role in GHLT occupancy of cabrucas currently. The few records we obtained of wild nonaerial predators were restricted to low activity cabruca plots near forest patches, suggesting that these species may be transient in cabrucas. An alternative explanation for this low detectability may be the elevated hunting pressure throughout this region (Cassano, Barlow, & Pardini, 2012).
Although we did not detect a direct effect of raptors on GHLT occupancy, it is important to consider that the predator-prey relationship could be quite different in highly intensified cabrucas. If cabrucas become even more simplified, that is with a lower density of shade trees and canopy connectivity, GHLTs will become more exposed to predators due to even more reduced canopy connectivity, lower midstory foliage density, and reduced availability of natural shelters. Besides, a decline in food resources could lead to longer travel distances to key food trees thereby exposing GHLTs to greater predation risk (Garber & Bicca-Marques, 2002). Accordingly, monitoring efforts of GHLT groups are required in highly intensified cabrucas to investigate these possible outcomes.

| CONSERVATION IMPLICATIONS
Considering the landscape composition within the current geographic range of GHLTs, one can easily conclude that a GHLT conservation plan that fails to explicitly consider cabrucas is likely doomed to failure. Unfortunately, the role of cabrucas in assisting GHLT conservation can be threatened by imminent prospects of widespread management intensification, as attempts to increase cacao yields typically involve thinning of shade canopy trees, which is likely to affect the main predictors of GHLT occupancy in this habitat.
However, some studies have shown that it is possible to combine high crop yields with high biodiversity levels in cacao agroforests (Clough et al., 2011). In southern Bahian cabrucas, shade cacao crop yields can increase two-fold compared to the regional average productivity by simply adjusting appropriate levels of mineral fertilizers and restricting overall canopy shading to 55%, without necessarily reducing shade tree density (Schroth et al., 2014).
Under current policies in which cabruca management intensification is incentivized, it has been proposed that exotic shade species, such as jackfruits, should be preferentially removed instead of native species, as exotic species are expected to have a lower ecological value for the native fauna (Schroth et al., 2014).
Considering previous findings    (Canale et al., 2016;. Currently, there is no official mechanism to regulate the thinning rate of exotic species from cabrucas, which is necessary as the removal of jackfruit trees can substantially reduce habitat carrying capacity for the attendant fauna (Gosper & Vivian-Smith, 2009). The impact of removing exotic species from intensified cabrucas, that is those containing a lowdensity of shade trees, can be even higher than in traditional cabrucas such as those in this study.
Finally, managing cacao farm landholdings to facilitate coexistence with GHLTs and other native wildlife is not just good conservation practice, it can also accrue additional economic benefits.
GHLTs play a critical role as natural seed dispersers, yet they do not raid cacao fruits nor damage cacao trees, thereby contributing to the regeneration and maintenance of traditional cabrucas (Catenacci et al., 2009). Also, the GHLT is a flagship species in the Atlantic Forest of southern Bahia, which can attract tourists to those cabrucas where they still occur. Although this tourism potential remains largely unexplored in this region, some producers are already using the public image of GHLTs in their commercial logos or exploring them as a focal species for ecotourism ventures. Primate watching can be both a profitable economic activity and a successful conservation strategy whenever benign tourism practices are adopted (Macfie & Williamson, 2010;Russon & Wallis, 2014). For example, the Lion F I G U R E 4 Probability of detecting golden-headed lion tamarins (Leontopithecus chrysomelas) at each playback point within cabrucas (p) as a function of the density of cacao trees (represented as the sum of density values per plot for each cabruca site). The dotted line and the color-coded area represent the estimates and the 95% confidence intervals, respectively F I G U R E 5 Occupancy of golden-headed lion tamarins (Leontopithecus chrysomelas) within cabrucas (Ψ) as a function of (a) jackfruit trees (Artocarpus heterophylus) Importance Value Index (IVI), and (b) the diameter at breast height (DBH) index (calculated as the sum of median DBH values for shade trees recorded per plot at each cabruca site). The dotted line and the color-coded area represent the estimates and the 95% confidence intervals, respectively Tamarin Association (http://www.micoleao.org.br/) has achieved very positive results from sustainable tourism activities focused on the endangered golden lion tamarin (Leontopithecus rosalia) showing that this activity has potential in other parts of Brazil. Promoting sustainable ecotourism as an alternative source of local income, combined with biodiversity conservation, has already been proposed by the state management decree. Explicitly linking regional economic development with biodiversity conservation, while maintaining the status of traditional cabrucas is, therefore, a wise strategy that can likely perpetuate these wildlife-friendly systems.