Anthropogenic modulators of species–area relationships in Neotropical primates: a continental-scale analysis of fragmented forest landscapes

Authors

  • Maíra Benchimol,

    Corresponding author
    1. School of Environmental Sciences, University of East Anglia, Norwich, UK
    2. Capes Foundation, Ministry of Education of Brazil, Brasília, DF, Brazil
    • Correspondence: Maíra Benchimol and Carlos A. Peres, School of Environmental Sciences, University of East Anglia, Norwich NR47TJ, UK.

      E-mails: m.souza@uea.ac.uk and c.peres@uea.ac.uk

    Search for more papers by this author
  • Carlos A. Peres

    Corresponding author
    1. School of Environmental Sciences, University of East Anglia, Norwich, UK
    • Correspondence: Maíra Benchimol and Carlos A. Peres, School of Environmental Sciences, University of East Anglia, Norwich NR47TJ, UK.

      E-mails: m.souza@uea.ac.uk and c.peres@uea.ac.uk

    Search for more papers by this author

Abstract

Aim

We conducted the first comprehensive quantitative review on the effects of habitat fragmentation on Neotropical primates to examine how both patch disturbance and landscape variables modulate species–area relationships (SARs) and species persistence in fragmented forest landscapes.

Location

Neotropical forests, from Mexico to Argentina.

Methods

We use species occupancy data from 705 forest fragments and 55 adjacent continuous forests nested within 61 landscapes, which we compiled from 96 studies reporting data on patch-scale species composition and patch size/location. Presence–absence data on 19 species functional groups and an index of hunting pressure and matrix type were assigned to each forest patch. We adopted a multilevel analysis, examining SARs and patterns of species retention coupled with the additive effects of hunting pressure and landscape connectivity both across all forest patches and 728 sites nested within 38 landscapes containing four or more sites.

Results

We uncovered a consistent effect of patch area in explaining primate species richness. Over and above area effects, however, SARs were strongly modulated by levels of hunting pressure at the landscape scale in predicting species occurrence and aggregate assemblage biomass. Matrix type was also a good predictor of both extant species richness and aggregate biomass when only non-hunted sites were considered, with patches in more permeable matrices containing more species.

Main conclusions

Although the importance of patch area in predicting species persistence is undeniable, we found that SARs were clearly affected by within-patch human exploitation of increasingly isolated primate populations. Both expanding the number of forest reserves and enforcing protection within nominal protected areas are therefore required to ensure the long-term persistence of full primate assemblages. We highlight the importance of considering multiple anthropogenic effects in assessing the synergistic effects of land use to explain patterns of species persistence in fragmented tropical forest landscapes.

Introduction

Tropical forests world-wide have been increasingly degraded by the relentless expansion of the twin processes of habitat loss and fragmentation (Heywood, 1995; Laurance & Peres, 2006). As these processes proceed, large tracts of pristine forest habitat are converted into smaller and more isolated forest patches embedded within a largely inhospitable matrix, with long-term detrimental effects on biodiversity (Turner, 1996; Fahrig, 2003). Although habitat isolation exacerbates species loss in most fragmented ecosystems, area effects play a prevailing role, with larger patches sustaining larger populations of a larger set of species (Watling & Donnelly, 2006; Ferraz et al., 2007). Species–area relationships (SARs) are therefore ubiquitous in most archipelagic landscapes and have been widely used to estimate local extinction rates associated with declining habitat areas (Angermeier & Schlosser, 1989; Lomolino, 2001; Drakare et al., 2006; He & Hubbell, 2011). In fact, SARs remain the most frequent approach in predicting biodiversity erosion in fragmented environments, despite outstanding uncertainties about the appropriateness of model fits (Lomolino, 2000; Tjorve, 2003, 2009; Triantis et al., 2012). However, SARs typically overlook synergistic interactions between area effects and external demographic stressors on (semi)isolated populations, which may accelerate local extinction rates. This includes anthropogenic forms of disturbance within fragmented landscapes such as edge-propagated wildfires and matrix mortality associated with hunting and roadkills (Woodroffe & Ginsberg, 1998; Peres, 2001; Cochrane & Laurance, 2002).

The landscape context may therefore be critical in determining the form of community disassembly within forest patches, yet these additive effects are rarely considered in empirical SARs. For instance, matrix permeability clearly affects patch occupancy, as species that can traverse the matrix or exploit its resources often occupy a larger number of patches (Gascon et al., 1999; Antongiovanni & Metzger, 2005; Ewers & Didham, 2006; Lees & Peres, 2009). Human perturbation also affects species persistence in forest patches and amplifies the detrimental impacts of fragmentation (Ewers & Didham, 2006). In Amazonian forest patches, hunting interacts synergistically with habitat fragmentation vastly reducing the large vertebrate species retention capacity of small patches (Peres, 2001). Most SAR applications to real-world fragmented forest landscapes would therefore show reduced explanatory power without explicitly considering the patch- and landscape-scale history of anthropogenic disturbance.

Neotropical primates comprise a unique taxonomic group to test SAR models in fragmented tropical forest landscapes as they are strict forest dwellers, highly conspicuous group-living species and highly variable in their tolerance to forest fragmentation (Harcourt & Doherty, 2005; Michalski & Peres, 2005). Forest primates play important roles in ecosystem structure and functioning because they account for a disproportionate share of the arboreal vertebrate biomass (Oates et al., 1990), often operate as central trophic species in forest food webs (Terborgh, 1983; Marsh, 2003) and are key seed dispersers (Link & Di Fiore, 2006). Moreover, primates are widely hailed as regional conservation icons, represent the best studied terrestrial mammal order (Reed & Fleagle, 1995), have been widely investigated throughout the Neotropics and account for much of the vertebrate data available from any tropical forest region. Neotropical forests contain the world's most diverse continental primate fauna (139 species in 19 genera: IUCN, 2008). Several studies have considered the detrimental effects of habitat fragmentation on primate species and assemblages (Estrada & Coates-Estrada, 1996; Marsh, 2003; Michalski & Peres, 2005). Yet no systematic review has attempted to examine the general continental-scale patterns of species persistence.

Here, we provide the first robust quantitative review of platyrrhine primate responses to habitat fragmentation throughout the Neotropics. In particular, we couple a patch- and landscape-level approach to variables describing the degree of anthropogenic disturbance to explain patterns of species persistence within forest isolates. We present the most comprehensive compilation on primate species occupancy of isolated forest patches from the northernmost (southern Mexico) to the southernmost (northern Argentina) Neotropical forest frontiers. These patches are embedded within a wide range of landscape contexts subjected to varying histories of human disturbance. We discuss the main drivers of primate local extinctions in fragmented forest landscapes, suggest how the fragmentation ecology research agenda could be enhanced and recommend priority conservation actions.

Methods

The dataset

We performed an exhaustive search of the formal and grey literature to identify all published and unpublished studies containing data on primate species composition within Neotropical forest fragments (defined as forest patches < 10,000 ha). To identify these studies, we first conducted a Web of Science and a Google Scholar search using the following keywords: habitat fragmentation, primates, mammals, landscape, Neotropical, with and without ‘Alouatta’ and ‘Cebus’, the most widely distributed genera. We then Google-searched these same keywords translated into Spanish and Portuguese to find publications in non-indexed journals, and undergraduate and postgraduate dissertations, thesis and monographs archived in all Meso and South American countries. We also searched and included the study bibliographies, unpublished studies we were aware of, and our own data from three fragmented landscapes (Michalski & Peres, 2005; Benchimol, 2009; Canale et al., 2012). We used a strict set of criteria to select the studies compiled in our initial database (Roberts et al., 2006). First, the study must have listed all primate species present within each patch, as obtained through field-verified interviews, linear-transect surveys, behavioural studies or any other documentation. Second, we selected only studies that explicitly provided either the size of fragments or exact geographical coordinates which enable us to measure patch size and assess the landscape context using Google Earth Pro tools. We set no restrictions on minimum numbers of sites surveyed and considered records of zero species richness (= 0) for previously occupied forest patches that no longer contained any species at the time of surveys.

Species data were tabulated following a functional classification of 24 platyrrhine primate ‘ecospecies’ (Peres & Janson, 1999). At each site, we assigned occupancy scores as either present [(1) for ecospecies that had been recorded at a patch] or absent [(0) when an ecospecies that was known to have occurred in the patch had been extirpated]. This enabled total primate species richness estimates (S) at any given patch. Post-fragmentation introductions of exotic and reintroductions of native species, as reported for a few sites, were excluded from our species richness estimates. Night monkeys (Aotus spp.), the only nocturnal platyrrhines, were added to our database for only those studies that explicitly documented nocturnal mammals using night surveys. At sites at which more than one independent data source describing the local species composition were available, we considered all species documented in at least one source.

We also compiled observed primate species composition data for large, unbroken forest areas adjacent to each fragment cluster (i.e. the best available continuous ‘pseudo-control’ sites), defined here as the nearest forest tracts > 10,000 ha that shared the same primate source fauna of adjacent study patches. We coupled these data with published NatureServe (Patterson et al., 2003) and IUCN (2008) range polygons, publications describing species ranges, and our own extensive joint personal knowledge (i.e. 42 years of fieldwork at > 120 Amazonian and Atlantic Forest landscapes) to estimate the baseline (pre-fragmentation) composition of the primate fauna at each site. This allowed derivation of the historical maximum primate species richness that would have once occurred at each site (hereafter, Smax). Because local primate species richness is widely variable throughout the Neotropics (Peres & Janson, 1999), we also calculated the proportion of extant species (pS) retained at each site as pS = S/Smax. We also calculated the aggregate assemblage biomass for each patch (B) by summing the estimated body masses of all extant sympatric species occurring at that patch and the estimated total biomass (Bmax) of all extant and extinct species that once occurred at the same patch. This allowed us to calculate the proportional extant biomass retained at each site as pB = B/Bmax.

For each patch, we recorded the geographical coordinates, the matrix type (i.e. the predominant land-cover class within a 1-km external buffer from each patch) and the level of hunting pressure on primates. Because the history and landscape context of forest patches are rarely characterized, we were unable to obtain these variables from most studies. However, we characterized landscape connectivity using two complementary approaches. Firstly, by assigning a matrix type index to each fragment based on its predominant components: (1) water; (2) pasture or cropland; or (3) secondary forest; and secondly, using Google Earth Pro tools to estimate the percentage of available forest cover outside fragments within 1 km from their perimeter. Suitable habitat cover within the matrix is widely recognized as robust indicators of landscape connectivity (Tischendorf et al., 2003). Although we were able to assign a matrix type index to all forest patches, we could only estimate matrix forest cover for 384 of all 705 patches, either because some studies failed to provide exact geographical coordinates of their study patch(es) or due to poor quality of the relevant images and/or severe cloud cover. Information on levels of subsistence, recreational and commercial hunting pressure was either obtained from the studies or information provided by the authors, who were approached individually in each case. We thus assigned historical levels of post-fragmentation hunting pressure at each site into three classes: (1) non-hunted; (2) lightly or occasionally hunted; and (3) heavily or persistently hunted.

Patches were considered to be spatially nested within landscapes for all landscape-scale analyses. Using ArcGIS (version 10; ESRI, 2011), we further combined all sites in our database with pan-Neotropical land-cover and digital elevation, The Shuttle Radar Topography Mission (SRTM) and distinguished landscapes within any given major region by overlaying spatial clusters of sites with layers of all significant geographical barriers including major rivers, montane areas, and large areas of historical deforestation. We deliberately avoided limiting the size threshold of a landscape as the number of patches they contained and their spatial configuration varied considerably.

From our comprehensive literature search, we included 96 publications in the final database, providing information on 760 sites (705 forest patches and 55 continuous forests) embedded within 61 landscapes scattered across 11 Neotropical countries from Mexico to Argentina (Fig. 1). The number of study sites per landscape varied from 1 to 144 and fragment sizes ranged from 0.1 to 9731 ha (see Appendix S1 in Supporting Information). Most sites examined here were exposed to light and heavy hunting pressure on primates (36.7% and 14.6%, respectively), and the predominant surrounding matrix was pasture and cropland (87.5%), compared with water (e.g. land-bridge islands within hydroelectric reservoirs) and secondary forests.

Figure 1.

Geographical locations of the 61 fragmented forest landscapes examined in this study. The circles are sized proportionally to the total number of forest sites surveyed within each landscape. Pie charts indicate the proportional mean and range of species retained (pS) within patches at each landscape.

Data analysis

To examine relationships between forest patch area and primate species richness (S), we first considered all 760 study sites and performed linearized forms of SARs using semi-log models, which perform well in explaining SARs (Rosenzweig, 1995; Lomolino, 2001). Because baseline species richness (Smax) in local primate faunas within each landscape varied widely across the entire continent (range = 1–8 species), we repeated these models considering the within-patch proportion of extant species (pS). We also assessed biomass–area relationships (BARs) between patch size and the total biomass (B) and the proportion of extant biomass retained in residual assemblages (pB) to examine broad patterns of species deletion across the body size spectrum available in each landscape. We performed nonlinear multimodel SARs for 38 of the 61 study landscapes where ≥ 4 forest sites had been surveyed. This was implemented by fitting nonlinear relationships based on eight possible SAR models using ‘mmSAR’ (Guilhaumon et al., 2010), including four convex (power, exponential, negative exponential and Monod) and four sigmoidal models (rational function, logistic, Lomolino and cumulative Weibull). A minimum of four sites per landscape was chosen because this is the smallest sample size required to run SARs using this R-package (Guilhaumon et al., 2010; http://mmsar.r-forge.r-project.org). We then used information theoretical analyses to evaluate model performance and parsimony using Akaike weights (Burnham & Anderson, 2002).

We further investigated the additive effects of external variables (hunting pressure and matrix type) on SARs, using the proportion of species (pS) and biomass (pB) retained at all 760 sites. Next, we ran all possible SAR semi-log regression models considering only 38 landscapes containing ≥ 4 study sites and examined how z-values (a direct measure of initial and overall slopes), intercepts and R2 values of landscape-specific SAR models responded to our external variables. ANCOVA was used to investigate the effects of matrix type on pS and pB, with patch area as a covariate. Finally, we used generalized linear mixed models (GLMMs) to relate variation in patch area, matrix type, forest cover and hunting pressure to the proportion of extant species (pS) and biomass (pB) retained within patches. This approach was the most appropriate to account for potential spatial autocorrelation (Bolker et al., 2009), with our global model incorporating a random term nesting ‘patches’ within ‘landscapes’, whereby same landscape patches shared the same source primate fauna. We also included Smax or Bmax values as offset variables in the proportion of species richness and biomass models, respectively, to account for patch-scale variation in the maximum species/biomass packing. We performed species richness and assemblage biomass GLMMs considering: (1) all 760 forest sites nested within 61 landscapes, with fixed effects available for each site – patch area, matrix type and hunting pressure; (2) only those forest sites for which we were able to obtain forest cover estimates, added as a 4th fixed effect (= 384 patches nested within 34 landscapes); and (3) only 728 forest sites nested within 38 landscapes containing ≥ 4 forest sites. Models were fitted using the ‘lme4’ package (Bates, 2007) within the R platform, and parameters were estimated using Laplace approximation as recommended by Bolker et al. (2009). We selected the ‘best’ models using the ‘MuMIn’ package (Bartón, 2009); examined every possible first-order variable combination; ranked them based on the Akaike Information Criterion (AIC; Burnham & Anderson, 2002); and determined the relative importance of each explanatory variable given their model frequency and cumulative Akaike weight.

Results

Species–area relationships

We found a clear positive patch area effect on total primate species richness across all 760 sites considering all landscapes (inline image = 0.299, < 0.001). This was confirmed by a positive area effect on the proportion of local species pools persisting within forest sites (inline image = 0.229, < 0.001, Fig. 2). Furthermore, increasingly larger forest patches retained a greater aggregate assemblage biomass (inline image = 0.223, < 0.001) and a larger proportion of the total biomass in the original primate fauna (inline image = 0.162, < 0.001, Fig. 2).

Figure 2.

Overall relationship between forest patch area and the proportion of primate species (a) and proportion of aggregate primate biomass (b) persisting within 760 sites investigated in this study. Circles are coloured according to levels of hunting pressure, with darker circles indicating more heavily hunted sites. Circles sizes are proportional to the pre-fragmentation maximum number of species that should have occurred at each patch, given historical records and any other information on species distributions.

Forest patch size was also a significant predictor of primate species richness at the landscape scale, explaining between 0% and 83.7% (mean inline image = 30.2%) of the variation in semi-log SAR models for the 38 landscapes containing at least four surveyed sites (Table 1). This explanatory power was further substantially improved with a nonlinear multimodel framework using raw (untransformed) data, which explained up to 61% of the landscape-scale SARs (mean inline image = 51.3%). The negative exponential model provided the most frequent ‘best-fit’ for SARs within those landscapes, followed by the power model. Nevertheless, we were unable to obtain proper convergence in nonlinear SAR models for 15 of the 38 landscapes, and this was independent of the number of patches investigated (r = −0.040, = 38).

Table 1. Coefficients (c), slopes (z) and adjusted R2 values of the ‘best’ nonlinear and semi-log species–area relationship (SAR) model for 38 Neotropical forest landscapes containing at least four study sites. Multimodel approach using ‘mmSAR’ was unable to run in landscapes containing only four or fewer sites
LandscapeRegionGeographical coordinatesNo of sites S max Hunting pressureMatrix typeBest modelSemi-Loga
Nameb c z inline image c s z s inline image
  1. a

    Formulae of the semi-log model: = cs + zslog(A). Here, cs = the intercept of the curve in arithmetic space and zs = a direct measure of the initial and overall slope.

  2. b

    Formulae of each model: Power – = cAz; Exponential – = c + zlog(A); Negative exponential (Negexpo) – = d(1−exp(−zA)); Monod – = d/(1cA−1); Rational function (Ratio) – = (c + zA)/(1 + dA); Logistic – = d/(1 + exp(−zA + f)); Lomolino – = d/1 + (zlog(f/A)); Cumulative Weibull (Weibull) – = d(1−exp(−zAf)); Invalid – analysis was not run successfully. S, number of species; A, area; c, z, f, d are fitted parameters (Guilhaumon et al., 2010).

  3. c

    No variation across the dataset.

Alta FlorestaBrazil (Amazon)56°05′ W, 09°54′ S1447NonePasture-CroplandRatio1.910.060.531.171.350.49
Los TuxtlasMexico90°48′ W, 16°14′ N882ModeratePasture-CroplandInvalid1.06−0.020.01
GuatemalaGuatemala89°32′ W, 16°58′ N502ModeratePasture-CroplandInvalid0.400.520.21
São JoãoBrazil (Atlantic Forest)42°01′ W, 22°27′ S493ModeratePasture-CroplandInvalid0.680.350.07
JequitinhonhaBrazil(Atlantic Forest)40°41′ W, 16°20′ S466HeavyPasture-CroplandPower0.850.180.41−0.101.050.39
Saint-EugeneFrench Guiana53°04′ W, 04°51′ N396NoneWaterPower0.590.260.55−0.271.630.64
South BahiaBrazil (Atlantic Forest)39°39′ W, 14°53′ S266HeavyPasture-CroplandPower1.570.050.101.500.270.07
BalbinaBrazil (Amazon)59°38′ W, 01°49′ S217NoneWaterLogistic6.220.020.87−1.062.380.83
PernambucoBrazil (Atlantic Forest)35°50′ W, 08°43′ S203NonePasture-CroplandInvalid1.21−0.190.00
Santa MariaBrazil (Atlantic Forest)53°42′ W, 29°43′ S201NonePasture-CroplandInvalid0.850.050.00
South-central AmazonBrazil (Amazon)54°53′ W, 02°50′ S177ModeratePasture-CroplandNegexpo5.270.020.183.920.320.00
Alter do ChãoBrazil (Amazon)54°57′ W, 02°29′ S168HeavyPasture-CroplandInvalid1.820.570.04
GuriVenezuela62°52′ W, 07°21′ N144NoneWaterPower0.670.180.540.170.880.61
Santa RosaCosta Rica85°39′ W, 10°50′ N133ModeratePasture-CroplandExponential−0.040.620.02−0.121.480.00
CampinasBrazil (Atlantic Forest)46°55′ W, 22°49′ S134NonePasture-CroplandExponential−0.500.680.62−0.691.660.58
Vale do TaquariBrazil (Atlantic Forest)52°02′ W, 29°41′ S121NonePasture-CroplandInvalid0.470.160.09
PDBFFBrazil (Amazon)59°52′ W, 02°24′ S116NoneSecondary forestWeibull5.490.170.901.601.290.61
Eastern AmazoniaBrazil (Amazon)47°47′ W, 02°33′ S117ModeratePasture-CroplandNegexpo6.360.000.522.271.060.32
BolíviaBolívia63°03′ W, 17°47′ S106ModeratePasture-CroplandNegexpo3.060.260.511.021.010.30
ChiapasMexico90°48′ W, 16°15′ N82ModeratePasture-CroplandLogistic2.010.070.591.100.270.38
MichelinBrazil (Atlantic Forest)39°19′ W, 13°46′ S84ModeratePasture-CroplandLogistic2.740.000.550.890.460.36
ViçosaBrazil (Atlantic forest)42°52′ W, 20°48′ S84ModeratePasture-CroplandInvalid1.40−0.090.00
Dois irmãosBrazil (Atlantic Forest)55°18′ W, 20°30′ S72NonePasture-CroplandInvalid−1.000.470.79
AugustoBrazil (Atlantic Forest)40°33′ W, 19°54′ S75NonePasture-CroplandPower1.740.120.190.671.150.03
Upper ParanáBrazil (Atlantic Forest)53°19′ W, 22°46′ S62HeavyPasture-CroplandPower0.750.140.240.450.490.04
SergipeBrazil (Atlantic Forest)37°14′ W, 11°12′ S63NonePasture-CroplandNegexpo3.150.040.91−0.721.890.84
North-eastern ColombiaColombia74°16′ W, 08°35′ N65ModeratePasture-CroplandPower1.040.170.700.491.020.57
MagdalenaColombia74°44′ W, 05°39′ N65ModeratePasture-CroplandInvalid4.03−0.330.00
Rio CascaBrazil (Atlantic Forest)42°44′ W, 20°04′ S55ModeratePasture-CroplandNegexpo2.600.060.121.440.520.00
MaranhãoBrazil (Atlantic Forest)48°08′ W, 05°00′ S55NonePasture-CroplandNegexpo4.350.040.920.551.340.54
CórregoBrazil (Atlantic Forest)39°50′ W, 18°24′ S54NonePasture-CroplandNegexpo4.090.000.67−4.972.270.48
ArarasBrazil (Atlantic Forest)44°15′ W, 22°25′ S55ModeratePasture-CroplandExponential−2.520.710.49−2.531.640.32
CorrientescArgentina58°50′ W, 27°30′ S51NonePasture-CroplandInvalid
La SuerteCosta Rica83°46′ W, 10°26′ N43NonePasture-Cropland Invalid1.960.410.10
TrinidadTrinidad & Tobago61°15′ W, 10°25′ N42ModeratePasture-CroplandInvalid−4.751.800.18
TucuruíBrazil (Amazon)49°30′ W, 04°16′ S47NoneWaterInvalid0.442.090.40
Barreiro RicoBrazil (Atlantic Forest)48°05′ W, 22°41′ S45NonePasture-CroplandInvalid−5.003.330.41

Are SARs modulated by hunting and connectivity?

Over and above patch area effects, we detected a significant effect of hunting pressure reported for individual sites in predicting primate species occupancy and biomass. The slopes of the relationships between patch area and the proportion of species (pS) and total biomass (pB) retained across all 760 forest sites were steeper in non-hunted sites than in lightly and heavily hunted sites (Fig. 3). In contrast, we failed to detect an effect of matrix type on the species richness and biomass retained considering all forest sites. Considering each landscape separately, the level of hunting pressure also profoundly affected the R2-values in semi-log SARs and BARs for the 38 landscapes containing at least four forest sites surveyed, whereas matrix type exerted no discernible influence (Fig. 4). Additionally, hunting pressure also affected the slopes of both SARs and BARs within patches sharing the same landscape (see Figs S2 & S3). The z-values and intercepts of these 38 SARs models did not show a clear trend in relation to hunting pressure and matrix type. However, when we considered only forest patches that had no history of hunting pressure, we detected a significant effect of matrix type on the relationship between patch area and the proportion of species (ANCOVA, F = 19.688, P < 0.001) and proportion of extant biomass retained (ANCOVA, F = 5.605, P = 0.018, Fig. 5).

Figure 3.

Relationship between forest patch size and the proportion of primate species (pS) and total biomass (pB) retained across 760 Neotropical forest sites, under varying levels of hunting pressure.

Figure 4.

Variation in R2 values, according to levels of hunting pressure and matrix type, in semi-log species–area relationships for 38 landscapes containing a minimum of four forest fragments surveyed. Matrix types are coded as: W, Water; P-C, Pasture and cropland; SG, Secondary growth.

Figure 5.

Relationship between forest patch size and the proportion of primate species (pS) and aggregate biomass (pB) retained within non-hunted forest isolates (= 383) surrounded by either water (i.e. true islands) or a matrix of cattle pastures and cropland.

Considering all 760 forest sites nested within 61 study landscapes, GLMMs showed that forest area (β = 0.723, P < 0.001) was a significant predictor of species richness, whereas forest area (β = 0.865, P < 0.001), level of hunting (β = −1.221, P < 0.001) and matrix type (β = 0.025, P < 0.001) were significant predictors of aggregate biomass (Table 2). The ‘best’ GLMM model explaining the proportion of species retained contained patch area, followed by a model including patch area and level of hunting pressure (∆AIC = 1.82, see Table S1). Considering the extant primate biomass, however, the top-ranking model included forest patch area and level of hunting pressure, followed by a model containing only patch area (∆AIC = 2.05). Given that pasture and cropland were the predominant matrix types surrounding forest patches across all studies, we ran all models again while excluding true islands and patches surrounded by secondary forest. However, this did not change the above pattern. We also found a similar pattern when performing GLMMs considering only the 728 forest sites nested within the 38 landscapes containing at least four forest sites. Patch area was again the strongest predictor of species richness (β = 0.798, P < 0.001), whereas forest area (β = 1.169, P < 0.001), level of hunting (β = −3.740, P < 0.001) and matrix type (β = 4.285, P < 0.001) were significant predictors of aggregate biomass. Likewise, we also detected the predominance of both area (β = 1.081, P < 0.001) and hunting pressure (β = −2.693, P < 0.05) effects on the proportion of species retained when GLMMs considered the additional connectivity effect of matrix forest cover. The best model included both patch area and level of hunting pressure, followed by a model containing patch area, hunting pressure and matrix forest cover (∆AIC = 1.68). When we ran the GLMMs explaining extant primate biomass, however, matrix forest cover was also a good predictor in the best model (β = 0.022, P < 0.05), in addition to patch area (β = 2.764, P < 0.001) and hunting pressure (β = −1.699, P < 0.01).

Table 2. Summary of generalized linear mixed models (GLMMs) of the proportion of primate species (pS) and extant biomass (pB) retained within 760 forest sites nested in 61 fragmented forest landscapes across the Neotropics, with Smax and Bmax as offsets
ParameterEstimateUnconditional SEConfidence intervalRelative importance
p S
Intercept−1.1370.345−1.812, −0.461
Hunting2−0.3190.322−0.950, 0.3110.24
Hunting3−0.5740.409−1.375, 0.227
Matrix20.9541.028−1.060, 2.9690.16
Matrix30.4750.445−0.398, 1.348 
Area (log x)0.7230.0970.533, 0.9141.00
p B
Intercept−1.1650.376−1.902, −0.429 
Hunting2−0.7080.411−1.514, 0.0970.74
Hunting3−1.2210.523−2.247, −0.195 
Matrix20.0250.261−0.490, 0.539 
Matrix30.2800.543−0.785, 1.344 
Area (log x)0.8650.1080.653, 1.0771.00

Discussion

This is the most comprehensive systematic review of the effects of habitat fragmentation on an entire suborder of tropical forest vertebrates and the first quantitative synthesis of community-wide Neotropical primate responses to anthropogenic forest fragmentation. A global-scale analysis on primate SARs (Harcourt & Doherty, 2005) considered data from only 7.4% of the forest fragments and 34% of the landscapes that we examined here. Primate fragmentation studies tend to focus on species responses to habitat size and isolation within single landscapes (Onderdonk & Chapman, 2000; Marsh, 2003; Marshall et al., 2010). Here, we used an extensive dataset comprising 760 forest patches embedded within 61 landscapes to examine which patch and landscape variables best predict primate species richness throughout the neotropics. We show that forest patch size is a robust predictor of primate species persistence, which is consistent with Harcourt & Doherty's (2005) global-scale analysis. We also show that patch area is a good predictor of our proxy of primate assemblage biomass. However, patch area effects, rather than operating in isolation, interact synergistically with game population depletion by hunters, who are highly selective and preferentially target large-bodied species in most forest landscapes (Peres, 1990, 2000), thereby driving non-random local extinctions. Maintaining large patches of primary/secondary forest, or increasing their size and/or connectivity through forest restoration, is therefore central to any conservation initiative, but this alone does not ensure the persistence of full primate assemblages in overhunted fragmented forest landscapes.

Several studies have shown the importance of patch area in retaining vertebrate species in tropical forest fragments (Ferraz et al., 2007; Michalski & Peres, 2007; Stouffer et al., 2009). This typically positive area effect can be attributed to a greater habitat diversity; larger pools of trophic and/or structural resources; and larger populations, all of which can operate independently to reduce local extinction rates (Connor & McCoy, 2001). Yet a single-minded focus on habitat patch area and isolation is not enough to maximize the biodiversity value of tropical forest remnants, as real-world ‘working’ landscapes are subjected to multifaceted natural and anthropogenic disturbances that interact synergistically with forest area (Laurance & Peres, 2006). These additive perturbation effects continue to be neglected in several fragmentation ecology studies, which largely focus on MacArthur & Wilson's (1967) original island biogeography paradigm, which that has single-mindedly stressed the effects of area and isolation while overlooking the influence of external stressors on species persistence within patches.

In addition to patch area, hunting pressure strongly affected the pattern of primate species persistence across all sites. Large-bodied primates comprise the most preferred prey items for indigenous groups in Neotropical forests (Redford & Robinson, 1987; Jerozolimski & Peres, 2003), and hunting is widely considered to be the most severe threat for many primate species (Milner-Gulland & Bennett, 2003), vastly surpassing the importance of habitat loss in the largest remaining tracts of tropical forests (Peres & Lake, 2003). Hunting interacts synergistically with forest fragmentation by facilitating physical access by hunters to prey populations, reducing local population sizes and precluding immigrants from rescuing sink populations, all of which can accelerate local extinction rates in forest isolates (Peres, 2001). Accordingly, we found clearly evidence that subsistence hunters had access to forest fragments within at least 27 landscapes. This aggravated the local extinction probability of mid-sized to large-bodied primate species, thereby reducing the explanatory power of species-area models in fragmented forest landscapes. Hunting also markedly inflated the minimum size of forest patches required to retain primate assemblages of any given size. For instance, forest fragments containing one half of the species in the original fauna were on average 233 ha in overhunted landscapes but only 34 ha in non-hunted landscapes. Conversely, retaining 90% of all species on average required a sevenfold increase in fragment size from 16,748 ha in non-hunted landscapes to 111,737 ha in landscapes where primates had been hunted.

Hunting pressure also significantly depressed the extant relative biomass persisting within forest patches, indicating that larger-bodied species had been disproportionately affected and driven to local extinction in overhunted fragments. This fits the broad patterns in Neotropical forests where primate species exceeding 3 kg are often considered fair game species, but smaller species are typically ignored by subsistence hunters (Redford & Robinson, 1987; Peres, 1990). Our body mass weighed SAR models predicted that retaining 50% of the total biomass of the baseline primate assemblage at each fragment would require a patch size increase from 28 ha at non-hunted sites to 1924 ha at heavily hunted sites. Likewise, retaining an almost complete (90%) assemblage biomass could be achieved by either setting aside non-hunted patches of c. 14,858 ha or hunted forest tracts approaching 1 million hectares, which are unrealistic in virtually all landscapes. Primate surveys throughout lowland Amazonia indicate that hunting significantly reduces the crude primate biomass in otherwise undisturbed continuous forest sites, where large-bodied atelines (i.e. Ateles, Lagothrix, Alouatta) succumb to steep population declines (Peres & Palacios, 2007). This can be extended to fragmented forest landscapes where hunters had driven large-bodied species to local extirpation way before the longer-term effects of patch area and/or isolation can operate (Peres, 2001). Likewise, our results clearly show that hunting pressure strongly affected both the proportion of species and the proportion of biomass retained within forest fragments, thereby flattening the slopes of both species–area and BARs and reducing their R2 values.

Although to a lesser extent, we also detected a discernible effect of landscape connectivity on overall species and biomass persistence. Matrix type exerted only a minor effect on species richness and aggregate assemblage biomass for all 760 forest sites examined here even though models containing ‘matrix type’ as a covariate appeared in the three top-ranking models explaining the proportion of extant primate species (see Table S1). However, we were able to detect a significant effect of matrix type on extant species richness and biomass when we restricted the analysis to non-hunted forest patches only. SAR slopes are good indicators of species persistence in island systems (Triantis et al., 2012), and we detected a significantly higher z-value for both relationships considering fragments isolated within a matrix of pastures or cropland. Moreover, once we added an independently derived metric of matrix forest cover into our analysis, we found that patches surrounded by large amounts of forest habitat also retained a higher proportion of extant primate species richness and biomass. Several studies have shown the importance of neighbouring habitats on the occupancy rate of birds and mammals in tropical forest fragments (Andrén, 1994; Antongiovanni & Metzger, 2005; Prugh et al., 2008; Lees & Peres, 2009), highlighting that enhancing matrix quality can facilitate movements across forest remnants (Franklin & Lindenmayer, 2009). The matrix plays a key role in both interpatch travel and foraging of forest primates in a fragmented landscape in Central Africa (Anderson et al., 2007), and large-bodied species were able to colonize Amazonian forest fragments by traversing a benign second-growth matrix (Boyle & Smith, 2010). Indeed, the structure of the matrix influences the likelihood of movements across forest patches, depending on patterns of locomotion and dispersal. For primates, open water seems to be more difficult to traverse than agropastoral and young secondary forest matrices, given that they can both serve as stepping stones or corridors for individual/group movements across forest patches. In the southern Amazonian landscape of Alta Floresta, for instance, a breeding group of spider monkeys occupying a c. 7-ha forest fragment for several years has successfully overcome a gap distance > 1 km by travelling on the ground through scrubby cattle pastures to reach a neighbouring fragment (C.A. Peres, unpublished data), underscoring the locomotion plasticity of even one of the most arboreal Neotropical mammals. However, once a landscape has been severely defaunated due to chronic overhunting, even relatively well-connected patches are likely to remain vacant as neighbouring source populations would remain unavailable for successful recolonization. Greater matrix permeability may therefore facilitate primate movements across forest patches, but this alone is not enough if hunting pressure continues to ravage populations unchecked.

Future directions

Most empirical applications of the SAR fail to consider mechanisms of species loss other than the classic area and isolation effects that have been so heavily revisited under the traditional island biogeography paradigm. Using a meta-analytical approach, we detected decisive interactions between the effects of habitat area and human-induced wildlife depletion of local populations in determining patterns of primate species persistence and assemblage biomass right across the New World tropics. Further studies should therefore consider the historical land use context contributing to the full mosaic of environmental perturbations in evaluating patterns of species persistence in fragmented landscapes. For tropical forest vertebrates in particular, future studies should consider the landscape structure in which fragments are embedded, rather than focusing entirely on patch-scale variables (Arroyo-Rodríguez & Mandujano, 2009). Improving the analytical power and policy utility of these studies will also require further details on the nature of historical anthropogenic disturbances affecting forest habitat isolates (e.g. wildfires, hunting, selective logging), larger spatial replication and sample sizes, and better measures of landscape connectivity between patches, all of which require better spatial data reported in individual studies. We also suggest that researchers should evaluate the interactions between landscape variables and species traits to enhance our understanding of species sensitivity to fragmentation and propose efficient mechanisms for species-specific conservation (Henle et al., 2004).

Conservation implications

Due to their universal charismatic appeal, non-human primates are widely recognized as conspicuous flagship species for biodiversity conservation (Mittermeier et al., 2013). Conservation strategies designed to retain full complements of primate species can therefore ensure the persistence of much of the co-occurring tropical forest biota. Based on our continental wide analysis on patterns of primate persistence in fragmented forest landscapes, we propose the following recommendations to inform conservation policy and action.

  1. Allocate higher conservation and research priorities to fragmented landscapes under a restoration paradigm. Because even small, isolated forest patches can retain a significant fraction of the original forest biota, protection of forest fragments becomes an imperative, particularly in highly deforested and/or semi-defaunated landscapes (Turner & Corlett, 1996; Canale et al., 2012). Expanding habitat area through forest restoration programmes or enhancing protection of both forest structure and composition in existing forest fragments should therefore be encouraged in all Neotropical landscapes. However, we found that persistence of at least 60% of the local pool of primate species requires forest patches of 100 ha or larger, suggesting that conservation efforts should prioritize patches considerably larger than this minimum size threshold. Patches ≥ 100 ha comprise only 11.5% of the 245,173 remaining fragments across the Brazilian Atlantic forest (Ribeiro et al., 2009) and c. 25% of existing fragments (including over 1.12 million km2 of forest) within four states of Brazilian Amazonia (Broadbent et al., 2008). These two forest regions encompassed most South American forests and contain the largest number of sites in our dataset (Fig. 1).
  2. Depletion of primate populations within fragments by subsistence hunters should be curbed or precluded. Hunting pressure had a decisive detrimental effect on large-bodied primate persistence within the fragments we investigated. This is consistent with other studies showing that hunting pressure vastly accelerates species loss from tropical forest fragments (Peres, 2001; Thornton et al., 2011). Enforcing hunting restrictions within forest fragments in both public protected areas and private landholdings and implementing education programs designed for local communities near those fragments would mitigate the pervasive effects of hunting and other forms of patch-scale disturbance on biodiversity.
  3. Re-establish connectivity between forest patches. In highly fragmented landscapes, enhancing the suitability of the surrounding habitats can facilitate matrix movements between fragments, thereby increasing patch occupancy in the long term (Andrén, 1994; Prugh et al., 2008). Setting aside or restoring riparian or upland forest corridors between remaining patches through land use subsidies should also be considered in mitigating biodiversity erosion in fragmented landscapes.

Acknowledgements

We thank all primatologists who helped us enhance our dataset by providing further details on their study landscapes, and three referees who provided constructive comments on the manuscript. M.B. is funded by a Brazilian Ministry of Education PhD studentship (CAPES, 080410/0). This manuscript was co-written during a CAPES-funded visiting fellowship by C.A.P. to Museu Paraense Emílio Goeldi, Brazil.

Biosketches

Maíra Benchimol is a PhD student at the Centre for Ecology, Evolution and Conservation at the University of East Anglia, UK. Her doctoral project seeks to understand the impacts of habitat fragmentation on biodiversity loss in Neotropical forests, considering both large vertebrate extinctions and phytodemographic transitions. She works at the interface between tropical ecology and conservation biology, encompassing the effects of anthropogenic disturbance on vertebrates, population and landscape ecology, mammal ecology and conservation, and community-based monitoring programs. Her PhD studies are funded by the Brazilian Ministry of Education (CAPES).

Carlos A. Peres is a Professor of Tropical Conservation Ecology at University of East Anglia, UK. His research interests include the ecology of tropical forest disturbance, the effects of harvesting on tropical forest wildlife, and the science, policy, and management of tropical forest resources. In particular, he runs a research program throughout a number of Amazonian forests to understand how faunal assemblages are structured in relation to key natural and anthropogenic environmental gradients at scales ranging from single landscapes to the entire basin.

Author contributions: M.B. and C.A.P. contributed equally to the conceptual ideas behind the manuscript, data acquisition and data analysis, and manuscript preparation.

Ancillary

Advertisement