Landscape-level human disturbance results in loss and contraction of mammalian populations in tropical forests

While the severity of the current biodiversity crisis is debated [1,2], defaunation, i.e., the loss of animals from local population declines, range contractions, and extinction, is widely documented [3,4]. Tropical forests represent the most species-rich terrestrial biome, yet human population growth and associated land conversion continue to make habitat loss, fragmentation, and unsustainable hunting major drivers of vertebrate decline [5,6], resulting in the highest concentration of threatened mammals relative to other biomes [4,7]. Some mammal species in tropical forests now survive almost exclusively in protected areas (hereafter PAs) [8] that often occur within extensively transformed and heavily populated landscapes; others depend on remote and sparsely populated areas [9,10]. However, it remains unclear how mammalian communities within tropical forest respond to pressures in the broader landscape in which they are embedded. Some evidence suggests that even biodiversity within PAs is threatened by higher anthropogenic threats outside their borders [10–13]. For example, forest loss in the 50 km surrounding tropical forest PAs contributed to a decline in tree species richness within PAs [11]; forest fragmentation in a 10 km buffer around monitoring sites reduced colonization rates of forest-specialist mammals [14]. Nevertheless, empirical evidence and assessments of the consequences of landscape-level human impacts on the richness and occupancy status of tropical forest wildlife communities are lacking. As governments commit to increasing the global PA extent to prevent further biodiversity loss following the Kunming-Montreal Global Biodiversity Framework [15], identifying the landscape-level conditions that can enhance wildlife conservation is paramount [16].

To date, data quality, geographic coverage, and methodological inconsistencies have prevented rigorous objective examination of the status of tropical forest wildlife. Large-scale assessments of mammalian species richness are typically based on range maps (e.g., [17,18]), which can overestimate species occurrence [19] and lack information on abundance. For mobile species, range maps confound absence with non-detection, potentially biasing the estimation of true community size [20]. Literature-based meta-analyses are prone to additional biases due to heterogeneity in data quality and spatiotemporal scales [21]. In contrast, in situ data collected systematically at broader scales produce robust inferences that are more appropriate for evaluating trends and testing hypotheses at continental and global scales [22]. Camera-trapping has revolutionized wildlife monitoring [23], with its cost-effectiveness and unmatched potential for scaling-up and standardization [24]. Here, we collected an extensive tropical data set of detections for 239 primarily ground-dwelling mammalian species in 37 forests throughout the tropics collected from 2,021 camera-traps (Fig 1). Sampled areas included those forming the TEAM Network [25] plus 20 additional areas that extend the geographic coverage and the gradients of anthropogenic pressure and legal protection status. Sampled area landscapes ranged from nearly intact to severely fragmented and human-dominated landscapes, and legal protection status ranged from PAs to multiple-use and unmanaged forests. This newly compiled and standardized data set of wildlife data collected throughout the tropics provides a unique opportunity to quantify the effects of anthropogenic disturbance on mammal richness and occupancy throughout tropical forests globally.

Fig 1. Map of the 37 tropical forest areas considered in the study.

(A) Inset maps represent 2 protected areas embedded in contrasting landscapes: landscapes with low human density (left) and landscapes with high human density (right). Green areas in the inset maps represent PA boundaries. (B) Bottom chart shows estimated species richness (black dots) and the observed number of detected species with camera-traps (blue triangles) in each forest, which are ordered by increasing human density. See S4 Table in the SI Appendix for the list of site codes. Forest cover layer derived from Hansen and colleagues [102], whereas human population density layer derived from Gridded Human Population of the World [89]. The data underlying this figure can be found in S1 Data. PA, protected area.

https://doi.org/10.1371/journal.pbio.3002976.g001

We test 4 non-mutually exclusive hypotheses for mechanisms linking variation in mammal community richness and occupancy to landscape-scale anthropogenic processes in tropical forests. Species richness represents the size of the mammalian community and here we focus on its geographic (among-area) variation, while community occupancy represents the central tendency of site use probability for all species belonging to a community. Occupancy is a measure of distribution [26,27] that accounts for imperfect detection and is a useful metric to overcome the inherent difficulties of abundance estimation [28]. It reflects site-level responses of wildlife to local pressures that can manifest potentially within a short time, and it ultimately affects species richness when a species no longer occupies an area. Hence, species richness and community occupancy provide complementary information on how mammalian communities respond to anthropogenic drivers of both local species extinction and population decline [29]. We test how they vary in response to (1) the amount of forest habitat loss; (2) degree of forest fragmentation; (3) distance to settlements; and (4) level of human density in the broader landscape (Fig 2) [11].

Fig 2. Hypothesized landscape-scale processes (H1– H4) that may influence the species richness and community occupancy of mammal communities.

The figure depicts the hypothesized influence of landscape-scale forest cover (H1), forest configuration (i.e., fragmentation and connectivity of remnant forest patches) (H2), distance to settlements (H3), and human density (H4). Landscape is defined as the area extending 50 km from camera-trap arrays [11]. Asterisks indicate predictions supported by the results.

https://doi.org/10.1371/journal.pbio.3002976.g002

Our first hypothesis is that as more forest cover is lost, remaining forest patches become isolated islands unable to sustain the pristine array of species due to species-area effects and altered capacity to support viable populations. Hence, where landscape forest cover is lower, we predict lower mammalian richness (Hypothesis (H) 1, Prediction (P) 1). Although Haddad and colleagues [30] have found no overall effect of area reduction on abundance, we hypothesize lower community occupancy where landscape forest cover is lower (H1, P2). Indeed, before a local extinction occurs, a population declines in abundance with consequent shrinkage of its distribution, resulting in a decrease in species-level occupancy. Second, we hypothesize that the size and shape of remnant forest patches in the landscape (i.e., habitat configuration) can affect population source-sink dynamics among monitoring sites and surrounding landscapes [31]. Forest loss has resulted in 70% of remaining tropical forests occurring within 1 km from an edge [30]. Reduced patch area and increasing isolation have been associated with reduced abundances for a range of taxa, including mammals [30]. Similarly, global assessments have shown reduced mammal distributions with higher fragmentation [32]. Hence, we predict reduced mammalian community occupancy in landscapes with more fragmented forest cover (H2, P1). We also predict lower species richness with increasing fragmentation (H2, P2) as a result of relict populations disappearing over time as a manifestation of the extinction debt [33]. Conversely, fragmentation may create more habitat heterogeneity colonized by generalist species [34]; hence, we alternatively predict an increase in species richness (H2, P3). Additionally, anthropogenic edges can promote the use of open habitats and agricultural matrices for dietary resources or refugia by some species [35], which may result in a lack of variation in community occupancy in response to the degree of fragmentation (i.e., reduced patch size and increased patch isolation) as some species have higher occupancy near edges while others spatially avoid them (H2, P4).

Third, we hypothesize that the presence of settlements and infrastructure in the landscape favors human permeability into forest interiors, driving the exploitation of resources and hunting [36]. We, therefore, predict lower mammalian community occupancy near infrastructure (H3, P1). We also predict lower species richness as the potential outcome of human-induced filtering of sensitive and misanthropic species [37] (H3, P2). Fourth, we assume that higher landscape-scale human population density represents both a direct stressor to wildlife and a proxy for threats that are unknown or difficult to measure, such as demand for bushmeat products [38] and forest resource extraction, which may create a landscape of fear [39,40]. Human density can affect species presence and distribution, resulting in either range expansions or contractions for species that exploit anthropogenic environments or avoid them, respectively [39,41]. While predicting an outcome is difficult given human density can serve as a proxy for multiple complex processes, we test the following alternative predictions for both community metrics: (1) an increase in abundance of synanthropic species or species not targeted by hunters via density compensation or ecological release offsets the decline of sensitive species [42,43]. We therefore predict no net change in average community occupancy in response to human density (H4, P1). (2) Alternatively, as a positive relationship has been found between increasing mammal densities and high level of Human Footprint Index globally [44], we hypothesize an increase in community occupancy (H4, P2) as the result of domination by human-tolerant species. For species richness, (3) we predict no variation (H4, P3) as the local disappearance of sensitive species may be compensated for by the colonization of new adaptable ones [1,45]. However, for 16 tropical forest PAs, higher human density in the immediate surroundings was associated with lower richness of area-demanding and habitat specialist guilds, particularly insectivores [46]. In US parks, greater human density in a 50 to 100 km buffer predicted mammal extinction inside PAs [47]. Thus, (4) we alternatively predict lower mammalian richness with higher surrounding human density as the outcome of species filtering, with loss of sensitive and less adaptable species (H4, P4).

Coupling this newly synthesized data set with state-of-the-art community modeling, we simultaneously test hypotheses about factors contributing to geographic variation in the absolute size (species richness) and distribution (community occupancy) of partially observed ecological communities. We test hypotheses about variation at both levels of community structure in relation to forest cover, habitat configuration, distance to settlements, and human density measured throughout the camera-trap monitoring sites and a 50-km buffer surrounding them for each forest (Figs 2 and S1). We also include the size and habitat productivity of the sampled areas, as well as the continent and the coefficient of variation (CV) of annual rainfall, to account for their potential influences on species richness. To our knowledge, this is the first study that simultaneously models species richness and community occupancy across many areas at once as complementary metrics of community structure, providing for a novel and powerful approach to studying how tropical forest wildlife communities respond to landscape-scale drivers of change globally.

We empirically demonstrated the negative impacts of anthropogenic threats occurring in the broader landscape on both forest mammal species’ richness and community occupancy throughout the tropics. Previous attempts to address related questions using the TEAM camera-trap data focused exclusively on occupancy trends [48] or dynamics [14,49], functional composition [46,50], or taxonomic and functional diversity [51]. In contrast, our approach allowed for the simultaneous quantification of multiple drivers at 2 levels of community organization (species richness and community occupancy) while accounting for sampling biases, and extending both the number of areas and the gradient of anthropogenic disturbance in surrounding landscapes. Using standardized and systematic data recorded within tropical forests under varying degrees of anthropogenic disturbance, we provide much needed empirical evidence on hypothesized drivers of mammal community size and distribution throughout the tropics. Our results indicate that human density, habitat loss, and fragmentation have detrimental consequences for tropical forest wildlife worldwide. Specifically, we found that human density was the single, strongest predictor of species richness, whereby mammal communities occurring within areas surrounded by higher human population density had significantly fewer species (supporting H4, P4), even though most areas sampled were within PAs. Moreover, average mammal occupancy was lower in areas characterized by smaller and more fragmented areas of remaining forest in the landscape (versus larger and more continuous, supporting H1, P2 and H2, P1). Therefore, addressing the drivers impacting tropical forest wildlife critically depends on mitigating threats from human density and anthropogenic pressures that originate well beyond where populations occur [52].

That mammal species richness decreased as human population density in the landscape increased (hypothesis H4, P4 supported) is arguably the most robust empirical evidence to date documenting the detrimental effects of human presence on tropical wildlife. Specifically, our model predicts a 1% decline in species richness with an increase of 16 persons/km² in the landscape. This result expands on the known negative effect of a sampling area’s human footprint on tropical mammal richness in Africa and Asia [51] to encapsulate impacts on species richness within tropical forests from well beyond sampling areas. High human density may lead to extirpation from overhunting [9] and species loss from other detrimental human activities [39]. Information on known local extinctions within recent decades independent of the camera-trap data analyzed was available for 11 of the monitored forests (S2 Table). The forests in this study where known local extinctions have occurred are among the areas under greatest anthropogenic pressure, including higher than average human density, further suggesting that contemporary anthropogenic threats have already impacted species richness. Across our data set, estimated species richness was among the lowest in Virunga National Park, Rwanda (18 species), surrounded by very high human population density (400 persons/km²) and a highly fragmented landscape. Indeed, the Virunga Massif has one of the highest human densities in the world [53]. In contrast, estimated species richness was among the highest in a sustainable logging concession in Gabon (PWG, 34 species), which is surrounded by very low human population (6.57 persons/km²). Furthermore, in the Udzungwa Mountains of Tanzania, the Udzungwa Scarp Nature Reserve has reduced mammal richness and occupancy when compared to the nearby Udzungwa Mountains National Park. Decades of isolation of the Nature Reserve’s forest, which has higher human density and more fragmented forest cover in the surrounding landscape relative to the National Park, led to the local extirpation of at least 3 species by the early 1970s, i.e., the buffalo (Syncerus caffer), the African elephant (Loxodonta africana), and the leopard (Panthera pardus) [54]. Surprisingly, we found that landscapes with more forest cover did not support richer communities of forest mammals, which contrasts with our hypothesis (H1, P1) and with the known positive relationship between the size of available habitat and species richness [55]. Additionally, species richness was not significantly associated with fragmentation, in contrast with H2, P2 and H2, P3. Rather than finding lower species richness in areas with greater habitat loss or fragmentation, the lack of significant effects may be due to turnover in species composition in modified environments, such as the replacement of forest-dependent species by habitat generalists that are able to colonize empty niches, resulting in a net-zero change in community size. Alternatively, the lack of a significant positive relationship between forest cover and species richness may suggest a degree of empty forest syndrome [56] in which forests remain but fauna have been lost for human disturbances other than habitat loss, such as hunting. We tested distance to settlements as a proxy for hunting because it reflects accessibility, but distance to settlement did not have a detectable effect on species richness. This result can be interpreted as a lack of evidence that accessibility reduced species richness or that the measure of distance to settlements may have been a poor hunting proxy.

In contrast to the negative effect of human density on species richness, human density did not predict variation in occupancy, supporting the prediction that high human density positively affects the distribution of some species while negatively affecting others, resulting in no net effect of human density on occupancy (supporting H4, P1). This density compensation mechanism may result from the net effect of negative human impact on misanthropic species and positive effect, via ecological release, on synanthropic species, or species not targeted by hunters. Such decoupling in the responses of richness and occupancy to human density may be a consequence of local extinction filtering, i.e., reduced distributions of the most vulnerable species and thriving of the more resilient ones [57]. The lack of support for higher community occupancy near high human density (H4, P2) does not preclude it from occurring if more sensitive species disappear over time under increasing pressure from human density. Critically moreover, mammal communities had higher occupancy when forest cover was more extensive (H1, P2) and less fragmented (H2, P1). Hence, larger and better-connected forest habitat sustains larger and more broadly distributed populations [58], whereas forest fragmentation contributes to lower mammal occupancy. That the effect of the landscape-scale habitat configuration on community occupancy was even stronger than that of forest cover may be attributable to fragmentation altering source-sink population dynamics (hypothesis H2, P1 supported), with a reduced capacity of some species to recolonize forest fragments after local populations are lost [59]. Interestingly, we note that while we addressed community-wide responses, the positive effects found in relation to both forest cover and forest patch size reveal that most species in the community benefit from increasing forest cover and patch size, even though these responses may be driven by the strict forest dependent species.

In contrast to our hypotheses, distance to settlements did not significantly predict variations in average community occupancy or species richness (hypothesis H3, P1 and H3, P2, rejected), suggesting that, similarly to the effect of human density on occupancy, infrastructure may positively affect the distribution and persistence of some species while negatively affecting others, resulting in no net effect [39,60]. While winners generally have faster life-history and more opportunistic diets when compared to losers [39,41], more work is required to determine which species characteristics best predict these outcomes [61]. Lastly, we found that landscape-scale anthropogenic pressures markedly affected species richness, while sampling area, precipitation, and NDVI did not, although larger sampling areas [55] and more productive forests [62] typically support more mammal species. This finding may reflect an adequate standardization of the camera-trap sampling design among areas while highlighting that landscape-scale anthropogenic impacts can override known ecological determinants of species richness such that the fundamental relationship between area and species richness was no longer detectable. In addition to the variables we examined, variation in richness among areas has likely been influenced by other factors, particularly biogeographic history [63]. In particular, the significantly lower richness estimated for the Neotropical fauna may reflect the disproportionally high Pleistocene extinctions, in which the continent lost ca 80% of large-bodied mammals [63].

By focusing on mammals in tropical forests, where similar environments support communities with similar functional composition and vulnerability to anthropogenic pressures [46,64], and by using in situ observational data from the largest number of sites and widest gradient of protection throughout the biome to date, we have shown that landscape-scale anthropogenic pressures are detrimental to wildlife, even within PAs. Indeed, 31 out of 37 sampled areas (84%) were within PAs, supporting earlier expert opinion-based evidence of their vulnerability to landscape threats [10]. For the 31 PAs, forest cover and forest patch size were significantly higher inside than outside PA borders (i.e., across the 50 km buffer; S2 Fig), further highlighting the role of landscape-scale habitat conditions for tropical mammals. The use of complementary metrics of community structure allowed us to determine that the effects of anthropogenic threats manifest differently for the total number of species supported compared to the distribution of the species present in the monitored forests. Our results demonstrate that the size and distribution of mammal communities in tropical forests are tied to human population density, forest cover loss, and degradation in the broader landscape where they occur.

Therefore, mammal conservation in tropical forests depends on mitigating the complex detrimental effects of anthropogenic pressures beyond PA borders [65–68]. Examples of such strategies include preventing further forest loss in the landscape, establishing buffer zones, and restoring habitat and connectivity in the wider matrix [68]. Most importantly, with half of the world population expected to reside in tropical regions by 2050 [6], our result of a marked impact of human density on wildlife indicates that holistic landscape-scale planning that harmonizes forest conservation with livelihood development becomes more imperative than ever. Multiple-use PAs and land-sharing approaches [68,69] are important strategies in this regard. Ninety percent of people living in extreme poverty in the tropics rely heavily on forest resources [70], thus improvement of living standards, promotion of alternatives to fuelwood for cooking and increased education can further allay tropical forest degradation [71]. In conclusion, our study warns that the creation of new PAs as provided by the Kunming-Montreal Global Biodiversity Framework may not achieve the desired biodiversity outcomes without concurrent investment in addressing landscape-scale threats. Towards this end, our results are relevant to the current UN Decade of Ecosystem Restoration [68], whereby priority locations for restoring tropical forests should be identified in landscapes surrounding isolated PAs, as these may promote persistence of tropical forests mammals.