A variety of factors can affect the biodiversity of tropical mammal communities, but their relative importance and directionality remain uncertain. Previous global investigations of mammal functional diversity have relied on range maps instead of observational data to determine community composition. We test the effects of species pools, habitat heterogeneity, primary productivity and human disturbance on the functional diversity (dispersion and richness) of mammal communities using the largest standardized tropical forest camera trap monitoring system, the Tropical Ecology Assessment and Monitoring (TEAM) Network. We use occupancy values derived from the camera trap data to calculate occupancy-weighted functional diversity and use Bayesian generalized linear regression to determine the effects of multiple predictors. Mammal community functional dispersion increased with primary productivity, while functional richness decreased with human-induced local extinctions and was significantly lower in Madagascar than other tropical regions. The significant positive relationship between functional dispersion and productivity was evident only when functional dispersion was weighted by species' occupancies. Thus, observational data from standardized monitoring can reveal the drivers of mammal communities in ways that are not readily apparent from range map-based studies. The positive association between occupancy-weighted functional dispersion of tropical forest mammal communities and primary productivity suggests that unique functional traits may be more beneficial in more productive ecosystems and may allow species to persist at higher abundances.

Study sites

We examined mammal communities from 15 moist tropical forest protected areas around the world, including sites in Africa, Asia, Madagascar, and the Neotropics (Table S1). All sites have been part of the Tropical Ecology Assessment and Monitoring (TEAM) Network and followed a standardized annual camera trapping protocol to monitor terrestrial (i.e. ground-dwelling) mammals (Jansen et al. 2014).

Data

Camera trapping and Occupancy

Sixty camera traps were deployed at each TEAM site during the site’s dry season, which was defined as months with <100 mm average rainfall, or the drier part of the year at sites with no dry season. The camera traps used to monitor the mammal community were deployed at a density of 1 camera per 1 to 2 km², depending on the size of the protected area. Camera trap points were each monitored for at least 30 consecutive days for multiple years at each site, spanning from 2007 to 2014. From the TEAM camera traps, we used species-specific protected area-level occupancy estimates from Beaudrot et al. (2016) as a proxy for abundances in calculating occupancy-weighted functional dispersion. Our use of occupancy-weighted functional diversity is particularly suitable for identifying underlying community assembly processes for two reasons. First, camera traps capture the realized local community whereas range maps typically overestimate the geographical distribution of species by including potentially non-suitable habitat areas (Pimm et al. 2017). Secondly, the occupancy estimates we used account for biases that result from imperfect detection and would otherwise skew results in favor of species that are more likely to be detected on camera traps. Species occupancy estimates were calculated as the median of posterior distributions of 1000 values estimated from single-species multi-year Bayesian dynamic occupancy models (from Beaudrot et al. 2016). For functional dispersion, we used species occupancy values from the most recent year with occupancy estimates available, which varied among sites from 2012 to 2014.

The realized mammal community within each TEAM study site consisted of mammal species observed by camera traps in the camera trap survey area, which covered about 20,000 hectares of the typical TEAM protected area (Range: 14,239- 27,445 hectares; Table S1). Specifically, we included ground-dwelling mammal species with an average body mass greater than one kilogram that were monitored by the TEAM network camera traps. We set a body mass threshold of one kilogram because other field techniques may be more effective for monitoring small mammal communities (Glen et al. 2013). Previous studies have shown that TEAM camera trap survey areas varied in the size of their realized mammal community from 5 to 31 species (Beaudrot et al. 2016, Rovero et al. 2020). One hundred twenty-seven species of mammals were observed (Table S2) and 310 populations of these species were monitored by TEAM camera traps with 77 species found at more than one site.

Functional traits and diversity

We collected functional trait data for all study species through a search of published literature and databases (Table 1, Table S2, Table S3). We chose six functional traits to calculate functional diversity: 1) average body mass, 2) diet composition, 3) social or asocial behavior, 4) scansorial or entirely terrestrial substrate use, 5) activity period, and 6) average litter size. These functional traits relate to both the response of the species to environmental conditions (response traits) and to the role of the species in an ecosystem (effect traits, Weiss & Ray 2019). Body mass affects the quality and quantity of resources necessary for survival, and also approximates the impact that the species may have on the ecosystem in terms of spatial range use, nutrient dispersal and trophic regulation. Diet composition characterizes the resources a species requires, but also identifies other taxa in the ecosystem with which a species potentially interacts. Social group size can alter species’ allocation of time to different behaviors including foraging, predator avoidance, and care of offspring, and also indicates how the species’ impact will be distributed in space. Substrate use characterizes where a species can obtain resources and where a species will directly interact with an ecosystem. Activity period characterizes when a species obtains resources and interacts with an ecosystem. Finally, litter size characterizes the life history strategy of a species and indicates how the ecological impact of a species can vary temporally based on population dynamics. As with any functional trait study, our results are in part dependent upon the traits considered and the diversity encompassed by functional metrics can change with the inclusion or exclusion of specific traits (45). Traits used in this study were gathered at the species-level and multiple traits were reduced to binary variables, which may reduce the breath of trait variation to some extent. Values for average body mass and average litter size were continuous and numeric. Diet composition and activity period involved a suite of binary categories, of which species could be assigned a positive value for more than one category. Values for sociality and substrate use were each a single binary category (Table 1). Average body mass and average litter size were log-transformed because their distributions tend to be log-normal and to prevent inflation of functional diversity in sites with species with large body and litter sizes.

To preserve total inertia and distance between the same species occurring in different assemblages (e.g. Villéger, Maire & Leprieur 2017), we calculated functional dispersion (Laliberté & Legendre 2010) and functional richness (Mason et al. 2005) for all sites and species pools in a single trait space (Fig. 1) using the dbFD function from the `FD` R package (Laliberté et al. 2014). We had no a priori expectations that certain traits were more important for functional diversity than others. Similarly, previous studies of mammal functional diversity have given equal weight to the functional traits included in this study (Oliveira et al. 2016, Safi et al. 2011, Penone et al. 2016). We therefore weighted traits with multiple categories so that the sum all of categories within a single trait was the same as the weight of a trait with a single category. Functional dispersion measures the distribution of species in trait space or how similar a community is in terms of its functional traits. Functional dispersion can either be unweighted, with all species accounted for equally, or weighted by species abundances, with distance from the community centroid to abundant species in trait space contributing more to the metric. To calculate occupancy-weighted functional dispersion, the occupancy data were arranged in a species by site matrix and coupled with the trait matrix. Functional richness was calculated using a matrix of all species and their traits and measured as the volume of the convex hull encompassing all species found at a given site in trait space. The functional richness metric is incapable of taking into account species abundances. 

Predictor variables

We quantified species pool functional richness, habitat heterogeneity, primary productivity and two measures of anthropogenic disturbance to use as predictors of functional richness and occupancy-weighted functional dispersion. The species pool included all forest-dwelling species that may inhabit the park based on their geographic ranges. We extracted species pools for each protected area using global mammal ranges from the IUCN Redlist (IUCN 2019). Protected area extents were determined from publicly available shapefiles from the United Nations Environment Program World Database on Protected Areas (UNEP-WCMC & IUCN 2020). For one site that did not have an accurate shapefile available on WDPA (Pasoh Forest Reserve), we used a two-kilometer buffer of the minimum convex-polygon surrounding the camera traps to define this boundary. Any forest-dwelling, ground-dwelling mammal species with an average body mass over one kilogram that had a range that overlapped with the protected area or was detected on camera traps within the park was considered part of the species pool. Species pool richness ranged from 5 to 40 species, with a total of 170 species composing the TEAM site species pools (Table S2). Species pool functional richness was calculated using the same function as realized community functional richness.

We calculated habitat heterogeneity as the Shannon diversity of landcover types within a two-kilometer buffer of the minimum convex polygon surrounding the camera traps at each site. Specifically, landcover classifications were obtained from Copernicus satellite data from 2015, and classifications were according to Copernicus Global Land Service Dynamic Landcover Map at 100-meter resolution (CGLS-LC100, Bucchorn et al. 2019). Landcover classifications for the habitat heterogeneity calculation included closed and open forest categories (evergreen broadleaf, deciduous broadleaf, mixed, and indeterminate for each), shrubs, herbaceous vegetation, sparse vegetation, and herbaceous wetlands. We removed landcover types associated with large bodies of water (permanent water bodies and oceans) as these would be unsuitable habitat for ground-dwelling mammals. We also removed anthropogenic landcover classifications from the calculation of this variable to prevent overlap with the anthropogenic disturbance variables.

We used published primary productivity and human disturbance values for TEAM sites from Rovero et al. 2020. Normalized difference vegetation index (NDVI), which measures plant reflectance of different wavelengths of light and indicates vegetation health, was used as a proxy of productivity and calculated over the camera trap array at each forest site. Human density was quantified within the Zone of Interaction, which incorporates land cover, elevation, waterways and anthropogenic developments to delineate the spatial extent expected to impact natural processes inside the protected area (DeFries et al. 2010).

Finally, we included the impact of local species extinction on functional richness as an additional measure of anthropogenic disturbance and as an offset to control for losses in functional richness due to extirpations that have been documented at four TEAM sites over recent decades. Specifically, Korup, Cameroon lost the leopard, golden cat and giant pangolin; Barro Colorado Island and Soberania National Park, Panama lost the white-lipped peccary, giant anteater and likely the jaguar; Bwindi Impenetrable Forest, Uganda lost the buffalo, leopard, and giant forest hog; Nam Kading, Laos lost both the tiger and leopard (38). We note that local extinctions may not necessarily impact functional diversity because of the functional redundancy that occurs in tropical forest mammal communities (48). We quantified local extinction for each TEAM site as the difference in functional richness between the functional richness of the current realized community and the functional richness of current realized community as well as the recently extirpated species.

Analysis

We modeled occupancy-weighted functional dispersion and community functional richness as a function of environmental and anthropogenic predictor variables using Bayesian generalized linear regression. None of the predictor variables were highly correlated (r < 0.6) and all continuous variables were scaled and centered to produce standardized beta coefficient estimates with a mean of zero and standard deviation of one. Given the left-skew in the distribution of both functional diversity metrics, we specified a Weibull distribution for both of our models. The Weibull distribution is an unbounded continuous probability distribution that can accommodate skews and long tails in the data.  The global model consisted of all potential predictor variables, plus a categorical fixed effect for biogeographic region (Neotropics, Africa, Asia, Madagascar) to control for region-specific patterns in functional diversity. The Neotropics was the category to which all other bioregions were compared. We used the brm function from the brms (Bürkner 2017) package in R to fit the models. We visually assessed model trace plots and used Rhat criteria for convergence. We interpreted the contribution of our predictor variables to the functional diversity metrics based on whether the 95% credible interval of the parameter’s posterior distribution included zero.

To assess the importance of occupancy in functional dispersion, we also ran regression models of two additional unweighted functional dispersion calculations. Unlike occupancy-weighted functional dispersion, the additional calculations treated all species equally and did not incorporate variation in species abundances. As a result, they likely omit important ecological information about the relative abundances of functional traits.  We calculated unweighted functional dispersion using 1) presence data of the observed community from camera trap data, and separately using 2) presence data extracted from IUCN range maps.

Citations

Beaudrot, L., Ahumada, J. A., O’Brien, T., Alvarez-Loayza, P., Boekee, K., Campos-Arceiz, A., … Andelman, S. J. (2016). Standardized Assessment of Biodiversity Trends in Tropical Forest Protected Areas: The End Is Not in Sight. PLOS Biology, 14(1), e1002357.

Buchhorn, M., Smets, B., Bertels, L., Lesiv, M., Tsendbazar, N., Herold, M., Fritz, S. (2019) Copernicus Global Land Service: Land Cover 100m: epoch 2015: Globe. Dataset of the global component of the Copernicus Land Monitoring Service 2019.

Bürkner, P. C. (2017). brms: An R package for Bayesian multilevel models using Stan. Journal of statistical software, 80(1), 1-28.

DeFries, R., Rovero, F., Wright, P., Ahumada, J., Andelman, S., Brandon, K., ... & Liu, J. (2010). From plot to landscape scale: linking tropical biodiversity measurements across spatial scales. Frontiers in Ecology and the Environment, 8(3), 153-160.

Glen, A. S., Cockburn, S., Nichols, M., Ekanayake, J., & Warburton, B. (2013). Optimising camera traps for monitoring small mammals. PloS one, 8(6), e67940.

Gorczynski, D., & Beaudrot, L. (2020). Functional diversity and redundancy of tropical forest mammals over time. Biotropica.

IUCN. (2019).  The IUCN Red List of Threatened Species. Version 2019‐2. 

Jansen, P. A., Ahumada, J., Fegraus, E., & O’Brien, T. (2014). TEAM: a standardized camera trap survey to monitor terrestrial vertebrate communities in tropical forests. Camera Trapping: Wildlife Research and Management, 263–270.

Laliberté, E., Legendre, P., Shipley, B., & Laliberté, M. E. (2014). Package ‘FD’. Measuring functional diversity from multiple traits, and other tools for functional ecology.

Laliberté, E., & Legendre, P. (2010). A distance‐based framework for measuring functional diversity from multiple traits. Ecology, 91(1), 299-305.

Mason, N. W., Mouillot, D., Lee, W. G., & Wilson, J. B. (2005). Functional richness, functional evenness and functional divergence: the primary components of functional diversity. Oikos, 111(1), 112-118.

Penone, C., Weinstein, B. G., Graham, C. H., Brooks, T. M., Rondinini, C., Hedges, S. B., ... & Costa, G. C. (2016). Global mammal beta diversity shows parallel assemblage structure in similar but isolated environments. Proceedings of the Royal Society B: Biological Sciences, 283(1837), 20161028.

Pimm, S. L., Harris, G., Jenkins, C. N., Ocampo‐Peñuela, N., & Li, B. V. (2017). Unfulfilled promise of data‐driven approaches: response to Peterson et al. Conservation Biology, 31(4), 944-947.

Rovero, F., Ahumada, J., Jansen, P. A., Sheil, D., Alvarez, P., Boekee, K., ... & Tenan, S. (2020). A standardized assessment of forest mammal communities reveals consistent functional composition and vulnerability across the tropics. Ecography, 43(1), 75-84.

Safi, K., Cianciaruso, M. V., Loyola, R. D., Brito, D., Armour-Marshall, K., & Diniz-Filho, J. A. F. (2011). Understanding global patterns of mammalian functional and phylogenetic diversity. Philosophical Transactions of the Royal Society B: Biological Sciences, 366(1577), 2536-2544.

UNEP-WCMC and IUCN (2020), Protected Planet: The World Database on Protected Areas (WDPA) [On-line], [insert month/year of the version downloaded], Cambridge, UK: UNEP-WCMC and IUCN. Available at: www.protectedplanet.net.

Villéger, S., Maire, E., & Leprieur, F. (2017). On the risks of using dendrograms to measure functional diversity and multidimensional spaces to measure phylogenetic diversity: a comment on Sobral et al.(2016). Ecology Letters, 20(4), 554-557.

Weiss, K. C., & Ray, C. A. (2019). Unifying functional trait approaches to understand the assemblage of ecological communities: synthesizing taxonomic divides. Ecography, 42(12), 2012-2020.

Additional files and descriptions are available on the github repository of the project.