Impacts of an indigenous settlement on the taxonomic and functional structure of dung beetle communities in the Venezuelan Amazon

Abstract

In the last 50 years, traditionally nomadic indigenous communities in Amazonia have increasingly adopted more sedentary lifestyles as a result of external influences. Permanent settlements lead to the concentration of disturbances (e.g., forest extraction and hunting) and threaten vulnerable species as well as those that provide important ecosystem services such as dung beetles. Here we evaluated the abundance, taxonomic, and functional structure (composition and diversity) of an ecological indicator group—dung beetles—along a disturbance gradient associated with a permanent settlement of the Jotï people in the Amazonian region of Venezuela. We applied generalized linear model to assess the response of dung beetle abundance to settlement distance and latent variable model to assess the influence of settlement distance on taxonomic diversity and functional structure. We found the abundance of roller-species increased but small-bodied beetles decreased away from the settlement. We found that proximity to the Jotï settlement did not affect metrics of taxonomic and functional diversity of the dung beetle assemblages in general, although functional evenness was lower away from the settlement. In contrast, we found impacts on the functional composition of dung beetles, with significant increase in the community-weighted means for roller species and large-bodied dung beetles away from Jotï settlement. Our findings suggest that the transition from nomadism to a more sedentary lifestyle has not caused widespread collapse in the diversity of dung beetle assemblages surrounding the settlement, however significant trends were observed in species-specific responses to human impact, and these responses were mediated by functional traits.

Appendices

Appendix 1

Justification for the use of negative binomial GLMM

The reasons for using the negative binomial distribution and a GLMM in our analyses are as follows. Initial analysis using a Poisson GLMM resulted in overdispersion due to the relatively large number of zeros in the beetle abundance (60%). Zero inflated Poisson models (Zuur and Ieno 2016b) were also overdispersed and we therefore applied the negative binomial GLMM.

Model equation for negative binomial GLMM

$$Beetles_{ij} \sim NB\left( {\mu_{ij} ,\theta } \right)$$

$$E\left( {Beetles_{ij} } \right) = \mu_{ij}$$

$$\mu_{ij} = e^{{Intercept + Covariates_{ij}+a_{i}+ s_{j}}}$$

$$a_{i} \sim N\left( {0,\sigma_{Trap}^{2}}\right)$$

$$s_{j} \sim N\left( {0,\sigma_{Species}^{2}}\right)$$

Beetles_ij is the number of beetles caught in trap i for species j, where i = 1,.., 55 and j = 1,..,26. To allow for dependency between multiple observations from the same trap, we included a random intercept site (a_i), which is assumed to be normal and independently distributed with mean 0 and variance σ²_Trap. We also included a random intercept species (b_j), which is assumed to be normal and independently distributed with mean 0 and variance σ²_Species. This random effect imposes a dependency structure between observed beetle numbers from the same species (i.e. observations from different traps from the same species).

Appendix 2

Model fitting using Bayesian context using Markov Chain Monte Carlo (MCMC) implemented in JAGS

Models were fit in a Bayesian context using Markov Chain Monte Carlo (MCMC) techniques implemented in JAGS (Plummer 2003) from within R (further details in Appendix 1). A burn-in of 80,000 iterations was used with 3 chains. We used a thinning rate of 100 and 4500 iterations were used for each posterior distribution. All continuous covariates were standardized. Once the models were fitted, model validation was applied to investigate the presence of any residual patterns. We plotted posterior mean Pearson residuals versus posterior mean fitted values, versus each covariate, and we also inspected the posterior mean residuals for any spatial dependency, and dependency between species.

The specification controlling the MCMC sampling for the LVM model uses the default values in the ‘boral’ package and are as follows:

• 1. Length of burn-in i.e., the number of iterations to discard at the beginning of the MCMC sampler was 1000.

• 2. Number of iterations including burn-in was 40,000.

• 3. Thinning rate was 30.

• 4. Seed for JAGS sampler was set to the value 123.

Appendix 3

Formulation of the latent variable model

The formulation of the LVM is as follows:

$$Y_{ij} \sim N\left( {\mu_{ij} ,\sigma_{j}^{2} } \right)$$

$$E\left( {Y_{ij} } \right) = \mu_{ij} = \alpha_{j} + X_{j} \times \beta_{j} + u_{ij}$$

The Y_ij is the value of the i^th observation on diversity or functional structure index j. There are 9 response variables, hence j = 1, …, 9. And there are 55 sites, which means that i = 1, …, 55. The x_j contains the 5 covariates. In an ordinary linear regression model, we assume that the residual terms u_ij are independent and normally distributed. In a multivariate GLMM we use a non-diagonal residual covariance matrix to model dependency between response variables. In the LVM we use the following construction.

$$u_{ij} = Z_{i}^{'} \times \lambda_{j}$$

The residuals u_ij are equal to a linear combination of typically 2 latent variables z_i (each of length 55), very much like axes in ordination techniques (e.g. principal component analysis, canonical correspondence analysis). The λ_js are factor loadings and tell us how the latent axes influence the response variables. Model validation followed the steps described in Hui et al. (2015). A more detailed description of LVM models can be found in Warton et al. (2015). We compared LVMs with 0, 1 and 2 latent axes using the widely applicable information criterion (WAIC), the expected Akaike information criterion (EAIC) and the expected Bayesian information criterion (BIC). See the ‘boral’ package documentation for or original references (Akaike 1974; Schwarz 1978; Watanabe 2013) explanation how these are calculated.

Appendix 4

Multicollinearity plot of the covariates

The multicollinearity plot of the covariates for three species traits (activity period, size, and nesting strategy), distance from settlement, and vegetation characteristics (plant species richness, and canopy openness). The plot includes values within the boxes that indicate variance inflation factor (VIF), which quantifies the severity of multicollinearity. VIF values indicate minimal collinearity among covariates. Boxes without numbers indicate the VIF values were very small. Following Zuur et al. (2010), we applied a VIF threshold value of 3 and below to support low levels of multicollinearity.

Appendix 5

Plots for model validation to verify the presence of any residual patterns

The following figures show a posterior mean Pearson residuals versus posterior mean fitted values; b posterior mean Pearson residuals versus covariates; c semivariogram plot of posterior mean residuals for any spatial dependency

.

Appendix 6

Pearson residuals for each dung beetle species combination

	AteC	CanthiG	CanthiK	Canthisp1	Canthisp2	CanthonQ	CanthonSemio	CanthonT	DelG	DelO	DelS	DichB	DichD
AteC		− 0.3	− 0.2	− 0.1	0.1	− 0.2	− 0.1	− 0.1	− 0.1	0.2	0.2	0.5	− 0.2
CanthiG			0.1	0.2	− 0.1	0.1	− 0.3	− 0.3	0.2	− 0.3	0.0	− 0.3	− 0.2
CanthiK				0.2	0.2	− 0.2	0.1	− 0.1	− 0.1	0.0	0.1	− 0.1	0.0
Canthisp1					0.3	− 0.2	− 0.2	− 0.3	0.1	− 0.2	0.1	− 0.2	− 0.2
Canthisp2						− 0.1	− 0.1	− 0.2	− 0.1	− 0.2	0.0	− 0.1	− 0.2
CanthonQ							0.2	0.1	− 0.1	0.3	0.0	0.0	0.0
CanthonSemio								0.4	0.1	0.1	− 0.1	− 0.1	0.1
CanthonT									− 0.1	0.0	− 0.3	− 0.1	0.0
DelG										− 0.1	0.2	− 0.1	0.2
DelO											0.2	0.4	− 0.1
DelS												0.4	− 0.2
DichB													− 0.2
DichD
DichL
DichM
DichP
EuryC
EuryH
EuryHy
OntR
OxyC
OxyF
OxyS
ScyC
SylB
Uro
AteC	DichL	DichM	DichP	EuryC	EuryH	EuryHy	OntR	OxyC	OxyF	OxyS	ScyC	SylB	Uro
CanthiG	0.6	0.0	− 0.1	− 0.1	− 0.2	0.3	0.4	0.0	0.2	0.1	− 0.1	− 0.3	− 0.1
CanthiK	− 0.1	− 0.3	− 0.1	0.1	− 0.1	− 0.3	− 0.1	− 0.2	0.1	0.1	0.1	0.0	− 0.3
Canthisp1	− 0.1	0.0	− 0.2	0.2	− 0.3	0.1	− 0.1	− 0.1	− 0.1	− 0.1	0.1	− 0.1	− 0.1
Canthisp2	0.1	− 0.2	0.2	0.4	0.0	− 0.2	− 0.1	− 0.2	− 0.1	− 0.1	0.2	0.0	0.0
CanthonQ	0.0	− 0.1	0.0	0.4	0.0	0.0	0.0	− 0.1	0.0	0.0	0.0	0.3	− 0.1
CanthonSemio	− 0.2	0.1	0.2	− 0.1	0.2	0.0	− 0.2	− 0.1	− 0.1	− 0.1	− 0.1	0.0	0.0
CanthonT	− 0.1	0.2	0.0	0.0	− 0.1	− 0.1	0.0	0.2	− 0.1	0.0	− 0.1	− 0.1	0.1
DelG	− 0.2	0.5	0.0	− 0.1	0.3	0.0	− 0.1	0.0	− 0.2	− 0.2	− 0.2	0.0	0.2
DelO	0.0	− 0.2	0.0	0.0	0.0	− 0.1	− 0.1	− 0.2	− 0.1	0.2	0.1	0.0	− 0.1
DelS	0.1	0.2	0.0	− 0.2	0.0	0.1	− 0.1	0.1	0.0	0.0	0.1	− 0.3	0.1
DichB	0.2	− 0.2	− 0.3	− 0.1	0.1	0.1	− 0.1	− 0.2	0.1	− 0.1	0.6	− 0.1	− 0.2
DichD	0.2	0.1	− 0.1	− 0.1	− 0.2	0.3	0.2	0.0	0.0	− 0.1	0.1	− 0.1	− 0.1
DichL	− 0.2	0.0	0.2	0.0	− 0.1	− 0.1	− 0.1	0.1	− 0.2	0.0	− 0.2	0.2	0.0
DichM		− 0.1	− 0.1	0.0	− 0.1	0.0	0.2	0.0	0.1	0.1	− 0.1	− 0.3	− 0.2
DichP			0.1	− 0.1	− 0.1	− 0.1	− 0.1	0.2	− 0.2	− 0.1	− 0.2	0.0	0.0
EuryC				0.0	0.0	− 0.1	− 0.1	0.0	0.0	0.1	− 0.1	0.1	− 0.1
EuryH					− 0.1	− 0.1	0.0	0.1	0.3	− 0.3	− 0.1	0.2	− 0.2
EuryHy						− 0.2	− 0.2	0.0	0.0	0.0	0.2	0.0	− 0.1
OntR							0.5	0.0	0.1	0.0	0.1	0.0	0.0
OxyC								0.1	0.2	0.0	− 0.2	0.1	− 0.1
OxyF									− 0.1	0.0	0.1	− 0.1	0.0
OxyS										− 0.1	0.0	0.0	− 0.1
ScyC											− 0.1	0.0	− 0.1
SylB												0.0	− 0.1
Uro													− 0.1

Appendix 7

Results of WAIC, EAIC, and BIC for the latent variable model

Results of the widely applicable information criterion (WAIC), the expected Akaike information criterion (EAIC) and the expected Bayesian information criterion (BIC) comparing the latent variable models with 0, 1 and 2 latent axes. Values with asterisk indicate the preferred latent variable model with 0, 1, or 2 axes.

	0 axis	1 axis	2 axes
WAIC	1034.081	778.5258	447.1811*
EAIC	1098.503	920.0688	758.1535*
EBIC	1363.391*	1454.0476	1557.0195