Univariate and Multivariate Spatial Modeling of Species Abundance.

spAbundance

spAbundance fits univariate (i.e., single-species) and multivariate (i.e., multi-species) spatial N-mixture models, hierarchical distance sampling models, and generalized linear mixed models using Markov chain Monte Carlo (MCMC). Spatial models are fit using Nearest Neighbor Gaussian Processes (NNGPs) to facilitate model fitting to large spatial datasets. spAbundance uses analogous syntax to its “sister package” spOccupancy (Doser et al. 2022). Below we provide a very brief introduction to some of the package’s functionality, and illustrate just one of the model fitting functions. For more information, see the resources referenced at the bottom of this page and the “Articles” tab at the top of the page.

Installation

You can install the released version of spAbundance from CRAN with

install.packages("spAbundance")

To download the development version of the package, you can use devtools as follows:

devtools::install_github("doserjef/spAbundance")

Note that because we implement the MCMC in C++, you will need a C++ compiler on your computer to install the package from GitHub. To compile C++ on Windows, you can install RTools. To compile C++ on a Mac, you can install XCode from the Mac app store.

Functionality

All model fitting functions allow for Poisson and negative binomial distributions for the abundance portion of the model. All GLM(M)s also allow for Gaussian and zero-inflated Gaussian models. Note the two functions for fitting spatailly-varying coefficient models are only available for Gaussian and zero-inflated Gaussian models.

Example usage

Load package and data

To get started with spAbundance we load the package and an example data set. We use data on 16 birds from the Disney Wilderness Preserve in Central Florida, USA, which is available in the spAbundance package as the neonDWP object. Here we will only work with one bird species, the Mourning Dove (MODO), and so we subset the neonDWP object to only include this species.

library(spAbundance)
# Set seed to get exact same results
set.seed(500)
data(neonDWP)
sp.names <- dimnames(neonDWP$y)[[1]]
dat.MODO <- neonDWP
dat.MODO$y <- dat.MODO$y[sp.names == "MODO", , ]

Fit a spatial hierarchical distance sampling model using spDS()

Below we fit a single-species spatially-explicit hierarchical distance sampling model to the MODO data using a Nearest Neighbor Gaussian Process. We use the default priors and initial values for the abundance (beta) and detection (alpha) coefficients, the spatial variance (sigma.sq), the spatial decay parameter (phi), the spatial random effects (w), and the latent abundance values (N). We also include an offset in dat.MODO to provide estimates of density on a per hectare basis. We model abundance as a function of local forest cover and grassland cover, along with a spatial random intercept. We model detection probability as a function of linear and quadratic day of survey and a linear effect of wind.

# Specify model formulas
MODO.abund.formula <- ~ scale(forest) + scale(grass) 
MODO.det.formula <- ~ scale(day) + I(scale(day)^2) + scale(wind)

We run the model using an Adaptive MCMC sampler with a target acceptance rate of 0.43. We run 3 chains of the model each for 20,000 iterations split into 800 batches each of length 25. For each chain, we discard the first 10,000 iterations as burn-in and use a thinning rate of 5 for a resulting 6,000 samples from the joint posterior. We fit the model using 15 nearest neighbors and an exponential correlation function. Run ?spDS for more detailed information on all function arguments.

# Run the model (Approx run time: 1 min)
out <- spDS(abund.formula = MODO.abund.formula,
            det.formula = MODO.det.formula,
            data = dat.MODO, n.batch = 800, batch.length = 25,
            accept.rate = 0.43, cov.model = "exponential",
            transect = 'point', det.func = 'halfnormal',
            NNGP = TRUE, n.neighbors = 15, n.burn = 10000,
            n.thin = 5, n.chains = 3, verbose = FALSE)

This will produce a large output object, and you can use str(out) to get an overview of what’s in there. Here we use the summary() function to print a concise but informative summary of the model fit.

summary(out)
#> 
#> Call:
#> spDS(abund.formula = MODO.abund.formula, det.formula = MODO.det.formula, 
#>     data = dat.MODO, cov.model = "exponential", NNGP = TRUE, 
#>     n.neighbors = 15, n.batch = 800, batch.length = 25, accept.rate = 0.43, 
#>     transect = "point", det.func = "halfnormal", verbose = FALSE, 
#>     n.burn = 10000, n.thin = 5, n.chains = 3)
#> 
#> Samples per Chain: 20000
#> Burn-in: 10000
#> Thinning Rate: 5
#> Number of Chains: 3
#> Total Posterior Samples: 6000
#> Run Time (min): 0.7641
#> 
#> Abundance (log scale): 
#>                  Mean     SD    2.5%     50%   97.5%   Rhat ESS
#> (Intercept)   -1.8186 0.3428 -2.5560 -1.8020 -1.1956 1.0692  64
#> scale(forest) -0.1999 0.2056 -0.5818 -0.2102  0.2443 1.0292 160
#> scale(grass)   0.1206 0.1939 -0.2720  0.1244  0.4938 1.0210 229
#> 
#> Detection (log scale): 
#>                    Mean     SD    2.5%     50%   97.5%   Rhat ESS
#> (Intercept)     -2.5392 0.1196 -2.7602 -2.5436 -2.2815 1.0850 204
#> scale(day)      -0.1658 0.0807 -0.3380 -0.1629 -0.0187 1.0341 364
#> I(scale(day)^2)  0.0011 0.0828 -0.1530 -0.0011  0.1648 1.0391 352
#> scale(wind)     -0.1352 0.0769 -0.2931 -0.1344  0.0126 1.0037 534
#> 
#> Spatial Covariance: 
#>            Mean     SD   2.5%   50%  97.5%   Rhat ESS
#> sigma.sq 0.4941 0.2648 0.1725 0.431 1.1929 1.0156 169
#> phi      0.0016 0.0018 0.0003 0.001 0.0072 1.0644 102

Posterior predictive check

The function ppcAbund performs a posterior predictive check on the resulting list from the call to spDS. We provide options to group, or bin, the data in different ways prior to performing the posterior predictive check, which can help reveal different types of inadequate model fit. Below we perform a posterior predictive check on the data grouped by site with a Freeman-Tukey fit statistic, and then use the summary function to summarize the check with a Bayesian p-value.

ppc.out <- ppcAbund(out, fit.stat = 'freeman-tukey', group = 1)
summary(ppc.out)
#> 
#> Call:
#> ppcAbund(object = out, fit.stat = "freeman-tukey", group = 1)
#> 
#> Samples per Chain: 20000
#> Burn-in: 10000
#> Thinning Rate: 5
#> Number of Chains: 3
#> Total Posterior Samples: 6000
#> 
#> Bayesian p-value:  0.535 
#> Fit statistic:  freeman-tukey

Model selection using WAIC

The waicAbund function computes the Widely Applicable Information Criterion (WAIC) for use in model selection and assessment.

waicAbund(out)
#> N.max not specified. Setting upper index of integration of N to 10 plus
#> the largest estimated abundance value at each site in object$N.samples
#>       elpd         pD       WAIC 
#> -167.74186   14.03248  363.54866

Prediction

Prediction is possible using the predict function, a set of covariates at the desired prediction locations, and the spatial coordinates of the locations. The object neonPredData contains percent forest cover and grassland cover across the Disney Wildnerness Preserve. Below we predict MODO density across the preserve, which is stored in the out.pred object.

# First standardize elevation using mean and sd from fitted model
forest.pred <- (neonPredData$forest - mean(dat.MODO$covs$forest)) /
               sd(dat.MODO$covs$forest)
grass.pred <- (neonPredData$grass - mean(dat.MODO$covs$grass)) /
               sd(dat.MODO$covs$grass)
X.0 <- cbind(1, forest.pred, grass.pred)
colnames(X.0) <- c('(Intercept)', 'forest', 'grass')
coords.0 <- neonPredData[, c('easting', 'northing')]
out.pred <- predict(out, X.0, coords.0, verbose = FALSE)

Learn more

The vignette("distanceSampling"), vignette("nMixtureModels"), and vignette("glmm") provide detailed descriptions and tutorials of all hierarchical distance sampling models, N-mixture models, and generalized linear mixed models in spAbundance, respectively. Given the similarity in syntax to fitting occupancy models in the spOccupancy package, much of the documentation on the spOccupancy website will also be helpful for fitting models in spAbundance.

Citing spAbundance

Please cite spAbundance as:

Doser, J. W., Finley, A. O., Kéry, M., and Zipkin, E. F. (2023). spAbundance: An R package for single-species and multi-species spatially-explicit abundance models. arXiv Preprint.

References

Doser, J. W., Finley, A. O., Kéry, M., and Zipkin, E. F. (2022). spOccupancy: An R package for single-species, multi-species, and integrated spatial occupancy models. Methods in Ecology and Evolution. https://doi.org/10.1111/2041-210X.13897.