Abstract
Successful management of wildlife populations requires accurate estimates of abundance. Abundance estimates can be confounded by imperfect detection during wildlife surveys. N-mixture models enable quantification of detection probability and, under appropriate conditions, produce abundance estimates that are less biased. Here, we demonstrate how to use the R-INLA package for R to analyze N-mixture models, and compare performance of R-INLA to two other common approaches: JAGS (via the runjags package for R), which uses Markov chain Monte Carlo and allows Bayesian inference, and the unmarked package for R, which uses maximum likelihood and allows frequentist inference. We show that R-INLA is an attractive option for analyzing N-mixture models when (i) fast computing times are necessary (R-INLA is 10 times faster than unmarked and 500 times faster than JAGS), (ii) familiar model syntax and data format (relative to other R packages) is desired, (iii) survey-level covariates of detection are not essential, and (iv) Bayesian inference is preferred.
Highlights
R-integrated nested Laplace approximation (INLA) was approximately 500 times faster than JAGS and 10 times faster than unmarked. This was the case despite the fact that unmarked produced maximum likelihood (ML) estimates and the JAGS analysis was run in parallel with each of three Markov chain Monte Carlo (MCMC) chains simulated on a separate virtual computing core
The data set is available as a demonstration data set in unmarked, so we compared the performance of R-INLA with that of unmarked using the analysis settings and model structure described in unmarked documentation
The purpose of this work was to detail the use of the R-INLA package (Rue et al 2017) to analyze N -mixture models and to compare analyses using R-INLA to two other common approaches: JAGS (Plummer 2003; Lunn, Jackson, Best, Thomas, and Spiegelhalter 2012), via the runjags package (Denwood 2016), which employs MCMC methods and allows Bayesian inference, and the unmarked package (Fiske and Chandler 2011), which uses maximum likelihood and allows frequentist inference
Summary
Successful management of wildlife species requires accurate estimates of abundance (Yoccoz, Nichols, and Boulinier 2001). This detection probability can be used to correct abundance estimates for imperfect detection (Royle 2004) Data resulting from this survey design are often modeled using an explicitly hierarchical statistical model referred to in the quantitative wildlife ecology literature as an N -mixture model (Royle and Nichols 2003; Dodd and Dorazio 2004; Royle 2004; Kéry, Royle, and Schmid 2005). P is commonly modeled as logit(pi,j) = α0 + α1xi,j, a logit-linear function of site-survey covariates This estimation approach can be extended to cover K distinct breeding or wintering seasons, which correspond with distinct years for wildlife species that are resident during annual breeding or wintering stages (Kéry, Dorazio, Soldaat, Van Strien, Zuiderwijk, and Royle 2009). The R-INLA approach is different from unmarked in that inference about model parameters falls within a Bayesian framework
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
