Bayesian Hierarchical Models bring new life to estimating species richness with non-linear functions
Talk, American Fisheries Society - Annual Meeting, 2023
Recommended citation: Toavs, T. R., Hasler, C. T., Suski, C. D., & Midway, S. R. (2022). "Bayesian Hierarchical Models bring new life to estimating species richness with non-linear functions." American Fisheries Society Annual Meeting. Grand Rapids, Michigan
Presenting the use of BHMs to estimate species richness.
Abstract
Species richness, the number of species in a defined environment, is often estimated from species accumulation curves (SAC), which are graphical representations of the aggregate number of unique species discovered in an environment with increasing sampling effort. Generally, SACs increase steeply with initial sampling effort and then rise slowly as the effort accumulated and is thought to be approaching the true species richness for the sampling location. The decrease in the accumulation rate allows for fitting a rarefaction curve to a SAC, even in situations with relatively low effort, which then can be extrapolated to an asymptote that serves as the species richness estimate. Different methods of rarefying SACs exist including conventional non-parametric estimators (e.g., Chao and Jackknife) and also non-linear asymptotic functions (NLF). NLFs have received criticism for performance, precision, and fit issues, but with increasing access to big data and advanced statistical methods, these concerns may be alleviated. Species richness estimates with NLFs, specifically when data is spatiotemporally rich, could potentially be enhanced through the use of Bayesian hierarchical models (BHM). More specifically, BHMs that add a hierarchical structure to NLFs with the addition of random effects at the watershed level. Using a novel, continental-scale fish database and simulated data we examined species richness estimation with both BHMs and conventional non-parametric estimators to explore the potential of BHMs for species richness estimates. Preliminary results show similar estimates and precision captured in BHMs and conventional non-parametric estimators, along with an increase in performance and precision with the addition of random effects compared to no hierarchical structure in NLF.