β-diversity in metacommunitiesSubmitted by editor on 4 September 2015.Get the paper!
This paper tests whether a statistical measure of community structure can actually differentiate between neutral and niche assembly processes. More broadly, we were interested in the connection between the ecological processes that determine community structure and statistical indicators of community structure that can be readily calculated from field observations. [Our conversations about the underlying ideas and issues that arose from discussions inspired an old blog post (at the EEB & Flow, http://evol-eco.blogspot.com/2013/11/community-structure-what-are-we-mis...).]
As an example, we focused on the “β-null deviation" measure, originally published by Jon Chase and collaborators (2011). It was initially developed as a null model approach that allowed differentiation between niche driven changes in β-diversity and changes resulting from random variability. However, β-null deviation—the difference between the observed β-diversity measure and the null expectation—is increasingly used to infer community assembly mechanisms (e.g. niche/neutral, stochastic/deterministic). At first glance, this approach seems reasonably intuitive—if β-diversity between communities is greater or lower than expected compared to the null model, then niche-driver processes (competition, environmental filtering) must be causing that deviation, whereas if β-diversity between communities is the same as the null (or random) expectation, community structure reflects neutral processes or stochasticity. Ecologists are increasingly trying to make inferences about the processes structuring communities using such high-level indicators. However, before we can confidently apply such approaches to field data they must at least pass a first-principles test with data generated from a known process. The most reliable way to do this is to generate artificial data using ecological models in which the process can be controlled.
We simulated metacommunities that were structured in various ways, from pure environmental filtering (i.e. each species has a unique niche) to complete neutrally (all species have identical). We considered both presence-absence and abundance-based formulations of the β-null deviation metric, and asked whether issues typical for observational data—sampling at only a single time point, presence of stochasticity in the system, incomplete sampling of the regional species pool, and changing assembly mechanism—altered the conclusions. Under ideal conditions, the β-null deviation metrics could be used to identify niche-based versus neutral processes. But, if stochasticity was present the presence-absence version failed. While the abundance-based formulation of the beta metric was more robust, it was sensitive to changes in abundance evenness, and was best interpreted through comparisons amongst multiple values, rather than in terms of its absolute value. These limitations are not un-typical of null models of all kinds and hopefully recognition of them will help direct future usage of this measure and development of other null models.
Caroline M. Tucker