Community ecology is a highly complex field in which very different processes can lead to similar patterns, therefore the processes governing community structure and assembly are difficult to infer. Though metacommunity theory has helped with this endeavor, it is still challenging to connect real world community diversity patterns with the processes that created them. We show that the unified theory of ecological communities suggested by Mark Vellend gives ecologists the tools necessary to infer process from patterns in community data by utilizing the Approximate Bayesian Computation (ABC) framework created for population genetics. Specifically, we adapted a multiallelic Moran model with selection and migration to simulate metacommunity dynamics. These simulations were then used to answer two questions: 1) Can population genetics models create community patterns expected under metacommunity theory, and 2) can Approximate Bayesian Computation be used to infer the metacommunity archetype working within communities? We used variation partitioning and null model analyses to see if simulations with different levels of selection, dispersal limitation, and spatial landscape type followed metacommunity theory. We then analyzed if ABC could accurately distinguish between Neutral and Species Sorting models and infer parameters using a variety of summary statistics common to community ecology.
Results/Conclusions
Variation partitioning showed that models under a Species Sorting archetype had significantly higher proportions of variation explained by environmental variables whereas Neutral models typically had higher proportions explained by space, as predicted by metacommunity theory. Additionally, models in which species were dispersal limited had higher proportions explained by space than did models which were non-dispersal limited. Fisher’s exact tests on results from coherence, turnover, and boundary clumping null models indicated that very similar patterns could be produced by both Species Sorting and Neutral simulations, suggesting that these null model methods are only indicative of pattern and not process. Finally, ABC was able to accurately select Neutral from Species Sorting and dispersal limited from non-dispersal limited models when conducting model selection on simulations. Furthermore, preliminary results suggest that dispersal (maximum dispersal distance, and probability of migrating) and community size parameters can be accurately predicted using ABC methods depending on the model type and the summary statistics used. These results indicate that an ABC framework has the potential to infer the processes governing community structure and assembly from real world community data, and therefore can be useful for management, conservation, and general ecological research.