Evaluating the congruence between distance-based and Bayesian multivariate models to predict oyster-reef community response to environmental stressors
A key objective of community ecology is to identify how processes and environmental constraints generate spatial patterns in community composition. However, the majority of approaches to answering this key question have ignored the fact that species are often jointly distributed, and species abundances can be constrained by not only physical conditions but also interactions such as competition. Estuarine habitats, in particular, are subject to a wide array of physical stressors, with changes in salinity and temperature being of paramount importance. Fluctuations in salinity can vary across an estuary as well as at a location within a single day due to tidal influences, subjecting oyster reef inhabitants to a wide range of physical conditions. The importance of these abiotic factors have been well-studied using distance-based analyses, although this may belie the role of biological interactions in structuring communities. We present a case study that uses a combination of multivariate approaches; particularly distance-based analyses and models, as well as a Bayesian joint species distribution model (JSDM), to evaluate the congruence of results between different statistical techniques. In particular, we apply these models to benthic oyster-reef communities to determine how environmental factors and biotic drivers co-regulate communities.
Both the distance-based and Bayesian multivariate approaches found significant effects of salinity and temperature on oyster-reef community composition. Joint biplots and correlation analyses indicated that of the measured environmental factors (q=6) temperature and the range of salinity experienced at a given site most strongly influence how samples ordinated. Whereas distance-based models provide a useful way to readily assess and visualize how overall species compositions of samples vary and overlap in response to salinity regime and temperature, its inability to handle joint absences as well as non-continuous covariates severely limited the analysis. The predicted values from Bayesian JSDM further indicated that salinity could be used to identify which species would be most sensitive to changes in environmental condition, and found that species compositions were similar when predicted via the empirical covariance matrix or the environmental response. Bayesian JSDM was more flexible in handling both zero-inflated data and multiple data types, but the model did not perform well when predicting high-abundance observations. Thus, the combined predictive power of both a classical and Bayesian approach provides a more holistic method of identifying which environmental factors are most influential in determining the distribution and abundance of co-occurring species.