COS 52-4
Using co-occurrence patterns within a Bayesian Joint Species Distribution Model to disentangle the effects of climate and biotic interactions on the distributions of Rocky Mountain trees

Tuesday, August 11, 2015: 2:30 PM
339, Baltimore Convention Center
Paige E. Copenhaver, Department of Botany, Program in Ecology, University of Wyoming, Laramie, WY
David M Bell, Pacific Northwest Research Station, U.S. Forest Service, Corvallis, OR
Daniel B. Tinker, Botany, University of Wyoming, Laramie, WY

The distributions of species are generally recognized to be driven by a combination of climatic, biotic, and historical factors, yet traditional niche-modeling approaches are limited in their ability to quantify the relative influences of each of these factors. A more precise understanding of how these factors contribute to species distribution patterns and how their relative influences vary across species is necessary to meet the need for accurate predictions of species and community responses to future environmental conditions. Bayesian Joint Species Distribution Models (JSDMs) offer a new approach to quantifying the contributions of climatic factors and biotic interactions to species distributions. In contrast to other community-based modeling approaches, JSDMs use a multivariate response distribution to characterize species co-occurrence, and model covariance may be used to assess correlation attributable to abiotic covariates as well as residual correlation, which may suggest biotic interactions when environmental envelopes are well described. Here, we apply a JSDM to occurrence data of 13 canopy tree species across the Rocky Mountain region. We predict species distributions using variables representing seasonal and annual temperature and precipitation and compare the relative strengths of environmental correlation to residual correlation to quantify the influences of climate and biotic interactions on distribution patterns.


Model performance was generally high for all species, with discrimination statistics indicating better-than-random performance (Kappa: 0.25-0.68; TSS: 0.46-0.78).  Including species co-occurrence in the model improved model fit to occurrence data over using only environmental covariates to predict occurrence probability.  For all species, environmental correlation was greater than residual correlation, suggesting that climate is acting as the dominant driver of broad-scale distribution patterns of canopy trees in this region. Residual correlation was notable for several species, and was generally greater for subalpine species than for species occupying lower elevations. Both positive and negative residual correlation were observed, indicating that both competition and facilitation may contribute to distribution patterns. Overall, these results indicate that this group of  tree species responds strongly to climatic conditions, but that, in some cases, interspecific interactions also have notable effects on distribution patterns. Our findings suggest that models that incorporate species co-occurrence, such as the JSDM presented here, may hold promise for producing improved forecasts of broad-scale species responses to climate change.