Integrating species interactions, climate, and spatial effects into species distribution models with breeding birds across the United States

Background/Question/Methods

Climate change is altering the environmental conditions that determine the distributions and abundances of species, and no-analog communities are expected to form as a result. Across taxa, significant distribution changes and local extinctions have already been documented. Models predicting the responses of species to climate change have relied on a static, single-species environmental niche or “climate envelope” approach that lacks biotic interactions. Biotic interactions are fundamental to species viability and thus have strong effects on patterns of species abundance and geographic distribution. Hence, models incorporating these important biotic interactions may provide more robust predictions about the effects of climate change on species distributions and abundances.

We apply a multivariate Bayesian spatial model to predict the distributions of breeding birds with point-level data from the US Breeding Bird Survey. We explicitly incorporate interactions among species through a species interaction matrix, account for spatial effects through spatial covariance matrices, and account for observation error in the data. As co-occurring species range in their interaction strength, we compared the performance of models with species known to interact (“interaction models”) to those without strong interactions (“co-occurring models”). We further controlled for landuse change and habitat type to evaluate the influence of species interactions on the distributions of species. For a given set of co-occurring or interacting species, we ran a suite of models: 1) with just climate predictors; 2) with just interaction matrices; 3) and with climate and interaction matrices.  These models were compared with DIC. Across communities, all co-occurring models were compared with all interaction models to evaluate support for including species interactions. 

Results/Conclusions

Incorporating species interactions improved model performance overall, though the influence varied by species. We found mixed results when comparing interacting versus co-occurring species. In some cases models with interacting species performed better than those with co-occurring species. Models with only climate predictors consistently ranked poorly when compared with models incorporating the interaction or co-occurrence of other species in addition to climate. Our results suggest that environment and climate variables alone are not sufficient for anticipating the effects of changing climate on breeding birds because they are not able to solely predict the distribution of bird abundances. This assessment and our other current research show that incorporating biotic interactions into the predictive modeling framework is a valuable step towards anticipating the effects of climate change on species and their ecological communities.