OOS 54-8
Incorporating known competitive and facilitiative interactions into SDMs for plant species: Evaluating patterns across spatial scales
Understanding whether and how local-scale processes influence broad-scale patterns has long been a central question in ecology. As species distribution models have been continually evolving with increasing pressure to include community and demographic processes, we were driven to ask the following: Are local species interactions reflected in large scale co-distributional patterns? We conducted a meta-analysis of 207 pairs of species in which competitive or facilitative interactions were measured experimentally at the plot scale. We then modelled each of these paired species’ distributions using generalized linear models (focal species ~ climate + interacting species). We assessed whether model predictions were improved by including interacting species as a co-variable, and more informatively, whether the magnitude of this improvement was correlated with the effect size of the interaction at the plot scale.
Results/Conclusions
We found that the effect sizes of biotic interactions at the plot scale were not correlated with the magnitudes of the improvement in model predictive accuracies (ΔAUC) at the broad scale. However, the majority of models’ predictive accuracies (AUC) were improved by simply including a known interacting species as a co-variable. This improvement was noticeable for species pairs that were either known competitors or facilitators at the plot scale. While we attempted to account for environmental factors, the question remains whether these models are improved solely as a function of including a species with shared environmental preferences. As questions about species interactions and scale are becoming increasingly relevant as researchers and land managers debate the value of including biotic interactions in species distribution modeling, our data suggest that the direction and magnitude of plot scale biotic interactions may not be represented in large scale distributional patterns; however, the inclusion of interacting species may improve each model’s predictive accuracy.