Null modelling methods have long been used to identify non-random patterns of species co-occurrence. However, clarity on the underlying ecological mechanisms is often limited because multiple mechanisms can produce similar co-occurrence patterns. This challenge is especially apparent when occurrence data span a heterogeneous landscape or a strong environmental gradient. In these cases, it can be particularly difficult to discriminate between the effects of environmental filtering and biotic interactions. Recent approaches have paired a null model pairwise co-occurrence approach with follow-up testing, using site environmental characteristics, to identify the likely driver. We developed a similar logical, hypothesis-testing framework but instead leveraged species trait similarity and hierarchical spatial sampling. Inferences of biotic interactions were based on similarity of diet and body size whereas habitat affinity and geographic range were used for environmental filtering. Our objective was to develop a trait-based framework that offers a general and versatile approach for inferring mechanisms from pairwise co-occurrence patterns of species across heterogeneous sites. We demonstrate our framework by analyzing the co-occurrence of small mammals over elevation in three independent mountain ranges in the Great Basin of the western United States.
Results/Conclusions
Our case study of small mammal assemblages along Great Basin elevational gradients demonstrates the effectiveness of our framework. We identified 52 non-random associations out of the 1,292 species pairs analyzed. Of these, 36 were aggregated and 16 were segregated. Using co-occurrence patterns alone, only five of the 52 significant pairwise associations could be attributed to a single mechanism. For the remaining 47 pairs, parsimonious mechanistic explanations were reached following functional trait-based hypothesis testing, including cases of both mechanisms of interest (environmental filtering and biotic interactions). Environmental filtering was inferred to be the mechanism responsible for all but four of the 52 significant pairs. The remaining four associations (two segregated and two aggregated) were consistent with expectations under biotic interactions. In addition to analyzing individual pairs, we used binomial tests of observed versus expected totals of intra- and inter-guild pairs to determine assemblage-wide deviations from random community structure for each mountain range. Signatures of environmental filtering were consistent across mountain ranges and scales. Despite differences in species composition and significant pairs among data sets, our approach revealed consistent mechanistic conclusions. Thus, it emphasizes the value of trait-based methods to co-occurrence and community assembly.