An increasingly popular approach to estimating species interaction networks is through the analysis of species pairwise co-occurrence patterns. Species that spatially or temporally co-occur at a site more (or less) often than expected by chance are thought to have positive (or negative) associations. The definition of an “association” varies widely among studies, but they are often directly referred to as species interactions, or as a proxy for species interactions. A negative association might indicate a trophic interaction or competition, while a positive association might indicate facilitation or mutualism. However, species interactions have a long history of experimental quantification, and are associated with a range of technical and mathematical definitions that facilitate the translation of experimental results to theoretical models. In this study, we quantify species association networks for a rocky intertidal ecosystem and compare them to a dataset of species interaction experiments from the same ecosystem.
Results/Conclusions
We assessed nine different statistical methods for their ability to estimate species associations, ranging from traditional metrics of spatial checkerboards to new machine learning and network methods for causal inference. Each method estimated a fundamentally different set of interacting species, suggesting that results from this diverse body of literature are not comparable. Further, no co-occurrence method matched estimated associations with experimental interactions at a rate better than 6%. In other words, spatial associations estimate a different process than species interaction experiments. A series of sensitivity analyses reveal that this result is robust to spatial grain size of observation, interaction type, guild, and taxon aggregation. A mechanistic community assembly simulation reveals the conditions under which species interactions may be estimated from species co-occurrence, in the presence of strong environmental gradients, dispersal, and trophic interactions. We suggest a series of steps forward, integrating the longstanding body of literature on species co-occurrence with modern trait and phylogenetic assembly theory, toward more robust pattern-process inference in community ecology.