Understanding the causes of ecosystem function is a central goal of ecology and of strong pragmatic importance, given widespread human-caused changes in the diversity and functioning of ecosystems. In part motivated by these anthropogenic impacts, substantial attention has been paid to the relationship between biodiversity and ecosystem function. However, it has been difficult to correctly attribute the causes of different measures of community structure to ecosystem function (e.g., functional traits, species diversity). These measures are correlated, either mathematically (i.e., calculated variables) or due to having shared, unmeasured causes. Here, we address these problems using graphical causal modeling (GCM) to derive a hypothesized causal model of community structure and its consequences for ecosystem function. GCM is multivariate network of causal relationships that gives a complete factorization of the joint probability distribution of the variables. Thus, it can be used to determine the necessary and sufficient conditions for identifiability of specific causal effects.
Results/Conclusions
The causal network relating community structure to ecosystem function, shows that measures of diversity and measures of ecosystem function are not causally connected, but rather correlated through the shared causal dependence on presence and abundance of the species in the community. Further, the relationship between abundance and function is mediated by the community-level composites of functional traits such as the mean value of a trait or variation in a trait or multiple traits. Using this framework to inform statistical analysis shows that, since the full set of traits driving function is unknown, and thus likely missing, species abundance measures must be included as predictors in the models of relationships between functional traits and function to obtain unbiased estimates. Finally, the GCM makes a number of testable predictions, which if refuted by data, have ecological interpretations that can help identify missing variables. For example, the structure of the causal relationship predicts that measures of species diversity and measures of ecological function should be independent given measures of community-level composites of traits. If a residual relationship between diversity and function is found, it suggests the presence of a missing variable specifically relating to trait functional diversity (as opposed to the community-weighted mean value of a particular missing trait). Thus, this work provides a conceptual framework for both directing research programs for the discovery of trait-function relationships in ecological systems as well as informing analyses to resolve the decades-long debate about the relationship between diversity and function.