Combining groups and niches in a model of food web structure

Background/Question/Methods

Food webs, networks of feeding links in an ecosystem, provide a valuable abstraction for understanding ecological communities. Two basic ideas have underpinned much of the discourse on food-web structure and their modelling: (1) the organization of species into a low-dimensional niche space of feeding interactions; and (2) the organization of the food web into compartments or groups. The use of probabilistic generative models has made it increasingly possible to directly test these and other hypotheses against food web data sets.

In previous studies, we proposed a group-based model and applied it to describe the overall architecture of a plant-mammal food web from the Serengeti ecosystem of Tanzania in terms of functional groups within trophic levels. In this study, we ask whether niche-based and group-based models can be combined to form a more effective description of the Serengeti network structure, and whether the niche and group patterns reflect evolutionary relationships. In order to do this, we employ a model that combines the coarse-grained structure of the group model with the constraints of a low-dimensional niche space using a hierarchical Bayesian formulation.

Results/Conclusions

We find that, among several model variants compared, a two-dimensional niche-based group model provided the best fit to the Serengeti food web. The identified groups were concordant with the previous analysis based on groups alone, but the clustering of feeding links in two-dimensional niche space improved the model fit substantially. In other words, a group-based model captured the high-level organization of the food web, but a detailed two-dimensional niche structure improved the fit, provided that groups were used to constrain the dimensionality of parameter space. Furthermore, the group structure and niche-space parameters, although inferred directly from the link data alone, reflect the biological traits of species as seen through evolutionary relationships.