Food webs are complex objects that depict who eats whom in a natural system. Several models have tried to capture their architecture, often implicitly implying latent traits, i.e., non-measured characteristics. Here we adopt a statistical approach where we use log body-size ratio of prey on predator as explanatory variable, using a dataset of 12 highly resolved food webs. Body size typically predicts 20% of trophic links.
Results/Conclusions
We introduce a method to explore the unexplained part of the structure, which is based on the estimation of latent traits. For each species, one foraging and one vulnerability trait are computed from the data, which dramatically improves the percentage of correctly predicted links (now 73% on average). These traits are useful to interpret the hidden structure of the food webs and can be related to external information (e.g. phylogeny, micro-habitat structure, or camouflage). We use taxonomic information on the species to show that phylogeny is in general closely linked to the latent traits.
