Dominique Caron , Ulrich Brose, Miguel Lurgi, Guillaume Blanchet, Dominique Gravel, Laura J. Pollock

Aim: Trophic interactions are central to our understanding of essential ecosystem functions as well as their stability. Predicting these interactions has become increasingly common due to the lack of empirical data on trophic interactions for most taxa in most ecosystems. We aim to determine whether and how accurately we can extrapolate to new communities both in terms of pairwise predator-prey interactions and higher level food web attributes (i.e., species position, food web-level properties).

Location: Canada, Europe, Tanzania.

Time period: Current.

Major taxa studied: Terrestrial vertebrates

Methods: We use a trait-based model of pairwise trophic interactions, calibrated independently on four different terrestrial vertebrate food webs (Canadian tundra, Serengeti, alpine south-eastern Pyrenees, and entire Europe) and assess the ability of each calibrated instance of the model to predict alternative food webs. We test how well predictions recover individual predator-prey interactions as well as higher level food web properties across geographical locations.

Results: We find that, given enough phylogenetic and environmental similarities between food webs, trait-based models predict most interactions and their absence correctly (AUC > 0.82), even across highly contrasting environments. However, network metrics were less well-predicted than single interactions by our models. Predicted food webs were more connected, less modular, and had higher mean trophic levels than observed.

Main conclusions: Theory predicts that the variability observed in food webs can be explained by differences in trait distributions and trait-matching relationships. Trait-based models can predict potential interactions amongst species in an ecosystem when calibrated using food web data from reasonably similar ecosystems. This suggests that food webs vary spatially primarily through changes in trait distributions. These models, however, are less good at predicting system level food web properties. We thus highlight the need for methodological advances to simultaneously address trophic interactions and the structure of food webs across time and space.