Tackling knowledge gaps about food webs with trait-based models

Background/Question/Methods

Food webs are central in ecology and conservation, shaping species coexistence, the functional roles of species, community stability and functioning. In parallel, human-induced disturbances are rewiring food webs driving them to become more simplified and homogeneous. Despite the importance and threats on food webs, knowledge about trophic relationships is lacking for most ecosystems and taxa. To address this knowledge gap, we need models that can predict both pairwise interactions and the food web structure with relatively sparse data. One family of these interaction models uses the functional match between the traits of predators and prey to predict feasible trophic interactions. However, it is unclear if these models are over-simplistic to make reliable predictions or too complex to be calibrated with limited data and make predictions in new systems. We designed a relatively simple model that uses species functional traits to predict all potential interactions between European terrestrial vertebrates. Our Bayesian generalized linear model uses predictors describing the vulnerability of the prey, the foraging ability of the predator and the trait-match of interacting species. We addressed two main questions: (1) How much data are needed to calibrate a useful model? (2) What are the taxonomic and spatial biases of model predictions.

Results/Conclusions

Overall, we estimated the entire European food web relatively well (AUC > 0.9) with models calibrated with a small fraction of all interactions (100 out of 71k). On average, predators were better predicted than prey, especially for some clades (e.g., storks and fowls). Also, the number of trophic interactions was consistently overestimated by 2 to 4 times. These results suggest that simplistic trait-based models can be used to create preliminary food webs even in regions where data are scarce, and help initiate targeted sampling or expert elicitation. We further explore how well our model can be used to transfer knowledge to entirely new ecosystems (e.g., African savanna, High Arctic) and the potential of updating predictions by constraining the structure of the predicted food web. Along with the other recent development in species interaction and ecological network predictions, these results show the potential of predictive models to allow the investigation of how food webs and species interactions vary in space and time, and to forecast the consequences of global changes on the composition and structure of ecosystems.