Incongruent drivers of network, species and interaction persistence in food webs

Communities can be described by species, interactions and the network of interactions which emerge from these building blocks. Networks are often summarized by diverse metrics which capture key components of network topology, for example, number of trophic levels, number of interactions per species (i.e. degree) and frequency of apparent competition modules (i.e. motifs). Network metrics are often used to predict whole community responses following a perturbation. Understanding whether our predictive capabilities at the network level are maintained at other levels of organization such as species and interaction levels is critical but rarely studied. Our objective was to determine whether we can use different network metrics (e.g. modularity, species roles) to predict which networks, species and interactions will persist following a perturbation. Given the nested structure of interactions, species and networks, we hypothesized that those metrics which were correlated with network persistence would be the same metrics correlated with which species and interactions persist and that the direction of these correlations would be maintained across organizational levels (i.e. network, species and interactions). We used numerical simulations of a dynamic food web model and model selection to test the relationships between network metrics (e.g. degree, modularity, motif profiles) and network, species and interaction-level persistence. We found that out degree and frequency of both apparent and exploitative competition were the best predictors of species persistence, however all metrics were weak predictors of species persistence. This demonstrates that metrics which predict network persistence are not always the best predictors of species and interaction level persistence. Additionally, metrics can work in opposition depending on the organizational scope examined; for example interaction strength is positively correlated with interaction persistence, but negatively correlated with species persistence. Multiscale analysis of network metrics such as this one, may provide critical insight for advancing network-based environmental management in the Anthropocene.