Quantifying the robustness of the food web to a perturbation
To evaluate the robustness of the network to taxa’s extinction, we simulated primary extinctions of taxa (i.e. removal of taxa from the network) according to various removal sequences and monitored the values of connectance and accumulated secondary extinctions (i.e. the extinction caused by the removal of all the prey of one taxon). Connectance is a good estimate of community sensitivity to a perturbation, and large values of connectance favour the spread of a perturbation (Martinez 1992; Delmas et al. 2019). On the other hand, secondary extinctions inform on robustness of the network, and is negatively correlated with it (Dunne et al. 2002). These simulations of extinctions were done by removing taxa in five different orders: (1) Sensitivity, from the highest to the lowest sensitivity score, (2) Centrality, from taxa with the highest to the lowest eigenvector centrality values, (3) Exposure, from taxa with the highest to the lowest exposure to fishing pressure, (4) Prey, from taxa with the highest to the lowest number of prey and (5) Predator, from taxa with the highest to the lowest number of predators. The connectance and accumulated secondary extinctions generated in each of these 5 removal scenarios were compared to a random mode in which taxa are randomly selected and removed from the network. This random removal was iterated 500 times. We followed the same procedure for modularity and nestedness (Fig. S3). To compare the robustness of this network with other networks, we computed the R50, defined as the proportion of taxa that have to be removed to result in a total taxa‘s loss ≥50% of the species in the original web (Dunne et al. 2004).
All analyses were conducted in R 4.0.2 (R Core Team 2020). The secondary extinction analysis was performed with modified functions from the NetworkExtinction package (Corcoran-Barrios et al. 2019).