Interaction generalisation and demographic feedbacks drive the resilience of plant-insect networks to extinctions.

Understanding the processes driving ecological resilience, that is the extent to which systems retain their structure while absorbing perturbations, is a central challenge for theoretical and applied ecologists. Plant-insect assemblages are well-suited for the study of ecological resilience as they are species-rich and encompass a variety of ecological interactions that correspond to essential ecosystem functions. Mechanisms affecting community response to perturbations depend on both the natural history and structure of ecological interactions. Natural history attributes of the interspecific interactions, for example whether they are mutualistic or antagonistic, may affect the ecological resilience by controlling the demographic feedbacks driving ecological dynamics at the community level. Interaction generalisation may also affect resilience, by defining opportunities for interaction rewiring, the extent to which species are able to switch interactions in fluctuating environments. These natural history attributes may also interact with network structure to affect ecological resilience. Using adaptive network models, we investigated the resilience of plant-pollinator and plant-herbivore networks to species loss. We specifically investigated how fundamental natural history differences between these systems, namely the demographic consequences of the interaction and their level of generalisation-mediating rewiring opportunities-affect the resilience of dynamic ecological networks to extinctions. We also create a general benchmark for the effect of network structure on resilience simulating extinctions on theoretical networks with controlled structures. When network structure was static, pollination networks were less resilient than herbivory networks; this is related to their high levels of nestedness and the reciprocally positive feedbacks that define mutualisms, which made co-extinction cascades more likely and longer in plant-pollinator assemblages. When considering interaction rewiring, the high generalisation and the structure of pollination networks boosted their resilience to extinctions, which approached those of herbivory networks. Simulation results using theoretical networks suggested that the empirical structure of herbivory networks may protect them from collapse. Elucidating the ecological and evolutionary processes driving interaction rewiring is key to understanding the resilience of plant-insect assemblages. Accounting for rewiring requires ecologists to combine natural history with network models that incorporate feedbacks between species abundances, traits and interactions. This combination will elucidate how perturbations propagate at community level, reshaping biodiversity structure and ecosystem functions.