Food web aggregation: effects on key positions

Food webs are often simulated dynamically to explore how trophic interactions influence resource and consumer abundances. As large trophic networks cannot be simulated in their original size – it would be too computationally expansive – they are shrunk by aggregating species together. However, key species may get lumped during this process, masking their unique role in their ecosystem. Therefore, a more systematic understanding of the aggregation effects on key positions is needed. Here, we study how six aggregation methods change 24 importance indices used to find key species in food webs. Our work was carried out on 76 aquatic food webs from the Ecopath with Ecosim database (EcoBase). The aggregation methods we considered were: 1) hierarchical clustering with the Jaccard index; 2) hierarchical clustering with the REGE index; 3) clustering within classic food web modules, which we refer to as ‘density-based' modules; 4) clustering within ‘predator-based modules' in which species fed on the same preys; 5) clustering within ‘prey-based modules' in which species are fed upon by the same predators; and 6) clustering within ‘groups' in which species share the same probability to interact with other groups. Hierarchical clustering with the REGE index produced the best results. Therefore, we recommend using it if we were interested in maintaining the identity of key species. The other algorithms could also be used to study specific network processes. However, we need to consider the bias they produce when masking important species.