Decoding information in multilayer ecological networks: The keystone species case

Abstract

The construction of a network capturing the topological structure linked to the interactions among species and the analysis of its properties constitutes a clarifying way to understand the functioning of an ecosystem at different scales of analysis. Here, we present a novel systematic procedure to profit from the enhanced information derived from considering its multiple levels and apply it to analyse the presence of keystone species. The proposed method presents a way to unveil the information stored in a network by comparing it to some randomised modification of itself. The randomising of the original network is done by swapping a controlled number of links while preserving the degree of the nodes. Then, we compare the modularity value of the original network with the randomised counterparts, which gives us a measure of the amount of relevant information stored in the first one. Once we have verified that the modularity value is meaningful, we use it to perform a community analysis and a characterisation of other topological properties in order to identify keystone species. We applied this method to a pollinator–plant–herbivore trophic network as a case study and we found that (a) the comparison between the modularity of the original and the randomised networks is a suitable tool to detect relevant information; and (b) identifying keystone species yields different results in bipartite networks from the ones obtained in networks of more than two trophic levels. We also analysed the effect of eliminating selected species from the system on the cohesion of the network. The selection of these species was made according to the centralities values, such as degree and betweenness, of the corresponding nodes. Our findings show that our analysis, mainly based on the measure of modularity is a reliable tool to characterise ecological networks. Additionally, we argue that since degree and betweenness are not always correlated, it is more reliable to measure both in an attempt to detect keystone species. The methodology proposed here to identify keystone species can be applied to other ecological networks currently available in the literature.