A new tool to improve the estimates of interaction rewiring considering the whole community composition

Abstract

Abstract Understanding temporal dynamics in ecological networks is crucial to predict their capability to cope with global changes. Despite this, proper quantification of network dynamics still remains a challenge. Temporal dynamics are typically studied using data of interaction networks over time, through the evaluation of interaction turnover and its two components: changes related to species turnover (species gains and losses) or rewiring (switching partners among the set of species shared over time). However, with this approach based exclusively on network data, dynamics are computed similarly for species that are truly missing from the community at a given temporal period, and for species occurring in the community but that do not interact with any other. This might lead to an underestimation of the real extent of rewiring, while overestimating the species turnover component of interaction turnover. We used data on 20 plant–pollinator communities to calculate interaction turnover components accounting also for the species that occurred in the communities at different temporal periods but did not appear in some of the temporal interaction networks (non‐interacting species) and then compared these estimates with conventional ones. Besides, we used empirical data and simulations to evaluate the extent to which dynamics estimates were affected by sampling effort when including and excluding non‐interacting species. As expected, disregarding the non‐interacting species that occurred in the communities at different temporal periods led to the underestimation of rewiring and the overestimation of species turnover as components of interaction turnover. Effect size was moderate when independent pollinator data were included, and large when including plants or both trophic levels. Simulations indicated that, in general, considering the non‐interacting species reduced biases at the time of identifying changes due to the different interaction turnover components. Accounting for non‐interacting species was particularly important to reduce bias when sampling effort was low and when dynamics were calculated seasonally. Despite sampling effort effects, phenology was the main determinant of species' rewiring frequencies. Our approach contributes to reducing biases and improving the estimates of interaction flexibility in networks, which are necessary to comprehend the response of communities in the face of global change.