No strong evidence that modularity, specialization or nestedness are linked to seasonal climatic variability in bipartite networks

Abstract

Aim: Given the influence of seasonality on most ecological systems, an emerging research area attempts to understand how community network structure is shaped by seasonal climatic variations. To do so, most researchers conduct their analyses using open networks due to the high cost associated with constructing their own community networks. However, unwanted structural differences from the unique sampling and construction methods used to create each open network likely make comparing these networks a difficult task. Here, with the largest set of open bipartite networks collected to date, we test whether seasonal climatic variations explain network structure while additionally accounting for construction/sampling differences between networks.Location: Trying to approach global. Time period: Contemporary.Major taxa studied: Plants and animals. Methods:Using 723 open bipartite networks, we test whether temperature and/or precipitation seasonality explains (un)weighted metrics of nestedness, modularity or specialization across plant-pollinator, seed-dispersal, plant-ant, host-parasite or plant-herbivore systems.Results: Generally, seasonality only weakly explained network structure: at most 16% of the variation in weighted metrics and 5% of the variation in unweighted metrics.Instead, a control for sampling bias in networks, sampling intensity, often better explained many of the network structural metrics. When limiting our analyses to only intensely sampled networks, however, about 33% of the variation in weighted modularity and specialization was explained by seasonality, but only in plant-pollinator networks.Main conclusions: Altogether, we do not find strong evidence that seasonality explains network structure. Our study also highlights the large amount of structural differences in open networks, likely from the many different sampling and network construction techniques adopted by researchers when constructing networks. Hence, a definitive test for the relationship between network structure and seasonality across large spatial extents will require a dataset free from sampling and other biases, where | 2511 BRIMACOMBE et al.

Shortcomings of reusing species interaction networks created by different sets of researchers

“…Ecologists commonly reuse species interaction networks created by different sets of researchers to test how ecological and environmental processes shape network topology across space and time [ 8 , 9 , 26 ]. However, unwanted topological differences as a result of the different ways in which researchers translate ecological communities into networks could inhibit their commensurability [ 10 , 23 , 24 ]. When assessing the degree of topological heterogeneity, i.e., the total amount of topological differences between a group of networks, we find that species interaction networks created by different sets of researchers are extremely topologically heterogeneous—about twice the amount than the next most heterogeneous network domain tested—and that this large heterogeneity is linked to the publication source of each network.…”

“…Except for sports networks that we constructed for this paper and whose data were obtained from Lahman (2021) [ 48 ], www.basketball-reference.com , and www.hockey-reference.com , all non-ecological networks were obtained from Michalska-Smith and Allesina (2019) [ 13 ]. Of the 723 species interaction networks used in this analysis (obtained from Brimacombe and colleagues (2022) [ 10 ]), 10 were ant–plant networks, 97 were host–parasite networks, 41 were plant–herbivore networks, 298 were plant–pollinator networks, and 277 were seed–dispersal networks. All networks that were included in our analysis were unweighted (i.e., interactions between nodes were binary).…”

“…com, and www.hockey-reference.com, all non-ecological networks were obtained from Michalska-Smith and Allesina (2019) [13]. Of the 723 species interaction networks used in this analysis (obtained from Brimacombe and colleagues (2022) [10]), 10 were ant-plant networks,…”

“…Consequently, it appears that publication has an even greater impact on the topology of species interaction networks than biological processes alone. This may occur since researchers of a given publication generally construct networks under parsimonious conditions [10], for example, by observing and characterizing ecological communities across the same time duration (e.g., Trøjelsgaard and colleagues (2015) [62]) or by classifying nodes across networks using the same protocol (e.g., Pereira Martins and colleagues (2020) [63]), and thus inadvertently control for many sources of topological heterogeneity. Of course, biological effects are likely influencing the topology of all PLOS BIOLOGY networks but they can be more easily obscured when analyzing networks across publications.…”

“…Due to the effort needed to observe species and their interactions in situ, creating species interaction networks requires a tremendous amount of resources for their adequate construction [6][7][8]. Given this requisite effort, instead of creating their own networks, ecologists often reuse available species interaction networks which happen to be created by different sets of researchers, to test their own ecological hypotheses [9,10]. These available species interaction networks have therefore been used in many ecological studies including when determining the complexity of networks [11], identifying common topological properties across networks [12,13], and evaluating how network topology is shaped by species traits [14], environmental factors/space [7,[15][16][17], and time [18][19][20].…”