The influence of pollinator traits on algorithm-based module detection in the H.J. Andrews plant-pollinator network

Modularity is an important structural pattern in plant-pollinator networks that, when identified and examined, can lead to better understanding of drivers behind plant-pollinator interactions and contributions to resilience. Most module-detection algorithms rely on network structure alone. However, Newman and Clauset’s (2016) annotated network algorithm is able to leverage additional information about the nodes to aid in module detection and division selection, making it a powerful tool to examine potential causes behind module formation in a network. This study uses the Newman/Clauset approach to test how pollinator functional traits are related to module divisions found by the algorithm in a plant-pollinator network in the H. J. Andrews Experimental Forest. Correlations between pollinator trait metadata and the module divisions found by the algorithm were variable and not particularly high for any of the subsets and traits tested. These results imply that even though small or specialist modules may be detectable using trait metadata, most structural patterns are likely driven by other factors such as generalist interactions and species abundance in the H. J. Andrews network. Introduction Analyzing higher-level ecological network structure is one way to reveal patterns in interactions among a wide range of species across space and time in a network. An important pattern is modularity, which occurs when groups of species associate more with each other than with others in the network. In a plant pollinator network, this would mean that a certain set of flowers is visited nonrandomly by a certain set of pollinators with lower numbers of interactions between sets. This pattern has been observed in some plant-pollinator networks and is thought to provide resilience to the overall network by limiting the effect of species loss or other disturbances to individual modules (Olesen et al. 2007, Dicks 2002). Understanding modules can also lead to evolutionary insight about the members or provide a frame for future study or conservation work. Many algorithms search for modules in a network based on structure, e.g. dense clusters with sparser connections to other clusters. In addition to detecting modules, the most powerful analysis would be able to contribute to narrowing down the factors that may be influencing those patterns. Several factors may influence modularity: habitat heterogeneity, where certain plants and pollinators are only found in certain conditions or places; taxonomic relatedness; timing of flower anthesis and pollinator flight; and shared “functional traits” that allow plants to exclude or attract certain pollinators or for pollinators to take advantage of specific flower types (Olesen et al. 2007; A. Moldenke, pers. comm.). The last point corresponds to the concept of a pollinator “syndrome” or set of traits that have evolved between plants and pollinators to be complementary, leading to specialization or exclusion of other pollinators (Dicks 2002). For example, long corolla tubes in a flower allow long-tongued pollinators to reach the nectar while excluding short-tongued pollinators. Combinations of traits also may lead to a degree of preference for a plant or pollinator that are not the result of tight evolutionary couplings. Either way, the functional traits of plants and pollinators are a compelling potential influence on interaction patterns and thus network structure. If modules are indeed influenced by something such as functional traits, providing trait information to an algorithm could enhance its ability to detect or choose between different divisions. The annotated network algorithm proposed by Newman and Clauset in “Structure and influence in annotated networks” is designed with this in mind (Newman and Clauset 2016). My study uses the annotated network approach provided by Newman and Clauset’s (2016) algorithm to examine the structure in a plant-pollinator network found in the H. J. Andrews Experimental Forest. I ask the question, How are network groupings in the Andrews influenced by pollinator traits? More specifically, are detected modules in the annotated network correlated to functional traits, and do these translate to meaningful ecological groups?