These studies reveal that while anthropogenic impacts often reduce species richness in networks as expected, properties of networks do not always respond to the loss of species as theory would predict ; such deviations from theoretical predictions call into question the robustness often attributed to plant-pollinator interaction networks . Thus, while general patterns have been uncovered regarding the overall structure of networks under relatively natural settings, empirical data are still needed to build consensus regarding how species loss impacts network structure and function . Here, we examine the structure of twelve plant-pollinator interaction networks in a species-rich ecosystem where we have previously documented a marked reduction in pollinator diversity associated with habitat fragmentation . Our system is also numerically dominated by the non-native western honey bee , providing the opportunity to assess the role of this widespread generalist species in networks in its introduced range. Our data allow us to address two hypotheses: pollinator diversity loss impacts the structural properties of networks so as to erode network robustness or functionality, and network structure is influenced by the introduction of the numerically dominant, generalized honey bee. Although there are now numerous metrics that describe different properties of networks , large plastic pots our test of the first hypothesis focuses on four metrics that are often reported and whose relationships with ecological function are mechanistically well-understood: nestedness, niche overlap, number of links per species, and generalization.
A network with a nested interaction structure is organized around a set of numerically abundant, generalized pollinator and plant taxa that are well connected and presumably provide a disproportionate amount of pollination services and food resources . Given previous findings that rare and specialized species, which tend to enhance nestedness , are more prone to local extinction in fragmented landscapes , we predict that pollinator diversity loss should decrease network nestedness. Low niche overlap may indicate effective resource partitioning among putative competitors or signify the presence of functionally diverse assemblages capable of supporting a high diversity of partners . Low niche overlap may also indicate a lack of functional redundancy, and therefore, a greater vulnerability of ecosystem function to the loss of species . We predict that pollinator diversity loss should increase niche overlap as more specialized species are removed, leaving behind more generalized pollinator species that visit the same set of generalized plant species. The number of links per species measures the degree to which possible interactions between plants and pollinators are realized and serves as a metric of the robustness of ecosystem function. Previous research has shown that links tend to be lost at a faster rate than species when habitat area is reduced ; therefore, we also predict that the number of links per species will decrease with pollinator diversity loss in our system.Generalization refers to the degree to which plant and pollinator species interact with many, rather than few, partners.
We expect pollinator diversity loss to increase network-level generalization via two distinct mechanisms. First, as discussed above, the preferential removal of specialists in typical scenarios of biodiversity loss should yield networks composed of more generalized species. Second, studies in both natural ecosystems and experimental mesocosms have shown that generalist pollinators may exhibit reduced selectivity when pollinator diversity is reduced, presumably because exploitative competition forces species to focus foraging efforts on only the food resources on which they forage with higher efficiency relative to their competitors . In our system, the non-native western honey bee reaches high abundances, largely due to the proliferation of feral, Africanized colonies . Generalists form the core of interaction networks , thus, the addition of an abundant super-generalist should have a disproportionate impact on network architecture. Specifically, we predict that honey bees should contribute to enhanced network nestedness, greater number of links per species, increased niche overlap, and increased generalization. Like many other studies taking place in fragmented or otherwise modified landscapes , our networks span a gradient of pollinator diversity. However, unlike most of these studies, we selected habitat fragments where plant communities have remained relatively intact, thus enabling us to study the effect of pollinator diversity loss in isolation from the effects of eroding entire networks. Additionally, previous empirical research investigating the impacts of exotic species on plant-pollinator interactions have mostly focused on non-native plants rather than pollinators .
Thus, our study also provides valuable data regarding the degree to which network structure responds to pollinator diversity loss in a system dominated by a non-native generalist .Between March and July of 2015 and 2016, we documented putative pollinators as they visited flowers of a set of focal plant species naturally growing in our study plots. Here, we defined putative pollinators as flying insects belonging to the orders Hymenoptera, Diptera, Lepidoptera, and Coleoptera . Since plant species in our system often exhibit clumped distributions, we performed pollinator surveys using a timed observation method . Compared to the frequently used transect method, the timed observation method grants superior ability to investigate network properties of rarer plant species , which comprise the majority of plant diversity at our sites. During each survey, we observed a single patch of plants for ca. 60 s, counting all putative pollinators already present on the patch as well as pollinators that arrived at the patch; after this time we moved on to the next patch. Patches ranged in size from a portion of inflorescences of one plant individual for large shrubs Abrams to several hundred individuals for annual forbs Greene, and was determined by our ability to keep track of putative pollinators in our field of view. Pollinators that are identifiable to species in the field were counted ; all others were collected, pinned, and individually identified in the laboratory. We documented each pollinator individual that appeared to contact reproductive parts of flowers as a single interaction, and took care not to count the same pollinator individual multiple times in the same patch. We collected roughly one third of all non-honey bee pollinator individuals we encountered ; and in the vast majority of cases, pollinator individuals that were counted but not collected were unambiguously matched with specimens collected from the same site on the same survey day. Because our survey protocol did not consistently allow detection of minute insects under field conditions, we only documented pollinators ≥ 2 mm, which represents the minimum size at which we can reliably spot insects. In 2015, we equally divided 120 min of survey time per study plot among all focal plant species in which more than ca. 5% of the flowers were in bloom. We limited survey time to 20 min per plant species when fewer than six plant species were blooming in a given plot. In 2016, we allocated 15 min to each plant species irrespective of how many plant species were in bloom at a given study plot. While our sampling methods prioritized standardizing sampling effort among study plots in 2015 and across all plant species in 2016, all plant species at a given study plot received the same sampling effort on any given day. Additionally, study plots received community-level plant-pollinator surveys for the same number of days in each year. Given that we always allocated the same amount of time to each plant species within each plot in a given visit, tall plastic pots the structure of networks and the relative interaction strengths of plants and pollinators documented at a given plot should be roughly comparable between years. The protocols and sampling effort we employed are comparable to those in other studies that have documented floral visitation networks in detail .Using data from the floral visitation surveys, we constructed plant-pollinator interaction networks where each plant and pollinator species is represented by a node, and the number of observations recorded between each unique pair of plant and pollinator taxa serves as a proxy for the strength of ecological relationships between the two partners . We pooled all observations from each study plot in each study year together into a single network, resulting in 24 total networks across the two study years. Then, we used package bipartite in program R to calculate network metrics to test predictions regarding the effects of pollinator diversity loss and exotic pollinator introduction on network structure.
To estimate nestedness, we calculated weighted nestedness based on overlap and decreasing fill , a metric that identifies truly nested patterns more consistently and precisely than competing metrics . To examine the degree of niche overlap, we calculated the weighted HornMorisita similarity index of interaction patterns among species as recommended by . Given that our focus was to examine the relationship between pollinator diversity and network structure, we report niche overlap for pollinators only. The number of links per species equals the number of distinct links divided by the sum of pollinator and plant species richness. Lastly, while not a traditionally considered network statistic per se, we also calculated the number of plot-level singleton species , as well as the proportion of species at each plotconsisting of singletons, to test the hypothesis that rare species are preferentially extirpated when species richness is reduced. To test the hypothesis that pollinator diversity loss leads to more generalized networks, we calculated H2′, a network-level metric of interaction selectivity integrated across both pollinators and plants . H2′ measures the degree to which plant and pollinator species in a network interact with specific sets of partners, as opposed to distributing their interactions among possible partners based on each partner’s relative abundance . As such, H2′ is relatively robust to variation in the number of species and individuals sampled, facilitating direct comparisons between networks .We constructed multiple linear regression models to examine how pollinator diversity loss influences network properties, where each measured network statistic is a dependent variable. Independent variables were selected among habitat category and the species richness of pollinators and plants recorded in each network. To control for among-site variation in the number of interactions documented, pollinator species richness was rarefied to the lowest site-level number of interactions documented each year , excluding honey bees. Plant species richness was included as a candidate covariate to control for the effects of increased network size on network structure . We constructed all combinations of the three independent variables using R package glmulti , and chose the model with the lowest corrected Akaike’s Information Criterion score. When multiple models achieved similar fit , we chose the minimum sufficient model, or the model that was selected as a top model in both study years. When no model achieved superior AICc scores compared to the null model, we report the results of the model with only rarefied pollinator richness included. Lastly, habitat category and rarefied pollinator richness are related to each other because the former was chosen to generate a gradient in the latter. Fragment plots had, on average, a 22.5% reduction in rarefied pollinator richness relative to reserves across the two study years . However, given the low variance inflation factor of these two variables , their relationship does not appear sufficiently strong for multicollinearity to affect our conclusions in cases where the best model includes both independent variables. In all analyses, data from the two study years were analyzed separately. To test the hypothesis that non-native honey bees modify the structure of plant pollinator interaction network structure in our system, we calculated all aforementioned network-level statistics after excluding honey bees from our dataset. Then, we compared honey bee-present versus honey bee-absent datasets with respect to each statistic using paired t-tests, combining networks from both reserves and fragments. We also calculated the proportion of all interactions consisting of honey bees at each site, and used the model selection process described above to examine whether the relative abundance of honey bees varies across the gradient of pollinator diversity loss. Data from the two study years were analyzed separately. Lastly, in addition to examining the impacts of pollinator diversity loss and introduced pollinators on the structural properties of network, we performed a correlation analysis to examine the degree to which each network statistic varied across the two years of our study. The structure of pollination networks is known to vary from year to year in a given locality ; our use of slightly different sampling methods in the two study years may also contribute to differences in the data. This analysis thus serves to identify network statistics that are robust to inter-annual variation in sampling methodology and population dynamics of plants and pollinators.