Flower visitors to plants in hedgerows and unrestored controls were netted for 1 h of active search time . All insect flower visitors that touched the reproductive parts of the flower were collected; however, here we focus only on wild bees, the most abundant and effective pollinators in the system . Bee specimens were identified to species by expert taxonomists. Surveys of the biotic and abiotic conditions were also conducted at each site throughout the flight seasons of the pollinators. At each site, each flowering plant in 50, one meter quadrats along the length of the hedgerow or control site was identified to species or morpho-species. The abundance of each plant species was estimated as the mean number of quadrats a species was present in, each year. In addition, in 2011 and 2012, we used the same quadrats to evaluate the physical characteristics of the site including the amount of vegetative cover and uncultivated, bare ground.To estimate the species turnover between sites of the same type , we used the variance in community composition as a measure of b-diversity . To calculate this metric, we first calculated the pairwise dissimilarity between sites within each year of the dataset using a dissimilarity estimator that incorporates species abundances, while also accounting for unobserved species . Dissimilarity estimates can be affected by the total number of species and individuals sampled at a site . For example, sampling from a fixed species pool,gutter berries the probability that two sites do not share any species is higher when there are few individuals at those sites.

By extending the method described by to include estimates of species’ abundances, we used null models to estimate the deviation of the observed dissimilarity from that which would be expected under a completely random community assembly process . With the corrected dissimilarity values, we then calculated the multivariate dispersion of communities as the variability inspecies composition within a site type . To investigate effects of site type, the b-diversity estimates were used as the response variable in a linear mixed model with site type as an explanatory variable along with random effects for both year and site . All analyses were conducted in R, version 3.1.1 .We next assessed which spatial pattern was most responsible for maintaining b-diversity within each site type in our landscape. Communities that turnover in species composition across space are thought to arise via two processes: species replacement and predictable species loss/gain . In the latter case, species-poor sites will often be subsets of species-rich sites, and thus, communities should exhibit some degree of nestedness. Such a pattern might occur when, for example, species assemble along a resource gradient . In contrast, species replacement should lead to communities that turnover in composition via substitution of species. This pattern could result when species track their preferred resource or, somewhat randomly via colonization and priority effects. Unlike species loss/gain, these communities would not be expected to exhibit any patterns in nestedness. Thus, to identify which of these two scenarios best describes the patterns in the landscape within each year, we determined whether our communities were significantly nested . We used the index NODF to measure nestedness . To further uncover the processes contributing to spatial heterogeneity, we asked whether the dissimilarity between pollinator communities within and between site types was related to the geographic distance between sites.

To do this, we compared the pollinator community dissimilarity matrix to the geographic distance between sites using Mantel tests. To assess the significance of the correlation, we permuted dissimilarity values among sites within each year to maintain the hierarchy of the data. We also looked for evidence that pollinator communities track resources across the landscape. One important such resource is floral hosts; if the majority of the pollinators track specific floral resources, differences in floral community composition between sites should generate corresponding differences in pollinator communities. To test this, we used Mantel tests to compare the pollinator community dissimilarity matrix to an analogous dissimilarity matrix for flowering plant species within and between site types. As we did for the bee community, we used an abundance-based measure to estimate the dissimilarity of the floral communities . Rather than tracking particular flowering plant species, bees may track floral resources generally. Therefore, we also characterized floral communities according to their species richness, diversity, and total floral abundance, all proxies for floral resource availability. We then used a Gower dissimilarity measure to characterize the changes in the floral resources between sites and then compared that to the pollinator community, again using Mantel tests to look for associations between and within site types. Lastly, both abiotic conditions and resources may affect which pollinator species are present. Bee species vary considerably in their nesting habits, and therefore, the availability of specific nesting materials may influence which species are able to occupy an area . To examine this, we characterized the nesting resources at each site. Specifically, we measured the mean and variability of the amount bare ground, dead wood, hollow stems, cracks in the soil, and vegetation cover .

We used Mantel tests to correlate pollinator community turnover with differences in the physical characteristics of sites, between and within site types, estimated using Gower dissimilarity.We determined whether agricultural areas act as an ecological filter on pollinator groups by comparing the trait distributions of pollinators found at unrestored controls to those found at hedgerows. Our unrestored control sites comprise a variety of unmanaged crop field edges and, therefore, represent the dominant conditions in our landscape. Consequently, the species visiting these sites are those that are likely present in the landscape prior to any restoration. To characterize the trait diversity of the bee communities, we computed three metrics that capture diversity, uniqueness, and distribution of trait values in the community: trait dispersion, divergence, and evenness . Trait dispersion is a measure of trait diversity, corrected for species richness ; trait divergence measures how species abundances are distributed within the trait space ; trait evenness measures the regularity with which traits are distributed across trait space, accounting for abundance . In combination, these metrics provide a relatively complete overview of the different aspects of species trait diversity . Selection of appropriate characters is essential to the characterization of the community’s distribution and diversity of traits . We selected resource capture and use traits that collectively influence the distribution of bee species as pollinators over space and time including resource specialization , body size sociality , nest location , and nest construction as described in more detail in Kremen & M’Gonigle . Each trait has the same weight in trait diversity metric estimation . Pollinator specialization was calculated using plant–pollinator interaction observations from a more extensive dataset from Yolo County that included both the data included in this study and additional data from sites where we collected flower visitors using the same methods . The specialization metric measures the deviation of the observed interaction frequency between a plant and pollinator from a null expectation where all partners interact in proportion to their abundances . It ranges from 0 for generalist species to 1 for specialist species. To determine whether trait evenness, dispersion,strawberry gutter system and divergence differed between controls and hedgerows at different stages of maturation, we used the trait diversity metrics as response variables in linear mixed models with site type as a fixed effect and year and site as random effects . If agriculture creates an ecological filter, the trait composition of agricultural bee communities should differ from that of a community that was randomly assembled from a shared meta-community. To test whether agriculture constitutes an ecological filter, we compared the observed trait values with the distribution of traits of randomly assembled communities.

Because species richness differs between hedgerow and control sites and furthermore, because differences in species richness may constrain the observed trait values and trait diversity , we randomly assembled communities of the same species richness as the observed communities. For quantitative traits, we focused on the mean trait value at a site weighted by abundance, and for categorical traits, we calculated the mean Simpson’s diversity of traits . To generate the randomized communities, we shuffled the species between sites while maintaining the species richness and the number of occurrences of a species within each year. We then re-calculated the mean trait value and Simpson’s diversity of traits for 9999 randomly assembled communities . Lastly, to calculate the probability of the observed trait value given a random assembly process, we computed the fraction of randomly assembled communities that had trait values greater than or equal to that of our observed community. For a given trait, if that probability was <0.025% , we concluded that site type exerted an ecological filter on that trait. To complement the previous analysis, we also asked whether the trait diversity and Simpson’s diversity of traits was significantly different between hedgerows and unrestored controls. We compared the mean trait value or Simpson’s diversity across site types using linear mixed models, with site status as an explanatory variable and site and year as random effects, as before . Lastly, we asked whether the pollinator composition of communities supported by between hedgerows and unrestored controls differed using a permutational multivariate analysis of variance . When comparing community composition, PERMANOVAs can be too liberal when the experimental design is unbalanced and the multivariate dispersions are heterogeneous because it is testing multiple hypotheses simultaneously . As the number of sites was nearly equal for hedgerows and controls within but not between years, we compared the community composition within each year.We have shown that on-farm restorations in the form of hedgerows, when replicated across a landscape, can promote the assembly of spatially heterogeneous and phenotypically diverse pollinator communities in intensively managed and simplified agriculture. Such restorations may thus help to slow or even reverse the biotic homogenization that is characteristic of such landscapes. Without hedgerows, intensive and simplified agriculture imposed a strong ecological filter that eroded patterns of spatial structuring between communities and diminished almost every aspect of community trait diversity and distribution that we investigated. This ecological filter affected a variety of phenotypic traits including nesting habits and also selected for smaller, less specialized bees. In concordance with a number of other studies conducted across a wide variety of taxa, we found that, by homogenizing communities, agriculture has the potential to affect the distribution of species over large scales . Loss of such diversity may impact the functioning and resilience of natural systems which could have profound implications for humans and wildlife. The provisioning of ecosystem services, such as pollination, requires a stable and diverse community of wild bees . These pollination services are critical both in natural communities and economically: 87% of all flowering plant species and 75%of crop species depend to some extent on animal pollinators in order to produce fruits or seeds . Animal-pollinated crops also supply a large proportion of essential nutrients to the human diet . Based on findings in other cropping systems, lower functional diversity, combined with the loss of key service providers, will likely negatively affect levels of pollination in both crops and wild plant populations . In addition, by reducing the size of the species pool, simplified agriculture may impact the stability of services and thus the reliability and predictability of plant reproduction and crop yields . Encouragingly, however, relatively small-scale restorations such as hedgerows can mitigate the homogenization caused by simplified agriculture, when replicated across landscapes. Hedgerows have also been shown to support other ecosystem services , so these small-scale, on-farm restoration measures may also provide an economic benefit to growers , although this is likely to be context dependent . We have shown that, in addition to supporting a higher diversity and abundance of pollinators , hedgerows also support approximately 14% higher b-diversity and approximately 10% more trait diversity, uniqueness, and evenness than unrestored field margins. In addition, because the trait diversity of the communities differed significantly between hedgerows and unrestored controls but community composition did not, the communities at controls are likely a subset of those at hedgerows. For example, 28% of the total species pool was found only at hedgerows, whereas only 13% of species were unique to unrestored controls .