Lastly, temporal beta diversity measures the degree to which individual temporal samples at a study site differ from one another with respect to the composition of taxa present, providing insight into the temporal turnover of the taxa that make up an assemblage . While some popular indices of beta diversity are mathematically derived from measures of alpha and gamma diversity , recent advancements in the field of statistics have enabled additional measures of beta diversity, such as multivariate dispersion , that are mathematically independent of measures of alpha and gamma diversity. Impacts of anthropogenic disturbance on temporal gamma diversity always result from changes in temporal alpha diversity, beta diversity, or both . Decreases in temporal alpha and beta diversity may be driven by different aspects of disturbance , and may have different implications for biological interactions and ecosystem function even if different patterns of temporal alpha and beta diversity loss lead to the same net change in temporal gamma diversity . Trends in temporal alpha and beta diversity may also act in opposition such that temporal gamma diversity remains unchanged in spite of the profound alteration to temporal assemblage structure . Thus, isolating the mechanisms through which disturbance impacts an assemblage requires an examination of all three components of temporal diversity . Such approaches may also serve to identify the ecological effects that result from disturbance .In this study, square pots plastic we investigated the impacts of urbanization-induced habitat fragmentation on the seasonal dynamics of a diverse native bee assemblage over a two-year period.

Bees represent an appropriate taxonomic group for studying how habitat fragmentation affects temporal dynamics because, like many other organisms that occupy seasonal environments, bees exhibit distinct periods of activity that differ among species with respect to both duration and timing of onset . Previous research has demonstrated that anthropogenic disturbance may differentially impact bee species active in different seasons , and that temporal turnover in bee assemblages can contribute to among-habitat differences in site-level bee species richness . Additionally, the key ecosystem function that bees perform is influenced by the season specific pollination effectiveness and temporal complementarity of individual bee species. An explicit consideration of temporal diversity patterns is thus necessary to assess how anthropogenic disturbance affects bee assemblage structure and to identify potential consequences for ecosystem function. Here, we explicitly examined the seasonal dynamics of our focal bee assemblages by simultaneously evaluating their temporal gamma, alpha, and beta diversity. Our use of linear mixed-effects models and analyses of multivariate dispersion distinguishes our study from previous work on temporal patterns in pollinator diversity, the majority of which has focused on quantifying the relative contributions of spatial versus temporal variation in structuring pollinator assemblages . Our approach enabled us to address the following research questions: does habitat fragmentation affect all three components of bee temporal diversity similarly? And how do the effects of habitat fragmentation vary with time? Addressing these research questions allowed us to scrutinize the impacts of habitat fragmentation with a temporal resolution that would be unachievable by pooling temporal samples within study sites.Study System: Between April and August of 2011 and 2012, we documented bee assemblages in the coastal sage scrub ecosystems of San Diego County, California, USA, a global hotspot of bee biodiversity with over 500 bee species documented in the surrounding areas .

We established 1-ha study plots in CSS habitat situated in large natural reserves , and well-preserved habitat fragments embedded within the residential, urban matrix. In 2011, we surveyed four study plots in reserves and four study plots in fragments. In 2012, we surveyed seven study plots in reserves and 11 study plots in fragments. Details regarding the location and treatment classification of each plot are provided in the Table 1-S1. Many of our study plots are located in the same system of reserves and fragments included in earlier studies on the ecological effects of urbanization-induced habitat fragmentation , including bees sampled incidentally in pitfall traps . Permission to conduct field research was obtained from the University of California, San Diego; the Otay-Sweetwater Unit and Tijuana River National Estuarine Research Reserve Unit of the US National Wildlife Refuge; the City of San Diego Open Space Parks Division and Real Estate Division; the City of La Mesa Open Space Division; and the City of Chula Vista Open Space Division. Data collection: We employed bowl trapping and aerial netting to sample bees at all study plots, on sunny days with light wind. Bowl traps consisted of plastic bowls 7 cm in diameter that were white or painted fluorescent blue or fluorescent yellow and filled with ca. 60 ml of unscented detergent solution. During each survey, 30 bowl traps were placed at a study plot before 0900 h and collected after 1500 h. Traps were placed on level ground in an alternating sequence of colors, deployed in two roughly linear transects originating from the corners of each plot and forming an “X” formation near the plot’s center. Traps were placed 5-10 m apart from one another and at least 1 m from the canopy of large shrubs to avoid being shaded. During aerial netting, one researcher walked throughout the study plot and examined blooming plants as well as presumed nesting substrates for bees. Non-Apis bee species were collected regardless of whether they were on flowers, in flight, or in the vicinity of presumed nesting substrates.

In 2011, surveys were performed ca. every 2-3 weeks at each study plot , during which time, 60-min bouts of netting were performed once between 0900 h and 1200 h and once between 1200 h and 1500 h . In 2012, in order to accommodate a larger number of study plots, surveys were performed ca. every 3-5 weeks and included only a single 60-min bout of netting at each plot during each survey. Although seven sites were sampled in both years , the level of sampling employed here seems unlikely to have altered bee assemblages during our study .All collected bees were individually mounted and identified to species or morphospecies within genus using taxonomic keys and the reference collections of the American Museum of Natural History, UC Riverside Entomology Research Museum, California Academy of Sciences, UC Berkeley Essig Museum of Entomology, and UC Davis Bohart Museum of Entomology. Additionally, we also categorized each bee species as a pollen generalist or a pollen specialist based on whether it is documented to exclusively collect pollen from a single plant family. Data used to classify bees as generalists or specialists come from literature accounts for the species  and its subgenus , as well as our own field observations. Bee assemblages often reflect the richness, abundance, and temporal dynamics of their host plant assemblages . Thus, concurrently with the bee sampling, we documented the identities of insect-pollinated native plant species present in each plot in each year; in 2012 we also counted the number of blooming individuals of each plant species in each plot during each survey. We documented blooming plants by walking through pre-planned paths that allowed the observer’s field of view to cover the entirety of the study plot, as in , because many key plant species in our system are patchily distributed and because the thick growth of large, woody shrubs prohibited the use of random linear transects at many of our plots. Statistical analyses: We compared native bee assemblages in reserve versus fragment plots with respect to their temporal gamma, alpha, and beta diversity. In order to avoid human biases associated with aerial netting , our analyses include only bee specimens collected by bowl traps; however, plastic grow pots inclusion of netted specimens in our analyses yielded qualitatively similar results. For analyses requiring species-level identification, we excluded 78 bee individuals not identifiable beyond genus. We also repeated all analyses at the genus level to ensure that particularly species-rich genera did not disproportionately influence our findings; the results of these additional analyses did not alter our main conclusions. Lastly, we verified that reserve and fragment plots did not differ with respect to the composition and temporal dynamics of insect-pollinated native plant assemblages, and that the plot-level compositions of bee assemblages were not spatially autocorrelated . All analyses were conducted in R version 3.3.1 ; packages vegan , MASS , car , and nlme were used in visualizing and analyzing data. Temporal gamma diversity: We define temporal gamma diversity as the diversity of bees at a single study plot, pooled across all temporal samples , with each sample representing the bee specimens collected at one study plot during a single day of data collection. We considered both species richness and assemblage evenness .

In addition, we examined the proportion of bee individuals represented by generalist species , as generalist bees can exhibit higher tolerance to anthropogenic disturbance compared to their specialist counterparts . Lastly, we also examined the temporal gamma component of bee abundance. We used rarefaction in our analyses of species richness and assemblage evenness to account for among-plot variation in the number of bees sampled. We used the lowest plot-level bee abundance recorded each year as the number of individuals to subsample in our rarefactions. Bee abundance was calculated as the total number of bee individuals collected at each plot averaged across the number of temporal samples. Assemblage evenness and generalist proportion were logit-transformed prior to analysis as recommended by , and bee abundance was cube root-transformed to improve normality. We used Welch’s two sample t-tests to compare fragment and reserve plots for all dependent variables listed above. Given the dependence of bee diversity on the diversity and assemblage composition of their host plant assemblages , we also repeated each analysis with the temporal gamma richness of native plants as an added independent variable . We then compared the corrected Akaike Information Criterion scores of each pair of models with or without plant richness added. Compared to original models that did not include plant richness, models that included plant richness yielded qualitatively similar results in all cases but had poorer or equivalent AIC scores; thus, we did not include plant richness in our final models. Temporal alpha diversity: We define temporal alpha diversity as the diversity of bees collected in a single temporal sample . As in our analyses of temporal gamma diversity, we examined species richness, logit-transformed assemblage evenness, logit-transformed generalist proportion, and cube root-transformed bee abundance. In our analyses of species richness and assemblage evenness, we rarefied each temporal sample to 20 bee individuals to allow for unbiased comparisons between treatments and across temporal samples. In analyses requiring rarefaction, we excluded one sample from the 2011 dataset and nine samples from the 2012 dataset . We chose to rarefy to 20 individuals in order to minimize the number of data points to exclude while retaining sufficient resolution in our data. To examine how bee assemblages in reserves and fragments differ over the course of the study period, we constructed linear mixed-effects models. This approach allowed us to quantify the direction of seasonal trends and to detect treatment-by-sample interactions, neither of which is possible for the additive diversity partitioning approach used by most published studies that examined bee temporal alpha diversity . In each model, treatment , temporal sample , and their interaction were included as fixed effects, and study plot identity was included as a random effect to control for repeated sampling as in . To account for possible non-linear relationships between dependent variables and Julian dates of temporal samples, we constructed second- and third-degree orthogonal polynomial models in addition to first-degree linear models for each dependent variable, and selected the model with the lowest corrected AIC score. When alternative models yielded equivalent AICc scores , the model with the lowest degree was chosen. Lastly, as with our analyses of temporalgamma diversity, we repeated all analyses with the temporal alpha richness of native plants as an added independent variable . Models that included plant richness yielded poorer AIC scores in all cases; thus, we did not include plant richness in our final models. Temporal beta diversity: We define temporal beta diversity as the multivariate dispersion of bee assemblages in distinct temporal samples from the same study plot. We chose this index because of its relative mathematical independence from measures of alpha and gamma diversity , as well as its capability to detect differences among assemblages in both species identity and relative abundance .