Landscape Integrity Eclipses Local Effects of Floral Resource Availability on Bumble Bee (Bombus Spp.) Abundance in a Water-Limited Island Ecosystem


 ContextHabitat loss threatens to exacerbate climate change impacts on pollinator communities, particularly in Mediterranean-type ecosystems where late season floral resource availability is limited by seasonal drought. While gardens have been found to supplement floral resources in water-limited urban landscapes, less is known about the role of natural habitat diversity in sustaining late season floral resources in more intact landscapes. ObjectivesWe investigated the importance of habitat integrity and diversity for bumble bees in a water-limited ecosystem, observing bumble bee community response to seasonal drought across gradients of disturbance and soil moisture.MethodsWe applied hierarchical models to estimate the effects of local site conditions versus landscape scale estimates of matrix habitat on bumble bee abundance. Floral resources, soil moisture, and other environmental variables were sampled along randomly distributed belt transects. Geospatial estimates of matrix habitat were derived from terrestrial ecosystem data. Bumble bees were sampled with blue vane traps.ResultsIn the late season we found that modified wet areas supported more floral resources and bumble bee workers as compared to dry semi-natural environments. Wetlands also supported more late season floral resources and bumble bee workers, though the latter effect was not significant. Despite higher levels of late season floral resources in modified wet environments, modified matrix habitat was negatively associated, and natural matrix positively associated, with workers in June and late-flying queens in July and August. We also detected differences in bumble bee community composition in disturbed versus undisturbed environments.ConclusionsThough wet modified habitats sustained the highest levels of late season floral resource availability and worker abundances in our study, bumble bee diversity and abundance were limited primarily by the availability of natural matrix habitat at the landscape scale. The conservation of natural habitat integrity and diversity can help support critical nesting and foraging habitat, and should be prioritized in efforts to foster the resilience of pollinator communities.


Introduction
Climate change has been implicated in the worldwide decline of bumble bees (Bombus Latreille, 1802) as temperature and precipitation begin to exceed the historically observed tolerances of species, and some species ranges recede to higher latitudes and elevations (Biella et  Increasing drought may cause reductions in oral resources supporting pollinators, resulting in a lower proportion of owers containing nectar, fewer owers per in orescence, and reduced oral richness and abundance (Phillips et al. 2018; Simon et al. 2021). Drought has also been found to advance owering phenology among annual herbaceous plants (König et al. 2017), and to cause stress or reduced productivity among perennial shrubs (Pérez-Camacho et al. 2012). In turn, overall species diversity in plant-pollinator communities has been found to decline under conditions of drought stress (Hoiss et al. 2015). These effects could potentially bene t generalist pollinators such as bumble bees, or, conversely, promote specialists with narrower foraging requirements (Minckley et al. 2013; Thomson 2016). Drought has moreover been found to exacerbate effects of interspeci c competition between honey bees (Apis mellifera Linnaeus, 1758) and bumble bees, reducing the foraging and reproductive success of native bumble bee populations (Thomson 2016).
Bumble bees are central place foragers limited by the availability of both nesting and foraging habitat.
Compositionally diverse landscapes offering a mosaic of herbaceous and forested habitats are known to promote the availability of foraging and nesting resources, and in turn, a greater abundance of bumble bees (Jha et  Research has tended to contrast agricultural and urban habitats with only one alternative semi-natural habitat type-generally in the context of highly fragmented landscapes (e.g., Heard et al. 2007 We investigated the importance of habitat integrity and diversity for bumble bees in a water-limited ecosystem by observing owering plant and bumble bee community response to seasonal drought across gradients of disturbance and soil moisture. We predicted that dry semi-natural habitats such as woodlands would host fewer oral resources and, in turn, fewer bumble bee workers than other habitat types during the dry late season. Conversely, we predicted that modi ed wet habitats such as gardens, and wet semi-natural habitats such as wetlands, would support greater oral resources, and, in turn, a greater abundance of workers in the late season. Disturbance was predicted to promote late season oral resource availability among exotic plant species, with bumble bee workers expected to persist longer in disturbed environments both wet and dry. We also tested whether surrounding natural and modi ed matrix habitat in uenced bumble bee abundance at the site level, analyzing the response of each species and caste. Finally, we tested whether different habitat types favoured certain species, with Bombus vosnesenskii expected to be more prominent in disturbed environments.

Study area
Galiano Island lies in the rain shadow of the Olympic Mountains and the Vancouver Island Ranges, in southern coastal British Columbia, Canada. This region is de ned by its temperate Mediterranean-type climate, with mild, wet winters and warm, dry summers (Klassen et al. 2015). The combined effects of low precipitation, warm temperatures, and high sunshine hours result in an annual moisture de cit during summer months, which varies to some extent depending on precipitation ( Galiano Island is relatively intact ecologically, with about 24% of its land base conserved in protected areas and a high percentage of forest cover, hosting vegetation communities typical of the Coastal Douglas-r Biogeoclimatic Zone (Nuszdorfer et al., 1991). Although the island has historically been subject to extensive industrial forestry practices, only ~11% of the landscape has been converted for present day human use (Madrone 2008). This largely forested study area thus presents a contrast to the fragmented environments previously studied by pollinator researchers in water-limited ecosystems (e.g., McFederick and LeBuhn 2005; Wray and Elle 2015).

Sampling
This study was based on a 2×2 factorial study design contrasting four conditions of disturbance and soil moisture availability: 1) dry semi-natural environments (woodland and associated rock outcrop communities); 2) wet semi-natural environments (wetlands); 3) dry modi ed environments (disturbed upland areas such as clear-cuts and hydro-line corridors); and, 4) wet modi ed environments (gardens, orchards and elds). Sites were strati ed based on available terrestrial ecosystem mapping data (Madrone, 2008), with 6 sites selected per condition (for a total of 24 sites). Limitations in existing site conditions and the logistics of site access resulted in an imbalance in the study design, with 4 sites representing the wet seminatural condition versus 8 sites representing the dry seminatural condition (and 6 sites representing the other two conditions).
Terrestrial ecosystem mapping data were used to obtain estimates of land cover falling within a 500 m buffer around each site, orienting around polygon centroids, including natural, modi ed, and forested matrix habitat (Fig. 1). Natural matrix habitat included coniferous and deciduous forest cover at various structural stages, as well as wetlands, woodlands, meadows, and rock outcrops. Modi ed matrix included rural, agricultural and developed land cover. Sites ranged in size from 0.21 to 6.3ha (Supplementary Information). While grouped site conditions circumscribed similar habitat types based on common soil moisture regimes, disturbance conditions were diverse, including forestry, re, landscaping, and other anthropogenic effects. Commonalities between site conditions (henceforth habitat types) are re ected in the similarity of vegetation communities, as shown in ordination plots (Supplementary Information). Additional environmental data were collected along transects at the site level, including estimates of canopy cover, slope, moss cover, coarse woody debris, and bare soil-variables known to correlate with bumble bee nesting habitat (Potts et al. 2005; Wray and Elle 2015).
To sample oral resource availability (FRA), we strati ed habitat types and randomly distributed 6-8 (2 × 15 m) belt transects throughout each site, using balanced acceptance sampling methods to ensure a random yet spatially balanced distribution of transects (van Dam-Bates et al., 2018). The number of transects was scaled roughly in proportion to the size of each site, with the aim of capturing variability in oral resource availability across the landscape. Floral resource availability was quanti ed as counts of owering shoots, recorded for each plant species at 1-m intervals, with each interval surveyed comprising a 1 × 2 m area spanning both sides of the transect line. Soil moisture was recorded at 5-m intervals as volumetric water content (%VWC) using a Field Scout Time-Domain Re ectometry probe. Sampling was conducted on a monthly basis from April through August 2018, resulting in ve samples per site, with each sample period covering 13 days.
To sample bumble bees, three blue vane traps (Stephen and Rao 2005) were placed at roughly equidistant intervals across each site. Traps were positioned to ensure their visibility given surrounding vegetation and landscape features, and to minimize public interference. Bumble bees captured by blue vane traps were pooled into one sample per site, resulting in 24 × 5 (120) samples divided into the four habitat types. Blue vane traps were vandalized in a dry semi-natural site in May, resulting in the loss of one sample (n=120-1=119). Bumble bees were identi ed following Williams et al. (2014).

Modelling
Statistical analyses were implemented in R Version 3.6.0 (R Core Team 2019). Linear mixed effects models (LMMs) were tted incorporating log-transformed soil volumetric water content (logVWC) as a response to contrasting habitat types ( xed effects), with transects nested within sites (random effects), to estimate differences in soil moisture between habitat types. To estimate differences in oral resource availability (FRA), we t negative binomial generalized linear mixed effects models (GLMMs) incorporating counts of owering shoots as a response to contrasting habitat types ( xed effects) and transects nested within sites (random effects) for discrete sample periods. We also modeled FRA as a seasonal pattern, by tting a GLMM incorporating time (julian days) as a quadratic term to estimate overall differences in FRA across site conditions. While time emerged as a signi cant term in these models, the quadratic term did not adequately describe the observed seasonal patterns in FRA, resulting in a poor t. Hence, our analyses have focused on discrete sample periods instead. Negative binomial generalized linear mixed effects models were also tted to model mean bumble bee abundance (counts) as a function of habitat type and sample periods ( xed effects), with a random intercept term corresponding to site. Other environmental variables incorporated as predictors included proportions of natural and modi ed matrix, forest cover, slope and moss cover.
Signi cant terms were evaluated based on model summaries, and top models selected based on AIC scores (see below).
We estimated differences in the total count of bumble bees sampled between habitat types for each sample period separately. Because our focus was on whether certain habitat types supported prolonged foraging under conditions of drought stress, we concentrated primarily on bumble bee worker abundance as a proxy for bumble bee foraging activity, though differences were also considered among drones and queensespecially late-ying queens, whose presence in the environment may be considered a proxy for the reproductive success of colonies. We tted GLMs to estimate differences in the abundance of bumble bees within isolated sample periods, as temporal autocorrelation was not problematic in these analyses. We also modelled the abundance of each species as a response to habitats and other environmental variables, both across sample periods (GLMMs) and within sample periods (GLMs). Linear mixed effects models were implemented using the R package 'nlme' (Pinheiro et

Onset of seasonal drought
Soil moisture declined with the onset of seasonal drought across all habitats (Fig. 2). Dry semi-natural and dry modi ed sites declined to 1.5 ±1.5% volumetric water content (VWC) by June, and had similarly low soil moisture conditions throughout the rest of the growing season. Wet semi-natural sites sustained relatively high soil moisture, declining from 20.6 ±1.1% VWC in June to 10.6 ±1.2% VWC in August. Wet modi ed sites were comparatively dry yet signi cantly wetter than dry sites from June (10.9 ±1.5% VWC) until August (4.6 ±1.5% VWC).

The foraging landscape
Plant community phenology varied signi cantly across habitats, resulting in distinct peaks and declines in oral resource availability (FRA) over the course of the growing season (Fig. 3). In dry semi-natural and dry modi ed habitats, FRA peaked early (in May), supporting signi cantly more FRA than wet semi-natural habitats from April through May. In wet semi-natural and wet modi ed habitats, FRA peaked later (in June). Both wet and disturbed conditions sustained FRA later into the season, with wet semi-natural, wet modi ed and dry modi ed habitats supporting signi cantly more FRA than dry semi-natural habitats from June through August. By August, signi cantly higher FRA was found in wet modi ed, wet semi-natural, and dry modi ed habitats versus dry semi-natural environments, in rank order of abundance, with wet modi ed environments supporting nearly twenty times more FRA than dry semi-natural environments (IRR WET.M .=17.70 DRY.N ., p<0.001).
Disturbance was also a critical factor promoting oral resource availability among exotic plants (Fig. 4) hosted signi cantly more bumble bee workers than wet semi-natural habitats (Fig. 5). In July, bumble bee worker abundance declined in both dry modi ed and dry semi-natural habitats. Meanwhile worker abundance peaked in wet semi-natural and modi ed environments. However, no signi cant differences in worker abundance was detected between habitats in July. Workers declined dramatically across all habitats by August (Fig. 6). During August, wet modi ed habitats supported signi cantly more workers than dry seminatural habitats (IRR WET.M .=3.07 DRY.N ., p=0.008), which supported the fewest workers. Wet semi-natural and dry modi ed habitats also supported more workers than dry semi-natural habitats in August, though these differences were not signi cant (Fig. 6).

Environmental predictors of bumble bees across the landscape
We modeled the response of bumble bees across all habitats and sample periods, as well as the response of discrete species and castes, as a function of several geospatial parameters and environmental variables known to be associated with foraging and nesting resources (Table 1). Floral resource availability (FRA) was the only signi cant predictor of bumble bee abundance across all habitat types (    . However, we also found evidence that semi-natural habitat diversity-and natural matrix habitat more broadly-correlated with greater abundances of bumble bees on the landscape in some cases. Worker abundance in dry semi-natural habitats was comparable to abundances found in dry disturbed habitats throughout the growing season, despite signi cantly higher levels of late season oral resources in the latter environments. Furthermore, in dry seminatural habitats, late season declines in worker abundance were not as severe as seen within more impacted ecosystems such as the fragmented oak savannahs of the nearby Saanich Peninsula on Vancouver Island. In this system, Wray and Elle (2015) found pollinator diversity and abundance to decline sharply in dry woodlands from June through August, tracking closely with decreasing oral resources-except where surrounding urban matrix habitat supplemented diminishing oral resources in these environments. We found a different pattern: high numbers of bumble bee workers were sustained in dry semi-natural habitats through June and July, despite a dramatic decline in oral resources. Moreover, modi ed matrix habitat was correlated with decreased abundance, and natural matrix habitat with increased abundance, of workers in June, and late-ying (reproductive) queens in July and August. Bumble bees thus appeared to be resilient to local declines in oral resources, with workers (in June) and late-ying queens occurring in greater numbers in sites surrounded by natural matrix habitat, independent of variation in local site conditions. This pattern suggests that, while gardens may play an important role supplementing late season oral resources in fragmented urban and agricultural landscapes, the availability of natural matrix habitat may be more limiting for bumble bees in more intact landscapes.
Bumble bees are known to forage at great distances, depending on resources available in the environment Wetlands as refuges for bumble bees during late season drought?
While recent research has found wetlands to be important for pollinators in agricultural landscapes (Vickruck et al. 2019), our study did not support the hypothesis that wetlands provide important foraging habitat for bumble bees in this water-limited ecosystem. Through the early season, oral resource availability (FRA) remained signi cantly lower in wetlands than other habitats, and wetlands were host to correspondingly low numbers of workers. Bumble bee nesting density is known to be directly related to oral abundance, especially early season FRA (Knight et al. 2009;Williams et al. 2012), which may have resulted in decreased abundance of bumble bee workers in these habitats. Yet it is also possible that wetlands are generally less habitable to ground-dwelling bumble bees than drier, upland habitats. Wet conditions have been observed to hamper bumble bee colony development (Harder 1986), and pollinator nesting is also known to be positively associated with moderately sloped upland habitats (Potts et al. 2005; Wray and Elle 2015). Wetlands may therefore be avoided by nesting queens, resulting in a lower density of colonies in these environments. In wetlands, worker abundance peaked in July-one month after FRA reached its peak-which suggests a lag in the productivity of local colonies. This pattern is also apparent across other habitat types, with worker abundance climaxing in the months following peak owering events.
Wetlands nevertheless supported greater oral resource availability than dry semi-natural habitats in the late season. Moreover, in contrast to all other habitats, where exotic plants became increasingly prevalent, wetland FRA was found chie y among native plants. These ndings indicate that natural habitat diversity may be especially important in intact landscapes, as the diversity emergent across soil moisture gradients may be necessary to support the availability of early and late season oral resources in native plant communities. Indeed, in the absence of major anthropogenic disturbance events, soil moisture gradients were likely crucial in shaping plant-pollinator interactions across the foraging landscapes of the past-particularly for insects having annual colony cycles such as bumble bees, which depend on the availability of both early and late season oral resources (Rundlöf et  , providing insight into these effects in the context of a Mediterranean-type ecosystem that exhibits distinct peaks and declines in oral resources due to seasonal drought. In this system, natural habitat diversity across the soil moisture gradient, from woodlands to wetlands, supported oral resources throughout the growing season. While human modi ed settings such as orchards and gardens supported greater oral resources and worker abundance in the late season, modi ed matrix habitat had a stronger negative effect overall on the occurrence of bumble bees. Our results thus underline an important lesson for pollinator conservation in the face of climate change and increasing habitat destruction. Across fragmented urban and agricultural landscapes, restoration interventions may be required to promote the foraging habitat necessary to foster diverse pollinator communities. However, such measures are of secondary importance to conservation in the context of more intact natural ecosystems. Where possible, the conservation of natural habitat integrity and diversity-across the spectrum of dry to wet habitat types-can help support critical nesting and foraging habitat, and should be prioritized in efforts to foster the resilience of pollinators on the landscape. Figure 1 Site locations on Galiano Island, BC, by habitat type and proportion of surrounding natural versus modi ed matrix habitat Figure 2 Mean soil moisture across each site class from April through August, shown with 95% con dence intervals (LMM estimates). Soil moisture was estimated as volumetric water content (%VWC), a ratio of water to soil (m3/m3), using a TDR probe. All differences in wet versus dry site conditions are signi cant at p = <.001-.001 Model outputs of counts of owering shoots across habitats and samples, with 95% con dence intervals. Signi cant differences between site conditions, both for independent samples and across sample ranges, are marked with letters (a is signi cantly different from b, x signi cantly different than y). All differences signi cant at p = <.001-.02 Mean Floral Resource Availability (FRA) among native versus exotic plants across habitats and samples, with 95% con dence intervals (GLMM estimates). Asterisks indicate signi cant differences between native versus exotic FRA at * (p=<0.05) and *** (p=<0.001). indicates signi cance could not be calculated because only exotic species are present Mean bumble bee worker abundance across habitats and samples, with 95% con dence intervals (GLM estimates). Signi cant differences between site conditions are marked with letters (a is signi cantly different from a, b signi cantly different from b…). All differences signi cant at p=<0.01