LETTER Interannual bumble bee abundance is driven by indirect
climate effects on floral resource phenology
Jane E. Ogilvie,1,2*
Sean R. Griffin,1,3
Zachariah J. Gezon,1,4,5
Brian D. Inouye,1,2
David W. Inouye1,6 and
Rebecca E. Irwin1,3
Abstract Climate change can influence consumer populations both directly, by affecting survival and repro- duction, and indirectly, by altering resources. However, little is known about the relative impor- tance of direct and indirect effects, particularly for species important to ecosystem functioning, like pollinators. We used structural equation modelling to test the importance of direct and indi- rect (via floral resources) climate effects on the interannual abundance of three subalpine bumble bee species. In addition, we used long-term data to examine how climate and floral resources have changed over time. Over 8 years, bee abundances were driven primarily by the indirect effects of climate on the temporal distribution of floral resources. Over 43 years, aspects of floral phenology changed in ways that indicate species-specific effects on bees. Our study suggests that climate- driven alterations in floral resource phenology can play a critical role in governing bee population responses to global change.
Keywords Bumble bee, Bombus, climate change, floral resources, phenology, pollinator, precipitation, snowmelt, structural equation model.
Ecology Letters (2017)
Climate change is affecting the abundance and distribution of organisms worldwide (Parmesan 2006; Van der Putten et al. 2010). Continuing changes in temperature, precipitation and the incidence of extreme weather events (IPCC 2013) can affect population sizes directly, by affecting survival and reproduction (Bale et al. 2002; Roland & Matter 2016), and indirectly, by altering resource availability and species interac- tions (Boggs & Inouye 2012; Høye et al. 2013; Kudo & Ida 2013). Because few studies examine both direct and indirect effects in a single analytical framework, we know little about the relative importance of these climate effects. It is critical to understand how the combined direct and indirect effects of climate affect populations to make better predictions of popu- lation responses to climate change, especially for species key to ecosystem functioning. Bumble bees (Bombus spp.) are functionally important
organisms because they are abundant and effective pollinators in both natural and agricultural systems (e.g. Hegland & Tot- land 2008; Garratt et al. 2014). Many bumble bee species are experiencing dramatic declines (Williams et al. 2009; Cameron et al. 2011). Although many factors are implicated – including habitat loss, agrochemicals and novel parasites – climate change is a likely driver of current and potential future decli- nes (Goulson et al. 2015; Kerr et al. 2015). Understanding the
effects of climate on high-elevation bumble bees is especially urgent because montane regions are expected to experience the most extreme changes in climate (Nogu�es-Bravo et al. 2007), and bumble bees are some of the primary pollinators in these areas (Bergman et al. 1996; Bingham & Orthner 1998). The available evidence suggests that bumble bee populations are responding to climate change: some species have under- gone recent latitudinal and elevational range shifts (Ploquin et al. 2013; Kerr et al. 2015; Pyke et al. 2016), others morpho- logical changes likely in response to changing floral resources (Miller-Struttmann et al. 2015), and some are active earlier in the season than in the past (Bartomeus et al. 2011). However, the relative importance of direct and indirect climate effects on bumble bee populations is poorly understood, which limits our ability to explain how climate change may affect these important pollinators. Bumble bees are eusocial insects whose population sizes rely
on continuous floral resource availability (abundance of flow- ers used by bees) for successive life stages during the flight season: overwintered queens start colonies in the spring, over- lapping worker cohorts forage during mid-season, and repro- ductive males and queens are produced late in the season (Goulson 2010). Though research is limited, climate may have negative direct effects on bumble bees if extreme temperatures or precipitation cause high mortality in overwintered queens or colonies (e.g. Vesterlund & Sorvari 2014; Oyen et al. 2016)
1The Rocky Mountain Biological Laboratory, Post Office Box 519, Crested
Butte, Colorado 81224 USA 2Department of Biological Science, Florida State University, 319 Stadium
Drive, Tallahassee, FL 32306 USA 3Department of Applied Ecology, North Carolina State University, 127 David
Clark Labs, Raleigh, NC 27695 USA
4Disney’s Animal Kingdom, Animal Programs Administration, PO Box 10000,
Lake Buena Vista, FL 32830 USA 5Department of Biology, Rollins College, 1000 Holt Avenue, Winter Park, FL
32789, USA 6Department of Biology, University of Maryland, College Park, MD 20742 USA
*Correspondence: E-mail: [email protected]
© 2017 John Wiley & Sons Ltd/CNRS
Ecology Letters, (2017) doi: 10.1111/ele.12854
or reduce critical foraging activity (e.g. Bergman et al. 1996). However, warmer temperatures may also have positive effects by increasing rearing temperatures and brood production (Holland & Bourke 2015), or foraging activity and provision- ing, as in mason bees (Forrest & Chisholm 2017). Climate may also have indirect effects by altering the abundance and phenology of vital floral resources (Thomson 2016). Bumble bee populations often increase with floral abundance (e.g. Inari et al. 2012; Crone & Williams 2016), and colony growth may be impeded by aspects of floral phenology, such as gaps in floral availability (Williams et al. 2012; Kudo 2014) and season length (Elliott 2009a). Flowering is strongly responsive to climate, and there are widespread reports of shifting phenologies and floral abundance declines over time or with climate extremes (e.g. Høye et al. 2013; Iler et al. 2013; Miller-Struttmann et al. 2015; Thomson 2016). In other polli- nator groups, such as flower-feeding butterflies, there is strong evidence for direct climate effects on populations (Roland & Matter 2016), as well as both direct and indirect climate effects via floral resource abundance (Boggs & Inouye 2012). However, the relative importance of direct and indirect cli- mate effects on bee populations remains unresolved. We examined the direct and indirect effects of climate on
the abundance of three subalpine bumble bee species in the Rocky Mountains of Colorado, USA (Fig. S1). At our study site over the last four decades, there has been documented earlier spring snowmelt, warmer spring and summer tempera- tures, and more frequent damaging spring frosts (Inouye 2008; Iler et al. 2013). Simultaneously, the flowering season is shifting earlier and extending (CaraDonna et al. 2014), and a mid-season floral decline is expanding (Aldridge et al. 2011). Given these complex changes that could influence bee popula- tions, we used long-term data on climate, flowers and bee abundance to ask: (1) whether the direct or indirect effects (via floral resources) of climate variation were more closely linked to changes in bee abundance over 8 recent years, and (2) whether these climate and floral resource variables – potentially important for bee abundance – have changed directionally over the last 43 years, suggesting future flower and bee responses to continued climate change. We used piecewise structural equation modelling (SEM; Lefcheck 2015) to disentangle the direct and indirect effects of climate on bee abundance (Fig. 1). We show that climate variation affects the abundance of three bumble bee species indirectly by alter- ing the temporal distribution of floral resources. Our study suggests that climate-driven alterations in floral resource phe- nology can play a critical role in governing bee population responses to global change.
Study system and approach
We used three long-term datasets collected near the Rocky Mountain Biological Laboratory (RMBL; 38°57.5 N, 106°59.3 W, 2900 m) in Colorado, USA. We combined bee, flower and climate datasets over 2009–2016 (8 years) to exam- ine the direct and indirect effects of climate on interannual bumble bee abundance, and flower and climate data over
1974–2016 (43 years) to examine the long-term trends in vari- ables potentially important for bumble bee populations. Bee data were collected from sites between 0.4 and 1.8 km from the central site where flowers and climate were measured, although the sites shared dominant habitats and plant species. The area has flower-rich subalpine meadows (c. 120 non-gra- minoid plant species occur in our floral plots), and up to 16 bumble bee species including two parasitic species. We focused on three common and distinguishable species, B. bi- farius, B. flavifrons and B. appositus (Fig. S1; Williams et al. 2014). They have short, medium and long tongues, respec- tively, which match differences in relative body size and the plant species they frequently visit (Inouye 1980; Pyke 1982). Flowering season timing and length is governed by climate: it begins when the winter snowpack melts (April 23 to June 19 over 1975–2016) and ends with the onset of frequent freezing temperatures (September–October).
The bumble bee abundance data are from an ongoing project described elsewhere (Gezon et al. 2015). Briefly, during the flowering season from 2009 to 2016, typically June through August, bumble bees were sampled at 2-week intervals at each of three sites (4–9 sampling periods totalling 20–54 observa- tion hours, mean 35 h, in each site and year). Each site was composed of three habitats representative of the meadows vis- ited by bees in the area (dry meadow, Salix spp.-dominated wet meadow and Veratrum tenuipetalum-dominated wet mea- dow). The three sites – A, B and D – were at 2980, 2930 and 3070 m elevation, respectively, and the area sampled at each site was c. 4500 m2. All three bee species were common at each site. On each sampling day, bumble bees were hand- netted from flowers for c. 1 h in each of the three habitat types in both the morning and afternoon (6 h per sampling period, although poor weather sometimes shortened sam- pling). Bee species were identified in the field based on distinc- tive pile colour patterns (Williams et al. 2014) and were marked with paint to avoid recounting individuals within days. We combined counts of the uncommon B. sylvicola and abundant B. bifarius to form a species complex, because they
Precipitation Floral phenology
Figure 1 Path diagram showing all hypothesized direct and indirect links
among climate variables, flower variables and bumble bee abundance.
Floral phenology and abundance may directly affect bee abundance
(purple arrows), while climate variables could affect bee abundance both
directly (green arrow to bee abundance) and indirectly through their
effects on floral phenology and abundance (green arrows to floral
© 2017 John Wiley & Sons Ltd/CNRS
2 J. E. Ogilvie et al. Letter
are difficult to differentiate in the field and both have similar flight phenologies, short tongues and visit similar plant species (Pyke 1982). On each sampling date at each bee site, we also recorded the plant species in bloom and the species of flower each bee was netted from, both of which informed our choice of floral resource variables. Our response variable for each of the three bumble bee spe-
cies was annual peak abundance (worker and male bees com- bined), which is an estimate of population size that was comparable across years. We could not use summed bee abun- dances across a season because the number of seasonal sam- pling periods varied across years. Peak abundances were unimodal and marked, which is the typical seasonal abun- dance curve of bumble bees (e.g., Pyke et al. 2011). Although peak abundances could be sensitive to weather conditions on a sampling day, we only sampled during suitable weather con- ditions. We pooled bees within sites because habitats were too close to be independent. For each site and species, peak abun- dance was the maximum number of bees caught/hour in a year. In the few site-years in which peak abundance occurred on the last sampling date, we are confident that we captured the peak due to a predictable seasonal decline in bumble bees that coincides with cooler overnight temperatures and floral abundance declines. Because peak bumble bee abundances occur late in the summer (Pyke et al. 2011), our analyses emphasize colony growth over a season. Moreover, in similar Colorado habitats, foraging worker numbers were highly cor- related with the number of colonies (Geib et al. 2015). The previous year’s peak bee abundance had no relationship with the current year’s peak (Fig. S2), so we did not include the previous year’s abundance in our analyses.
To measure bumble bee floral resources, we used a detailed flower community dataset collected from 1974 to 2016 at the RMBL (Inouye 2008; CaraDonna et al. 2014), a site central to the bee sampling sites. These data are housed at the Open Science Framework (https://doi.org/10.17605/osf.io/jt4n5). Within permanent 2 9 2 m plots, we counted all open flowers approximately every second day throughout the growing sea- son. Individual flowers were the unit counted, except for plants in the Asteraceae for which we counted capitula. Over 2009–2016, data are from 30 plots (7 in dry meadow, 21 in wet meadow including that dominated by Salix spp. and Ver- atrum tenuipetalum and 2 in aspen forest), while over 1974– 2016, data are from 23 plots (7 in dry meadow, 14 in wet meadow and 2 in aspen forest). From this dataset, we com- piled a separate plant list for each bee site that matched the plant species recorded at that site. From each site-specific list, we made separate lists of plant species heavily visited by each of the three bee species (pooling plant species visited by B. bi- farius and B. sylvicola for the B. bifarius complex). We com- bined bumble bee netting data from our bee sites and visitation data from near the RMBL to determine heavily vis- ited plant species (J. E. Ogilvie, unpublished data). The plant species included in each list together comprised 92–95% of the total flower visits made by each bee species (20–24 plant
species were included in the site-specific B. bifarius complex, B. flavifrons and B. appositus lists; Table S3). We calculated two variables to describe floral resource
abundance and timing, referred to hereafter as annual floral sum and the number of floral days. To calculate annual floral sum, using the site- and bee species-specific plant lists, we pooled flower counts across plant species and plots on every sampling date, and calculated the sum of flowers from first flower until 80% of flowers had accumulated. We used this measure because peak bumble bee abundance tended to occur near the date that 80% of a season’s flowers accumulated and we wanted the floral resource variables to be defined consis- tently across years and independently of the bee abundance variable. Number of floral days was the number of days above a low flower threshold (0.75 flowers/m2 or 3 flowers per 2 9 2 m plot) between the first flower date and the date on which 80% of the season’s flowers had accumulated. This variable reflects the time span of floral availability for bumble bees – days of very few flowers are likely of poor foraging value. The relationship between bumble bee abundance and the number of floral days was consistent across a range of low flower thresholds from 0.5 to 1 flower/m2. The variance inflation factors (VIFs) between floral sum and the number of floral days for all three bee species over both time periods (2009–2016, 1974–2016) were 1.19–2.5 indicating low collinearity.
We used two climate variables, date of snowmelt and cumula- tive precipitation from May through July, because they have been shown to relate strongly to flower abundance and phe- nology (e.g. Lambert et al. 2010; Iler et al. 2013), and could also affect bumble bee populations directly. For queens that hibernate underground, snowmelt signals the potential start of the flight season, while precipitation may govern the time available for foraging. We selected these two variables by first creating a list of a priori climate variables, removing those with high VIFs (indicating multicollinearity), and finally removing those with strong one-way correlations with others. We considered temperature variables, though temperature measurements from the RMBL of accumulated degree-days above 0°C in June and July were negatively correlated with snowmelt date and May–July precipitation, respectively (Pear- son r = �0.67 and �0.54, P < 0.05). Annual snowmelt dates were the day of year that a permanent 5 9 5 m plot was bare of snow, recorded 1975–2016 at the RMBL. Daily precipita- tion, including both snow and rain, was measured in cm water content, and then summed over the period May through July. Precipitation over 2009–2016 was measured at the RMBL, while data over 1975–2016 are from the Crested Butte National Oceanic and Atmosphere Administration weather station (ca. 9 km south of the RMBL), because monthly sum- mer precipitation data for the RMBL do not extend as far. 2000–2016 precipitation data from the RMBL and Crested Butte were highly correlated (Pearson r = 0.78, P < 0.001). Climate measurements at the RMBL were taken by long-time resident, billy barr.
© 2017 John Wiley & Sons Ltd/CNRS
Letter Climate and flower effects on bumble bees 3
Direct and indirect climate effects on bees
To disentangle the direct and indirect effects of climate on interannual bumble bee abundance over 8 recent years, we combined the datasets on bees, flowers and climate and used piecewise SEMs (Lefcheck 2015). Piecewise SEMs are con- ceptually similar to classical path analysis, but rather than use global estimation from a single variance–covariance matrix, piecewise SEMs solve each component model sepa- rately. Thus, piecewise SEM allows for models with their own sampling distributions and can operate with smaller sample sizes (Lefcheck 2015). For each bee species, we statis- tically compared a set of SEMs to determine the key direct and/or indirect climate effects that drive bee abundance. To do so, we first constructed a full causal path model with all hypothesized relationships among our variables using knowl- edge of the study system (Fig. 1; Grace et al. 2012). We then fit the three component models (response variables: number of floral days, annual floral sum and bee abundance) as lin- ear (LMs) or generalized linear models (GLMs) each with their appropriate distributions (negative binomial or Pois- son). From the full causal model, we then removed paths to bee abundance to create a set of SEMs with every possible combination of paths that were biologically plausible (16 SEMs, including the full model). In the series of SEMs, each component model had site (three levels) as a predictor to account for variation among sites, and the two floral vari- able models always had direct climate effects (as in Fig. 1), given well-known effects of climate on flowering. The response variable for the bee abundance model was the num- ber of bees caught on the peak sampling day, with the sam- pling effort (in hours) included as an offset term to account for different sampling durations. To assess the overall fit of each SEM, we used Shipley’s test
of d-separation (Shipley 2009, 2013), which tests whether model fit would be improved by the inclusion of identified missing paths. The d-separation test generates a Fisher’s C test statistic, which can be used to assess overall fit of the SEM and to calculate Akaike’s information criterion cor- rected for small sample sizes (AICc) for model selection (Ship- ley 2009, 2013). For each bee species, we, therefore, used a two-part selection process, in which we first only considered SEMs with P-values derived from Fisher’s C of > 0.05. Of the SEMs with no significant missing paths, we then selected the SEM with the lowest AICc value by a difference of at least two points. For each best-fit SEM, we extracted all coeffi- cients to determine the strength of paths. We used the Ben- jamini-Hochberg procedure to correct P-values for multiple comparisons (Smith & Cribbie 2013). All statistical analyses were conducted in R version 3.3.2 (R Core Team 2016). Tests of d-separation and extraction of coefficients were done using the R package ‘piecewiseSEM’ (Lefcheck 2015) and compo- nent negative binomial GLMs using ‘MASS’ (Venables & Ripley 2002).
Long-term climate and flower trends
To test whether climate (snowmelt date and May–July precipi- tation) and the assemblage of flowers visited by bumble bees
(annual floral sum and the number of floral days) have chan- ged directionally over time, we examined the long-term cli- mate and flower datasets. For the flower dataset spanning 1974–2016, we excluded the years 1976–1978, 1990 and 1994 due to missed sampling (N = 38 years; 23 2 9 2 m plots/ year). For the climate dataset spanning 1975–2016, precipita- tion data from 1978 and 1979 were excluded because of miss- ing data (N = 40 and 42 years for precipitation and snowmelt date, respectively). To examine how the climate and floral variables have changed over four decades, we used simple lin- ear models for each response variable.
Direct and indirect climate effects on bees
Over 8 years, 2009–2016, there was substantial variation in peak abundance of the three bumble bee species (B. bifarius: 0–17, B. flavifrons: 0.17–13 and B. appositus: 0–8, bees caught/hour). The assemblage of flowers visited by the three bee species also varied over the same time period, both in annual floral sum, an estimate of cumulative floral abundance (B. bifarius: 60.7–897.4; B. flavifrons: 36.7–850.5 and B. ap- positus: 26.3–810.8 flowers/m2 accumulated to 80% of the season total) and in the number of floral days, an estimate of the season time span of available floral resources (B. bifarius: 14–47, B. flavifrons: 20–40 and B. appositus: 14–34, days above the flower threshold of 0.75/m2). Likewise, there was also substantial variation in the climate variables, snowmelt date (April 23 to June 7) and May–July precipitation (11.00–33.63 cm). For each bumble bee species, we compared a series of
piecewise SEMs that varied in the presence of direct and indi- rect paths of climate variables to bee abundance. Each com- parison yielded a single best-fitting SEM (Fig. 2a–c; Table S4). In the best-fitting model for each of the three spe- cies, bee abundance was driven most strongly by the indirect effects of precipitation and snowmelt date on the number of floral days (Fig. 2a–c; Table S4). Bee abundance increased with more floral days (Fig. 3), while the number of floral days increased with greater summer precipitation and later snowmelt dates (Fig. 2a–c; Table S5). Annual floral sum was included but not significant in the model for B. appositus, and had a weak negative effect on B. bifarius abundance (Table S5). There were no significant directional trends through time in peak bee abundance, number of floral days, and annual floral sum for all three bee species over 2009–2016 (Fig. S6).
Long-term climate and flower trends
Over the 42 years (1975–2016), there was a trend for snow to melt 12.8 � 7.1 days earlier (LM, F1,40 = 3.3, P = 0.0781; Fig. 4a), while May–July precipitation showed no consistent pattern (LM, F1,38 = 0.2, P = 0.671; Fig. 4b). Over 1974–2016 (43 years), the number of floral days increased for the B. flav- ifrons floral assemblage by 5.9 � 2.7 days (LM, F1,36 = 5.0, P = 0.032; Fig. 4c), while the number of floral days did not change significantly for the B. bifarius and B. appositus floral
© 2017 John Wiley & Sons Ltd/CNRS
4 J. E. Ogilvie et al. Letter
assemblages (LMs, B. bifarius: F1,36 = 1.1, P = 0.31; and B. appositus: F1,36 = 2.7, P = 0.11; Fig. 4c). In addition, the annual floral sum of each bee species’ plant assemblage did not change directionally over the four decades (LMs, B. bifar- ius: F1,36 = 0.3, P = 0.57; B. flavifrons: F1,36 = 0.03, P = 0.85; and B. appositus: F1,36 = 0.8, P = 0.37; Fig. 4d). Additional analyses showed that the total floral season length – which differed from the number of floral days by including all days from first flowers to the date on which 80% of flowers had accumulated – increased over 43 years by 20.9 � 5.2 and 20.8 � 5.3 days for the B. bifarius and B. flavifrons floral assemblages, respectively (LMs with year as the fixed effect; B. bifarius: F1,36 = 15.9, P = 0.0003; B. flavifrons: F1,36 = 15.6, P = 0.0003) and marginally so for the B. apposi- tus floral assemblage by 10.4 � 5.4 days (F1,36 = 3.6, P = 0.064; Fig. S7). Furthermore, the number of days below the flower threshold (low floral days) significantly increased through time for the B. bifarius and B. flavifrons floral assem- blages by 12.4 � 3.5 and 8.7 � 3.1 days, respectively (LMs with year as the fixed effect, B. bifarius: F1,36 = 12.7, P = 0.001; B. flavifrons: F1,36 = 7.8, P = 0.008), and margin- ally so for the B. appositus flowers by 6.15 � 3.4 days (F1,36 = 3.2, P = 0.081; Fig. S7). Thus, the trend for earlier snowmelt dates is increasing the length of the floral season; however, those additional days are of low floral abundance for B. bifarius and B. appositus, while there are days both below and above the flower threshold for B. flavifrons. In the 8-year dataset, abundance of all three bee species was nega- tively affected by the number of low floral days (negative binomial GLMs with number of days ≤ 0.75 flowers/m2 and site as fixed effects, all three species: P < 0.0001, N = 24 site-years; Fig. S8), though this effect was only significant with the inclusion of an extreme year with many low floral days.
Precipitation Floral days
Bombus bifarius abundance
Precipitation Floral days
Bombus appositus abundance
Precipitation Floral days
Bombus avifrons abundance
P < 0.05 P < 0.005
P < 0.0005
Figure 2 Path diagrams showing the climate and flower variables that
govern interannual variation in the peak abundance of (a) Bombus
bifarius, (b) B. flavifrons and (c) B. appositus. Paths between variables
included in the best-fitting piecewise structural equation models are
shown. The faint arrow indicates an insignificant path included in a
model; solid and dashed arrows indicate a positive and negative effect of
a variable on another, respectively; and the arrow thickness indicates the
significance level of the path adjusted for multiple comparisons (thick:
P < 0.0005, medium: P < 0.005, thin: P < 0.05). Bee data are from three sites in each of 8 years (2009–2016, N = 24); flower and climate data are from a single central site, though the floral variables were created from
plant species lists specific to the three bee sampling sites.
10 20 30 40 50
10 20 30 40 50 10 20 30 40 50
. b ee
(a) (b) (c) Site A Site B
Figure 3 The relationship between annual peak abundance (bees netted/hour) and the number of days above a flower threshold (0.75/m2) in a season for
three bumble bee species, (a) Bombus bifarius, (b) B. flavifrons and (c) B. appositus. The number of floral days was identified as the key driver of bumble
bee abundance for all three bee species in our structural equation models (Fig. 2). Bee data are from three sites over 8 years, and floral data are from a
central site with plant species lists specific to the three bee sampling sites (2009–2016, N = 24). The different shaped points are different sites (squares = site A, triangles = site B, circles = site D).
© 2017 John Wiley & Sons Ltd/CNRS
Letter Climate and flower effects on bumble bees 5
There is evidence that climate change has influenced some bumble bee populations (e.g. Ploquin et al. 2013; Kerr et al. 2015), but the mechanisms underlying those changes – whether direct or indirect climate effects – remain unclear. We found that interannual abundances of three subalpine bumble bee species were driven by the indirect effects of climate on the temporal distribution of floral resources. In particular, bee abundance was most strongly positively related to the number of days above a low flower threshold. Our study suggests that climate-driven alterations in floral phenology can play a criti- cal role in governing bumble bee population responses to ongoing global change.
Direct and indirect climate effects on bees
Although animal populations can show direct responses to cli- mate variation (e.g. Bale et al. 2002; Roland & Matter 2016), there is growing recognition that the indirect effects on species interactions are more common (Ockendon et al. 2014). Our results add to these prior studies by demonstrating strong indirect effects of climate on bumble bees, providers of
important pollination services. Our SEM approach did not detect any direct climate effects on interannual bumble bee abundances; instead, climate effects were all indirect. Precipi- tation and temperature (which strongly co-varied with snow- melt date and precipitation) can influence daily bumble bee foraging activity (e.g. Bergman et al. 1996) and may cause direct mortality in extreme events (Oyen et al. 2016), while snowmelt date contributes to overwintered queen emergence (Kudo & Ida 2013). However, the interannual variation in these climate measures did not directly affect bee abundances in our study. Much more is known about how climate affects plants than how climate affects bees – in part because of a lack of long-term data on bee populations – so it is possible that our climate variables did not capture aspects most impor- tant to bumble bees. This may be unlikely, however, because many of the climate variables we considered in preliminary analyses were correlated, so additional variables are likely to provide similar conclusions. Furthermore, bumble bees can tolerate some climate variation: individuals can fly over a broad temperature range (Heinrich 1979), and underground colonies are buffered from temperature and precipitation extremes. This is unlike egg or larval butterflies that may be exposed to and affected by temperature extremes (Boggs &
1975 1985 1995 2005 2015
1975 1985 1995 2005 20151975 1985 1995 2005 2015
1975 1985 1995 2005 2015 Year
Figure 4 Patterns in (a and b) climate variables and (c and d) the bumble bee-visited floral assemblages spanning 43 years (1974–2016). Long-term climate panels show (a) day of year of snowmelt and (b) cumulative precipitation (rain and snow) from May to July (N = 42 years). The bumble bee floral assemblage panels show (c) the number of days above a flower threshold (0.75/m2) and (d) the cumulative sum of flowers (to 80% of the season total), in
which triangles are Bombus bifarius, circles are B. flavifrons, and crosses are B. appositus (N = 38 years for each species; flower counts made approximately every second day pooled across 23 2 9 2 m plots). The lines are fitted from linear models – marginally significant in (a) (P = 0.0781) and significant in (c) for B. flavifrons (P = 0.032).
© 2017 John Wiley & Sons Ltd/CNRS
6 J. E. Ogilvie et al. Letter
Inouye 2012; Roland & Matter 2016). The observed climate variation did, however, strongly affect the abundance and phenology of the floral assemblages, consistent with other studies on single plant species (Iler et al. 2013) and the com- munity at our site (CaraDonna et al. 2014). Although climate- driven variation in floral resources is known to influence bee abundance (Thomson 2016), we show for the first time that the effects of climate on floral days and floral abundance have a stronger influence on bee abundance than direct climate effects. Although research has highlighted the positive effect of flo-
ral abundance on bee populations (reviewed in Roulston & Goodell 2011), few studies have explicitly considered the effect of within-season temporal resource distribution (floral resource phenology) on bee population size or reproductive output (Crone 2013). Those studies that have considered tem- poral resource distribution effects find slower or reduced bum- ble bee brood production with periods of low resources (Schmid-Hempel & Durrer 1991; Kudo 2014). We found that the number of days above a flower threshold – a measure of the time available with sufficient floral resources – had a strong positive effect on the interannual abundance of all three bees, while annual floral sum had a weak effect on only two species. This effect could be weak because our floral abundance measure did not come from the exact sites where bees were sampled. However, the sites at which flowers and bees were sampled contained the same habitats and plant spe- cies, and we are confident that the among-year variation in floral abundance is greater than the among-site variation. Future studies of floral resource effects on pollinators need to consider more nuanced ways of deconstructing flowering phe- nology, as we have done, because temporal resource distribu- tion is multifaceted and the critical components will be organism-specific. The time available for foraging may limit animal reproduc-
tive output more commonly than is appreciated (Rose & Lyon 2013). Subalpine bumble bees have short seasons within which to grow and reproduce: there are typically 10 weeks between queen emergence and the appearance of males at our site (Elliott 2009b), and our study supports suggestions that bum- ble bees at high altitude may be limited by the time to exploit floral resources (Pyke 1982; Elliott 2009a). The more days with sufficient flowers, the more workers can forage and pro- vision brood and colonies can grow. In a similar vein, with many days below a flower threshold, bee abundances are lower, perhaps because resource gaps cause spring queens to starve or result in insufficient floral resources to maintain col- ony growth (Kudo 2014). Indeed, bumble bees should be vul- nerable to periods of resource shortage because they have limited food storage for withstanding low floral abundance periods (Goulson 2010). Experimental studies that examine how the timing and magnitude of resource gaps affect colony initiation, worker production and reproductive output are needed to understand how global change will influence bum- ble bees. Two caveats are important to consider when interpreting
our study. First, a single extreme year was important in deter- mining the observed patterns. As extreme weather is predicted to increase with climate change, there is a critical need for
continued long-term monitoring of bee populations that allow us to capture these rare but increasing important events. Sec- ond, climate variation may have other indirect effects on bee populations that we did not measure. Populations can be lim- ited by top-down instead of bottom-up forces, and this rela- tionship may be modified by climate (Hoekman 2010). For example, with warmer temperatures, the benefit of increased foraging opportunities for mason bees was negated by increases in wasp parasitism (Forrest & Chisholm 2017). Bum- ble bee populations could also be influenced by predators and intra- and interspecific competition, factors we were unable to capture here.
Changes in spring snowmelt timing and measures of floral phenology over four decades (see also Aldridge et al. 2011; CaraDonna et al. 2014) suggest species-specific and potentially conflicting effects on bumble bee populations. First, based simply on long-term trends in floral resources, our results sug- gest that B. bifarius and B. appositus populations may have remained consistent over the last four decades given the lack of directional change in the number of floral days of their flo- ral assemblages. In comparison, increases in the number of floral days for B. flavifrons floral assemblages suggest the potential for increases in its populations. Thus, the reshaping of flowering communities with climate change (CaraDonna et al. 2014) may also indirectly contribute to the reshaping of pollinator communities, alongside climate-induced changes in pollinator distributions (Kerr et al. 2015). Second, however, advancing snowmelt dates have lengthened the flowering sea- son, and the number of low floral abundance days have simul- taneously increased, perhaps, in part, due to increased incidences of damaging spring frosts (Inouye 2008), or increasing summer temperatures and drought conditions (Aldridge et al. 2011). Because resource gaps may negatively affect bumble bee populations, and potentially those of other pollinators that have long foraging seasons (e.g. broad-tailed hummingbirds), continued increases in low resource days may negate any benefits of an extending flowering season. In our system, such low floral resource days occur when bumble bees may be vulnerable to resource deficits: in the spring, when queen bees are initiating nests (Schmid-Hempel & Durrer 1991; Williams et al. 2012; Kudo 2014), and in the mid-season before the summer peak of flowers, when colonies are provi- sioning for reproductive brood. It is unknown which floral resource variable is most important to bumble bee abundance – days above or days below a flower threshold – but given that floral phenology is changing through time, addressing this question should be a priority for future research. We sug- gest that resource phenology is likely to affect consumer pop- ulations more broadly than is appreciated, especially in the context of climate change.
We thank the many field assistants who helped collect data; the exceptional billy barr for use of his climate data; the Irwin, Inouye, and Underwood lab groups and Paul
© 2017 John Wiley & Sons Ltd/CNRS
Letter Climate and flower effects on bumble bees 7
CaraDonna for advice; Jonathan Lefcheck for statistical help; and Ignasi Bartomeus and two anonymous reviewers for insightful comments that improved the manuscript. Funding was provided by the National Science Foundation grants DEB-9408382, IBN-9814509, and DEB-0238331 to DWI; DEB-0922080 to DWI and REI; DEB-1354104 to DWI, REI, BDI and NU; and funds from North Carolina State Univer- sity to REI. We thank the Rocky Mountain Biological Labo- ratory for logistical support, and the John Tuttle family and the Gunnison National Forest for access to some study sites. Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.
DWI and REI designed research; ZJG, SRG, DWI, REI and JEO performed research; JEO, SRG, BDI and NU conducted statistical analyses; JEO and SRG wrote the first draft of the manuscript; all authors provided feedback on analyses and the manuscript.
DATA ACCESSIBILITY STATEMENT
The primary data used in this study are archived at Figshare: http://dx.doi.org/10.6084/m9.figshare.5400796 (Ogilvie et al. 2017).
Aldridge, G., Inouye, D.W., Forrest, J.R.K., Barr, W.A. & Miller-
Rushing, A.J. (2011). Emergence of a mid-season period of low floral
resources in a montane meadow ecosystem associated with climate
change. J. Ecol., 99, 905–913. Bale, J.S., Masters, G.J. & Hodkinson, I.D. (2002). Herbivory in global
climate change research: direct effects of rising temperature on insect
herbivores. Glob. Change Biol., 8, 1–16. Bartomeus, I., Ascher, J.S., Wagner, D., Danforth, B.N., Colla, S.,
Kornbluth, S. et al. (2011). Climate-associated phenological advances
in bee pollinators and bee-pollinated plants. Proc. Natl Acad. Sci. USA,
108, 20645–20649. Bergman, P., Molau, U. & Holmgren, B. (1996). Micrometeorological
impacts on insect activity and plant reproductive success in an alpine
environment, Swedish Lapland. Arct. Alp. Res., 28, 196–202. Bingham, R.A. & Orthner, A.R. (1998). Efficient pollination of alpine
plants. Nature, 391, 238–239. Boggs, C.L. & Inouye, D.W. (2012). A single climate driver has direct
and indirect effects on insect population dynamics. Ecol. Lett., 15, 502– 508.
Cameron, S.A., Lozier, J.D., Strange, J.P., Koch, J.B., Cordes, N., Solter,
L.F. et al. (2011). Patterns of widespread decline in North American
bumble bees. Proc. Natl Acad. Sci. USA, 108, 662–667. CaraDonna, P.J., Iler, A.M. & Inouye, D.W. (2014). Shifts in flowering
phenology reshape a subalpine plant community. Proc. Natl Acad. Sci.
USA, 111, 4916–4921. Crone, E.E. (2013). Responses of social and solitary bees to pulsed floral
resources. Am. Nat., 182, 465–473. Crone, E.E. & Williams, N.M. (2016). Bumble bee colony dynamics:
quantifying the importance of land use and floral resources for colony
growth and queen production. Ecol. Lett., 19, 460–468. Elliott, S.E. (2009a). Surplus nectar available for subalpine bumble bee
colony growth. Environ. Entomol., 38, 1680–1689.
Elliott, S.E. (2009b). Subalpine bumble bee foraging distances and
densities in relation to flower availability. Environ. Entomol., 38, 748– 756.
Forrest, J.R.K. & Chisholm, S.P.M. (2017). Direct benefits and indirect
costs of warm temperatures for high-elevation populations of a solitary
bee. Ecology, 98, 359–369. Garratt, M.P.D., Coston, D.J., Truslove, C.L., Lappage, M.G., Polce, C.,
Dean, R. et al. (2014). The identity of crop pollinators helps target
conservation for improved ecosystem services. Biol. Conserv., 169, 128– 135.
Geib, J.C., Strange, J.P. & Galen, C. (2015). Bumble bee nest abundance,
foraging distance, and host-plant reproduction: implications for
management and conservation. Ecol. App., 25, 768–778. Gezon, Z.J., Wyman, E.S., Ascher, J.S., Inouye, D.W. & Irwin, R.E.
(2015). The effect of repeated, lethal sampling on wild bee abundance
and diversity. Methods Ecol. Evol., 6, 1044–1054. Goulson, D. (2010). Bumblebees: Behaviour, Ecology, and Conservation,
2nd edn. Oxford University Press, Oxford.
Goulson, D., Nicholls, E., Bot�ıas, C. & Rotheray, E.L. (2015). Bee
declines driven by combined stress from parasites, pesticides, and lack
of flowers. Science, 347, 1255957.
Grace, J.B., Schoolmaster, D.R. Jr., Guntenspergen, G.R., Little, A.M.,
Mitchell, B.R., Miller, K.M. et al. (2012). Guidelines for a graph-
theoretic implementation of structural equation modeling. Ecosphere, 3,
Hegland, S.J. & Totland, Ø. (2008). Is the magnitude of pollen limitation
in a plant community affected by pollinator visitation and plant species
specialisation levels? Oikos, 117, 883–891. Heinrich, B. (1979). Bumblebee Economics. Harvard University Press,
Hoekman, D. (2010). Turning up the heat: temperature influences the
relative importance of top-down and bottom-up effects. Ecology, 91,
2819–2825. Holland, J.G. & Bourke, A.F.G. (2015). Colony and individual life-
history responses to temperature in a social insect pollinator. Funct.
Ecol., 29, 1209–1217. Høye, T.T., Post, E., Schmidt, N.M., Trøjelsgaard, K. & Forchhammer,
M.C. (2013). Shorter flowering seasons and declining abundance of
flower visitors in a warmer Arctic. Nat. Clim. Change, 3, 759–763. Iler, A.M., Høye, T.T., Inouye, D.W. & Schmidt, N.M. (2013). Nonlinear
flowering responses to climate: are species approaching their limits of
phenological change? Philos. Trans. R. Soc. Lond. B Biol. Sci., 368,
Inari, N., Hiura, T., Toda, M.J. & Kudo, G. (2012). Pollination linkage
between canopy flowering, bumble bee abundance and seed production
of understorey plants in a cool temperate forest. J. Ecol., 100, 1534– 1543.
Inouye, D.W. (1980). The effect of proboscis and corolla tube lengths on
patterns and rates of flower visitation by bumblebees. Oecologia, 45,
197–201. Inouye, D.W. (2008). Effects of climate change on phenology, frost
damage, and floral abundance of montane wildflowers. Ecology, 89,
353–362. IPCC. (2013). Summary for Policymakers. In: Climate Change 2013: The
Physical Science Basis. Contribution of Working Group I to the Fifth
Assessment Report of the Intergovernmental Panel on Climate Change
(eds Stocker, T.F., Qin, D., Plattner, G.K., Tignor, M., Allen, S.K. &
Boschung, J. et al. ). Cambridge University Press, Cambridge, United
Kingdom and New York, NY, pp. 3–29. Kerr, J.T., Pindar, A., Galpern, P., Packer, L., Potts, S.G., Roberts, S.M.
et al. (2015). Climate change impacts on bumblebees converge across
continents. Science, 349, 177–180. Kudo, G. (2014). Vulnerability of phenological synchrony between plants
and pollinators in an alpine ecosystem. Ecol. Res., 29, 571–581. Kudo, G. & Ida, T.Y. (2013). Early onset of spring increases the
phenological mismatch between plants and pollinators. Ecology, 94,
© 2017 John Wiley & Sons Ltd/CNRS
8 J. E. Ogilvie et al. Letter
Lambert, A.M., Miller-Rushing, A.J. & Inouye, D.W. (2010). Changes in
snowmelt date and summer precipitation affect the flowering phenology
of Erythronium grandiflorum (glacier lily; Liliaceae). Am. J. Bot., 97,
1431–1437. Lefcheck, J.S. (2015). piecewiseSEM: piecewise structural
equation modelling in R for ecology, evolution, and systematics.
Methods Ecol. Evol., 7, 573–579. Miller-Struttmann, N.E., Geib, J.C., Franklin, J.D., Kevan, P.G., Holdo,
R.M., Ebert-May, D. et al. (2015). Functional mismatch in a bumble
bee pollination mutualism under climate change. Science, 349, 1541– 1544.
Nogu�es-Bravo, D., Ara�ujo, M.B., Errea, M.P. & Mart�ınez-Rica, J.P.
(2007). Exposure of global mountain systems to climate warming
during the 21st Century. Glob. Environ. Change, 17, 420–428. Ockendon, N., Baker, D.J., Carr, J.A., White, E.C., Almond, R.E.A.,
Amano, T. et al. (2014). Mechanisms underpinning climatic impacts on
natural populations: altered species interactions are more important
than direct effects. Glob. Change Biol., 20, 2221–2229. Ogilvie, J.E., Griffin, S.R., Gezon, Z.J., Inouye, B.D., Underwood, N.,
Inouye, D.W. & Irwin, R.E. (2017). Data from: Interannual bumble
bee abundance is driven by indirect climate effects on floral resource
phenology. Figshare, doi:10.6084/m9.figshare.5400796.
Oyen, K.J., Giri, S. & Dillon, M.E. (2016). Altitudinal variation in
bumble bee (Bombus) critical thermal limits. J. Therm. Biol, 59, 52–57. Parmesan, C. (2006). Ecological and evolutionary responses to recent
climate change. Annu. Rev. Ecol. Evol. Syst., 37, 637–669. Ploquin, E.F., Herrera, J.M. & Obeso, J.R. (2013). Bumblebee
community homogenization after uphill shifts in montane areas of
northern Spain. Oecologia, 173, 1649–1660. Pyke, G.H. (1982). Local geographic distributions of bumblebees near
Crested Butte, Colorado: competition and community structure.
Ecology, 63, 555–573. Pyke, G.H., Inouye, D.W. & Thomson, J.D. (2011). Activity and
abundance of bumble bees near Crested Butte, Colorado: diel, seasonal,
and elevation effects. Ecol. Entomol., 36, 511–521. Pyke, G.H., Thomson, J.D., Inouye, D.W. & Miller, T.J. (2016). Effects
of climate change on phenologies and distributions of bumble bees and
the plants they visit. Ecosphere, 7, 1–19. R Core Team. (2016). R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria.
Available at: http://www.R-project.org/. Last accessed 23 January 2017.
Roland, J. & Matter, S.F. (2016). Pivotal effect of early-winter
temperatures and snowfall on population growth of alpine Parnassius
smintheus butterflies. Ecol. Monogr., 86, 412–428. Rose, A.P. & Lyon, B.E. (2013). Day length, reproductive effort, and the
avian latitudinal clutch size gradient. Ecology, 94, 1327–1337.
Roulston, T.H. & Goodell, K. (2011). The role of resources and risks in
regulating wild bee populations. Annu. Rev. Entomol., 56, 293–312. Schmid-Hempel, P. & Durrer, S. (1991). Parasites, floral resources and
reproduction in natural populations of bumblebees. Oikos, 62, 342–350. Shipley, B. (2009). Confirmatory path analysis in a generalized multilevel
context. Ecology, 90, 363–368. Shipley, B. (2013). The AIC model selection method applied to path
analytic models compared using a d-separation test. Ecology, 94, 560– 564.
Smith, C.E. & Cribbie, R.A. (2013). Multiplicity control in structural
equation modeling: incorporating parameter dependencies. Struct. Equ.
Modeling, 20, 79–85. Thomson, D.M. (2016). Local bumble bee decline linked to recovery of
honey bees, drought effects on floral resources. Ecol. Lett., 19, 1247– 1255.
Van der Putten, W.H., Macel, M. & Visser, M.E. (2010). Predicting
species distribution and abundance responses to climate change: why it
is essential to include biotic interactions across trophic levels. Philos.
Trans. R. Soc. Lond. B Biol. Sci., 365, 2025–2034. Venables, W.N. & Ripley, B.D. (2002). Modern Applied Statistics with S,
4th edn.. Springer, New York.
Vesterlund, S.-R. & Sorvari, J. (2014). Longevity of starved bumblebee
queens (Hymenoptera: Apidae) is shorter at high than low
temperatures. Eur. J. Entomol., 111, 217–220. Williams, P., Colla, S. & Xie, Z. (2009). Bumblebee vulnerability:
common correlates of winners and losers across three continents.
Conserv. Biol., 23, 931–940. Williams, N.M., Regetz, J. & Kremen, C. (2012). Landscape-scale
resources promote colony growth but not reproductive performance of
bumble bees. Ecology, 93, 1049–1058. Williams, P.H., Thorp, R.W., Richardson, L.L. & Colla, S.R. (2014).
Bumble bees of North America: An identification guide. Princeton
University Press, Princeton.
Additional Supporting Information may be found online in the supporting information tab for this article.
Editor, Andrew Bourke Manuscript received 10 April 2017 First decision made 21 May 2017 Manuscript accepted 31 August 2017
© 2017 John Wiley & Sons Ltd/CNRS
Letter Climate and flower effects on bumble bees 9