The effect of repeated, lethal sampling on wild bee
abundance and diversity
Zachariah J. Gezon
, John S. Ascher
Rebecca E. Irwin
Department of Biological Sciences, Dartmouth College, 78 College St, Hanover, NH 03755, USA;
Rocky Mountain Biological
Laboratory, Crested Butte, CO 81224, US A;
Division of Invertebrate Zoology, American Museum of Natural History, New
York, NY 1 0024, USA;
Department of Biological Sciences, National University of Singapore, 14 Science Drive 4, Singapore
117543, Singapore; and
Department of Biology, University of Maryland, College Park, MD 20742, USA
1. Bee pollinators provide a critical ecosystem service to wild and agricultural plants but are reported to be
declining world-wide due to anthropogenic change. Long-term data on bee abundance and diversity are scarce,
and the need for additional quantitative sampling using repeatable methods has been emphasized. Recently,
monitoring programmes have begun using a standardized method that employs a combination of pan traps and
sweep netting, resulting in lethal sampling of bees. This standardized method can remove a large number of b ees
from sites during each sampling day, raising concern that the sampling itself could have a negative eﬀect on bee
2. We conducted an experiment to assess whether lethal sampling for bees using pan traps and netting aﬀected
bee abundance and diversity when done every two weeks throughout a season and over multiple years. We com-
pared bee abundance, richness, evenness and functional group composition between sites that had been sampled
every two weeks from 2009 to 2012 to similar sites not previously sampled.
3. We found that the standardized method for sampling bees, with specimens from 132 morphospecies, did not
aﬀect bee communities in terms of abundance, rareﬁed richness, evenness, or functional group composition.
Thus, our results indicate that the bee communities we sampled are robust to such sampling eﬀorts, despite
removing an average of 2862 bees per season.
4. We discuss several explanations for why sampling did not aﬀect bee abundance or community structure,
including a density-dependent response to reduced competition for resources.
5. These results suggest that bee monitoring programmes sampling once every two weeks with pan traps and net-
ting will not aﬀect bee community structure. We urge researchers monitoring bees to utilize standardized proto-
cols so that results can be compared across space and time.
Key-words: bee bowl, native bees, over-collecting, over-sampling, pan traps, Pollination, pollinator
Environmental monitoring programmes are fundamental to
ecological and environmental sciences, and play a crucial role
in informing science-based environmental policy (Lovett et al.
2007). Environmental monitoring programmes are commonly
used to assess environmental change, such as the eﬀects of sub-
urbanization on pollinator communities (Carper et al. 2014),
the eﬀects of habitat restoration on insect food webs (Albrecht
et al. 2007), and the impact of dams on river benthic inverte-
brate communities (Statzner et al. 2001; Chaves-Ulloa, Uman-
a-Villalobos & Springer 2014). Many environmental
monitoring programmes require the collection and killing of
organisms due to the diﬃculty of identifying specimens in the
ﬁeld (Reynoldson & Metcalfe-Smith 1992; Hawkins et al.
2000; Schonberg et al. 2004), creating concern that the moni-
toring itself may negatively impact the study taxa or system,
and that changes do cumente d by monit oring prog rammes
may be caused by the monitoring eﬀorts themselves. Monitor-
ing of wild bees is of particular interest due to the pollination
services they provide (Gallai et al. 2009) and the perception
that bees are in decline (Potts et al. 2010), but most bees are
diﬃcult to identify in the ﬁeld, necessitating collection to make
or conﬁrm species identiﬁcation (Tepedino & Stanton 1981;
Gibbs et al. 2013). The aim of this study was to determine
whether a commonly employed bee monitoring protocol using
lethal sampling techniques aﬀected bee abundance and diver-
sity when used repeatedly over several years.
Pollination by insects is an essential ecosystem service in
both natural and agricultural systems. Approximately 90% of
ﬂowering species world-wide rely on pollination by insects and
other animals for reproduction (Ollerton, Winfree & Tarrant
*Corresponding author. E-mail: firstname.lastname@example.org
© 2015 The Authors. Methods in Ecology and Evolution © 2015 British Ecological Society
Methods in Ecology and Evolution 2015 doi: 10.1111/2041-210X.12375
2011), and pollinators provide approximately $202 billion per
year globally in pollination services to agricultural systems
(Gallai et al. 2009). Despite the ecological and economic
importance of insect pollinators, and of bees in particular, little
is known about their population trends beyond what can be
learned from comparisons to historical records or museum
specimens (Bartomeus et al. 2013; Burkle, Marlin & Knight
2013), prompting the development of standardized protocols
for sampling bees (e.g. LeBuhn et al. 2003; Westphal et al.
2008). Concerns about global pollinator declines along with
the development of standardized protocols have motivated the
collection of long-term bee pollinator data using monitoring
programmes (Committee on the Status of Pollinators in North
America, & National Research Council 2007; Westphal et al.
2008) and the proposal of large-scale, long-term community
eﬀorts (Lebuhn et al. 2013).
One commonly used standardized bee catching protocol
(outlined in Materials and Methods) (e.g. Wilson, Messinger &
Griswold 2009; Neame, Griswold & Elle 2013; Winfree et al.
2014) employs a combination of sweep netting and pan traps
(small plastic b owls pain ted ﬂ uorescent yellow, ﬂuorescent
blue, or white, ﬁlled with soapy water). Pan traps typically
account for a large percentage of bees caught (Wilson, Gris-
wold & Messinger 2008; Popic, Davila & Wardle 2013) and, by
design, kill the sampled bees. Thus, a large number of bees can
be collected each time the protocol is used, and bees are typi-
cally sampled repeatedly in the same site over the course of a
season and sometimes for multiple years (e.g. Wilson, Gris-
wold & Messinger 2008; Kearns & Oliveras 2009; Jha & Van-
dermeer 2010). A concern is that high bee capture rates of
certain taxa, such as primitively eusocial Lasioglossum (Dialic-
tus), obtained using pan traps and netting could aﬀect bee
abundance, diversity, and community composition. Uninten-
tionally aﬀecting wild bee populations would be problematic
because th e me thods could (i) drive long-term patterns in mea-
sured bee abundance, diversity, and community composition,
thus sampling would not accurately reﬂect how unsampled bee
populations and communities would have responded to their
natural environment, and (ii) negatively impact pollination in,
and near, the study sites. Although concern has been voiced
about the potential eﬀects of over-collecting bees (Shepherd
et al. 2003; Tepedino et al. 2015), little work to date has inves-
tigated the long-term impacts of pan trapping and netting.
The purpose of this study was to assess whether repeated use
of pan trapping and netting had an impact on abundance and
diversity of bee populations and communities. We used two
approaches. First, we compared trends in bee abundance over
5 years of sampling from sites that were sampled every
2 weeks throughout the ﬂight season for bees. We predicted
that if repeated sampling negatively aﬀected bee populations,
bee abundance would decrease both over the course of a sam-
pling season, and from one year to the next. Secondly, we used
an experiment in which we compared bee abundance, species
richness and evenness at sites that had been sampled twice per
week for 4 years to sites that had never been sampled before.
We predicted that if bee communities were aﬀected by repea-
ted pan trapping and netting, bee abundance and diversity
(richness and evenness) would be lower in sites that had been
repeatedly sampled than in newly sampled sites.
Materials and methods
We conducted this study in the West Elk Mountains of southwest
Colorado, in Gunnison County near the Rocky Mountain Biological
Laboratory (RMBL). The RMBL (2886 m elevation, 38°57
W) is highly seasonal, with a growing season typically from late
April or May until September (CaraDonna, Iler & Inouye 2014). Open
meadow habitats around the RMBL are dominated by numerous
perennial ﬂowering plant species, many of which are visite d by bee p oll-
inators (Forrest, Inouye & Thomson 2010). Pollinator communities
around the RMBL are dominated by native bees (Inouye 1977; Forrest
& Thomson 2011); the European Honey Bee Apis mellifera (Hymenop-
tera: Apidae) is absent in this study system, and other non-native bee
species nearly so.
We studied a total of 26 sites in three river valleys: East River, Slate
River, and Washington Gulch (Appendix S1 and Fig. S1). To test
whether bee sampling aﬀected intra- and inter-annual variation in bee
abundance, we used nine sites, all located in the East River valley
(Appendix S1). For ﬁve consecutive years (2009–2013), these sites were
sampled every two weeks throughout the ﬂowering season. Each site
was classiﬁed as either wet or dry (3 sites) meadow, and wet meadows
were further divided into wet-Veratrum (3 sites) and wet-Salix (3 sites)
based on dominant vegetation (Veratrum tenuipetalum and shrubby
Salix spp ., respectively). Wet and d ry meadow are two dominant habi-
tat types around the RMBL (Langenheim 1962). To test whether sites
that were sampled repeatedly every two weeks for three years versus
only sampled once diﬀered in bee abundance and diversity in 2012, we
used 23 sites. Seventeen were sampled only once (11 dry meadow and
six wet-Veratrum sites), and were located in the East, Slate, and Wash-
ington Gulch river valleys (Appendix S1). We compared these ‘single-
sample’ sites to sites of similar habitats that had been sampled every
2 weeks from 2009 to 2012 (the three wet-Veratrum and three dry mea-
dow sites described above, Appendix S1). D uring the summer of 2012,
each time we sampled a repeat-sample site, we would sample a compa-
rable single-sample site within two days. We then assigned a numeric
‘sample week’ to the single-sample and repeat-sample sites that were
blocked in time, for a total of seven sample weeks.
Each site consisted of two, ﬁxed 45 m transects at approximately right
angles. Pan traps were made from 9 cm diameter plastic bowls (Fisher-
; Hexagonal Polystyrene Weighing Dishes, Pittsburgh, PA,
USA) painted ﬂuorescent yellow or ﬂuorescent blue using Krylon
Fluorescent Paint (colour numbers 3104 and 3109, respectively), or left
unpainted (white). On each sample day, pan traps ﬁlled with soapy
water (water with a small amount of Dawn
Ultra Dishwashing soap,
original scent) were placed every 3 m alon g each transect with pan-trap
colours assigned at random (as in LeBuhn et al. 2003). Pan traps were
left in place from ~0800 to 1700, the peak hours of pollinator activity
around the RMBL, after which the contents were strained and stored
in 75% EtOH until processing. All bees were washed in water, dried
with a hair drier, pinned and labelled as in LeBuhn et al. (2003).
© 2015 The Authors. Methods in Ecology and Evolution © 2015 British Ecological Society, Methods in Ecology and Evolution
2 Z. J. Gezon et al.
Pan traps typically catch many smaller bees but fail to consistently
capture larger bodied, stronger ﬂying bees including some Megachili-
dae and Apidae (Cane, Minckley & Kervin 2000). Thus, in conjunction
with the pan traps, we systematically walked each transect, netting for
bees for one hour each in the morning and afternoon when pan traps
were in place (as outlined in LeBuhn et al. 2003). Netted bees that
could not be identiﬁed to species in the ﬁeld were c ollected, pinned, and
labelled. We could identify most bumblebees (Bombus)tospeciesinthe
ﬁeld based on pile colours, with reference to the RMBL insect collec-
tion. Ide ntiﬁed bumblebees were marked with a dot of indelible ink on
their thorax to avoid recounting released individuals.
Pinned and labelled bees from the pan traps and netting were identi-
ﬁed to species or morphospecies, although we were unable to identify
some cryptic taxa beyond genus or subgenus, notably Lasioglossum
(Dialictus)andL. (Evylaeus sensu lato).
To control for potential diﬀerences in biotic and abiotic conditions
between sites sampled repeatedly versus once, we collected the follow-
ing covariates each time we sampled for bees. First, along the transects
at each site, we measured ﬂoral abundance using three, ﬁxed, 4 m
ﬂower census plots (six per site). On the same days that the sites were
sampled for bees, ﬂower censuses were conducted within all plots by
recording all ﬂowering plant species, number of ﬂowers per plant for
up to ten plants per species, and number of plants per plot. Secondly,
we recorded daily maximum temperatures during each sampling period
from a c entrally located weather station near the RMBL (38
36″W, 2945 m elevation). Temperature can vary dramatically
at these subalpine sites, with strong eﬀects on insect ﬂight and activity
(Corbet et al. 1993).
We performed separate analyses for bees that were captured and killed
(hereafter referred to as ‘vouchered’), bees that were netted, identiﬁed
in the ﬁeld and released alive (hereafterreferredtoas‘released’),aswell
as with all bee data combined (hereafter referred to as ‘combined’).
Only vouchered specimens are included in the analyses described below
except where otherwise indicated. Due to extreme drought conditions
in 2012, abundance of Bombus workers was extremely low at all sites
(both those that were sampled repeatedlyaswellasthoseonlysampled
once); thus, an analysis of only the re leased bees was not possible. For
calculated as bees caught per hour (bees/hour). For vouchered bees, we
averaged the capture rates for bees caught in pan traps and bees caught
by net. Floral density (ﬂowers m
) was calculated by multiplying the
average num ber of ﬂowers per plant for each species by the number of
plants per plot, summing across species, and dividing by four (each plot
was 4 m
). Plot means were then averaged at each site for each sample
bout to calculate site mean ﬂoral density on each sampling day. Bee
species richness was rareﬁed to the lowest common abundance across
the combined data using the vegan package in R version 3.0.3 (R
Development Core Team 2011; Oksanen et al. 2012), and evenness
was calculated as E
(Smith & Wilson 1996). E
ranges from 0 to 1,
with 0 representing minimum evenness and 1 maximum evenness
(Smith & Wilson 1996).
We also wanted to determine whether community composition of
particular bee functional groups (Winfree, Griswold & Kremen 2007;
Williams et al. 2010) was aﬀected by pan trapping. We categorized
the nesting habits of eac h species (soil, wood, pithy stems, constructed
cells, cavity or hive), diet (pollen generalists vs. pollen specialists), and
sociality (eusocial vs. not eusocial) based on information compiled
from published literature (Mitchell 1960; Hurd 1979; Cane, Griswold
& Parker 2007; Michener 2007) and from extensive ﬁeld experience by
J.A. and E.W. We used Carper et al. (2014) as the basis for our nest-
ing substrate designations, with the addition of the category ‘con-
structed cells’ for bees that construct cells on exposed surfaces using
resins, masticated leaves, and other materials. We diﬀerentiated
between bees that nest in existing cavities in wood (‘cavity’) and those
that excavate nests in this substrate (‘wood’), although the results were
not qualitatively aﬀected when ‘cavity’ and ‘wood’ were combined
(analyses not shown). We classiﬁed all non-parasitic Bombus as ‘hive’
nesters due to the heterogeneity of nesting substrates used by Bombus,
even within species. Parasitic bees were included in the nesting cate-
gory of their host. The category ‘diet’ refers to pollen specialists versus
pollen generalists, with ‘specialist’ referring to known oligolectic spe-
cies,aswellastaxaforwhichavailable evidence suggests oligolecty or
mesolecty. ‘Sociality’ was broken into two categories: ‘eusocial’ and
‘not eusocial’, with ‘eusocial’ containing only non-parasitic Bombus
spp. Species that could potentially be eusocial based on their phyloge-
netic placement, such as Halictus and Lasioglossum (Dialictus), were
considered ‘not eusocial’ because eusociality is unlikely to be realized
in high-elevation habitats with short growing seasons (Kocher et al.
2014), such as subalpine Colorado. Nonetheless, we also conducted a
separate analysis of sociality, with Bombus, Halictus and Lasioglossum
(Dialictus) treated as ‘eusocial’ given that in lower elevation areas
not sampled in this study, Halictus and/or Lasioglossum (Dialictus)
can be eusocial (Kocher et al. 2014). Parasitic bees were excluded
from analyses of diet and sociality. We also categorized bee species by
size using intertegular (IT) span to estimate dry body mass (Cane
1987; Greenleaf et al. 2007). We measured IT span using a stereo-
scope and calibrated ocular micrometer for up to ﬁve randomly
selected female bees from each morphospecies, using only worker bees
for social species. We classiﬁed species as small (<4mg),medium
(4–16 mg), or large (>16 mg) (Winfree, Griswold & Kremen 2007).
We performed separate analyses for bee abundance (estimated as the
bee catch rate), rareﬁed richness, evenness, and functional groups. We
also calculated mean expected foraging distance for the bees using
overall mean IT span and the statistical relationship between body
size and foraging distance presented by Greenleaf et al. (2007).
To test whether bee abundance varied over the course of each sam-
pling season and from year to year, we used generalized linear mixed
models. We used AICc values to rank and compare multiple
hypotheses and pick the model best supporte d by th e data. The candi-
date models included meadow type (dry, wet-Veratrum,orwet-Salix),
mean ﬂoral density, and daily maximum temperature as factors. Site,
year and sampling week were included as random eﬀects. Sampling
week was deﬁned as the number of times a site was sampled within a
season, which allowed for diﬀerences in timing of snowmelt (and thus
accesstothesiteforsampling)among sites and years. For example,
sampling at higher elevation sites began slightly later than lower eleva-
tion sites due to later snowmelt and phenology, but the ﬁrst time each
site was sampled was lumped into a single phen ologically similar sam-
ple week. Likewise, years with early snowmelt and phenology (e.g.
2012) had more sampling weeks than years with late snowmelt and phe-
nology (e.g. 2011) due to a longer ﬂight season. We log-transformed
bee capture rates to fulﬁl the assumption of normality (here and below).
We used the lmer function in th e lme4 package to construct the models
and the aictab function in the AICcmodavg package to rank and com-
pare models using R version 3.0.3 (R Development Core Team 2011;
Mazerolle 2013; Bates et al. 2014). We used type III sums of squares
and Satterthwaite approximation of degrees of freedom to calculate F
and P values.
To test for diﬀerences in bee abundance, rareﬁed richness, and even-
ness between single-sample sites and repeat-sample sites, we used nested
© 2015 The Authors. Methods in Ecology and Evolution © 2015 British Ecological Society, Methods in Ecology and Evolution
Eﬀect of lethal sampling on bee populations 3
ANCOVAs with treatment (single or repeated sampling), block (week),
and river valley as factors, and mean ﬂoral density and daily maximum
temperature as covariates. We included block as a random eﬀect to
control for the week the samples were collected, and we tested treat-
ment over the block term (here and below). To test whether the eﬀect of
repeated sampling varied with bee functional group, we used general
linear mixed models with functional group, treatment (single or
repeated sampling), block (week), and river va lley as factors, and mean
ﬂoral d ensity and daily maximum temperature as covariates. We
included site as a rando m eﬀect in the functional group analysis to con-
trol for variation among sites. Four separate analyses were performed
for eac h functional group: nesting habits, diet, sociality, and body size.
We used JMP 11.0 (SAS Institute 2013) to perform the nested
and functional group analyses. To visualize potential diﬀerences in
community composition between single and repeat-sampled sites, we
used non-metric multidimensional scaling (NMDS), using the Bray–
Curtis dissimilarity index, using the vegan package (Oksanen et al.
2012) in R version 3.0.3 (R Development Core Team 2011). Stress val-
ues from models with two to six dimensions were compared to deter-
mine th e most appropriate summary of the relationship between
sampling and community composition.
Across ﬁve years and 26 sites, we vouchered 14 312 pan-
trapped bees, netted-and-vouchered 995 bees, and netted-
and-released 2489 bees. Collectively, these bees belonged to
ﬁve families, 28 genera, and 132 species/morphospecies
including 26 county records and three state records (Appen-
dix S2), with a mean IT span of 139 mm and mean expected
foraging range of 66 m (Greenleaf et al. 2007). Approxi-
mately 28% of the bees caught were male, and 40% of the
bees caught were Lasioglossum (Dialictus)sweatbees.For
bees identiﬁed to species, the two most common species were
the sweat bees Halictus virgatellus and Halictus rubicundus,
comprising 1131% and 2 52% respectively of all bees
vouchered. In addition, we captured 35 species represented by
a single individual. In 2012, we collected 100% of the genera
and 98% of the morphospecies that we would eventually
catch after just 62 of the 95 total sampling events. A rarefac-
tion curve of the 2012 sampling data (Fig. S2) revealed a satu-
rating function, suggesting that our sampling eﬀort
adequately characterized common bee taxa but that we would
likely continue to capture new, rarer species with additional
INTER- AND INTRA-ANNUAL VARIATION
Although we predicted that repeated sampling of sites would
result in a decrease in bee catch rate both among years and
within seasons, we found no evidence for either pattern.
Among years, total bee catch rates (vouchered and released
combined, divided by total sampling eﬀort) ranged from
633 bees per hour in 2011 to 814beesperhourin2010
(Fig. 1). Catch rates for vouchered bees only were consistent
across years, except for 2012, when the catch rate nearly
doubled. Moreover, when we examined intra-annual varia-
tion in vouchered bee catch rates, we found no general pat-
tern within years (assessed visually from Fig. 2, S3 and S4).
The best-ﬁt model for analysing inter- and intra-annual vari-
ation in bee abundance was supported by 88% of the
Akaike weights and included signiﬁcant interactions between
ﬂower density and maximum daily temperature, with site as
a random eﬀect. The model also allowed the intercepts (but
not slopes) to vary among years and among sample weeks
within years (nested random eﬀects) (Appendix S3). In addi-
tion, we found signiﬁcant main eﬀects of both ﬂower density
and maximum daily temperature on vouchered bee catch
= 983 , P < 0002 and F
P < 0001, respectively), with bee catch rate increasing with
increasing ﬂower density and temperature (b 1SE:
048 015 and 008 002, respectively). We also found a
Fig. 1. Interannual variation in mean bee
catch rates ( 1 SE)forbeesthatwerecaught
in pan traps and vouchered (black bars),
caught in nets and vouchered (white bars),
(gray bars). Only Bombus spp. were identiﬁed
in the ﬁeld and released. Bee catch rates varied
among years with no consistent pattern over
© 2015 The Authors. Methods in Ecology and Evolution © 2015 British Ecological Society, Methods in Ecology and Evolution
4 Z. J. Gezon et al.
signiﬁcant interaction between ﬂower density and maxi-
mum daily temperature on vouchered bee catch rate (b =
002 0007, F
= 945, P = 00023).
REPEATED VERSUS SINGLE-SAMPLED SITES
The catch rate of vouchered bees was 3128% higher on aver-
age in repeat-sampled sites versus single-sampled sites,
although the diﬀerence was not statistically signiﬁcant
(Table 1, Fig. 3A). We also found no diﬀerences in rareﬁed
richness or evenness (E
) between repe at-samp led and single-
sampled sites (Table 1, Fig. 3B and 3C, respectively). River
valley had a signiﬁcant eﬀect on rareﬁed species richness, with
the East River valley having higher rareﬁed richness (Least
Squares Means SE: 568 022 species) than Slate River
valley (442 058 species) and Washington Gulch
(364 056 species). Bee catch rate was 5954% higher in
wet-Veratrum sites than dry meadow sites (1264 503 and
792 309, respectively). We also found a weak, positive
relationship between rareﬁed richness and evenness and maxi-
mum daily temperature, and a weak, positive relationship
between catch rate of vouchered bees and ﬂower density
(Table 1). Results were qualitatively similar for combined bee
data (Fig. S5 and Appendix S4).
We also found no signiﬁcant main eﬀect of treatment
(repeated versus single-sample) when we subdivided the bees
by functional group (nesting habits, diet, sociality, and body
size) (Table 2, Fig. 4). Likewise, we found no signiﬁcant inter-
action between treatment and functional group for nesting
habits and body size (Table 2, Fig. 4a,d). When only Bombus
(and not Halictus and Lasioglossum (Dialictus)) were treated as
eusocial, we found no signiﬁcant interaction between treat-
ment and sociality (Table 2, Fig 4c; see Appendix S5, Fig. S6
for analysis of these taxa categorized as eusocial). We found a
signiﬁcant interaction between treatment and diet (Table 2,
Fig. 4b), with catch rates of pollen specialists being 4388%
higher, but catch rates of pollen generalists being 795% lower,
in repeated- than single-sample sites. Bee catch rate varied by
Fig. 2. Bee catch rates (mean bees caught per
hour) for vouchered bees (collected using pan
traps and netting) varied from year to year,
with no con sistent pattern related to the num-
ber of times the sites were sampled. Data rep-
resent mean ( 1 SE) catch rates across all
sites after repeated sampling within the season.
Years with earlier phenology (e.g., 2012) had
more sampling bouts due to the longer ﬂying
Eﬀect of lethal sampling on bee populations 5
functional group classiﬁcation (Table 2). Soil-nesting bees
comprised 6663% of all vouchered bees, and were between
406 and 3542-times more common than bees of any other
nesting habits (Fig. 4a). The majority of vouchered specimens
were pollen generalists (84 81%, Fig. 4b) and not eusocial
(9562%, assuming our montane Halictus and Lasioglossum
(Dialictus) were not eusocial, Fig. 4c). Small- (4891%) and
medium-sized (3875%) bees were both more common than
large bees (12 33%, Fig. 4d).
NMDS indicated overlap in bee community composition
between repeated versus single-sampled sites (Fig. 5a),
among river valleys (Fig. 5b), and between wet versus dry
habitats (Fig. 5c). Community composition changed dramat-
ically over the course of the summer, as indicated by the dis-
tinct clusters of points when separated by sample week
(Fig. 5d). The overall stress of the two-dimensional NMDS
was 018, indicating fair conﬁdence in the ordination dis-
tances. Models with higher dimensionality resulted in lower
stress (e.g. three-dimensional solution, stress = 0126), but
the results were qualitatively similar (Fig. S7a–d) to the two-
bee catch rate or diversity, even after sampling the same sites
every other week during the ﬂight season for three consecutive
years. This lack of response was consistent across all bee func-
tional groups with the exception of diet. Although we caught
more pollen specialists in repeated-sample sites than in single-
sample sites (and the opposite pattern for pollen generalists),
the magnitude of eﬀect associated with diet was relatively small
and did not represent a large change in catch rate between
treatments (Fig. 4b). Taken together, these results suggest that
the bee populations sampled in this study are robust to
repeated, lethal sampling twice monthly for multiple, consecu-
Several non-mutually exclusive mechanisms could explain a
lack of sampling eﬀect, four of which we explain here. The ﬁrst
two mechanisms reﬂect underlying ecological mechanisms
on species life histories. First, compensatory mortality and
natality could explain the lack of sampling eﬀect. Bees often
compete for limiting resources, including ﬂoral food resources
(Minckley et al. 1994; Roulston & Goodell 2011) and nesting
Table 1. Nested ANCOVAs testing the eﬀect of treatment (repeated vs.
single sampling) on (A) vouchered bee catch rate [log(bees/hour)], (B)
rareﬁed richness and (C) evenness (measured as E
). Type refers to
the habitat type (dry or wet-Veratrum), valley is the river valley sites
were located in (East River, Slate River, or Washington Gulch), and
max temp is maximum temperature
DF F Ratio P
(A) Catch rate
Treatment 1, 893 027 0619
Type 1, 2141 6 31 0020
Valley 2 , 979 227 015 5
Flower density 1, 2146 353 0074
Max temp 1, 1725 138 0255
(B) Rareﬁed richness
Treatment 1, 822 003 0868
Type 1, 2643 0 57 0457
Valley 2, 1434 534 0019
Flower density 1, 2651 039 0538
Max temp 1, 1576 411 0060
Treatment 1, 643 016 0698
Type 1, 2432 1 11 0325
Valley 2, 1094 3
Flower density 1, 2434 014 0713
Max temp 1, 1260 380 0074
Fig. 3. Least Square Means (LSM) from ANCOVAs of vouchered bees
niﬁcant diﬀerence in (a) bee ca tch rate (bees per hour), (b) rareﬁed rich-
ness, or (c) evenness between single and repeated sampling sites.
6 Z. J. Gezon et al.
sites (e.g. Hubbell & Johnson 1977; Potts & Willmer 1997; Stef-
fan-Dewenter & Schiele 2008). Reducing bee populations
through le thal sampling may re duce competition for these
resources and therefore increase the population growth rate
via density-dependent processes. Compensation is a funda-
mental principle of wildlife management (Bartmann, White &
Carpenter 1992), and although bee population dynamics are
not well understood (Steﬀan-Dewenter & Schiele 2008), the
same principles that allow for sustainable harvesting of ﬁsh
(Worm et al. 2009) and hunting of wildlife (Boyce, Sinclair &
White 1999) could apply to ‘harvesting’ of insects in environ-
mental monitoring programmes. Secondly, lethal sampling
may have no eﬀect on subsequent bee population size if sam-
pling simply removed individuals from the population that
would not have had an opportunity to reproduce anyway. If
bees are controlled by bottom-up rather than top-down forces
(e.g. Steﬀan-Dewenter, Potts & Packer 2005), removing adult
bees from the population via lethal sampling may not reduce
population size as long as enough bees remain to utilize all lim-
iting resources. Relative to the ﬁrst mechanism proposed, here
we would not expect to ﬁnd a change in bee population g rowth
rate following lethal sampling.
Thirdly, the reproductive habits of bees may lessen the
eﬀects of sampling. Solitary bees construct and mass provision
brood cells as they forage (e.g. Eickwort et al. 1996); thus,
many individual bees may be sampled after they have mated
and carried out some reproduction, which could mitigate the
eﬀect of sampling among years. Unlike many vertebrate taxa,
solitary bees do not require ongoing parental care after the
female provisions and seals larval brood cells (sealing the
entrance of a completed nest may or may not be essential for
brood survival). Thus, the reproductive output of a vouchered
specimen may be reduced but not eliminated. Moreover, leth-
ally sampling some workers from eusocial species may have
minimal eﬀect on the colony level production of males and
daughter queens (e.g. Harbo 1986). Finally, if males are
removed from the population after mating, or females are not
mate-limited, then sampling males may also have no eﬀect on
univoltine bee populations and communitie s in subsequent
The fourth mechanism we outline concerns the short ﬂight
season of most solitary bees, and thus high species turnover
during a typical summer. In cases where bees have short ﬂight
seasons, sampling every two weeks may not result in within-
season declines in bee abundance if new species or populations
are being sampled during each sampling event. For example,
Halictus rubicundus and other typically social halictine bees
may have solitary populations in high-elevation regions of the
Rocky Mountains due to short ﬂight season, with individual
bees only foraging for approximately two weeks (Eickwort
et al. 1996). Our data support observations of short ﬂight peri-
ods by Eickwort et al. (1996) in that detection of individual
species within our sites spanned 1188 082 days on average
( SE). Thus, in systems with high species (and population)
turnover, a biweekly sampling scheme may allow ample time
for brood provisioning between sampling events.
Although our study speciﬁcally tested whether lethal sam-
pling negatively aﬀected bee populations, our results may be
generalizable to other entomological systems. For example,
Hunter et al. (2014) found that 90% of Finnish moth popula-
tions were stable or increasing during 32 years of repeatedly
sampling 11 sites using light traps. Unfortunately, experimen-
tal data on the eﬀects of sampling arthropods are extremely
limited, and most accounts of negative eﬀects of sampling only
address the over-collection of rare and valuable specimens for
the black market, such as alpine Lepidoptera in Asia (Hama,
Ishii & Sibatani 1989). The dearth of information on the eﬀects
of sampling insects is unfortunate given the concerns about
specimen collection in general (Wagner 1991; Minteer, Collins
& Puschendorf 2014b; Minteer et al. 2014a) and of bees in par-
ticular (Tepedino et al. 2015), and the utility of insect collec-
tions for ecological research (e.g. Benton et al. 2002; Novotny
et al. 2007).
Catch rates for both vouchered and released bees varied dra-
matically among years (Fig. 2 and S3, respectively), and there
was no consistent decrease in vouchered bee catch rates
from one year to the next as we would expect if lethal samp ling
Table 2. Nested ANCOVAs testing the eﬀect of treatment (repeated vs.
single sampling) on vouchered bee catch rate [log(bees/hour+1)] by
functional group. Type refers to habitat type (dry or wet-Veratrum
meadow). Separate analyses were performed for each of four functional
groups: (A) nesting substrate, (B) diet, (C) sociality and (D) body size.
Body size was treated as a three-level categorical variable (small, med-
ium and large)
DF F Ratio P
(A) Nesting substrate
Type 1, 12 9618 0 027
Treatment 1, 844 0 15 0711
Valley 2 , 987 1 29 0318
Flower density 1, 3553 2 79 0104
Max temp 1, 1547 0 37 0552
Nesting 5, 191310158 <00001
Nesting*Trmt 5, 1913047 0801
Type 1, 13 32 5 19 0040
Treatment 1, 877 0 31 0593
Valley 2 , 992 2 86 0105
Flower density 1, 3423 2 30 0138
Max temp 1, 1556 1 83 0196
Diet 1, 49 94 18917 <00001
Diet*Trmt 1, 49 94 6 11 0017
Type 1, 1232 3 78 0075
Treatment 1, 7513 0 43 0531
Valley 2 , 988 2 75 0113
Flower density 1, 3184 1 17 0287
Max temp 1, 1301 1 46 0248
Sociality 1, 5186 32625 <00001
Sociality*Trmt 1, 51 86 0 95 0335
(D) Body size
Type 1, 13 67 3 53 0082
Treatment 1, 919 0 18 0678
Valley 2 , 989 2 38 0143
Flower density 1, 3760 3 05 0089
Max temp 1, 1818 0 77 0
Size 2, 83 57 53 72 <00001
Size*Trmt 2, 8357 1 49 0231
Eﬀect of lethal sampling on bee populations 7
negatively impacted bee populations. Rather, the interannual
variation in bee catch rates was likely related to seasonal diﬀer-
ences among years. For example, high snowfall and late phe-
nology in 2011 resulted in extremely high ﬂoral resources, and
may have caused the solitary bee populations to increase,
resulting in high vouchered bee capture rates in the subsequent
year 2012. Likewise, extreme drought and freezing events (Ino-
uye 2008) in 2012 resulted in extremely low ﬂoral resources
during that summer, which may have caused the near absence
of Bombus spp. workers. However, the capture rates of both
vouchered and released bees returned to more average num-
bers in 2013, which was a more average year in terms of snow-
pack, phenology and ﬂoral abundance (Z. Gezon, personal
observation). Our results reﬂect the extreme variability and
unpredictability of ﬂoral resources observed in many systems,
along with the corresponding eﬀects on pollinator abundance
and plant–pollinator interactions (Tepedino & Stanton 1980;
on, Waser & Ollerton 2008; Olesen et al. 2008).
Five caveats are important to the interpretation of this
study. First, our results apply to sampling bees every two
weeks and cannot be used to assess whether more frequent
sampling, for example weekly or even daily, may negatively
aﬀect bee populations or communities. Secondly, the distances
between the repeated-sample sites (min = 125 m,
max = 2400 m) were large compared to the mean expected
foraging distance of the bees we sampled (66 m), and thus the
maximum area that we could theoretically depopulate through
sampling. Smaller distances between sampling sites in relation
to bee foraging areas could aﬀect the ability of bees to recolo-
nize sampled areas and could therefore aﬀect bee populations
in some areas. Thirdly, the ecosystem investigated in this study
(subalpine Rocky Mountain West) is highly seasonal with a
short ﬂight-season; thus, there is high species turnover when
sampling once every two weeks. How sampling aﬀects bee pop-
ulations and communities in habitats with less seasonal turn-
over requires further investigation. Fourthly, we did not test
for taxon-speciﬁc declines in abundance over sampling sea-
sons. Looking at within-site patterns would be most appropri-
ate for determining whether sampling aﬀects individual
species, but the probability of catching the same taxa within a
site over multiple sampling periodswaslowevenfortheabun-
dant taxa because of the infrequency of sampling compared to
the short ﬂight season of most species within our sites. Fifthly,
we only explored whether repeated sampling aﬀected bee catch
Fig. 4. Least Square Means ( LSM) of vouc hered bee catch rates for each of four functional groups, (a) nesting habits, (b) diet, (c) sociality, and (d)
body size. In the sociality analysis, only Bombus were categorized as eusocial. The only functional group classiﬁcation that interacted signiﬁcantly
with treatment (single versus repeated sampling) was diet. Note the diﬀerent scale on the y-axes.
8 Z. J. Gezon et al.
rate and community structure and did not assess its impact on
ﬂoral visitation and plant reproduction. Bees are critical in per-
forming pollination services, and complete removal of even a
single species of bee can have impacts on plant–pollinator
interactions and plant reproduction (Brosi & Briggs 2013), so
future studies should investigate whether intensive sampling
results in pollen limitation of melittophilous plants.
Given the economic importance of bee pollinators and the
mounting threat of anthropogenic disturbances such as cli-
mate and land-use change (Klein et al. 2007; Potts et al.
2010; Ollerton, Winfree & Tarrant 2011), bee inventorying
and monitoring programmes are critically important for doc-
umenting the biodiversity of ecosystem service providers.
Long-term and repeated sampling is needed to monitor bee
populations across space and time (Committee on the Status
of Pollinators in North America, & National Research Coun-
cil 2007; Winfree 2010; Lebuhn et al. 2013), but implement-
ing and funding such a programme remains a challenge due,
in part, to divergent views on the relevance and statistical
power of the proposed approaches, and the ability to over-
come the taxonomic impediments (Lebuhn et al. 2013; Tepe-
dino et al. 2015; and response by Lebuhn et al. 2015). The
very eﬀectiveness (lethality) of proposed trapping measures
has also been cited as a concern, but our research provides
compelling evidence that lethal sampling of bees using pan
traps and netting every two weeks does not negatively aﬀect
bee populations in terms of abundance, species richness,
evenness and functional group composition. While our study
does not directly address the mechanisms that make bee pop-
ulations robust to lethal sampling, we suggest several plausi-
ble explanations that draw from well-known ecological
theories and species life histories. However, the generality of
our results remains to be tested for bee populations and com-
munities in other regions. Nonetheless, the lack of environ-
mental impact of bee sampling we document is encouraging,
suggesting that pan trapping and netting bi-monthly may be
a sustainable method to inventory bee diversity and to docu-
ment long-term trends in bee populations and communities.
We would like to thank M. G. Rightmyer for identiﬁcation of Osmia spp.; T. L.
Griswold for identiﬁcation of Stelis spp.; R. Brennan, R. Polanco, A. Slominski,
S. Turner and J. Welch for ﬁeld assistance; S. Sprott for making the map of the
ﬁeld sites; and R. Chaves-Ulloa, members of the Irwin Lab and two anonymous
reviewers for providing valuable feedback on the manuscript. The RMBL and
the Gunnison National Forest provided access to ﬁeld sites. This work was sup-
ported by grants from the National Science Foundation (DEB-0841862 and
DEB-0922080). Any opinions, ﬁndings and conclusions or recommendations
expressed in this material are those of the authors and do not necessarily reﬂect
the views of the National Science Foundation.
All supplemental ﬁgures and appendices have been uploaded as supporting infor-
mation. Specimen-related data are available at the RMBL website www.rmbl.org
and will be archived at the American Museum of Natural History, which deposits
their data in the Arthropod Easy Capture website and shares it with biodiversity
portals including www.discoverlife.org, and, in the near future, with idigbio.
All other data related to the experiment are available from the Dryad Digital
Repository (Gezon et al. 2015).
on, R., Waser, N.M. & Ollerton, J. (2008) Year-to-year variation in the
topology of a plant-pollinator interaction network. Oikos, 117, 1796–1807.
Albrecht, M., Duelli, P., Schmid, B. & Muller, C.B. (2007) Interaction diversity
within quantiﬁed insect food webs in restored and adjacent intensively man-
aged meadows. Journal of Animal Ecology, 76, 1015–1025.
Fig. 5. Two-dimensional non-metric multidi-
mensional scaling (NMDS) ordination plots
indicating overlap in bee communities when
separated by (a) treatment (S = single-sam-
pled sites, R = repeat-sampled sites), (b) river
valley (E = East River, S = Slate River,
WG = Washington Gulch), (c) habitat type
(D = dry meadow, W = wet-Veratrum), and
(d) sample week. Abbreviations are located at
the grand mean for each group, and the perim-
eter and area of each group is outlined and
shaded. Note that only vouchered bees were
Eﬀect of lethal sampling on bee populations 9
Bartmann, R.M., Whit e, G.C. & Carpenter, L.H. (1992) Compensatory mortal-
ity in a Colorado mule deer population. Wildlife Monographs, 121,3–39.
Bartomeus, I., Ascher, J.S., Gibbs, J., Danforth, B.N., Wagner, D.L., Hedtke,
S.M. & Wi nfree, R. (2013) Historical changes in northeastern US bee pollina-
tors related to shared ecological traits. Proceedings of the National Academy of
Sciences of the United States of America, 110, 4656–4660.
Bates, D., Maechler, M., Bolker, B. & Walker, S. (2014) lme4:Linear mixed-
eﬀects models using Eigen and S4. R package version 1.1-5.
Benton, T.G., Bryant, D.M., Cole, L. & Crick, H.Q.P. (2002) Linking agricul-
tural practice to insect and bird populations: a historical study over three dec-
ades. Journal of Applied Ecology, 39,673–687.
Boyce, M.S., Sinclair, A.R.E. & White, G.C. (1999) Seasonal compensation of
predation and harvesting. Oikos, 87, 419–426.
Brosi, B.J. & Briggs, H.M. (2013) Single pollinator species losses reduce ﬂoral
ﬁdelity and plant reproductive function. Proceedings of the National Academy
of Sciences of the United States of America, 110, 13044–13048.
Burkle, L.A., Marlin, J.C. & Knight, T.M. (2013) Plant-pollinator interactions
over 120 years: loss of species, co-occurrence, and function. Science, 339,
Cane, J.H. (1987) Estimation of bee size using intertegular span (Apoidea). Jo ur-
nal of the Kansas Entomological Society, 60, 145.
Cane, J.H., Griswold, T. & Parker, F.D. (2007) Substrates and materials used for
nesting by North American Osmia bees (Hymenoptera: Apiformes: Megachili-
dae). Annals of the Entomological Society of America, 100, 350–358.
Cane, J.H., Minckley, R.L. & Kervin, L.J. (2000) Sampling bees (Hymenoptera:
Apiformes) for pollinator community studies: pitfalls of pan-trapping. Journal
of the Kansas Entomological Society, 73, 225–231.
CaraDonna, P.J., Iler, A.M. & Inouye, D.W. (2014) Shifts in ﬂowering phenology
reshape a subalpine plant community. Proceedings of the National Academy of
Sciences of the United States of America, 111, 4916–4921.
urbanization on forest bee communities. Environmental Entomology, 43,
Chaves-Ulloa, R., Umana-Villalobos, G. & Springer, M. (2014) Downstream
eﬀects of hydropower production on aquatic macroinvertebrate assemblages
intworiversinCostaRica.Revista de biologia tropical, 62
(Suppl 2), 179–201.
Committee on the Status of Pollinators in North America, and National
Research Council. (2007) Status of Pollinators in North America. National
Academies Press, Washington, DC.
K. (1993) Temperature and the pollinating activity of social bees. Ecological
Eickwort, G.C., Eickwort, J.M., Gordon, J. & Eickwort, M.A. (1996) Solitary
behavior in a high altitude population of the social sweat bee Halictus rubicun-
dus (Hymenoptera: Halictidae). Behavioral Ecology and Sociobiology, 38,227–
Forrest, J., Inouye, D.W. & Thomson, J.D. (2010) Flowering phenology in subal-
pine meadows: does climate variation inﬂuence community co-ﬂowering pat-
terns? Ecology, 91,431–440.
Forrest, J.R.K. & Thomson, J.D. (2011) An examination of synchrony between
insect emergence and ﬂowering in Rocky Mountain meadows. Ecological
Monographs, 81, 469–491.
Gallai, N., Salles, J.M., Settele, J. & Vaissiere, B.E. (2009) Economic valuation of
the vulnerability of world agriculture confronted with pollinator decline. Eco-
logical Economics, 68, 810–821.
Gezon, Z.J., Wyman, E.S., Ascher, J.S., Inouye, D.W. & Irwin, R.E. (2015) Data
from: The eﬀect of repeated, lethal sampling on wild bee abundance and diver-
sity. Dryad Digital Repository, doi:10.5061/dryad.5385j.
Gibbs, J., Packer, L., Dumesh, S. & Danforth, B.N. (2013) Revision and reclassi-
ﬁcation of Lasioglossum (Evylaeus), L. (Hemihalictus) and L. (Sphecodoga-
stra) in eastern North America (Hymenoptera: Apoidea: Halictidae). Zootaxa,
Greenleaf, S.S., Williams, N.M., Winfree, R. & Kremen, C. (2007) Bee foraging
ranges and their relationship to body size. Oecologia, 153, 589–596.
Hama, E., I shii, M. & Sibatani, A. (1989) Nipponsan tyoorui no suiboo to hogo
(Decline and conservation of butterﬂies in Japan), Part 1. Nippon Rinsi Gak-
kai (Lepidopterological Society of Japan).
Harbo, J.R. (1986) Eﬀect of population-size on brood production, worker sur-
vival and honey gain in colonies of honeybees. Journal of Apicultural Research,
Hawkins, C.P., Norris, R.H., Hogue, J.N. & Feminella, J.W. (2000) Develop-
ment and evaluation of predictive models for measuring the biological integrity
of streams. Ecological Applications,
Hubbell, S.P. & Johnson, L.K. (1977) Competition and nest spacing in a tropical
stingless bee community. Ecology, 58, 949–963.
Hunter, M.D., Kozlov, M.V., Itamies, J., Pulliainen, E., Back, J., Kyro, E.M. &
Niemela, P. (2014) Current temporal trends in moth abundance are counter to
predicted eﬀects of climate change in an assemblage of subarctic forest moths.
Global Change Biology, 20, 1723–1737.
Hurd, P.D. Jr (1979) Superfamily Apoidea. Catalog of Hymenoptera in Amer-
ica north of Mexico.Vol.2 (eds K.V. Krombein, P.D. Hurd Jr, D.R.
Smith & B.D. Burks), pp. 1741–2209. Smithsonian Institution Press,
Inouye, D.W. (1977) Species structure of bumblebee communities in North
America and Europe. The Role of Arthropods in Forest Ecosystems (ed. W.J.
Mattson), pp. 35–40. Springer Verlag, New York.
Inouye, D.W. (2008) Eﬀects of climate change on phenology, frost damage, and
ﬂoral abundance of montane wildﬂowers. Ecology, 89,353–362.
Jha, S. & Vandermeer, J.H. (2010) Impacts of coﬀee agroforestry management
on tropical bee communities. Biological Conservation, 143,1423–1431.
Kearns, C.A. & Oliveras, D.M. (2009) Environmental factors aﬀecting bee diver-
sity in urban and remote grassland plots in Boulder, Colorado. Journal of
Insect Conservation, 13,655–665.
Klein, A.M., Vaissiere, B.E., Cane, J.H., Steﬀan-Dewenter, I., Cunningham,
S.A., Kremen, C. & Tscharntke, T. (2007) Importance of pollinators in chang-
ing landscapes for world crops. Proceedings of the Royal Society B-Biological
Sciences, 274, 303–313.
Kocher, S.D., Pellissier, L., Veller, C., Purcell, J., Nowak, M.A., Chapuisat, M. &
Pierce, N.E. (2014) Transitions in social complexity along elevational gradients
reveal a combined impact of season length and development time on social
evolution. Proceedings of the Roy al Society B-Biological Sciences, 281,
Langenheim, J.H. (1962) Vegetation and environmental patterns in Crested Butte
area, Gunnison County, Colorado. Ecological Monographs, 32,249–285.
LeBuhn, G., Griswold, T., Minckley, R., Droege, S., Roulston, T., Cane, J. et al.
(2003) A standardized method for monitoring bee populations – The Bee
Inventory (BI) Plot.
Lebuhn, G., Droege, S., Connor, E.F., Gemmill-Herren, B., Potts, S.G., Minck-
ley, R.L. et al. (2013) Detecting Insect Pollinator Declines on Regional and
Global Scales. Conservation Biology, 27,113–120.
Lebuhn, G., Droege, S., Connor, E.F., Gemmill-Herren, B., Potts, S.G., Minck-
ley, R.L. et al. (2015) Evidence-based conservation: reply to Tepedino et al.
Conservation Biology, 29,283–285.
Lovett, G.M., Burns, D.A., Driscoll, C.T., Jenkins, J.C., Mitchell, M.J., Rustad,
L., Shanley, J.B., Likens, G.E. & Ha euber, R. (2007) Who needs environmen-
tal monitoring? Frontiers in Ecology and the Environment, 5,253–260.
Mazerolle, M.J. (2013) AICcmodavg: model selection and multimodel inference
based on (Q)AIC(c). R package version 1.35.
Michener, C.D. (2007) TheBeesoftheWorld, 2nd edn. Johns Hopkins University
Press, Baltimore, MD.
Minckley, R.L., Wcislo, W.T., Yanega, D. & Buchmann, S.L. (1994) Behavior
and phenology of a specialist bee (Dieunomia) and sunﬂower (Helianthus) pol-
len availability. Ecology, 75, 1406–1419.
Minteer, B.A., Collins, J.P. & Puschendorf, R. (2014b) Specimen collection: an
essential tool–response. Science (New York, NY), 344, 816.
Minteer, B.A., Collins, J.P., Love, K.E. & Puschendorf, R. (2014a) Ecology.
Avoiding (re)extinction. Science (New York, NY), 344,260–261.
Mitchell, T.B. (1960) Bees of the Eastern United States. North Carolina Agricul-
tural Experiment Station.
Neame, L.A., Griswold, T. & Elle, E. (2013) Pollinator nesting guilds respond dif-
ferently to urban habitat fragmentation in an oak-savannah ecosystem. Insect
Conservation and Diversity, 6,57–66.
(2007) Low beta diversity of herbivorous insects in tropical forests. Nature,
Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O’Hara,
R.B. et al. (2012) vegan: Community Ecology Package.
Olesen, J.M., Bascompte, J., Elberling, H. & Jordano, P. (2008) Temporal
dynamics in a pollination network. Ecology, 89, 1573–1582.
Ollerton, J., Winfree, R. & Tarrant, S. (2011) How many ﬂowering plants are pol-
linated by animals? Oikos, 120, 321–326.
Popic, T.J., Davila, Y.C. & Wardle, G.M. (2013) Evaluation of common methods
for sampling invertebrate pollinator assemblages: net sampling out-perform
pan traps. PLoS ONE, 8, e66665.
Potts, S.G. & Willmer, P. (1997) Abiotic and biotic factors inﬂuencing nest-site
selection by Halictus rubicundus, a ground-nesting halictine bee. Ecological
Potts, S.G., Biesmeijer, J.C., Kremen, C., Neumann, P., Schweiger, O. & Kunin,
W.E. (2010) Global pollinator declines: trends, impacts and drivers. Trends in
Ecology & Evolution, 25, 345–353.
10 Z. J. Gezon et al.
R Development Core Team (2011) R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria.
Reynoldson, T.B. & Metcalfe-Smith, J.L. (1992) An overview of the assessment
of aquatic ecosystem health using benthic invertebrates. Journal of Aquatic
Ecosystem Health, 1, 295–308.
Roulston, T.A.H. & Goodell, K. (2011) The role of resources and risks in regulat-
ing wild bee populations. Annual Review of Entomology, 56,293–312.
SAS Institute. (2013) JMP, Version 11.0.0. SAS Institute, Cary, North Carolina,
Schonberg, L.A., Longino, J.T., Nadkarni, N.M., Yanoviak, S.P. & Gering, J.C.
(2004) Arboreal ant species richness in primary forest, secondary forest, and
pasture habitats of a tropical montane landscape. Biotropica, 36, 402–409.
Shepherd, M., Buchmann, S.L., Vaughan, M. & Black, S.H. (2003) Pollinator
Conservation Handb ook: A Guide to Und erstanding, Protecting, and Prov iding
Habitat for Native Pollinator Insects. The Xerces Society, Portland.
Smith, B. & Wilson, J.B. (1996) A consumer’s guide to evenness indices. Oikos,
Statzner, B., Bis, B., Doledec, S. & Usseglio-Polatera, P. (2001) Perspectives for
biomonitoring at large spatial scales: a uniﬁed measure for the functional com-
position on invertebrate communities in European running waters. Basic and
Applied Ecology, 2,73–85.
Steﬀan-Dewenter, I., Potts, S.G. & Packer, L. (2005) Pollinator diversity and crop
pollination services are at risk. Trends in Ecology & Evolution, 20,651–652.
Steﬀan-Dewenter, I. & Schiele, S. (2008) Do resources or natural enemies drive
bee population dynamics in fragmented habitats? Ecology, 89, 1375–1387.
Tepedino, V.J. & Stanton, N.L. (1980) Spatiotemporal variation in phenology
and abundance of ﬂoral resources on shortgrass prairie. The Great Basin Natu-
ralist, 40, 197–215.
Tepedino, V.J. & Stanton, N.L. (1981) Diversity and competition in bee-plant
communities on short-grass prairie. Oikos, 36,35–44.
Tepedino, V.J., Durham, S., Cameron, S.A. & Goodell, K. (2015) Documenting
bee decline or squandering scarce resources. Conservation Biology, 29,280–
Wagner, D.H. (1991) The “1-in-20 rule” for plant collectors. Plant Science Bulle-
tin, 37, 11.
Westphal, C., Bommarco, R., Carre, G., Lamborn, E., Morison, N., Petanidou,
T. et al. (2008) Measuring bee diversity in diﬀerent European habitats and bio-
geographical regions. Ecological Monographs, 78, 653–671.
Williams, N.M., Crone, E.E., Roulston, T.H., Minckley, R.L., Packer, L. &
Potts, S.G. (2010) Ecological and life-history traits predict bee species
responses to environmental disturbances. Biological Conservation, 143, 2280–
Wilson, J.S., Griswold, T. & Messinger, O.J. (2008) Sampling bee communities
(Hymenoptera: Apiformes) in a desert landscape: are pan traps suﬃcient?
Journal of the Kansas Entomological Society, 81,288–300.
Wilson, J.S., Messinger, O.J. & Griswold, T. (2009) Variation between bee com-
munities on a sand dune complex in the Great Basin Desert, North America:
implications for sand dune conservation. Journal of Arid Environments, 73,
Winfree, R. (2010) The conservation and restoration of wild bees. Year in Ecology
and Conservation Biology, 2010, 169–197.
Winfree, R., Griswold, T. & Kremen, C. (2007) Eﬀect of human disturbance on
bee communities in a forested ecosystem. Conservation Biology, 21,213–223.
Winfree, R., Williams, N.M., Dushoﬀ, J. & Kremen, C. (2014) Species abun-
dance, not diet breadth, drives the persistence of the most linked pollinators as
plant-pollinator networks disassemble. American Naturalist, 183, 600–611.
(2009) Rebuilding global ﬁsheries. Science, 325, 578–585.
Received 22 January 2015; accepted 10 March 2015
Handling Editor: Jana Vamosi
Additional Supporting Information may be found in the online version
of this article.
Fig. S1. Map of the sampling sites. Treatment refers to sites that were
sampled twice-monthly from 2009 to 2013 (‘repeated’) or only once in
Fig. S2. Rarefaction curve for the number of morphospecies caught
over the course of the 2012 sampling season.
Fig. S3. Intra-annual variation in bee catch rates (mean bees caught per
hour) for caught-and-released beesvariedfromyeartoyear,withno
Fig. S4. Patterns of vouchered bee catch rates varied from year to year,
with a trend toward higher catch rates early and late in the season,
except in the year 2012.
Fig. S5. Least Square Means (LSM) from
chered plus caught-and-relea sed bees) between single-sampled and
Fig. S6. Least Square Means (LSM) from nested
ANCOVA testing the
eﬀect of treatment (repeated vs. single sampling) on vouchered bee
catch rate [log(bees/hour + 1)] by sociality.
Fig. S7. Three-dimensional non-metric multidimensional scaling
(NMDS) ordination plots indicating overlap in bee communities when
separated by (a) treatment, (b) river valley, (c) habitat type, and (d)
Appendix S1. Sampling site information and locations.
Appendix S2. Species, morphospecies, and functional group designa-
tions used in analyses.
Appendix S3. Table comparing apriorimodels of bee catch rate within
and among years in the repeatedly sampled sites.
Appendix S4. Nested ANCOVAs testing the eﬀect of treatment
(repeated vs. single sampling) on (A) bee catch rate [log(bees/hour)],
(B) rareﬁed richness, and (C) evenness (measured as E
combined (vouchered and catch-and-release).
Appendix S5. Nested
ANCOVA testing the eﬀect of treatment (repeated
vs. single sampling) on vouchered bee catch rate [log(bees/hour + 1)]
Eﬀect of lethal sampling on bee populations 11