native bees buffer the negative impact of climate …...pollination of watermelon crops, we predict...

Native bees buffer the negative impact of climatewarming on honey bee pollination of watermelon cropsROMINA RADER * , JAMES RE I LLY * † , I GNAS I BARTOMEUS * ‡ and RACHAEL WINFREE*†

*Department of Entomology, Rutgers, The State University of New Jersey, 93 Lipman Drive, New Brunswick, NJ 08901, USA,

†Department of Ecology, Evolution and Natural Resources, Rutgers, The State University of New Jersey, 93 Lipman Drive,

New Brunswick, NJ 08901, USA

Abstract

If climate change affects pollinator-dependent crop production, this will have important implications for global food

security because insect pollinators contribute to production for 75% of the leading global food crops. We investigate

whether climate warming could result in indirect impacts upon crop pollination services via an overlooked mecha-

nism, namely temperature-induced shifts in the diurnal activity patterns of pollinators. Using a large data set on bee

pollination of watermelon crops, we predict how pollination services might change under various climate change sce-

narios. Our results show that under the most extreme IPCC scenario (A1F1), pollination services by managed honey

bees are expected to decline by 14.5%, whereas pollination services provided by most native, wild taxa are predicted

to increase, resulting in an estimated aggregate change in pollination services of +4.5% by 2099. We demonstrate the

importance of native biodiversity in buffering the impacts of climate change, because crop pollination services would

decline more steeply without the native, wild pollinators. More generally, our study provides an important example

of how biodiversity can stabilize ecosystem services against environmental change.

Keywords: climate change, ecosystem function, ecosystem service, insurance, resilience, response diversity, stability, tempera-

ture, wild bee

Received 17 March 2013 and accepted 19 April 2013

Introduction

The ability of agricultural production to keep up with

the growing human population (Gregory & George,

2011; Tilman et al., 2011) may be further taxed by cli-

mate change because rising temperatures are predicted

to reduce yields for several important food crops

(Lobell et al., 2011). In addition to these direct effects

acting through plant physiology (Porter & Semenov,

2005), climate change may have indirect effects on crop

production mediated by insect pollinators, which con-

tribute to production for 75% of the leading global food

crops (Klein et al., 2007).

Climate change might affect crop pollination through

its effects on managed and/or wild pollinators. The

honey bee, Apis mellifera, is by far the most versatile

and ubiquitous managed pollinator, increasing yield in

96% of animal-pollinated crops (Klein et al., 2007).

However, domestic honey bee stocks in the USA and

central Europe (van Engelsdorp et al., 2008; Potts et al.,

2010b), and also feral honey bee colonies (Kraus &

Page, 1995), are declining in many regions due to the

ectoparasitic mite Varroa destructor (Mesostigmata:

Varroidae) and other pests and diseases. This depen-

dence of agricultural crops on pollination by a single

species is thus an increasingly risky strategy (Winfree,

2008; Potts et al., 2010a). Wild pollinator species can

perform equal to or better than the honey bee as pollin-

ators for some crops (Jarlan et al., 1997; Winfree et al.,

2007; Jauker & Wolters, 2008; Rader et al., 2009), and

furthermore contribute to crop production even when

honey bees are present (Garibaldi et al., 2013).

It is biologically reasonable to expect that climate

warming will have strong effects on insects including

pollinators. Climate change impacts on insects are fre-

quently calculated at broad temporal (e.g. date of first

seasonal emergence, Sparks & Yates, 1997) or spatial

scales (e.g. geographical shifts in range, Parmesan,

2006). Such studies typically find that insects are

already responding strongly to the changing climate

(Deutsch et al., 2008; Morris et al., 2008). Smaller scale

changes are also likely, however, because most insects

are primarily ectotherms (Willmer & Unwin, 1981), and

even small changes in temperature can alter their activ-

ity and foraging behaviour (Heard & Hendrikz, 1993;

Stone, 1994). Although temperature can be positively

related to pollinator activity at flowers (Herrera, 1997;

‡ Present address: Department of Ecology, Swedish University of

Agricultural Sciences, Uppsala, SE-75007, Sweden

Correspondence: Romina Rader, Department of Physical Geogra-

phy and Quaternary Geology, Stockholm University, Stockholm,

SE 10691, Sweden, tel. +46 707558530, fax +46 8164671,

e-mail: [email protected]

© 2013 John Wiley & Sons Ltd 3103

Global Change Biology (2013) 19, 3103–3110, doi: 10.1111/gcb.12264

Willmer & Stone, 2004), important pollinators such as

bees reach high temperature thresholds after which

they provide few pollination services, because they

spend less time foraging and more time cooling

(Heinrich, 1979, 1980, 1993; Cooper et al., 1985). Differ-

ent species are likely to respond to temperature in

different ways, due to variation in thermoregulatory

ability, reflectance and body size (Heinrich, 1974;

Willmer & Unwin, 1981; Bishop & Armbruster, 1999).

If pollinator species respond differentially to climate

warming, then crop plants that are pollinated by a diver-

sity of species may be buffered against the effects of cli-

mate change. Broadly, this idea is known as response

diversity, and could occur with respect to many types of

environmental changes and for different ecosystem ser-

vices (Walker et al., 1999; Elmqvist et al., 2003). In the case

of pollination, environmental factors including tempera-

ture, precipitation andwind speed affect pollinator forag-

ing and thus could lead to response diversity (Bluthgen&

Klein, 2011). For example, in North American almond

orchards, honeybees were sensitive to high winds and

foraged less, whereas wild pollinators were less sensitive

and provided pollination services even under high wind

conditions (Brittain et al., 2013).

In this study, we use a long-term data set to develop

daily temperature-dependent activity surfaces for eight

bee taxa that pollinate watermelon (Citrullus lanatus).

We investigate the importance of fine-scale interactions

between pollinator-dependent crop plants and their

pollinators in response to rising temperatures using

IPCC projections (IPCC, 2007). We determine whether

daily activity patterns differ by bee species, and use this

information to predict the likely impacts of rising tem-

peratures upon pollinator activity patterns and thus

crop pollination function. We also explore the potential

for wild pollinators to stabilize any changes in function

associated with climate change.

Materials and methods

Crop species

We used watermelon C. lanatus as a model crop because (i) it

has unisexual flowers and thus is completely dependent on

insect pollination to set fruit (Delaplane & Mayer, 2005); (ii) its

flowers attract a diverse pollinator assemblage (Winfree et al.,

2007); and (iii) watermelon flowers are open for only 1 day

such that lifetime pollination to individual flowers can be

measured in a single day’s field work.

Pollinator activity at crop flowers

To measure pollinator visitation rate to flowers, we estab-

lished one 50-m transect of crop row within each of 18 water-

melon farms. Data were collected during the peak bloom of

watermelon at each farm, from 29 June to 20 August in 2005,

2007, 2008 and 2010. All farms were situated within a 90 by

60 km region of central New Jersey and eastern Pennsylvania,

USA. Data were collected at each of the 18 farms on two sepa-

rate days in 2005, on 1 day at each farm in 2007 and 2008, and

on 3 days at each farm in 2010. Pollinator visitation rate to

flowers was calculated by conducting 45-s surveys of groups

of flowers at 40 equally spaced points along each transect. At

each point, we observed visits to as many flowers as we could

view simultaneously. We surveyed each transect three times

per day between 8:00 hours and 13:00 hours, which is the per-

iod during which watermelon flowers are open. For purposes

of relating our measures of bee activity to hourly temperature

records, we recorded our flower visitation rate data in units of

bee visits per flower per time. Data were not collected when it

was rainy, or when wind speed was >4.6 m s�1 (90% of

records were collected at wind speeds below 2.4 m s�1). After

collecting the flower visitation rate data, we collected bees

from watermelon flowers with hand nets for 30 min and used

these specimens for species identification, as further detailed

below.

Pollinator species identification

In the field, we recorded all pollinator visits to flowers and

visually identified each individual pollinator ‘on the wing’ to

one of twelve species groups. In this study, we analyse data

for eight dominant species groups, which in combination

account for 98% of all pollinator visits and an estimated 95%

of all pollination function, across our entire data set. Three of

our analysed groups contain a single species that is readily

identifiable on the wing: A. mellifera, Melissodes bimaculata and

Peponapis pruinosa. The remaining five groups each contained

3–14 species, based on species-level identifications of speci-

mens netted from the same sites on the same days (Tables S1

and S2). Four groups were dominated by one or two species:

the group Bombus impatiens was dominated by B. impatiens

which accounted for 98% of records; the group Ceratina was

dominated by Ceratina calcarata/dupla (females of these two

species are not separable even under a dissecting microscope)

which accounted for 82% of records; the group ‘green bees’

was dominated by Augochlora pura which accounted for 89%

of records; and the group ‘tiny dark bees’ was dominated by

Lasioglossum imitatum which accounted for 65% of records. The

last group, ‘small dark bees’, was the most speciose and com-

prised 14 species including L. versatum (40% of records) and L.

pilosum (14%). Hereafter, we refer to both individual species

and groups as ‘taxa’. See Data S1 for further details on the spe-

cies composition of the observation groups and for tests of the

assumptions made by grouping species (Tables S1–S3).

Pollen deposition

To measure the number of watermelon pollen grains depos-

ited on stigmas during a single pollinator visit, we bagged

unopened, virgin female watermelon flowers with pollinator-

exclusion mesh and later offered these flowers individually to

bees foraging on watermelon flowers. After the bee visited the

© 2013 John Wiley & Sons Ltd, Global Change Biology, 19, 3103–3110

3104 R. RADER et al.

flower, the flower was protected from further pollinator visits,

placed in a floral tube and allowed to sit at room temperature

for ca. 24 h to allow the pollen to adhere to the stigma. Stig-

mas were then removed, softened in 10% KOH, stained with

1% fuchsin and prepared as microscopic slides so that the

number of watermelon pollen grains on the stigma could be

counted with a compound microscope. Control flowers were

left bagged until the end of the field day, and contained few

pollen grains (mean = 1, mode = 0, N = 95 stigmas).

Temperature records

For each hour and day that bee visitation rate was recorded at

each farm, we assigned a temperature based on temperature

records accessed from local weather stations. We obtained

records from weather stations that were positioned close to

each farm (median = 6 km, range 1–14 km). We used the

same weather stations for each year of data collection to

reduce variability among weather temperature records per

farm and across time. Temperature data were also collected

during our data collection at each farm, and these tempera-

tures were correlated with weather station records and

explained a large proportion of the variation (R2 = 0.59). The

transect-level records, however, diverged from the weather

station records at high temperatures (above 30 °C), likely

because our field thermometers overestimated high tempera-

tures due to the instruments not being insulated. Thus,

because precision at high temperatures is important for our

analyses and weather station measurements are less likely to

be influenced by measurement error, we used weather station

data instead of field measurements to reduce the potential for

measurement and instrument bias.

Generating the pollinator response surfaces

Both temperature and time of day are important determinants

of pollinator activity but they are not independent. Therefore,

we included both time of day and temperature to calculate

pollinator response surfaces. Each response surface consists of

a two dimensional grid with time of day on the x axis, temper-

ature on the y axis and the flower visitation rate for the bee

taxa being analysed as the z axis (outcome variable). For each

pollinator taxon, the flower visitation rate was calculated as

the total number of visits from that taxon/total flowers

observed during one transect observation period (visits per

flower per hour). Hourly visitation rates were pooled by taxon

across farms and years to estimate a taxon-specific mean and

variance for each cell in the temperature-time grid. We scaled

1 h of time to 2 °C of temperature as this combination roughly

matched the variation in time and temperature within our sys-

tem. For example, on a typical day, temperature increases by

roughly 10 °C over 5 h from 8:00 hours to 13:00 hours. As a

form of sensitivity analysis, we generated surfaces for three

different distance metrics: 2 °C = 1 h, 3 °C = 1 h, and

1 °C = 1 h. The resulting surfaces were very similar, espe-

cially in the densely sampled central region (Data S1).

To generate a mean and a variance for the flower visitation

rate within each cell of the temperature-time grid, we used the

k-nearest neighbour algorithm (Bremner et al., 2005). Prelimin-

ary analyses indicated that k = 30 points provided a good bal-

ance between sample size and spatial resolution (Data S1). We

then used a 2-D local polynomial smoothing function (loess

function in the ‘stats’ package, R Development Core Team,

2012) to simplify the response surfaces, thereby making the

simulation less sensitive to, for example, the particular times

of day that were sampled within each hour.

Current and future temperature predictions

To compare pollinator activity and pollination function

between current and future temperature scenarios, we used

the Intergovernmental Panel on Climate Change (IPCC) sce-

narios specific to the USA east coast region (NECIA, 2007) for

both the present (i.e. 2005–2010) and future (i.e. 2050–2055 and

2094–2099) time periods. In accordance with recommenda-

tions from the IPCC data provider (Hayhoe et al., 2008), we

used the IPCC-based temperature predictions in all of our

simulations for the current period (i.e. 2005–2010) instead of

estimating our own from temperature data measured at our

field sites. This method controls for the prediction process

itself because it compares the future climate predictions to the

current predictions based on the same model. We investigated

two scenarios proposed by the IPCC in its fourth and latest

assessment report (IPCC, 2007). The A1F1 (hereafter referred

to as A1) is the most extreme climate change scenario and

predicts a global mean warming from 2.4 to 6.4 °C by 2099,

whereas the B1 scenario is the least extreme and predicts a

mean warming from 1.1 to 2.9 °C.The IPCC predictions provide an estimated daily maximum

and minimum temperature for the month of July (which

encompassed 83% of our data records and which we thus

used as the basis of our simulation). To incorporate daily and

yearly variations in temperature, we simulated changes in

temperature (i) among the hours within a given day;

(ii) among the days within the month; and (iii) among years

over the 5-year time period. We estimated these in a several-

step process that incorporated multiple sources of uncertainty.

First, for each simulated day we used the July minimum and

maximum temperatures to calculate daily temperature pro-

files using a truncated sine curve (Parton & Logan, 1981),

which provides the best fit to the way in which temperature

changes across the course of a day at our study sites (Data S1;

Fig. S1). Second, to incorporate the empirical day-to-day vari-

ability in minimum and maximum temperatures, we used a

day-to-day variance estimated from our data for each day in

July from two representative weather stations (KLOM and

KLU) for 2 years of our study (2007 and 2010). Finally, to

account for variation in temperature among years, we calcu-

lated the yearly variance in the mean temperatures for each of

the 5 years from the IPCC-based temperature predictions. We

did this both at the time over which our bee data were

collected (i.e. 2005–2010), and at two future time periods (i.e.

2050–2055 and 2094–2099). In this way, we draw a minimum

and maximum temperature for each simulated day while

accounting for within – day, within – month and among year

variability.


NATIVE BEES BUFFER CLIMATE WARMING 3105

Our modelling approach investigates climate change solely

via increases in temperature. Even though changes in rainfall

during the bloom period could affect pollination in principle,

changes in summer rainfall patterns are expected to be modest

relative to increases in temperature in our study region

(NECIA, 2007).

Estimating pollination function per flower visit

For each bee taxon, we estimated single-visit pollen deposition

as a function of time of day. We include time of day in our

estimates of pollen deposition as the quantity of watermelon

pollen available declines with time of day, due to removal by

bees (Stanghellini et al., 2002). We estimated a value for both

the mean and the variance for a given time of day from a dis-

tribution based on the mean and SE predicted for that time of

day by a linear regression analysis. This technique incorpo-

rates our uncertainty about the actual value of pollen deposi-

tion at that time of day, given the variability in the data (Data

S1; Fig. S2).

Estimating pollination services

To estimate the pollination function contributed by each polli-

nator taxon under both the present and future conditions, we

developed a Monte Carlo model that combines our various

forms of data. We started with a simpler model that had been

developed to estimate present-day pollination function for a

subset of the data analysed here (Winfree et al., 2007). In this

study, we modify this basic model to incorporate input

parameters related to the temperature-dependent activity sur-

face for each pollinator species group, combined with current

and future temperature predictions.

The model input is a three-stage process that incorporates

future temperature predictions, the number of visits by each

taxon and the amount of pollen deposited per visit (Data S1).

Each iteration of the simulation represents the lifetime of one

female watermelon flower (because flowers are open for only

1 day, this is one flower-day), and estimates the pollination

services delivered to that flower as the number of watermelon

pollen grains deposited on the stigma. Pollen deposition by

each bee taxon is simulated separately (because taxa have dif-

ferent parameter values) and then summed to estimate the

total pollination per flower.

Each iteration of the simulation proceeds as follows. First,

for each simulated flower-day we draw an hourly temperature

profile (see above). Second, the model simulates the flower

visitation rate for each bee taxon. Here, we overlay the tem-

perature profile on the response surfaces (mean and variance;

see above) to obtain expected means and variances for the

number of visits that flowers receive as a function of time of

day and temperature (Fig. 1; Fig. S3). This mean and variance

is used to draw a simulated number of visits by a particular

Tem

pera

ture

°C

Time of day

(a) (b) (c) (d)

(e) (f) (g) (h)

Fig. 1 Response surfaces of all pollinator species groups in relation to temperature and time of day. Activity is measured as visits per

flower per hour, with regions of higher activity plotted as red and lower activity as yellow. The markings on the x and y axes (the ‘rug’)

indicate sampling effort; areas with low sampling effort are an extrapolation from the data. Black lines represent mean predicted tem-

perature for A1 climate change scenario; grey lines represent mean predicted temperature for B1 climate change scenario. Solid lines

are mean predicted temperatures for time period 2005–2010; dashed lines are for time period 2050–2055 and dotted lines are for time

period 2094–2099. Note changes in scale of legend among taxa.



taxon in a given hour from a negative binomial distribution.

Third, hourly measures are then converted into estimates of

the total pollination services delivered to a given flower over

its 1 day lifetime, by summing the pollen delivered by each

bee taxon over the 6 h flowers are open, from 7.30 to 13.30

(Data S1).

We ran 5000 iterations (i.e. 5000 flowers) of the simulation

to obtain the mean number of pollen grains deposited by each

bee taxon, and by all bees in combination for current and pro-

jected future temperatures (for either 2049–2055 or 2094–2099).

We calculated SE by using all the error propagated through

the model for each form of input data, then taking the SE over

the number of either present or future days over which mean

pollination per day was being estimated (Data S1).

Because within-species variation in bee abundance across

sites could potentially bias our estimates of bee activity as a

function of temperature, we took various steps to ensure that

our estimates of visitation were based on taxa known to be

present, and hence potentially active, at each specified temper-

ature and time interval (Data S1). Finally, as a way of contex-

tualizing the predicted changes in pollination services that we

report for climate warming, we calculated present-day spatial

variation in pollination services for comparative purposes

(Data S1).

Results

At the 18 farms we surveyed over 4 years, we found 31

species of insect pollinators visiting watermelon flow-

ers, observed 16 708 pollinator visits to 47 216 flowers

and obtained 417 records of pollen deposition in single

visits to previously unvisited flowers.

Our response surfaces, which represent pollinator

activity patterns, showed that taxa exhibit diverse

responses to both temperature and time of day (Fig. 1).

For example, under current conditions, A. mellifera is

most active at temperatures between 24 and 30 °C and

from 9:00 to 11:00 hours. In contrast, M. bimaculata is

most active above 30 °C and before 9:00 hours (Fig. 1).

Pollination efficiency, as measured by pollen deposition

in single visits to experimental flowers, also differed

among bee taxa (Fig. 2, Kruskal–Wallis v2 = 80.6,

df = 7, P < 0.0001; Data S1; Table S4).

When pollen deposition was combined with visita-

tion rate in accordance with temperature in the simula-

tion model, the resulting prediction was that some taxa

increase in their contribution to future pollination func-

tion under climate change whereas others decrease

(Fig. 3). For example, total pollen deposition by Ceratina

is predicted to increase by 86.4% under the most

extreme A1 climate change scenario for 2094–2099. Incontrast, total pollen deposition by the honey bee, A.

mellifera, is predicted to decline by 14.5% under the

same extreme A1 scenario (Fig. 3).

Even though the predicted changes for particular

individual taxa were large in some cases, the aggregate

change in pollination services to watermelon flowers

was only +4.6% (2094–2099, A1 scenario). This is

because declines in pollen deposition by some taxa

were offset by increases in other taxa. Pollination

services provided by the managed honey bee are pre-

dicted to decline under climate warming, whereas six

of the seven wild bee taxa are predicted to increase

their pollination services.

Small dark

Tiny dark

Ceratina

Green bees

Apis mellifera

Bombus

Melissodes

Peponapis

0 200 400 600 800 1000 1200Number pollen grains per visit

Fig. 2 Pollen deposited in a single visit by each species group.

Box indicates quartiles with median marked as a horizontal line;

points are outliers and whiskers (error bars) above and below

the box indicate the 90th and 10th percentiles.

Mea

n po

llen

depo

sitio

n pe

r vis

it

Year

Fig. 3 Mean day-long pollen deposition per flower �SE as esti-

mated by the simulation for each taxon.



Altogether, in the absence of taxa that increase under

future climate change, pollination services would be

expected to decline by 15.3% (Fig. 4). When for compar-

ative purposes we calculated present-day variation in

pollination services across space, we found that the

mean percentage difference between farms was 47%

(Data S1).

Discussion

It is critical to understand how global food needs will be

met in the coming decades, given the predictions for

human population growth, intensifying land use, and

climate change (Foley et al., 2011; Tilman et al., 2011). To

date, estimates of how climate change might affect crop

production have focused on the direct effects of climate

warming on crop plants (e.g. Long et al., 2006; Tubiello

et al., 2007; Georgescu et al., 2011). Potential indirect

effects on insect pollination of crops have been little

studied, although most crop plants benefit from animal

pollination (Klein et al., 2007). Here, we show that cli-

mate warming will alter pollinator activity patterns, and

hence pollination, of a model pollinator-dependent crop

plant. However, despite large percentage changes in the

pollination provided by particular taxa, the aggregate

pollination services are predicted to change little (ca.

5%). This is because differential responses to climate

warming by diverse taxa pollinating watermelon may

buffer pollination services against climate change.

Our results are consistent with the biological insur-

ance hypothesis, which proposes that the maintenance

of ecosystem functions and services will be enhanced by

having a diverse assemblage of species providing the

service (Lawton & Brown, 1993). Specifically, our results

suggest that a mechanism known as response diversity

may stabilize pollination services against climate

change. Response diversity occurs when multiple spe-

cies providing the same ecosystem service respond dif-

ferentially to environmental change, thus buffering the

aggregate service (Winfree & Kremen, 2009; Lalibert�e

et al., 2010; Karp et al., 2011). Specifically, we found that

pollination services from the sole managed agricultural

pollinator in this system, the honey bee, are predicted to

decline under warming, but these decreases are offset

by increases in the pollination provided by wild taxa

(Fig. 3). The buffering effects of wild pollinators are par-

ticularly strong because in our study system, aggregate

pollination services provided by wild taxa exceed the

services provided by honey bees (Fig. 4).

The honey bee is the main pollinator of agricultural

crops in many parts of the world, thus changes in the

pollination services it provides will have large ramifi-

cations for crop production (Free, 1993; Klein et al.,

2007). A. mellifera populations have already been nega-

tively affected by a number of pests and diseases

worldwide (Oldroyd, 2007; van Engelsdorp et al.,

2008). Our results predict that climate warming will

further reduce the pollination services provided by

honey bees to midsummer crops in our climatic zone,

because we found that honey bees were most active on

flowers at the relatively cool temperatures of 24–30 °C.Our models predict that future warmer temperatures

will be beyond the optimal activity range for honey

bees (Fig. 1), causing a 14.5% reduction in the pollina-

tion services they provide. The reduced flower visita-

tion by honey bees that we observed above 35 °C is

unlikely to be directly due to thermoregulatory ability

because A. mellifera can tolerate temperatures above

40 °C (Cooper et al., 1985). Rather, the decrease is

likely to be indirect due to the behavioural changes

associated with high temperatures, including reduced

foraging duration and flight distances as bees spend

more time collecting water for hive thermoregulation

(Cooper et al., 1985).

(a)

(b)

Fig. 4 Mean day-long pollen deposition per flower �SE as esti-

mated by the simulation for all taxa, wild bees and Apis mellifera.



As our model focuses on the plant–insect interactionas opposed to the response by either partner (i.e. poten-

tial changes in the watermelon plant’s flowering phe-

nology independent of its pollinators, or the activities

of the insect pollinators not connected with watermelon

flowers), there are several potential effects that are not

included in our estimates of pollen deposition. First,

just as the seasonal phenologies of plants and pollina-

tors have already shifted with climate warming (Bar-

tomeus et al., 2011), plant and pollinator diurnal

phenologies may also adapt to enable activity at an ear-

lier, cooler time of day. In plants, this response has

recently been demonstrated in rice whereby early

flower opening in response to warmer air temperatures

helped to avoid sterility caused by heat stress at anthe-

sis (Kobayasi et al., 2010). Although we do not include

plant or pollinator diurnal adaptation in our model,

our results are conservative with respect to this possi-

bility. If watermelon plants shift their diurnal phenol-

ogy to earlier in the day, in parallel to the changes we

predict for pollinators, then changes in future aggregate

pollination function are likely to be even smaller than

the modest changes predicted here, which are them-

selves smaller than the present-day variation in pollina-

tion services among sites (Data S1).

Second, our approach does not fully incorporate

long-term adaptations that may occur once farmers

adjust their expectations of future climate. These

changes could include increasing the number of honey-

bee hives supplied to insect-pollinated crops, expansion

of crop area into cooler regions, planting of new crop

varieties (Liu et al., 2010) or altering of planting dates

(Lauer et al., 1999). Variation in the seasonal availability

of crop flowers could, in turn, result in different polli-

nator species being present, as could geographical

range shifts of pollinators in response to climate change

(Parmesan et al., 1999; Memmott et al., 2007; Deutsch

et al., 2008; Bartomeus et al., 2013). Either scenario

could secondarily affect the diurnal timing of pollinator

activity, but such changes are beyond our ability to

predict at present.

Our investigation of behaviourally mediated changes

in pollination services predicts that climate warming

will affect pollinator species’ activity patterns differen-

tially, causing some to increase and others to decrease

in their provision of pollination services. Due to the dif-

ferential responses among taxa and the diversity of taxa

that pollinate watermelon, aggregate pollination ser-

vices delivered to crop flowers are predicted to increase

slightly, even under the most extreme climate change

scenario. Because in the absence of wild taxa, pollina-

tion services would be predicted to decline, our results

demonstrate the importance of native biodiversity in

buffering the impacts of environmental change.

Acknowledgements

Data collection was funded by an Endeavour Australia Postdoc-toral Research Fellowship (RR), a USDA-AFRI grant #2009-65104-05782 to R Winfree (PI) and N Williams (Co PI), a NSFBIO DEB collaborative grant #0554790/0516205 to C. Kremen(PI), N. M. Williams (PI), and R. Winfree (Co-PI), and RutgersUniversity faculty start-up funds to R Winfree. I.B. was fundedby a postdoctoral fellowship EX2009-1017 from the SpanishMinistry for Education. We are grateful to S. Cunningham,D. Cariveau, and two anonymous reviewers for insightful com-ments on previous drafts of the manuscript, and to J. Ascherand J. Gibbs for species identification of our bee specimens.

References

Bartomeus I, Ascher JS, Wagner D, Danforth BN, Colla S, Kornbluth S, Winfree R

(2011) Climate-associated phenological advances in bee pollinators and bee-

pollinated plants. Proceedings of the National Academy of Sciences, 108, 20645–20649.

Bartomeus I, Ascher JS, Gibbs J, Danforth BN, Wagner DL, Hedtke SM, Winfree R

(2013) Historical changes in northeastern US bee pollinators related to shared eco-

logical traits. Proceedings of the National Academy of Sciences, 110, 4656–4660.

Bishop JA, Armbruster WS (1999) Thermoregulatory abilities of Alaskan bees: effects

of size, phylogeny and ecology. Functional Ecology, 13, 711–724.

Bluthgen N, Klein AM (2011) Functional complementarity and specialisation: the role

of biodiversity in plant-pollinator interactions. Basic and Applied Ecology, 12, 282–

291.

Bremner D, Demaine E, Erickson J, Iacono J, Langerman S, Morin P, Toussaint G

(2005) Output-sensitive algorithms for computing nearest-neighbour decision

boundaries. Discrete and Computational Geometry, 33, 593–604.

Brittain C, Kremen C, Klein A-M (2013) Biodiversity buffers pollination from changes

in environmental conditions. Global Change Biology, 19, 540–547.

Cooper PD, Schaffer WM, Buchmann SL (1985) Temperature regulation of honey bees

(Apis mellifera) foraging in the Sonoran desert. Journal of Experimental Biology, 114,

1–15.

Delaplane KS, Mayer DF (2005) Crop Pollination by Bees. CABI Publishing, Walling-

ford, United Kingdom.

Deutsch CA, Tewksbury JJ, Huey RB, Sheldon KS, Ghalambor CK, Haak DC, Martin

PR (2008) Impacts of climate warming on terrestrial ectotherms across latitude.

Proceedings of the National Academy of Sciences, 105, 6668–6672.

Elmqvist T, Folke C, Nystrom M, Peterson G, Bengtsson J, Walker B, Norberg J (2003)

Response diversity, ecosystem change and resilience. Frontiers in Ecology and Envi-

ronment, 1, 488–494.

Foley JA, Ramankutty N, Brauman KA et al. (2011) Solutions for a cultivated planet.

Nature, 478, 337–342.

Free JB (1993) Insect Pollination of Crops. Academic Press, Harcourt Brace Jovanovich,

Publishers, London.

Garibaldi LA, Steffan-Dewenter I, Winfree R et al. (2013) Wild pollinators enhance

fruit set of crops regardless of honey bee abundance. Science, 339, 1608–1611.

Georgescu M, Lobell D, Field K (2011) Direct climate effects of perennial bioenergy

crops in the United States. Proceedings of the National Academy of Sciences of the Uni-

ted States of America, 108, 4307–4312.

Gregory PJ, George TS (2011) Feeding nine billion: the challenge to sustainable crop

production. Journal of Experimental Botany, 62, 5233–5239.

Hayhoe K, Wake C, Anderson B et al. (2008) Regional climate change projections for

the Northeast USA. Mitigation and Adaptation Strategies for Global Change, 13, 425–

436.

Heard TA, Hendrikz JK (1993) Factors influencing flight activity of colonies of the

stingless bee Trigona carbonaria (Hymenoptera, Apidae). Australian Journal of Zool-

ogy, 41, 343–353.

Heinrich B (1974) Thermoregulation in endothermic insects. Science, 185, 747–756.

Heinrich B (1979) Keeping a cool head: honeybee thermoregulation. Science, 205,

1269–1271.

Heinrich B (1980) Mechanisms of body-temperature regulation in honey bees, Apis

mellifera. I. Regulation of head temperature. Journal of Experimental Biology, 85, 61–

72.

Heinrich B (1993) The hot-Blooded Insects: Strategies and Mechanisms of Thermoregulation.

Harvard University Press, Cambridge, MA.

Herrera CM (1997) Thermal biology and foraging responses of insect pollinators to

the forest floor irradiance mosaic. Oikos, 78, 601–611.



IPCC (2007) The Physical Science Basis. Contribution of Working Group I to the Fourth

Assessment Report of the Intergovernmental Panel on Climate Change. (ed. Solomon S,

Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL). IPCC,

Cambridge, UK.

Jarlan A, Oliveira DD, Gingras J (1997) Pollination by Eristalis tenax (Diptera: Syrphi-

dae) and seed set of greenhouse sweet pepper. Journal of Economic Entomology, 90,

1646–1649.

Jauker F, Wolters V (2008) Hover flies are efficient pollinators of oilseed rape. Oecolo-

gia, 156, 819–823.

Karp DS, Ziv G, Zook J, Ehrlich PR, Daily GC (2011) Resilience and stability in bird

guilds across tropical countryside. Proceedings of the National Academy of Sciences,

108, 21134–21139.

Klein AM, Vaissiere BE, Cane JH, Steffan-Dewenter I, Cunningham SA, Kremen C,

Tscharntke T (2007) Importance of pollinators in changing landscapes for world

crops. Proceedings of the Royal Society B-Biological Sciences, 274, 303–313.

Kobayasi K, Matsui T, Yoshimoto M, Hasegawa T (2010) Effects of temperature, solar

radiation, and vapor-pressure deficit on flower opening time in rice. Plant Produc-

tion Science, 13, 21–28.

Kraus B, Page RE (1995) Effect of Varroa jacobsoni (Mesostigmata: Varroidae) on feral

Apis mellifera (Hymenoptera: Apidae) in California. Environmental Entomology, 24,

1473–1480.

Lalibert�e E, Wells JA, Declerck F et al. (2010) Land-use intensification reduces func-

tional redundancy and response diversity in plant communities. Ecology Letters,

13, 76–86.

Lauer JG, Carter PR, Wood TM, Diezel G, Wiersma DW, Rand RE, Mlynarek MJ

(1999) Corn hybrid response to planting date in the northern corn belt. Agronomy

Journal, 91, 834–839.

Lawton JH, Brown VK (1993) Redundancy in ecosystems. In: Biodiversity and Ecosys-

tem Function (eds Schluze ED, Mooney HA), pp. 255–270. Springer, New York.

Liu Y, Wang E, Yang X, Wang J (2010) Contributions of climatic and crop varietal

changes to crop production in the North China Plain, since 1980s. Global Change

Biology, 16, 2287–2299.

Lobell DB, Schlenker W, Costa-Roberts J (2011) Climate trends and global crop

production since 1980. Science, 333, 616–620.

Long SP, Ainsworth EA, Leakey ADB, N€osberger J, Ort DR (2006) Food for thought:

lower-than-expected crop yield stimulation with rising CO2 concentrations.

Science, 312, 1918–1921.

Memmott J, Craze PG, Waser NM, Price MV (2007) Global warming and the

disruption of plant-pollinator interactions. Ecology Letters, 10, 710–717.

Morris WF, Pfister CA, Tuljapurkar S et al. (2008) Longevity can buffer plant and

animal populations against changing climatic variability. Ecology, 89, 19–25.

NECIA (2007) Synthesis report of the Northeast Climate Impacts Assessment

(NECIA). In: Confronting Climate Change in the U.S. Northeast: Science, Impacts, and

Solutions (eds Frumhoff PC, McCarthy JJ, Melillo JM, Moser SC, Wuebbles DJ),

Union of Concerned Scientists (UCS), Cambridge, MA.

Oldroyd BP (2007) What’s killing American honey bees? PloS Biology, 5, e168.

Parmesan C (2006) Ecological and evolutionary responses to recent climate change.

Annual Review of Ecology, Evolution, and Systematics, 37, 637–669.

Parmesan C, Ryrholm N, Stefanescu C et al. (1999) Poleward shifts in geographical

ranges of butterfly species associated with regional warming. Nature, 399, 579–583.

Parton WJ, Logan JA (1981) A model for diurnal variation in soil and air temperature.

Agricultural and Forest Meteorology, 23, 205–216.

Porter JR, Semenov MA (2005) Crop responses to climatic variation. Philosophical

Transactions of the Royal Society B: Biological Sciences, 360, 2021–2035.

Potts SG, Biesmeijer JC, Kremen C, Neumann P, Schweiger O, Kunin WE (2010a) Glo-

bal pollinator declines: trends, impacts and drivers. Trends in Ecology & Evolution,

25, 345–353.

Potts SG, Roberts SPM, Dean R, Marris G, Brown M, Jones R, Settele J (2010b)

Declines of managed honeybees and beekeepers in Europe? Journal of Apicultural

Research, 49, 15–22.

R Development Core Team (2012) R: A Language and Environment for Statistical Com-

puting. R Foundation for Statistical Computing, Vienna, Australia. Available at:

www.R-project.org.

Rader R, Howlett BG, Cunningham SA et al. (2009) Alternative pollinator taxa are

equally efficient but not as effective as the honeybee in a mass flowering crop. Jour-

nal of Applied Ecology, 46, 1080–1087.

Sparks TH, Yates TJ (1997) The effect of spring temperature on the appearance dates

of British butterflies 1883-1993. Ecography, 20, 368–374.

Stanghellini MS, Schultheis JR, Ambrose JT (2002) Pollen mobilization in selected

Cucurbitaceae and the putative effects of pollinator abundance on pollen depletion

rates. Journal of the American Society for Horticultural Science, 127, 729–736.

StoneGN (1994) Activity patterns of females of the solitary beeAnthophora plumipes in rela-

tion to temperature, nectar supplies and body size.Ecological Entomology, 19, 177–189.

Tilman D, Balzer C, Hill J, Befort BL (2011) Global food demand and the sustainable

intensification of agriculture. Proceedings of the National Academy of Sciences, 108,

20260–20264.

Tubiello FN, Soussana J-F, Howden SM (2007) Crop and pasture response to climate

change. Proceedings of the National Academy of Sciences, 104, 19686–19690.

van Engelsdorp D, Hayes J, Underwood RM (2008) A survey of honey bee colony

losses in the U.S., Fall 2007 to Spring 2008. PLoS ONE, 3, e4071.

Walker B, Kinzig A, Langridge J (1999) Plant attribute diversity, resilience, and

ecosystem function: the nature and significance of dominant and minor species.

Ecosystems, 2, 95–113.

Willmer PG, Stone GN (2004) Behavioral, ecological, and physiological determinants

of the activity patterns of bees. Advances in the Study of Behavior, 34, 347–466.

Willmer PG, Unwin DM (1981) Field analyses of insect heat budgets: reflectance, size

and heating rates. Oecologia, 50, 250–255.

Winfree R (2008) Pollinator-dependent crops: an increasingly risky business. Current

Biology, 18, R968–R969.

Winfree R, Kremen C (2009) Are ecosystem services stabilized by differences among

species? A test using crop pollination. Proceedings of the Royal Society B, 276, 229–237.

Winfree R, Williams NM, Dushoff J, Kremen C (2007) Native bees provide insurance

against ongoing honey bee losses. Ecology Letters, 10, 1105–1113.

Supporting Information

Additional Supporting Information may be found in theonline version of this article:

Data S1. Supplementary details concerning methodology.Figure S1. Plot of temperature and time of day for the 91sampling days.Figure S2. Regression lines demonstrating the decline inpollen deposition with time.Figure S3. Time-temperature response surfaces of varianceamong the eight species.Table S1. Species composition of the five taxonomic groups.Table S2. Error rate assigning pinned specimens to speciesgroups.Table S3. Differences in pollen deposition within speciesgroups.Table S4. Differences in pollen deposition among speciesgroups.



TEXT S1 - SUPPORTING METHODS AND RESULTS

MANUSCRIPT ID GCB-13-0326

MANUSCRIPT TITLE: Native bees buffer the negative impact of climate warming on honey

bee pollination of watermelon crops

Pollinator species and species groups

The species composition of each “on the wing” observation group was determined based on the

physical specimens collected at the same sites on the same days. The number of species per

group varied from 3-15 (Table S1). Note the honey bee A. mellifera is not included as it was not

netted for the purposes of identification, but this is a uniquely easy species to identify to the

species level on the wing as it is non-native and has no congeners in North America.

We measured our error in the definition of the observation groups by having field workers assign

pinned specimens to observation groups prior to the species-level identification being done.

Error rates on the species group assignations were 0.5-18.8% (see Table S2). The only groups

with >5% error were small dark and tiny dark, for which the errors were primarily due to mis-

assignations between these two categories. This is not surprising as these two groups are created

from continuous variation in body size among morphologically similar species. Nonetheless, we

maintained separate groupings for our analyses as body sizes were sufficiently different among

small and tiny dark taxa to expect potential variation in activity patterns. Further, the tiny dark

group was dominated by a single species which was readily identifiable to this group, hence a

separate tiny dark group likely better represents these taxa than if the group were merged with

small dark.

¨

Assumptions made by using observation groups

For the five observation groups that contain more than one species, we assume the species within

a given species group have similar values of single visit pollen deposition. This assumption

seems reasonable given that temperature-dependent activity and pollen deposition are likely

functions of phylogeny and morphology (body size, hairiness), and species grouped into the

same observation group are usually both phylogenetically and morphologically similar. We

tested this assumption about similarity of pollen deposition among species within a group by

collecting species-level data for a subset of species. We did this by catching each bee after it

visited the single, experimental flower, and then comparing species-level pollen deposition

among species within the same observation group. We used a general linear model with “pollen

deposition” as the log-transformed response variable and “species” as the fixed factor.

To determine differences among species we used Tukey’s Honestly Significant Difference

(HSD) post-hoc test, adjusted for multiple comparisons (“multcomp” package, R Development

Core Team (2012). All comparisons of within-group taxa were statistically insignificant (Table

S3). In contrast, comparisons in single-visit pollen deposition across groups were often

significant (see Table S4).

The second assumption made by the use of on-the-wing observation groups is that diurnal

activity patterns are similar for the species within a group. However, as our capacity to identify

taxa to species was limited, particularly for the more species-rich groups, we do not know if

daily activity patterns differ among taxa within species observation groups. Although we

obtained records of species abundances using netted specimens, we did not believe this was

sufficiently representative of visitation rate. This is because some bees are easier to capture than

others (hence behavioral differences could bias sampling toward particular species) and

individual abundances obtained from using net data, assume each pollinator has made a single

visit and does not account for multiple visits by each individual. This second assumption,

however, is conservative with respect to our main result, because it may result in an

underestimate of the buffering capacity provided by native bees in our study, if in reality some

differential responses and thus buffering occurs within the species groups.

Differences among the eight taxonomic groups in single visit pollen deposition

To determine if species differed in single visit pollen deposition we used a Kruskal-Wallis rank

sum test and post-hoc pairwise comparisons with Behrens-Fisher- and Steel-type nonparametric

multiple comparisons (“npmc” package, R Development Core Team 2012). Taxa differed in

pollen deposition (Kruskal-Wallis chi-squared = 80.5626, df = 7, P value = 1.058e-14, Table

S4).

Creating visitation surfaces and estimating diurnal temperature profiles

We chose k= 30 points to calculate the nearest neighbor algorithm as this provided a good

balance between sample size and spatial resolution. When we used fewer points (k=15), this

resulted in a higher variance because each within cell value is based on only a few empirical

measures of visitation rate. Greater than 30 points resulted in visitation rates at times and

temperatures too far away from the cell being estimated. Picking the nearest 30 points requires

calculating distances between the cell of interest and all available data points. For the purposes

of calculating these distances on the map we scaled 1 hr of time to 2 °C of temperature as this

combination roughly matched the variation of time and temperature within our system (on a

typical day, temperature increases by roughly 10 °C over 5 hrs from 8.00 to 13.00 hrs). As a

form of sensitivity analysis, we generated surfaces for three different distance metrics: 2 °C = 1

hr, 3 °C = 1 hr, and 1 °C =1 hr. The resulting surfaces were very similar, especially in the

densely sampled central region. Full runs of the model with these three algorithms differed

marginally in exact predicted percentage increases and decreases for particular taxa, but the

overall conclusions of the study were unchanged.

We estimated daily temperature profiles using a truncated sine curve, following the methods of

Parton and Logan (1981). Shape parameters of the curve were estimated by fitting a dataset of

hourly temperature measurements from the nearest weather station to each sampling site on the

day that it was visited for sampling, resulting in a curve with a maximum at 15.00 and a

minimum at 6.00 (~sunrise). This curve achieved a much better fit to our data than a simple sine

curve with the same minimum and maximum (Figure S1).

Spatial variation in pollinator activity

Accounting for variation in bee abundance across sites

Within-species variation in bee abundance across sites could potentially bias our estimates of bee

activity as a function of temperature. We took several steps to minimize this possibility. First,

we visited all farms at both high and low temperatures to ensure equal representation of

pollinator visits across the full temperature range of our study. Second, prior to analysis for each

species, we excluded farm-year combinations in which that species was not present at all. We

did this to ensure our estimates of visitation were based on taxa known to be present, and hence

potentially active, at each specified temperature and time interval. Third, we incorporated

variation in the visitation rate at a given time-temperature combination using the same nearest-

neighbor method used to calculate the mean visitation rates (see visitation surface maps). Thus

we had two separate visit surfaces for each species, one for means (Fig. 1) and one for variances

(Fig. S3). Both the mean and variance visitation surfaces were used to obtain expected means

and variances for the number of visits that flowers receive as a function of time of day and

temperature. In this way any variability in background population abundance across sites would

be incorporated into to the uncertainty we propagate through our estimates of the temperature-

bee activity relationship.

Present-day spatial variability in pollinator visitation rates

To contextualize our predictions for changes in pollination function due to climate change, we

estimated the present-day variability in pollination function across space. To do this, we used a

simpler version of the model (Winfree et al., 2007) to calculate the mean number of pollen grains

deposited per flower on a per-site basis. We used the largest of our four annual data sets for this,

the 23 sites studied in 2005. We then calculated the absolute value of the percentage change in

pollination function between all possible pairs of sites. The mean of these percentage changes

between sites was 47%.

Preparation of model input data

Single-visit pollen deposition

We combined the pollen deposition data across farms on the assumption that per-visit pollen

deposition by a bee foraging within a patch of watermelon is a function of bee morphology and

behaviour, i.e. species identity, rather than site, as has been assumed by other studies using

similar methods (Greenleaf & Kremen, 2006, Kremen et al., 2002, Vazquez et al., 2005,

Winfree, 2007). Because single-visit pollen deposition per visit declined over the course of the

day (morning versus afternoon number of pollen grains deposited per visit, z = 4.05, P= 0.0001,

Fig. S2), likely due to the depletion of pollen in anthers dehisced in the morning (Stanghellini et

al., 2002), we included time of day in our estimates of pollination function. For each bee taxon,

we estimated single visit pollen deposition as a function of time of day using linear regression.

The pollen data were log(x+1) transformed to achieve normality of residuals and

homoscedasticity. Subsequently, each value of per-visit pollen deposition as used in the

simulation was obtained via the following three-step process:

1) To incorporate uncertainty about mean pollen deposition amounts, a mean pollen

deposition value was drawn from a normal distribution with a mean equal to the predicted

pollen deposition amount at a given time of day from the regression line, and a standard

error equal to the standard error of the residuals.

2) To incorporate the individual variability in pollen deposition between visits, we drew a

single pollen deposition amount to go with each individual visit (at the given time of day)

from a gamma distribution with the drawn mean and a standard deviation equal to the

standard deviation (representing real variability) of the residuals.

3) Because the above analysis was done using log-transformed data, the pollen amount was

then back-transformed to the original scale.

In sum, instead of drawing an actual data value for pollen grains per visit from a particular time

of day, we drew a value from a distribution based on the mean and standard error predicted for

that time of day by a linear regression analysis. This technique incorporates our uncertainty

about the actual value of pollen deposition at that time of day, given the variability in the data.

Calculations of standard error for pollen deposition estimates

We used the following formula to calculate the standard error (SE) of our pollen deposition

estimates where 175 = (35 days per season multiplied by 5 years; this is the number of days over

which we are estimating mean pollen deposition) and the variance in pollen deposition per

flower (σ2p) is calculated across the 5000 iterations (days) of the simulation.

SE = sqrt (σ2p / 175 )

References Cited: Greenleaf S. S., Kremen C. (2006) Wild bees enhance honey bees' pollination of hybrid

sunflower. Proceedings of the national academy of sciences of the United States of America, 103, 13890-13895.

Kremen C., Williams N. M., Thorp R. W. (2002) Crop pollination from native bees at risk from agricultural intensification. Proceedings of the national academy of sciences of the United States of America, 99.

Parton W. J., Logan J. A. (1981) A model for diurnal variation in soil and air temperature. Agricultural Meteorology, 23, 205-216.

R Development Core Team (2012) R: A language and environment for statistical computing, Vienna, Australia, R Foundation for Statistical Computing. www.R-project.org.

Stanghellini M. S., Schultheis J. R., Ambrose J. T. (2002) Pollen mobilization in selected Cucurbitaceae and the putative effects of pollinator abundance on pollen depletion rates. Journal of the American Society for Horticultural Science, 127.

Vazquez D. P., Morris W. F., Jordano P. (2005) Interaction frequency as a surrogate for the total effect of animal mutualists on plants. Ecology Letters, 8, 1088-1094.

Winfree R., Griswold, T. And 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. (2007) Native bees provide insurance against ongoing honey bee losses. Ecology Letters, 10, 1105-1113.

Figure S1. Plot of temperature and time of day values for the 91 sampling days in our

bee visitation dataset. The black line is a loess smooth line through the data points,

which rises approximately linearly through 11:00 and levels off to a maximum at 15:00.

Solid red and blue lines show the fitted truncated sine curves using minimum and

maximum temperature values predicted by IPCC scenarios A1 (blue) and B1 (red) for

the same time period covered by our data (2005-2010). Thus, the truncated sine curve

provides a good fit to the data. In contrast, a simple sine curve, represented by the

dotted lines, does not provide a good fit.

Fig. S2: Regression lines for each of the eight species groups demonstrating the

decline in single-visit pollen deposition with time of day.

Fig. S3: Time-temperature visitation surfaces of variance among the eight species.

Activity is measured as visits per flower per hour, with regions of higher activity plotted

as red and lower activity as yellow. The markings on the x and y axes (the “rug”)

indicates sampling effort; areas with low sampling effort are an extrapolation from the

data. Black lines represent A1 and grey lines represent B1 climate change scenario.

Solid lines are 2005-2010, dashed lines are 2050-2055, dotted lines are 2094-2099.

Table S1: Species composition of the five groups that comprised more than one

species based on physical specimens collected at the same sites on the same days.

Species ID group Species composition Abundance of each species

Bombus impatiens B. impatiens B. fervidus B. bimaculatus B. griseocollis B. vagans B. perplexus

1155 8 3 4 4 1

Ceratina C. calcarata/dupla C. strenua

287 57

Green bees Augochlora pura Augochlorella aurata Augochloropsis metallica

616 113 3

Small dark Lasioglossum versatum Halictus confusus L. pilosum L. oceanicum L. zephyrum L. ephialtum L. genuinum L. obscurum L. pectorale L. bruneri L. laevissiumum L. cressonii L. zophops L. pectinatum

292 169 175 29 19 14 13 10 10

9 7 4 3 3

Tiny dark L. imitatum L. hitchensi L. tegulare L. weemsi L. paradmirandum L. illinoense L. ellisiae

545 152 69 41 38 16 7

Table S2: The percentage error rate of field workers assigning pinned specimens to

species groups.

Species ID group Number of specimens identified

Error rate of field workers

Bombus impatiens 1083 1.8% Ceratina 319 1.9% Green bees 619 0.5% Melissodes bimaculata 108 4.6% Peponapis pruinosa 89 1.1% Small dark 588 12.2% Tiny dark 695 18.8%

Table S3: Analyses of within-group differences in single visit pollen deposition. Two

species groups, Ceratina and green bees, could not be analyzed in this way due to

insufficient species-specific records.

Small dark species group

Taxon 1 Taxon 2 Z-value P-value Lasioglossum cressonii Halictus confusus 1.810 0.572 Lasioglossum pilosum Halictus confusus 0.788 0.992 Lasioglossum zephyrum Halictus confusus 2.721 0.098 Lasioglossum pilosum Lasioglossum cressonii -1.446 0.808 Lasioglossum zephyrum Lasioglossum cressonii 0.744 0.994 Halictus versatum Lasioglossum pilosum -1.502 0.777 Lasioglossum zephyrum Lasioglossum pilosum 2.431 0.197

Tiny dark species group

Taxon 1 Taxon 2 Z-value P-value Lasioglossum hitchensi Lasioglossum imitatum -1.412 0.473 Lasioglossum tegulare Lasioglossum imitatum -1.428 0.463 Lasioglossum tegulare Lasioglossum hitchensi -0.501 0.955

Table S4: Differences in pollen deposition among species groups.

Taxon 1 Taxon 2 Statistic P value Small dark Tiny dark 1.58 0.621 Small dark Ceratina 0.66 0.987 Small dark Green bees 3.91 0.001 Small dark Apis mellifera 4.68 <0.001 Small dark Bombus impatiens 7.39 <0.001 Small dark Melissodes bimaculata 3.21 0.016 Small dark Peponapis pruinosa 4.33 <0.001 Tiny dark Ceratina -0.21 1.000 Tiny dark Green bees 1.86 0.435 Tiny dark Apis mellifera 2.97 0.034 Tiny dark Bombus impatiens 5.05 <0.001 Tiny dark Melissodes bimaculata 2.78 0.059 Tiny dark Peponapis pruinosa 3.60 0.004 Green bees Ceratina 1.16 0.864 Apis mellifera Ceratina 1.97 0.360 Bombus impatiens Ceratina 3.00 0.031 Melissodes bimaculata Ceratina 2.32 0.184 Peponapis pruinosa Ceratina 2.91 0.040 Apis mellifera Green bees 1.63 0.585 Bombus impatiens Green bees 3.71 0.002 Melissodes bimaculata Green bees 2.24 0.214 Peponapis pruinosa Green bees 2.90 0.042 Bombus impatiens Apis mellifera 1.57 0.626 Melissodes bimaculata Apis mellifera 1.61 0.598 Peponapis pruinosa Apis mellifera 1.86 0.429 Melissodes bimaculata Bombus impatiens 1.03 0.914 Peponapis pruinosa Bombus impatiens 0.81 0.966 Peponapis pruinosa Melissodes bimaculata -0.47 1.000

native bees buffer the negative impact of climate …...pollination of watermelon crops, we predict...

Documents