Ecology, 92(3), 2011, pp. 687–698� 2011 by the Ecological Society of America

Daily temporal structure in African savanna flower visitationnetworks and consequences for network sampling

KATHERINE C. R. BALDOCK,1,3 JANE MEMMOTT,2 JUAN CARLOS RUIZ-GUAJARDO,1 DENIS ROZE,1,4

AND GRAHAM N. STONE1,5

1Institute of Evolutionary Biology, University of Edinburgh, Kings Buildings, West Mains Road, Edinburgh EH93JT United Kingdom2School of Biological Sciences, University of Bristol, Woodland Road, Bristol BS8 3PZ United Kingdom

Abstract. Ecological interaction networks are a valuable approach to understandingplant–pollinator interactions at the community level. Highly structured daily activity patternsare a feature of the biology of many flower visitors, particularly provisioning female bees,which often visit different floral sources at different times. Such temporal structure impliesthat presence/absence and relative abundance of specific flower–visitor interactions (links) ininteraction networks may be highly sensitive to the daily timing of data collection. Further,relative timing of interactions is central to their possible role in competition or facilitation ofseed set among coflowering plants sharing pollinators. To date, however, no study hasexamined the network impacts of daily temporal variation in visitor activity at a communityscale. Here we use temporally structured sampling to examine the consequences of dailyactivity patterns upon network properties using fully quantified flower–visitor interaction datafor a Kenyan savanna habitat. Interactions were sampled at four sequential three-hour timeintervals between 06:00 and 18:00, across multiple seasonal time points for two sampling sites.In all data sets the richness and relative abundance of links depended critically on when duringthe day visitation was observed. Permutation-based null modeling revealed significanttemporal structure across daily time intervals at three of the four seasonal time points, drivenprimarily by patterns in bee activity. This sensitivity of network structure shows the need toconsider daily time in network sampling design, both to maximize the probability of samplinglinks relevant to plant reproductive success and to facilitate appropriate interpretation ofinterspecific relationships. Our data also suggest that daily structuring at a community levelcould reduce indirect competitive interactions when coflowering plants share pollinators, as iscommonly observed during flowering in highly seasonal habitats.

Key words: Africa; competition; ecological networks; facilitation; Kenya; mutualism; pollination;savanna; temporal structure; visitation webs.

INTRODUCTION

Pollination mutualisms drive one of the most impor-

tant ecosystem services on Earth and underlie much of

the planet’s terrestrial biodiversity (Kearns et al. 1998,

Bascompte and Jordano 2007). A growing body of work

addresses the community scale of pollination processes

by applying interaction network (web) approaches to

entire plant–pollinator communities. These approaches

allow quantification of both direct and indirect interac-

tions within and between trophic levels, allowing

examination of issues such as the basis of species

coexistence and the consequences of species addition

or loss (Memmott and Waser 2002, 2004, Traveset and

Richardson 2006, Bascompte and Jordano 2007, Lopez-

araiza-Mikel et al. 2007, Aizen et al. 2008).

Generating plant–pollinator interaction networks is

labor intensive and involves the summation of flower

visitation data collected over periods of days, weeks,

months, and occasionally years. In interpreting such

networks, it is important to remember first that not all of

the represented links between plants and pollinators

occur at the same time; and second, that for a given

network, the impact of some plant–pollinator interac-

tions depends critically on their relative timing. For

example, if shared pollinators visit sympatric plant taxa

flowering at the same time (coflowering), plants may

compete for pollinators (Waser and Real 1979, Rathcke

1983, Stone et al. 1996, 1998). However, if visitation by

shared pollinators is partitioned in time, then coflower-

ing plants may facilitate each others’ reproduction by

maintaining a larger pollinator population than any

could alone (Moeller 2004). Thus the interaction

between the two plant species can be negative or

positive, depending on the timing of their interactions

with pollinators.

Manuscript received 3 June 2010; revised 25 August 2010;accepted 2 September 2010. Corresponding Editor: R. A.Raguso.

3 Present address: School of Biological Sciences, Wood-land Road, Bristol BS8 1UG United Kungdom.

4 Present address: Station Biologique de Roscoff, CentreNational de la Recherche Scientifique, Adaptation etDiversite en Milieu Marin (UMR 7144), Place GeorgesTeissier, BP 74, 29682 Roscoff Cedex, France.

5 Corresponding author. E-mail: Graham.Stone@ed.ac.uk

687

There is widespread evidence for seasonal partitioning

of visitation by shared pollinators (Stiles 1977, Waser

and Real 1979, Pleasants 1980, Sargent and Ackerly

2008), and interaction network studies have addressed

temporal structure over the same timescale (Lundgren

and Olesen 2005, Basilio et al. 2006, Medan et al. 2006,

Kaiser-Bunbury et al. 2010). The consequences of

summing interactions over a field season were explored

by Jordano et al. (2003) in terms of ‘‘forbidden links,’’

specific plant–pollinator interactions that cannot happen

due to differences in plant and pollinator seasonal

phenology or structural incompatibilities despite their

apparent plausibility in a summed seasonal network.

For this reason Basilio et al. (2006) advise against

representing communities as a single network, but as a

seasonal series. However, with a single exception (Olesen

et al. 2008), network studies have yet to address the finer

scale temporal dynamics of plant–pollinator networks.

In particular, no study has yet examined variation in

network structure over time within a single day.

Whether such structure exists has implications for our

understanding of both the sign and magnitude of direct

and indirect interactions between species both in

pollination networks and more generally. Moreover it

also could inform and influence the collection and

interpretation of interaction data collected over longer

timescales.

There are good reasons to expect daily temporal

structure in plant–pollinator interactions. Many studies

show that different pollinator taxa are active at different

times of day in a given community (e.g., Herrera 1990,

Willmer and Stone 2004) and that a given pollinator visits

different plants at different times of day (e.g., Armbruster

and Herzig 1984, Stone 1994, Stone et al. 1998). Such

variation is driven by the interaction between ‘‘bottom-

up’’ influences of floral resource provision (e.g., daily

timing of flower opening and provision of floral rewards;

Armbruster and Herzig 1984, Stone et al. 1999) and ‘‘top-

down’’ influences of pollinator biology, including thermal

physiology (Stone 1994, Herrera 1995, Bishop and

Armbruster 1999, Stone et al. 1999), water balance

(Willmer 1988), sexual interactions (Stone 1995, Stone

et al. 1995), predation risk (Munoz and Arroyo 2004),

nesting cycles (Stone 1994, Willmer and Stone 2004), and

phylogenetic inertia (Cunningham 1991). Daily temporal

structuring of activity is particularly well-studied in bees,

whose endothermic flight physiology and need to forage

beyond their individual needs makes them uniquely

sensitive to both bottom-up and top-down structuring

factors (Herrera 1990, Stone 1994, Willmer and Stone

2004).

Whether plant–pollinator interactions show daily

temporal structuring at a community level has impor-

tant consequences for the interpretation of interactions

among coflowering plants that share pollinators. For

pollinators such as bees that regularly remove pollen

from their bodies, daily temporal partitioning of

visitation among coflowering plants can turn a poten-

tially negative interaction (competition for pollination)

into a positive interaction (facilitation of fruit set)(Armbruster and Herzig 1984, Stone et al. 1998, Raine et

al. 2007). Which of these scenarios holds generatesdiametrically opposing predictions of the impact of

species gain (by invasion or habitat restoration) or loss(by extinction). For example, loss of a cofloweringspecies can result in either competitive release or

facilitation collapse. Sampling schemes that potentiallyallow discrimination between these alternatives are

clearly valuable in predicting the pollination servicesimpact of species loss or gain on a community.

Further, if daily temporal structuring exists, recordingvisitation for a limited daily time period risks missing a

subset of plant–pollinator interactions, influencing boththe richness of links recorded and their relative abun-

dance. Although some network studies have recordedpollinator activity through much of the day (Dupont et

al. 2003, Lundgren and Olesen 2005, Basilio et al. 2006,Morales and Aizen 2006, Aizen et al. 2008), very few

specify when visitation to each plant taxon was recorded.Many network studies provide no information on the

daily timing of data collection (e.g., Forup and Memmott2005, Gibson et al. 2006, Nielsen and Bascompte 2007,

Olesen et al. 2008, Petanidou et al. 2008). Suchinformation is central to understanding many of theprocesses contributing to the structure of such networks,

in asking, for example, whether observation at a specificplant matched timing of floral reward release.

Here we use time-structured sampling to assess thesensitivity of inferred flower visitation webs to the daily

timing of data collection and to assess the evidence fordaily temporal structure at a community level. To do

this we constructed networks for four diurnal timeintervals (06:00–09:00, 09:00–12:00, 12:00–15:00, and

15:00–18:00) at four seasonal time points for twoflowering plant communities in a Kenyan savanna

habitat. Our first objective was to quantify variationbetween daily time intervals in flower visitation by

specific taxa. We used permutation-based null modelingto test the hypothesis that activity is randomly

distributed across the four time intervals. Rejection ofthis hypothesis implies sensitivity of network contribu-

tions of such taxa to daily timing of data collection. Oursecond objective was to examine the extents to which

any signature of daily temporal structure, and the taxacontributing such structure, are consistent throughseasonal time and across sites and years. Our third

objective was to reveal the impact of daily sampling timeon the richness of links sampled and on the proportion

of links missed when sampling is restricted to specifictime intervals.

METHODS

Study location and sampling dates

Data were collected in 2004 between 3 May and 31August at Mpala Research Centre (08170 N, 378520 E,

altitude 1800 m), Laikipia District, central Kenya (Ruiz-

KATHERINE C. R. BALDOCK ET AL.688 Ecology, Vol. 92, No. 3

Guajardo et al. 2010). Sampling was conducted at two

0.5-ha plots (50 3 100 m), named Turkana Boma (TB)

and Junction (JN), located 6 km apart in open Acacia/

Commiphora savanna bushland. These sites were select-

ed to differ in topography and flowering plant species,

allowing us to examine the evidence for daily temporal

structure in communities with differing species compo-

sitions. Site JN was on a rocky and dry escarpment with

sparse vegetation, while TB was on flatter, better

watered ground with a thicker vegetation cover (Bal-

dock 2007, Ruiz-Guajardo 2008, Ruiz-Guajardo et al.

2010). Flowering plant species richness was similar at

the two sites across the four sampling months (species

totals ¼ 64 at TB and 48 at JN), although only 40% of

flowering species were common to both sites (see

Appendix A). Data were collected at TB for four

separate monthly flower–visitor networks in May, June,

July, and August and at JN for three networks in June,

July, and August. These sampling periods were selected

to coincide with three distinct phases of the annual

flowering season at Mpala. Sampling in May and June

targeted the major annual flowering period following the

long seasonal rains; sampling in July targeted the main

dry season when flowering plant species richness was

low; and sampling in August targeted a short period of

seasonal rainfall when floral species richness was again

high (Baldock 2007, Ruiz-Guajardo 2008).

Sampling rationale

Our approach was to first examine in detail the

species-rich June networks at TB and JN, assessing the

evidence for daily temporal structure in two sites with

differing flowering plant communities (objective 1). We

constructed fully quantified interaction networks for

these data sets and used a null modeling approach (see

Analysis below) to quantify the extent to which activity

by specific taxa was randomly distributed through daily

time and hence sensitive to time of sampling. We then

summarize the results of methodologically identical null

model analyses for all monthly data sets at each site to

assess the consistency of daily temporal structure across

seasonal time and to identify those taxa showing

significant structure (objective 2). Finally, to illustrate

the effect that choice of observation time interval has on

network structure (objective 3), we summarize variation

in network parameters across time intervals for each

monthly network and time interval network. We identify

the number of unique links per time interval (i.e.,

specific flower–visitor interactions observed in no other

time interval in that network) for all monthly networks

at both sites and quantify the impact, in terms of

numbers of links missed, of restricting sampling to each

single three-hour time interval.

Data collection

Data for each network were collected over two

consecutive weeks, the minimum required to devote

adequate observation effort to each floral source, with at

least two weeks between successive monthly networks.

Floral resource abundance in each plot was quantified at

the start of each week by counting the number of floral

units (defined as an individual flower or collection of

flowers that an insect of ;0.5 cm body length could

walk within or fly between [Saville 1993]) for each plant

species present. Flowering plants were identified to

species using keys and descriptions in Blundell (1992)

and Agnew and Agnew (1994) or to genus and

morphospecies.

In each plot and month, visitation was quantified for

all plant species with at least 10 floral units. Insect visits

to flowers of each plant species were recorded during 20-

min observations in each of four daily sampling intervals

(06:00–09:00, 09:00–12:00, 12:00–15:00, and 15:00–

18:00) encompassing the 12-h period between dawn

and dusk. While high levels of elephant activity

precluded network sampling at night, we checked

nocturnal visitation to the species present in our plots

for individuals near the Mpala Research Centre.

Flowers of most species were closed at night and only

two species, Carissa edulis (Apocynaceae) and Turraea

mombassana (Meliaceae), were visited (by hawkmoths,

shortly after dusk). The night-flowering species are

considered in more detail in the Discussion.

Each plant species was allocated one 20-min obser-

vation period in each daily time interval in which its

flowers were open, in each week that it flowered. This

resulted in between 20 and 160 min of observation per

species per monthly network. Only one set of observa-

tions could be conducted in August, and each species in

this month received a maximum of 80 min of

observation. To avoid bias in the timing of observations

for each plant species, order of observation periods in

each time interval was randomized across plant species.

Sampling was carried out simultaneously at TB and JN

in June, July, and August with sampling days alternating

between the two plots. Visitation sampling totaled 306 h

across the four months of data collection.

For each plant species and time interval, visitation was

observed for a recorded number of floral units occupying

;1 m2. One visit was recorded every time a visitor made

contact with the sexual parts of a floral unit. Return visits

to the same floral unit were counted as a second visit. All

visitors were insects in the orders Hymenoptera, Diptera,

Coleoptera, and Lepidoptera. Species were identified

visually or captured and identified either to species by

specialist taxonomists or to morphospecies using museum

collections. Butterflies were identified to species using

Larsen (1991). Ant visits were recorded, but were very

rare and are excluded from analyses due to low

taxonomic resolution. Insects that escaped before being

identified were lumped at the taxonomic level for which

identity was certain (family or order).

Analysis

Quantitative flower–visitor interaction networks.—Ful-

ly quantified flower–visitor interaction networks for the

March 2011 689DAILY FLOWER-VISITOR NETWORK STRUCTURE

complete (whole-day) and daily time interval June data

sets for each site were generated using software

developed by J. Memmott in Mathematica (Wolfram

Research, Champaign, Illinois, USA). Food web linkage

statistics adapted for use in plant–pollinator networks

were calculated for each monthly and time interval

network to examine the effect of daily time on network

structure. Linkage for plants and visitors represents the

mean number of links for plant species and visitor taxa,

respectively, and is a measure of plant and visitor

generalization. Linkage was calculated for visitor taxa as

lv ¼ L/V and plant species as lp ¼ L/Pv, respectively,

where L is the number of links recorded, V is the number

of visitor taxa recorded in the time interval, and Pv is the

number of plant species visited.

Null model analysis of temporal variation in flower–

visitor links.—Our underlying rationale is to compare

our field data to a null model that specifically precludes

daily temporal structure but is otherwise realistic. If

there is no daily temporal structure in our data, then the

number of different flower–visitor link types observed in

a given time interval should match that observed for an

equal quantity of data drawn at random from the

equivalent whole-day data set (i.e., across all four time

intervals). In contrast, if activity is highly structured in

daily time, specific links will be restricted to specific time

intervals, and the number of different link types

observed in specific time intervals will be lower than

the null expectation. For each time interval we

compared therefore the number of observed links with

that present in randomly generated data sets drawn from

the whole-day data set for the same month at the same

site. One thousand sets of four randomized time interval

networks were generated from each whole-day data set

using software written in Cþþby D. Roze. In each set of

random networks, each observed visit in the whole-day

data set was randomly assigned to one of the four time

intervals, subject to two constraints: (1) The total

number of visits per time interval matched the observed

value, thus preserving variation in absolute levels of

visitation through the day. (2) Only flower–visitor links

observed in the whole-day data set were permitted, since

good biological reasons may preclude particular links

(forbidden links sensu Jordano et al. 2003).

Null models were developed for two data sets for each

month at each site: one incorporated only visits made by

taxa identified to species, while a second incorporated all

recorded visits. The first approach is conservative in that

it minimizes any potentially artifactual impact of

lumped multispecies groupings, but it also necessitates

exclusion of data in all monthly networks (11–44% of

visits). The second approach included lumped taxa,

reflecting both the difficulty of visitor identification in

biodiverse tropical habitats and the desire to maximize

data content. For each time interval and site, departure

of the observed numbers of links from null expectations

was assessed using a conservative two-tailed approach,

i.e., we rejected the null hypothesis if the observed

frequency fell in the bottom 2.5% or above the top97.5% of the randomly generated values. Results are

presented for all visitor taxa combined and separatelyfor links involving bees, wasps, flies, beetles, and

butterflies and moths. This subdivision allows us toask which taxa contribute to any signature of temporalstructuring observed and reflects the fact that while the

timing of visitation may be highly structured for sometaxa (particularly bees; Willmer and Stone 2004) it may

not be for others.This null modeling approach involves four tests (one

per time interval) for each data set. Though thresholdsignificance (alpha) values are routinely adjusted for

multiple tests, the application of such proceduresremains an area of active debate (Perneger 1998, Moran

2003, Nakagawa 2004). Commonly applied corrections(such as Bonferroni ) can be overly conservative and

increase the risk of making a type II error. We thereforepresent the results of null model analyses both with

unadjusted threshold significance levels (i.e., P , 0.05, P, 0.01) and indicate those that remain significant after

the Dunn-Sidak adjustment (Quinn and Keough 2002)of the P , 0.05 threshold to P , 0.0127. We used the

Dunn-Sidak rather than the Bonferroni correction asthis approach slightly improves the power for eachcomparison (Quinn and Keough 2002). However, with

four tests, the adjusted significance level is very similarfor the standard Bonferroni correction (0.0125 with

Bonferroni correction). Thus all null model analysesoriginally significant at P , 0.01 remain significant after

adjustment (including all those on which our conclu-sions are based), while analyses significant only at P ,

0.05 are no longer significant. We present unadjustedresults, highlighting those for which the Dunn-Sidak

adjustment of threshold significance levels has anyimpact.

RESULTS

Objective 1: flower–visitor interactions are structuredin daily time

The June whole-day and time interval flower–visitornetworks are shown for TB in Fig. 1, and qualitatively

similar patterns are presented for JN in Fig. 2. Summarystatistics for these networks are provided in Table 1. We

recorded 174 links between 40 plant species and 98 insecttaxa across both sites. Bees were dominant flower

visitors (49% of recorded visits at TB and 81% at JN),contributing more visits than other taxa in time intervals

2, 3, and 4 at both sites and in time interval 1 at JN(Figs. 1 and 2). Bees were also the most species-rich

visitor taxon (48% of visitor species at TB and 54% atJN) and were involved in the greatest number of links

(53% at TB and 62% at JN).Visual comparisons of the time interval networks

show obvious temporal variation in flower–visitorinteractions (Figs. 1 and 2). Visit numbers, visitor

taxon richness, and link richness were all greater in

KATHERINE C. R. BALDOCK ET AL.690 Ecology, Vol. 92, No. 3

time intervals 2 (09:00–12:00) and 3 (12:00–15:00) than

in time intervals 1 (06:00–09:00) and 4 (15:00–18:00) at

both sites, with values consistently lowest in time

interval 1 (Table 1). The null model results show that

activity by visitor species is highly structured in daily

time: excluding time interval 1 at TB (and time interval

1 at JN when Dunn-Sidak correction is applied), the

numbers of links in each time interval were significant-

ly lower than in randomly assembled networks at both

sites (Fig. 3). This reflects the fact that few visitor

species were observed in more than one time interval

(18% and 15% of species at TB and JN, respectively).

The observed temporal structuring is due predomi-

nantly to temporal patterns in bee activity: numbers of

links involving bees were consistently below null

expectations (Fig. 3). Of the other visitor groups,

butterflies and moths (TB) and wasps (JN) also showed

significant departures from null expectations (Fig. 3

and see Appendix B for taxa with nonsignificant

results). Very similar patterns and conclusions were

reached in analyses including lumped taxa (see

Appendix C).

FIG. 1. Quantitative flower visitation networks for theTurkana Boma site at Mpala Research Centre, central Kenya inJune: (a) whole-day network (all data collected between 06:00and 18:00), (b) time interval 1 (06:00–09:00), (c) time interval 2(09:00–12:00), (d) time interval 3 (12:00–15:00), (e) time interval4 (15:00–18:00). In each network the rectangles represent insecttaxa (top row) and plant species (bottom row), and theconnecting triangles represent links between taxa. The widthof the rectangle for each plant species represents the number offloral units present in the plot. The width of the rectangle foreach visitor taxon represents the total number of visits to allplants made, and the widths of the connecting lines representthe number of visits observed for that link. Visitor taxa arecolor-coded as follows: red, bees (Hymenoptera); medium blue,wasps (Hymenoptera); light blue, flies (Diptera); dark blue,beetles (Coleoptera); yellow, butterflies and moths (Lepidop-tera). All networks are drawn to the same scale.

FIG. 2. Quantitative flower visitation networks for theJunction site at Mpala Research Centre, central Kenya in June:(a) whole-day network (all data collected between 06:00 and18:00), (b) time interval 1 (06:00–09:00), (c) time interval 2(09:00–12:00), (d) time interval 3 (12:00–15:00), (e) time interval4 (15:00–18:00). See legend to Fig. 1 for a full explanation of thenetwork diagrams.

March 2011 691DAILY FLOWER-VISITOR NETWORK STRUCTURE

Objective 2: daily temporal structure persiststhrough seasonal time

At both TB and JN, plant and visitor taxa in thenetworks varied across seasonal time (full details of

flower–visitor links observed in all networks can befound in Appendix D). Over the four months of

sampling, we recorded 267 links between 46 plantspecies and 123 insect taxa at TB and 124 links between30 plant species and 80 insect taxa at JN. Summary

statistics for all time interval and whole-day networks inMay, July, and August are provided in Appendix E. In

all months, patterns paralleled those seen in the Junenetworks: bees consistently made more visits than anyother taxon (40–71% of total visits) and were involved in

the greatest number of links (43–61%). Numbers ofvisits, visitor taxa richness, and link richness were

highest in either time interval 2 or 3 and lowest in timeinterval 1 in all networks (Table 1; Appendix E).

Activity by visitor species was structured in daily timein all networks except the July network at TB (and theJuly network at JN when Dunn-Sidak correction is

applied; Appendix F), due primarily to patterns in beeactivity; numbers of links involving bees were more

consistently below null expectations than for other

visitor groups (Fig. 3 and see Appendix G). Again, very

similar patterns were found in analyses including

lumped taxa (data not shown).

Objective 3: impact of observation time interval

on the network

Sampling of flower visitors was highly sensitive to the

timing of observation, with few species observed in

multiple time intervals. The most notable exception was

the honey bee, Apis mellifera, which was recorded in

time intervals 2–4 in three monthly networks (Appendix

H). In contrast, many plants were visited in more than

one time interval (43–78% at TB, 20–63% at JN), and a

notable minority were visited in at least three time

intervals (14–27% at TB, 31% at JN in June; see

Appendix H for examples). Thus, in most networks,

choice of recording time interval had a stronger impact

on the recorded species richness of visiting insects than

of visited plants. These specific patterns are reflected in

linkage summary statistics, which were consistently

higher for the overall network than for any constituent

time interval network in plants and visitors (Table 1;

Appendix E).

TABLE 1. Network summary statistics for the whole-day and time interval networks at the (a) Turkana Boma (TB) and (b)Junction (JN) sites at Mpala Research Centre, central Kenya in June 2004.

Network property

Time interval

Whole-day1 (06:00–09:00) 2 (09:00–12:00) 3 (12:00–15:00) 4 (15:00–18:00)

a) TB June

Visitor taxa 1 36 34 16 60Visitor species� 1 29 25 9 50Plant species observed (P) 23 31 32 23 32Plant species visited (Pv) 1 22 24 8 27Flower visits 1 118 199 54 372Flower–visitor links 1 48 50 17 95Connectance, C (%) 4.34 4.30 4.60 4.62 4.95Plant linkage 1.00 2.27 2.17 2.38 3.62Visitor linkage 1.00 1.35 1.49 1.19 1.58Unique links� 1 33 35 11 ���Percentage of unique links§ 100 69 70 65 ���Links missed if sampling restricted to interval 94 47 45 78 ���Percentage of links missed if sampling restricted

to interval99 49 47 82 ���

b) JN June

Visitor taxa 4 25 35 13 58Visitor species� 2 20 32 8 52Plant species observed (P) 29 33 34 25 35Plant species visited (Pv) 4 18 18 10 26Flower visits 11 129 247 118 505Flower–visitor links 4 30 49 14 89Connectance, C (%) 3.44 3.64 4.12 4.31 4.38Plant linkage 1.00 1.67 2.72 1.40 3.42Visitor linkage 1.00 1.20 1.40 1.08 1.53Unique links� 4 25 45 10 ���Percentage of unique links§ 100 83 92 71 ���Links missed if sampling restricted to interval 85 59 40 75 ���Percentage of links missed if sampling restricted

to interval96 66 55 84 ���

� Visitor taxa identified to species or morphospecies.� Defined as a link observed in no other time interval in the same whole-day network.§ Calculated as a proportion of the total links per time interval.

KATHERINE C. R. BALDOCK ET AL.692 Ecology, Vol. 92, No. 3

All time intervals in each whole-day network con-

tained unique links that were not observed in any othertime interval in the same network (Table 1; Appendix E).

Time intervals 2 and 3 contained the greatest numbers ofunique links in all monthly networks. However, consid-

eration of unique links as a proportion of the total

number of links in each time interval shows them to forma high proportion of total links in almost all time

intervals in each network (65–100%, see Appendix E).Exclusion of any time interval during sampling will thus

miss specific links. To investigate the effect of sampling

over a shorter daily time window, we calculated the

proportion of links that would be missed if sampling was

restricted to a single time interval (Table 1; Appendix E).At least 45% of links from the whole-day network would

be missed if sampling was restricted to any single 3-htime interval (Appendix E), with the impact minimized if

sampling is restricted to time intervals 2 or 3 as these

contained greater numbers of unique links.

DISCUSSION

Our results show that daily timing of data collection

has a significant impact on inferred structures of flower

visitation networks for these savanna communities.

FIG. 3. Comparisons between the number of flower–visitor links actually observed in each time interval (white) and the mean(695% CI) observed in 1000 randomized networks (gray) for visitors identified to species in the Turkana Boma (TB) and Junction(JN) June networks: (a) all visitor species at TB, (b) all visitor species at JN, (c) bees at TB, (d) bees at JN, (e) butterflies and mothsat TB, and (f ) wasps at JN. Significant differences between observed values and null expectations are indicated by asterisks. Resultssignificant at P , 0.01 remain significant with Dunn-Sidak correction of the threshold P , 0.05 for multiple tests. Equivalentfigures for nonsignificant taxa are shown in Appendix B.

* P , 0.05; ** P , 0.01 (non-adjusted significance levels).

March 2011 693DAILY FLOWER-VISITOR NETWORK STRUCTURE

With the three-hour interval resolution used here, most

flower–visitor links were restricted to a single time

interval. Visit frequency and link richness were both

highest in the middle of the day, although lower rates of

visitation earlier and later in the day both contributed

novel links to the networks. This temporal structuring

meant that restriction of observations to any single

three-hour time interval resulted in failure to detect a

large proportion of links observed through the day

(Table 1; Appendix E). We first consider the limitations

of our approach, then the causes and consequences of

temporal structure both in pollination networks and

more generally in other networks. We end by addressing

the implications of our results for pollination ecologists

with regard to both sampling networks in the field and

analyzing multiple data sets gathered from the literature.

Limitations

There are three limitations to our approach. First, we

could not generate high-resolution network data for

nocturnal flower visitation, when nocturnal pollinators

such as moths and bats would have been active. This

decision was driven by safety considerations and does

not weaken our conclusion that timing of data collection

matters in network construction. It is already well

known that plant–pollinator interactions are temporally

structured by night and day (Fenster et al. 2004) and

where nocturnal visitation is substantial there will be an

even greater need to incorporate all 24 hours of the

diurnal cycle into an appropriate sampling strategy if the

causes and consequences of network structure are to be

understood. To our knowledge, only one network study

(Devoto et al., in press) has considered nocturnal plant–

pollinator interactions (but see Clinebell et al. [2004] for

a multispecies approach incorporating 24-h sampling).

Second, we combined laboratory and field identification

of flower visitors. A drawback of laboratory identifica-

tion of captured specimens is that they cannot return to

make future visits. Other studies have removed this bias

by catching all insects (e.g., Lopezaraiza-Mikel et al.

2007, Forup et al. 2008), but we argue that to do so

would have caused local extinction of some rare flower

visitors in our networks. Finally, although all problem-

atic species were morphotyped by taxonomists, identi-

fication to species was not possible in all cases because

the insect fauna of Kenya is still not completely

described. The lack of any contrast between analyses

using specimens identified to species and those including

lumped morphotaxa suggests that uncertainty in species

identities has not significantly influenced our results or

associated inference.

Extent and potential causes of observed daily

temporal structure

Our null modeling approach revealed nonrandom

distribution of specific flower–visitor links through daily

time in almost all of our seasonal networks, a result to

be expected from the large literature on pollinator

activity patterns (e.g., Stone et al. 1988, Willmer and

Stone 2004) but rarely explicitly considered in the

network literature. The impact of daily time on network

structure was consistent across two sites differing in

floral community over six of seven seasonal data sets,

suggesting that it is a consistent feature at least of

savanna flower–visitor interaction networks. We do not

suggest that network data collection strategies are the

best way to demonstrate such structure, but it is clear

that it can be demonstrated in such data and should

therefore be acknowledged in sampling for interaction

networks. Bees showed the most consistent evidence of

daily temporal structure among forager taxa, consistent

with the dependence of female reproductive success on

both thermal physiology and floral resource availability

(Herrera 1990, 1995, Stone 1994, Stone et al. 1998,

Willmer and Stone 2004). The strength of the patterns in

bees may also reflect their dominance in our data, as in

many other flower–visitor networks (e.g., Memmott and

Waser 2002, Forup et al. 2008). More detailed analyses

of daily activity patterns in non-bee visitors would assist

our understanding of this observed contrast between

bees and other taxa.

Understanding the causes of temporal structure

requires understanding the relative contributions of

sampling error and the top-down and bottom-up

influences driving the timing of flower–visitor interac-

tions. Revealing temporal structure imposed by top-

down aspects of visitor biology requires finer resolution

activity pattern and behavioral data than collected in

this study (e.g., Stone et al. 1995, 1998), but our data do

reveal bottom-up constraints imposed by floral resourc-

es. Floral imposition of bottom-up control is sometimes

very obvious: in our networks flowers of the morning

glory Ipomoea kituiensis (Convolvulaceae) opened from

08:00 and were shriveled by noon, while flowers of Sida

ovata and S. schimperiana (Malvaceae) opened only

from 11:00 to 15:00. For these plants, visitation outside

these times is impossible, representing a daily temporal

equivalent of forbidden links sensu Jordano et al. (2003).

In other species, however, the timing of resource

provision is not apparent externally. For example,

pollen release in open flowers of the acacia Senegalia

(Acacia) brevispica (Fabaceae) at Mpala takes place

around noon each day (Baldock 2007), a process that is

clearly reflected in shifts in the visitors with links to this

species. Prior to pollen release, most visitation is by

flower-feeding beetles and nectar feeders, while after

pollen release most vists are by pollen-collecting bees

(Appendix H). As a result, the sets of links involving S.

brevispica before and after pollen release are very

different. (See also Stone et al. [1996, 1998] for similar

patterns for a set of acacia species in Tanzania.) Timing

of pollen release in acacias and other plants is sensitive

to microclimate and can vary at the same site between

days, with knock-on effects on the activity of associated

pollinators (e.g., Stone et al. 1998, 1999, Raine et al.

2007). Which component of the visitor community for

KATHERINE C. R. BALDOCK ET AL.694 Ecology, Vol. 92, No. 3

S. brevispica is detected will thus depend on when it is

observed. Examples of bottom-up structuring are also

apparent in many temperate communities, particularly

in bees (Willmer and Stone 2004). Examples include

crepuscular or nocturnal visitation of evening primroses

following late afternoon anthesis (Moody-Weis and

Heywood 2001) and tracking of pollen availability

within and among plant species by Anthophora solitary

bees (Stone 1994, Stone et al. 1999).

Repeatability of patterns across years and seasonal

temporal patterns

Though we generated detailed network data for one

year only, data collected in subsequent years suggest

that daily temporal structure is likely to persist across

years as well as across seasons and sites. We continued

to collect monthly data on floral abundance using the

same protocol for the TB and JN sites throughout 2005

and 2006 and found similar groups of plant species to

coflower in a given month over all three years (2004–

2006; Appendix A). Though equivalent community-level

visitation data are not available for subsequent years,

data collected in TB plot in June 2006 for 12 of the

flowering species sampled in the June 2004 network

support the conclusions based on the 2004 data (217

hours of observation; see Appendix I). Bees dominated

numbers of visits and flower–visitor links as in 2004, and

a high proportion (59%) of flower–visitor links occurred

in only one of the four time intervals. Null model

analysis (Appendix I) also revealed evidence of daily

temporal structure in the 2006 data, with the numbers of

bee and lepidopteran flower–visitor links significantly

lower than null expectations for at least two time

intervals. The repetition of these patterns across years is

consistent with their generation by fundamental aspects

of floral and visitor biology, as we suggest above.

Our seasonal data (Appendix D; to be analyzed in

detail elsewhere) show that in addition to the daily

temporal patterns discussed in detail below there is also

seasonal variation in the species and links involved.

Seasonal variation means that the plants interacting via

shared pollinators, positively or negatively, change over

time. Summing the likely selective impact of such

interactions thus requires incorporation of both seasonal

and daily timescales (Rathcke 1983, Stone et al. 1998).

Consequences of daily temporal structure

for network studies

Where detailed information on the timing of dehis-

cence and release of floral rewards exists, it is possible to

quantify biologically relevant visitation (i.e., by pollen

vectors) in appropriate limited daily time intervals (e.g.,

Armbruster and Herzig 1984, Stone et al. 1998, 1999,

Willmer and Stone 2004). However, few pollination

network studies sample visitation with such knowledge,

beyond the obvious constraints imposed by floral

opening and closure (as for our Ipomoea and Sida

examples). Sampling of visitation for a limited daily time

window without such information runs the risk of

recording visitation at times when floral rewards are

absent or outside periods when pollen can be harvested

or deposited on receptive stigmas. Visits recorded at

these times may miss pollen vectors altogether. Where

network studies have sampled for only a small portion

of the day in the absence of specific information on the

plants involved, we cannot assume that recorded links

constitute those associated with pollen dispersal and

seed set, with obvious implications for ecosystem service

conclusions made from such networks. In our study

failure to record visitation in any time interval would

omit a subset of link types. Unsurprisingly, the impact is

greatest for the two intervals during the middle of the

day, when visit frequency and visitor richness are

greatest. The numbers of interval-specific links are

lowest for the first and last daily time intervals (Table

1; Appendix E), but ignoring these two intervals would

still result in failure to detect certain links.

Studies that do sample across a broad daily time

window still risk misinterpretation of the relationships

among plants and pollinators if they do not consider the

existence of daily temporal structure. For example, Apis

mellifera visited four plant species at JN in June:

Ipomoea kituiensis (Convolvulaceae) in time interval 2,

Emilia discifolia (Asteraceae) and Hypoestes forskahlii

(Acanthaceae) in time interval 3, and Leucas glabrata

(Lamiaceae) in time interval 4 (Appendix H). The

summation of these data over a daily timescale could

lead to the conclusion that these four plant species are

competing for visits from the same pollinator and are at

risk from interspecific pollen transfer. However, our

data show that A. mellifera visited different plant species

at different times of day, as observed for honey bees in

other studies (e.g., Stone et al. 1996, 1998). Given that

honey bees regularly remove pollen from their bodies,

these plants could instead potentially facilitate each

other’s pollination (Rathcke 1983, Moeller 2004).

Daily temporal structure in species interactions could

have consequences for other types of ecological net-

works. The extent to which daily temporal variation is

important will depend on the timescale over which

interactions occur and whether species have daily

activity patterns that restrict the times at which they

are available to interact. For example, interactions

between parasitoids and their hosts are unlikely to be

structured in daily time since parasitism takes place over

hours or days. Predator–prey interactions, including

those between mutualistic ant guards and other insects,

are more likely to be structured in daily time since

interactions occur over short time periods and both

interacting species can have specific daily activity

patterns (Willmer and Stone 1997, Raine et al. 2002,

Kronfeld-Schor and Dayan 2003). Hori and Noda’s

(2001) study of intertidal rocky shore food webs

provides an elegant example of the impact of daily

variation in the physical environment on taxa involved

in trophic links, with major shifts between periods

March 2011 695DAILY FLOWER-VISITOR NETWORK STRUCTURE

during which the shore is either submerged or exposed.

However, while rapid pollen removal means that thedistinction between competition for pollination and

facilitation mediated by shared pollinators can beinfluenced by daily time, interactions in predator–prey

networks have longer-term dynamics.

Implications for sampling plant–pollinator networks

We propose that sampling for plant–pollinatorinteraction networks should ideally target all daily time

intervals in which flowers of each component species areopen. This approach should guarantee sampling of

pollen vectors among the general visitor assemblage,although subsequent analyses (for example, of correla-

tions between visitor activity and timing of floral rewardprovision and stigma sensitivity) are likely to be

necessary to identify key interactions for the speciesinvolved. While this sets the ideal, compromises do have

to be made in many sampling programs. For example, itwould be difficult to sample multiple distant sites over

entire field seasons using this approach. It is possiblethough to always clearly state the timing of data

collection. Moreover, sampling can be rotated throughdifferent time intervals at each field site in order toreduce bias. For example, Henson et al. (2009) rotated

sampling at 12 field sites through three daily timeintervals over a four-month field season. However, care

still needs to be taken to check that the factorsstructuring daily activity patterns (such as ambient

temperature or timing of floral resource provision) arethe same across field sites.

Concluding remarks

This study shows the impact of daily temporalstructure in flower–visitor interactions on network

structure for two communities in an African savannahabitat. Revealing the extent to which top-down and

bottom-up effects contribute to this structure requiresfurther detailed study of floral reward provision and

daily activity patterns of visitor taxa. We recommendthat future plant–pollinator network studies employexplicitly time-structured sampling to maximize the

probability that interactions central to the biology ofthe taxa involved are appropriately observed. The

benefits of this approach would be threefold: (1) itwould improve our understanding of the factors

structuring pollination networks; (2) it would allow usto better appreciate the impact of environmental change

on pollination networks, and (3) it would improve ourability to restore damaged pollination networks.

ACKNOWLEDGMENTS

We thank John Deeming, Connal Eardley, George Else,David Greathead, Josef Gusenleitner, Michael Kuhlmann,Alain Pauly, Adrian Pont, Woijiech Pulawski, RaymondWahis,and Andrew Whittington for taxonomic expertise, and theNatural History Museum, London, and the National Museumsof Kenya for access to collections. We are grateful to MpalaResearch Centre and Wanja Kinuthia at the National Museumsof Kenya for logistical support. We thank Raphael Erengai,

Patrick Lenguya, Pat Willmer, Andrew Schnabel, AdrianaOtero Arnaiz, and Anna Watson for additional support in datacollection and field identification. We are grateful to AndrewSchnabel and three anonymous reviewers for constructivecomments on the manuscript. This work was supported by aNERC quota studentship to the University of Edinburgh(K. C. R. Baldock) and by U.S. National Science Foundationgrant number DEB-0344519 (G. N. Stone). We thank theKenyan Government for permission to conduct this research(Research Clearance Permit number MOEST 13/001/33C 116).

LITERATURE CITED

Agnew, A. D. Q., and S. Agnew. 1994. Upland Kenya wildflowers. East Africa Natural History Society, Nairobi,Kenya.

Aizen, M. A., C. L. Morales, and J. M. Morales. 2008. Invasivemutualists erode native pollination webs. PLoS Biology6:396–403.

Armbruster, W. S., and A. L. Herzig. 1984. Partitioning andsharing of pollinators by four sympatric species of Dale-champia (Euphorbiaceae) in Panama. Annals of the MissouriBotanical Garden 71:1–16.

Baldock, K. C. R. 2007. Multispecies pollination interactions ina Kenyan savannah ecosystem. Thesis. University ofEdinburgh, Edinburgh, UK.

Bascompte, J., and P. Jordano. 2007. Plant–animal mutualisticnetworks: the architecture of biodiversity. Annual Review ofEcology, Evolution, and Systematics 38:567–593.

Basilio, A. M., D. Medan, J. P. Torretta, and N. J. Bartoloni.2006. A year-long plant-pollinator network. Austral Ecology31:975–983.

Bishop, J. A., and W. S. Armbruster. 1999. Thermoregulatoryabilities of Alaskan bees: effects of size, phylogeny andecology. Functional Ecology 13:711–724.

Blundell, M. 1992. Wild flowers of East Africa. HarperCollins,London, UK.

Clinebell, R. R., A. Crowe, D. P. Gregory, and P. C. Hoch.2004. Pollination ecology of Gaura and Calylophus (Ona-graceae, tribe Onogreae) in western Texas, USA. Annals ofthe Missouri Botanical Garden 91:369–400.

Cunningham, S. A. 1991. Experimental evidence for pollinationof Banksia spp. by non-flying mammals. Oecologia 87:86–90.

Devoto, M., S. Bailey, and J. Memmott. In press. The ‘nightshift’: nocturnal pollen-transport networks in a boreal pineforest. Ecological Entomology.

Dupont, Y. L., D. M. Hansen, and J. M. Olesen. 2003.Structure of a plant–flower–visitor network in the high-altitude sub-alpine desert of Tenerife, Canary Islands.Ecography 26:301–310.

Fenster, C. B., W. S. Armbruster, P. Wilson, M. Dudash, andJ. D. Thomson. 2004. Pollination syndromes and floralspecialization. Annual Review of Ecology, Evolution, andSystematics 35:375–403.

Forup, M. L., and J. Memmott. 2005. The restoration of plant–pollinator interactions in hay meadows. Restoration Ecology13:265–274.

Forup, M. L., K. S. E. Henson, P. G. Craze, and J. Memmott.2008. The restoration of ecological interactions: plant–pollinator networks on ancient and restored heathlands.Journal of Applied Ecology 45:742–752.

Gibson, R. H., I. L. Nelson, G. W. Hopkins, B. J. Hamlett, andJ. Memmott. 2006. Pollinator webs, plant communities andthe conservation of rare plants: arable weeds as a case study.Journal of Applied Ecology 43:246–257.

Henson, K. S. E., P. G. Craze, and J. Memmott. 2009. Therestoration of parasites, parasitoids and pathogens toheathland communities. Ecology 90:1840–1851.

Herrera, C. M. 1990. Daily patterns of pollinator activity,differential pollinating effectiveness, and floral resourceavailability, in a summer-flowering mediterranean shrub.Oikos 58:277–288.

KATHERINE C. R. BALDOCK ET AL.696 Ecology, Vol. 92, No. 3

Herrera, C. M. 1995. Floral biology, microclimate, andpollination by ectothermic bees in an early blooming herb.Ecology 76:218–228.

Hori, K., and T. Noda. 2001. Spatio-temporal variation ofavian foraging in the rocky intertidal food web. Journal ofAnimal Ecology 70:122–127.

Jordano, P., J. Bascompte, and J. M. Olesen. 2003. Invariantproperties in coevolutionary networks of plant–animalinteractions. Ecology Letters 6:69–81.

Kaiser-Bunbury, C. N., S. Muff, J. Memmott, and C. B. Muller.2010. The robustness of pollination networks to the loss ofspecies and interactions: a quantitative approach incorporat-ing pollinator behaviour. Ecology Letters 13:442–452.

Kearns, C. A., D. W. Inouye, and N. M. Waser. 1998.Endangered mutualisms: the conservation of plant–pollina-tor interactions. Annual Review of Ecology, Evolution, andSystematics 29:83–112.

Kronfeld-Schor, N., and T. Dayan. 2003. Partitioning of timeas an ecological resource. Annual Review of Ecology,Evolution, and Systematics 34:153–181.

Larsen, T. B. 1991. The butterflies of Kenya and their naturalhistory. Oxford University Press, Oxford, UK.

Lopezaraiza-Mikel, M. E., R. B. Hayes, M. R. Whalley, and J.Memmott. 2007. The impact of an alien plant on a nativeplant–pollinator network: an experimental approach. Ecolo-gy Letters 10:539–550.

Lundgren, R., and J. M. Olesen. 2005. The dense and highlyconnected world of Greenland plants and their pollinators.Arctic Antarctic and Alpine Research 37:514–520.

Medan, D., A. M. Basilio, M. Devoto, N. J. Bartoloni, J. P.Torretta, and T. Petanidou. 2006. Measuring generalizationand connectance in temperate, year-long active systems.Pages 245–259 in N. M. Waser and J. Ollerton, editors.Plant–pollinator interactions: from specialization to general-ization. University of Chicago Press, Chicago, Illinois, USA.

Memmott, J., and N. M. Waser. 2002. Integration of alienplants into a native flower–pollinator visitation web.Proceedings of the Royal Society B 269:2395–2399.

Memmott, J., N. M. Waser, and M. V. Price. 2004. Toleranceof pollination networks to species extinctions. Proceedings ofthe Royal Society B 271:2605–2611.

Moeller, D. A. 2004. Facilitative interactions among plants viashared pollinators. Ecology 85:3289–3301.

Morales, C. L., and M. A. Aizen. 2006. Invasive mutualismsand the structure of plant–pollinator interactions in thetemperate forests of north-west Patagonia, Argentina.Journal of Ecology 94:171–180.

Moran, M. D. 2003. Arguments for rejecting the sequentialBonferroni in ecological studies. Oikos 100:403–405.

Moody-Weis, J. M., and J. S. Heywood. 2001. Pollinationlimitation to reproductive success in the Missouri eveningprimrose Oenothera macrocarpa (Onograceae). AmericanJournal of Botany 88:1615–1622.

Munoz, A. A., and M. T. K. Arroyo. 2004. Negative impacts ofa vertebrate predator on insect pollinator visitation and seedoutput in Chuquiraga oppositifolia, a high Andean shrub.Oecologia 138:66–73.

Nakagawa, S. 2004. A farewell to Bonferroni: the problems oflow statistical power and publication bias. BehavioralEcology 15:1044–1045.

Nielsen, A., and J. Bascompte. 2007. Ecological networks,nestedness and sampling effort. Journal of Ecology 95:1134–1141.

Olesen, J. M., J. Bascompte, H. Elberling, and P. Jordano.2008. Temporal dynamics in a pollination network. Ecology89:1573–1582.

Perneger, T. V. 1998. What’s wrong with Bonferroni adjust-ments. British Medical Journal 316:1236–1237.

Petanidou, T., S. Kallimanis, J. Tzanopoulos, S. P. Sgardelis,and J. D. Pantis. 2008. Long-term observation of apollination fluctuation in species and interactions, relative

invariance of network structure and implicates for estimatesof specialization. Ecology Letters 11:564–575.

Pleasants, J. M. 1980. Competition for bumblebee pollinatorsin Rocky Mountain plant communities. Ecology 61:1446–1459.

Quinn, G. P., and M. J. Keough. 2002. Experimental designand data analysis for biologists. Cambridge University Press,Cambridge, UK.

Raine, N. E., A. S. Pierson, and G. N. Stone. 2007. Plant–pollinator interactions in a Mexican Acacia community.Arthropod–Plant Interactions 1:101–117.

Raine, N. E., P. Willmer, and G. N. Stone. 2002. Spatialstructuring and floral avoidance behavior prevent ant–pollinator conflict in a Mexican ant-acacia. Ecology83:3086–3096.

Rathcke, B. 1983. Competition and facilitation among plantsfor pollination. Pages 305–338 in L. Real, editor. Pollinationbiology. Academic Press, New York, New York, USA.

Ruiz-Guajardo, J. C. 2008. Community–plant pollinatorinteractions on a Kenyan savannah. Thesis. University ofEdinburgh, Edinburgh, UK.

Ruiz-Guajardo, J. C., A. Schnabel, R. Ennos, S. Preuss, A.Otero-Arnaiz, and G. N. Stone. 2010. Landscape genetics ofthe key African acacia species Senegalia mellifera (Vahl): theimportance of the Kenyan Rift Valley. Molecular Ecology19:5126–5139.

Sargent, R. D., and D. D. Ackerly. 2008. Plant–pollinatorinteractions and the assembly of plant communities. Trendsin Ecology and Evolution 23:123–130.

Saville, N. M. 1993. Bumblebee ecology in woodlands andarable farmland. Thesis. University of Cambridge, Cam-bridge, UK.

Stiles, F. G. 1977. Coadapted competitors: the floweringseasons of hummingbird-pollinated plants in a tropicalforest. Science 198:1177–1178.

Stone, G. N. 1994. Activity patterns of females of the solitarybee Anthophora plumipes in relation to temperature, nectarsupplies and body size. Ecological Entomology 19:177–189.

Stone, G. N. 1995. Female foraging responses to harassment inthe solitary bee Anthophora plumipes. Animal Behaviour50:405–412.

Stone, G. N., J. N. Amos, T. F. Stone, R. J. Knight, H. Gay,and F. Parrott. 1988. Thermal effects on activity patterns andbehavioural switching in a concourse of foragers onStachytarpheta mutabilis (Verbenaceae) in Papua NewGuinea. Oecologia 77:56–63.

Stone, G. N., F. Gilbert, P. G. Willmer, S. Potts, F. Semida,and S. Zalat. 1999. Windows of opportunity and thetemporal structuring of foraging activity in a desert solitarybee. Ecological Entomology 24:208–221.

Stone, G. N., P. M. J. Loder, and T. M. Blackburn. 1995.Foraging and courtship behaviour in males of the solitary beeAnthophora plumipes (Hymenoptera: Anthophoridae): ther-mal physiology and the role of body size. EcologicalEntomology 20:169–183.

Stone, G. N., P. G. Willmer, and S. Nee. 1996. Dailypartitioning of pollinators in an African Acacia community.Proceedings of the Royal Society B 263:1389–1393.

Stone, G. N., P. G. Willmer, and J. A. Rowe. 1998. Partitioningof pollinators during flowering in an African Acaciacommunity. Ecology 79:2808–2827.

Traveset, A., and D. M. Richardson. 2006. Biological invasionsas disruptors of plant reproductive mutualisms. Trends inEcology and Evolution 21:208–216.

Waser, N. M., and L. A. Real. 1979. Effective mutualismbetween sequentially flowering plant-species. Nature281:670–672.

Willmer, P. G. 1988. The role of insect water balance inpollination ecology: Xylocopa and Calotropis. Oecologia76:430–438.

March 2011 697DAILY FLOWER-VISITOR NETWORK STRUCTURE

Willmer, P. G., and G. N. Stone. 1997. Ant deterrence inAcacia flowers: how aggressive ant-guards assist seed-set.Nature 388:165–167.

Willmer, P. G., and G. N. Stone. 2004. Behavioral, ecological,and physiological determinants of the activity patterns ofbees. Advances in the Study of Behavior 34:347–466.

APPENDIX A

Flowering plant species present at the two sites during network sampling in 2004 and for equivalent months in 2005 and 2006(Ecological Archives E092-057-A1).

APPENDIX B

Comparisons of numbers of flower–visitor links per time interval between observed networks and null model results for visitorspecies with no significant differences between observed values and null expectations at both study sites in June 2004 (EcologicalArchives E092-057-A2).

APPENDIX C

Comparisons of numbers of flower–visitor links per time interval between observed networks and null model results for visitortaxa at both study sites in June 2004 (Ecological Archives E092-057-A3).

APPENDIX D

Flower–visitor links recorded for each monthly network (Ecological Archives E092-057-A4).

APPENDIX E

Network summary statistics for each monthly network (Ecological Archives E092-057-A5).

APPENDIX F

Comparisons of numbers of flower–visitor links per time interval between observed networks and null model results for allvisitor species combined for each monthly network (Ecological Archives E092-057-A6).

APPENDIX G

Comparisons of numbers of flower–visitor links per time interval between observed networks and null model results for eachvisitor taxon with significant results across all monthly networks (Ecological Archives E092-057-A7).

APPENDIX H

Examples of variation across time intervals in the links involving specific taxa (Ecological Archives E092-057-A8).

APPENDIX I

Results of null model analyses of flower–visitor data from Turkana Boma site in 2006 (Ecological Archives E092-057-A9).

KATHERINE C. R. BALDOCK ET AL.698 Ecology, Vol. 92, No. 3