Ecology, 95(2), 2014, pp. 466–474
Ó2014 by the Ecological Society of America
Land-use impacts on plant–pollinator networks: interaction
strength and specialization predict pollinator declines
CHRISTIANE NATALIE WEINER,
KARL EDUARD LINSENMAIR,
AND NICO BLU
Department of Animal Ecology and Tropical Biology, University of Wu
Ecological Networks, Biology, Technische Universita
¨t Darmstadt, Germany
Abstract. Land use is known to reduce the diversity of species and complexity of biotic
interactions. In theory, interaction networks can be used to predict the sensitivity of species
against co-extinction, but this has rarely been applied to real ecosystems facing variable land-
use impacts. We investigated plant–pollinator networks on 119 grasslands that varied
quantitatively in management regime, yielding 25 401 visits by 741 pollinator species on 166
Species-speciﬁc plant and pollinator responses to land use were signiﬁcantly predicted by
the weighted average land-use response of each species’ partners. Moreover, more specialized
pollinators were more vulnerable than generalists. Both predictions are based on the relative
interaction strengths provided by the observed interaction network. Losses in ﬂower and
pollinator diversity were linked, and mutual dependence between plants and pollinators
accelerates the observed parallel declines in response to land-use intensiﬁcation. Our ﬁndings
conﬁrm that ecological networks help to predict natural community responses to disturbance
and possible secondary extinctions.
Key words: Biodiversity Exploratories; biodiversity loss; co-extinction; interaction strength; mutualistic
networks; plant–animal interactions; pollination crisis; specialization.
The ongoing large- and small-scale changes in
anthropogenic land use are known to deplete biodiver-
sity (Duraiappah and Naeem 2005). A major goal of
biodiversity research is to understand how complex
networks of functional interactions between species
respond to disturbance and how a gradual loss of
biodiversity may affect overall ecosystem function
(Loreau et al. 2001, Koh 2004, Tylianakis et al. 2007).
These questions are of particular concern for the
pollination of ﬂowering plants since about 87.5%of the
angiosperms, among them many agricultural crops,
depend on animal pollination (Ollerton et al. 2011).
Several studies indicate that agricultural intensiﬁcation
triggers losses in the diversity of plant and pollinator
communities due to habitat conversion and fragmenta-
tion, fertilization, and pesticide use (Cunningham 2000,
Burkle and Irwin 2010, Brittain and Potts 2011).
Moreover, a high functional diversity of pollinators
may sustain a high plant diversity and lead to higher
pollination success and seed set of individual plants
(Klein et al. 2003, Hoehn et al. 2008). Among-plant
competition for limited pollinators may lead to reduc-
tion in per capita services to plants in relatively dense or
diverse populations (Vamosi et al. 2006 ). On the other
hand, visitors to dense populations are expected to be
more ﬂower constant, increasing the chance of pollen
transfer between conspeciﬁcs (Kunin 1997), and polli-
nation may be more reliable in dense plant populations
(Bernhardt et al. 2008). Outcrossing by pollinators is
important in the long term where inbreeding negatively
affects population viability and increases local extinction
risks (Byers 1995). In turn, high plant diversity is
assumed to promote pollinator richness and functional
diversity (Kwaiser and Hendrix 2007). Consequently,
experiments with manipulated plant species diversity
(Ebeling et al. 2008) and comparisons across different
¨nd et al. 2010) demonstrated positive
relationships between plant diversity and pollinator
diversity and abundance. Additionally, analyses of
historic data from Britain and the Netherlands revealed
parallel diversity declines in bees and insect-pollinated
plants (Biesmeijer et al. 2006 ).
These results lead to the hypothesis that losses in plant
and ﬂower-visitor diversity might be causally linked,
e.g., a consequence of mutual dependence. Such
dependency on certain partners implies that interaction
partners are specialized to a considerable degree. To
understand land-use effects on interacting species, it is
thus crucial to investigate their degree of specialization
and the identity of each species’ partners. This may
allow predictions of how land-use-induced changes in
species composition would affect natural communities
and their functions. Network analysis provides a useful
framework for characterizing specialization and predict-
Manuscript received 20 March 2013; revised 27 June 2013;
accepted 23 July 2013. Corresponding Editor: B. D. Inouye.
ing vulnerability of resource–consumer relationships or
mutualisms to species loss (Montoya et al. 2006 ).
Some studies suggested that specialist species are
prone to disturbance, while generalists beneﬁt from it
(McKinney 1997, Aizen et al. 2012; but see Va
Simberloff 2002, Winfree et al. 2007). While some
approaches have predicted the vulnerability of complex
communities based on simulated extinctions or dynamic
population modeling (Memmott et al. 2007, Pocock et
al. 2012), such changes have been rarely tested in real-
world systems. Methods used in modeling approaches
are controversial (Benadi et al. 2012, James et al. 2012),
and conﬂicting conclusions based on empirical data may
be partly explained by the fact that specialization
metrics differ in their sensitivity to sampling effects
¨thgen 2010). Since the number of links (observed
interaction partners) increases with the number of
observations, rarity and specialization are confounded
unless corrected by appropriate network metrics (Blu
gen et al. 2007).
In the present study, we focus on specialization and
changes in plant–pollinator interactions in grasslands
along a gradient of increasing land-use intensity. We
hypothesized that (1) land-use intensiﬁcation triggers a
decline in plant diversity and, consequently, a plant-
mediated decline in the diversity of ﬂoral resource
consumers. Moreover, we expected (2) stronger effects
of land-use intensiﬁcation on specialized plant and
pollinator species, which are more dependent on their
speciﬁc partners than generalists are. However, we
assumed that (3) pollinator-mediated declines in plant
species are less pronounced than resource-mediated
declines of pollinators, since many plant species are
not obligatory insect pollinated and are capable of
Study area and land-use intensity
The large-scale Biodiversity Exploratories represent
three bioregions in Germany located in the Schorfheide-
Chorin (Sch), Hainich-Du
¨n (Hai ), and Schwa
(Alb) (Fischer et al. 2010). Each of the three Explor-
atories covers a connected area of 422 to ;1300 km
land and each comprises 50 grassland plots. These plots
are situated within a matrix of agricultural land in use
and measure 50 350 m each. The minimum distance
between the outer edges of two plots is 200 m and each
grassland plot is at least 30 m away from the nearest
forest edge. A detailed description of all selection criteria
for experimental plots is given by Fischer et al. (2010).
The plots represent a broad gradient of land-use
intensity, ranging from near-natural, protected sites to
intensively fertilized, mown, or grazed meadows and
pastures (sheep, horses, cattle).
Qualitative categorization of land use such as
meadow/pasture or fertilized/unfertilized obscures the
variation of intensities within a category, e.g., differenc-
es in grazing intensity or fertilizer application. We
therefore used a continuous land-use-intensity index for
grasslands that incorporates the three variables fertil-
ization, mowing, and grazing intensity (Blu
¨thgen et al.
2012). For each plot k, the land-use intensity L
deﬁned as the square root of the sum of these three
variables, each of which was standardized by its regional
mean (i.e., the mean of each Exploratory):
is the fertilization level (kg nitrogenha
is the frequency of mowing per year, and G
livestock density (livestock unitsd
) on the
site. Due to the standardization by ratios, L
dimensionless. We used the mean L
of the three years
2006–2008 for all correlations; although L
only to a small degree between years, this mean value
best captures previous and ongoing management which
may both effect plants and pollinators. L
shown to predict responses in the vegetation, namely the
plants’ nitrogen indicator values, nitrogen and phos-
phorous contents in plant and soil, as well as plant
¨thgen et al. 2012).
Between May and August 2008, we investigated
plant–ﬂower-visitor networks on 119 different experi-
mental grassland plots (Alb, 39; Hai, 39; Sch, 41). Of
these, 29 plots were investigated repeatedly, up to four
times (in Alb, 15 plots were surveyed repeatedly, nine
plots two times, three plots three times, and three plots
four times; in Hainich, eight plots were surveyed
repeatedly, four plots two times, four plots three times;
in Schorfheide, six plots were surveyed repeatedly, ﬁve
plots two times, one plot three times), resulting in 162
surveys done in total (Alb, 63; Hai, 51; Sch, 48). Each
survey covered a time span of six hours between
morning and afternoon and an area of 200 33m
(length 3width) along the edge of the square
experimental grassland plot. For this transect, which
we walked three times during one survey (three rounds,
two hours each), we documented all plant–ﬂower-visitor
interactions. We recorded each insect that visited a
ﬂower as well as the ﬂower species on which it was
observed, but disregarded those insects that were sitting
on the outer petals obviously not feeding on pollen or
nectar. Specimens that we could not identify in the ﬁeld
were collected and later identiﬁed to species level with
the help of experts (see Acknowledgments).
To gain independent data on ﬂower abundance, we
ﬁrst counted the number of ﬂowering units per plant
species and transect or, in highly abundant species,
extrapolated it from a smaller area. One ‘‘ﬂowering
unit’’ was deﬁned as a unit of one (e.g., Ranunculaceae)
or more ﬂowers (e.g., Asteraceae) demanding an insect
to ﬂy in order to switch to another unit (Dicks et al.
2002). To incorporate differences in ﬂowering area, we
assessed ﬂower diversity by multiplying the number of
February 2014 467PLANT–POLLINATOR NETWORKS AND LAND-USE
ﬂowering units of a species by its average ﬂowering area
. In zygomorphic ﬂowers, ﬂowering area was
calculated as a rectangle based on ﬂower length and
width, while in actinomorphic ﬂowers ﬂowering area
was calculated as a circle based on the ﬂower diameter.
In umbels, we divided the diameter of a ﬂowering unit
by two before calculating the ﬂowering area, as
ﬂowering units are much less compact here than other
ﬂowers. This is reasonable, since ﬂower display size is
related to pollinator attraction (Grindeland et al. 2005)
and also predicted the pollen volume per ﬂower for a
subset of the investigated plants for which we have
sampled pollen (Pearson, r¼0.62, P¼0.00002; N¼40
plant species; data not shown). We obtained data on
plant species breeding systems i.e., whether a plant
species is potentially self-compatible (autogamous spe-
cies and species with mixed mating) or not (xenogamous
species) from the BiolFlor database (Klotz et al. 2002).
Forty-seven plant species are self-incompatible, 12
species show mixed mating, ﬁve are autogamous, and
two species have an apomictic breeding system.
From each survey, a single interaction network was
compiled and analyzed separately. Use of short-term
interaction networks allowed us to record a uniquely
high number of network replicates as well as to avoid
confounding effects by seasonal variation and nonover-
lapping phenology. We analyzed all ﬂower visits from
insect ﬂower visitors belonging to the orders of Diptera,
Hymenoptera, Lepidoptera, and Coleoptera. All these
visitor taxa are generally known to pollinate and are
thus termed ‘‘pollinators’’ in accordance with previous
studies, although they may not pollinate each ﬂower on
which they forage. We excluded generally non-pollinat-
ing taxa (grasshoppers, spiders), but also Nitidulidae
from analysis, as they occurred in particularly high
numbers and are easily overlooked and under-sampled
in structurally complex ﬂowers, which would bias the
While in the Hainich and Schorfheide Exploratories
we left a minimum interval of 30 days before repeatedly
surveying a plot, in the Alb Exploratory regarding 13
repeatedly sampled plots, we had conducted a total of 27
surveys within 30 days (12 plots were sampled two times,
one plot three times within 30 days). To avoid
phenologically similar replicates per plot, we calculated
mean values from these surveys per plot for each of the
variables below before correlating them to land use. This
reduced the number of independent replicates on the Alb
to 49 instead of 63. The dissimilarity of plant and
pollinator assemblages across the remaining repeated
surveys from the same plots was high. Repeated surveys
from the same plot showed the same or an even higher
level of species turnover than surveys from different
plots (Appendix: Table A1). Mantel tests (Spearman,
) based on Bray-Curtis distance and 1 310
permutations showed a strong correlation between
plant/insect species composition (based on relative
abundances) and sampling date (plants, all r
all P0.0003, n¼49 Alb, 51 Hai, and 48 Sch; insects,
0.23, all P0.0001, n¼49 Alb, 51 Hai, and 48
Sch). In contrast, the spatial arrangement of the plots
did not affect our data (plants, all r
0.04, all P
0.15; pollinators, all r
0.03, P0.32). Moreover,
land-use intensity neither correlated with sampling date
nor spatial distance in any Exploratory (all r
all P0.25). Therefore, despite pronounced temporal
variation, we consider our analyses of land-use effects
unbiased by spatial and temporal effects.
Hitherto most studies have investigated specialization
and predicted possible consequences for co-extinctions
based on qualitative metrics, i.e., the number of links of
each species (‘‘species degree’’). Moreover, pooled data
over longer temporal or spatial scales were used (e.g.,
Memmott et al. 2007). Such metrics are prone to
variation in sampling effort (Va
´zquez et al. 2009) and
disregard differences in the proportional distribution of
species. Species with few observations inevitably have
few links, hence specialization of many rare species is
severely overestimated due to several undetected links.
This bias has been demonstrated for pollinators when
other sources of information of ﬂower use were
employed (Dorado et al. 2009). Pooling data over large
areas or over long time periods also produces many
zeros due to ‘‘forbidden links’’ produced by spatial or
temporal nonoverlap, which hampers the interpretation
of specialization. Therefore, it is important to carefully
deﬁne specialization based on quantitative metrics
independent of sampling effort and species abundances
(Dormann et al. 2009) in order to compare the species’
responses to disturbance. We thus calculated comple-
mentary specialization of plants and pollinators em-
ploying the information-theoretical indices H0
¨thgen et al. 2006) for each of our short term
2speciﬁes the degree of complementary
specialization in the entire network, while d0
izes the specialization of each species ias its quantitative
non-conformity, e.g., its deviation in ﬂower visitation
from the distribution of all pollinators. Both indices
vary between 0 and 1, with high values corresponding to
more pronounced niche complementarity. While H0
iare mathematically independent of the total observa-
tion frequency per species and per network, due to the
standardization based on marginal totals, other network
metrics such as species degree, dependence, connectance,
and nestedness directly reﬂect variation in species’ total
frequencies as well as sampling effort (Blu
This bias is also evident in our dataset, where species
degree and generality strongly increased with number of
observations, whereas d0
iwas unaffected (Appendix: Fig.
2was tested against Pateﬁeld’s null model,
running 10 000 randomizations (Blu
¨thgen et al. 2006).
We used the weighted mean d0
iof each species iacross
all networks (weighted by the total interaction records of
iper plot k) as well as a weighted mean d0
CHRISTIANE NATALIE WEINER ET AL.468 Ecology, Vol. 95, No. 2
groups of ﬂower visitors, namely bees, other hymenop-
terans, beetles, butterﬂies, syrphids, and other dipterans.
To provide a weighted mean for such a group in each
plot k, each species iwas weighted by its total number of
individuals recorded in k. We segregated bees from other
hymenopterans and syrphids from other dipterans, as
both are commonly used bioindicator taxa (Biesmeijer et
Our goal was to distinguish effects of niche properties,
e.g., specialization and speciﬁc partner identity, on
species’ responses to land use from the effects of species
abundances. We thus also tested species abundances
(i.e., total number of individuals observed during ﬂower
visits, or total ﬂower area for plants) for land-use effects
For each pollinator species i, we identiﬁed their
general response to land-use intensity (r
). To quantify
the sign and magnitude of r
, we used a Spearman
correlation coefﬁcient (r
) between the relative abun-
dance of species iper plot k(percentage of total
individuals) and land-use intensity L
across all plots,
including cases where iwas absent. The same method
was applied to quantify the response of each plant
species j(replace iby jabove, see Fig. 1 and Appendix:
In addition to the degree of specialization of a species,
the identity of its partners may be important. The land-
use response of an animal may be determined by the
land-use response of its associated plant species,
weighted by the plant’s relative importance for its
partner (interaction frequency) provided in the network.
Each plant species jof Jtotal plant species can be
described by its land-use response r
. The average land-
use response of all the food plants frequented by
pollinator species i(E
), weighted by the number of
between iand j, is then
Inversely, the average land-use responses of all the
pollinator species ivisiting a plant species jis
If the partner identities and interaction strengths of
interactions in a network determine the average land-use
response of species in a community, we expect a positive
correlation between actual species responses and the
average responses of their speciﬁc partners. Hence, there
should be a positive relation r
across all Iﬂower
visitors if plants determine the responses of visitors, and
across all Jplants, if pollinators determine the
responses of plants. We tested the determinants of those
using ANCOVA (type II SS)
including the three predictors E
and pollinator group (categorical) for pollinators and E
j(both continuous), and breeding system (categorical)
for plants. Alternatively to data from our ﬂower surveys,
we used binominal vegetation survey data collected on 4
34 m quadrats per plot (see Blu
¨thgen et al. 2012) to
calculate logistic regressions (see Appendix: Fig. A1b)
and used the odds instead of Spearman to calculate r
. The alternative approach yielded the same
overall results (Fig. A2). All statistics were conducted in
R 2.15.1 (R Development Core Team 2012).
Our networks document 25 401 interactions between
166 plant species and 741 pollinator species. We
identiﬁed 115 bee species, including 25 pollen specialists
(oligolectic bees), 48 other hymenopterans, 50 butter-
ﬂies, 104 beetles, 103 syrphids, and 321 other dipteran
species. A full list of species is provided in the Appendix:
Table A2 and A3.
Plant responses.—Plant species richness (Spearman
¼0.22, P¼0.007, N¼148 networks) and
Shannon diversity (r
¼0.21, P¼0.01) declined with
increasing land-use intensity. The average land-use
response of a plant species (r
) was predicted by the
weighted response of its pollinator species (E
neither differed with plant specialization nor between
self-compatible and self-incompatible plants (Table 1a).
Yet, average specialization in plants was very high (d0
0.55 60.22 [mean 6SD]).
Moreover, plant responses to land-use intensity were
related to their relative abundance: rare plants (in terms
of their proportional coverage of ﬂoral area) showed a
stronger decline with increasing land use than more
abundant ones (r¼0.22, P¼0.005).
Pollinator responses.—Neither total pollinator species
¼0.08, P¼0.32, N¼148), abundance (r
0.001, P¼0.99), or Shannon diversity (r
0.07) was correlated to land-use intensity. Pollinator
species composition corresponded to ﬂower composition
(Mantel tests based on Bray-Curtis distance, all r
0.22, P0.0001). Moreover, responses to land-use
intensity were independent of pollinator abundance (r¼
0.035, P¼0.34, N¼741 pollinator species; Appendix:
Responses of pollinators to land use (r
depend on their association with speciﬁc ﬂowers, i.e., the
weighted mean responses of their plant species visited
; Table 1b). Pollinator specialization (d0
i) was a
signiﬁcant predictor of r
if treated as the sole variable,
but not in the mixed model, where it signiﬁcantly
interacted with E
(Table 1b). Moreover, pollinator
group identity had a signiﬁcant inﬂuence on pollinator
) to land use (Table 1). Regarding the
interaction term between d0
, land-use responses
of highly (d0
i0.6, n¼38) and intermediately
specialized pollinators (0.2 d0
i,0.6, n¼261) were
more strongly driven by the responses of their preferred
plants than in more generalized pollinators (d0
442, Fig. 2). However, plant species that support
unspecialized and intermediately specialized pollinators
February 2014 469PLANT–POLLINATOR NETWORKS AND LAND-USE
were more vulnerable to land use than plant species
supporting highly specialized pollinators: there was a
negative relationship between pollinator specialization
and the land-use response of their resources (r
P,0.0001, n¼741). For highly specialized pollinators
the trend had an opposite direction (Appendix: Fig. A4).
Regarding the interaction term between pollinator
group identity and E
, bees and other hymenopterans,
butterﬂies, beetles, and ﬂies (excluding syrphids) strong-
ly reﬂected the land-use response of the plant species
they visited in their own relative abundances. In
contrast, syrphids seemed to respond to land-use
changes independently from the responses displayed by
the plants they visited (Table 1, Appendix: Fig. A5).
With increasing land-use intensity the proportion of
syrphid species increased, whereas the proportion of
butterﬂy and hymenopteran species (excluding bees)
decreased. The proportion of bee, beetle, and dipteran
species (excluding syrphids) did not show signiﬁcant
trends across the Exploratories (Table 2), although bees
signiﬁcantly declined and dipterans signiﬁcantly in-
creased with land-use intensity in the Alb (r
P¼0.007 and r
¼0.47, P,0.001, respectively).
Plant–pollinator networks deviated signiﬁcantly from
random associations and were highly structured (mean
network specialization H0
2¼0.63 60.17[mean 6SD], N
¼148). Most networks were signiﬁcantly different from
Pateﬁeld’s null model of random interactions (P,0.001
for 130 networks and P,0.05 for additional nine
FIG. 1. Land-use response of (a) Lotus corniculatus in terms of relative ﬂower cover and (b) one of its visitors, Thymelicus
sylvestris, in terms of relative abundance. In theory, the land-use response of a pollinator may be predicted by the land-use response
of its food plants, if land use affects pollinators mainly indirectly via changes in food resources.
CHRISTIANE NATALIE WEINER ET AL.470 Ecology, Vol. 95, No. 2
TABLE 1. (a) Statistical model to predict species-speciﬁc plant responses to land use (r
) based on the weighted average
pollinator response E
), specialization (d0
j), and breeding system of each plant species and (b) model to
predict pollinator responses (r
) based on weighted average plant responses E
), specialization (d0
pollinator group identity.
Complete model Univariate model
df FPdf FP
a) Species-speciﬁc plant responses
1 19.24 0.00002 1 19.75 0.000016
j1 0.49 0.4857 1 0.49 0.48
Breeding system 1 0.26 0.6140 1 0.23 0.69
j1 1.56 0.2141
3breeding system 1 0.42 0.5178
j3breeding system 1 0.02 0.8817
Error 159 164
b) Pollinator responses
1 228.28 ,0.00001 1 265.43 ,0.000001
i1 0.00 0.9547 1 13.51 0.000255
Pollinator group 5 4.08 0.0012 5 9.90 ,0.000001
i1 6.79 0.0093
3pollinator group 5 3.89 0.0018
i3pollinator group 5 1.21 0.3030
Error 722 735
Note: Complete model and main factors in three univariate models are shown (ANCOVA; Type II SS).
FIG. 2. Interaction strengths in quantitative networks predict indirect effects of land-use intensiﬁcation. Pollinator abundances
decrease in response to declines of their most frequently visited plant species. Therefore, for the sensitivity of a species, it is not only
important how specialized it is, but also on whom it is specialized. Regression lines are shown for pollinators with low (d0
intermediate (0.2 d0
i,0.6), and high (d0
i0.6) degree of specialization.
February 2014 471PLANT–POLLINATOR NETWORKS AND LAND-USE
networks). Network specialization was not consistently
related to land-use intensity (r
¼0.11, P¼0.22). Species
idiffered between pollinator groups
¼55.50, P,0.0001). It was
strongest for bees and butterﬂies, intermediate for
beetles and hymenopterans, and lowest for syrphids
and other dipterans (Table 2).
Our results demonstrate four important land-use
effects on plant–pollinator interactions. (1) Land-use
intensiﬁcation primarily triggers losses in ﬂower diver-
sity, which could lead to nonrandom and resource-
mediated declines in certain pollinators. Overall polli-
nator diversity is not signiﬁcantly affected by land use,
but pollinator composition is. (2) Although responses of
the pollinators visiting a plant species may also inﬂuence
plant abundance, this effect is weaker. (3) Land-use
intensiﬁcation has a disproportionate impact on the
abundance of more specialized pollinators, (4) but not
on the abundance of specialized plant species.
The linkage between a pollinator’s response and the
response of the plant species it visits potentiates for
specialized pollinators, i.e., specialists on plants that
proﬁt from land use are increasing, while those on
negatively affected plant species decrease accordingly. A
strong dependence of pollinators on a narrow set of
plant species is associated with higher co-extinction risk,
since these plant species cannot be functionally replaced
by others (Praz et al. 2008). Moreover, in communities
characterized by low response diversity and low
functional redundancy, resilience after disturbance and
the ability to buffer environmental changes are reduced
(Elmqvist et al. 2003, Laliberte et al. 2013). Negative
impacts of specialization may be partly compensated by
a higher efﬁciency of specialists, e.g., specialist bees are
very effective in ﬁnding ﬂowers, pollen collection, and
digestion (Strickler 1979, Dobson and Peng 1997), but
the general extent of such compensation is unknown.
Our ﬁndings are consistent with the hypothesis that
pollinator declines are driven by the disappearance of
their important host plants, while the reciprocal effects
of pollinators on plants are weaker. In this type of
mutualism, the composition of plant communities may
be relatively robust against losses of particular polli-
nators, at least in the short term covered by our study
(Kalisz et al. 2004 ). Most grassland plants involved in
our study are self-compatible and/or have vegetative
reproduction modes (Klotz et al. 2002) and thus our
surveys may not be suitable to detect effects of reduced
genetic diversity in plant populations that may occur
with pollinator losses. While plant reproductive ﬁtness
and outcrossing may be at risk over longer time spans,
the immediate effects on the ﬁtness and/or local
distribution of pollinator communities may be more
severe when their important ﬂoral resources become
unavailable (Biesmeijer et al. 2006, Goulson et al.
2008). The asymmetry in local extinction risks may be
increased by the fact that pollinators typically provide
several times more species per network than plants and
¨thgen et al. 2007), also
mirrored in our data (ﬂowering plants, 8.4 64.4
species; visiting pollinators, 31.9 615.2 species; n¼148
networks). The mutual specialization and thus depen-
dence between pollinator and plant species may lead to
parallel regional declines in historical comparisons
(Biesmeijer et al. 2006, Fru
¨nd et al. 2010). Correspond-
ingly, the more generalized syrphids suffered less from
regional extinctions (and often even gained a higher
diversity in some regions) in recent decades than the
more specialized bee species (Biesmeijer et al. 2006,
Jauker et al. 2009). These trends are also reﬂected in
their responses to land use in our study. Land-use
intensiﬁcation not only causes a loss in plant diversity,
but also translates into pronounced changes in polli-
nator communities. The changes in pollinator compo-
sition, the dominance of ﬂies and declines in many
other taxa, correspond to a biotic homogenization
(Filippi-Codaccioni et al. 2010) on high-intensity
grasslands. Species richness and abundance of syrphids
was also positively inﬂuenced by land-use intensiﬁca-
tion in other studies (Biesmeijer et al. 2006, Ebeling et
al. 2008, Jauker et al. 2009), whereas bee diversity and
abundance declined (Biesmeijer et al. 2006, Le Fe
al. 2010). This process is easily overlooked when
focusing on total biodiversity only. Land-use intensi-
ﬁcation reduces taxonomic breadth and functional
diversity, which could conversely affect plant repro-
ductive success, species richness, and functional diver-
sity (Klein et al. 2003, Hoehn et al. 2008). In a South
African ecosystem, Pauw (2007 ) showed that, among
seven species of orchids, those that were more
specialized suffered severely from the loss of the single
Dipteran pollinators showed the lowest specialization
for plant species, whereas bees, other hymenopterans,
butterﬂies, and beetles were more specialized (see also
Weiner et al. 2011), conﬁrming that specialization
represents a risk that renders species more vulnerable
to co-extinction (McKinney 1997, Va
´zquez and Simber-
loff 2002, Winfree et al. 2007, Aizen et al. 2012, Pocock
et al. 2012). Correspondingly, investigations on butter-
ﬂies (Tudor et al. 2004 ), beetles (Kotze and O’Hara
TABLE 2. Land-use responses (changes in species richness with
increasing land-use intensity, Spearman’s r
) and ﬂower
i) of six pollinator groups.
Pollinator group r
Bees 0.04 0.64 0.39 0.22
Other hymenopterans 0.21 0.01 0.28 0.23
Butterﬂies 0.28 ,0.0005 0.33 0.24
Beetles 0.10 0.24 0.27 0.21
Syrphids 0.21 0.01 0.24 0.16
Other dipterans 0.09 0.26 0.25 0.19
CHRISTIANE NATALIE WEINER ET AL.472 Ecology, Vol. 95, No. 2
2003), and bumblebees (Kleijn and Raemakers 2008)
demonstrated that many specialized species are of
conservation concern and have undergone a consider-
able decline in the last decades.
In addition to indirect effects via ﬂower composition
and availability, land use may affect pollinators directly,
e.g., via disruption of life cycles (Johst et al. 2006 ) or
supply of appropriate nesting resources (Potts et al.
2005) or larval habitats. Many bees and beetles show
preferences for certain environmental conditions, larval
sites, or nesting sites, and their abundance depends on
certain habitats and landscape structures (Gathmann
and Tscharntke 2002). On the other hand, generalized
ﬂower visitors like most syrphids and other dipterans are
not restricted to certain landscape structures and may
proﬁt from diverse larval habitats (Jauker et al. 2009).
Over longer time spans, such direct land-use effects on
pollinators may transform into pollinator-mediated
effects on plant communities. However, in the short
term covered by our study land-use effects on plants and
plant-mediated effects on pollinators played a greater
role than vice versa.
Our ﬁndings emphasize how systems based on
mutualism may undergo severe transformation due to
land-use intensiﬁcation. Agricultural management is a
major factor driving the change of ﬂoral and faunal
richness in anthropogenic landscapes. Network analy-
ses, particularly the degree of complementary speciali-
zation and the quantitative interaction strength, may
provide important tools to predict how different species
respond to disturbance and biodiversity changes in
For identiﬁcation of insects, we thank W. Adaschkiewitz, R.
Heiß, G. Merkel-Wallner, B. Merz, V. Michelsen, S. Prescher,
H. G. Rudzinski, A. Stark, K. Szpila, M. Tospann, M. von
Tschirnhaus, and H. P. Tschorsnig (Diptera); D. Doczkal
(Apidae, Syrphidae); M. Fellendorf, M. Hermann, V. Mauss,
and H. Schwenninger (Apidae); K. Horstmann and S.
Klopfstein (Ichneumonidae); L. Hubweber and P. Sprick
(Coleoptera); M. Krauss and B. Wende (Symphyta); R. Schultz
(Formicidae); and M. Wo
¨lﬂing (Lepidoptera). D. Va
provided helpful comments. We thank the managers of the
three Exploratories, S. Renner, S. Gockel, K. Wiesner, and M.
Gorke, for their work in maintaining the plot and project
infrastructure; S. Pfeiffer and C. Fischer for support through
the central ofﬁce; M. Owonibi for managing the central
database; and M. Fischer, D. Hessenmo
¨ller, J. Nieschulze, D.
Prati, I. Scho
¨ning, F. Buscot, E.-D. Schulze, W. W. Weisser,
and the late E. Kalko for their role in setting up the Biodiversity
Exploratories project. The work has been funded by the DFG
Priority Program 1374 ‘‘Infrastructure-Biodiversity-Explorato-
ries’’ (LI 150/20-1 and BL 960/2-1). Field work permits were
issued by the responsible state environmental ofﬁces of Baden-
¨ringen, and Brandenburg (according to §72
Aizen, M. A., M. Sabatiano, and J. M. Tylianakis. 2012.
Specialization and rarity predict nonrandom loss of interac-
tions from mutualist networks. Science 335:1486–1489.
Benadi, G., N. Blu
¨thgen, T. Hovestadt, and H. J. Poethke.
2012. Population dynamics of interacting plant and pollina-
tor communities: stability reconsidered. American Naturalist
Bernhardt, C. E., R. J. Mitchell, and H. J. Michaels. 2008.
Effects of population size and density on pollinator
visitation, pollinator behavior, and pollen tube abundance
in Lupinus perennis. International Journal of Plant Science
Biesmeijer, J. C., et al. 2006. Parallel declines in pollinators and
insect pollinated plants in Britain and the Netherlands.
¨thgen, N. 2010. Why network analysis is often disconnected
from community ecology: a critique and an ecologist’s guide.
Basic and Applied Ecology 11:185–195.
¨thgen, N., et al. 2012. A quantitative index of land-use
intensity in grasslands: Integrating mowing, grazing and
fertilization. Basic and Applied Ecology 13:207–220.
¨thgen, N., F. Menzel, and N. Blu
¨thgen. 2006. Measuring
specialization in species interaction networks. BMC Ecology
¨thgen, N., F. Menzel, T. Hovestadt, B. Fiala, and N.
¨thgen. 2007. Specialization, constraints, and conﬂicting
interests in mutualistic networks. Current Biology 17:341–
Brittain, C., and S. G. Potts. 2011. The potential impacts of
insecticides on the life-history traits of bees and the
consequences for pollination. Basic and Applied Ecology
Burkle, L. A., and R. E. Irwin. 2010. Beyond biomass:
measuring the effects of community level nitrogen enrich-
ment on ﬂoral traits, pollinator visitation and plant
reproduction. Journal of Ecology 98:705–717.
Byers, D. L. 1995. Pollen quantity and quality as explanations
for low seed set in small populations exempliﬁed by
Eupatorium (Asteraceae). American Journal of Botany 82:
Cunningham, S. A. 2000. Depressed pollination in habitat
fragments causes low fruit set. Proceedings of the Royal
Society B 267:1149–1152.
Dicks, L. V., S. A. Corbet, and R. F. Pywell. 2002.
Compartmentalization in plant–insect ﬂower visitor webs.
Journal of Animal Ecology 71:32–43.
Dobson, H. E. M., and Y.-S. Peng. 1997. Digestion of pollen
components by larvae of the ﬂower-specialist bee Chelostoma
ﬂorisomne (Hymenoptera: Megachilidae). Journal of Insect
Dorado, J., D. P. Va
´zquez, E. Stevani, and N. Chacoff. 2009.
Rareness and specialization in plant–pollinator networks.
Dormann, C., J. Fru
¨nd, N. Blu
¨thgen, and B. Gruber. 2009.
Indices, graphs and null models: analysing bipartide ecolog-
ical networks. Open Ecology Journal 2:7–24.
Duraiappah, A., and S. Naeem. 2005. Ecosystems and human
well-being: biodiversity synthesis. World Resources Institute,
Washington, D.C., USA.
Ebeling, A., A.-M. Klein, J. Schumacher, W. W. Weisser, and
T. Tscharntke. 2008. How does plant richness affect
pollinator richness and temporal stability of ﬂower visits?
Elmqvist, T., C. Folke, M. Nystro
¨m, G. Peterson, J. Bengtsson,
B. Walker, and J. Norberg. 2003. Response diversity,
ecosystem change, and resilience. Frontiers in Ecology and
the Environment 1:488–494.
Filippi-Codaccioni, O., V. Devictor, Y. Bas, and R. Julliard.
2010. Toward more concern for specialisation and less species
diversity in conserving farmland biodiversity. Biological
Fischer, M., et al. 2010. Implementing large-scale and long-
term functional biodiversity research: the Biodiversity
Exploratories. Basic and Applied Ecology 11:473–485.
February 2014 473PLANT–POLLINATOR NETWORKS AND LAND-USE
¨nd, J., K. E. Linsenmair, and N. Blu
¨thgen. 2010. Pollinator
diversity and specialization in relation to ﬂower diversity.
Gathmann, A., and T. Tscharntke. 2002. Foraging ranges of
solitary bees. Journal of Animal Ecology 71:757–764.
Goulson, D., G. C. Lye, and B. Darvill. 2008. The decline and
conservation of bumblebees. Annual Review of Entomology
Grindeland, J. M., N. Sletvold, and R. A. Ims. 2005. Effects of
ﬂoral display size and plant density on pollinator visitation
rate in a natural population of Digitalis purpurea. Functional
Hoehn, P., T. Tscharntke, J. M. Tylianakis, and I. Steffan-
Dewenter. 2008. Functional group diversity of bee pollina-
tors increases crop yield. Proceedings of the Royal Society B
James, A., J. W. Pitchford, and M. J. Plank. 2012. Distangeling
nestedness from models of ecological complexity. Nature
Jauker, F., T. Dieko
¨tter, F. Schwarzbach, and V. Wolters.
2009. Pollinator dispersal in an agricultural matrix: opposing
responses of wild bees and hoverﬂies to landscape structure
and distance from main habitat. Landscape Ecology 24:547–
Johst, K., M. Drechsler, J. Thomas, and J. Settele. 2006.
Inﬂuence of mowing on the persistence of two endangered
large blue butterﬂy species. Journal of Applied Ecology 43:
Kalisz, S., D. W. Vogler, and K. Hanley. 2004. Context-
dependent autonomous self-fertilization yields reproductive
assurance and mixed mating. Nature 430:884 –887.
Kleijn, D., and I. Raemakers. 2008. A retrospective analysis of
pollen host plant use by stable and declining bumble bee
species. Ecology 89:1811–1823.
Klein, A.-M., I. Steffan-Dewenter, and T. Tscharntke. 2003.
Fruit set of highland coffee increases with the diversity of
pollinating bees. Proceedings of the Royal Society B 270:955–
Klotz, S., I. Ku
¨hn, and W. Durka. 2002. Bioﬂor: eine
Datenbank mit biologisch-o
¨kologischen Merkmalen zur
Flora von Deutschland. Bundesamt fu
¨r Natuschutz, Bonn,
Koh, L. P. 2004. Species coextinction and the biodiversity crisis.
Kotze, D. J., and R. B. O’Hara. 2003. Species decline—but
why? Explanations of carabid beetle (Coleoptera, Carabidae)
declines in Europe. Oecologia 135:138–148.
Kunin, W. E. 1997. Population size and density effects in
pollination: pollinator foraging an plant reproductive success
in experimental arrays of Brassica kaber. Journal of Ecology
Kwaiser, K. S., and S. D. Hendrix. 2007. Diversity and
abundance of bees (Hymenoptera: Apiformes) in native and
ruderal grasslands of agriculturally dominated landscapes.
Agriculture, Ecosystems and Environment 124:200–204.
Laliberte, E., et al. 2013. Land-use intensiﬁcation reduces
functional redundancy and response diversity in plant
communities. Ecology Letters 13:76–86.
´on, V., A. Schermann-Lgionnett, Y. Delettre, S. Aviron,
R. Billeter, R. Bugter, F. Hendrickx, and F. Burel. 2010.
Intensiﬁcation of agriculture, landscape composition and
wild bee communities: a large scale study from European
countries. Agriculture, Ecosystems and Environment 137:
Loreau, M., et al. 2001. Biodiversity and ecosystem functioning:
current knowledge and future challenges. Science 294:804–
McKinney, M. L. 1997. Extinction vulnerability and selectivity:
combining ecological and paleontological views. Annual
Review of Ecology and Systematics 28:495–516.
Memmott, J., P. G. Craze, N. M. Waser, and M. V. Price. 2007.
Global warming and the disruption of plant–pollinator
interactions. Ecology Letters 10:710–717.
Montoya, J. M., S. L. Pimm, and R. V. Sole
´. 2006. Ecological
networks and their fragility. Nature 442:259–264.
Ollerton, J., R. Winfree, and S. Tarrant. 2011. How many
ﬂowering plants are pollinated by animals? Oikos 120:321–
Pauw, A. 2007. Collapse of a pollination web in small
conservation areas. Ecology 88:1759–1769.
Pocock, M. J. O., D. M. Evans, and J. Memmott. 2012. The
robustness and restoration of a network of ecological
networks. Science 335:973–977.
Potts, S. G., B. Vulliamy, S. Roberts, C. O’Toole, A. Dafni, G.
Ne’eman, and P. Willmer. 2005. Role of nesting resources in
organising diverse bee communities in a Mediterranean
landscape. Ecological Entomology 30:78–85.
Praz, C. J., A. Mu
¨ller, and S. Dorn. 2008. Specialized bees fail
to develop on non-host pollen: Do plants chemically protect
their pollen? Ecology 89:795–804.
R Development Core Team. 2012. R 2.15.1. R Project for
Statistical Computing, Vienna, Austria. www.r-project.org
Strickler, K. 1979. Specialization and foraging efﬁciency of
solitary bees Hoplitis anthocopoides. Ecology 60:998–1009.
Tudor, O., R. L. H. Dennis, J. N. Greatorex-Davies, and T. H.
Sparks. 2004. Flower preferences of woodland butterﬂies in
the UK: nectaring specialists are species of conservation
concern. Biological Conservation 119:397–403.
Tylianakis, J. M., T. Tscharntke, and O. T. Lewis. 2007.
Habitat modiﬁcation alters the structure of tropical host–
parasitoid food webs. Nature 445:202–205.
Vamosi, J., T. M. Knight, J. A. Steets, S. J. Mazer, M. Burd,
and T. L. Ashmann. 2006. Pollination decays in biodiversity
hotspots. Proceedings of the National Academy of Sciences
´zquez, D. P., N. Chacoff, and L. Cagnolo. 2009. Evaluating
multiple determinants of the structure of mutualistic net-
works: comment. Ecology 90:2039–2046.
´zquez, D. P., and D. Simberloff. 2002. Ecological speciali-
zation and susceptibility to disturbance: conjectures and
refutations. American Naturalist 159:606– 623.
Weiner, C., M. Werner, K. E. Linsenmair, and N. Blu
2011. Land-use intensity in grasslands: changes in biodiver-
sity, species composition and specialization in ﬂower-visitor
networks. Basic and Applied Ecology 12:292–299.
Winfree, R., T. Griswold, and C. Kremen. 2007. Effect of
human disturbance on bee communities in a forested
ecosystem. Conservation Biology 21:213–223.
Specialization metrics, land use responses, and species lists (Ecological Archives E095-041-A1).
CHRISTIANE NATALIE WEINER ET AL.474 Ecology, Vol. 95, No. 2