Content uploaded by Anna Traveset
Author content
All content in this area was uploaded by Anna Traveset on Sep 29, 2017
Content may be subject to copyright.
EV-1
Effects of habitat loss on the plant – fl ower visitor network structure
of a dune community
Anna Traveset , Roc í o Castro-Urgal , Xavier Rotll à n-Puig and Amparo L á zaro
A. Traveset (http://orcid.org/0000-0002-1816-1334) (atraveset@uib.es), R. Castro-Urgal, X. Rotll à n-Puig (http://orcid.org/0000-0003-2046-
0621) and A. L á zaro, Mediterranean Inst. for Advanced Studies - Biodiversity and Conservation, c/Miquel Marqu é s 21, ES-07190 Esporles,
Mallorca, Balearic Islands, Spain.
Pollination is a valuable ecosystem service, and plant – pollinator interactions in particular are known to play a crucial role
in conservation and ecosystem functioning. ese mutualisms, like other ecological interactions, are currently threatened
by diff erent drivers of global change, mainly habitat loss, fragmentation, or modifi cation of its quality. Most studies so
far have focused on the impact of such disturbances on particular species interactions and we thus need more empirical
evidence on the responses at a community-level. Here we evaluated how habitat loss infl uenced the pattern of interactions
between plants and their fl ower visitors in a coastal dune marshland community. Using data from four years (2008 – 2011),
we assessed the eff ect of a large disturbance in the area (occurring in 2010) that represented the loss of more than 50% of
the vegetation cover. We found a considerable decrease in species richness and abundance of fl ower visitors, which resulted
in a lower number of interactions after the disturbance. Not all functional groups, however, responded similarly. Contrary
to the expected from previous fi ndings, bees and wasps were less negatively infl uenced than beetles, fl ies and ants, possibly
due to their higher movement capacity. Species interactions in the community were more specialized after habitat loss,
resulting in a lower level of network nestedness and a higher modularity. At a species level, the number of fl ower visitors
per plant decreased after the disturbance, and plants were visited by less abundant fl ower visitors. Our fi ndings lead us to
predict that the overall plant – fl ower visitor network became less robust and resilient to future perturbations. However, the
fact that each functional group responds distinctly to disturbances makes it more diffi cult to foresee the fi nal consequences
on community composition and ecosystem functioning.
e interactions between plants and their pollinators play a
crucial role in biodiversity, conservation and ecosystem func-
tioning. Habitat loss, fragmentation and changes in habi-
tat quality, and in landscape structure in general, represent
major threats to such interactions and thus to both plant
and pollinator species persistence in the communities. Stud-
ies examining such threats at a community level, however,
are still few and we thus have rather little empirical evidence
on the fi nal consequences of such habitat and landscape
changes for the functioning of this important ecosystem
service (Klein et al. 2007, Hagen et al. 2012, Ferreira et al.
2013, Nielsen and Totland 2014). Previous work has shown
that a reduction in habitat quality and landscape heterogene-
ity cause species losses and leads to changes in the pattern of
interactions among species, i.e. in the interaction network
structure (Tylianakis et al. 2007, Gonz á lez et al. 2011). By
reducing pollinator availability and diversity due to decreased
fl oral resource supplies as well as nesting sites, habitat
modifi cations can infl uence the levels of cross-pollination
and, ultimately, fruit and seed production (Aguilar et al.
2006, Winfree et al. 2011, Hagen et al. 2012, Viana et al.
2012, Ferreira et al. 2013, Vanbergen et al. 2014). Likewise,
variation in conspecifi c plant densities may aff ect plant
reproductive success by changing the pollinator-mediated
connectivity between individuals in a plant population at
diff erent spatial scales (Hegland et al. 2014, Vanbergen et al.
2014). is indicates that, by altering interspecifi c interac-
tions at a plant community-level, habitat disturbance can cas-
cade down aff ecting the patterns of gene fl ow across levels of
biological organization and potentially driving evolutionary
changes (Eckert et al. 2010, Ferreira et al. 2013).
Rare and specialized interactions have shown to be
the fi rst to disappear after habitat reduction, and thus an
increase in the frequency of generalist plants and/or pollina-
tor species is usually observed (Ashworth et al. 2004, Aizen
et al. 2012, Vanbergen et al. 2014). A decrease in network
nestedness in disturbed habitats has been reported in several
systems (Vanbergen et al. 2014, Moreira et al. 2015, Revilla
et al. 2015), which has led authors to predict reductions
in the number of coexisting species (Bastolla et al. 2009),
and in the robustness and resilience of plant – pollinator net-
works to further perturbations (Bascompte 2009, Fortuna
et al. 2013). e loss of species and their interactions in a
disturbed network can also lead to the formation of isolated
compartments within the network (Spiesman and Inouye
2013) which run a higher risk of disappearing after future
disturbances than if species are connected in a cohesive
network. Diff erent models have shown that the distribution
© 2017 e Authors. Oikos © 2017 Nordic Society Oikos
Subject Editor: Paulo Guimar ã es Jr. Editor-in-Chief: Dries Bonte. Accepted 29 May 2017
Oikos 000: 001–010, 2017
doi: 10.1111/oik.04154
EV-2
of number of interactions becomes more skewed when mov-
ing from pristine to disturbed systems, and that mutualis-
tic networks might collapse at critical habitat destruction
thresholds (Keitt 2009, Kaiser-Bunbury et al. 2010, Viana
et al. 2012, Fortuna et al. 2013).
Habitat degradation may also cause homogenization of
the plant – pollinator networks by promoting higher link-
diversity but lower link-turnover in disturbed sites compared
to undisturbed ones (Nielsen and Totland 2014). Moreover,
the particular species ’ network functional role can change
notably along a disturbance gradient. us, a plant species
can act as a hub (being at the network core) in one site but as
a specialist (being at the network periphery) in another site
(Campos-Navarrete et al. 2013, Nielsen and Totland 2014).
Module and network hubs (i.e. species highly connected
within their modules and with other species in other mod-
ules, respectively), together with connectors (species that
link diff erent modules), are considered keystone species for
sustaining network structure and thus their removal due to a
disturbance would have the strongest eff ects and might even
collapse the network (Olesen et al. 2007, Kaiser-Bunbury
et al. 2010, Fortuna et al. 2013).
Changes in pollinators ’ diversity are frequently reported
mostly due to increased isolation of habitat patches and
reduced landscape complexity caused by environmental
simplifi cation (Ferreira et al. 2013). However, not all pol-
linator species respond similarly to habitat changes. Social
bees, for instance, are known to be sensitive to changes in the
distribution of nesting and foraging habitats in the landscape
(Williams et al. 2010, Carvell et al. 2012, Kennedy et al.
2013, Garibaldi et al. 2014); thus, land cover changes
can directly aff ect individual survival probability, locally
reducing species abundance (Ferreira et al. 2015). Solitary
bees, however, may be more aff ected by habitat destruction
as they are more specialized in food resources or nesting
sites than social bees (Williams et al. 2010, Ferreira et al.
2015). By contrast, non-social insects with free-living prog-
eny (e.g. dipterans, coleopterans) may be less aff ected by
distance between resource patches, as they do not need to
return to their brood cells repeatedly after foraging (Jauker
et al. 2009, Parsche et al. 2011). Moreover, fl ower visitor
abundance and species richness have been shown to increase
with fl oral abundance (Hegland and Boeke 2006, Hagen
and Kraemer 2010) and plant diversity (Potts et al. 2003,
Ghazoul 2006, Bl ü thgen et al. 2007, Ebeling et al. 2008).
In general, there is still scarce information on how diff er-
ent pollinator functional groups can respond to habitat dis-
turbance (Burkle et al. 2013, Aguirre-Guti é rrez et al. 2015,
L á zaro et al. 2016) and how they change their interaction
patterns with plants in the community (e.g. their rewiring
capacity within the network).
In this study, we aimed at evaluating the impact of habitat
disturbance (habitat loss, in particular) on the patterns of
plant – fl ower visitor interactions in a coastal dune marshland
community at the north of Mallorca (Balearic Islands, western
Mediterranean Sea). e plant – fl ower visitor network of this
community was monitored for four consecutive years, from
2008 to 2011. After the fl owering season of 2010, the study
area was greatly disturbed due to the construction of a golf
course that caused the loss of ca 50% of the vegetation cover,
leaving the bare soil (Fig. 1). is provided an opportunity
to assess the extent to which substantial habitat loss altered
the interactions between plants and their fl ower visitors.
Our specifi c questions were the following: 1) did network
structural properties change after habitat disturbance more
than the expected from temporal changes in the previous
years? 2) To what extent were fl oral resources and species
richness and abundance of fl ower visitors aff ected by the dis-
turbance? 3) At the species level, how consistent across years
were degree (linkage level), contribution to nestedness, level
of selectiveness, strength, and weighted closeness centrality,
and did these parameters change notably after the distur-
bance? 4) Which fl ower visitors ’ functional groups and which
plant species (regarding traits such as fl ower abundance and
fl oral symmetry) experienced the greatest changes in species
level parameters after the disturbance? 5) If networks had a
modular structure, how consistent in time were species ’ roles
regarding modularity, and did they change more after the
disturbance relative to previous years?
Methods
Study site
e study was carried out in Son Bosc (39 ° 46 ′ 28.11 ″ N ,
3 ° 07 ′ 45.34 ″ E), a diverse dune marshland in northern
Mallorca, adjacent to S ’ Albufera Natural Park. e
predominant vegetation consists of Daucus carota (Apiaceae),
Helichrysum stoecha s (Asteraceae), Lotus corniculatus
Figure 1. Aerial photograph showing the study area before (A) and
after (B) the disturbance (soil removal for a golf course construction).
e green line marks the study area, the red line the disturbed area
and the vertical black and white pattern the study area damaged.
EV-3
(Fabaceae), Lotus cytisoides (Fabaceae), Scabiosa atropurpurea
(Dipsacaceae) and Teucrium dunense (Lamiaceae) and over
80 fl owering species have been recorded in the area, mostly
annual plants although also some shrubs like Cistus salviifolius
(Cistaceae) and Myoporum tenuifolium (Myoporaceae). Such
a high diversity of fl owers allows maintaining an also high
diversity of fl ower-visitors (ca 125 spp.), a good fraction of
which nest in the sandy soils of this area. Specifi cally, this
area bears the highest bee species richness of Mallorca Island
(D. Baldock pers. comm.).
We performed censuses in Son Bosc during four consecu-
tive years, from 2008 to 2011. During the summer of 2010,
an area of ca 2.3 ha was disturbed due to the construction
of a golf course, representing 52.3% of the total area covered
in our study (ca 4.5 ha; delimited in green in Fig. 1). Given
that the largest disturbance occurred when most plants had
already fl owered in 2010, we expected the highest impact on
the plant – fl ower visitor network the following year. us,
during 2011, we kept censusing all fl owering plants in the
remaining unaltered area including as well other plant
species that were still present in the surroundings of the
disturbed area (bare soil).
Sampling methods
All plants in bloom were monitored throughout the fl owering
season, from early April to the end of July. Once or twice
per week, we made insect censuses on fl owers from
haphazardly selected individuals from all fl owering plant
species. Censuses were done from 10:00 a.m. to 17:00 p.m.
on sunny and non-windy days. Insect visits to fl owers were
recorded from a distance of approximately 1 m to minimize
interference with insect behavior. We recorded contacts of
insect visitors to fl owers during 3 – 5 min periods. Due to
the small fl ower size of most species, insects nearly always
touched the reproductive parts of the fl ower, although we
did not record this or their behaviour. Hence, we use the
term fl ower - visitation networks, regardless of the effi ciency
of each insect visitor in the pollination process. We must
note, however, that considering such effi ciency and distin-
guishing between true pollinators from those that act as
cheaters might lead to a diff erent network structure, e.g. the
network might be more specialized (Alarc ó n 2010, Genini
et al. 2010). During each census we recorded: 1) identity of
fl owering plant species; 2) number of open fl owers of each
individual plant observed; 3) identity of each fl ower visitor;
4) number of individuals of each species visiting fl owers; and
5) number of fl owers visited by each fl ower visitor. When
fl ower visitors could not be identifi ed in the fi eld, these were
collected (usually after fi nishing the census) for identifi ca-
tion by taxonomists. We categorized fl ower visitor species
into the following functional groups (as done in previous
studies; Fenster et al. 2004): ants, bees, beetles, hoverfl ies,
fl ies (mainly muscoid fl ies), butterfl ies, wasps and others
(mostly hemiptera).
Time spent censusing fl ower visitors along the entire
season was on average 36.3 h. Most intensive sampling
was from 2009 to 2011, when we also estimated fl ower
abundance fortnightly at each site. In each fl ower census,
we recorded the number of all open fl owers of each fl ower-
ing plant encountered within permanent belt transects; we
surveyed 13 transects (50 2 m) in 2009 and 10 transects
in 2010 and 2011, covering a total area of 1300 m
2 and
1000 m
2 , respectively. Further details on sampling can be
found in Castro-Urgal et al. (2012).
Network parameters
We built four quantitative interactions matrices (one for
each year) using the number of visits per unit time as link
weight. For each network, we calculated the most widely
used quantitative descriptors of the structure of weighted
ecological interaction networks (Tylianakis et al. 2010).
At network level, these were: connectance (C), weighted
nestedness (WNODF), complementary specialization H 2 ′ ,
interaction evenness (IE) and quantitative modularity (Q).
At species level, we obtained the following metrics both for
each plant and fl ower visitor species in the networks: degree,
strength, species selectiveness (d ′ ) (termed index of special-
ization in other studies; Bl ü thgen et al. 2006), weighted
closeness centrality (wCC), contribution to nestedness ni (see
Supplementary material Appendix 1 for defi nitions of each
parameter), standardized connection ‘ c ’ and participation
values ‘ z ’ . We used the bipartite package ver. 1.18 (Dormann
et al. 2009) run in R to obtain all these network metrics,
except WNODF and contribution to nestedness which were
obtained with the software NODF ver. 2.0 ( < www.keib.
umk.pl/nodf/ > )(nestedness based on overlap and decreas-
ing fi ll; Almeida-Neto and Ulrich 2011). e signifi cance
of WNODF values was assessed against 100 randomizations
using the ‘ rc ’ and ‘ p ’ null models; the ‘ rc ’ model resamples
with row/column weights fi xed, while the ‘ p ’ model ran-
domizes proportional to the respective marginal distribution
(Almeida-Neto and Ulrich 2011).
Quantitative modularity ( Q ) was estimated using the
QuanBIMo algorithm (Dormann and Strauss 2014), which
is implemented in R . It consists of a recurrent Markov chain
Monte Carlo (MCMC) algorithm to fi nd the best division
of nodes (species) into modules. A total of 10
6 MCMC
steps were used with a tolerance level of 10
10 . As Q values
can vary among diff erent runs, we repeated the calculations
100 times for each network – using the computeModules
function – and selected the iteration with maximum likeli-
hood as the best estimation of Q. To account for Q ’ s depen-
dence on network size and test the signifi cance of modularity
values, we calculated a z-score for each network by running
the same algorithm in 100 random networks with identi-
cal marginal totals as the empirical network (using the null
model ‘ r2d ’ ; Guimer à and Amaral 2005) and comparing the
modularity values between random and empirical networks.
Such tests were done in the bipartite package (Dormann and
Strauss 2014).
Following Guimer à and Amaral (2005), we identifi ed
species with important roles in the network by computing
standardized connection and participation values ( c and z ,
respectively). While c refers to the even distribution of links
across modules, z refers to within-module degrees.
Statistical analysis
All analyses were conducted in R ver. 3.1.2 ( < www.r-project.
org > ). To compare species richness of fl ower visitors among
EV-4
habitat disturbance (2011) than the three previous years
(Table 1). e number of plant species was also reduced
after the disturbance but to a lower extent than the number
of fl ower visitors (Table 1). All networks showed a highly
consistent connectance around 5%. e networks were sig-
nifi cantly nested (p 0.001), meaning that the partners of
the most specialized species are a subset of those that interact
with the most generalist species (Table 1). e lower nested-
ness in 2011 might thus result from the loss of some generalist
species after the disturbance and/or from an increase in
specialized interactions; the latter is actually supported by
the higher H 2 ′ value in 2011 (Table 1). Interaction evenness
was moderate (ca 0.50) across the four years of the study
(Table 1). Finally, the networks were signifi cantly modu-
lar each year; however, while both modularity (Q) and the
number of modules were higher in 2011 than the other
years, this was not the case for z-scores (Table 1).
Overall, species richness of fl ower visitors was lower in
2011 (mean SE: 10.75 3.65) compared to the previous
years (14.71 2.23) ( χ 2 8.25, df 3, p 0.046). How-
ever, this diff erence was mostly due to a decrease in beetle
and fl ie species richness (Table 2). e other groups barely
changed across the four years of the study. Regarding fl ower
visitor abundance, the best model showed that it varied sig-
nifi cantly among all the study years, consistently among
functional groups (year: χ 2 72.45, df 3, p 0.0001).
It was lowest in 2011 (visits min
– 1 : 0.09 0.02), highest
in 2010 and 2009 (1.10 0.11 and 1.12 0.004,
respectively), and intermediate in 2008 (0.18 0.02).
years, we performed a generalized linear model (GLM) using a
Poisson distribution and log as link function. In this analysis,
year was included as a fi xed categorical factor whereas the
number of species in each fl ower visitor functional group as
sampling units. e interannual variations in species level
network parameters and abundances were analysed by means
of generalized linear mixed models (GLMM, package lme4 )
that included species as a random factor to avoid pseudorep-
lication. We used separate models for plants and fl ower visi-
tors, and for each network parameter. e models for plants
only included year as fi xed categorical predictor variable,
whereas those for fl ower visitors also included functional
group and its interaction with year. If the interaction was
non-signifi cant, we run the models with the fi xed variables
separately and chose the best model based on AIC. All func-
tional groups were included in the models, except for the
‘ others ’ group owing to its low species number. Due to the
nature of the data, we used: 1) Poisson distribution and log
link functions for the degree analyses, after checking for the
absence of overdispersed data (Zuur et al. 2009); 2) Gaussian
distribution and log link function for the models of selec-
tiveness; and 3) gamma distribution and log link function
for the rest of the variables. e consistency among years in
species roles within the network was also analysed by means
of GLMMs, including c and z as response variables, year as
fi xed factor, and species as random factor. Plant species were
the sampling units, and data were adjusted to a gamma dis-
tribution in each model.
As we found signifi cant diff erences in both plant degree
and selectiveness between 2011 and the average of the three
previous years, we further assessed whether such diff erences
were associated to changes in the prevalence of plant spe-
cies with diff erent fl ower symmetry and diff erent fl ower
abundances. For this, we performed two separate GLMs
to analyse the after-disturbance change in degree and selec-
tiveness (calculated as the diff erence between the degree/
selectiveness in 2011 and the average degree/selectiveness
in the previous years) as response variables, and fl ower
symmetry (zygomorphic versus actinomorphic) and fl ower
abundance (average from 2009 – 2011) as independent pre-
dictor variables. In both models, sampling units were the
study plant species, and a Gaussian distribution was used
given that the response variables fullfi lled the assumptions
of normalilty.
Post hoc analyses to test for diff erences among levels of
a signifi cant factor were conducted using Tukey a posteriori
tests (package multicomp in R).
Data deposition
Data available from the Fileshare Repository: < https://
csicannatraveset.sharefile.com/d-sf34b761ef9a4607a > .
(Traveset et al. 2017).
Results
Overall community structure
e number of fl ower visitor species, number of links and
weighted nestedness were much lower the year following
Table 1. Network parameters for each study year. WNODF: weighted
nestedness, H
2 ’ : index of specialization (selectiveness), IE: interac-
tion evenness. For modularity, the z-score is given, as the Q observed
is compared to that expected with a null model based on marginal
totals (representing abundance distributions of plants and fl ower
visitors; see Dormann and Strauss 2014 for further details).
2008 2009 2010 2011
No. plants 56 68 67 52
No. fl ower visitors 120 110 123 86
No. links 347 390 494 248
Connectance 0.052 0.052 0.060 0.055
WNODF 7.698 8.67 9.412 5.456
H
2 ’ 0.589 0.618 0.547 0.685
IE 0.532 0.529 0.514 0.497
Modularity (Q) 0.307 0.370 0.368 0.568
Number of modules 8 8 6 13
Modularity z-score 316.60 760.34 722.79 303.58
Table 2. Species richness in each fl ower visitor group across the four
study years. The groups in which species richness was considerably
reduced after habitat disturbance are marked with an asterisk.
Flower visitor group 2008 2009 2010 2011
Ants 6 3 4 2
Bees 29 35 38 31
Beetles 25 24 29 18
*
Butterfl ies 7 5 3 3
Flies 25 18 23 11
*
Hoverfl ies 5 6 6 5
Wasps 13 13 14 15
Others 11 6 6 1
EV-5
observed in weighted closeness centrality and contribution
to nestedness, but these did not seem to be related to the
disturbance, as both parameters were signifi cantly diff erent
between 2008 and the other years (Fig. 2C – D), whereas the
values in 2011 did not diff er signifi cantly from those in other
years (Fig. 2C – D).
Flower abundance positively infl uenced the after-
disturbance change in plant species degree ( χ 2 4.57,
df 1, p 0.03) but not in selectiveness ( χ 2 0.60, df 1 ,
Species-level network properties
Plant species
At the species level, network parameters diff ered signifi cantly
among years, except plants ’ strength (Table 3). In 2011, plants ’
degree was signifi cantly lower (Fig. 2A) and selectiveness was
signifi cantly higher (Fig. 2B) than the previous years; that
is, the number of fl ower visitors per plant decreased after
the disturbance, and plants were visited by less abundant
fl ower visitors. Signifi cant diff erences among years were also
Table 3. Results of the generalized linear models comparing species level network parameters among the study years for plants and fl ower
visitors. When the interaction between year and functional group was signifi cant, the LRT for the interaction is given, but both factors were
also included in the models.
Species-level network index
Plants
Year
Flower visitors
Year Flower visitor group
Degree
χ
3
2 55.14, p 0.0001 χ
18
2 36.34, p 0.006
Strength χ
3
2 5.14, p 0.162
χ 18
2 36.40, p 0.006
Selectiveness (d ’ )
χ 3
2 27.59, p 0.0001 χ
18
2 64.09, p 0.0001
Weighted closeness centrality χ
3
2 139.78, p 0.0001 χ
18
2 51.78, p 0.0001
Contribution to nestedness χ
3
2 15.4, p 0.001
χ 18
2 51.79, p 0.0001
Standardized connection ( c )
χ 3
2 10.55, p 0.014
χ
18
2 34.99, p 0.009
Participation values ( z ) χ
3
2 4.07, p 0.25
χ 18
2 20.37, p 0.31
Figure 2. Mean ( SE) values of (A) degree; (B) selectiveness (d ′ ); (C) weighted closeness centrality; and (D) contribution to nestedness for
plants across the study years. Diff erent letters indicate signifi cant diff erences among years. Note that the higher the negative value, the more
the species contributes to nestedness (following defi nition by Almeida-Neto and Ulrich 2011).
EV-6
on the after-disturbance change in either degree ( χ 2 0.82,
df 1, p 0.37) or selectiveness ( χ 2 0.81, df 1 ,
p 0.37), i.e. both actinomorphic and zygomorphic species
responded similarly to the disturbance regarding these two
metrics.
p 0.44). In other words, those species producing more
fl owers were those most aff ected by the decrease in number
of fl ower visitors after the disturbance, although they were
not necessarily those showing a higher selectiveness. On the
other hand, fl ower symmetry did not have a signifi cant eff ect
Figure 3. Mean ( SE) values of (A) degree, (B) strength, and (C) selectiveness (d ′ ) across years for each fl ower visitor group. In all cases,
the interaction between year and fl ower visitor group was signifi cant at p 0.05. Diff erent letters indicate signifi cant diff erences among
years within each functional group.
EV-7
Lastly, fl ies and hoverfl ies tended to contribute more,
but ants less, to nestedness in 2011 than the other years,
though diff erences were not signifi cant. e rest of groups
showed interannual variations that were not related to the
disturbance (Fig. 4B).
Species roles in the networks
Standardized connection, c , varied among years for plant
species (Table 3), although such variation was unrelated to
the disturbance (Fig. 5A). By contrast, the loss of habitat did
infl uence c for fl ower visitors, although this was contingent
upon the functional group (Table 3, Fig. 5B). Both ants and
butterfl ies showed lower c in 2011 compared to the previ-
ous years whereas the other fl ower visitors showed either no
signifi cant variation across years or variation was not related
to the disturbance (Fig. 5B).
On the contrary, participation values, z , showed low
temporal variation for both plants (Table 3) and fl ower
visitors (year: χ 2 2.92, df 3, p 0.40; functional group:
χ 2 5.11, df 6, p 0.53), and there was not signifi cant
interaction year functional group; Table 3).
Flower visitor species
For all species level network parameters, a signifi cant inter-
action was found between year and fl ower visitor functional
group (Table 3), indicating that such groups do not vary
consistently along time. Ants, beetles, and butterfl ies showed
lower degrees in 2011 compared to the other years, whereas
the rest of functional groups either showed no annual
diff erences in their degree, or these were not due to the
disturbance (Fig. 3A). Ants also showed a reduced strength
in 2011 compared to the previous years (Fig. 3B), whereas
the other groups showed either no variation among years or
the variation was not related to the disturbance (Fig. 3B).
e loss of habitat led to a higher insect selectiveness,
d ′ , in all functional groups, although only ants and beetles
were signifi canty more selective after the disturbance than
the previous years (Fig. 3C).
Regarding weighted closeness centrality, ants and
butterfl ies showed lower values in 2011 than the previous
years, although diff erences were signifi cantly only for ants
(Fig. 4A). e other fl ower visitor groups showed either an
increase (bees, beetles and fl ies) or no interannual variations
in this metric (Fig. 4A).
Figure 4. Mean ( SE) values of (A) weighted closeness centrality, and (B) contribution to nestedness across years for each fl ower visitor
group. Note that, according to the defi nititon of contribution to nestedness (Almeida-Neto and Ulrich 2011), a species with negative values
contributes more than one with positive values. In all cases, the interaction between year and fl ower visitor group was signifi cant at
p 0.05. Diff erent letters indicate signifi cant diff erences among years within each functional group.
EV-8
in 2011 was in fact higher when compared to the previous
years, what would support the lower nestedness values of that
year. Other studies have also documented decreases in net-
work nestedness in disturbed habitats (Vanbergen et al. 2014,
Moreira et al. 2015, Revilla et al. 2015) though not always
(Spiesman and Inouye 2013). A reduced nestedness is often
associated with lower stability and resilience of plant – pollina-
tor networks to perturbations (Bastolla et al. 2009, Fortuna
et al. 2013), although there is controversy on this (James et al.
2012, Saavedra and Stouff er 2013, Rohr et al. 2014).
Greater modularity in disturbed habitats compared to
undisturbed ones has also been reported (Spiesman and
Inouye 2013). A more modular network is thought to
reduce the opportunity for species to facilitate one another
by sharing mutualistic partners and thus to have a destabi-
lizing eff ect ( é bault and Fontaine 2010). Dormann and
Strauss (2014) showed that quantitative modularity (Q) was
positively related to complementary specialization H 2 ′ , using
22 quantitative pollination networks. We thus expected that
an increase in H 2 ′ after habitat disturbance might result in a
higher modularity. Both Q and the number of modules were
actually higher in 2011 than the previous years, supporting
the expectation. Nevertheless, when comparing the z-scores,
the temporal diff erences disappeared.
Discussion
e habitat loss in our study area showed to notably impact
some of the structural properties of the plant – fl ower visitor
network. Although most metrics varied across years, the
number of fl ower visitors and the number of links in the
network decreased much more after the disturbance than
the three previous years. is is concordant with results from
other studies showing a reduction in pollinator availability
and diversity attributed to a decrease in fl oral resource sup-
plies as well as nesting sites after disturbance (Winfree et al.
2011, Hagen et al. 2012, Ferreira et al. 2013, Vanbergen
et al. 2014). Despite this, network connectance was highly
consistent in time, suggesting that the number of links
changes with a similar proportion as the number of species
does; this has also been reported in other studies that have
examined temporal variation in the structure of pollination
networks (Petanidou et al. 2008). Moreover, the nested pat-
tern of interactions was weaker the year following the dis-
turbance, which suggests that some generalist species (either
plants or fl ower visitors) disappeared or were less abundant –
and thus likely had fewer interactions – and/or that more
specialized interactions appeared among the prevalent species
in the community. e level of network specialization ( H 2 ′)
Figure 5. Mean ( SE) s tandardized connection, c, across years in (A) plant species, and (B) fl ower visitor functional groups. e interaction
between year and functional group was signifi cant at p 0.05. Diff erent letters indicate signifi cant diff erences among years (A), or among
years within each functional group (B).
EV-9
resources availability, together with a likely reduction in avail-
able nesting sites for some insect species, led to a reduction
in species richness and abundance of fl oral visitors, which
translated in turn to a less nested and more modular network
composed of more specialized interactions. Not all fl ower-
visitor functional groups were similarly aff ected; beetles, fl ies
and ants were more negatively infl uenced by the disturbance
than other groups such as bees and wasps, what we attribute
to their overall lower mobility. ese fi ndings do not sup-
port, thus, the idea that non-social insects with free-living
progeny are less infl uenced by habitat destruction than
social bees (Jauker et al. 2009, Parsche et al. 2011). Future
empirical studies from other systems are necessary to under-
stand the mechanisms by which diff erent functional groups
of fl ower visitors respond to disturbances and to assess the
consequences of such diff erent responses on ecosystem
functioning.
Acknowledgements – We thank Jaume Reus, Pep Mora, Joan
Torrandell and Zeeba Khan for assistance in the fi eld, and David
Gibbs, David Baldock, Jordi Ribes, Marcos B á ez, M. Carles-Tolr á ,
Paco Laroche, Pedro Orom í , Xavier Canyelles and Xavier Espadaler
for insect identifi cations.
Funding – is study is framed within projects CGL2010-18759/
BOS and CGL2013- 44386-P fi nanced by the Spanish Ministry of
Economy and Competitiveness. RCU was supported by a predoc-
toral grant from the Spanish Government (Ministry of Economy
and Competitiveness and Ministry of Education, Culture and
Sport), whereas AL was supported by a postdoctoral contract
co-funded by the Regional Government of the Balearic Islands and
the European Social Fund 2014 – 2020.
Permissions – e Servei de Protecci ó d ’ Esp è cies, Espais de Natura
Balear (Conselleria de Agricultura, Medi Ambient and Territori)
provided permissions to work at the study site.
References
Aguilar, R. et al. 2006. Plant reproductive susceptibility to habitat
fragmentation: review and synthesis through a meta-analysis.
– Ecol. Lett. 9: 968 – 980.
Aguirre-Guti é rrez, J. et al. 2015. Susceptibility of pollinators to
ongoing landscape changes depends on landscape history.
– Divers. Distrib. 21: 1129 – 1140.
Aizen, M. A. et al. 2012. Specialization and rarity predict nonran-
dom loss of interactions from mutualist networks. – Science
335: 1486 – 1489.
Alarc ó n, R. 2010. Congruence between visitation and pollen-
transport networks in a California plant – pollinator community.
– Oikos 119: 35 – 44.
Almeida-Neto, M. and Ulrich, W. 2011. A straightforward
computational approach for measuring nestedness using quan-
titative matrices. – Environ. Model. Sofware 26: 173 – 178.
Ashworth, L. et al. 2004. Why do pollination generalist and
specialist plant species show similar reproductive susceptibility
to habitat fragmentation? – J. Ecol. 92: 717 – 719.
Bascompte, J. 2009. Mutualistic networks. – Front. Ecol. Environ.
7: 429 – 436.
Bastolla, U. et al. 2009. e architecture of mutualistic networks
minimizes competition and increases biodiversity. – Nature
458: 1018 – 1020.
Bl ü thgen, N. et al. 2006. Specialization, constraints, and confl icting
interests in mutualistic networks. – Curr. Biol. 17: 341 – 346.
Burkle, L. A. et al. 2013. Plant – pollinator interactions over 120
years: loss of species, co-occurrence and function. – Science
339: 1611 – 1615.
At the species level, most network metrics varied across
years. For plants, only strength was found to be consistent in
time, indicating that the relative importance of a particular
plant species for fl ower visitors was rather constant across
the four study years, also after perturbation. is is indeed
interesting as it suggests that despite some plants ’ traits can
change over time (e.g. its abundance or its degree), the sum
of dependencies of each fl ower visitor species on that plant
is not modifi ed. By contrast, the temporal changes observed
in plant degree, weighted closeness centrality, and contri-
bution to nestedness indicate that a plant ’ s position in the
network depend upon how generalist it is and on how gen-
eralists their fl ower visitors are in a particular year. Likewise,
given the high temporal fl uctuations in insect populations,
the levels of plant selectiveness (d ′ ) are expected to be highly
variable over time; the highest values observed in 2011 prob-
ably refl ect that the lowest fl ower visitor abundances were
found that year.
Regarding fl ower visitors, all species-level metrics varied
across years, including species strength. In this case, the
importance of a particular insect species for the entire plant
community is likely to vary much depending upon its abun-
dance and generalization level, which might be more vari-
able than for plant species. In fact, we found large diff erences
across years in fl ower visitor abundances, the year after habitat
disturbance showing the lowest values. Disturbance did not
only infl uence fl ower visitor abundance but also species rich-
ness, although this was contingent upon functional group.
Species richness of beetles and fl ies were the most negatively
aff ected, which might be because they are the most impor-
tant functional groups in terms of percentage of pollinator
visits, with 53.4% and 19.9%, respectively, (average for the
four years). Moreover, ants, which represent a 4.7% of the
total of visits, were the functional group most infl uenced
by the disturbance regarding generalization level, strength,
selectiveness, and closeness centrality. ey also played a less
important role as module connectors after the disturbance.
is is most likely attributable to the shorter foraging range,
compared to the other functional groups, specifi cally fl ying
insects. It is unknown why the other groups, especially bees
and wasps, which are also abundant in the area (14.5% and
3.8% of the visits, respectively), were not infl uenced to the
same extent. One possible reason is that these groups depend
upon nesting sites and food resources that were not a limit-
ing factor, and despite fl ower visitor abundance was reduced
the disturbance did not lead any species to local extinction.
In addition, bees and wasps might have a greater mobility
(Torn é -Noguera et al. 2014) and thus are less infl uenced
by local disturbances. We need more information on the
rewiring capacity of diff erent fl ower visitor functional groups
in face of a habitat disturbance (Burkle et al. 2013, Aguirre-
Guti é rrez et al. 2015, L á zaro et al. 2016) as well as on the
ultimate consequences of such rewiring for plant repro-
ductive success. From a conservation viewpoint, monitor-
ing these changes in natural communities could be highly
relevant to the proposal of eff ective conservation strategies
(Campos-Navarrete et al. 2013).
In short, our fi ndings showed that the large distur-
bance occurred in the study area altered to a large extent
the structure of the plant – fl ower visitor network, making
it more vulnerable to future perturbations. e lower fl oral
EV-10
Keitt, T. H. 2009. Habitat conversion, extinction thresholds, and pol-
lination services in agroecosystems. – Ecol. Appl. 19: 1561 – 1573.
Kennedy, C. M. et al. 2013. A global quantitative synthesis of
local and landscape eff ects on wild bee pollinators in agroeco-
systems. – Ecol. Lett. 16: 584 – 599.
Klein, A. M. et al. 2007. Importance of pollinators in changing
landscapes for world crops. – Proc. R. Soc. B 274: 303 – 313.
L á zaro, A. et al. 2016. Moderation is best: eff ects of grazing
intensity on plant – fl ower visitor networks in Mediterranean
communities. – Ecol. Appl. 26: 796 – 807.
Moreira, E. F. et al. 2015. Spatial heterogeneity regulates plant –
pollinator networks across multiple landscape scales. – PLoS
One 10: e0123628.
Nielsen, A. and Totland, Ø . 2014. Structural properties of
mutualistic networks withstand habitat degradation while
species functional roles might change. – Oikos 123: 323 – 333.
Olesen, J. M. et al. 2007. e modularity of pollination networks.
– Proc. Natl Acad. Sci. USA 104: 19891 – 19896.
Parsche, S. et al. 2011. Experimental environmental change and
mutualistic vs antagonistic plant fl ower – visitor interactions.
– Perspect. Plant Ecol. Evol. Syst. 13: 27 – 35.
Petanidou, T. et al. 2008. Long-term observation of a pollination
network: fl uctuation in species and interactions, relative
invariance of network structure and implications for estimates
of specialization. – Ecol. Lett.11: 564 – 575.
Potts, S. G. et al. 2003. Linking bees and fl owers: how do fl oral
communities structure pollinator communities? – Ecology 84:
2628 – 2642.
Revilla, T. A. et al. 2015. Robustness of mutualistic networks under
phenological change and habitat destruction. – Oikos 124:
22 – 32.
Rohr, R. P. et al. 2014. On the structural stability of mutualistic
systems. – Science 345: 1253497.
Saavedra, S. and Stouff er, D. B. 2013. “ Disentangling nestedness ”
disentangled. – Nature 500: E1 – E2.
Spiesman, B. J. and Inouye, B. D. 2013. Habitat loss alters the
architecture of plant – pollinator interaction networks. – Ecology
94: 2688 – 2696.
é bault, E. and Fontaine, C. 2010. Stability of ecological
communities and the architecture of mutualistic and trophic
networks. – Science 329: 853 – 856.
Torn é -Noguera, A. et al. 2014. Determinants of spatial distribution
in a bee community: nesting resources, fl ower resources, and
body size. – PLoS One 9: e97255.
Traveset, A. et al. 2017. Data from: Eff ects of habitat loss on the
plant – fl ower visitor network structure of a dune community.
– Fileshare Repository: < https://csicannatraveset.sharefi le.com/
d-sf34b761ef9a4607a > .
Tylianakis, J. M. et al. 2007. Habitat modifi cation alters the structure
of tropical host – parasitoid food webs. – Nature 445: 202 – 205.
Tylianakis, J. M. et al. 2010. Conservation of species interaction
networks. – Biol. Conserv. 143: 2270 – 2279.
Vanbergen, A. J. et al. 2014. Grazing alters insect visitation networks
and plant mating systems. – Funct. Ecol. 28: 178 – 189.
Viana, B. F. et al. 2012. How well do we understand landscape
eff ects on pollinators and pollination services? – J. Pollination
Ecol. 7: 31 – 41.
Williams, N. M. et al. 2010. Ecological and life-history traits
predict bee species responses to environmental disturbances.
– Biol. Conserv. 143: 2280 – 2291.
Winfree, R. et al. 2011. Native pollinators in anthropogenic
habitats. – Annu. Rev. Ecol. Evol. Syst. 42: 1 – 22.
Zuur, A. F. et al. 2009. Mixed eff ects models and extensions in
ecology with R. Book series: statistics for biology and health.
– Springer, ISBN 9780387874579.
Campos-Navarrete, M. J. et al. 2013. Structure of plant – Hymenoptera
networks in two coastal shrub sites in Mexico. – Arthropod
Plant Interactions. 7: 607 – 617.
Carvell, C. et al. 2012. Molecular and spatial analyses reveal links
between colony-specifi c foraging distance and landscape-level
resource availability in two bumblebee species. – Oikos 121:
734 – 742.
Castro-Urgal, R. et al. 2012. How diff erent link weights aff ect the
structure of quantitative fl ower – visitation networks. – Basic
Appl. Ecol. 13: 500 – 508.
Dormann, C. F. and Strauss, R. 2014. A method for detecting
modules in quantitative bipartite networks. – Methods Ecol.
Evol. 5: 90 – 98.
Dormann, C. F. et al. 2009. Indices, graphs and null models:
analyzing bipartite ecological networks. – Open Ecol. J. 2:
7 – 24.
Ebeling, A. et al. 2008. How does plant richness aff ect pollinator
richness and temporal stability of fl ower visits? – Oikos 117:
1808 – 1815.
Eckert, C. G. et al. 2010. Plant mating systems in a changing
world. – Trends Ecol. Evol. 25: 35 – 43.
Fenster, C. B. et al. 2004. Pollination syndromes and fl oral
specialization. – Annu. Rev. Ecol. Evol. Syst. 35: 375 – 403.
Ferreira, P. A. et al. 2013. What do we know about the eff ects of
landscape changes on plant – pollinator interaction networks?
– Ecol. Indic. 31: 35 – 40.
Ferreira, P. A. et al. 2015. Responses of bees to habitat loss in
fragmented landscapes of Brazilian Atlantic Rainforest.
– Landscape Ecol. 30: 2067 – 2078.
Fortuna, M. A. et al. 2013. Habitat loss and the disassembly of
mutalistic networks. – Oikos 122: 938 – 942.
Garibaldi, L. A. et al. 2014. From research to action: enhancing
crop yield through wild pollinators. – Front. Ecol. Environ.
12: 439 – 447.
Genini, J. et al. 2010. Cheaters in mutualistic networks. – Biol.
Lett. 6: 494 – 497.
Ghazoul, J. 2006. Floral diversity and the facilitation of pollination.
– J. Ecol. 94: 295 – 304.
Gonz á lez, A. et al. 2011. e disentangled bank: how loss of
habitat fragments and disassembles ecological networks. – Am.
J. Bot. 98: 503 – 516.
Guimer à , R. and Amaral, L. A. N. 2005. Functional cartography
of complex metabolic networks. – Nature 433: 895 – 900.
Hagen, M. and Kraemer, M. 2010. Agricultural surroundings
support fl ower-visitor networks in an Afrotropical rain forest.
– Biol. Conserv. 143: 1654 – 1663.
Hagen, M. et al. 2012. Biodiversity, species interactions and
ecological networks in a fragmented world. – Adv. Ecol. Res.
46: 89 – 210.
Hegland, S. J. and Boeke, L. 2006. Relationships between the density
and diversity of fl oral resources and fl ower visitor activity in a
temperate grassland community. – Ecol. Entomol. 31: 532 – 538.
Hegland, S. J. et al. 2014. Floral neighbourhood eff ects on
pollination success in red clover are scale- dependent. – Funct.
Ecol. 28: 561 – 568.
James, A. et al. 2012. Disentangling nestedness from models of
ecological complexity. – Nature 500: 227 – 230.
Jauker, F. et al. 2009. Pollinator dispersal in an agricultural matrix:
opposing responses of wild bees and hoverfl ies to landscape
structure and distance from main habitat. – Landscape Ecol.
24: 547 – 555.
Kaiser-Bunbury, C. N. et al. 2010. e robustness of pollination
networks to the loss of species and interactions: a quantitative
approach incorporating pollinator behaviour. – Ecol. Lett. 13:
442 – 452.
Supplementary material (available online as Appendix oik-
04154 at < www.oikosjournal.org/appendix/oik-04154 > ).
Appendix 1.
