Mutually beneficial pollinator diversity
and crop yield outcomes in small
and large farms
Lucas A. Garibaldi,
*Luísa G. Carvalheiro,
Bernard E. Vaissière,
Breno M. Freitas,
Hien T. Ngo,
Fermín J. Chamorro García,
Fabiana Oliveira da Silva,
Márcia de Fátima Ribeiro,
Maria C. Gaglianone,
Alípio J.S. Pacheco Filho,
Lucia H. Piedade Kiill,
Guiomar Nates Parra,
Ranbeer S. Rawal,
Antonio M. Saraiva,
Blandina F. Viana,
Ecological intensification, or the improvement of crop yield through enhancement of
biodiversity, may be a sustainable pathway toward greater food supplies. Such sustainable
increases may be especially important for the 2 billion people reliant on small farms,
many of which are undernourished, yet we know little about the efficacy of this approach.
Using a coordinated protocol across regions and crops, we quantify to what degree
enhancing pollinator density and richness can improve yields on 344 fields from
33 pollinator-dependent crop systems in small and large farms from Africa, Asia, and Latin
America. For fields less than 2 hectares, we found that yield gaps could be closed by a
median of 24% through higher flower-visitor density. For larger fields, such benefits only
occurred at high flower-visitor richness. Worldwide, our study demonstrates that ecological
intensification can create synchronous biodiversity and yield outcomes.
More than 2 billion people are reliant on
smallholder agriculture (farms with less
than 2 ha) in developing nations, repre-
senting 83% of the global agricultural
population (1,2). In such countries, hu-
man population is growing faster than in devel-
oped nations, while many of the rural inhabitants
are poor, undernourished, and live in conditions
where the environment is either degraded or
being degraded (2–4). As a result, improving
the livelihoods of smallholders through higher
and more stable crop yields, while minimizing
negative environmental impacts, is essential for
achieving global food security and poverty re-
duction (3,5). Ecosystem services enhanced
through biodiversity (such as nutrient cycling,
biotic pollination, or pest control) can replace,
complement, or interact synergistically with ex-
ternal inputs (such as fertilizers, introduction of
pollinator colonies, and pesticides) and should
create mutually beneficial environmental and food-
supply scenarios (6,7). Despite advocacy for such
“ecological”intensification (6–8), its effectiveness
in small versus large holdings is largely unknown.
Moreover, smallholding crop systems in devel-
oping countries have been largely neglected in
ecosystem-services research (2,4).
Yield gaps, defined here as the difference in
crop yield between high- and low-yielding farms
of a given crop system (Fig. 1), are pervasive for
small holdings in many developing countries (7–9).
This definition of yield gaps is particularly rel-
evant for smallholders, as the attainable yields in
field trials and research centers usually result
from applying different technologies (e.g., nu-
trients provided as manure in crop-livestock
smallholding systems versus synthetic fertilizers
used in large monocultures in research centers)
(3,7). Such empirical estimates of attainable yields
are more conservative than modeled potential
yields (10), but they are likely achievable with
current technology (9). Indeed, the marginal re-
turns from additional inputs can make modeled
potential yields nonprofitable for farmers (9).
Yield gaps can be partially closed through the
provision of optimal amounts and quality of
resources, such as water, nutrients, and pollen
(9,11). Although fruit or seed set of many crops
relies on wild pollinators (12), management for
improved pollination services is uncommon in
these systems (13), likely contributing to yield
gaps globally (11). Indeed, pollination has been
neglected even in studies analyzing the con-
tinental or global drivers of yield gaps (5,7,9,10).
Pollinator deficits may be more relevant than
before, as (i) other resources (e.g., nutrients)
are increasingly provided (e.g., fertilizers) to crops
(6,8); (ii) cultivated area of pollinator-dependent
crops is expanding more rapidly than the area of
pollinator-independent crops (11); (iii) cultivated
area of pollinator-dependent crops is also expand-
ing more rapidly than the stock of managed
honey bee colonies (14); and (iv) populations of
wild pollinators are increasingly threatened (15,16).
Furthermore, pollinator-dependent crops provide
gions of the world where micronutrient deficien-
cies are common (4). To date, it is uncertain to
whatdegree localpopulations of pollinators need
to be enhanced (“flower-visitor density gap”),
and how much of the yield gaps (kg ha
be closed by such management (Fig. 1).
388 22 JANUARY 2016 •VOL 351 ISSUE 6271 sciencemag.org SCIENCE
Instituto de Investigaciones en Recursos Naturales, Agroecología
y Desarrollo Rural (IRNAD), Sede Andina, Universidad Nacional de
Río Negro (UNRN) and Consejo Nacional de Investigaciones
Científicas y Técnicas (CONICET), Mitre 630, CP 8400, San
Carlos de Bariloche, Río Negro, Argentina.
Ecologia, Universidade de Brasília, Campus Universitário Darcy
Ribeiro, Brasília - DF, 70910-900, Brazil; Centre for Ecology,
Evolution and Environmental Changes (CE3C), Faculdade de
Ciências da Universidade de Lisboa 1749-016 Lisboa, Portugal &
Naturalis Biodiversity Center, postbus 9517, 2300, RA, Leiden,
Institut national de la recherche agronomique,
UR406 Abeilles et Environnement, 228 route de l'Aérodrome,
CS40509, F84914, Avignon Cedex 9, France.
Agriculture Organization of the United Nations, Viale delle Terme
di Caracalla 00153, Rome, Italy.
Departamento de Zoologia,
Universidade Federal da Bahia, Instituto de Biologia, Rua Barão
de Geremoabo, S/N, Campus de Ondina, CEP 40170110,
Salvador, BA, Brazil.
Departamento de Zootecnia–Centro de
Ciências Agrárias, Universidade Federal do Ceará, Campus
Universitário do Pici, CEP 60021970, Fortaleza, CE, Brazil.
Secretariat, Intergovernmental Platform on Biodiversity and
Ecosystem Services (IPBES), UN Campus, Platz der Vereinten
Nationen 1, D-53113, Bonn, Germany.
Universidad Nacional del Comahue-CONICET, Instituto de
Investigaciones en Biodiversidad y Medioambiente, Quintral 1250,
CP 8400, San Carlos de Bariloche, Río Negro, Argentina.
Norwegian Institute for Nature Research, Post Office Box 5685
Sluppen, NO-7485, Trondheim, Norway.
Key Laboratory for
Insect-Pollinator Biology of the Ministry of Agriculture, Institute of
Apicultural Research, Chinese Academy of Agricultural Sciences,
100093, Beijing, China.
Pontifícia Universidade Católica do Rio
Grande do Sul (PUCRS), Av. Ipiranga, 6681, CEP 90619900,
Porto Alegre, RS, Brazil.
Department of Plant Protection, Faculty
of Agriculture, Bogor Agricultural University. Jln. Kamper,
Darmaga, Bogor 16680, West Java, Indonesia.
investigaciones en Abejas (LABUN), Departamento de Biología,
Universidad Nacional de Colombia, Sede Bogotá, CP11001,
Departamento de Educação em Ciências
Agrárias e da Terra, Universidade Federal de Sergipe, Campus do
Sertão, Rodovia Engenheiro Jorge Neto. Silos KM 0, CEP
49680000, Nossa Senhora da Gloria, SE, Brazil.
Agriculture and Animal Science, Rampur, Chitwan, Nepal.
Embrapa Semiárido, BR 428, Km 152, C.P. 23, zona rural, CEP
56302970, Petrolina, PE, Brazil.
Jardim Botânico do Rio de
Janeiro (JBRJ), Rua Pacheco Leão 915, CEP 22460030, Rio de
Janeiro, RJ, Brazil.
Laboratório de Ciências Ambientais,
Universidade Estadual do Norte Fluminense Darcy Ribeiro, CEP
28013620, Campos dos Goytacazes, RJ, Brazil.
Zimbabwe, Faculty of Agriculture, Crop Science Department, Post
Office Box MP167, Mt Pleasant, Harare, Zimbabwe.
Conservation and Management of Pollinators for Sustainable
Agriculture through Ecosystem Approach project, Honey Bee
Research Institute, National Agricultural Research Centre, Park
Road, Post Office Box 44000, Islamabad, Pakistan.
Agricultural and Livestock Research Organisation-Sericulture, Post
Office Box 7816 code 01000 Thika, Kenya.
Agriculture and Natural Sciences, School of Biological Sciences,
University of Cape Coast, Cape Coast, Ghana.
Recursos Genéticos e Biotecnologia, Parque Estação Biológica,
W5 Norte (final), CEP 70770917, Brasília, DF, Brazil.
Meio Ambiente e Recursos Hídricos (INEMA)–UR Extremo Sul,
Rua Viena, no. 425, Bairro Dinnah Borges, CEP 45820970,
Eunápolis, BA, Brazil.
G.B. Pant Institute of Himalayan
Environment and Development, Kosi-Katarmal, Almora–263 643,
Department of Plant Pest Diseases, Faculty
of Agriculture, University of Brawijaya. Jl. Veteran, Malang 65145,
East Java, Indonesia.
Universidade de São Paulo, Escola
Politécnica, Av. Prof. Luciano Gualberto Travessa 3, n.158, CEP
South African National
Biodiversity Institute, Kirstenbosch Research Centre, Private Bag
X7, Claremont, 7735, South Africa.
Conservation Ecology and
Entomology, Stellenbosch University, Private Bag X1, 7602,
Matieland, South Africa.
Centro de Pesquisa Emílio Schenk,
Fundação Estadual de Pesquisa Agropecuária (Fepagro Vale do
Taquari), 1° Distrito, Fonte Grande, Caixa Postal 12, CEP
95860000, Taquari, RS, Brazil.
*Corresponding author. E-mail: firstname.lastname@example.org
We recorded flower-visitor density, flower-visitor
richness, and crop yield in 344 fields of 33 crop
systems across small and large holdings in Africa,
Asia, and Latin America (figs. S1 and S2). To
avoid the limitations of different methodolo-
gies, and considering the global nature of our
focus, we performed coordinated experiments
(17) over a 5-year period (2010–2014)—a col-
laborative approach that encompassed large
geographic ranges involving a standardized pro-
tocol. This sampling protocol (18) used fields with
contrasting flower-visitor density and richness
not confounded with management variables
other than the ones that were employed to in-
fluence flower-visitor assemblages (table S1).
Therefore, crop systems are defined as a crop
species in a particular year and region subject
to similar management, except for flower-visitor
density and richness (table S1). Following this
protocol (18), flower-visitor density was measured
by scan sampling a fixed number of open floral
units (hereafter “flowers”) in each of four subplots
in each field, on at least four dates during the
main flowering period (19). Flower-visitor spe-
cies richness was measured by netting all visitors
of crop flowers along six 25-m-long and 2-m-wide
transects for herbaceous crops (or six pairs of
adjacent trees for orchard crops). Crop yield was
measured by harvesting all the fruits or seeds of
) and then mul-
or by harvesting 1 to 5 m
, according to the
crop (18). Crop yield (log
) was analyzed
through (hierarchical) mixed-effects models with
fields nested within crop systems. Fixed effects
were flower-visitor density (number of visitors in
SCIENCE sciencemag.org 22 JANUARY 2016 •VOL 351 ISSUE 6271 389
Fig. 1. Pollinator deficit is defined here as the
amount of yield gap that can be accounted by
closing flower-visitor density gap. Worl dwide, for
<2-ha fields, our study shows that yield gaps could
be closed by a median of 24% (mean = 31%) through
higher flower-visitor density (table S2). For larger
fields, such a level of yield benefits only occurred if
they sustained high flower-visitor richness (Fig. 2).
Although the relation between crop yield and flower-
visitor density is expected to be positive but as-
ymptotic (11), our study supports a linear relation,
demonstrating that the highest levels of flower-
visitor density observed around the world are still at
Fig. 2. Worldwide, the benefits of flower-visitor
density to crop yield are greater for smaller than
larger holdings, and when flower-visitor richness
is higher. Moreover, high richness can compensate
this negative influence of field size. Each point is a field
within a crop system; lines are the fixed-effect pre-
dictions from the best hierarchical model without co-
variables. Small (<2 ha) versus large holdings, and low
(<3 species) versus high richness, are categories only
for graphical purposes, while the model considers
field size and species richness as quantitative varia-
bles. By using the same protocol, we could express
density as number of visitors in 100 crop flowers, avoid-
ing standardizations to integrate results from different
crop systems. Because yield (kg ha
different magnitudes for different crop species (e.g.,
coffee versus tomatoes), we present the crop yield
after subtracting the random intercept for each crop
100 crop flowers), flower-visitor richness (number
of species per field in 30 min of net sampling),
field size (log
ha), and their interactions (19).
Random effects were intercepts and slopes for
each crop system for the relation between crop
yield and flower-visitor density and richness. Al-
though our focus was on developing countries,
research partners from Norway followed the same
protocol in three crop systems, and their data were
included in the analyses for comprehensiveness.
Globally, yield gaps were large and common
across fields in each crop system (Fig. 1 and table S2).
Crop yield in low-yielding fields (10th percentile)
was, on average, only 47% of the value in high-
yielding fields (90th percentile; see table S2 for
). Differences in flower-visitor density (i.e.,
flower-visitor density gaps) were similarly large
(Fig. 1 and table S2). The fields with low flower-
visitor density (10th percentile: 2.5 flower visitors
in 100 flowers on average across crop systems) had
only 44% of the individuals of the fields with high
values (90th percentile: 5.5 flower visitors in 100
flowers on average across crop systems). These re-
sults indicate that even for crops of a given variety
planted within a particular region and year, and
managed similarly, there are large opportunities
to increase flower-visitor densities and yields to
the values of the best farms (90th percentile).
The effects of flower-visitor density on crop
yield were largely influenced by field size (which
ranged from 0.1 to 327.2 ha in our study) and
flower-visitor richness (which ranged from 0 to
11 species in our study), as reflected by a three-
way interaction (Fig. 2 and table S3). For small-
holdings worldwide, crop yield increased linearly
with flower-visitor density, suggesting that in-
adequate pollination quantity and/or quality is
partly responsible for yield gaps (11,20). These
benefits were irrespective of flower-visitor rich-
ness. In contrast, for larger holdings, the benefits
of flower-visitor density oncropyieldweregreater
in fields with higher flower-visitor richness (Fig. 2
and table S3). Therefore, greater flower-visitor
richness could compensate the negative influence
that field size had on the relationship (slope)
between crop yield and flower-visitor density.
For example, in fields with only one flower-visitor
species, the increase in crop yield per unit of
flower-visitor density was 106% higher for fields
of 2 ha than for those of 20 ha. However, this
difference was reduced to only 16% when four
flower-visitor species were present. Globally, our
results suggest that the effectiveness of ecological
intensification (represented here by flower-vi sitor
density) differs between small and large hold-
ings, being greater for small holdings and when
species richness is enhanced.
To test if these results could be explained by
environmental and management aspects that
covary with flower-visitor density, flower-visitor
richness, or field size (table S1), we added to the
previous mixed-effects model the following fixed
effects: level of conventional intensification (a
quantitative index based on the presence of mono-
cultures, synthetic fertilizers, herbicides, pesticides,
and fungicides) (19); isolation from seminatural
or natural habitats (log
km); crop pollinator
dependence (%); latitude (decimal degrees); long-
itude (decimal degrees); baseline level of
flower-visitor density (10th percentile: number
per 100 flowers); magnitude of yield gap (%); mag-
nitude of flower-visitor gap (%); and the two-way
interactions between each of these covariables and
flower-visitor density (19). The best-fitting model
(i.e., lower corrected Akaike’s information crite-
rion, AICc) (19) was then derived from evaluatio n
of all possible combinations of predictors and
covariables, including a model without predictors.
The influences of flower-visitor density, flower-
visitor richness, field size, and their interactions
were still included as predictors of crop yield in
the best model, in addition to the intensification
level, isolation from natural habitats, and f lower-
visitor gap (table S3). Importantly, fixed-effect
values (and standard errors) for these predictors
were of similar magnitude in the models with and
without covariables (table S3), reflecting their
independent contribution from the covariables
in predicting crop yield [see also VIF (variance
inflation factor) values in table S3]. The sum
of the AICc weights of all the models for each
predictor and covariable was used as an estimate
of its relative importance (19). Notably, among
all the variables we tested, flower-visitor den-
sity was the most important predictor of crop
yield (Fig. 3). As expected (21), the level of
conventional intensification was an important
predictor of crop yield (Fig. 3), showing a
positive relation (table S3). Crop yield decreased
with isolation from natural habitats, and more
so when flower-visitor density was lower (table
S3). Worldwide, our data show that effects of
flower-visito r density, flower-visitor richness, and
field size are highly relevant in the context of, and
not confounded by, other environmental and
management variables affecting crop yield.
Our best-fitting model (table S3) allows the
estimation of the degree to which yield gaps
) can be closed by enhancing local pop-
ulations of flower visitors for a given field size
and several other key management and environ-
mental covariables (note the high coefficient of
less than 2 ha, the enhancement of flower-visitor
density in fields with the lowest values (10th
percentile) to those of the best fields (90th per-
centile) should close yield gaps by a median of
24% (Fig. 1 and table S2). The remaining 76% of
oriented to the optimal provision and efficient
use of other resources (e.g., radiation, nutrients,
water), including sowing date, plant density, gen-
etic material, conservation agriculture, and in-
tegrated pest management, among many others
(5–7,9,10). In contrast, for larger fields, such level
of yield benefits from enhancement of flower-
visitor density occurs only if these fields have high
flower-visitor richness (Fig. 2 and table S2). In our
study, the influences of field size were not con-
founded by several important environmental and
management variables affecting crop yield (table
S3). Lower benefits from flower-visitor density in
390 22 JANUARY 2016 •VOL 351 ISSUE 6271 sciencemag.org SCIENCE
Fig. 3. Flower-visitor density (D) was the most important predictor of crop yield for pollinator-
dependent crops globally.The relative importance is the sum of the Akaike information criterion weights
of the models with each predictor. Inten, level of conventional intensification; Isolation, distance to
seminatural or natural areas; Vis gap, magnitude of flower-visitor gap; F size, field size; Richness, flower-visitor
richness; Dependence, crop pollinator dependence; Vis base, baseline level of flower-visitor density.
pollinated by flower visitors with large foraging
ranges, which are usually generalist species, such
as honey bees (12). In accordance with this hypo-
thesis, we found greater dominance of Apis spp.
in larger holdings regardless of species richness
(fig. S3), and that flower-visitor density effects
were enhanced when richness increased in large
fields (Fig. 2). Such synergistic influences among
pollinator species on crop yield (kg ha
due to different nonexclusive mechanisms (22),
including pollination niche complementarity (23,24),
interspecific interactions (25,26), or raising the
chances of providing effective pollinator species
(i.e., sampling effects of biodiversity) (27,28).
Pollinator deficits have been neglected from
previous global or continental yield gap analyses
(5,7,9,10). However, here we found that they are
pollinator-dependent crops in small holdings
(Fig. 1 and table S2), even after considering sev-
eral environmental and management predictors
of crop yield (Fig. 3). Indeed, flower-visitor den-
sity was the most important predictor of crop
yield. Closing flower-visitor density gaps is a re-
alistic objective, as our figures are based on the
densities observed in real-world farms (i.e., the
difference between the 90th and 10th percen-
tiles). Unfortunately, recent studies suggest that
flower-visitor assemblages in agroecosystems are
increasingly threatened because of declining flo-
ral abundance and diversity, as well as increasing
exposure to pesticides and parasites (15,16). Such
trends can be reversed by a combination of prac-
tices, the effectiveness of which is context depen-
dent, including sowing flower strips and planting
hedgerows, providing nesting resources, more
targeted use of pesticides, and/or restoration of
seminatural and natural areas adjacent to crops
(table S1) (13,29).
Enhancing smallholder livelihoods through
greater crop yields while reducing negative en-
vironmental impacts from agriculture is one of
the greatest challenges for humanity (3,5). More-
over, from a food-security point of view, pollinator-
dependent crops provide essential micronutrients
to human health where needed (4). Our data in-
dicate that the effectiveness of ecological inten-
sification through pollination services was greater
for small, rather than large, holdings. Using pol-
lination services as a case study, we demonstrated
that ecological intensification can create mutually
beneficial scenarios between biodiversity and crop
REFERENCES AND NOTES
1. S. K. Lowder, J. Skoet, S. Singh, “What do we really know about
the number and distribution of farms and family farms
worldwide?”Background paper for the State of Food and
Agriculture 2014 (Food and Agriculture Organization of the
United Nations, Rome, 2014).
2. P. R. Steward et al., Agric. Food Security 3, 5 (2014).
3. M. Herrero et al., Science 327, 822–825 (2010).
4. R. Chaplin-Kramer et al., Proc. R. Soc. B Biol. Sci. 281,
5. H. C. J. Godfray et al., Science 327, 812–818 (2010).
6. R. Bommarco, D. Kleijn, S. G. Potts, Trends Ecol. Evol. 28,
7. P. Tittonell, K. E. Giller, Field Crops Res. 143,76–90 (2013).
8. J. A. Foley et al., Nature 478, 337–342 (2011).
9. N. D. Mueller et al., Nature 490, 254–257 (2012).
10. D. B. Lobell, K. G. Cassman, C. B. Field, Annu. Rev. Environ.
Resour. 34, 179–204 (2009).
11. L. A. Garibaldi, M. A. Aizen, A. M. Klein, S. A. Cunningham,
L. D. Harder, Proc.Natl.Acad.Sci.U.S.A.108,5909–5914
12. L. A. Garibaldi et al., Science 339, 1608–1611 (2013).
13. L. A. Garibaldi et al., Front. Ecol. Environ. 12, 439–447
14. M. A. Aizen, L. D. Harder, Curr. Biol. 19, 915–918 (2009).
15. D. Goulson, E. Nicholls, C. Botías, E. L. Rotheray, Science 347,
16. M. Rundlöf et al., Nature 521,77–80 (2015).
17. L. H. Fraser et al., Front. Ecol. Environ. 11,147–155
18. B. E. Vaissière, B. M. Freitas, B. Gemmill-Herren, Protocol to
detect and assess pollination deficits in crops: a handbook for
its use (Food and Agriculture Organization of the United
Nations, Rome, 2011).
19. See supplementary materials on Science Online.
20. M. A. Aizen, L. D. Harder, Ecology 88, 271–281 (2007).
21. V. Seufert, N. Ramankutty, J. A. Foley, Nature 485, 229–232
22. T. Tscharntke, A. M. Klein, A. Kruess, I. Steffan-Dewenter,
C. Thies, Ecol. Lett. 8, 857–874 (2005).
23. P. Hoehn, T. Tscharntke, J. M. Tylianakis, I. Steffan-Dewenter,
Proc. R. Soc. B Biol. Sci. 275, 2283–2291 (2008).
24. J. Fründ, C. F. Dormann, A. Holzschuh, T. Tscharntke, Ecology
94, 2042–2054 (2013).
25. L. G. Carvalheiro et al., Ecol. Lett. 14, 251–259 (2011).
26. C. Brittain, N. Williams, C. Kremen, A. M. Klein, Proc. Biol. Sci.
280, 20122767 (2013).
27. B. J. Cardinale et al., Nature 443, 989–992 (2006).
28. M. Schleuning, J. Fründ, D. García, Ecography 38, 380–392
29. D. Kleijn et al., Nat. Commun. 6, 7414 (2015).
ACKNO WLEDGM ENTS
Funding was provided by the Global Environment Fund, United
Nations Environment Program, United Nations Food and
Agriculture Organization (GEF/UNEP/FAO) Global Pollination
Project, with additional support to the Food and Agriculture
Organization of the United Nations from the Norwegian
Environment Agency for a project on “Building Capacity in the
Science-Policy Interface of Pollination Services,”and from
the International Fund for Agricultural Development for the
development of the sampling protocol (18). Other funding:
Conselho Nacional de Desenvolvimento Científico e Tecnológico
Brazil, CONICET Argentina (PIP 114-201101-00201), Norwegian
Environment Agency (2012/16642), The Research Council of
Norway (225019), and Universidad Nacional de Río Negro
Argentina (PI 40-B-259, PI 40-B-399). The data used in the
analyses are available in the supplementary materials. L.A.G.,
L.G.C., and J.H. compiled data; L.A.G. and L.G.C. analyzed data;
L.A.G. wrote the first version of the manuscript; the authors named
between L.A.G. and J.Ås. discussed and revised earlier versions
of the manuscript. The authors named between J.An and H.Z.
are listed alphabetically, as they contributed equally, gathering
field data, providing corrections to subsequent manuscript drafts,
and discussing ideas. G. Andersson, B. Geslin, and P. Steward
provided insightful comments on previous versions of this paper.
Materials and Methods
Figs. S1 to S3
Tables S1 to S3
22 June 2015; accepted 15 December 2015
Biogenesis and function of tRNA
fragments during sperm maturation
and fertilization in mammals
*Colin C. Conine,
*Jeremy M. Shea,
Alan G. Derr,
Xin Y. Bing,
Ryan W. Serra,
Benjamin R. Carone,
Emiliano P. Ricci,
†Xin Z. Li,
Melissa J. Moore,
Craig C. Mello,
Oliver J. Rando
Several recent studies link parental environments to phenotypes in subsequent
generations. In this work, we investigate the mechanism by which paternal diet affects
offspring metabolism. Protein restriction in mice affects small RNA (sRNA) levels in
mature sperm, with decreased let-7 levels and increased amounts of 5′fragments of
glycine transfer RNAs (tRNAs). In testicular sperm, tRNA fragments are scarce but
increase in abundance as sperm mature in the epididymis. Epididymosomes (vesicles that
fuse with sperm during epididymal transit) carry RNA payloads matching those of mature
sperm and can deliver RNAs to immature sperm in vitro. Functionally, tRNA-glycine-GCC
fragments repress genes associated with the endogenous retroelement MERVL, in both
embryonic stem cells and embryos. Our results shed light on sRNA biogenesis and its
dietary regulation during posttesticular sperm maturation, and they also link tRNA fragments
to regulation of endogenous retroelements active in the preimplantation embryo.
Accumulating evidence indicates that paren-
tal environments can affect the health of
offspring. For example, paternal nutrition
influences offspring metabolism in mam-
mals (1). Our prior published work showed
that male mice consuming a low-protein diet
fathered offspring exhibiting altered hepatic cho-
lesterol biosynthesis, relative to the offspring
of control males (2). The mechanisms by which
paternal conditions reprogram offspring phe-
notype remain elusive, as males can influence
offspring via the sperm epigenome, microbiome
SCIENCE sciencemag.org 22 JANUARY 2016 •VOL 351 ISSUE 6271 391
Supplementary Materials for
Mutually beneficial pollinator diversity and crop yield outcomes in
small and large farms
Lucas A. Garibaldi,* Luísa G. Carvalheiro, Bernard E. Vaissière, Barbara Gemmill-
Herren, Juliana Hipólito, Breno M. Freitas, Hien T. Ngo, Nadine Azzu, Agustín Sáez,
Jens Åström, Jiandong An, Betina Blochtein, Damayanti Buchori, Fermín J. Chamorro
García, Fabiana Oliveira da Silva, Kedar Devkota, Márcia de Fátima Ribeiro, Leandro
Freitas, Maria C. Gaglianone, Maria Goss, Mohammad Irshad, Muo Kasina, Alípio J.S.
Pacheco Filho, Lucia H. Piedade Kiill Peter Kwapong, Guiomar Nates Parra, Carmen
Pires, Viviane Pires, Ranbeer S. Rawal, Akhmad Rizali, Antonio M. Saraiva, Ruan
Veldtman, Blandina F. Viana, Sidia Witter, Hong Zhang
*Corresponding author. E-mail: email@example.com
Published 22 January 2016, Science 351, 387 (2016)
This PDF file includes:
Materials and Methods
Figs. S1 to S3
Tables S1 to S3
Caption for Database S1
Other Supplementary Materials for this manuscript include the following:
(available at www.sciencemag.org/content/351/6271/387/suppl/DC1)
Materials and Methods
We sampled 344 fields of 33 crop systems in 12 countries (fig. S1 and table S1).
Crop systems were defined as a given crop species, in a particular region and year,
subject to similar management, except for flower-visitor density and richness (table S1).
The crops considered include a wide array of annual and perennial fruit, seed, nut, and
stimulant crops that are pollinator dependent to some degree. Crops pollinated primarily
by wind or autonomous self-pollination were not studied. Crop systems were selected to
represent the spectrum of management practices (traditional, intensive agriculture,
organic agriculture), landscape settings (cleared, simple, complex landscapes), crop
species, crop varieties (growth form, breeding system, pollinator dependence), abiotic
and biotic variables, and we also included crops in their native and non-native (exotic
crops) range. Some crops were sampled for one year, whereas others were sampled for up
to three years, depending on the funding of each research partner (table S1). Fields were
selected to encompass the environmental and management realities of the different
producers within each crop system. Sampling plots were selected within each field
following the same protocol (18) in all crop systems.
In multiple fields of each of the 33 animal-pollinated crop systems, we measured
flower-visitor density, flower-visitor richness, and crop yield using the same protocol in
landscapes dominated by small- or large-holdings (18). All these variables were
measured in the same plots (50 x 25 m), located in the center of small fields, and halfway
between the center and border of large fields. Given that we measured crop yield in
several entire plants or plots per field subjected to open pollination, our results properly
represent average field conditions and are not biased by resource translocation within the
plants to different flowers (18). The same harvesting method was employed within each
crop system. Our focus on crop yield at a relevant farmer level (kg ha-1) prevents the use
of hand pollination as a way to achieve maximum pollination because it is practically
impossible for most crops to hand pollinate all the flowers of a plant. Furthermore, hand
pollination typically is performed with pure pollen sources from a compatible individual,
with pollen capable of successfully fertilizing the ovum of the female flower. Under
natural conditions, however, pollinators deposit a mix of pollen from various sources,
including the same individual or other individuals of the same variety (30). Therefore,
hand pollination may represent an unattainable goal under natural conditions, leading to
estimates of pollen deficits that are not relevant for natural management or economic
crop production. The pollination treatment to assess deficits was thus performed
indirectly by manipulating the flower-visitor fauna.
Flower-visitor density was measured by scan sampling a fixed number of open
floral units (hereafter “flowers”) in each of four subplots in each field, on at least four
dates during the main flowering period. By using the same protocol, we could express
density directly as no. of visitors in 100 crop flowers, avoiding standardizations to
integrate results from different crop systems. Flower-visitor species richness was
measured by netting all visitors along six 25 m long and 2 m wide transects for
herbaceous crops (or six pairs of adjacent trees for orchard crops) for 5 minutes per
transect. This gives 30 minutes of active net sampling per field, with the clock stopped
each time a captured insect is being handled, which implies at least two hours considering
active sampling plus insect handling, further repeated on at least four dates during the
main flowering period (i.e. 8 hours per field). Taking into account that the flowering
period of most crops lasts only two or three weeks and that researchers need to sample
several sites on the same date when weather is favorable, we consider this to be a high
sampling effort. A few research partners sampled more than six transects per field. In
these cases, we randomly sampled six transects to express the number of species per field
in 30 minutes of net sampling and ensure that all sites were at a comparable level of
We also gathered information on several other potential predictor variables (table
S1), including (a) the level of conventional intensification, a quantitative index ranging
from -3 to 5, constructed as the balance between 5 variables of conventional
intensification each adding 1 to the index (presence of monoculture, synthetic fertilizers,
herbicides, pesticides, and fungicides) and 3 agroecological variables each adding -1 to
the index (presence of polyculture, organic certification, and organic fertilizers); (b)
isolation from semi-natural or natural habitats (log10 km; we classified natural habitats as
in (31)); (c) crop pollinator dependence (%), based on (32), which was updated with
information from pollinator exclusion experiments on local varieties when available; (d)
latitude (decimal degrees); (e) longitude (decimal degrees); (f) baseline level of flower-
visitor density (10th percentile: no. 100 flowers-1); (g) yield gap (10th/90th percentile); and
(h) flower-visitor gap (10th/90th percentile).
Crop yield (log10 kg ha-1) was modeled through a general linear mixed-effects
approach in R software (version 2.15.1, lme4 package, lmer function, Gaussian error
distribution). Mixed-effects models produce similar results to Bayesian hierarchical
models when uninformative priors are employed, especially with large samples, as in our
case (33–36). Fixed-effects included flower-visitor density (no. flower visitors in 100
flowers), flower-visitor richness (no. species in 30 minutes), field size (log10 ha), their
two-way interactions, and their three-way interaction. When a two-way interaction is
significant, it means that both predictor variables have an effect on the response variable.
There is no need for a main effect to be significant. Indeed, as stated throughout
statistical literature (e.g. see pages 718-720 in (37)), when interactions are present,
interpreting main effects in isolation is misleading, as the effect of one predictor on the
response variable depends on the level of the other predictor. Similarly, when a three-way
interaction is present, such in our case (table S3), interpreting two-way interactions or
main effects in isolation could be misleading. Finally, the hierarchical data structure
(fields nested within crop systems) was accounted for by including crop system as a
random-effect. In particular, our model estimated different intercepts and slopes of the
influences of flower-visitor density and richness for each crop system.
Based on the corrected Akaike’s Information Criterion (AICc), we selected the
best model, after evaluating the models resulting from all possible combinations of the
predicting variables (flower-visitor density, flower-visitor richness, and field size) and
their interactions (MuMIn package, dredge function) (38). We found no clear
improvement (lower AICc) when considering curvilinear relations, and therefore we
present only models with linear form. AICc values were obtained from maximum
likelihood estimates of regression coefficients, whereas parameter estimates for final
models were obtained using the restricted maximum likelihood method (39). We also
estimated R2 based on the square of the Pearson's correlation coefficient between
observed and predicted (considering both fixed- and random-effects) crop yields (table
S3). To understand if observed responses in crop yield could be explained by
environmental and management aspects that co-varied with flower-visitor density,
flower-visitor richness, or field size (table S1), we added several co-variables to the
previous mixed-effects model (see main text and Variables section above). We tested the
Gaussian and homoscedasticity assumptions for the standardized residuals of the best
model (39) and found that these assumptions were valid. Furthermore, we found no
evidence of multicollinearity among predictor variables (table S3). We also performed
the analyses with and without three potential outliers (i.e., the most extreme fields in
terms of flower-visitor density) and our results remained the same.
Examples of crop systems sampled in our study. (A) Turnip rape in China (0.12 ha),
(B) Cucumber in Indonesia (0.16 ha), (C) Turnip rape and buckwheat in Nepal (0.30 ha),
(D) Tomato in Brazil (1.05 ha), (E) Apple in Norway (2.58 ha), (F) Oil seed rape in
Brazil (11 ha), (G) Coffee in Brazil (25 ha), (H) Apple in Brazil (43 ha).
Global distribution of the 33 crop systems.
Worldwide, larger holdings (≥ third quartile: 14 ha) have greater dominance of Apis
spp. than smaller holdings (≤ first quartile: 0.5 ha), regardless of species richness.
Small vs. large holdings, and low vs. high richness, are categories only for graphical
purposes, while a mixed-effects model adjusted to the logit transformation of Apis
dominance considered field size and species richness as quantitative variables. This
model included fields nested within crop systems, and random intercept and slopes for
crop systems. The inclusion of a two-way interaction between field size and species
richness did not improve model fit (i.e. lower AICc).
Characteristics of the crop systems sampled. Average values are provided for all variables. Dependence = pollinator dependence. HB =
Crop (variety) Scientific
ha Decimal degrees Index km %
Hedge plants and
flower patches in the
Hedge plants and
flower patches in the
natural areas, density
of HB hives,
natural areas and
density of HB hives
Density of HB hives
natural areas, density
of HB hives,
intensity of forest
intensity of forest
intensity of forest
Red clover seed
sowing of flower
Red clover seed
sowing of flower
Oil seed rape
natural areas and
density of HB hives
natural areas and
Density of Melipona
and HB hives
Density of Melipona
and HB hives
Density of Melipona
and HB hives
Crop yield and flower-visitor density gaps (defined as the difference between 90 and 10th percentiles) observed across fields.
Pollinator deficit is defined as the degree of yield gap that can be reduced by increasing flower-visitor density from the 10th to the 90th
percentile according to our best mixed-effects model with co-variables (see main text, Fig. 1, and table S3). The benefits from
increasing flower-visitor density varies according to field size, flower-visitor richness, and isolation from natural areas (table S3).
Average yields (2012 and 2013) at the world and national levels from FAOSTAT. Some national yields were obtained from other
sources (indicated with asterisk) when FAO's data were not available. NA = data not available.
Crop yield Flower-visitor density Pollinator deficit
gap ratio 90th
no. species 30
min-1 kg ha-1 10th / 90th no. in 100 flowers 10th / 90th kg
% of yield
6.3 NA –
2800* 4887 2717 2170 0.56 4.10 1.40 2.70 0.34 449 21
2.5 13759 –
9277 8423 3323 5100 0.39 11.20 4.32 6.88 0.39 589 12
2.5 33113 –
10081 28095 16300 11795
0.58 10.06 4.78 5.29 0.47
1.6 13759 –
9277 2843 964 1879 0.34 10.22 4.10 6.12 0.40 830 44
12 7 5 0.55 5.83 3.68 2.15 0.63 0,4 7
298 186 111 0.63 5.64 3.29 2.35 0.58 16 14
3.3 5784 –
4500* 6890 2801 4089 0.41 16.72 6.06 10.66
0.36 3209 78
881 476 405 0.54 10.13 5.92 4.21 0.58 229 57
– Nepal –
2.8 975 –
955 599 409 190 0.68 5.36 3.28 2.08 0.61 86 45
3.9 15239 –
6491 49027 17285 31742
0.33 7.09 2.97 4.12 0.42
4.6 33701 –
63734 63000 43500 19500
0.69 1.81 0.32 1.49 0.18 5085 26
3.6 15239 –
6491 45491 14284 31208
0.31 4.68 1.10 3.58 0.24 6694 21
0,70 0,18 0,52 0.26 7.41 4.52 2.90 0.61 -0,2 -37
3.2 33701 –
63734 74200 50000 24200
0.67 0.92 0.23 0.69 0.25 2725 11
2.4 15239 –
7427 50598 12915 37683
0.26 2.13 0.78 1.35 0.37 2073 6
– India –
5.3 NA –
219* 487 243 244 0.50 5.99 4.54 1.45 0.76 46 19
– India -
5.8 NA –
219* 509 244 265 0.48 7.95 4.00 3.95 0.50 115 43
– India -
5.8 NA –
219* 384 219 165 0.57 3.75 2.65 1.10 0.71 30 18
0.1 896 –
1427 1910 799 1111 0.42 8.83 0.49 8.34 0.06 -130 -12
795 485 310 0.61 1.74 0.74 1.00 0.43 10 3
949 456 493 0.48 2.55 1.12 1.42 0.44 20 4
1.8 1939 –
1426 6370 2745 3625 0.43 2.39 1.92 0.47 0.80 16 0
6.2 7870 –
9769 8920 6410 2510 0.72 0.14 0.06 0.08 0.44 112 4
2.4 7870 –
12700 15548 1586 13961
0.10 15.42 8.36 7.06 0.54 2824 20
1.6 809 –
132 51 7 43 0.14 0.71 0.48 0.24 0.67 0 0
1.6 809 –
132 104 32 72 0.31 0.83 0.54 0.29 0.65 -1 -1
1.8 15239 –
33398 15633 5903 9730 0.38 10.20 4.19 6.01 0.41 -2208
1.0 15239 –
33398 39537 21357 18180
0.54 7.69 4.33 3.36 0.56 -2312
1.2 15239 –
33398 40770 20124 20646
0.49 3.42 1.90 1.52 0.56 -257 -1
0.1 896 –
1427 1749 400 1349 0.23 2.43 0.43 2.01 0.17 -254 -19
0.7 783 –
1347* 5464 3042 2423 0.56 0.68 0.10 0.58 0.15 -380 -16
0.2 783 –
1347* 5560 4234 1326 0.76 0.08 0.00 0.08 0.00 -72 -5
8.0 1623 –
1181 1541 763 778 0.50 2.40 0.67 1.73 0.28 942 121
*Alternative sources for national yield averages:
Turnip rape-China: (40)
Cotton-Brazil: (43); WORLD: (44)
† Kg per plant for Agraz because it is not cultivated as a crop but harvested as wild.
Akaike’s Information Criterion (AIC), corrected AIC (AICc), and fixed effects (standard
errors in parentheses) for mixed-effects models of the influences on crop yield (all effects
tested are listed). The best models were derived from comparing AICc values of all
possible combinations of predicting variables with or without co-variables (see methods).
In bold, values for which the 95 % confidence interval do not overlap with zero. The
table shows a significant Density x Richness x Field size interaction, which means that
the three variables are relevant predictors of crop yield. It also implies that two-way
interactions and main effects should be retained in the best model but cannot be
interpreted in isolation. Fixed-effect values and their standard errors for Density,
Richness, and Field size are very similar in models with and without co-variables
showing their independent effects on crop yield. Highest Variance Inflation Factor
(VIFmax) observed across all the variables of each model shows absence of
multicollinearity. The magnitude of the fixed effects cannot be compared among
predictor variables because they are expressed in different units.
Best without co-
Best with co-
Flower-visitor density (no. 100 flowers-1)
Flower-visitor richness (no. species 30 min-1)
Field size (log10 ha)
Density x Richness
Density x Field size
Richness x Field size
Density x Richness x Field size
Pollinator dependence (%)
Latitude (decimal degrees)
Longitude (decimal degrees)
Baseline level of flower-visitor density (10th percentile: no. 100 flowers-1)
Yield gap (%)
Flower-visitors gap (%)
Density x Intensification
Density x Isolation
Density x Pollinator dependence
Density x Latitude
Density x Longitude
Density x Baseline flower-visitor density
Density x Yield gap
Density x Flower-visitors gap
Additional Data table S1 (separate file)
Database_S1.txt: Data used in the analyses of this article.
References and Notes
1. S. K. Lowder, J. Skoet, S. Singh, What do we really know about the number and distribution of
farms and family farms worldwide? Background paper for the State of Food and Agriculture
2014 (Food and Agriculture Organization of the United Nations, Rome, Italy, 2014).
2. P. R. Steward, G. Shackelford, L. G. Carvalheiro, T. G. Benton, L. A. Garibaldi, S. M. Sait,
Pollination and biological control research: Are we neglecting two billion smallholders?
Agric. Food Security 3, 5 (2014). doi:10.1186/2048-7010-3-5
3. M. Herrero, P. K. Thornton, A. M. Notenbaert, S. Wood, S. Msangi, H. A. Freeman, D. Bossio,
J. Dixon, M. Peters, J. van de Steeg, J. Lynam, P. P. Rao, S. Macmillan, B. Gerard, J.
McDermott, C. Seré, M. Rosegrant, Smart investments in sustainable food production:
Revisiting mixed crop-livestock systems. Science 327, 822–825 (2010). Medline
4. R. Chaplin-Kramer et al., Global malnutrition overlaps with pollinator-dependent micronutrient
production. Proc. R. Soc. B Biol. Sci. 281, 20141799 (2014).
5. H. C. J. Godfray, J. R. Beddington, I. R. Crute, L. Haddad, D. Lawrence, J. F. Muir, J. Pretty, S.
Robinson, S. M. Thomas, C. Toulmin, Food security: The challenge of feeding 9 billion
people. Science 327, 812–818 (2010). Medline
6. R. Bommarco, D. Kleijn, S. G. Potts, Ecological intensification: Harnessing ecosystem services
for food security. Trends Ecol. Evol. 28, 230–238 (2013). Medline
7. P. Tittonell, K. E. Giller, When yield gaps are poverty traps: The paradigm of ecological
intensification in African smallholder agriculture. Field Crops Res. 143, 76–90 (2013).
8. J. A. Foley, N. Ramankutty, K. A. Brauman, E. S. Cassidy, J. S. Gerber, M. Johnston, N. D.
Mueller, C. O’Connell, D. K. Ray, P. C. West, C. Balzer, E. M. Bennett, S. R. Carpenter, J.
Hill, C. Monfreda, S. Polasky, J. Rockström, J. Sheehan, S. Siebert, D. Tilman, D. P. Zaks,
Solutions for a cultivated planet. Nature 478, 337–342 (2011). Medline
9. N. D. Mueller, J. S. Gerber, M. Johnston, D. K. Ray, N. Ramankutty, J. A. Foley, Closing yield
gaps through nutrient and water management. Nature 490, 254–257 (2012). Medline
10. D. B. Lobell, K. G. Cassman, C. B. Field, Crop yield gaps: Their importance, magnitudes, and
causes. Annu. Rev. Environ. Resour. 34, 179–204 (2009).
11. L. A. Garibaldi, M. A. Aizen, A. M. Klein, S. A. Cunningham, L. D. Harder, Global growth and
stability of agricultural yield decrease with pollinator dependence. Proc. Natl. Acad. Sci.
U.S.A. 108, 5909–5914 (2011). Medline doi:10.1073/pnas.1012431108
12. L. A. Garibaldi, I. Steffan-Dewenter, R. Winfree, M. A. Aizen, R. Bommarco, S. A.
Cunningham, C. Kremen, L. G. Carvalheiro, L. D. Harder, O. Afik, I. Bartomeus, F.
Benjamin, V. Boreux, D. Cariveau, N. P. Chacoff, J. H. Dudenhöffer, B. M. Freitas, J.
Ghazoul, S. Greenleaf, J. Hipólito, A. Holzschuh, B. Howlett, R. Isaacs, S. K. Javorek, C.
M. Kennedy, K. M. Krewenka, S. Krishnan, Y. Mandelik, M. M. Mayfield, I. Motzke, T.
Munyuli, B. A. Nault, M. Otieno, J. Petersen, G. Pisanty, S. G. Potts, R. Rader, T. H.
Ricketts, M. Rundlöf, C. L. Seymour, C. Schüepp, H. Szentgyörgyi, H. Taki, T. Tscharntke,
C. H. Vergara, B. F. Viana, T. C. Wanger, C. Westphal, N. Williams, A. M. Klein, Wild
pollinators enhance fruit set of crops regardless of honey bee abundance. Science 339,
1608–1611 (2013). Medline
13. L. A. Garibaldi, L. G. Carvalheiro, S. D. Leonhardt, M. A. Aizen, B. R. Blaauw, R. Isaacs, M.
Kuhlmann, D. Kleijn, A. M. Klein, C. Kremen, L. Morandin, J. Scheper, R. Winfree, From
research to action: Enhancing crop yield through wild pollinators. Front. Ecol. Environ. 12,
439–447 (2014). doi:10.1890/130330
14. M. A. Aizen, L. D. Harder, The global stock of domesticated honey bees is growing slower than
agricultural demand for pollination. Curr. Biol. 19, 915–918 (2009). Medline
15. D. Goulson, E. Nicholls, C. Botías, E. L. Rotheray, Bee declines driven by combined stress
from parasites, pesticides, and lack of flowers. Science 347, 1255957 (2015). Medline
16. M. Rundlöf, G. K. Andersson, R. Bommarco, I. Fries, V. Hederström, L. Herbertsson, O.
Jonsson, B. K. Klatt, T. R. Pedersen, J. Yourstone, H. G. Smith, Seed coating with a
neonicotinoid insecticide negatively affects wild bees. Nature 521, 77–80 (2015). Medline
17. L. H. Fraser, H. A. L. Henry, C. N. Carlyle, S. R. White, C. Beierkuhnlein, J. F. Cahill Jr., B. B.
Casper, E. Cleland, S. L. Collins, J. S. Dukes, A. K. Knapp, E. Lind, R. Long, Y. Luo, P. B.
Reich, M. D. Smith, M. Sternberg, R. Turkington, Coordinated distributed experiments: An
emerging tool for testing global hypotheses in ecology and environmental science. Front.
Ecol. Environ. 11, 147–155 (2013). doi:10.1890/110279
18. B. E. Vaissière, B. M. Freitas, B. Gemmill-Herren, Protocol to detect and assess pollination
deficits in crops: a handbook for its use (Food and Agriculture Organization of the United
Nations, Rome, Italy, 2011). www.fao.org/docrep/013/i1929e/i1929e00.htm
19. See supplementary materials on Science Online.
20. M. A. Aizen, L. D. Harder, Expanding the limits of the pollen-limitation concept: Effects of
pollen quantity and quality. Ecology 88, 271–281 (2007). Medline doi:10.1890/06-1017
21. V. Seufert, N. Ramankutty, J. A. Foley, Comparing the yields of organic and conventional
agriculture. Nature 485, 229–232 (2012). Medline doi:10.1038/nature11069
22. T. Tscharntke, A. M. Klein, A. Kruess, I. Steffan-Dewenter, C. Thies, Landscape perspectives
on agricultural intensification and biodiversity – ecosystem service management. Ecol. Lett.
8, 857–874 (2005). doi:10.1111/j.1461-0248.2005.00782.x
23. P. Hoehn, T. Tscharntke, J. M. Tylianakis, I. Steffan-Dewenter, Functional group diversity of
bee pollinators increases crop yield. Proc. R. Soc. B Biol. Sci. 275, 2283–2291 (2008).
24. J. Fründ, C. F. Dormann, A. Holzschuh, T. Tscharntke, Bee diversity effects on pollination
depend on functional complementarity and niche shifts. Ecology 94, 2042–2054 (2013).
25. L. G. Carvalheiro, R. Veldtman, A. G. Shenkute, G. B. Tesfay, C. W. Pirk, J. S. Donaldson, S.
W. Nicolson, Natural and within-farmland biodiversity enhances crop productivity. Ecol.
Lett. 14, 251–259 (2011). Medline doi:10.1111/j.1461-0248.2010.01579.x
26. C. Brittain, N. Williams, C. Kremen, A. M. Klein, Synergistic effects of non-Apis bees and
honey bees for pollination services. Proc. Biol. Sci. 280, 20122767
27. B. J. Cardinale, D. S. Srivastava, J. E. Duffy, J. P. Wright, A. L. Downing, M. Sankaran, C.
Jouseau, Effects of biodiversity on the functioning of trophic groups and ecosystems. Nature
443, 989–992 (2006). Medline doi:10.1038/nature05202
28. M. Schleuning, J. Fründ, D. García, Predicting ecosystem functions from biodiversity and
mutualistic networks: An extension of trait-based concepts to plant-animal interactions.
Ecography 38, 380–392 (2015). doi:10.1111/ecog.00983
29. D. Kleijn, R. Winfree, I. Bartomeus, L. G. Carvalheiro, M. Henry, R. Isaacs, A. M. Klein, C.
Kremen, L. K. M’Gonigle, R. Rader, T. H. Ricketts, N. M. Williams, N. Lee Adamson, J. S.
Ascher, A. Báldi, P. Batáry, F. Benjamin, J. C. Biesmeijer, E. J. Blitzer, R. Bommarco, M.
R. Brand, V. Bretagnolle, L. Button, D. P. Cariveau, R. Chifflet, J. F. Colville, B. N.
Danforth, E. Elle, M. P. Garratt, F. Herzog, A. Holzschuh, B. G. Howlett, F. Jauker, S. Jha,
E. Knop, K. M. Krewenka, V. Le Féon, Y. Mandelik, E. A. May, M. G. Park, G. Pisanty, M.
Reemer, V. Riedinger, O. Rollin, M. Rundlöf, H. S. Sardiñas, J. Scheper, A. R. Sciligo, H.
G. Smith, I. Steffan-Dewenter, R. Thorp, T. Tscharntke, J. Verhulst, B. F. Viana, B. E.
Vaissière, R. Veldtman, C. Westphal, S. G. Potts, Delivery of crop pollination services is an
insufficient argument for wild pollinator conservation. Nat. Commun. 6, 7414 (2015).
30. J. D. Thomson, Using pollination deficits to infer pollinator declines: Can theory guide us?
Conserv. Ecol. 5, 6 (2001).
31. L. A. Garibaldi, I. Steffan-Dewenter, C. Kremen, J. M. Morales, R. Bommarco, S. A.
Cunningham, L. G. Carvalheiro, N. P. Chacoff, J. H. Dudenhöffer, S. S. Greenleaf, A.
Holzschuh, R. Isaacs, K. Krewenka, Y. Mandelik, M. M. Mayfield, L. A. Morandin, S. G.
Potts, T. H. Ricketts, H. Szentgyörgyi, B. F. Viana, C. Westphal, R. Winfree, A. M. Klein,
Stability of pollination services decreases with isolation from natural areas despite honey
bee visits. Ecol. Lett. 14, 1062–1072 (2011). Medline doi:10.1111/j.1461-
32. A. M. Klein, B. E. Vaissière, J. H. Cane, I. Steffan-Dewenter, S. A. Cunningham, C. Kremen,
T. Tscharntke, Importance of pollinators in changing landscapes for world crops. Proc. Biol.
Sci. 274, 303–313 (2007). Medline doi:10.1098/rspb.2006.3721
33. R Development Core Team, R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria (2013); available at www.r-
34. A. Gelman, J. Hill, Data analysis using regression and multilevel/hierarchical models
(Cambridge Univ. Press, Cambridge, UK, 2007).
35. D. Bates, M. Maechler, B. Bolker, S. Walker, lme4: Linear mixed-effects models using Eigen
and s4 (2014), (available at http://cran.r-project.org/package=lme4).
36. S. S. Qian, T. F. Cuffney, I. Alameddine, G. McMahon, K. H. Reckhow, On the application of
multilevel modeling in environmental and ecological studies. Ecology 91, 355–361 (2010).
37. D. R. Anderson, D. J. Sweeney, T. A. Williams, Statistics for business and economics IIe
(South-Western Cengage Learning, Mason, OH, 2011).
38. K. Bartoń, MuMIn: Multi-model inference. R package version 1.10.0 (2014); available at
39. A. F. Zuur, E. N. Ieno, N. J. Walker, A. A. Saveliev, G. M. Smith, Mixed effects models and
extensions in ecology with R (Springer, New York, 2009).
40. S. Wang, Effects of climate change and management practices on rapeseed production in
Australia and China (Northwest A&F University, Australia, 2014).
41. IICA, Desarrollo territorial con enfoque de sistemas agroalimentarios localizados (AT-SIAL):
La Comarca Andina del paralelo 42, Argentina (Instituto Interamericano de Cooperación
para la Agricultura (IICA), Mexico, 2013).
42. B. A. Gudade, P. Chhetri, U. Gupta, T. N. Deka, K. Vijayan, Traditional practices of large
cardamom cultivation in Sikkim and Darjeeling. Life Sci. Leaflets 9, 62–68 (2013).
43. CONAB, Levantamentos de safra (2015); available at
44. U.S. Department of Agriculture, Cotton: World markets and trade (2015); available at