Po

I

a

b

RA

A

sfi(calcbpn

Z

aaRbnusb

h1

Basic and Applied Ecology 16 (2015) 737–745

ollinators, pests and soil properties interactively shapeilseed rape yield

gnasi Bartomeusa,b,∗, Vesna Gagica, Riccardo Bommarcoa

Swedish University of Agricultural Sciences, Department of Ecology, SE-75007 Uppsala, SwedenEstación Biológica de Donana (EBD-CSIC), Dpto. Ecología Integrativa, ES-41092 Sevilla, Spain

eceived 24 November 2014; accepted 23 July 2015vailable online 31 July 2015

bstract

Pollination, pest control, and soil properties are well known to affect agricultural production. These factors might interactivelyhape crop yield, but most studies focus on only one of these factors at a time. We used 15 winter oilseed rape (Brassica napus L.)elds in Sweden to study how variation among fields in pollinator visitation rates, pollen beetle attack rates and soil propertiessoil texture, pH and organic carbon) interactively determined crop yield. The fields were embedded in a landscape gradient withontrasting proportions of arable and semi-natural land. In general, pollinator visitation and pest levels were negatively correlatednd varied independently of soil properties. This may reflect that above- and below-ground processes react at landscape andocal spatial scales, respectively. The above-ground biotic interactions and below-ground abiotic factors interactively affectedrop yield. Pollinator visitation was the strongest predictor positively associated with yield. High soil pH also benefited yield,ut only at lower pest loads. Surprisingly, high pest loads increased the pollinator benefits for yield. Implementing managementlans at different spatial scales can create synergies among above- and below-ground ecosystem processes, but both scales areeeded given that different processes react at different spatial scales.

usammenfassung

Bestäubung, Schädlingskontrolle und Bodeneigenschaften beeinflussen die Agrarproduktion. Diese Faktoren könnten inter-gierend den Ernteertrag beeinflussen, aber die meisten Studien konzentrieren sich auf nur einen Faktor. Wir untersuchtenuf 15 Winterrapsfeldern (Brassica napus L.) in Schweden, wie die von Feld zu Feld variierenden Bestäuberbesuchsraten,apsglanzkäfer-Befallsraten und Bodeneigenschaften (Bodentextur, pH, organischer Kohlenstoff) wechselwirkend den Ertrag

estimmten. Die Felder repräsentierten einen Landschaftsgradienten mit unterschiedlichen Anteilen von Agrarflächen undaturnahen Arealen. Allgemein waren Bestäuberbesuch und Schädlingsbefall negativ miteinander korreliert, und sie variiertennabhängig von den Bodeneigenschaften. Dies könnte anzeigen, dass oberirdische Prozesse und Prozesse im Boden auf der Land-chaftsebene bzw. der lokalen Ebene reagieren. Die oberirdischen biotischen Interaktionen und die abiotischen Bodenfaktoren eeinflussten wechselwirkend den Ertrag. Der Bestäuberbesuch war der stärkste positiv mit dem Ertrag verknüpfte Faktor.

∗Corresponding author at: Estación Biológica de Donana (EBD-CSIC), Dpto. Ecología Integrativa, ES-41092 Sevilla, Spain. Tel.: +34 954466700.E-mail address: nacho.bartomeus@gmail.com (I. Bartomeus).

ttp://dx.doi.org/10.1016/j.baae.2015.07.004439-1791/© 2015 Gesellschaft für Ökologie. Published by Elsevier GmbH. All rights reserved.

7

Ehaf©

K

I

iaottcnis

p((taloITs&ttPccah1

tttacmas(eAm

38 I. Bartomeus et al. / Basic and Applied Ecology 16 (2015) 737–745

in hoher pH-Wert begünstigte ebenfalls den Ertrag, aber nur bei geringem Schädlingsbefall. Überraschenderweise, steigerteoher Schädlingsbefall die positive Wirkung des Bestäuberbesuchs auf den Ertrag. Das Aufstellen von Bewirtschaftungsplänenuf unterschiedlichen räumlichen Skalen kann Synergien zwischen oberirdischen und unterirdischen Ökosystemprozessenreisetzen, aber beide Skalen werden benötigt, da unterschiedliche Prozesse auf unterschiedlichen Skalen reagieren.

2015 Gesellschaft für Ökologie. Published by Elsevier GmbH. All rights reserved.

es; Pol

ai

atttGCKWeaosaetagw(winB

atlpp(ncelpctot

eywords: Ecosystem services; Above- and below-ground process

ntroduction

Future agriculture needs to be productive to sustain thencreasing human population, while conserving biodiversitynd the environment. A suggested solution is to stabilizer increase crop yields by maximizing the use of ecosys-em services provided by biodiversity, thereby decreasinghe dependence on external inputs of agrochemicals in agri-ulture (Bommarco, Kleijn, & Potts 2013). For this, weeed to resolve how different biotic and abiotic processesnteractively shape yield, something that is poorly under-tood.

Crop pollination is a key ecosystem service that sup-orts crop yield quantity (Garibaldi et al. 2013) and qualityBartomeus et al. 2014) in three quarters of all crop speciesKlein et al. 2007). In contrast, herbivory by pest insectsypically reduces yields in all major crops by 5–15% on aver-ge (Oerke & Dehne 2004), and in individual cases yieldosses can be far higher (e.g., pollen beetle yield losses inilseed rape fields may reach up to 80%, Nilsson 1987).n addition, several soil properties affect crop production.here is solid evidence from agricultural trials showing thatoil texture is associated with water retention (Rawls, Gish,

Brakensiek 1991). Soil organic carbon (SOC) increaseshe stability of several soil properties, such as increasinghe soil cation-exchange capacity (Campbell 1978; Johnston,oulton, & Coleman 2009; Lal 2010, 2011). Soil pH islosely linked to biological activity of below-ground soilommunities and positively related to nutrient availabilitynd soil fertility (Foth & Ellis 1997), which may translate toigher crop yield (Dick 1992; Barszczak, Barszczak, & Foy993).

Despite the widely acknowledged importance of pollina-ion, pest herbivory and soil properties for shaping yield,he information we have on the joint effects of these fac-ors on yields is fragmentary at best. Hence processes above-nd below-ground are generally studied in isolation, and itsontribution to plant growth and crop yield can only be esti-ated if we consider them to be additive. However, this

ssumption has been challenged in small-scale experimentshowing complex interactions between both compartmentsVan der Putten, Vet, Harvey, & Wäckers 2001; Bezemer

t al. 2005; Barber, Adler, Theis, Hazzard, & Kiers 2012).bove- and below-ground communities can be powerfulutual drivers, with both positive and negative feedbacks

bAs

lination; Pollen beetles; Oilseed rape; Soil organic carbon; pH

mong them (Strauss & Irwin 2004; Wardle et al. 2004). Suchnteractions remain unexplored for larger spatial scales.

For instance, pollination has most often been studied as context-independent process, but recent findings suggesthat pollination success and resulting crop yield are linkedo other factors such as fruit predation, irrigation or fer-ilization (Bos et al. 2007; Boreux, Kushalappa, Vaast, &hazoul 2013; Lundin, Smith, Rundlöf, & Bommarco 2013;lassen et al. 2014; Klein, Hendrix, Clough, Scofield, &remen 2015; Motzke, Tscharntke, Wanger, & Klein 2014;ielgoss et al. 2015). For example, Lundin et al. (2013)

xperimentally showed that pollination and pest control of seed predator interact synergistically. The combined effectf these ecosystem services on the yield was higher than theum of the isolated services. Local crop management canlso interact synergistically with pollination. There is recentvidence that irrigation positively affects the net benefit tohe plant from pollinators in two contrasting crops, coffeend almond (Boreux et al. 2013; Klein et al. 2015). Moreenerally, it is expected that below-ground soil properties, asell as related ecosystem services provided by soil organisms

Wagg, Bender, Widmer, & van der Heijden 2014), changeater retention and nutrient assimilation, and hence should

nterplay with above-ground biotic interactions such as polli-ation and pest damage in deciding the yield (e.g., Williams,irkhofer, & Hedlund 2014).Most evidence about interactive effects on yield from

bove- and below-ground processes comes from experimen-al studies (e.g., Barber et al. 2012; Lundin et al. 2013). Weack detailed data on how crop yield is affected by multi-le processes at the scales at which crop cultivation takeslace–in the arable field and in the surrounding landscapebut see Boreux et al. 2013). For example, pollinators andatural enemies to crop pests are both affected by landscapeomposition at scales up to several kilometers (Shackelfordt al. 2013), whereas soil properties are mostly affectedocally by management of the individual arable field. Hence,olicy-relevant assessments of ecosystem services in agri-ultural landscapes cannot rely on the simple assumptionhat a certain land-use results in a given service supply. Notnly local field management, but also the composition ofhe surrounding landscape is an important determinant of

iodiversity and ecosystem services (Gabriel et al. 2010).ttempts to maximize the production of a single ecosystem

ervice can result in substantial declines in the provision of

Applied

oR

Lplraitiosn

M

S

etaaranaafiT2ba(caa3

wab8rmnir(

S

b

qtFapottivabbt(vss

tAtttfl

1A5IdbiI(

bptpfiastlotisrfr

I. Bartomeus et al. / Basic and

ther ecosystem services (Bennett, Peterson, & Gordon 2009;audsepp-Hearne, Peterson, & Bennett 2010).Here, we use fifteen winter oilseed rape (Brassica napus

.) fields situated in a landscape gradient with contrastingroportions of arable and semi-natural land to study naturalevels of variation in pollinator visitation rates, pest attackates and soil properties. Rather than testing specific mech-nism by which this above and below-ground componentsnteract, we aim to describe the observed patterns and assesshe relative importance of each factor for yield formationn an important field crop, as well as potential interactionsccurring among them. These results will generate morepecific hypothesis for the mechanisms underlying possibleon-additive effects on yield.

aterials and methods

tudy sites

Fifteen conventional winter oilseed rape (B. napus, vari-ties Excalibur and Compass) fields were selected in 2013 inhe Västergötland region (see Appendix A: Fig. 1), Sweden,long a landscape gradient with contrasting proportions ofrable and semi-natural land. Both varieties were well rep-esented along the landscape gradient. All sites were locatedt least 3 km apart from each other. Västergötland is domi-ated by arable land, mainly cereals, and woodlands, with

small fraction of pastures and meadows. Percentage ofrable land was used as a proxy of agricultural intensi-cation (Steffan-Dewenter, Munzenberg, Burger, Thies, &scharntke 2002; Thies, Steffan-Dewenter, & Tscharntke003; Fahrig 2013) and was measured at multiple scales (seeelow) using information on land-use characteristics avail-ble from the Integrated Administration and Control SystemIACS), a data base developed by the Swedish Board of Agri-ulture. The landscape gradient ranged from 20% to 80% ofrable land in all radii considered. In each field we sampledn area not sprayed with insecticides of 40 × 70 m, situated0 m from the edge into the field to avoid edge effects.Oilseed rape is partially self-compatible, and can set fruit

ithout pollinators, but pollinators increase yield quantitynd quality (Bartomeus et al. 2014). Its main pest is the polleneetle (Meligethes aeneus F.), that can cause yield losses up to0% (Nilsson 1987; Alford, Nilsson, & Ulber, 2003). Oilseedape has high demands of phosphorous and sulphur, which isodulated by soil pH, which affects the availability of many

utrients to plants. For example, at low pH-values sulphurs less available for plants. Positive relationships of oilseedape yield with pH have been shown for two Swedish regionsMattsson 2008).

ampling

Pollinator abundance was sampled twice during peakloom. For each site and round, we established three 0.5 m2

fpdn

Ecology 16 (2015) 737–745 739

uadrats randomly placed along a 50 m transect centered inhe non-sprayed area, parallel to its length (see Appendix A:ig. 1). We observed each quadrat for 5 min and recordedll pollinators visiting flowers. To record a flower visitor as aollinator, the insect had to have contact with the central partsf the flower, i.e., the anthers or stigma. We counted pollina-ors once regardless of how many flowers they visited. Sincehere is no way of telling whether an observed pollinator vis-ted a patch for the first time or had returned for a secondisit, pollinators may have been recorded twice occasion-lly. Insects were assigned to one of the following categoriesy visual inspection: Honey bee (Apis mellifera L.), bum-le bees (Bombus sp.), wild bees (diverse species, mostly inhe genus Andrena), hoverflies (Syrphidae) and other speciesmostly Diptera, Hymenoptera and Lepidoptera). All obser-ations were done by a single observer. Pollinators were onlyampled on days with sun or scattered clouds and at windpeeds <15 km/h.

Pollen beetles were counted at four sampling plots alonghe central transect and 10 m apart from each other (seeppendix A: Fig. 1). Adult pollen beetles were counted on

en plants at each sampling plot (i.e., on 40 plants per field inotal). Counts were done three times in the season betweenhe pollen beetle colonization in the green bud stage and untilowering was over.To measure soil properties, we collected five random

5 cm deep soil cores (6 cm diameter) at each site (seeppendix A: Fig. 1). Cores were mixed and transported at◦C and protected from sunlight. We determined pH (SS-

SO 10390), proportion of soil organic carbon (SOC) afterry combustion (SS-ISO 10694) and soil texture, measuredy determination of percent clay and percent sand particlesn mineral soil material after sieving and sedimentation (SS-SO 11277). All soil analyses were done by Agrilab, Uppsalahttp://www.agrilab.se).

Yield was measured as total seed weight per plant justefore harvesting. Number of pods was counted on 5 plantser plot, using the same four plots as used for pollen bee-les counts (i.e., 20 plants per field). Number of seedser pod was counted on 20 pods randomly chosen fromve plants at each sampling plot (80 pods per field) andveraged per plot. Weight of 100 seeds from randomlyelected pods was measured three times per sampling ploto obtain the mean weight of one seed. Yield was calcu-ated at the plant level as pods per plant × mean numberf seeds per pod × mean weight of one seed. We estimatedotal crop yield as weight of seed obtained per plant, becauset integrates fruit quality and seed set. Yield as weight ofeeds per plant is moderately correlated with yield harvesteported by farmers in kg/ha (Pearson r = 0.48, n = 9 fieldsor which we can contact farmers). Note that our resultsefer to biologically relevant changes in yield per plant, but

or management decision-making further work and morerecise estimation of the relation between interactions ofifferent ecosystem services and area-based yield will beeeded.

7 Applied

S

t(outaaiapshdim

vlc2rw

gmapphnamot(Wt&pwiaaaA(peAamhe

Vtm(Ia

wcuwtefiwsso

n

R

ddhtv7S

eaa(iamavoicc

gpovr

40 I. Bartomeus et al. / Basic and

tatistical analysis

To understand how different above and below-ground fac-ors covary, we first present a principal component analysisPCA) which defines an orthogonal coordinate system thatptimally describes the variance in our data and that wassed to visually represent synergies and trade-offs amonghe variables. Visitation of each pollinator guild, total pestbundance and the three soil properties measured (pH, SOC,nd soil texture measured as clay and sand percentage) werencluded in the analysis. We only used one data point per sitend variable measured by summing the total number of visitser pollinator guild, or total number of pests across plots andampling rounds per site. All variables were scaled before-and to allow meaningful comparisons among variables withifferent units. In addition, we present in the supplementarynformation pairwise Pearson correlations among all factorseasured (see Appendix A: Text 1 and Table 1).Furthermore, we explored at which landscape scale each

ariable individually responded to the percentage of arableand. Each of the above variables was regressed against per-entage agricultural land at increasing radius ranging from50 m to 3 km using linear models at the field scale. Theadius at which most variation was explained by the modelas selected based on maximized r2 values.Second, we assessed the influence of the above- and below-

round factors on crop yield. We used general mixed effectsodels with crop yield per plant as the response variable

nd total pollinator visits, pest levels and soil properties asredictors. For each soil property, we had only one estimateer field. Percentage of sand was not included because it isighly correlated with clay percentage. We pooled all polli-ator visits per field; pollinators move freely among plants,nd the total visitation abundance in a field is a relevanteasure to relate to yield. To avoid over-parameterization

f the statistical models, we pooled all guilds and analyzedotal visitation as it provides a reliable proxy of pollinationVazquez, Morris, & Jordano 2005; Garibaldi et al. 2013).

e used pollen beetle counts per plot because pollen bee-les are less mobile and can be patchily distributed (Williams

Ferguson 2010). Finally, we used yield measured in fivelants in each plot. Hence, in all models, “plot” nestedithin “field” was included as random factor. The full model

ncluded the total pollinator visits, the pest counts per plot,nd the three soil properties (pH, SOC and clay percent as

measure of texture). We included all pairwise interactionsnd selected the best models based on AICc (Burnham &nderson 2002) using the dredge function in package MuMin

Bartón 2014). We limited the amount of parameters includeder model to five to avoid over-parameterization of the mod-ls. We averaged among models within two AICc points.ll variables were centered beforehand to make the inter-

ctions easier to interpret (Cleasby & Nakagawa 2011). All

odels were visually inspected for normality of errors and

eteroscedasticity. We checked for collinearity in the mod-ls by estimating the variance inflation factors (VIF). All

(a3r

Ecology 16 (2015) 737–745

IFs were below 3, hence, there was no strong collinearity inhe models. We tested for spatial autocorrelation of the best

odel residuals using Moran’s index in the ‘spdep’ packageBivand et al. 2012). Spatial autocorrelation was low (Moran = −0.07) and hence was not taken into account in furthernalysis.

Although nitrogen may have important effects on yield,e did not take it into account in our main analysis. We

ould only retrieve information on total nitrogen fertilizerssed per field in 9 farms. To test whether our results changehen N fertilization is included, in a separate analysis we

ested the effect of nitrogen fertilization and all the above-xplained predictors in crop yield per plant for this subset ofelds. In addition to the above-described crop yield model,e present in the appendix equivalent models with the three

ubcomponents of yield (number of pods, seeds per pod andeed weight) as response variables. This allows an evaluationf the relative importance of each separate yield component.

All analyses were done in R, using the base package andlme (Pinheiro, Bates, DebRoy, & Sarkar 2014).

esults

All fields where visited by a diverse pollinator communityominated by bees and hoverflies. Total pollinator abun-ances ranged from 4 to 12.6 visits per square meter perour across fields. Pollen beetle abundance ranged from 14o 173 individuals (mean 48 ± 11) per plot. Soil propertiesaried considerably, with pH values ranging between 5.6 and.6, clay percentage values ranging between 1% and 30% andOC ranging between 1.4% and 4.2%.The PCA showed that, overall, sites with lower pest lev-

ls tended to have more pollinators, and that those variablesre independent of soil properties (Fig. 1). The first twoxes of the PCA explained together 55% of the variance31% and 24% respectively), with subsequent axes explain-ng less than 15% each. We found a trade-off between pestsnd pollinators, with sites with lower pest levels, havingore pollinators. The less abundant wild bees and hoverflies

re independent of honeybee and bumblebee visits, and co-ary in opposite directions among them. Along axis 1, totalrganic carbon and pH have high loadings and sand contents represented in both principal components. As expected,lay content follows an opposite trend compared with sandontent.

To further explore this pattern of variation, we investi-ated at which scale each variable responded. As expected,ollinators in general responded negatively to percentagef agriculture in the landscape (estimate of total pollinatorisits at 3000 m radius = −0.07 ± 0.03, p = 0.03), but guildsesponded, albeit non-significantly, at contrasting scales

wild bees at 250 m radius = 0.01 ± 0.01, p = 0.22; hoverfliest 2000 m radius = −0.02 ± 0.01, p = 0.22; bumblebees at000 m radius = -0.03 ± 0.02, p = 0.25; honeybees at 2500 madius = −0.07 ± 0.04, p = 0.09); with wild bees responding

I. Bartomeus et al. / Basic and Applied Ecology 16 (2015) 737–745 741

Fig. 1. Covariation patterns between different above and below-ground factors affecting yield. The first component of thePCA shows a trade off between main pollinators and pests,while the second component describes mainly soil proper-ties. PCA loadings: honey bee (PC1 = −0.28, PC2 = 0.63),wild bees (PC1 = −0.49, PC2 = 0.03), hoverflies (PC1 = 0.93,PC2 = 0.03), bumble bees (PC1 = 0.06, PC2 = 0.62), other polli-nators (PC1 = 0.32, PC2 = −0.07), pollen beetles (PC1 = −0.55,PC2 = −0.76), SOC (soil organic matter; PC1 = 0.61, PC2 = −0.19),pH (PC1 = 0.72, PC2 = −0.06), clay percent (PC1 = −0.64,P

wetbattNo

epacpscb(ot02wt

Fig. 2. Big bees and pests respond to landscape configurations atlarge scales. Explanatory power of percent of agriculture in the land-scape at different scales for (A) pollinators (honey bee in black, wildbees in blue, bumble bees in red, hoverflies in green and the overallresponse in gray), (B) pollen beetles and (C) soil properties (totalorganic carbon in black, pH in red and percentage of clay in blue).(t

ft

twere affected by the interaction between pH and pest lev-

C2 = 0.63), sand percent (PC1 = 0.34, PC2 = −0.79).

eakly and at a very small radius, while bumblebees and hon-ybees responded at radii up to 2.5–3 km (Fig. 2A). Overall,otal pollinator visits peaked at 3 km radius because honey-ees and bumblebees are more abundant than the wild beesnd drive the overall response. Pollen beetles responded posi-ively to percent agriculture at a scale of 2.5 km (Fig. 2B), buthe trend was not significant (estimate = 4.1 ± 2.29, p = 0.09).one of the soil properties were affected by the percentagef arable land at any scale (Fig. 2C; all models p > 0.2).When analyzing the effect on yield, we found seven mod-

ls within two AICc points (see Appendix A: Table 2) withollinators, pH and pests retained in most models. The aver-ged model (Table 1) shows that pollinators are positivelyorrelated with yield and that there is an interaction with theest. At high pest numbers, the relationship with pollinators isteeper (Fig. 3A). This interaction should be interpreted withare, given that there are few data points with high levels ofoth, and because they are weakly, but negatively correlatedVIF < 3). Interestingly, pH had a positive effect on yield, butnly when pest levels were low. At high pest levels, the rela-ionship disappeared (Fig. 3B). The best model marginal r2 is.20, while the conditional r2 is 0.55 (Nakagawa & Schielzeth012). Including nitrogen fertilization for the subset of fields

ith this information does not change the results showing

hat pollinators and their interaction with pests are the mainem

For interpretation of the references to color in this figure legend,he reader is referred to the web version of this article.)

actors driving yield. Nitrogen itself is not significantly linkedo yield, but it interacts with pH (Table 2).

The different subcomponents of yield show that pollina-ors are most important for fruit set, while seeds per pod

ls. Interestingly SOC, a variable not retained in the overallodel, is also positively correlated with seeds per pod. Finally

742 I. Bartomeus et al. / Basic and Applied Ecology 16 (2015) 737–745

Table 1. Pollinators are the most important factor affecting crop yield. Model-averaged coefficients of the model predicting oilseed rapeyield. The relative importance indicates the proportion of models containing each predictor; “pollinators” was the only variable retained inall models.

Relative variable importance Estimates Std. error z-Value p-Value

Pests 0.34 0.01 0.01 1.2 0.23pH 0.79 0.65 0.39 1.53 0.13Pollinators 1 0.16 0.06 2.33 0.02Pests × pH 0.23 −0.02 0.02 2.09 0.04Pests × pollinators 0.23 0.004 0.002 2.09 0.04SOC 0.23 0.46 0.39 1.05 0.29pH × pollinators 0.11 0.11 0.14 0.72 0.47

Fig. 3. Pollinators and soil pH interact with pests in their effect on yield, such that (A) the pollination benefit increases with pest attack and(B) yield increases with pH, but only at low pest attacks. Black lines are estimate predictions for the average level of pests. Red lines arepredictions for low (10% quantile) and blue lines for high (90% quantile) levels of pests respectively. (For interpretation of the references tocolor in this figure legend, the reader is referred to the web version of this article.)

Table 2. Model-averaged coefficients of the model predicting oilseed rape yield for a subset of 9 fields for which we have nitrogen fertilizerinformation. The relative importance indicates the proportion of models containing each predictor.

Relative variable importance Estimates Std. error z-Value p-Value

Pests 1.00 0.02 0.01 1.78 0.07pH 0.45 14.72 3.92 3.07 0.002Pollinators 0.55 0.18 0.05 2.87 0.004N 0.82 −0.003 0.01 0.19 0.84P 0.N −0.

sp

D

ofhpa

yicri(oe

ests × pollinators 0.55

:pH 0.45

eed weight is mainly affected by the interaction betweenollinators and pests (see Appendix A: Table 3).

iscussion

Crop yield is shaped by combinations of biotic and abi-tic factors. Identifying the main above- and below-ground

actors that ensure high yields requires an examination ofow they naturally co-vary in the landscape. We show thatollinators, pest levels and soil properties (mainly soil pHnd nitrogen fertilization) are key factors determining the

twtr

007 0.003 2.45 0.0109 0.02 3.06 0.002

ield in winter oilseed rape. These have been independentlydentified to affect yield in a number of crops, but they oftenorrelate weakly with yield. For instance, even if there is aobust general trend that yield increases with pollinator vis-tation, there remains a great deal of unexplained variationGaribaldi et al. 2013). Sites with similar pollinator levelsften differ substantially in yield. Studies addressing severalcosystem services and abiotic factors simultaneously have

he potential to explain more of this variation. Importantly,e show that such factors can interact, thereby modifying

he outcome of the main effects. Our study thereby adds toecent experimental evidence that the response of yield to one

Applied

ftaeppa

lnsbtpieb(esai

tepbtorbgttF1bta

p(iIdltbf1hsa

pc

yfbc2

etaaiasitpmFwpp(d2isfeibtaa

fTaf(tsgtioy

A

a

I. Bartomeus et al. / Basic and

actor or resource, such as pollinators, depends on other fac-ors, such as pest control levels, and that their effects are notdditively contributing to yield (Lundin et al. 2013). How-ver, in our dataset, even after accounting for pollinator visits,est attack rates and several soil properties, the fixed factorsredict only 20% of the variance, while the random factorsssociated with unmeasured field variables explain up to 55%.

We identified that pollen beetles and the most abundant pol-inators (i.e., honey bees and bumble bees) naturally co-variedegatively. This is partially explained by the landscape analy-is, as both respond to percent of arable land at similar scales,ut in opposite directions (although the trend for pollen bee-les is non-significant). One explanation for this pattern is thatollinators respond positively to an increased amount of feed-ng and nesting resources in complex landscapes (Kennedyt al. 2013), and that pollen beetle abundances are suppressedy natural enemies that are benefited by such landscapesChaplin-Kramer, O’Rourke, Blitzer, & Kremen 2011). How-ver, given that pollen beetles feed on flower buds and aretill active on flowers during the pollination period, they maylso have a direct effect by deterring pollinators from heavilynfested fields.

Interestingly, we show an interaction between pollen bee-les and pollinators in their impact on yield. Contrary toxpected, at the same pollinator visits level, the pollinators’ositive effect on yield is higher when abundances of polleneetles are high. Hence, rather than pollen beetles loweringhe visitation efficiency (e.g., by reducing pollen availability)r directly damaging the plant (e.g., increasing fruit abortionates; Alford et al. 2003), it seems that the observed polleneetle damage to buds results in considerable compensatoryrowth by oilseed rape. For example, it has been reportedhat moderate feeding damage to the terminal raceme leadso increased production of new side racemes (Williams &ree 1979; Tatchell 1983; Lerin 1987; Axelsen & Nielsen990). It is interesting that this compensatory growth is onlyeneficial under high pollinator abundance, and may indicatehat the benefit may only arise if the newly produced branchesre well pollinated.

We also show that soil properties vary across sites inde-endently to the proportion of arable land in the landscapeWilliams & Hedlund 2013). Soil pH seems to be the mostmportant soil factor for explaining yield in our analyses.nterestingly, the positive effect of soil pH on yield is onlyetectable at low pest levels. This implies that at high pestevels, the benefits from increasing pH and thereby soil fer-ility do not translate to increased yield, but may insteade lost to pest damage or invested into plant defenses. Inact, soil fertility can increase plant defenses (Coley et al.985) and we found that fields with a high pH tended toave rather low pest levels. This pattern was weak, but isuggested by the PCA analysis (Fig. 1) and deserves future

ttention.

Surprisingly, soil texture (i.e., proportion clay), which isositively related to water retention and nutrient exchangeapacity, was not retained in any of the best models explaining

tepF

Ecology 16 (2015) 737–745 743

ield. This indicates that water was probably not a limitingactor in this year and region. However, clay contents maye important in years with low precipitation, and for otherlimatic regions or crops (see Boreux et al. 2013; Klein et al.015).

Our results support recent claims that interactions amongcosystem services are to be expected, but the importance ofhe key above- and below-ground variables affecting yieldnd their interactive effects are likely to be crop-specificnd to vary between sites and years. Our results are notntended to be used for predictive purposes, given that theyre based on a limited number of fields, but will allow testingpecific hypothesis behind the mechanisms generating thenteractions reported here for oilseed rape. Moreover, withhe accumulation of more case studies on a variety of crop-ing systems and ecosystems, we may be able to reach aore general understanding of interactive effects on yield.or example, the degree of plant dependency on pollinatorsill determine the potential benefit that can be achieved byollinators. However, even in plants with high rates of self-ollination, yield quality is enhanced with insect pollinationBartomeus et al. 2014). Herbivores that affect the repro-uctive parts of the plant, such as seed weevils (Lundin et al.013) or pollen beetles (this study) are more likely to directlynteract with the benefits from pollinators. Herbivore plantuckers or defoliators can be nutrient sinks that affect fruitormation, even when sufficient pollination is achieved (Bost al. 2007). Plant species-specific pathways to absorb, assim-late and mobilize nutrients will determine how above- andelow-ground factors interact. For example, coffee planta-ions can trigger one or two flowering peaks a year clearlyffecting pollinator responses, and this depends on nutrientnd water availability (Boreux et al. 2013).

The strength and shape of the relationships between dif-erent above- and below-ground processes is poorly known.his is partly because we lack information about synergiesnd trade-offs in the management of multiple processes andew studies have simultaneously considered effects of localfield) and landscape scale land use on multiple ecosys-em functions (Bianchi, Booij, & Tscharntke 2006). Wehow that interactions between biotic and abiotic factors canive rise to scale-dependent synergies when managing mul-iple ecosystem services. Hence, both above-ground bioticnteractions regulated at large scales and below-ground abi-tic factors managed at local scales interact to form cropield.

cknowledgments

Field work was conducted by Oskar R. Rubbmark, Ger-rd Malsher and Laura Riggi. Audrey St-Martin helped with

he soil samples. Ola Lundin and three anonymous review-rs provided valuable comments on an earlier draft of thisaper. Funding was provided by the Swedish research councilORMAS.

7 Applied

A

fj

R

A

A

B

B

B

B

B

B

B

B

B

B

B

B

C

C

C

C

C

D

F

F

G

G

J

K

K

K

L

L

L

L

M

44 I. Bartomeus et al. / Basic and

ppendix A. Supplementary data

Supplementary data associated with this article can beound, in the online version, at http://dx.doi.org/10.1016/.baae.2015.07.004.

eferences

lford, D. V., Nilsson, C., & Ulber, B. (2003). Biocontrol of oilseedrape pests. Insect pests of oilseed rape crops. Oxford, UK: Black-well Science.

xelsen, J., & Nielsen, P. (1990). Compensation in spring sownoilseed rape after attack by pollenbeetles (Meligethes aeneusF.). Tidsskrift Planteavl, 195–199.

arber, N. A., Adler, L. S., Theis, N., Hazzard, R. V., & Kiers,E. T. (2012). Herbivory reduces plant interactions with above-and belowground antagonists and mutualists. Ecology, 93,1560–1570.

arszczak, Z., Barszczak, T., & Foy, C. D. (1993). Effect of mois-ture, nitrogen rates, and soil acidity on seed yield and chemicalcomposition of winter oilseed rape cultivars. Journal of PlantNutrition, 16(1), 85–96.

artomeus, I., Potts, S. G., Steffan-Dewenter, I., Vaissière, B. E.,Woyciechowski, M., Krewenka, K. M., et al. (2014). Contribu-tion of insect pollinators to crop yield and quality varies withagricultural intensification. PeerJ, 2, e328.

artón, K. (2014). MuMIn: Multi-model inference. R Package (ver-sion 1.7.7)

ennett, E. M., Peterson, G. D., & Gordon, L. J. (2009). Understand-ing relationships among multiple ecosystem services. EcologyLetters, 12, 1394–1404.

ezemer, T. M., De Deyn, G. B., Bossinga, T. M., Van Dam, N. M.,Harvey, J. A., & Van der Putten, W. H. (2005). Soil communitycomposition drives aboveground plant–herbivore–parasitoidinteractions. Ecology Letters, 8, 652–661.

ianchi, F. J., Booij, C. J. H., & Tscharntke, T. (2006). Sustainablepest regulation in agricultural landscapes: A review on landscapecomposition, biodiversity and natural pest control. Proceedingsof the Royal Society B, 273, 1715–1727.

ivand, R., Altman, M., Anselin, L., Assuncão, R., Berke, O.,Bernat, A., et al. (2012). spdep: Spatial dependence: Weightingschemes, statistics and models. R Package (version 0.5-46).

ommarco, R., Kleijn, D., & Potts, S. G. (2013). Ecological intensi-fication: Harnessing ecosystem services for food security. Trendsin Ecology & Evolution, 28, 230–238.

oreux, V., Kushalappa, C. G., Vaast, P., & Ghazoul, J.(2013). Interactive effects among ecosystem services andmanagement practices on crop production: Pollination incoffee agroforestry systems. Proceedings of the NationalAcademy of Sciences of the United States of America, 110,8387–8392.

os, M. M., Veddeler, D., Bogdanski, A. K., Klein, A.-M., Tscharn-tke, T., Steffan-Dewenter, I., et al. (2007). Caveats to quantifyingecosystem services: Fruit abortion blurs benefits from crop pol-lination. Ecological Applications, 17, 1841–1849.

urnham, K. P., & Anderson, D. R. (2002). Model selection andmultimodel inference. New York: Springer.

ampbell, C. A. (1978). Soil organic carbon, nitrogen and fertility.Soil Organic Matter, 173–200.

M

Ecology 16 (2015) 737–745

haplin-Kramer, R., O’Rourke, M. E., Blitzer, E. J., & Kremen, C.(2011). A meta-analysis of crop pest and natural enemy responseto landscape complexity. Ecology Letters, 14, 922–932.

lassen, A., Peters, M. K., Ferger, S. W., Helbig-Bonitz, M.,Schmack, J. M., Maassen, G., et al. (2014). Complementaryecosystem services provided by pest predators and pollinatorsincrease quantity and quality of coffee yields. Proceedings ofthe Royal Society B, 281, 20133148.

leasby, I. R., & Nakagawa, S. (2011). Neglected biological pat-terns in the residuals. Behavioral Ecology and Sociobiology, 65,2361–2372.

oley, P. D., Bryant, J. P., & Chapin, F. S. (1985). Resource avail-ability and plant antihervibory defense. Science, 895–899.

ick, R. P. (1992). A review: Long term effects of agricultural sys-tems on soil biochemical and microbial parameters. AgricultureEcosystems & Environment, 40, 25–36.

ahrig, L. (2013). Rethinking patch size and isolation effects:The habitat amount hypothesis. Journal of Biogeography, 40,1649–1663.

oth, H. D., & Ellis, B. G. (1997). Soil fertility. Boca Raton, USA:CRC Press Inc.

abriel, D., Sait, S. M., Hodgson, J. A., Schmutz, U., Kunin, W. E.,& Benton, T. G. (2010). Scale matters: The impact of organicfarming on biodiversity at different spatial scales. Ecology Let-ters, 13, 858–869.

aribaldi, L. A., Steffan-Dewenter, I., Winfree, R., Aizen, M. A.,Bommarco, R., Cunningham, S. A., et al. (2013). Wild pollina-tors enhance fruit set of crops regardless of honey bee abundance.Science, 339, 1608–1611.

ohnston, A. E., Poulton, P. R., & Coleman, K. (2009). ‘Soil organicmatter: Its importance in sustainable agriculture and carbon diox-ide fluxes.’ In D. L. Sparks (Ed.). Advances in Agronomy, 101,1–57.

ennedy, C. M., Lonsdorf, E., Neel, M. C., Williams, N. M., Rick-etts, T. H., Winfree, R., et al. (2013). A global quantitativesynthesis of local and landscape effects on wild bee pollinatorsin agroecosystems. Ecology Letters, 16, 584–599.

lein, A.-M., Hendrix, S. D., Clough, Y., Scofield, A., & Kremen,C. (2015). Interacting effects of pollination, water and nutrientson fruit tree performance. Plant Biology, 17, 201–208.

lein, A.-M., Vaissiere, B. E., Cane, J. H., Steffan-Dewenter, I.,Cunningham, S. A., Kremen, C., et al. (2007). Importance ofpollinators in changing landscapes for world crops. Proceedingsof the Royal Society B, 274, 303–313.

al, R. (2010). Beyond Copenhagen: Mitigating climate change andachieving food security through soil carbon sequestration. FoodSecurity, 2, 169–177.

al, R. (2011). Sequestering carbon in soils of agro-ecosystems.Food Policy, 36, S33–S39.

erin, J. (1987). Compensation in winter rape following simulatedpollen beetle damage. Bulletin OILB SROP, 10, 57–63.

undin, O., Smith, H. G., Rundlöf, M., & Bommarco, R. (2013).When ecosystem services interact: Crop pollination benefitsdepend on the level of pest control. Proceedings of the RoyalSociety B, 280, 20122243.

attsson, L. (2008). Höjt pH ökar fröskörden. Svenskraps(http://www.svenskraps.se/kunskap/pdf/00768.pdf)

otzke, I., Tscharntke, T., Wanger, T. C., & Klein, A.-M. (2014).

Pollination mitigates cucumber yield gaps more than pesticideand fertilizer use in tropical smallholder gardens. Journal ofApplied Ecology, 52, 261–269.

Applied

N

N

O

P

R

R

S

S

S

T

T

V

V

W

W

W

W

W

W

I. Bartomeus et al. / Basic and

akagawa, S., & Schielzeth, H. (2012). A general and simplemethod for obtaining R 2 from generalized linear mixed-effectsmodels. Methods in Ecology and Evolution, 2, 133–142.

ilsson, C. (1987). Yield losses in summer rape caused by pollenbeetles (Meligethes aeneus). Swedish Journal of AgriculturalResearch, 17, 105–111.

erke, E.-C., & Dehne, H.-W. (2004). Safeguarding production– losses in major crops and the role of crop protection. CropProtection, 23, 275–285.

inheiro, J., Bates, D., DebRoy, S., Sarkar, D., & EISPACK authors.(2014). nlme: Linear and nonlinear mixed effects models. RPackage.

audsepp-Hearne, C., Peterson, G. D., & Bennett, E. M. (2010).Ecosystem service bundles for analyzing tradeoffs in diverselandscapes. Proceedings of the National Academy of Sciences ofthe United States of America, 107, 5242–5247.

awls, W. J., Gish, T. J., & Brakensiek, D. L. (1991). Estimating soilwater retention from soil physical properties and characteristics.Advances in Soil Science, 16, 213–234.

hackelford, G., Steward, P. R., Benton, T. G., Kunin, W. E., Potts, S.G., Biesmeijer, J. C., et al. (2013). Comparison of pollinators andnatural enemies: A meta-analysis of landscape and local effectson abundance and richness in crops. Biological Reviews of theCambridge Philosophical Society, 88, 1002–1021.

teffan-Dewenter, I., Munzenberg, U., Burger, C., Thies, C., &Tscharntke, T. (2002). Scale-dependent effects of landscape con-text on three pollinator guilds. Ecology, 83, 1421–1432.

trauss, S. Y., & Irwin, R. E. (2004). Ecological and evolutionaryconsequences of multispecies plant-animal interactions. AnnualReviews in Ecology, Evolution and Systematics, 35, 435–466.

atchell, G. (1983). Compensation in spring sown oil seed rape(Brassica napus L.) plants in response to injury to their flower

buds and pods. Journal of Agricultural Science, 101, 565–573.

hies, C., Steffan-Dewenter, I., & Tscharntke, T. (2003). Effects oflandscape context on herbivory and parasitism at different spatialscales. Oikos, 101, 18–25.

W

Available online at www.s

ScienceD

Ecology 16 (2015) 737–745 745

an der Putten, W. H., Vet, L. E. M., Harvey, J. A., & Wäckers, F.L. (2001). Linking above- and belowground multitrophic inter-actions of plants, herbivores, pathogens, and their antagonists.Trends in Ecology & Evolution, 16, 547–554.

azquez, D. P., Morris, W. F., & Jordano, P. D. (2005). Interactionfrequency as a surrogate for the total effect of animal mutualistson plants. Ecology Letters, 8, 1088–1094.

agg, C., Bender, S. F., Widmer, F., & van der Heijden, M. G. A.(2014). Soil biodiversity and soil community composition deter-mine ecosystem multifunctionality. Proceedings of the NationalAcademy of Sciences of the United States of America, 111,5266–5270.

ardle, D. A., Bardgett, R. D., Klironomos, J. N., Setala, H., vander Putten, W. H., & Wall, D. H. (2004). Ecological linkagesbetween aboveground and belowground biota. Science, 304,1629–1633.

ielgoss, A., Tscharntke, T., Rumede, A., Fiala, B., Seidel, H.,Shahabuddin, S., et al. (2015). Interaction complexity matters:disentangling services and disservices of ant communities driv-ing yield in tropical agroecosystems. Proceedings of the RoyalSociety B, 281, 20132144.

illiams, I. H., & Ferguson, A. W. (2010). Spatio-temporal distri-butions of pests and their parasitoids on the oilseed rape crop. InBiocontrol-based integrated management of oilseed rape pests.Netherlands: Springer.

illiams, I., & Free, J. (1979). Compensation of oil-seed rape (Bras-sica napus L.) plants after damage to their buds and pods. Journalof Agricultural Sciences, 92, 53–59.

illiams, A., & Hedlund, K. (2013). Indicators of soil ecosystemservices in conventional and organic arable fields along a gradi-ent of landscape heterogeneity in southern Sweden. Applied SoilEcology, 65, 1–7.

illiams, A., Birkhofer, K., & Hedlund, K. (2014). Above-

and below-ground interactions with agricultural management:Effects of soil microbial communities on barley and aphids.Pedobiologia, 57, 67–74.

ciencedirect.com

irect