Local resources, linear elements and mass-flowering crops determine bumblebee occurrences in moderately intensified farmlands

(1)

Local resources, linear elements and mass- ﬂ owering crops determine bumblebee occurrences in moderately intensi ﬁ ed farmlands

Eveliina Kallioniemi

^a,b,

*, Jens Åström

^a

, Graciela M. Rusch

^a

, Sondre Dahle

^a

, Sandra Åström

^a

, Jan Ove Gjershaug

^a

aNorwegianInstituteforNatureResearch,P.O.Box5685Sluppen,7485Trondheim,Norway

bNorwegianInstituteforBioeconomyResearch,P.O.Box115,1431Ås,Norway

ARTICLE INFO

Articlehistory:

Received13July2016

Receivedinrevisedform22December2016 Accepted24December2016

Availableonlinexxx

Keywords:

Bombus Pollinators Linearfeatures Mass-ﬂoweringcrops Landscapeecology Agriculturallandscapes Ecosystemservices

ABSTRACT

Thereisconsensusthatland-usechangeisamaindriverbehindtherecentdeclinesofmanypollinator populationsinEurope.However,itisstillnotadequatelyunderstoodhowthelocalresourcequalityand landscapecompositioninfluencepollinators,andifandhowtheeffectsvaryinspaceandtime.We analysed the influenceof landscape- (2kmradius) and local scale- (50m transects) resourceson bumblebee species richness and abundance during two years in South-eastern Norway, where agricultureishighlymodernisedbutlandscapesstillshowlimitedspatialhomogenization.Localflower densityandspeciesrichnesswerestronglypositivelyassociatedwithbumblebeedensitiesandspecies richness, buthigher landscape-levelflower speciesrichnesswerelinkedtolowerlocalbumblebee abundances.Earlyandlatemassfloweringcropshadclear,butcontrasting,effects.Thetotalareaofearly floweringcropshadaconsistentnegativeimpactonbumblebeedensityandspeciesrichnessthroughout theseason,whilelatefloweringcropshadapositiveimpactinthebeginningoftheseasonbeforetheir bloom,suggestingacarry-overeffectfrompreviousyears.Thenegativeeffectsofearlyfloweringcrops couldbeduetocompetitionofbumblebeeswithhoney-bees,whicharewidelyusedinthesecrops.

Bumblebeedensityandspeciesrichnesswereclearlynegativelycorrelatedwiththetotalareaofforest and flower-poor land use areas,including grass fields and cereals. In contrast,bumblebees were positivelyassociated withmost linearelements inthelandscape (especiallypasture andcropland verges),exceptforroads,whichnegativelyaffectedbumblebeedensities,possiblyduetoincreased mortality,sincethequalityoftheflowerresourcesdidnotdifferfromotherlinearelements.Ourresults showthatthequalityandthespatialandtemporaldistributionofflowerresourceswithinthelandscape areimportantdriversforbumblebees,butcancreatecounterintuitivedistributionpatternsdependingon thetemporalandspatialresolutionofthesurvey.Increasingflowerresourcesinlinearelementsandthe amountoflatemass-floweringcropsmaybeviablemanagementmeasurestoimproveconditionsfor bumblebeesinmoderatelyintensifiedlandscapes.

1.Introduction

Crop pollination is increasingly recognised as a major component of globalfood security. Pollinatorconservation and statusassessmentsarenowreceivingconsiderableattentiondue toincreasingthreats to pollinatorsand reports of considerable pollinatorpopulationdeclines(Pottsetal.,2016,2010;Vanbergen etal.,2013)togetherwithestimatedpollinationdeﬁcits(Garibaldi et al., 2016). Bee density and diversity are important for the

deliveryofa resilientpollinationservicetofloweringcropsand wildplants(Garibaldietal.,2014;Rogersetal.,2014).Ithasbeen observedthathigherpollinatordiversityleadstoincreasesinfruit andseedsetoffocalplantsandisanimportantpredictorofcrop yieldsworldwide(Garibaldietal.,2016;Lowensteinetal.,2015), possiblythroughimprovedmatchingbetweendifferentpollinator and crop species (Cardinale et al., 2006; Rosenfeld,2002).Bee populationsare alsosensitive toweatherconditions which can resultinlargeyeartoyearvariationinpopulationsizes.Mediated by species-specific responses, pollinator diversity helps to maintain stable pollination services by buffering against this variation, since it increases the likelihood that some species respond favourably to the fluctuating weather conditions (c.f

*Corresponding author at: Norwegian Institute of Bioeconomy Research, Vinnavegen38,Stjørdal7512,Norway.

E-mailaddress:[email protected](E.Kallioniemi).

http://dx.doi.org/10.1016/j.agee.2016.12.039

ContentslistsavailableatScienceDirect

Agriculture, Ecosystems and Environment

j o u r n al h o m e p a g e: w w w . el s e v i e r . c o m / l o c at e / a g e e

(2)

“responsediversity”inElmqvistetal.,2003;Garibaldietal.,2014;

Kremen et al., 2002). Accordingly, simpliﬁcation of pollinator communities has been linked to decreased stability of seed production(Bommarcoetal.,2012),andmaybeonereasonfora lowerstabilityofyieldsinpollinator-dependentcropscomparedto othercrops(Pottsetal.,2016).

Simpliﬁedlandscapecompositionthatresultfromagricultural intensiﬁcation(FjellstadandDramstad,1999;Rickettsetal.,2008;

Steffan-Dewenteretal.,2002;Tscharntkeetal.,2005)isamongthe critical factors that affectbeepopulations in the industrialized world.

Highercoveroflargeandhomogenouscroplandareasislinkedto pollinatorpopulationdeclines(Pottsetal.,2010;Senapathietal., 2015;Vanbergenetal.,2013)andmoreheterogeneouslandscapes areassociatedwithhigherbumblebeespeciesrichnessanddensities (Rundlöf et al., 2008), for example through provisioning of complementary ﬂoral resources (Mallinger et al., 2016).

Furthermore, many studies showthat higher proportions of cropland anddecreasedsemi-natural habitatsresultin lowerbee species richness(Garibaldietal.,2011;Rickettsetal.,2008).

Severalstudiesfocusontheeffectsoflanduseonpollinator densityanddiversity(e.g.Carreetal.,2009;Goulsonetal.,2010;

Rickettsetal.,2008)aswellasontheirforagingbehaviour(e.g.Jha andKremen,2013).Thesehowever,haveoftenbeenconductedin highlyhomogenouslandscapeswithintenseagriculturalproduc- tion (but see Diaz-Forero et al., 2013). Still, many agricultural landscapes in the Western world consist of long established patchworksofcroplandandotherlanduses,oftenconstrainedby abioticfactorssuchastopography.Thisisespeciallytrueinour study area in Norway, where the spatial simplification and homogenizationoftheagriculturallandscapehasbeenrelatively limited.Insuchsettings,otherfactors,suchashabitatqualityand thecontinuityoffoodresourcescouldbemorerelevantthanthe amountofavailablenestingsitesorforagingdistances(Garibaldi et al.,2011; Rickettset al., 2008), which has beenemphasised previously(Lonsdorfetal.,2009).AsinmuchoftheWesternworld, the quality of the landscape elements has also been highly transformedinNorway;pastures,lays,andmeadowshavebeen largelyconvertedintocereal or grassproduction usingmodern techniques,andmanysmallfieldshavebeenconglomeratedinto largerunits(FjellstadandDramstad,1999).Hence,currentland uses within the established agricultural landscape and their impacts on habitatquality, including thecomposition of crop- fields, are likely important drivers of pollinator occurrences (Kennedyetal.,2013;Ricouetal.,2014).

Forinstance,inadditiontonon-crophabitats,ﬂoweringcrops areanimportantresourceforpollinators(Rundlöfetal.,2014),a factor consideredalso in spatial models of pollination services (Zulian et al., 2013). Mass-ﬂowering crops constitute a pulse resource, highly concentrated in time, with strong effects on pollinator population structure (Diekötter et al., 2010, 2014;

Hanleyetal.,2011;Holzschuhetal.,2013,2016).Thiscanresultin temporaleffectsbothbetween(Rundlöfetal.,2014)andwithin years(Riedingeretal.,2015),aswellasspatialeffects(Montero- Castañoetal.,2016).ThecropphenologyinNorwayallowsusto studytheeffectsofearlyandlatemass-ﬂoweringcropsseparately.

Themainearlymass-floweringcropsbloominearlytolateMay andcanprovidelargepollensourcesfornewlyemergingqueens, whencommunitiesofworkersarestillrelativelysmall.Theseare likely important early season resources for bumblebees in the region of our study. Early pollen sources are often cited as importantfoodsourcesforbumblebees(O’Rourkeetal.,2014)but thecommonlyemployedexampleofwillows(Salixspp.)areoften sparsely distributed throughout an entire region, making it difficulttoassesstheirimportance.Theothermainmass-flowering cropsflowerinthemiddleofJuly–August,offeringresourcesina periodclearlyseparatedfromtheearlyones.

Despitetheattentiongivenrecentlytotheeffectofthequality oflandscapeelementsonpollinators(Kennedyetal.,2013;Ricou etal.,2014),currentknowledgeisinsufficienttoprovidereliable modelsofpollinationservicestosupportlocaldecisions,which oftenrelyonexpert-basedscoringofflowerresourcesuitability (Lonsdorf et al., 2011; Zulian et al., 2013). Frequently, natural habitats such as wetlands, heathland and woodland are given equalimportance(Steffan-Dewenteretal.,2002;Woodcocketal., 2013),whichmaybeonereasonbehindthelackofcorrespondence betweenbeerichnessandtheareaofsemi-naturalhabitatinthese studies, and which is typically foundin studies that aggregate land-usestoalesserdegree(Garibaldietal.,2011;Lonsdorfetal., 2009). Further, recent studies highlight the importance of botanicalattributesoflandscapeelementsforpollinators,showing forinstance,thatricherplantassembliesinriparianmarginscan supportmorepollinatorsthangrasslandfields(Coleetal.,2015).

In modern agricultural landscapes, the linear elements that borderthemajorlandusetypes—ﬁeldmargins,roadvergesand forestedges—aregenerallyconsideredtoprovidealargeproportion ofthefoodresources(HanleyandWilkins,2015)andnestingplaces forbees,andhavebeenusedasspatialindicatorsofthecapacityof agro-ecosystemstogeneratepollinatorservices(Zulianetal.,2013).

However,thepositiveimpact oftheirpresenceon thelandscapelevel abundance of pollinators have not been widely conﬁrmed. In addition,althoughpollinatorstudiesoftenconsidervariousspatial scales,rangingfromfarmtolandscapeleveleffects(e.g.Kennedy etal.,2013;Kovacs-Hostyanszkietal.,2013),theverylocal(transect level)effectshaveseldombeenanalysedtogetherwithlandscape leveleffects(butseeDiaz-Foreroetal.,2013).

Bumblebees form an important pollinator group in Norway (Totland et al., 2013) and are abundant enough to enable the collectionofsufficientdataforseveralspecies.Sincebumblebees haveanimpressiveabilitytolocatesuitableflowerresources(Olsson etal.,2015)andcanflyuptoseveralkilometrestoforage(Osborne etal.,2008), weexpectbumblebeeassembliestobeaffectedby differentfeaturesofthelandscape,withimpactsatdifferentscales.

In this study, we investigate the importance of landscape configurationand resource quality onbumblebees in relatively spatially heterogeneous, but highly modernised agricultural landscapes. We map the flower resources both at the local (transect)andlandscapescale,andusestudylandscapescentred oneitherearlyorlatemass-floweringcrops(ornoneforcontrol).

Bysimultaneouslyaddressingthetransectandlandscapelevels, withrepeatedsamplesspanningtwoyears,weareabletoexplore potential spatialand temporal aggregationeffects. We hypoth- esisedthatthe1)habitatqualityintermsoflocalflowerresources andtheamountofresourceswithinalandscapewouldbemain determinantsofbumblebeedensityandspeciesrichness,andthat 2) early and late mass flowering crops would synergistically enhancebumblebeepopulations.Duetothetemporalvariationof flower resources and the active forage seeking behaviour of bumblebees,wefurtherhypothesisedthat3)temporalandspatial aggregationeffectswouldinfluencethebumblebeedistributions.

Lastly, we wanted to explore to what extent 4) land-use heterogeneity inﬂuenced bumblebees in these moderately homogenisedlandscapes.

2.Materialandmethods 2.1.Studyareaandﬁeldsites

Theareastudiedwasthesouth-easternpartofNorwayinthe counties of Vestfold, Telemark and Buskerud (see Fig. 1). This regioncontainsarangeoflandscapetypes,includingsomeofthe most intensiﬁed agricultural landscapes in Norway, which are dominatedbycerealproduction,interspersedwithvegetablesand

(3)

fruitandberryproduction,as wellaslandscapesdominatedby forests,withinterspersedagriculturalactivity.Thelandscapeunits inthestudy(seebelow)wereclusteredintwogeographicalareas, theWestwithabundantfruitproductioninadditiontocrops,and the East containing mostly agricultural crops, but with fruit productioninsomelandscapes.

Weuseddigitalland-covermaps(BjørdalandBjørkelo,2006)to delimit26circularareasof2kmradius(hereon,‘landscapes’)with varying composition of land uses. Although maximum flight distancesofbumblebeescanbeseveralkilometres,mostforaging movementsrecordedarelessthan1km(Hagenetal.,2011 and referencestherein),making 2km radiia reasonablechoice.The centrepointsofthelandscapeswereselectedaccordingtotheir closenesstoappleorchardsandredcloverseedproductionfieldsin 2013, representing early and late mass-flowering crops, respectively.Only6.1%oftheareaofearlymass-floweringcrops wasmadeupofberrieswhichbloomoveranextendedperiodof time,sotheseagriculturalunitsrepresentanearly-seasonresource thatisdistinctfromthelatemass-floweringcropsofredcloverand spring-sownrapeseed.Theindividuallandscapestherebysamplea gradientoflandscapeswithhighagriculturalactivity(maximumof 58%agriculturallandand23%forest)tolandscapesdominatedby spruceandpineforest(minimumof2%agriculturallandand80%

forest).Intotal,thestudyareacomprised327squarekilometres, whichweremappedaccordingtotheirlanduse.Bumblebeeswere recordedalongtransectwalks(seebelow)coveringapproximately 11ha. It is possible that the individual transects were also inﬂuencedbyfeaturesoutsidethedelimitedlandscapeunitarea of2kmradius,astransectswerespreadwithinthesecirclesand couldbelocatedclosetotheborder.However,this wouldhave addedtotheunexplained variabilityin thedata,but not likely introduced systematic biases. Distances between the study landscape centres ranged from 2.5 to 64km. The minimum

distancebetweenthetwoclusterswere33kmandthedistance betweenthecentres57km.Furtherdetailsofthelandscapesare availableintheSupplementarymaterial.

2.2.Bumblebeeandtransectsampling

Withineachlandscape,weestablished50mtransectslocated alonglinearfeatures,withtransectstypesclassiﬁedasroad,ﬁeld, forestandpastureedgesbasedontheNorwegianLandResource Map 1:5000 AR5 classes 12,21&22, 30, and 23, respectively (BjørdalandBjørkelo,2006).Thenumberoftransectsandtransect typeswithineachlandscapewasproportionaltothetotallengthof eachlinearfeaturetypeinalandscape,varyingbetween6and30 transectsperlandscape,withameanof21,and535transectsin totalforthestudy.Thesamplingeffortwastherebyscaledtothe amountof linear elementsin thelandscape.Transect positions wererecordedwithaGPSandmappedtoenablethesamelocation to be revisited. We conducted a total of seven visits to each transect,onevisiteachmonthbetweenMayandAugustin2013 andoneinMay,JulyandAugustin2014.Therecordingwasdone duringdryweatherconditionsbetweenthehours09:00and18:00.

At each transect we walked approximately 5min per visit, allowingforadditionaltime for specimenhandlingandspecies identiﬁcation,and recordedthenumberandspeciesofbumble- bees within 2m on both sides. Individuals of B. lucorum, B.

soroeensis,B.magnus,B.cryptarumandB.terrestriswereaggregated intoonegroupfortheanalysesbecauseitwasnotalwayspossible to distinguish the species identity accurately in the ﬁeld.

Additionalindividualsforwhichspeciescouldnotbedetermined inthefieldwerecollectedandidentifiedinthelaboratory.Wealso recordedwhenbumblebeeswereobservedonflowersandwhich flowerspecies(totalof4919observationson104differentflower species/families). This data was used later for weighting the Fig. 1.Mapwithlocationsofthe26studylandscapesof2kmradius(plottedtoscale)insouthernNorwaycoveringintotal327km².Thestudylocationsweredividedintoeast andwestclustersfortheanalysisasshown.

(4)

importance of ﬂower species when calculating the density of ﬂowerresources(seebelow).

At5mfromthestartingpointofeachtransectwedelimiteda plot of 25m, in which we recorded all open flowers. The registeredflowerunitsdiffereddependingonthetypeofflower butthesameflowerunitwasconsistentlyrecordedwithinspecies.

Flower units included individual flowers (e.g. Convolvulaceae), flowerclusters(e.g.Apiaceae),flowerheads(e.g.Asteraceaeand clovers), racemes (e.g. other Fabaceae), and flower stalks (e.g.

Campanulaceae and Ranunculaceae). When flower density was veryhighandhomogeneousintermsofspeciescompositioninthe 25mplots,weuseda12msub-plotforflowercounts,andthe valueswereextrapolatedtothe25mplot.Inthecaseofshrubs, werecordedtheirheightand theareatheycoveredwithin the 25mplot,andwhetherthespecieswasflowering.Afewspecies wererecordedatagenusorfamilylevelduetodifficultyinreliable identification inthefield.Withinthesegroups,thespecies had similarflowertraits.

2.3.Variables

Weusedthetotalnumberofbumblebeesandthetotalnumber of bumblebeespecies recorded per transect visit as dependent variablesinthestatisticalmodels.

Wecalculatedthelengthofthelinearelements(road,field-, forest-and pasture edges,seeabove) and theareaof land use classes using digital land use/land-cover maps (Bjørdal and Bjørkelo,2006). Inaddition, weground-truthed thelandscapes in2014andrecordedthecropspeciesintheagriculturalfields.The landuseswerecombinedintofivedifferentmaincategories; 1) earlymass-floweringcrops(includingfruitorchardsandcultivated berries)2)latemass-floweringcrops,(includingpeas,cloverand oilseedrape)3) forests(including allforests classes inAR5) 4) flower-poor areas (including grass fields and cereal crops) 5) grasslands(includingpastures,meadowsand fallows).Weused these same categories also to calculate a Shannon’s land use diversityindexforeachlandscapewhichsummarizesthenumber oflandusecategoriesandevennessofthetotalareaofthedifferent landusetypeswithineachlandscape.

We calculated the flower resources both at transect and landscape level. Flower species richness at transect level was calculatedasanumberof speciesfloweringduringthetransect visit.Onlyopen,non-witheredflowerswererecorded.Becauseof the varying numbers of transects in different landscapes, we estimatedthetotalspeciesrichnesswithinthelandscapes,using the“specpool”functionfromtheVeganpackage(Oksanenetal., 2016) in R (R Core Team, 2016). These estimates were not correlatedwiththenumberoftransectsinalandscape,andcould therefore beused as anunbiased estimateof thetotal species richnesswithinalandscape.

Due to the differences in flower morphology and counting methods(seeabove),flowercountswerenotdirectlycomparable betweenspecies,butthemeasureiscomparableamongtransects andlandscapes(Rundlöfetal.,2014).Westandardizedtheamount offlowerresourcesperspeciesineachtransectinrelationtothe totalnumberofflowercountsofthatspeciesinthesamesampling period (month of sampling). To account for differences in the qualityof plant species as flower resource,we weighted these standardizedflowercountsbythetotalbumblebeevisitationrate foreach plantspecies/familyduring thesamesamplingperiod.

Thesestandardizedandweightedvalueswerethensummedupto representaflowerdensityestimateateachtransect.Similarly,at thelandscapelevel,wesummeduptherelativenumberofflowers ofeachflowerspecieswithinthelandscapeduringthesampling period,weightedthemwiththeplantspecificvisitsduringthat samplingperiodandthensummedtheseupforeachlandscape.

Local daily temperature averages were obtained from the NorwegianMeteorologicalInstitute(http://www.met.no),which areinterpolatedfora11kmgridbasedonlocalweatherstations (Mohr,2008;Tveitoetal.,2000).

2.4.Statisticalanalyses

Webuiltseparatemodelsforbumblebeespeciesrichnessand densityusingtheglmmADMBpackage(Fournieretal.,2012;Skaug etal.,2016)inR(RCoreTeam,2016).Weusedanegativebinomial distribution and log link function which is suitable for over- dispersed countdataand for modellingthe zero-inﬂation from transect samples with no bumblebee observations. We ﬁtted modelsthatincludedallexplanatoryvariablesofinterestapriori:

i.e.areasofthefivedifferentlanduses(seeabove) withineach landscape,Shannon’sland useindexforeach landscape,flower speciesrichnessanddensityatlandscapeandtransectlevel,the totallengthofthefourdifferentlinearelements(field,forestand pasture edges and roads) per landscape, and temperature asa covariate.Duetothespatialdivisionofthestudylandscapes,we dividedthelandscapesintoawesternandaneasternclusterand added thisasa factor inthemodels.Wereportand discussall parameterestimatestogetherwiththeiruncertainty,astheywere considered interesting a priori, and to avoid problems with spurious results associated with rule-based model selection (BurnhamandAnderson,2002;GelmanandHill,2007;Reichert and Omlin, 1997). We included transect_id, landscape_id and sampling period as random effects to account for spatial aggregation effects and repeated observations. In addition, we fittedseparatemodelstoinvestigate possibleseasonaleffectsof earlyandlatemass-floweringcropsonbumblebees.Themodels includedthetotalareaperlandscapeofeitherearlyorlatemass- floweringcrops,theirinteractionwithsamplingperiod,andthe east-west clustering factor as explanatory variables, using glmmADMBwithnegativebinomialdistributionandtransect_id and landscape_id as random effects. With all analyses, we accountedforzeroinflationinthemodelswhenitimprovedthe model fit based on AIC values. Similarly, we used families as

“nbinom”or“nbinom1”dependingonwhichproducedlowerAIC values.Allexplanatoryvariableswerestandardized(using^Xⁱ^X

sdðXÞ)to improve model ﬁt and interpretability (Zuur et al., 2009).

Furthermore,wecheckedthegeneralisedVarianceInﬂationFactor (VIF)forallmodelstotestcollinearityofthevariables(Foxand Monette,1992).Thesewerefound tobe<1.38forall variables which are consideredacceptaple. Spatialautocorrelation of the measuredvariableswascalculatedasMoran’sIusingtheRpackage

“ape”andwasverylow(<0.08)andnotsigniﬁcantforanyofthe models.

3.Results

3.1.Flowerresources

Localﬂowerspeciesrichnessanddensityhadpositiveeffecton both bumblebee densityand species richness (Fig. 2, Table 1).

Transectswithonestandarddeviationmoreflowerspecieshadon average 38% (95% conf.int: 28% to 48%) more individuals of bumblebeesand29%(95%conf.int:23%to35%)morebumblebee species. Landscape level flowerdensity likewise had estimated positiveeffectsonlocalbumblebeedensityandspeciesrichness althoughthe95%confidenceintervalsoverlappedzero.Interest- ingly, after accounting for the local flower resources, a higher flowerspeciesrichness inthesurroundinglandscapenegatively affectedlocalbumblebeedensity.Therewasasimilartendencyfor species richness, although this effect was more uncertain.

(5)

Landscapeswithonestandarddeviationmoreﬂowerspecieshad on average 16% (95% conf.int: 28% to 4%) less bumblebee individualsand 5.6%(95%conf.int:13to3%)fewerbumblebee speciespertransect.Theresultswerequalitativelythesameina modelcontainingonlylandscapelevelexplanatoryvariables(not

shown),indicating that this isnot anartefact ofthemultilevel model.

3.2.Landuses

Bumblebee density and species richness were not clearly relatedtolandusediversity(Shannonlandusediversityindex) within the landscapes, and the parameter estimate showed considerable uncertainty. Adding second degree polynomials, allowingforahumped-backedeffect,didnotimprovethemodels andarenotreportedfurther.Incontrast,bothbumblebeedensity andspeciesrichnesswereclearlynegativelycorrelatedwiththe totalareaofforestandflower-poorlanduseareas,includinggrass fieldsandcereals(Fig.2).Flower-poorareasshowedthestrongest effectofallexplanatoryvariables;landscapeswithonestandard deviationmoreflower-poorlanduseshad onaverage45%(95%

conf.int:56%to31%)lessbumblebeeindividualsand32%(95%

conf.int:42%to21%) lessbumblebeespecies. Inthepresent study,flower-poorlandareastogethercompriseonaverage21%(sd 12) of the total area in the landscapes and make up the vast majorityofthecroplands.Areaswithonestandarddeviationmore forest had 19% (95% conf.int: 29 to 8%) less bumblebee individuals. There was alsoa tendencythat bumblebeedensity andspeciesrichnesswerenegativelyinfluencedbythetotalareaof pasture,but this effectwas uncertainwithconfidenceintervals overlappingzero.

3.3.Massfloweringcropsandinteractionwithsamplingtime Earlyandlatemassfloweringcropshadclear,butcontrasting, effects; the total area of early flowering crops within the landscapes had a negative impact on bumblebee density and speciesrichness,whilelatefloweringcropshadapositiveimpact onbumblebeedensity(Fig.2,Table 1).Theeffectoflatemass- floweringcropsonbumblebeespeciesrichnesswasalsopositive butmoreuncertain.Landscapeswithonestandarddeviationmore early mass-flowering cropshadonaverage21%less bumblebee individuals (95%conf.int:33%to6%)and 20% (95%conf.int:

28% to 10%) less bumblebee species. Landscapes with one standarddeviationmorelatemass-ﬂoweringcropshadonaverage 22%morebumblebeeindividuals(95%conf.int:5%to42%)and7%

(95%conf.int:3%to19%)morebumblebeespecies.

Theeffectoftheareaoflatemass-floweringcropswithineach landscapeinteractedclearlywithsamplingtime:largerareasof latemass-floweringcropswereassociatedwithhigherbumblebee speciesrichnessanddensityonlyintheearlyregistrationsinMay (Fig.3, Table 2). Laterin the season,from June toAugust, the relationshipwasingeneralnegative,withclearnegativeeffectson bumblebeediversityinJulyandAugust,andonbumblebeedensity in July. In contrast, there were no signs of interaction effects betweenearlymass-floweringcropsandsamplingperiod(Table2).

Thus,therewasnoevidencethatthenegativeeffectofearlymass- ﬂoweringcropswasdependentonsamplingperiod.

3.4.Linearfeatures

Both the total length of pasture and cropland edges were positively related to bumblebee density and species richness, althoughthe95%conﬁdenceintervaloverlappedzerofortheeffect ofcroplandedgeonbumblebeespeciesrichness.Landscapeswith onestandarddeviationmorepastureedgeormorecroplandedge had15%(95%conf.int:2%to30%)and41%(95%conf.int:5%to89%) morebumblebeeindividuals,and11%(95%conf.int:3%to21%)and 18% (95% conf.int: 3% to 44%) more bumblebee species, respectively.In contrast,thelengthofforestedgeshada lower Fig.2.Resultsfromthegeneralisedlinearmixedmodelsofbumblebeedensity(a)

anddiversity(b)withnegativebinomialdistributionandloglinkfunction.Allfixed explanatoryvariablesincludedinthemodelareshown.Thedotsshowtheexponent ofthemodelestimateforeachvariable,correspondingtoamultiplicativemodel, suchthate.g.avalueof1.5indicatesapositiveeffectof150%andavalueof0.5 representsanegativeeffectof50%.Thelinesrepresentthe95%confidenceintervals, andvariablesnotcrossing1(representingthecutbetweenpositiveornegative influenceinthemodel)arepresentedwiththickerlinesandlargerdots.EarlyMass cropandLateMasscroprefertoearlymass-floweringcropandlatemass-flowering areas,respectively.‘Cluster’indicatesWestandEastlandscapeclusters.Landuse diversitycorrespondsto theShannon indexcalculatedonlanduse/land-cover types.

(6)

estimatedinﬂuence,withtheuncertaintyoverlappingzerobroadly forbothbumblebeedensityandspeciesrichness.

Unexpectedly,thetotallengthofroadshadanegativeeffecton bumblebeedensity.Therewasacorrespondingestimatednegative effect onbumblebee species richness, although the conﬁdence intervalhereoverlappedzero.Landscapesthathadonestandard deviationmorelengthofroadshadonaverage15%(-25%to2%) lessbumblebeeindividualsand8%(95%conf.int:17%to0.7%)less bumblebee species. None of these effects appear tobe clearly relatedtotheﬂowerresourcesofthedifferentlinearlandscape elements, or the local bumblebee densities in each respective

linearelementcategory(Supplementarymaterial),suggestingthat thesewerenotdrivenbyaggregationeffects.

3.5.Randomeffectsandspatialclustering

Therandomeffectofsamplingperiodaccountedformostofthe variation with a variance of 0.25 and 0.17 for thedensity and speciesrichnessmodels,respectively.Individualtransectsinturn, had an estimated variance of 0.14 and 0.03, and individual landscapes0.005and0.002forthedensityandspeciesrichness models, respectively. The west-east clustering showed clearly higher abundances and species richness in thewestern cluster thanintheeastern(Fig.2).

4.Discussion 4.1.Flowerresources

Wefoundclearevidencethatbumblebeedensityandspecies richnessarepositivelyaffectedbythelocalflowerspeciesrichness anddensity,whichiscongruentwithmostpreviousfindings(Cole etal.,2015;Ebelingetal.,2008;Pottsetal.,2003).Forexample, Pottsetal.(2003)foundthatbeespeciesrichnesswaslinkedto floralspeciesrichness,andflowerabundancehasbeenfoundtobe a goodindicatorofbeepatchuse(Bennettetal.,2014).Ebeling etal.(2008)alsofoundthatincreasingplantdiversityenhances and stabilizespollinatorvisits.Thelocalflowerspeciesrichness was a particularly important determinant of local bumblebee richness,suggestingthatbumblebeeshavetheabilitytoeffectively locatetheirpreferredflowerresources.Takingintoaccountthese strong effects of the localflower resources,the landscape level densityofflowerresourcesshowedanadditionalbutweakerand more uncertain positive effect onbumblebees. In otherwords, increasingtheamountoffloweringplantsleadstogreaterdensity andspeciesrichnessofbumblebees,andinadditionitislikelythat thereisaspillovereffectfromthewildflowerresourcesinthe surroundinglandscape.Thiscouldactbothwithinthecurrentyear, withtheregionalleveloffloweringresourcesboostingpopulation numbers,orbetweenyearsasaresultofincreasedreproduction.

Since foraging is directed toward the most rewardingforaging habitatpatchesasdeterminedbyqualityanddistance,theactual foraginghabitatwilllikelyincludeawiderangeofforagequalities closetothenest,buta muchnarrowerrangefartheraway.Our results support previous evidence for spatial aggregation at Table1

Resultsofthegeneralisedlinearmixedmodelsforbumblebeedensityandbumblebeespeciesrichness.Year,transectIDandSamplingtimewereincludedinthemodelas randomfactors.

Bumblebeedensity Bumblebeespeciesrichness

Estimate Std.Error zvalue Pr(>|z|) Estimate Std.Error zvalue Pr(>|z|)

Flowerspeciesrichnesslocal 0.32 0.04 8.57 0.00 0.25 0.02 10.72 0.00

Flowerdensitylocal 0.27 0.04 7.08 0.00 0.10 0.02 5.70 0.00

Flowerspec.richn.landscape 0.17 0.07 2.56 0.01 0.06 0.05 1.26 0.21

Flowerdensitylandscape 0.10 0.05 1.84 0.07 0.06 0.04 1.82 0.07

Flower-poorm² 0.59 0.11 5.18 0.00 0.39 0.08 4.85 0.00

Forestm² 0.26 0.09 2.90 0.00 0.21 0.06 3.20 0.00

Pasturem² 0.10 0.06 1.56 0.12 0.07 0.04 1.57 0.12

Latemass-ﬂowercropm² 0.20 0.08 2.54 0.01 0.07 0.05 1.31 0.19

Earlymass-ﬂowercropm² 0.23 0.08 2.75 0.01 0.22 0.06 3.79 0.00

Pasturelinear 0.14 0.06 2.22 0.03 0.11 0.04 2.56 0.01

Forestlinear 0.02 0.08 0.24 0.81 0.05 0.05 0.87 0.38

Croplinear 0.34 0.15 2.28 0.02 0.16 0.10 1.61 0.11

Roads 0.16 0.07 2.33 0.02 0.09 0.05 1.82 0.07

Landusediversity 0.10 0.14 0.72 0.47 0.11 0.09 1.15 0.25

Temperature 0.10 0.05 1.92 0.05 0.06 0.04 1.40 0.16

Cluster 0.55 0.14 3.88 0.00 0.39 0.10 3.88 0.00

Intercept 0.09 0.25 0.34 0.73 0.47 0.21 2.20 0.03

Fig. 3.Results of generalised linear mixed models exploring the interaction betweensamplingtimeandtheamountoflateandearlymass-floweringcropsin thelandscapesforbumblebeedensity(a)andspeciesrichness(b)withnegative binomialdistributionwithloglink.LateMassCroprefertolatemass-flowering crops.Thedotsshowtheexponentofthemodelestimateforeachvariable,similar toFig.2.Thelinesrepresentthe95%confidenceintervals,andvariablesnotcrossing the1(representingthecutbetweenpositiveornegativeinfluenceintheloglinked model)arepresentedwiththickerlinesandlargerdots.

(7)

attractive resources (e.g. Walther-Hellwig and Frankl, 2000;

Osborneetal.,2008;Redheadetal.,2016)andpollinationmodels that incorporate foraging behaviour, which assume that bees selectivelyusethose partsofthelandscapethatenhancestheir fitness,avoidingforaginginotherpartsofthelandscape(Olsson et al., 2015).Still, theresults indicate that increasingthe total amount of flower resources within a landscape will increase landscape level pollinator density and species richness, which couldinturnleadtohigherflowervisitationandlikelymorestable andeffectivepollinationofbothcropsandwildplans(Bommarco etal.,2012;Rogers etal.,2014).For example,arecent studyby Garibaldietal.(2016)showedthebenefitofincreasingpollinator densityandrichnessoncropyieldinawiderangeofagricultural crops.

Interestingly,aftertakingintoaccountthestrongpositiveeffect oflocalflowerdensityandspeciesrichness,therewasanegative influenceofthelandscapelevelflowerspeciesrichnessonlocal bumblebee densities, and a similarly negative,but less certain effectonspeciesrichness. Thisadditionalbutopposite effectto localflowerrichnessiscounterintuitiveatfirstglance.However, bumblebees actively search the landscape and are likely to aggregateatparticularlyattractivelocalflowerresources,asthis studyindicates.JhaandKremen(2013),forinstance,documented longermean foraging distances in patches where flower plant richnesswashighandvariabilityinflowercoverbetweenpatches withinlandscapelow.Thismeansthatadistributedhighdiversity offlowerresourceswithinalandscapecouldleadtoadilutionof bumblebee at thelocal scale. Other studies also providesome evidenceofimpactsofflowerresourcesonthespatialaggregation ofbees(e.g.Osborneetal.,2008;Redheadetal.,2016;Walther- HellwigandFrankl,2000).Bennettetal.(2014)foundanegative influenceofflowerrichnessontotalbeevisitationrate,butthey suggestedthatthenegativeeffectofflowerrichnessatsmallscales couldbeanartefactofotherfactorsthatarecorrelatedwithflower richness,andwecannotexcludethatthiscouldalsobethecase here.

4.2.Landscapecomposition

Earlierstudieshaveassociatedhabitatandlandscapehetero- geneitywithhigherbiodiversity(Bentonetal.,2003;Holzschuh etal.,2007;Martinset al.,2014).In ourstudy, whiletherewas positiveestimatedeffectsoflanduseheterogeneityonbumblebee speciesrichnessanddensity,theseestimateswereratherlowand

had considerable uncertainty. A possible reason for this result couldbethecoarsescaleofthelandusecategoriesthatwereused forcalculatingthelanduseheterogeneityindex(ﬁvecategories).A more detailed description of the habitats might beneeded for revealing a hypothetical positive effect. However, given the relativelyhighheterogeneityoftheentireregion,itisreasonable thatvariationinlandscapeheterogeneitywillherehavelesseffect.

Incontrast,wefoundstrongnegativeeffectsofflower-poorareas— here representing mainly cereal crops and grass fields — on bumblebeespeciesrichnessand density,indicating thatatleast this category is relevant despite being broad. Typically, this categoryrepresentslargemonoculturefieldswithfewflowersand otherresourcesforbumblebeesandareexpectedtoshownegative correlationwithpollinatorsonalandscapescale(Holzschuhetal., 2007;Steffan-Dewenteretal.,2002).

Theamountofforestedareawithinalandscapehadanegative effectonbumblebees.Thiscontrasts withearlierstudieswhere forestedareasareconsideredasnaturalorsemi-naturalhabitats andaregenerallyexpectedtohavepositiveimpactsonnumbers anddiversityofbees(Garibaldietal.,2011;Kremenetal.,2004;

Zulian et al., 2013). However, some studies have indicated a negativecorrelationbetweenforestcoverandbeeabundanceand diversity(Carreetal.,2009;Mandeliketal.,2012;Winfreeetal., 2007).Possibledifferencesbetweenthesestudiescouldbeboth thequantityandqualityofforestswithinthelandscapes,andthe kindofresourcesthey provide.Foreststandsarethoughttobe relativelylowqualityforforagingbutpossiblyprovidingnesting substrate(RoulstonandGoodell,2011).Inthepresentstudyarea, eventhemostintensivelyfarmedlandscapeshaveafairamountof forestleft;theforestcoverinthelandscapesvariedbetween23 and80percent.Consideringtherelativelyhighdispersalabilities ofbumblebees(Dramstad,1996;Osborneetal.,2008),thiscould mean that resources in forests might not be limiting for bumblebees in these landscapes. In addition, most bumblebee species found in the agricultural landscapes are not forest specialists,butratherassociatedwithagriculturalmanagement, andthereforeanegativeassociationwithforestsisnotunexpected (Ødegaardetal.,2015;Åströmetal.,2016).

Lastly,theareaofpastureshadweaknegative,anduncertain effect on bumblebee species richness and density. Pasture is deﬁnedintheNorwegianLandResourceMap asanagricultural areathatcouldbeusedforpasture,butnotharvestedbymachine, and istherefore abroad category. Overall,therewereveryfew semi-naturalgrasslandswithahighabundanceofﬂoweringplants Table2

Resultsfromthegeneralisedlinearmixedmodelsofbumblebeedensityandbumblebeespeciesrichnesswithinteractionbetweensamplingtimeandmass-ﬂoweringcrops asexplanatoryvariables.YearandtransectIDwereincludedinthemodelasrandomfactors.

Bumblebeedensity Bumblebeespeciesrichness

Estimate Std.Error zvalue Pr(>|z|) Estimate Std.Error zvalue Pr(>|z|)

Intercept 0.54 0.11 4.80 0.00 0.79 0.09 8.90 0.00

Cluster 0.34 0.15 2.25 0.02 0.24 0.11 2.22 0.03

LateMasscropMay 0.22 0.10 2.34 0.02 0.20 0.07 2.74 0.01

LateMasscropJune 0.03 0.10 0.28 0.78 0.10 0.09 1.11 0.27

LateMasscropJuly 0.25 0.10 2.65 0.01 0.29 0.07 4.22 0.00

LateMasscropAugust 0.18 0.10 1.80 0.07 0.18 0.07 2.44 0.02

June 0.24 0.11 2.18 0.03 0.31 0.10 3.17 0.00

July 1.57 0.10 16.11 0.00 1.10 0.08 14.68 0.00

August 0.96 0.10 9.47 0.00 0.69 0.08 8.59 0.00

EarlyMasscropMay 0.11 0.10 1.01 0.31 0.01 0.09 0.11 0.91

EarlyMasscropJune 0.09 0.12 0.76 0.45 0.00 0.11 0.01 0.99

EarlyMasscropJuly 0.03 0.10 0.32 0.75 0.07 0.08 0.88 0.38

EarlyMasscropAugust 0.02 0.10 0.18 0.86 0.04 0.08 0.49 0.63

Intercept 0.54 0.11 4.80 0.00 0.79 0.09 8.90 0.00

(8)

in thestudy area, and it is likely that theseareas classiﬁedas pasturesrepresentingeneralpoorforagingareasforbumblebees.

4.3.Linearelements

The borders between two land use types delimit distinct habitatsthroughoutthelandscapes,which,together withroads, constitute what we have called ‘linear elements’. In modern agricultural landscapes, which are mostly devoid of non-crop ﬂoweringplants,theselinearelementsconstituteamajorresource for pollinators. Our data show varying effects of these linear elementswhichindicatescleardifferencesinthewaythedifferent typesoflinearelementsinﬂuencebumblebees.

Weobservedpositive effectsof theamount(totallength)of pasturebordersinthelandscapesonbothbumblebeedensityand speciesrichness. Pasturebordersdidnothaveparticularly high levels of ﬂower resources or bumblebee occurrences (see Supplementarymaterial),suggesting thattheirbeneﬁttobum- blebeesmaynotbeprimarilyasaproviderofforageareas.Onecan speculate that these linear elements provide suitable nesting habitatsforbumblebees,aspasturesarelessfrequentlysprayed with insecticides, and not shaded by forest, thus providing favourablemicro-climates(Herrera,1995).

Forestedges,inturn,hadweakinﬂuenceonbothbumblebee density and species richness. This is in contrast to common conceptions,asearlierstudies(Bennettetal.,2014)havefound thatbeesoftenforageinareasclosetoforestedges.Theweakeffect offorest edgesmaybea resultof thatforests are nota scarce resourceintheseregions,that mostbumblebeespeciesarenot forestspecialists,orthat forestedgesareassociatedwithforest area,whichhadaclearnegativeinﬂuenceonbumblebees.

Cropbordershadinourstudyapositivebutvariableeffecton bumblebeedensity,withtheeffectonbumblebeespeciesrichness beingpositive,butmoreuncertain.Theuncertaininfluenceofcrop borders is possibly due to different management regimes of croplands. To our knowledge, after consulting with the local agriculturalextensionservice, there areno agro-environmental schemesimplementedintheareatoenhancethequalityoffield marginsaspollinatorresources,andasaresult,theyaremostly verynarrowwithwildflowersgrowingoutsidethemanagedarea.

However,thetypeofcropandtheuseofpesticidesorherbicides vary,aswellaslocalgroundconditions,whichresultsinvariability amongﬁeldedgesintheirqualityaspollinatorresourceproviders.

Still,theseresultsindicatethatﬁeldbordershavethepotentialto providevaluableresourcesforbumblebees,ifproperlymanaged.

Incontrasttoallotherlinearelements,theamountofroadshad a negative inﬂuence on local bumblebee density, and also a negativebutmoreuncertaineffectonbumblebeespeciesrichness.

The negative influence of roads is not likely explained by differences in flower resources along roads compared to other linearelementsinthelandscape,northatbumblebeesaggregate especiallytoroadsides.Roadsidesshowedflowerresourcesonpar withcroplandedgesandhad bumblebeeoccurrencesin similar rangeastheothertransecttypes(seeSupplementarymaterial).In contrast, Hanley and Wilkins (2015) attributed higher local bumblebeeabundancetohigherflowerabundancesalongroads.

Thenegativeinfluenceofroadscouldhavevariousexplanations, includingincreasedmortalityduetotrafficcollisions,sub-optimal timingofmowing,herbicideapplication,saltspreadingatwinter time, orpollution. Theseresultsindicate thatthere might bea trade-offbetweenincreasingbumblebeesbyaugmentingroadside flowerresourcesandpotentiallossofbumblebeesbyunforeseen negativeeffects.Giventhecurrentattentionthatroadsideflower resourcemanagementisreceivingasapotentialtooltoimprove forageresourcesforbees(e.g.Hopwood,2008;MacCana,2013), theseunexpectednegativeeffectsoughttobeinvestigatedfurther.

4.4.Mass-ﬂoweringcropsandseasonalvariation

Unexpectedly,wefoundaclearnegativeeffectofthetotalarea ofearlymass-floweringcropswithinthelandscapesonbumblebee density and species richness, and this effect was consistent regardless of sampling time. This consistencysuggest that the negativeeffectis notdue totemporary aggregationin thefruit orchards duringbloom. Wealsoobserved a highdominanceof managed honey bees within the fruit orchards (unpublished), indicating that this resource was relatively little used by bumblebees.Thenegativeinfluenceofearlymass-floweringcrops on bumblebees could thus be the result of competition from managedhoneybees.Almosteveryappleorchardkepthoneybees tosecureefficientpollination,andtheforagingrangeofhoneybees aresufficienttoinfluencetheentirelandscapeofstudy.Itisalso possiblethatbumblebeesareaffectedbyinsecticidesprayingsor other agrochemicals commonly used at the orchards, but this seems less likely due to the seeming lack of aggregation of bumblebees withinthe orchards. Insecticide sprayingsare also relativelylimited,especiallyduringthebloomingperiod(personal communicationwithapplefarmers).Further,earlymass-flowering cropsmayhavelittleimpactonbumblebeecolonysizeearlyinthe season,and mayactas ‘flower-poor’ elementsin thelandscape after the blooming period, depending on how the orchard understoreyismanaged(i.e.compositionofthegrassland).

Incontrast,therewasastrongpositiveeffectoftheareaoflate mass-ﬂowering cropswithin landscapesonbumblebeedensity, andsimilarbutlesscertaineffectsonbumblebeespeciesrichness.

Thepositiveeffectoflatemass-floweringcropsiscongruentwith findings that mass-flowering crops increase colony size of B.

terrestris,B.vosnesnskiiandB.pascuorum(Herrmannetal.,2007;

Westphaletal.,2009;Williamsetal.,2012).

Therewasclearevidenceofaninteractionbetweensampling timeandthepresenceoflatemass-floweringcrops,showingthat landscapes with more late mass-flowering crops had higher bumblebeedensityandspeciesrichnessearlyintheseasoninMay, while we documented lower bumblebee density and species richness fromJunetoAugust,whenredcloverandspring-sown oilseedrapeareinbloom.ThiswasespeciallyevidentfortheJuly sampling,whenthered-cloverstartstobloom.Wesuggestthatthe positiveinfluenceoflatemass-floweringcropsearlyintheseason is a carry-over effect from higher bumblebee reproduction in previous years. Although farmers in ourstudy region typically rotate crops betweenyears and fields, this rotation is centred aroundthefarmsandcroptypeswithinalandscapearerelatively stable between years. As a result, we cannot specifically disentangle theeffect of previous and current yearcrops. This wouldrequireahigherturnoverofareaofmass-floweringcropsin the landscapes, or considerably more landscape replicates.

However, a recent study from Sweden also shows a positive correspondenceofbumblebeequeendensitywiththeareaofalate mass-ﬂoweringcrop,redclover(Trifoliumpratense)(Rundlöfetal., 2014). The suggested mechanism is that red clover provides resourceslateintheseasonwhencolonysizesarelargestandthe production of reproducing individuals takes place. Therefore resource demand is high during this time and late-season ﬂowering resources can be a bottleneck for reproduction.

Riedingeretal.(2015)alsodocumentedhigherdensitiesofbees thefollowingyearinlandscapeswithhighcoverofoilseedrape.

We attribute the negative association between bumblebee occurrencesandlatemass-ﬂoweringcropareasinJune-August,to that thelatemass-ﬂoweringcropsattractbumblebeesfromthe surroundinglandscape,wherethesurveytransectswerelocated.

Notethatthiseffectisindependentofthequalityorquantityofthe localresourcesattransectsintheselandscapes.Thisisinlinewith theaggregationeffectsofnon-bombusbeesshownbyRiedinger

(9)

et al. (2015) in landscapeswith increasing oilseed rape cover.

Rundlöfetal.(2014)alsofoundlargerdensitiesofbumblebeesin themass-floweringfieldsthaninthefieldbordersduringmass- flower crop bloom, and the attractiveness of red clover to bumblebees well known (e.g. Carvell et al., 2006; Fussell and Corbet,1992;Pywelletal.,2005).Anotherstudyhasshownlower bumblebeeabundancesingrasslandsinlandscapeswithoilseed rapeduringblooming(Holzschuhetal.,2011)whileHanleyetal.

(2011)documentedincreasedbumblebeenumbersduringfield- beanbloomingintransectsadjacenttothefields.Allthesestudies indicatepowerfulaggregationeffectsofmassfloweringcropsthat maytemporarilyaffectthedensitiesinthesurroundinglandscape.

Therearealsoreportsthatwildplantsarelikelytosharethe samepollinatorsasoilseedrape(StanleyandStout,2014),thus possiblyincreasingbetween-plantcompetitionforpollinators.Our resultsthereforeraise questionsaboutthe impactof themass- floweringcropsonwildflowersthatfloweratthesametime,ifthe mass-floweringcropsaredrawingindividualsfromthelandscape (Diekötteretal.,2010;Holzschuhetal.,2011;Rundlöfetal.,2014).

5.Conclusions

Thisstudyshowsthattheabundanceandspeciesrichnessof bumblebees in moderately modified agricultural landscapes is stronglyinfluencedbothbythelocalandthedistributed flower resourcesthroughoutthelandscape.Thelocalamountanddiversity ofwildflowers,whichto ahigh degreearelocatedalonglinear elementsinthelandscape,aswellastheareaoflatemass-flowering crops, was found to have the strongest positive influence on bumblebeedensitiesandspeciesrichness.Increasingtheamountof these resourcesthroughout thelandscapesappears to beviable managementtargetsforbumblebees,alsoinmoderatelyhomoge- nisedlandscapes.Thehighlymobileforagingbehaviourofbum- blebeescausedspatialaggregationatattractiveflowerresources, whichdilutedthebumblebeesinlandscapesrichinresources,with counterintuitiveresponsestoregionalflowerdiversityasaresult.

Theseaggregationeffectsalsohadatemporalsignal,corresponding to the bloom of late mass-flowering crops, causing temporary, negativerelationshipsbetweentheobservednumbersofbumble- beesandtheamount offlower resources within thelandscapeduring bloom.Thesefindingshighlightthecomplexnatureofpollinator surveysandwarnsagainstmakingconclusionsbasedonsurveysof insufficientspatialandtemporalspan.

Somewhat worryingly, early mass-flowering crops, here representedbyfruitorchardsthatbloomaroundtheemergence ofhibernatingqueens,hadnegativeeffectsonbumblebees.Alikely mechanismforthisresultiscompetitionbymanagedhoneybees centredat theearlymass-floweringcrops, butthis needstobe further studied to be concluded. Consequently, pollination- dependentcrops both benefit fromthe landscape-level species poolofpollinators,butalsogreatlyinfluencewildpollinators.At the same time, this valuable resource for wild pollinators is affectedalsobyfactorsbeyondtheindividualfarmer’smanage- mentcontrol.

Apartfrommass-floweringcrops,thebulkofthefoodresources forpollinatorswheresituatedalonglinearelementsthatdissect thelandscape,andthedominatingcropfieldsofgrassandcereal negativelyaffectedbumblebeedensityand diversity.Simplifica- tionofthelandscapewithflower-poormonoculturesandmerging oflandparcelsthereforehasnegativeconsequencesforpollinators alsoinrelativelyheterogeneousagriculturallandscapessuchasthe onesinourstudy. Lastly,theunexpectednegativeeffectofthe amount of roads within the landscapes on bumblebees raise questionsoftheappropriatenessofincreasingflowerresourceson roadvergesas aconservationmeasure, and encouragesfurther study.

Acknowledgements

This studywas funded by: the Research Council of Norway (NINA – Strategic Institute Program 208434/F40), PolliClover project, 225019 and BIOSMART project 244608/050) and the Norwegian Environment Agency (2012/16642). We thank two anonymous referees for their constructive suggestions which substantially improved the manuscript, and Heidi Myklebost, ArnsteinStaverløkk,EmmaBengtssonandalargenumberofﬁeld assistantsthathelpedinthenumerousﬁeldcampaigns,andFrode Ødegaardfordiscussionsduringtheplanningphase.

AppendixA.Supplementarydata

Supplementarydataassociatedwiththisarticlecanbefound,in theonlineversion,athttp://dx.doi.org/10.1016/j.agee.2016.12.039.

References

Åström,J.,Åström,S.,Bøhn,K.,Gjershaug,J.O.,Staverløkk,A.,Ødegaard,F.,2016.

NasjonalOvervåkingavDagsommerfuglerogHumleriNorge.Oppsummering avAktivitetenI2015.NI..

Bennett,J.A.,Gensler,G.C.,Cahill,J.F.,2014.Small-scalebeepatchuseisaffected equallybyﬂoweravailabilityandlocalhabitatconﬁguration.BasicAppl.Ecol.

15,260–268.doi:http://dx.doi.org/10.1016/j.baae.2014.03.004.

Benton,T.G.,Vickery,J.A.,Wilson,J.D.,2003.Farmlandbiodiversity:ishabitat heterogeneitythekey?TrendsEcol.Evol.18,182–188.doi:http://dx.doi.org/

10.1016/S0169-5347(03)00011-9.

Bjørdal,A.,Bjørkelo,K.,2006.AR5klassiﬁkasjonssystem.Klassiﬁkasjonav arealressurser(No.01/2006).Skogoglandskap..

Bommarco,R.,Lundin,O.,Smith,H.G.,Rundlöf,M.,2012.Drastichistoricshiftsin bumble-beecommunitycompositioninSweden.Proc.R.Soc.B-Biol.Sci.279, 309–315.doi:http://dx.doi.org/10.1098/rspb.2011.0647.

Burnham,K.P.,Anderson,D.R.,2002.ModelSelectionandMultumodelInference:a PractivalInformation-theoreticApproach,2nded.Springer.

Cardinale,B.J.,Srivastava,D.S.,Duffy,J.E.,Wright,J.P.,Downing,A.L.,Sankaran,M., Jouseau,C.,2006.Effectsofbiodiversityonthefunctioningoftrophicgroups andecosystems.Nature443,989–992.doi:http://dx.doi.org/10.1038/

nature05202.

Carre,G.,Roche,P.,Chifﬂet,R.,Morison,N.,Bommarco,R.,Harrison-Cripps,J., Krewenka,K.,Potts,S.G.,Roberts,S.P.M.,Rodet,G.,Settele,J.,Steffan-Dewenter, I.,Szentgyorgyi,H.,Tscheulin,T.,Westphal,C.,Woyciechowski,M.,Vaissiere,B.

E.,2009.Landscapecontextandhabitattypeasdriversofbeediversityin Europeanannualcrops.Agric.Ecosyst.Environ.133,40–47.doi:http://dx.doi.

org/10.1016/j.agee.2009.05.001.

Carvell,C.,Westrich,P.,Meek,W.R.,Pywell,R.F.,Nowakowski,M.,2006.Assessing thevalueofannualandperennialforagemixturesforbumblebeesbydirect observationandpollenanalysis.Apidologie37,326–340.

Cole,L.J.,Brocklehurst,S.,Robertson,D.,Harrison,W.,McCracken,D.I.,2015.

Riparianbufferstrips:theirroleintheconservationofinsectpollinatorsin intensivegrasslandsystems.Agric.Ecosyst.Environ.211,207–220.doi:http://

dx.doi.org/10.1016/j.agee.2015.06.012.

Diaz-Forero,I.,Kuusemets,V.,Maend,M.,Liivamaegi,A.,Kaart,T.,Luig,J.,2013.

Inﬂuenceoflocalandlandscapefactorsonbumblebeesinsemi-natural meadows:amultiple-scalestudyinaforestedlandscape.J.InsectConserv.17, 113–125.doi:http://dx.doi.org/10.1007/s10841-012-9490-3.

Diekötter,T.,Kadoya,T.,Peter,F.,Wolters,V.,Jauker,F.,2010.Oilseedrapecrops distortplant-pollinatorinteractions.J.Appl.Ecol.47,209–214.doi:http://dx.doi.

org/10.1111/j.1365-2664.2009.01759.x.

Diekötter,T.,Peter,F.,Jauker,B.,Wolters,V.,Jauker,F.,2014.Mass-ﬂoweringcrops increaserichnessofcavity-nestingbeesandwaspsinmodernagro-ecosystems.

GCBBioenergy6,219–226.doi:http://dx.doi.org/10.1111/gcbb.12080.

Dramstad,W.E.,1996.Dobumblebees(Hymenoptera:apidae)reallyforagecloseto theirnests?J.InsectBehav.9,163–182.doi:http://dx.doi.org/10.1007/

BF02213863.

Ebeling,A.,Klein,A.-M.,Schumacher,J.,Weisser,W.W.,Tscharntke,T.,2008.How doesplantrichnessaffectpollinatorrichnessandtemporalstabilityofﬂower visits?Oikos117,1808–1815.doi:http://dx.doi.org/10.1111/j.1600-

0706.2008.16819.x.

Elmqvist,T.,Folke,C.,Nyström,M.,Peterson,G.,Bengtsson,J.,Walker,B.,Norberg,J., 2003.Responsediversity,ecosystemchange,andresilience.Front.Ecol.Environ.

1,488–494.doi:http://dx.doi.org/10.1890/1540-9295(2003)001[0488:RDECAR]

2.0.CO;2.

Fjellstad,W.J.,Dramstad,W.E.,1999.Patternsofchangeintwocontrasting Norwegianagriculturallandscapes.Landsc.UrbanPlan.45,177–191.doi:http://

dx.doi.org/10.1016/S0169-2046(99)00055-9.

Fournier,D.A.,Skaug,H.J.,Ancheta,J.,Ianelli,J.,Magnusson,A.,Maunder,M., Nielsen,A.,Sibert,J.,2012.ADModelBuilder:usingautomaticdifferentiation forstatisticalinferenceofhighlyparameterizedcomplexnonlinearmodels.

Optim.MethodsSoftw.27,233–249.