1

Geophysical Research Letters

Supporting Information for

TheimportanceofENSOphaseduringvolcaniceruptionsfordetectionandattribution

Flavio Lehner1, Andrew P. Schurer2, Gabriele C. Hegerl2, Clara Deser1, and Thomas L. Frölicher3

1ClimateAnalysisSection,NationalCenterforAtmosphericResearch,Boulder,USA2School of Geosciences, University of Edinburgh, Edinburgh, UK

3Environmental Physics, Institute of Biogeochemistry and Pollutant Dynamics, ETH Zürich, Switzerland

Content

1 CoincidenceofvolcaniceruptionsandElNiñoeventsinpaleoclimaterecords2 Alternativeobservationaldatasets(FigureS1)3 Detectionandattributionextendedfigure(FigureS2)4 Detectionandattributionsensitivityanalysis(FigureS3andS4)5 Uncertaintyrangesandcontrolsubsampling(FigureS5)6 ResultswithGFDLmodels(FigureS6)7 AdditionalfigureforCESM1‘Pacemaker’simulations(FigureS7)8 AdditionalfigureincludingtheSantaMariaeruptionin1902(FigureS8)9 Tablesonmodels,simulations,andsubsampling(TablesS1,S2,S3)

2

1CoincidenceofvolcaniceruptionsandElNiñoeventsinpaleoclimaterecordsReconstructions of volcanism (Crowley and Unterman 2013) and ENSO (Li et al. 2013)suggestthatthecoincidenceofElNiñoeventswiththreeconsecutivevolcaniceruptionsofthe sameor greatermagnitude asAgunghas not occurred in at least the past 700 years.ThiswastestedbysearchingforcoincidencesoferuptionsandElNiñoeventsinthesameorfollowingyear.

2Alternativeobservationaldatasets

FigureS1:Globalmeansurfacetemperature(GMST)anomalyrelativetothemeanofthe5years preceding the eruption start date (dashed vertical line) averaged over the threerecent volcanic eruptions (Agung, El Chichon and Pinatubo) from different observationaldatasets.Timeserieshavebeensmoothedwithatriangular1-2-1filterforvisualpurposesonly.

3

3DetectionandattributionextendedfigureFigureS2isanextentionofFigure3ofthemainpaper.FigureS2bandS2dareidenticaltoFigure3aand3b.

FigureS2: Estimated contribution by natural (green) and anthropogenic (red) forcing toglobalmean surface temperature anomaliesusing (a,b) all CMIP5 simulations and (c,d)only thosesimulationswith theobservedENSOphase. Bold linesaremulti-modelmeansmultiplied by the best-fit scaling factor. This scaling factor is also plotted as a thickhorizontal line in (b) and (d). The shading shows the models multiplied by the 5-95%uncertainty range as plotted by the bar in (b) and (d). Dashed line shows the unscaledmodelresults(‘historicalNat’only).Anomaliesaretakenfromthemeanofthewholeperiod(a)orthemeanofeachsegment(c).Theverticaldashedlinesindicatetheeruptiondatesofthethreemajorvolcanoes.Timeserieshavebeensmoothedwithatriangular1-2-1filterforvisualpurposesonly.

4

4DetectionandattributionsensitivityanalysisAseriesofanalyseswerecarriedouttotestthesensitivityoftheresultsshowninFigure3andFigureS2tothedifferentanalysissteps.FiguresS3aandS3fshowthebestfitscalingfactorfortheunsampled,continoustimeseriesas also shown in Figure 3a and3b in themain paper. They show that the anthropogenicforcingisdetectable(valuesalwayssignificantlygreaterthanzero)andconsistentwiththeobservations(uncertaintyrangeencompassingone),whilethenaturalforcingisdetectablebut has a stronger response in models than in the observations (uncertainty rangesignificantlylessthanzero).Forthefinalsubsampledanalysis(FigureS3eandS3j;Figure3aand3binthemainpaper,respectively)thenumberofmodelsimulationsusedtoformthemulti-modelmeanandthenused in the regression analysis, was limited to 35 ‘historical’ simulations and 13‘historicalNat’simulations.Totesttheeffectofrestrictingthenumberofmodels,FigureS3bandS3g show the rangeof scaling factors for100 randomcombinationsof 35 ‘historical’and13 ‘historicalNat’simulations(withoutsplittingthetimeseries intosegmentsyet).Asexpected, the ranges of the scaling factors show some sensitivity to the subset ofmodelschosen,yet themainconclusionsareunchangedfromthefullsetofsimulations(compareFigureS3btoS3aandFigureS3gtoS3f).In the final sub-sampled analysis, the GMST time series is split into three equal-lengthsegments.Theeffectofthisontheresultsfor35randomlyselected‘historical’simulationsand13randomlyselected ‘historicalNat’ simulations is shown inFigureS3candS3h.Thebest-fit anthropogenic scaling factors increase slightlywhen the time series are split intothreesegments,likelyduetotheshorteningofthetimeseries.Takinganomaliesoverthreeshortsegmentsinsteadoftheentiretimeperioddecreasestheimportanceoftheendpointsintheregressionanalysis,inparticulariftheregressionslopehasacontinuoustrend,asisthecasefortheanthropogenically forcedresponse.Foragivensegment, thisalsoreducesthe signal-to-noise ratio andwith that the sensitivity to the choice of randomly selectedmodels. This results in a slightly noisier uncertainty range in Figure S3c and S3h.Importantlyforthemainconclusionsofthestudy,thescalingfactorsforthenaturalforcingresponseremainunchanged(compareFigureS3gandS3h).ThefinalsensitivitytestaddressesthesubsamplingofthemodelstoselectonlythosewithanElNiñoeventintheborealwinteraftereachofthelargevolcaniceruptions.TheresultsareshowninFigureS3dandS3ifor100differentrandomcombinationsofthesubsampledmodels(again,foreachofthe100randommodelcombinationswepick35from‘historical’and13 from ‘historicalNat’).Thebest-fitanthropogenicscaling factorsareverysimilarastheones fromthefullsetofmodelsimulations(compareFigureS3dtoS3c).There is lessdependence on the choice of model, i.e., the uncertainty range is less noisy, in thesesubsampled results because there are generally fewermodels to choose from due to the

5

ENSOphasesubsamplingcriteria.Thescaling factors for thenatural forcingareshown inFigure S3i, in which, crucially, the scaling factor range now always encompasses one,regardlessofmodelselection,whileifthemodelsarenotsub-sampled(FigureS3f,g,h)theuncertainty range is always significantly less than one. This demonstrates that ourconclusionsarenotaconsequenceofthemodelselection.One finalconcernregarding themodelselection is the largenumberofsimulations in theCMIP5 archive from one particularmodel family, namely theNASAGoddard Institute forSpace Studies (GISS) models: 45 out of 198 ‘historical’ simulations and 20 out of 68‘historicalNat’simulationsarefromGISS-E2-H-CC,GISS-E2,GISS-E2-R-CC,orGISS-E2-R.Wererantheanalysis inFigureS3ontheCMIP5archiveexcludingallGISSmodelsandfoundnot discernable difference to the results using all CMIP5models (compare Figure S3 andS4).Theresultspresented in themainpartof thepaperare thecombinationofall thescalingfactorscalculated foreachof the100modelcombinationswhichareshowninFigureS3dandS3i.ThesecombinedrangesareshowninFigureS3eandS3j.

FigureS3:Scalingfactors fortheresponsetoanthropogenicandnatural forcings.(a)and(f) are based on the continuous timeseries from the unsampled multi-model meanensembles;(b) and(g) arebasedon100different combinationsof35 randomly selected‘historical’ and 13 ‘historicalNat’ simulations; (c) and (h) are based on 100 differentcombinations of 35 randomly selected ‘historical’ and 13 ‘historicalNat’ simulations splitintothreesegments;(d)and(i)arebasedonthesamecriteriaas(c)and(h)butrestricted

6

to those simulationswhich fulfill the ENSO phase selection criteria. (e) and (j) show thecombinedresults forallmodelcombinationsshownin(d)and(i).Colorshadingindicatesthe5-95%range,andthesolidcurvesshowthebest-fitscalingfactor.

FigureS4:SameasFigureS3,butexcludingallGISSmodels.

7

5UncertaintyrangesandcontrolsubsamplingTo calculate the uncertainty range for the scaling factors, internal variability samples areadded to the noise-reduced fingerprints (themulti-model mean from the ‘historical’ and‘historicalNat’ simulations) and observations. For each combination of internal variabilitysamplesascaling factor is calculated.Theuncertaintyrange is thendetermined fromthisdistributionofscalingfactors(seeFigureS5a).Toestimateanuncertaintyrangeonthescalingfactorsof thefullCMIP5modelset,1,000different combinations of ‘piControl’ simulation samples are used to calculate a 5-95%rangeforeachoftheforcingscalingfactors.Fortheanalysisonthesubsampledsimulationswe repeat this procedure 100 times with 1,000 different combinations of ‘piControl’simulations. A 5-95% uncertainty range is then calculated from the combination of all100,000calculatedscalingfactors.Another concern regarding the estimation of an uncertainty range is the question ofwhetherthepresenceofanElNiñoduringaneruptionsystematicallychangesthecharacteroftheinternalvariability.Ifso,thismightaffecttheuncertaintyrangeonthescalingfactorfornaturalforcing.Theinternalvariabilitysamplesarerealizationsofthevariabilityaroundthemeanstateofaclimateintheabsenceofexternalforcing.Usually,samplesfromcontrolsimulations (e.g., ‘piControl’ simulations) are used for this. In the case here, where themodel fingerprints have been subsampled to only include thosewith the observed ENSOphaseduringpost-eruptionwinters, the internalvariabilitymightbedifferent than in theunconstrainedcase.Toinvestigatethis,wesubsamplethe‘piControl’samplesinexactlythesamewayasinthe‘historical’and‘historicalNat’simulations(FigureS5b).Unsurprisingly,theglobalmeansurfacetemperatureincreasesfollowingthethreevirtualeruptionsduetothe presence of an El Niño event, which occurs, by definition, in every sample. Cruciallythough,thevariabilityaroundthemeanisnotnoticeablydifferentfromtheunconstrainedcase, as can be shown by subtracting the multi-model mean time series (Figure S5c).Consequently, in the main paper the uncertainties as derived from the unsampled‘piControl’simulationsareusedforsimplicity.

8

FigureS5:GMSTanomaliesfordifferentsamplesof‘piControl’simulations.(a)Unsampled,‘piControl’simulations.(b) ‘piControl’simulationssubsampledtohaveanElNiñoeventinthe first virtual post-eruptionwinter (vertical dashed line). (c) Same as (b) butwith themulti-model mean subtracted. No effect in the variability due to subsampling followingvolcaniceruptionscanbeseen.Notethat thecyclicalvariability is theseasonalcyclewithGMSTinborealwinterbeingmorevariablethaninborealsummer.

9

6ResultswithGFDLmodelsInadditiontotheCESM1LargeEnsemble,weinvestigatethe30-memberGeophysicalFluidDynamicsLaboratory(GFDL)ESM2MLargeEnsemble,setup inaverysimilarwayas theCESM1LargeEnsemble(Rodgersetal.2015).TheGFDLESM2MLargeEnsembleisinitiatedin1950andalsofollowstheconventional‘historical’forcingprotocolofCMIP5.Finally,weinvestigate a 10-member ‘Pacemaker’ ensemblewithGFDLCM2.1, a GFDLmodel versionthat performs very similarly to GFDL-ESM2M (Dunne et al. 2012). This GFDL‘Pacemaker’ensemblewasalsoconfiguredfollowingKosakaandXie2013.FigureS6showsresultsfromtheGFDLESM2Mfree-running30-memberLargeEnsembleaswell as the 10-member GFDL CM2.1 ‘Pacemaker’ ensemble. The 30-member LargeEnsemblehasanaveragecoolingof−0.32°Caftervolcaniceruptions(measuredasthefirstpost-eruptionminimum in GMST). By subsampling the simulations according towhethertheycontainanElNiñointhefirstwinterafteraneruption,itcanbeseenthatthevolcaniccoolingisinhibitedbytheElNiño,therebyimprovingthetimingofthecoolingascomparedtoobservations.However,inalmostallofthesesubsampledsimulationsthereisastrongLaNiña2-3yearsafter theeruption,unlikeobservations, leading toanunrealistic coolingof−0.41°C.ThisbiasiscausedbythesystematicoccurrenceofLaNiñaafteranElNiñointhismodel version (Dunne et al. 2012). Besides being too regular, the ENSO amplitudes arelikelytoostronginthismodel,leadingtounrealisticamplitudesinGMST.UsinginsteadtheGFDL ‘Pacemaker’ simulations, inwhich theENSO chronologymatches that in reality,wefindthatwhilethepacemakerrunsareattheupperendoftheentiremodeldistribution,themodel stilloverestimates thepost-eruptioncooling, indicating that itmay indeedhaveanoverly strong response to volcanic forcing and that the discrepancy with observationscannot be fully explained by the ENSO phase sampling. Investigating the reasons for thisoverestimatedcoolingintheGFDLmodelsisbeyondthescopeofthisstudy.

10

Figure S6: Observed and simulated global mean surface temperature (GMST) anomalyrelativetothemeanofthe5yearsprecedingtheeruptionstartdate(dashedverticalline)averaged over the three recent volcanic eruptions (Agung, El Chichon and Pinatubo). (a)GFDL-ESM2Mlargeensembleand‘Pacemaker’simulations.Theaverageover3eruptionsisshown(seetextfordetails).Thecoloredshadinggivesthe5-95%confidenceinterval.Thenumberof ensemblemembers considered is given inbrackets. Themonthsduringwhichtheblueandredcurvesaresignificantlydifferent(t-test,95%confidence)areindicatedinbrightgreenalongthex-axis.Timeserieshavebeensmoothedwithatriangular1-2-1filterforvisualpurposesonly.

11

7AdditionalfigureforCESM1‘Pacemaker’simulations

FigureS7:CESM1‘Pacemaker’resultsfortheindividualeruptionsof(a,d,g)Agung,(b,e,h) El Chichon, and (c, f, i) Pinatubo. (Top row) Same as Figure 1. (Middle row) Globalmean surface temperature (GMST) response relative to the five years preceding theeruptionmonthfromobservations(black)andtheCESM1‘Pacemaker’simulations(green:shading indicatesminimumandmaximumacross the10-member ‘Pacemaker’ ensemble).Timeserieshavebeensmoothedwithatriangular1-2-1filterforvisualpurposesonly(onefilter iteration for models, and five iterations for observations to make spectralcharacteristics comparable between models and observations). (Bottom row)Temperaturedifference (observationsminus ‘Pacemaker’ ensemblemean)during a threemonthsperiodfollowingtheeruptionstartdate,centeredonthepeakpositiveanomalyinobservations.Numberinbottomrightcornerofeachmapgivesthespatialmeanofthemap,i.e.,howmuchwarmerobservationsarerelativetothe‘Pacemaker’simulations.

12

8AdditionalfigureincludingtheSantaMariaeruptionin1902

FigureS8:GlobalmeansurfacetemperatureresponsetothefoureruptionsofSantaMaria,Agung, El Chichon, and Pinatubo combined. The colored shading gives the 5-95%confidenceinterval.Thenumberofensemblemembersconsideredisgiveninbrackets.Themonths during which the blue and red curves are significantly different (t-test, 95%confidence)areindicatedinbrightgreenalongthex-axis.Timeserieshavebeensmoothedwith a triangular 1-2-1 filter for visual purposes only. Note that the total number ofsimulations (187) is slightly smaller than in Figure 2 (198), as not all simulations cover1902.

13

9Tablesonmodels,simulations,andsubsamplingThe number of simulationswhichmeet the ENSO selection criteria for each of the threevolcaniceruptionsisgiveninTableS1(formoredetailsseeTablesS2andS3).Table S1: Available simulations. Total number of ensemble members available for eachexperiment(column2)andnumberofensemblemembersthatmatchtheobservedENSOstateduringtheborealwinterfollowingtheeruption(columns3to5).Experiment Numberofensemblemembers

All Agung ElChichon PinatuboCMIP5historical 198 35 55 42CMIP5historicalNat 68 17 20 13CESM1 LargeEnsemble

40 9 12 7

GFDLLargeEnsemble 30 8 3 1TableS2:‘historical’simulationsusedforentireanalysis(column2)andthosethatmatchtheobservedENSOevolutionforindividualvolcaniceruptions(columns3to5).FGOALS-g2 and the CMCCmodels were excluded, as they do not include volcanic forcing for thehistoricalperiod. EnsemblemembershistoricalModel All Agung ElChichon PinatuboACCESS1-0 r1i1p1 ACCESS1-0 r2i1p1 ACCESS1-0 r3i1p1 r3i1p1 ACCESS1-3 r1i1p1 r1i1p1 ACCESS1-3 r2i1p1 ACCESS1-3 r3i1p1 r3i1p1bcc-csm1-1 r1i1p1 r1i1p1bcc-csm1-1 r2i1p1 r2i1p1 bcc-csm1-1 r3i1p1 r3i1p1bcc-csm1-1-m r1i1p1 bcc-csm1-1-m r2i1p1 r2i1p1 bcc-csm1-1-m r3i1p1 BNU-ESM r1i1p1 r1i1p1 r1i1p1CanESM2 r1i1p1 r1i1p1 CanESM2 r2i1p1 r2i1p1 CanESM2 r3i1p1 CanESM2 r4i1p1 CanESM2 r5i1p1 r5i1p1 CCSM4 r1i1p1 r1i1p1 CCSM4 r1i2p1

14

CCSM4 r1i2p2 r1i2p2 CCSM4 r2i1p1 CCSM4 r3i1p1 CCSM4 r4i1p1 CCSM4 r5i1p1 r5i1p1 CCSM4 r6i1p1 r6i1p1 CESM1-BGC r1i1p1 CESM1-CAM5 r1i1p1 CESM1-CAM5 r2i1p1 CESM1-CAM5 r3i1p1 r3i1p1 CESM1-CAM5-1-FV2 r1i1p1 CESM1-CAM5-1-FV2 r2i1p1 CESM1-CAM5-1-FV2 r3i1p1 r3i1p1CESM1-CAM5-1-FV2 r4i1p1 CESM1-FASTCHEM r1i1p1 CESM1-FASTCHEM r2i1p1 r2i1p1 CESM1-FASTCHEM r3i1p1 CESM1-WACCM r1i1p1 r1i1p1 CESM1-WACCM r2i1p1 CESM1-WACCM r3i1p1 CESM1-WACCM r4i1p1 CESM1-WACCM r5i1p1 CESM1-WACCM r6i1p1 r6i1p1 CESM1-WACCM r7i1p1 CNRM-CM5 r10i1p1 r10i1p1 CNRM-CM5 r1i1p1 CNRM-CM5 r2i1p1 r2i1p1 CNRM-CM5 r3i1p1 r3i1p1 r3i1p1 r3i1p1CNRM-CM5 r4i1p1 CNRM-CM5 r5i1p1 r5i1p1 r5i1p1 r5i1p1CNRM-CM5 r6i1p1 CNRM-CM5 r7i1p1 r7i1p1 CNRM-CM5 r8i1p1 r8i1p1 r8i1p1CNRM-CM5 r9i1p1 CNRM-CM5-2 r1i1p1 r1i1p1 CSIRO-Mk3-6-0 r10i1p1 CSIRO-Mk3-6-0 r1i1p1 CSIRO-Mk3-6-0 r2i1p1 r2i1p1 r2i1p1CSIRO-Mk3-6-0 r3i1p1 r3i1p1 CSIRO-Mk3-6-0 r4i1p1 r4i1p1 r4i1p1CSIRO-Mk3-6-0 r5i1p1

15

CSIRO-Mk3-6-0 r6i1p1 CSIRO-Mk3-6-0 r7i1p1 CSIRO-Mk3-6-0 r8i1p1 r8i1p1 CSIRO-Mk3-6-0 r9i1p1 CSIRO-Mk3L-1-2 r1i2p1 CSIRO-Mk3L-1-2 r2i2p1 r2i2p1CSIRO-Mk3L-1-2 r3i2p1 r3i2p1 EC-EARTH r12i1p1 r12i1p1 EC-EARTH r1i1p1 EC-EARTH r2i1p1 EC-EARTH r6i1p1 r6i1p1EC-EARTH r8i1p1 r8i1p1EC-EARTH r9i1p1 FIO-ESM r1i1p1 FIO-ESM r2i1p1 r2i1p1 FIO-ESM r3i1p1 r3i1p1 r3i1p1GFDL-CM2p1 r10i1p1 GFDL-CM2p1 r1i1p1 GFDL-CM2p1 r2i1p1 GFDL-CM2p1 r3i1p1 r3i1p1 GFDL-CM2p1 r4i1p1 r4i1p1 r4i1p1GFDL-CM2p1 r5i1p1 GFDL-CM2p1 r6i1p1 GFDL-CM2p1 r7i1p1 GFDL-CM2p1 r8i1p1 GFDL-CM2p1 r9i1p1 r9i1p1GFDL-CM3 r1i1p1 GFDL-CM3 r2i1p1 GFDL-CM3 r3i1p1 r3i1p1 GFDL-CM3 r4i1p1 r4i1p1 r4i1p1 GFDL-CM3 r5i1p1 r5i1p1GFDL-ESM2G r1i1p1 GFDL-ESM2M r1i1p1 r1i1p1 r1i1p1GISS-E2-H r1i1p1 GISS-E2-H r1i1p2 r1i1p2GISS-E2-H r1i1p3 r1i1p3 r1i1p3GISS-E2-H r2i1p1 GISS-E2-H r2i1p2 r2i1p2 GISS-E2-H r2i1p3 GISS-E2-H r3i1p1 GISS-E2-H r3i1p2 r3i1p2

16

GISS-E2-H r3i1p3 GISS-E2-H r4i1p1 GISS-E2-H r4i1p2 GISS-E2-H r4i1p3 r4i1p3 GISS-E2-H r5i1p1 r5i1p1 GISS-E2-H r5i1p2 GISS-E2-H r5i1p3 GISS-E2-H r6i1p1 r6i1p1GISS-E2-H r6i1p2 GISS-E2-H r6i1p3 r6i1p3GISS-E2-H-CC r1i1p1 r1i1p1 r1i1p1 GISS-E2-R r1i1p1 GISS-E2-R r1i1p12

1

GISS-E2-R r1i1p122

GISS-E2-R r1i1p124

r1i1p124 r1i1p124

GISS-E2-R r1i1p125

r1i1p125

GISS-E2-R r1i1p126

GISS-E2-R r1i1p127

r1i1p127

GISS-E2-R r1i1p128

r1i1p128

GISS-E2-R r1i1p2 GISS-E2-R r1i1p3 GISS-E2-R r2i1p1 GISS-E2-R r2i1p2 r2i1p2GISS-E2-R r2i1p3 r2i1p3 GISS-E2-R r3i1p1 r3i1p1 r3i1p1GISS-E2-R r3i1p2 r3i1p2 GISS-E2-R r3i1p3 r3i1p3 GISS-E2-R r4i1p1 r4i1p1GISS-E2-R r4i1p2 r4i1p2 GISS-E2-R r4i1p3 GISS-E2-R r5i1p1 GISS-E2-R r5i1p2 r5i1p2GISS-E2-R r5i1p3 GISS-E2-R r6i1p1 r6i1p1 GISS-E2-R r6i1p2 r6i1p2

17

GISS-E2-R r6i1p3 r6i1p3 GISS-E2-R-CC r1i1p1 r1i1p1 HadCM3 r10i1p1 r10i1p1HadCM3 r1i1p1 HadCM3 r2i1p1 HadCM3 r3i1p1 HadCM3 r4i1p1 r4i1p1 HadCM3 r5i1p1 r5i1p1 r5i1p1 HadCM3 r6i1p1 HadCM3 r7i1p1 r7i1p1 r7i1p1HadCM3 r8i1p1 r8i1p1 r8i1p1 r8i1p1HadCM3 r9i1p1 r9i1p1 r9i1p1HadGEM2-AO r1i1p1 HadGEM2-CC r1i1p1 r1i1p1 HadGEM2-ES r1i1p1 HadGEM2-ES r2i1p1 HadGEM2-ES r3i1p1 r3i1p1 HadGEM2-ES r4i1p1 r4i1p1 HadGEM2-ES r5i1p1 inmcm4 r1i1p1 r1i1p1IPSL-CM5A-LR r1i1p1 r1i1p1 r1i1p1 IPSL-CM5A-LR r2i1p1 r2i1p1IPSL-CM5A-LR r3i1p1 r3i1p1 IPSL-CM5A-LR r4i1p1 IPSL-CM5A-LR r5i1p1 r5i1p1 IPSL-CM5A-LR r6i1p1 r6i1p1 IPSL-CM5A-MR r1i1p1 r1i1p1 IPSL-CM5A-MR r2i1p1 r2i1p1IPSL-CM5A-MR r3i1p1 IPSL-CM5B-LR r1i1p1 MIROC4h r1i1p1 r1i1p1MIROC4h r2i1p1 r2i1p1 MIROC4h r3i1p1 MIROC5 r1i1p1 MIROC5 r2i1p1 MIROC5 r3i1p1 r3i1p1 MIROC5 r4i1p1 MIROC5 r5i1p1 MIROC-ESM r1i1p1 MIROC-ESM r2i1p1 MIROC-ESM r3i1p1 r3i1p1 r3i1p1

18

MIROC-ESM-CHEM r1i1p1 MPI-ESM-LR r1i1p1 r1i1p1 MPI-ESM-LR r2i1p1 MPI-ESM-LR r3i1p1 r3i1p1 MPI-ESM-MR r1i1p1 MPI-ESM-MR r2i1p1 r2i1p1MPI-ESM-MR r3i1p1 r3i1p1 MPI-ESM-P r1i1p1 r1i1p1 r1i1p1MPI-ESM-P r2i1p1 r2i1p1MRI-CGCM3 r1i1p1 r1i1p1 r1i1p1MRI-CGCM3 r2i1p1 r2i1p1 r2i1p1MRI-CGCM3 r3i1p1 r3i1p1MRI-CGCM3 r4i1p2 MRI-CGCM3 r5i1p2 r5i1p2 MRI-ESM1 r1i1p1 r1i1p1NorESM1-M r1i1p1 NorESM1-M r2i1p1 r2i1p1 NorESM1-M r3i1p1 r3i1p1 NorESM1-ME r1i1p1 r1i1p1 NorESM1-ME r1i1p2 r1i1p2 Totalnumber 198 35 55 42Table S3: historicalNat simulations used for entire analysis (column 2) and those thatmatchtheobservedENSOevolutionforindividualvolcaniceruptions(columns3to5). EnsemblemembershistoricalNatModel All Agung El

ChichonPinatubo

bcc-csm1-1 r1i1p1 BNU-ESM r1i1p1 CanESM2 r1i1p1 r1i1p1 CanESM2 r2i1p1 r2i1p1 r2i1p1 CanESM2 r3i1p1 r3i1p1 r3i1p1 CanESM2 r4i1p1 r4i1p1 r4i1p1CanESM2 r5i1p1 r5i1p1 CCSM4 r1i1p1 CCSM4 r2i1p1 r2i1p1 CCSM4 r4i1p1 CCSM4 r6i1p1 r6i1p1CESM1-CAM5 r1i1p1 CESM1-CAM5 r2i1p1 r2i1p1

19

CESM1-CAM5 r3i1p1 CESM1-CAM5-1-FV2 r1i1p1 r1i1p1 CESM1-CAM5-1-FV2 r3i1p1 r3i1p1 CESM1-CAM5-1-FV2 r4i1p1 CNRM-CM5 r1i1p1 CNRM-CM5 r2i1p1 r2i1p1 r2i1p1CNRM-CM5 r3i1p1 CNRM-CM5 r4i1p1 r4i1p1 CNRM-CM5 r5i1p1 CNRM-CM5 r8i1p1 CSIRO-Mk3-6-0 r1i1p1 CSIRO-Mk3-6-0 r2i1p1 r2i1p1CSIRO-Mk3-6-0 r3i1p1 CSIRO-Mk3-6-0 r4i1p1 CSIRO-Mk3-6-0 r5i1p1 GFDL-CM3 r1i1p1 r1i1p1GFDL-CM3 r3i1p1 r3i1p1 GFDL-CM3 r5i1p1 r5i1p1GFDL-ESM2M r1i1p1 r1i1p1 GISS-E2-H r1i1p1 r1i1p1 GISS-E2-H r1i1p3 r1i1p3 r1i1p3GISS-E2-H r2i1p1 GISS-E2-H r2i1p3 r2i1p3 r2i1p3GISS-E2-H r3i1p1 r3i1p1 GISS-E2-H r3i1p3 r3i1p3 r3i1p3 GISS-E2-H r4i1p1 GISS-E2-H r4i1p3 GISS-E2-H r5i1p1 GISS-E2-H r5i1p3 GISS-E2-R r1i1p1 r1i1p1 GISS-E2-R r1i1p3 r1i1p3 GISS-E2-R r2i1p1 GISS-E2-R r2i1p3 r2i1p3 GISS-E2-R r3i1p1 GISS-E2-R r3i1p3 r3i1p3 r3i1p3 GISS-E2-R r4i1p1 r4i1p1 GISS-E2-R r4i1p3 r4i1p3 GISS-E2-R r5i1p1 GISS-E2-R r5i1p3 r5i1p3HadGEM2-ES r1i1p1 r1i1p1 HadGEM2-ES r2i1p1

20

HadGEM2-ES r3i1p1 r3i1p1HadGEM2-ES r4i1p1 r4i1p1 IPSL-CM5A-LR r1i1p1 IPSL-CM5A-LR r2i1p1 r2i1p1 IPSL-CM5A-LR r3i1p1 r3i1p1 r3i1p1IPSL-CM5A-MR r1i1p1 r1i1p1IPSL-CM5A-MR r2i1p1 IPSL-CM5A-MR r3i1p1 r3i1p1 MIROC-ESM r1i1p1 r1i1p1 MIROC-ESM r2i1p1 MIROC-ESM r3i1p1 r3i1p1 MIROC-ESM-CHEM r1i1p1 r1i1p1 MRI-CGCM3 r1i1p1 r1i1p1 NorESM1-M r1i1p1 r1i1p1Totalnumber 68 17 20 13References

Crowley,T.J.,andM.B.Unterman,2013:Technicaldetailsconcerningdevelopmentofa1200yrproxyindexforglobalvolcanism.EarthSyst.Sci.Data,5,187–197,doi:10.5194/essd-5-187-2013.

Dunne,J.P.,andCoauthors,2012:GFDL’sESM2GlobalCoupledClimate–CarbonEarthSystemModels.PartI:PhysicalFormulationandBaselineSimulationCharacteristics.J.Clim.,25,6646–6665,doi:10.1175/JCLI-D-11-00560.1.

Kosaka,Y.,andS.-P.Xie,2013:Recentglobal-warminghiatustiedtoequatorialPacificsurfacecooling.Nature,501,403–407,doi:10.1038/nature12534.http://www.ncbi.nlm.nih.gov/pubmed/23995690.

Li,J.,andCoauthors,2013:ElNiñomodulationsoverthepastsevencenturies.Nat.Clim.Chang.,3,822–826,doi:10.1038/nclimate1936.http://www.nature.com/doifinder/10.1038/nclimate1936\nhttp://www.nature.com/nclimate/journal/v3/n9/full/nclimate1936.html(AccessedAugust10,2014).

Rodgers,K.B.,J.Lin,andT.L.Frölicher,2015:EmergenceofmultipleoceanecosystemdriversinalargeensemblesuitewithanEarthsystemmodel.Biogeosciences,12,3301–3320.