HAL Id: hal-01684042
https://hal.sorbonne-universite.fr/hal-01684042
Submitted on 15 Jan 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Distributed under a Creative Commons Attribution| 4.0 International License
Tentative reconstruction of the 1998–2012 hiatus in global temperature warming using the IPSL–CM5A–LR
climate model
Didier Swingedouw, Juliette Mignot, Éric Guilyardi, Sébastien Nguyen, Lola Ormières
To cite this version:
Didier Swingedouw, Juliette Mignot, Éric Guilyardi, Sébastien Nguyen, Lola Ormières. Ten- tative reconstruction of the 1998–2012 hiatus in global temperature warming using the IPSL–
CM5A–LR climate model. Comptes Rendus Géoscience, Elsevier, 2017, 349 (8), pp.369 - 379.
�10.1016/j.crte.2017.09.014�. �hal-01684042�
Review paper
Tentative reconstruction of the 1998–2012 hiatus
in global temperature warming using the IPSL–CM5A–LR climate model
Didier Swingedouw
a,*, Juliette Mignot
b, Eric Guilyardi
b, Se´bastien Nguyen
b, Lola Ormie`res
aaEnvironnementsetpale´oenvironnementsoce´aniquesetcontinentaux(EPOC),UMRCNRS5805EPOC–OASU–universite´ deBordeaux, alle´eGeoffroy-Saint-Hilaire,33615Pessac,France
bLOCEAN/IPSL(Sorbonneuniversite´s,UPMC–CNRS–IRD–MNHN),4,placeJussieu,75005Paris,France
1. Introduction
Thesurfacetemperatureaveragedoverthewholeglobe hasunequivocallyincreasedoverthelast century(IPCC, 2013). It is plotted in Fig. 1, using the HadCRUT4 compilation(Moriceetal.,2012)forillustration.Thereis neverthelesssubstantialvariabilityaroundtheveryclear positivetrendovertheperiod1870–2015,withdecadesof acceleratedincreaseandothersofdecreasingtendencies.
Thesedecadalfluctuationsareprimarilythemanifestation ofthenaturalvariability,whichisrelatedtotheeffectof natural externalforcingoftheclimatesystem, likesolar
irradiancevariationsandvolcaniceruptions,aswellasto the intrinsic variability of the ocean–atmosphere–ice coupled system. The climate system indeed fluctuates, evenifexternalforcingdoesnotvary.
The anthropogenic forcing of the climate system, mainly associated with greenhouse gas concentrations andaerosolsemissions,isevolvingverysmoothly(Fig.2).
Itcanthushardlyexplainsuchdecadalfluctuations,except maybefor the rapid increase in aerosols release at the beginning of the 1960s, which may partly explain the coolingexperiencedatthatmoment(Stottetal.,2000),as wellastheincreaseinaerosolsemissionsinAsiafromthe 1990s,whichmayinducecomplexteleconnectionpatterns (Smithetal.,2016).Naturalvariabilityotherwiseremains themaincandidateforexplainingthedecadalfluctuations occurringontopofthelong-termwarmingtrend.
ARTICLE INFO
Articlehistory:
Received28September2017
Acceptedafterrevision28September2017 Availableonline15November2017 HandledbyVincentCourtillot
Keywords:
Climateandoceandynamics Globalwarming
Hiatus
Pacemakerexperiments Climatemodels Surfacenudging Globaltemperaturetrends
ABSTRACT
Theperiod running from1998 to 2012 hasexperienceda slowerincreasein global temperatureatthesurfaceoftheEarththanthedecadesbefore.Severalexplanationshave beenproposed,rangingfrominternalvariabilityoftheclimatesystemtoacontributionof thenaturalexternalforcing.Inthisstudy,weusetheIPSL-CM5A-LRclimatemodeltotest these different hypotheses. We consider historical simulations, including observed externalforcing,inwhichnudgingtowardsobservedseasurfacetemperaturehasbeen appliedtodifferentregionsoftheoceantophasethedecadalvariabilityoflarge-scale modesintheAtlanticandthePacifictoobservations.Wefindthatphasingthetropical Pacificisreducingthewarmingtrenddetectedinhistoricalsimulationsbyafactoroftwo, buttheremainingtrendisstilltwiceaslargeastheobservedone.Combiningthetropical Pacificphasingandthepotentialeffectofrecenteruptionsallowsustofullyreproducethe observedhiatus.Conversely,nudgingtheAtlanticdoesnotdriveanyhiatusinthismodel.
C 2017Acade´miedessciences.PublishedbyElsevierMassonSAS.Thisisanopenaccess articleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/
4.0/).
* Correspondingauthor.
E-mailaddress:didier.swingedouw@u-bordeaux.fr(D.Swingedouw).
ContentslistsavailableatScienceDirect
Comptes Rendus Geoscience
w ww . sc i e nce d i re ct . co m
https://doi.org/10.1016/j.crte.2017.09.014
1631-0713/C 2017Acade´miedessciences.PublishedbyElsevierMassonSAS.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://
creativecommons.org/licenses/by-nc-nd/4.0/).
Theperiodbetween1998and2012experienceda particularly weak warming trend as compared to the precedingdecades,whiletheanthropogenicforcingwasas strongasduringthepreviousdecade,ifnotstronger.This eventhasbeendubbedasahiatus,giventhatthesimple directresponsetoanthropogenicincreaseingreenhouse gasesemissionsshouldhavebeenaquasi-linearwarming overthelast30years.Manystudieshavetriedtoattribute iteithertonaturalvariabilityortoanthropogenicaerosols emissions. In this sense, this fluctuation has been a stimulatingtestbedtoevaluateourlevelofunderstanding ofthesourcesoffluctuationsoftheclimatesystem.
Severalinterestinghypotheseshaveemergedtoexplain thishiatusperiod.Santeretal.(2014)haveproposedthat theaccumulationof small(tropospheric) volcanicerup- tionsover thisperiodhasledtoa significantloadingof aerosolsintheatmosphere,whichhasincreaseditsalbedo and thus reflected a large amount of solar short-wave radiation.ThiseffectmayhavemitigatedtheEarthsurface temperatureincreasedue totheincrease ingreenhouse gasesconcentrationsintheatmosphere.Inadditiontothis naturalfluctuationintheexternalforcing,otheranalyses have scrutinized the specific spatial pattern of the temperature changes over this time period (Kaufmann etal.,2011;KosakaandXie,2013).Theyhavehighlighted thefactthat thePacificbasin, whichaccountsformore
thanathirdoftheEarthsurface,hasexperiencedacooling trendinthe2000sinmanylocations,followingapattern resemblinga well-knownmodeofdecadalvariability in thisbasin,theso-calledPacificDecadalOscillation(PDO, Mantuaetal.,1997).Meehletal.(2011)showedthat,in climate modelsprojections ofthe 21stcentury,periods withsmalldecadalwarmingtrendsaretypicallycharac- terizedbyanegativephaseofthePDO.Indeed,duringsuch phases, the equatorial upwelling of cold deep water is enhanced, while warmwaters from the western warm poolsubduct.Consequently,theadditionnalheatreceived intheEarthsystemduetoincreasedgreenhouseeffectis buriedinsubsurface,whilethedeeperoceanisproviding coldwateratthesurface,whichcoolsalargepartofthe upperPacificbasin(Balmasedaetal.,2013;Douvilleetal., 2015).ThisspecificstateinthePacificOceaninthe2000s hasbeenrelatedwithabnormallystrongeasterlywindsin thewesternpartofthetropicalPacificOcean,whichmay ultimatly explainthe adjustmentof this basin(England et al., 2014) and the negative phase of the PDO.
Nevertheless,theoriginofthisextremeanomalouswind remainsunknown.
Kosaka and Xie (2013) have shown that restoring a climatemodeltowardstheobservedseasurfacetempera- ture(SST)trendinasmallregionoftheeasternequatorial Pacific only representing 8.2% of the Earth surface is sufficienttoreproducemostofthespatialfeaturesofthe anomalies of the hiatus period over the whole Pacific, Fig. 1. (a) Monthly mean of global surface temperature anomalies
(referenceperiodisthewholeperiod)fromHadCRUT4untilMay2015.
(b)Fifteen-yeartrendsince1850fromHadCRUdatausingaone-year slidingwindowtocomputethelineartrendfrommonthlymeandata.The redlinein(b)isathree-yearrunningmeanoftheblackline.
Fig.2.(a)Externalforcingappliedinthedifferentsimulations(seeTable 1),and(b)definitionoftheregionwherepartialnudginghasbeenapplied (tropicalEastPacificingreen,Atlanticinblue).
includingwindvariations.Indeed,thechangeinthezonal temperaturegradient of thetropicalbasinactivatesthe Bjerknesfeedback,amechanismbywhichtheeast–west SSTgradientinthetropicalareadrivesthesurfacewinds, which then feed back positively on the temperature gradient.Furthermore,theassociatedmodificationofthe Walker cell, the zonal longitude-altitude atmospheric large-scalerotatingtropicalcell,canalsoimpactthedeep atmosphericconvectionzoneinthewesternPacific.This provides a vorticitysourcetoexcite Rossbywaves,that leadtoteleconnectionpatternstowardsthehighlatitudes.
Thisnotablyexplainsthecoolingpatternobservedinthe Northeast Pacific during the hiatus (similarly to what happensduringaLaNin˜ aevent).KosakaandXie(2013) thusshowedthatactivatingakeyaspectofthischainof coupledmechanismscanbesufficienttoreproducemost of theobservedpattern.In thiscontext,a continuumof stochastic perturbations of the wind field fed by tele- connection and then oceanic adjustment in the high latitudes can be sufficient to induce low-frequency variabilityinthePacificbasin(Newmanetal.,2016).
Another idea that hasbeen proposed toexplain the increaseofeasterlywindinthePacificisbasedoninter- oceanicbasininteractionsinthetropicalband.McGregor etal.(2014)arguedthatthePacificbasinmayhavebeen influenced by the large warming experienced in the tropicalAtlanticinthe2000s,whichwouldhavestrongly increased the atmospheric deep convection over the AtlanticOcean,andtherebymodifiedremotelytheWalker circulationoverthePacific.Nevertheless,theexperimental design theyused to test this hypothesis wasbased on atmospheric-only model simulations or simulations partiallycoupledwithamixedlayeroceanicmodel.These simplifications maysignificantly affect theexact repre- sentation of the mechanisms at play, and notably the ocean-atmosphere couplingat therootofthesemecha- nisms.Thus,whileMcGregoretal.(2014)arguedthatthe Atlantic could be the general pacemaker of the multi- decadalvariabilityovertheglobe,Trenberthetal.(2014) showedthatvariationsinthePacificOceancouldalsovery wellimpacttheAtlanticOcean,notably throughRossby waves, highlighting theinherent coupling between the differentbasins.
Finally, Smithet al.(2016) recentlyshowedthatthe hiatusperiodwasalsorelatedtothedelayedrecoveryof theclimatesystemfromtheeruptionofMountPinatubo that occurred in1991, and tothemorerecent effectof regional changes in the distribution of anthropogenic aerosolswhichlikelyinfluencedthePDO(seeTakahashi andWatanabe,2016).
Someofthedifferentsourcesofmultidecadalvariabili- tydescribedabovehavebeenshowntoaddupinorderto quantitatively represent the recent hiatus (Huber and Knutti,2014;MarotzkeandForster,2014),buttheexact originofthePacificvariationsisstillnotclearandrequires further analysis to decipher and quantify the best hypothesestoexplainit.
In this paper, we attempt to reproduce the hiatus characteristics using classical nudging techniques of SST anomalies applied in various versions, and using different representations of the external forcing in the
IPSL–CM5A–LR model. This allows us to evaluate the robustness of the different mechanisms proposed to explain the hiatus period in an independent state-of- the-artclimatemodel.
2. Data,modelandmethods 2.1. Descriptionofthemodel
The ocean–atmosphere coupled model used in this studyistheIPSL-CM5A(Dufresneetal.,2013)initslow- resolution (LR) version as developed for CMIP5. The atmosphericmodelisLMDZ5(Hourdinetal.,2006)with a9696L39regulargrid(horizontalresolutionaround 2.58 in latitude and 3.75 in longitude) and the oceanic modelis NEMO(Madec,2008)withan182149L31 non-regular grid (horizontal resolution around 28, with refinementupto0.58notablyattheequator),inversion 3.2, including the LIM-2 sea ice model (Fichefet and Morales Maqueda, 1997) and the PISCES (Aumont and Bopp,2006)moduleforoceanicbiogeochemistry.
2.2. Simulations
The historical simulations use a prescribed external radiativeforcingdeducedfromtheobservedincreasein greenhousegasesandaerosolconcentrationsaswellasthe ozone changes (Fig.2a) and the land-usemodifications (notshown,seeDufresneetal.,2013).Theyalsoinclude estimates of solar irradiance variations and of tropical stratosphericvolcaniceruptions,representedasadecrease inthetotalsolarirradiance(dependingontheintensityof thevolcaniceruptions,Fig.2a,cf.Dufresneetal.,2013), overthehistorical period,here defined as[1850–2005].
Wehave runanensembleoffivehistorical simulations, differingbytheir initialconditions,takenfromdifferent dates of a 1000-year control simulation under pre- industrial conditions, each separated by 10 years. This control pre-industrial simulation is itself starting after thousandsofyearsofspin-upprocedure.
Eachhistoricalsimulationisextended after2005fol- lowingaRCP4.5emissionscenario.Thesesimulationsare projectionsaimingatevaluatingthepotentialimpactof futureanthropogenicemissionsonclimateuntil2100.To accountforpossiblefuturevolcaniceruptions,whichmay cooltheclimate,aconstantbackgroundvolcanicforcing hasbeen addedto theexternal forcing starting in year 2006.Itrepresentstheaverageofvolcanicforcingoverthe period1860–2000,andisequalto 0.25W/m2ofglobal radiative forcing. This anomalous forcing due to back- groundvolcaniceruptionisnotaimedtoberealisticonthe 2006–2012periodanddoesnotincludeanyinformation onthetimingoftheobservedforcingeffectfromvolcanic eruptionsoverthistimeframe,asdescribedinSanteretal.
(2014).Nevertheless,thecumulativeradiativeforcingover theanalysed period isof thesame orderof magnitude, sinceSanteretal.(2014)evaluatedtheradiativeimpactof tropicaleruptionstobearound0.25W/m2perdecadefora trendcomputedoverJanuary2001toDecember2012.The historical plus RCP45 scenario including background
volcanicforcingensembleiscalledHisRbginthefollowing (cf. Table1)and comprisesfivemembers.Thesearethe historical simulations that have been included in the CMIP5 database. To test the effect of the recent weak volcaniceruptionsontheclimatevariability,andalsothe effectoftheartificialshiftofradiativeforcingof 0.25W/
m2 imposed in 2006, we have also performed RCP4.5 simulationswherethebackgroundvolcaniceruptionsare notconsidered,namedHisRnobg(cf.Table1).
Thefive-memberensembleofnudgedsimulationsover thewholeocean(exceptbelowseaice),calledNudGlo,has thesameforcingasHisRnobg,andincludesanudgingterm towards the observed anomalous monthly SST (Smith etal.,2008).Eachsimulationstartson the1stofJanuary 1949 from one of the historical simulations presented above(Fig.2).Thenudgingtechniqueconsistsinaddinga heatfluxtermQtotheSSTequationundertheformQ=
g
(SST’mod SST’obs), whereSST’mod standsforthemodelled anomalousSSTateachtimestepandgridpoint,andSST’obs
for the anomalous observed SST (Smith et al., 2008).
Anomalies are computed with respect to the monthly climatology of SST over the period 1949–2005 in the corresponding historical simulation and in theobserva- tions, respectively. We use a restoring coefficient
g
of 40 Wm 2K 1 corresponding to a physically based (Frankignoul and Kestenare,2002)relaxingtimescaleof around 60 days over a 50 m-deep mixed layer (see Swingedouwetal.,2013orOrtegaetal.,2017forfurther details).Strongervaluesofg
,asusedinmanyotherstudies (Keenlysideetal.,2008;KosakaetandXie,2013;Luoetal., 2005; Pohlmann et al., 2009) have the potential to distortthe higher-frequencyocean–atmosphere interac- tion (Cassou, pers. comm.) or tocreate spurious water massesintheocean.Inthatsense,this studyallowsthe evaluationofnewmodelresultsobtainedwiththischoice ofg
aswillbeusedinupcomingintercomparisonprojects (cf.Boeretal.,2016).Wealsoconsiderpartiallynudgedsimulationswhere we use similar experimental setup as in NudGlo, but imposingtherestoringonlyinlimitedoceanicareas.We first perform a similar experiment as Kosaka and Xie (2013),whereSSTnudgingisonlyappliedinthetropical EastPacific(called NudPac). Therestoringconstant used here (40 Wm 2K 1, as in NudgGlo) is nevertheless six timeslowerthantheoneusedinKosakaandXie(2013).To evaluatethepotentialimpactoftheAtlanticbasin,wealso producesimulations(calledNudAtl)withsimilarexternal forcingandpartialnudging,butonlyappliedintheAtlantic basin (cf. Fig. 2b). Both NudPac and NudAtl types of simulationsaremainlydesignedtoreproducethehiatus period.Therefore,theystartin1990fromglobalnudged
simulations.TheareaoftheEarthsurfaceconcernedbythe restoringterminthedifferentexperimentisaround70%of theglobeinNudGlo,8%inNudPacand15%inNudAtl.
Weperformafive-memberensembleforeachtypeof simulation. This is designed to remove most of the remaininginternalvariabilityfromthemodelsimulations.
Indeed,eventhoughourpartiallynudgedsimulationsare tryingtocapturethenaturalandinternalvariabilityfrom therealsystem,eachsimulationisalsoaffectedbyitsown internalvariability,arisingfromtheatmosphere,thedeep ocean,andnon-nudgedregions,whicharenottheonesof interesthere.
2.3. Observationaldataproducts
Tocompareoursimulationswithrecenttrends,weuse the HadCRUT4 surface temperature reconstruction (Moriceetal.,2012).ThisreconstructionusesSeaSurface Temperature from HadSST3 (Kennedy et al., 2011) and atmospheric temperatureover landat around2mfrom CRUTEM4(OsbornandJones,2014).Thesedatasetshave beendevelopedbytheClimaticResearchUnit(University ofEastAnglia)inconjunctionwiththeHadleyCentre(UK MetOffice).Theyaregivenon a5858grid,wheregrid pointswherenotenoughdataareavailableareassignedto the ‘‘unknown’’ value.To assess the uncertainty of this reconstruction,weuseafive-memberensembleprovided byHadCRUT4.
Toanalysetheatmosphericcirculationchangesduring the hiatus period, we use the recent 20th-century reanalysis (20CR)Projectversion2(Compoetal.,2011), consisting of an ensemble of 56 reconstructions with 2828 gridded 6-hourly weather data from 1871 to 2010. Producing such a large ensemble is aimed at removing the internal variability from the model and better stick to the observed signals. Each ensemble member was performed using the NCEP/GFS (National CenterforEnvironmentalPrediction/GlobalForecastSys- tem) atmospheric model, prescribing the monthly sea surfacetemperatureandseaicechangesfromHadISSTas boundaryconditions,andassimilatingsealevelpressure data from the International Surface Pressure Databank version2(http://www.rda.ucar.edu/datasets/ds132.0).We usetheensemblemeantoperformalltheanalysis.
3. Results
3.1. Analysisoftheglobal15-yeartrends
To compare the simulated temperature to available temperatureobservationsand,in particular,tocorrectly Table1
Descriptionofthefive-memberensemblesimulations.
Simulations Name Period Forcing Restoring
Historical+rcp45withoutbgVol. HisRnobg 1850–2030 Allbutbackgroundvolcaniceruptionsfrom2006 No NudgedGlob. NudGlo 1949–2015 Allbutbackgroundvolcaniceruptionsfrom2006 GlobalSST
NudgedPac. NudPac 1991–2012 Allbutbackgroundvolcaniceruptionsfrom2006 TropicalEastPacificSST NudgedAtl. NudAtl 1991–2012 Allbutbackgroundvolcaniceruptionsfrom2006 WholeAtlantic Historical+rcp45withbgVol. HisRbg 1850–2030 Allincludingbackgroundvolcaniceruptionsin2006 No
account for missing data in theobservations in spatial average, weinterpolate themodel outputson thesame gridasHadCRUT4.Furthermore,sinceHadCRUT4dataare representative ofSST when locatedover the ocean and atmospheric2-metertemperaturefortherestoftheglobe (Richardsonetal.,2016),weaccountforthisinourmodel datacomparison. Thus, weusesimulatedSSTwhenthe consideredgridpointislocatedovertheoceanmaskand 2matmospherictemperaturewhenlocatedovertheland oricemask.
Fig.1ashowsthattheglobaltemperatureatthesurface oftheEarthisvaryingatthedecadalscalesince1850.The 15-yearlineartrendcomputedwithaslidingwindowof one year (Fig. 1b) illustrates this large multi-decadal variability. Although it is clear from Fig. 1b that in general, the 15-year trend of global temperature is positive,theperiodfromthe1940stothe1970sexhibits negative15-yeartrends.Thisperiodhasbeenanalysedby Booth et al. (2012), who related it notably with large anthropogenicaerosolsemissionsatthistime.Afterwards, thetrendsstronglyincreaseandarealwayspositive.The recentperiod1998–2012isshowingtheweakest15-year trendsince1980.Thisiswhatwewilldefineasthehiatus periodintherestofthepaper.Inthefollowing,wewill thereforemainlyfocusonthetrendoverthisperiod,trying tounderstandthedynamicalprocessesthatcanexplainits lowvalue.
The time evolution of the global mean surface temperature simulated for the ensemble mean of the differentmodelingset-upsisplottedinFig.3a.Itisclear thatthetrendoverthe1998–2012periodislargestinthe free fully coupled HisRcp ensemble, while it is most reduced in the globally constrained NudGlo ensemble, enclosing in its uncertainty the HadCRUT4 dataset.
PartiallynudgedsimulationsNudPacisinbetweenwhile the NudAtl ensemble is very similar to the historical simulations.Thehiatusperiodisthusbestreproducedin the NudGlo ensemble, consistently with the strongest constraint ontheobservedSSTs imposedin thissetup, whileNudAtlandHisRcpreproduceittheleast.
The linear trend over the 1998–2012 period is extracted for each set of simulations and shown in Fig.3b,togetherwiththeassociatederrorbar.Itconfirms the visual inspection of Fig. 3a: only the trend in the NudGlo ensemble lies within the error bar of the observations,alltheothersshowa largerwarmingover thisperiod.TheNudPacensembleisneverthelesscaptur- ing a weaker trend than the others, especially when compared withtheotherpartiallynudged NudAtlsimu- lations.Theseresultssuggestthat intheIPSL–CM5A–LR model, the equatorial Pacific partly paces the global temperature (notably through global teleconnections), while it is not the case for the Atlantic. Furthermore, NudPacsimulationsstilloverestimatetheobservedtrend by 0.18C/15 years. We find that the inclusion of the background volcanic forcing precisely induces such a reduction in the trend, as illustrated by the difference HisRbg minus HisRnobg. Thus, correct variations in the eastern tropical Pacific and volcanic forcing could be sufficient,underlinearassumption,tocorrectlyreproduce thehiatusinthismodel.
3.2. Spatialpatternoftemperaturetrendsfortheperiod 1998–2012.
Togoonestepfurther,wecomputethe15-yeartrendof thetargetperiodateachgridpointandcomparethespatial structure of the local trends of temperature in the observationandthedifferentsimulations(Fig.4).Wefirst checkthatthespatialaverageoftheselocaltrendsequals thetrendofthespatialaveragediscussedinFig.3b,since non-linearity could question the usefulness of this diagnostic.Thedifferenceamountstolessthan5%ofthe globaltrendfor allthesimulationsand HadCRUT4 data whenusingannualmeanvalue.Thus,wearguethatthe mapoflocaltrendsisarelevantdiagnostictoexplainthe trendoftheglobaltemperature.
The HadCRUT4 pattern (Fig. 4a) shows that the observed hiatus is primarilyrelated tothe coolingof a Fig. 3.(a)Timeevolution of themonthly meanglobaltemperature centeredovertheperiod1990–1994.InredistheHadCRUT4observation- baseddataset,inblackisHisRbgensemble,inbluetheNudAtl,ingreenthe NudPac, and in pink the NudGlo (cf. Table 1 for the name of the experiments).Theerrorbars arecomputedasthespreadofthefive membersofeachensemblearoundtheensemblemean.Allsimulation dataset has been masked to be comparable to observations (cf.
experimentaldesign).A24-monthrunningmeanhasbeenappliedto alltimeseriesforreadability.(b)Trendinglobaltemperatureoverthe period1998–2012intheHadCRUT4observationandinthedifferent simulations(Table1).Theerrorbariscomputedfromthefivemembers considered in each ensemble ofexperiments. To compare correctly HadCRUT4andthesimulations,weapplythesamespatialmaskinthe simulationsfromthatavailableforthedata,andweconsiderSSTwhen overtheoceanand2-metertemperaturewhenoverthelandassurface temperature,followingwhathasbeendoneinHadCRUT4dataset.
largeportionofthePacificOceanaswellasalargecooling overAsia.Incontrast,thetwodifferentsetsofhistorical simulations,whichshowverysimilarpatternsoftemper- aturetrends(Fig.4eandf),exhibitalargewarmingofthe NorthernHemisphereandtropicalarea.Theyshowasmall warming of the Southern Hemisphere, which is even experiencing a cooling in several areas of the mid- latitudes.Suchapatternisconsistentwithglobalwarming (Stocker et al., 2013) or response to solar variations (Swingedouwetal.,2011).Arecentstudyhighlightedthe role of the Deacon cell in this hemispheric adjustment
(Armouretal.,2016).Theseresultsshowthattheobserved cooling in the North Pacific is generally not driven by externalforcing,exceptperhapsfortheNortheastPacific region, which shows a slight cooling trend in these simulations.
InNudGlo,thetemperaturepatternisgenerallysimilar totheobservedone(Fig.4b),withacoolingintheWest Pacificbothintropicalandhighlatitudesareas,awarming inthewesthighlatitudesPacific,awarminginthetropical AtlanticandacoolinginthecentralNorthAtlantic.This indicates that NudGlo succeeds in capturing the main Fig.4.Spatialmapofsurfacetemperaturetrend(in0.18C/yr)overtheperiod1998–2012.Thesignificanttrendsfollowingatwo-sidedstudenttestare stippledinblack.(a)HadCRUT4data,(b)NudGloensemblemean,(c)NudPacensemblemean,(d)NudAtlensemblemean,(e)HisRnobgensemble,and(f) HisRbgensemblemean.
featureof SSTtrendsforthe1998–2012period. NudPac shows nolarge-scalesignificantwarmingintheeastern tropicalPacific,inagreementwithNudGloandHadCRUT4, and contrarily to NudAtl and historical ensembles. In NudAtl,thegeneraltemperaturetrendpatternobservedin the Atlantic, where nudging is applied, is reproduced, althoughthecoolingareainthenorthernmid-latitudesis not entirely captured. This may be due to the strong internal variability in this region of deep mixed layer, which can overwhelm the nudging constraint (Ortega et al.,2017).Thetropical Pacificisstronglywarmingas comparedtoNudPac.ThisresponseofthetropicalPacific mayexplainthelargewarmingattheglobalscaleobserved intheNudAtlensembleofsimulations(Fig.3b).
Thus,whilethetropicalPacificsucceedsinattenuating theglobalmeansurfacewarming(NudPac),theAtlanticis notcapableofdrivingtheobservedchangesinthePacific OceanandNudAtlisthereforenotreproducingthehiatus.
3.3. Dynamicalresponse
Thetrendsofsea-levelpressure(SLP)andsurfacewind velocity (Fig. 5) in the 20CR reanalysis show a clear increaseintheeasterliesovermostofthetropicalPacific Oceanfortheperiod1998–2012,associatedwithapositive SLPtrendinthesoutheasternPacific.IntheAtlanticand Indian sector, the SLP is anomalously negative. An anomalouseasterly trend is alsodetected in theIndian
Fig.5.Spatialmapofsea-levelpressuretrend(indPa/yr)overtheperiod1998–2012aswellastrendinwindvelocityasarrows.(a)NOAA-20CRreanalysis data,(b)NudGloensemblemean,(c)NudPacensemblemean,(d)NudAtlensemblemean,(e)HisRnobgensemble,and(f)HisRbgensemblemean.Thegreen boxin(a)correspondstotheaverageareainFig.6.
Ocean. We notice a pattern resembling a Rossby wave emanatingfromthetropicalPacifictowardsAntarcticaand uptothesouthernIndianOcean,withalternatepositive andnegativeSLPpoles.IntheNorthernHemisphere,sucha pattern inthe trendsis less clear,but we notea trend towardshigh pressure over the Aleutian low, and over Iceland, while trends over the Azores high is rather negative.Thisresemblesatrendtowardsanegativephase oftheArcticOscillation.
Noneofthetwohistoricalensembles,whichshowonce again very similar patternsat all latitudes, capture the amplitude of SLP and wind anomalies in the tropical Pacific.On theother hand,they reproduce thepositive SLPtrendoftheAleutian low,associated withthelocal coolingtrendseeninFig.4viaadvectionofcoldArcticair in the central Pacific. This confirms that the external forcing(mostprobablyanthropogenicaerosols,cf.Smith et al., 2016,Takahashi and Watanabe, 2016)may have playeda roleinthesesimulationsaswellasinthereal system.
The signs of the SLPtrends in the different tropical oceanicbasinsarecapturedinNudGlo.Theteleconnection withtheAleutianlowisrelativelywell-reproducedinthe Northern Hemisphere, in contrastto theteleconnection pattern in the Southern Hemisphere, where few direct observations are available. The NudPac ensemble can reproducethepositivetrendinthesubtropicalregionsof thePacificandthenegativeoneintheIndianOcean.The SLP trend over the tropical Atlantic basin is less well reproduced, as well as the wave patterns towards midlatitudesinbothhemispheres.Athighlatitudes,the response is characterized by a negative trend over the NorthPoleandapositivetrendovertheSouthPole,ina fashionsimilartowhatisfoundinbothhistoricalsets,but differingfromtheobservations.SSTnudgingintheNorth Atlantic(NudAtl)leadstoaSLPtrendofoppositesigninall basins,whiletheanomalyinthesouthernOceanisagainas inthehistoricalsimulations.
Wealsofindanincreaseinthetradewinds(easterlies) in the tropical West Pacific in most of the ensembles, exceptNudAtl. Focusingontheboxanalysed inEngland etal.(2014,cf.fig.5),Fig.6confirmsthisincreaseinzonal windspeed.Nevertheless,theamplitudeofthetrendover the1990sandthe2000sismorethanfourtimesweakerin thetwohistoricalensemblesascomparedtoobservations.
This is smaller than the ratio found by Takahashi and Watanabe(2016),whoweresuggestingthattheanthro- pogenicaerosolsaccountforaboutonethirdofthewind stresstrendoverthesametimeperiod,butofmagnitude comparable tothe resultsof England et al. (2014).The amplitudeofthetrendisalsounderestimatedinNudGlo andNudPacbyaboutafactoroftwo.Thismayindicatethat theanomalouswindchangesareonlypartiallyforcedby theSSTgradient andthatunforcedstochastic variations mayplayforabouthalf(accordingtooursimulations)of theobservedwindtrendinthetropicalarea.Alternatively, an underestimation of the SST-wind coupling in this model, or a too weak constrain applied on the SST as comparedtoKosakaandXie(2013),couldbeattheorigin ofthelackofzonalwindintensificationintheequatorial Pacificinournudgedsimulations.
4. Discussionsandconclusions
Whilethehiatusinglobalmeantemperaturehasbeen analysedfordifferenttimeperiodsinpreviousstudies,we findhere thatthe1998–2012 oneisthelowest15-year trendofthelastfewdecades,andthattrendscomputed over20yearsshowaweakerhiatus(notshown).Wehave therefore focused ouranalysis on this period. We have compared, using the IPSL–CM5A–LR model, historical simulations with different external forcing, accounting (HisRbg)or not (HisRnobg) for the effectof background volcanic eruptions from 2006,and globally(NudGlo)or partially (NudPacintheEast tropicalPacificand NudAtl overthewholeAtlantic)SST-nudgedsimulationstotrack theprocessespotentiallyexplainingthishiatusperiod.
Wefindthathistoricalsimulationswithoutbackground volcanoes overestimate the observed trend in global temperaturebyafactorof4,andonlyof3ifbackground volcanoes areincorporatedfrom2006onward.Nudging overthewholeoceanissufficienttoreproducethehiatus, whichisreasonablegiventhattheoceancovers70%ofthe Earth.SSTnudgingonlyimposedovertheeasternPacific (8% of the Earth) is increasing the agreement with observations(dividingthetrendofHisRbgbyafactorof two),sothat accountingboth for variability intheEast tropical Pacific and including background volcanoes of similar amplitude as that observed,can succeed, under linearassumption,inreproducingthehiatustrend.Onthe contrary,NudAtlissimilartoHisRnobg,sothatnudgingthe Atlanticbasin(15%oftheEarth)hasnoeffectontheglobal averagerepresentationofthehiatusintheIPSL–CM5A–LR model. In this respect, nudging over an almost twice- smallerregion(NudPacascomparedtoNudAtl)canhave verydifferenteffectsontheglobaltemperature,illustrat- ingthecomplexityofthedynamicsoftheEarthsystem.
It is striking in our simulations that the Pacific adjustment in NudAtlis opposedtotheoneimposed in NudGloandNudPac.TheAtlanticadjustmentinNudPacis alsoverydifferentfromtheoneimposedin NudAtl.One couldthussuspectabiasedlinkbetweentheAtlanticand Fig.6.WindvelocityinthewesterntropicalPacific(68S–68N,1508W–
1808W, cf. Fig. 5a) in the different simulations, following the box definitionfromEnglandetal.(2014).A5-yearrunningmeanhasbeen appliedonalltimeseries,andtheyhavebeencenteredoverthemeanof theyears1990–1994.
thePacific inthemodel. Thelink betweenAtlantic and Pacific main modes of variability (AMV–Atlantic Multi- decadalOscillation–andPDO)hasbeenanalysedbyMarini andFrankignoul(2014)intheobserveddataset.Theyfind asignificantlinkwhereapositiveAMVleadsbyonedecade a negative PDO, although this link is at the limit of statisticalsignificanceduetotheshortinstrumentaltime frame.IntheIPSL-CM5A-LRmodel,wefoundinthe1000- yearlongpre-industrial simulationa maximumcorrela- tion in phase, with a positive AMV associated with a positive PDO (not shown). This may confirm that the mechanismsrelating thetwobasinsarenot wellrepre- sentedinthisparticularmodel,undertheassumptionthat theobservedsignalissignificant.Inthiscontext,wecannot conclude on the possible teleconnections between the Atlantic and thePacific basinat play duringthe hiatus period.
ThelinkbetweentheAtlanticandthePacificcanalsobe modulatedbythestateoftheIndianOcean.Indeed,Terray etal.(2016)andKajtaretal.(2016)showedonashorter timeframethattheinfluenceoftheAtlanticOceanonthe ElNin˜ oSouthernOscillation(ENSO)dynamicsis largely dependentontheIndianOceanstateaswell.SincethePDO is related with the low-frequency variations of ENSO (Newmanetal.,2016),thestateoftheIndianOceanmay alsohaveplayedaroleduringthehiatusperiod(e.g.,Luo et al., 2012; Mochizuki et al., 2016). An interesting additionalpacemakerexperimentwouldthusbetodrive theIndianOceanaswell,orthetropicalIndian,Atlantic andPacificoceanstogether,toexploreiftheinfluenceon thetropicalPacificis clearerwhentropical Atlanticand Indianoceansarebothalsoconstrained.
Another possibility to explain the lack of hiatus in NudAtl is the adjustment of the IPSL–CM5A–LR climate model to zonal SST gradients between the different tropical basins. The dynamics induced by the Atlantic–
Pacificgradientoverthelastthreedecadeshasbeennicely illustrated in Li et al. (2016). They showed that in simulations of the CESM–CAM climate model, a large warmingintheAtlanticsectorfirstlyinducesachangein windvelocityoverthetropicalareathroughtheso-called Gill(1980)response(seefig.3fromLietal.,2016).Indeed, asshowninGill(1980),adiabaticheatinganomalylocated in theAtlanticisinducingtheemissionofKelvinwaves east oftheanomaly, whichweaken thewestwardwind velocity over the equatorial Indian Ocean and slightly enhancetheeasterliesoverthewesternPacific.Through changesinlatentheatfluxes,thistendstowarmtheIndian Ocean andslightlycoolthewesternPacific.Westofthe diabaticheating,twoRossbywavepacketsareemittedon eachsideoftheequator(cf.Gill,1980).Thesewavestendto increase the off-equatorial westerlies and decrease the equatorialeasterlies(cf.fig.1ofGill,1980,andfig.2aofLi etal.,2016).Attheequator,thedecreaseinupwellingand latentheatfluxshouldincreasethesurfacetemperature, whileofftheequatortheincreaseinlatentheatfluxshould leadtocooling.
Therefore, in a first phase, the response to diabatic heatingintheAtlanticiscomplexandleadstowarmingin theEastequatorialPacificandcoolingintheWestPacific, which wouldtendtoactivatea Nin˜ o-likeresponse (e.g.,
Lengaigneetal.,2004).Nevertheless,thewarmingofthe Indian and thecoolingoff equatorial EastPacific, when theypropagatetoWestPacific andEast Pacific,respec- tively,should,ontheopposite,favouraNin˜ a-likeresponse.
Thisis thesubtleinterplaybetweenthesetwoopposing signaturesthatwillactivateinasecondphasetheBjerknes feedbackintoonedirection(Nin˜ a-like)ortheother(Nin˜ o- like). While Li et al. (2016) argued that the interplay betweenthesetwoeffectswouldmostlyfavouraNin˜ a-like signature, it is possible that a slight difference in the impactofthediabaticheatingfromtheAtlanticcouldas wellactivateaNin˜ o-likeresponse.Thisseemstobewhat happened in our NudAtl ensemble. Consequently, the responsetodiabaticheatingintheAtlanticmaynotbeso straightforward,andslightdifferencesinthebackground states (for instance from the Indian Ocean) can lead a responseoppositetotheoneshowninLietal.(2016).
Thepoorperformanceintermsofhiatusreproduction ofNudAtlisthusduetoanoppositelinkbetweenAtlantic andPacificOceansascomparedtotheonesuggestedin other studies (Mcgregor et al., 2014; Takahashi and Watanabe,2016)orinobservations(Mariniand Franki- gnoul, 2014). Thisshows that this supposed link is not capturedinthismodel.Nevertheless,weinsistherethat theobservedlinkbetweendecadalvariationoftheAMV andPDO(MariniandFrankignoul,2014)wasfoundona veryshortrecordforobservations(aroundonecentury), whichquestionsthestatisticalsignificanceofthisresultas stated byMarini and Frankignoul (2014).Analysing the Atlantic–Pacificconnectionsonalongertimeframethan the last century will be necessary to better assess the existenceoftheAtlantic-Pacificlinkintherealsystem.A bettercharacterisationoftheprocessesatplayduringthe hiatus,andthepossiblelinkbetweenPacificandAtlantic basins,willstronglybenefitfromon-goingCMIP6inter- comparisonprojectdedicatedtopacemakerexperiments (Boeretal.,2016),whereeithertheAtlanticorthePacific willbe nudgedtowardsobservations (dcppC–pac–pace- makerand dcppC–atl–pacemaker–see Boeretal.,2016).
Furthermore,thepotentialroleplayedbytheIndianOcean forthehiatusperiodremainstobeevaluated.
Acknowledgments
ThisresearchwaspartlyfundedbytheANRMORDICUS project(ANR-13-SENV-0002-02).Itisalsofundedbythe SPECS project funded by the European Commission’s SeventhFrameworkResearchProgrammeunderthegrant agreementNo.308378.Thisworkwasgrantedaccessto theHPCresourcesofTGCCundertheallocationNo.2016- 017403.WethankPatrickBrockmannforhelpwithfigure design.ThispaperwassolicitedbytheChiefEditoraspart of a series of invited papers from laureates of the 2015prizesoftheFrenchAcademyofSciences.
References
Armour,K.C.,Marshall,J.,Scott,J.R.,Donohoe,A.,Newsom,E.R.,2016.
Southernoceanwarming delayedby circumpolar upwelling and equatorwardtransport.Nat.Geosci.9,549–554, http://dx.doi.org/
10.1038/ngeo2731.