anthropogenic influence on the 2018 summer warm spell in...
TRANSCRIPT
1
Anthropogenic influence on the 2018
summer warm spell in Europe: the impact
of different spatio-temporal scales
NicholasLeach1*,SihanLi2,3,SarahSparrow2,GeertJanvanOldenborgh4,FraserC.Lott5,
AntjeWeisheimer1,6,7,MylesR.Allen1,3,
1DepartmentofPhysics,AtmosphericOceanicandPlanetaryPhysics,UniversityofOxford,United
Kingdom.
2Oxforde-ResearchCentre,DepartmentofEngineeringScience,UniversityofOxford,United
Kingdom.
3EnvironmentalChangeInstitute,SchoolofGeographyandtheEnvironment,UniversityofOxford,
UnitedKingdom.
4KoninklijkNederlandsMeteorologischInstituut,DeBilt,TheNetherlands.
5UKMetOfficeHadleyCentre,Exeter,UnitedKingdom.
6DepartmentofPhysics,NationalCentreforAtmosphericScience(NCAS),UniversityofOxford,
UnitedKingdom.
7EuropeanCentreforMedium-RangeWeatherForecasts(ECMWF),Reading,UnitedKingdom.
*Correspondingauthoremail:[email protected]
2
Abstract
Wedemonstratethatinattributionstudies,eventsdefinedoverlongertimescalesgenerally
producehigherprobabilityratiosduetolowerinterannualvariability,reconcilingseemingly
inconsistentattributionresultsofEurope's2018summerheatwavesinreportedstudies.
Introduction
Thesummerof2018wasextremelywarminpartsofEurope,particularlyScandinavia,the
IberianPeninsula,andCentralEurope,witharangeofall-timetemperaturerecordsset
acrossthecontinent(Johnston,2018;NESDIS,2018).ImpactswerefeltacrossEurope,with
wildfiresburninginSweden(Krikken,Lehner,Haustein,Drobyshev,&vanOldenborgh,
2019;Watts,2018),heatstrokedeathsinSpain(Publico,2018),andwidespreaddrought
(Harris,2018).Duringthesummer,theWorldWeatherAttribution(WWA)initiative
releasedananalysisoftheheat-spell(WorldWeatherAttribution,2018)basedon
observations/forecastsandmodelsinspecificlocations(Dublin,Ireland;DeBilt,
Netherlands;Copenhagen,Denmark;Oslo,Norway;Linkoping,Sweden;Sodankyla,Finland;
Jokionen,Finland),whichconcludedthattheincreaseinlikelihoodduetohuman-induced
climatechangewasatleasttwotofivetimes.InDecember,theUKMetOffice(UKMO)
statedthattheyfoundthe2018UKsummertemperaturesweremadethirtytimesmore
likely(PressOffice,2018;McCarthy,etal.,2019).Thesetwoestimatesappearto
quantitativelydisagree,howeverweshowtheycanbereconciledbyinvestigatingthe
effectsofusingdifferentspatialdomainsandtemporalscalesintheeventdefinition.We
alsodemonstratethatprescribedSSTmodelsimulationscanunderrepresentthevariability
oftemperatureextremes,especiallynearthecoast,withimplicationsforanyderived
attributionresults.
3
Event Definition
Weconsidervarioustemperature-basedeventdefinitionstodemonstratetheimpactofthis
choiceinattributionassessments,andassesstowhatextenthumaninfluenceaffectedthe
seasonalandpeakmagnitudesofthe2018summerheat-eventonarangeofspatialscales.
Thestatisticweuseistheannualmaximumofthe1,10and90dayrunningmeanofdaily
mean2mtemperature,(hereafterTM1x,TM10xandTM90xrespectively).Weanalysethree
spatialscales:modelgridbox,regionalandEuropean.ForregionalandEuropeanevent
definitions,thespatialmeaniscalculatedbeforetherunningmean.Regionalextentsare
takenfromChristensenandChristensen(2007),andEuropeanextentistheE-OBS(Cornes,
vanderSchrier,vandenBesselaar,&Jones,2018)domain[landpointsin25N-71.5N,25W-
45E].TheWWAusedtheannualmaximaof3-daymeandailymaximumtemperaturesat
specificgridpointsforitsconnectiontolocalhealtheffects(D’Ippoliti,etal.,2010),whilethe
UKMOusedtheJJAmeantemperatureovertheentireUKinordertoanswerthequestion
ofhowanthropogenicforcingshaveaffectedthelikelihoodofUKsummersaswarmas
2018.Thesamejustificationscanbeusedhere,thoughweaddthatdifferentheat-event
timescalesareimportanttodifferentgroupsofpeople,andassuchusingseveraltemporal
definitionsmayincreaseinterestinheat-eventattributionstudies.However,werecognise
thatotherdefinitionsthanthoseusedheremaybemorerelevanttosomeimpacts
observed(suchasdefiningtheeventinthecontextoftheatmosphericflowpatternand
droughtwhichaccompaniedtheheat),andotherlinesofreasoningforselectingone
particulareventdefinitionexist(Cattiaux&Ribes,2018).
4
Model Simulations and Validation
ThreesetsofsimulationsfromtheUKMOHadleyCentreHadGEM3-Aglobalatmospheric
model(Christidis,etal.,2013;Ciavarella,etal.,2018)areused.Theseareahistorical
ensemble(1960-2013,Historical),andfactual(ACT)andcounterfactual(a``natural"world
withoutanthropogenicforcings,NAT)ensemblesof2018.Wecompareresultsfromthis
factual-counterfactualanalysiswiththosefromatrend-basedanalysisofHistorical,
ensemblesfromEURO-CORDEX(Vautard,etal.,2013;Jacob,etal.,2014;Vrac,Vaittinada
Ayar,&Ayar,2017)(1971-2018)andRACMO(Aalbers,Lenderink,vanMeijgaard,&vanden
Hurk,2018;Lenderink,etal.,2014)(1950-2018),andobservationsfromE-OBS(1950-2018).
AfullmodeldescriptionisprovidedintheSupplementaryInformation.Initially,we
performedouranalysiswiththeweather@homeHadRM3PEuropean-25kmsetup(Massey,
etal.,2015),butfoundthismodeloverestimatesthevariabilityoverallEuropefordaily
throughseasonalscaleeventstatisticsandsowasomitted.
Methodology
Wecalculatethereturnperiod,RP,forthe2018eventinadistributionfittoE-OBSusing
thegeneralisedextremevalue(GEV)distributiontomodelTM1xandTM10x,andthe
generalisedlogisticdistributiontoempiricallymodelTM90xthroughout.Sincethe
distributionoftemperatureextremeschangesastheclimatedoes,toaccountforthenon-
stationarityofthetimeserieswefirstremovethetrendattributabletolow-pass-filtered
globally-averagedmeansurfacetemperature[GMST,fromBerkeleyEarth(Rohde,etal.,
2013)]inanordinary-least-squaresregression[theregressioncoefficientortrendisshown
inSIFig.1(Diffenbaugh,etal.,2017)].Wethenfindthetemperaturethreshold
correspondingtoRPinadistributionfittothemodel’sclimatology.Inthefactual-
5
counterfactualanalysis,wedothisbyfittingparameterstoadetrended(againstGMST,
trendsshowninSIFig.2c7-c9)climatologicalensembleofHistoricalplus15randomly
sampledmembersofACT.Wefinallycalculatetheprobability(𝑃)ofexceedingthis
climatologicaltemperaturethresholdindistributionsfittotheACTandNATensemblesand
calculatetheprobabilityratio,PR = 𝑃!"#/𝑃!"#,representingtheincreasedlikelihoodof
the2018eventinthefactualcomparedtothecounterfactualworld.Usingestimatedevent
probabilitiesratherthanobservedmagnitudesconstitutesaquantilebiascorrection(Jeon,
Paciorek,&Wehner,2016),minimizingmodelbiasesinthemeanandvariabilityofthe
temperaturesanalysed.Adescriptionofuncertaintycalculationandthetrend-based
analysisdiscussedbelowisincludedintheSI.
Results
Extremedailyheat-events,measuredbyTM1x,aredistributedheterogeneouslythroughout
Europe(SIFig.1i).Thisisparalleledinthefactual-counterfactualPRsseeninFig.1a,with
largeproportionsofIberia,theNetherlands,andScandinaviaexperiencingeventsthatwere
highlyunlikelyinaclimatewithoutanthropogenicinfluence.Asimilarresultisfoundonthe
regionalscale(Fig.1d)withScandinaviaandtheIberianPeninsularespectivelyexperiencing
1-in-150[26,26000]1and1-in-30[9,550]yeareventsinthecurrentclimatethatwere
highlyunlikelyinthenaturalclimatesimulatedinNAT.Theremainingregionsrecord
maximumdailytemperatureslikelytoberepeatedwithin4years.Consideringthewholeof
Europe,thelikelihoodofthe2018maximumofdailyEuropeanmeantemperatureoccurring
withoutclimatechangeiszero.ThisresultisconsistentwithUhe,etal.,(2016)andAngélil,
1Numbersin[]representa90\%confidenceinterval
6
etal.,(2018),whoshowedthatincreasingspatialscaletendstoincreasetheprobability
ratio.
Extreme10-dayheat-events,TM10x,werealsowidespreadinEurope,withthemost
extremeoccurringinScandinavia(SIFig.1j).Regionally,thePRsbecomemoreuniform(Fig.
1d),thoughScandinaviaandtheIberianPeninsulastillhaveveryhighbest-estimatePRsof
185[17,infinite]and110[18,56000]respectively.Thebest-estimatePRfortheaverageof
Europeisstillformallyinfinite.
ThePRmapforseason-longheat-eventsmeasuredbyTM90xismoreuniformthroughout
Europe(Fig.1c).Scandinavia,BritishIsles,France,andCentralandEasternEuroperegions
allexperiencedontheorderof1-in-10yearevents(SIFig.1l),andthecorrespondingbest-
estimatePRsarebetween10and100forallregions(Fig.1d),includingthosewithlower
returnperiods.ThePRfortheEuropeanaverageis1000[500-2000].
Trend-basedanalysis,SIFigs.1m-p(observations)and2b(models),yieldssimilarresults,
thoughwenotethatforHadGEM-3AthisresultsingenerallyhigherPRs,duetothelinear
trendwithGMSTintheclimatologybeinggreaterthanthedifferencebetweenthetwo
ensemblesusedinthefactual-counterfactualanalysis.Observationalandmodelanalysis
contradictinsomegridboxesinNorthernScandinaviaforTM1xandTM10x,sincethe
observedbest-estimatetrendagainstGMSTisnegative,reducingtheeventprobabilityfor
thepresent-daycomparedtothepre-industrialclimate,thereforeyieldingPRsoflessthan
1.Comparingregionalfactual-counterfactualmodelwithobservationalanalysis(Fig.1dvsSI
Fig.1p)showsthatthelargeobservationaluncertaintiesoverlapwiththemodelresults:the
differencecouldbeduetonaturalvariabilityaffectingthesmallobservationalsamplesize.
7
However,wearecautiousofdrawinganyconclusionsregardingthechangeinlikelihoodof
extremeheat-eventsasdefinedherefortheselocations.
ThePRincreaseswiththeeventstatistictimescaleforthemajorityofgridpointsandregions
(showninFig.1).Fig.2illustratesthecauseusingtheBritishIslesregion:asthetimescale
increases,theeventstatisticdistributionvariancedecreases,whilethemeanshiftbetween
thefactualandcounterfactualdistributionsremainsconstant.SIFig1tshowsthatthe
similarityintrendswithGMSTbetweenthethreetimescalesisalsotrueforthe
observations.ThedecreaseinvarianceusuallyresultsinhigherPRs,givenaparticularevent
returntime,forthelongertimescales.Thereareexceptionsduetotheboundeduppertail
ofaGEVdistributionwithanegativeshapeparameter,resultingintheveryhighPRsfor
TM1xinScandinavia,IberiaandtheNetherlands.Thesolidanddottedblacklinescompare
thetemperaturethresholdswhenusingeventreturnperiodstoanomalymagnitudesinE-
OBS.ThisexplainswhytheTM90xPRismuchhigherthantheothertimescalesforthe
BritishIsles:inadditiontothedecreasedvariance,theseasonal-scaleheat-eventwasmore
unusualthantheothertimescales,withalongerreturnperiod(10.6[5.7,21]years)than
TM10x(2.6[1.8,3.9]years)andTM1x(3.6[2.5,6.2]years).Thesefactorstogetherresultin
PRsof3.6[2.9-4.8]forTM1xand43[27-84]forTM90x.Wesuggestthatthechangein
variancebetweenthetimescalesusedlargelyreconcilesthedifferencesbetweenthe``two
tofive"and``thirty"timesincreasesinlikelihoodfoundbytheWWAandUKMOreports,
withothermethodologicalfactorsplayingaminorrole.Althoughthehigherreturnperiod
forTM90xhassomeimpact,itislesssignificantthatthechangeinvariability.
Fig.2alsodemonstratesarelevantdeficiencyinthemodel:themodeldistributionsare
narrowerthantheobserveddistributions,meaningthemodelhaslowervariabilitythanthe
8
realworld.Thisreducedvariancehasasignificantimpactonattributionresults(Bellprat,
Guemas,Doblas-Reyes,&Donat,2019),andmeansthatthePRsfortheBritishIsles
presentedhere,especiallyforTM90x,arelikelytobeoverestimated.Underrepresented
variabilityoftenoccursinprescribedSSTmodels(Fischer,Beyerle,Schleussner,King,&
Knutti,2018),andisvisibleinHadGEM-3AformanycoastallocationsoverEurope(SIFig.
2a7-a9).Fig.2dshowsthepowerspectrumofJJAsummertemperaturesovertheBritish
Isles,indicatingthatHadGEM3-AhassimilarspectralcharacteristicstoE-OBS,but
underrepresentstheintraseasonal2mtemperaturevariabilityatalmostallfrequencies,
whichwilllikelyresultinoverestimatedPRs.Powerspectraforothermodelensemblesare
shownforcomparison,demonstratingthatthefullybias-correctedEURO-CORDEXensemble
hasthesamevariabilitycharacteristicsandmagnitudeastheobservations.
Discussion
Ouranalysishighlightsakeypropertyofextremeweatherattribution:thevarianceofthe
eventdefinitionused,bothintermsofthestatisticitselfanditsrepresentationwithinany
modelsused.Theuseoflongertemporaleventscalesingeneralincreasesboththespatial
uniformityandmagnitudeoftheprobabilityratiosfound,consistentwithKirchmeier-Young,
Wan,Zhang,&Seneviratne(2019),duetoadecreaseinvariancecomparedtoshorter
scales.Thedifferenceintemporalscalebetweentworeportsconcerningthe2018summer
heatissufficienttoexplainthelargediscrepancyinattributionresultbetweenthem.We
findthatseveralEuropeanregionsexperiencedseason-longheat-eventswithapresent-day
returnperiodgreaterthan10years.Thepresent-daylikelihoodofsucheventsoccurringis
approximately10to100timesgreaterthana``natural"climate.Theattributionresultsalso
showthattheextremedailytemperaturesexperiencedinpartsofScandinavia,the
9
NetherlandsandIberiawouldhavebeenhighlyunlikelywithoutanthropogenicwarming.
TheprescribedSSTmodelexperimentsusedheretendtounderestimatethevariabilityof
temperatureextremesnearthecoast,whichmayleadtotheattributionresultsoverstating
theincreaseinlikelihoodofsuchextremesduetoanthropogenicclimatechange(Bellprat,
Guemas,Doblas-Reyes,&Donat,2019).Weaimtoproperlyquantifytheimpactofthe
underrepresentedvariabilityinfurtherwork.Althoughherewehaveusedanunconditional
temperaturedefinitionforconsistencywiththestudieswetrytoreconcile,weplanto
furtherinvestigatetheeffectofincludingboththeatmosphericflowcontextandother
impactrelatedvariablessuchasprecipitationintheeventdefinition,andaddressissues
modelsmighthavewithrealisticallysimulatingthephysicaldriversofheatwaves.
Acknowledgements
WeacknowledgetheE-OBSdatasetfromtheEU-FP6projectUERRA(http://www.uerra.eu)
andtheCopernicusClimateChangeService,andthedataprovidersintheECA\&Dproject
(https://www.ecad.eu).Wewouldliketothankallofthevolunteerswhohavedonatedtheir
computingtimetoClimateprediction.netandWeather@Home.
References
Aalbers,E.E.,Lenderink,G.,vanMeijgaard,E.,&vandenHurk,B.J.(2018,jun).Local-scale
changesinmeanandheavyprecipitationinWesternEurope,climatechangeor
internalvariability?ClimateDynamics,50(11-12),pp.4745-4766.
Angélil,O.,Stone,D.,Perkins-Kirkpatrick,S.,Alexander,L.V.,Wehner,M.,Shiogama,H.,...
Christidis,N.(2018,apr).Onthenonlinearityofspatialscalesinextremeweather
attributionstatements.ClimateDynamics,50(7-8),pp.2739-2752.
10
Bellprat,O.,Guemas,V.,Doblas-Reyes,F.,&Donat,M.G.(2019,dec).Towardsreliable
extremeweatherandclimateeventattribution.NatureCommunications,10(1),p.
1732.
Cattiaux,J.,&Ribes,A.(2018,aug).DefiningSingleExtremeWeatherEventsinaClimate
Perspective.BulletinoftheAmericanMeteorologicalSociety,99(8),pp.1557-1568.
Christensen,J.H.,&Christensen,O.B.(2007,may).AsummaryofthePRUDENCEmodel
projectionsofchangesinEuropeanclimatebytheendofthiscentury.Climatic
Change,81(S1),pp.7-30.
Christidis,N.,Stott,P.A.,Scaife,A.A.,Arribas,A.,Jones,G.S.,Copsey,D.,...Tennant,W.J.
(2013,may).ANewHadGEM3-A-BasedSystemforAttributionofWeather-and
Climate-RelatedExtremeEvents.JournalofClimate,26(9),pp.2756-2783.
Ciavarella,A.,Christidis,N.,Andrews,M.,Groenendijk,M.,Rostron,J.,Elkington,M.,...
Stott,P.A.(2018,jun).UpgradeoftheHadGEM3-Abasedattributionsystemtohigh
resolutionandanewvalidationframeworkforprobabilisticeventattribution.
WeatherandClimateExtremes,20,pp.9-32.
Cornes,R.C.,vanderSchrier,G.,vandenBesselaar,E.J.,&Jones,P.D.(2018,sep).An
EnsembleVersionoftheE-OBSTemperatureandPrecipitationDataSets.Journalof
GeophysicalResearch:Atmospheres,123(17),pp.9391-9409.
D’Ippoliti,D.,Michelozzi,P.,Marino,C.,De’Donato,F.,Menne,B.,Katsouyanni,K.,...
Perucci,C.A.(2010).Theimpactofheatwavesonmortalityin9Europeancities:
ResultsfromtheEuroHEATproject.EnvironmentalHealth:AGlobalAccessScience
Source,9(1).
11
Diffenbaugh,N.S.,Singh,D.,Mankin,J.S.,Horton,D.E.,Swain,D.L.,Touma,D.,...
Rajaratnam,B.(2017,may).Quantifyingtheinfluenceofglobalwarmingon
unprecedentedextremeclimateevents.ProceedingsoftheNationalAcademyof
SciencesoftheUnitedStatesofAmerica,114(19),pp.4881-4886.
Fischer,E.M.,Beyerle,U.,Schleussner,C.F.,King,A.D.,&Knutti,R.(2018,aug).Biased
EstimatesofChangesinClimateExtremesFromPrescribedSSTSimulations.
GeophysicalResearchLetters,45(16),pp.8500-8509.
Harris,C.(2018,aug).Heat,hardshipandhorribleharvests:Europe'sdroughtexplained.
Euronews.Retrievedfromhttps://www.euronews.com/2018/08/10/explained-
europe-s-devastating-drought-and-the-countries-worst-hit
Jacob,D.,Petersen,J.,Eggert,B.,Alias,A.,Christensen,O.B.,Bouwer,L.M.,...Yiou,P.
(2014,apr).EURO-CORDEX:newhigh-resolutionclimatechangeprojectionsfor
Europeanimpactresearch.RegionalEnvironmentalChange,14(2),pp.563-578.
Jeon,S.,Paciorek,C.J.,&Wehner,M.F.(2016,jun).Quantile-basedbiascorrectionand
uncertaintyquantificationofextremeeventattributionstatements.Weatherand
ClimateExtremes,12,pp.24-32.
Johnston,C.(2018,aug).Heatwavetemperaturesmaytop45CinsouthernEurope.The
Guardian.Retrievedfrom
https://www.theguardian.com/world/2018/aug/04/temperatures-in-southern-
europe-top-45-heatwave-spain-portugal
12
Kirchmeier-Young,M.C.,Wan,H.,Zhang,X.,&Seneviratne,S.I.(2019,sep).Importanceof
framingforextremeeventattribution:theroleofspatialandtemporalscales.
Earth’sFuture.
Krikken,F.,Lehner,F.,Haustein,K.,Drobyshev,I.,&vanOldenborgh,G.J.(2019,aug).
AttributionoftheroleofclimatechangeintheforestfiresinSweden2018.Natural
HazardsandEarthSystemSciencesDiscussions,pp.1-24.
Lenderink,G.,vandenHurk,B.J.,KleinTank,A.M.,vanOldenborgh,G.J.,vanMeijgaard,E.,
deVries,H.,&Beersma,J.J.(2014,nov).Preparinglocalclimatechangescenariosfor
theNetherlandsusingresamplingofclimatemodeloutput.EnvironmentalResearch
Letters,9(11),p.115008.
Massey,N.,Jones,R.,Otto,F.E.,Aina,T.,Wilson,S.,Murphy,J.M.,...Allen,M.R.(2015,
jul).weather@home-developmentandvalidationofaverylargeensemblemodelling
systemforprobabilisticeventattribution.QuarterlyJournaloftheRoyal
MeteorologicalSociety,141(690),pp.1528-1545.
McCarthy,M.,Christidis,N.,Dunstone,N.,Fereday,D.,Kay,G.,Klein-Tank,A.,...Stott,P.
(2019,November07).DriversoftheUKsummerheatwaveof2018.Weather,74,
390-396.
NESDIS.(2018,aug).RecordSummerHeatBakesEurope.NESDIS.Retrievedfrom
https://www.nesdis.noaa.gov/content/record-summer-heat-bakes-europe
PressOffice.(2018,dec).Chanceofsummerheatwavesnowthirtytimesmorelikely.Met
OfficeNews.Retrievedfromhttps://www.metoffice.gov.uk/about-us/press-
office/news/weather-and-climate/2018/2018-uk-summer-heatwave
13
Publico.(2018,aug).NeuvefallecidosporlaoladecalorenEspana.Publico.Retrievedfrom
https://www.publico.es/sociedad/nueve-fallecidos-ola-calor-espana.html
Rohde,R.,Muller,R.A.,Jacobsen,R.,Muller,E.,Perlmutter,S.,Rosenfeld,A.,...Wickham,
C.(2013).ANewEstimateoftheAverageEarthSurfaceLandTemperatureSpanning
1753to2011.Geoinformatics&Geostatistics:AnOverview,1,p.1.
Uhe,P.,Otto,F.E.,Haustein,K.,vanOldenborgh,G.J.,King,A.D.,Wallom,D.C.,...Cullen,
H.(2016,aug).Comparisonofmethods:Attributingthe2014recordEuropean
temperaturestohumaninfluences.GeophysicalResearchLetters,43(16),pp.8685-
8693.
Vautard,R.,Gobiet,A.,Jacob,D.,Belda,M.,Colette,A.,Déqué,M.,...Yiou,P.(2013,nov).
ThesimulationofEuropeanheatwavesfromanensembleofregionalclimate
modelswithintheEURO-CORDEXproject.ClimateDynamics,41(9-10),pp.2555-
2575.
Vrac,M.,VaittinadaAyar,P.,&Ayar,P.V.(2017,jan).InfluenceofBiasCorrectingPredictors
onStatisticalDownscalingModels.JournalofAppliedMeteorologyandClimatology,
56(1),pp.5-26.
Watts,J.(2018,jul).WildfiresrageinArcticCircleasSwedencallsforhelp.TheGuardian.
Retrievedfromhttps://www.theguardian.com/world/2018/jul/18/sweden-calls-for-
help-as-arctic-circle-hit-by-wildfires
WorldWeatherAttribution.(2018,jul).HeatwaveinnorthernEurope,summer2018.
worldweatherattribution.org.Retrievedfrom
14
https://www.worldweatherattribution.org/attribution-of-the-2018-heat-in-
northern-europe/
Figure Captions
Figure1:Probabilityratiosforthe2018summerheat-eventderivedfromHadGEM3-A
factual-counterfactualsimulations.Panelsa-cshowmapofincreasedlikelihoodinthereal
worldatgridboxscaleforthethreeeventtimescalesanalysedrespectively.Notethatthe
upperlimitonthecolorscaleis1000andgridboxeswithaninfiniteprobabilityratioare
shownindarkred.Paneldshowstheregionalprobabilityratiosforthethreetimescales;
thecrossesdenotethebest-estimate,andthebarsdenotethe5th-95thpercentiles.Note
thatthebest-estimateprobabilityratiosforTM1xinScandinaviaandtheIberianPeninsula
wereinfiniteand2000respectively.
Figure2:Panelsa-c,probabilitydensityfunctionsofthethreetemporalscalesofevent
statisticfortheBritishIsles,showingHadGEM3-AACT(factual)andNAT(counterfactual)
simulations,andobservationsfromE-OBS;allasanomaliesabovethemodelorobserved
1961:1990meanclimatology.Thickblacklinesshowthe2018eventdefinedprobabilistically
astheHadGEM3-AHistoricaltemperaturethresholdcorrespondingtotheE-OBSreturn
period,dottedblacklinesshowtheeventdefinedintermsofthemagnitudeobserved
directlyfromE-OBS.Paneld,periodogramsofJJAdailymeantemperatureintheBritishIsles
(seasonalityandmeanremoved)calculatedasthemeanofintraseasonalperiodogramsfor
allavailableyears.TheHadGEM3-ApowerspectrumiscalculatedfromtheHistorical
ensemble.
15
Figures
Figure1:Probabilityratiosforthe2018summerheat-eventderivedfromHadGEM3-A
factual-counterfactualsimulations.Panelsa-cshowmapofincreasedlikelihoodinthereal
worldatgridboxscaleforthethreeeventtimescalesanalysedrespectively.Notethatthe
upperlimitonthecolorscaleis1000andgridboxeswithaninfiniteprobabilityratioare
shownindarkred.Paneldshowstheregionalprobabilityratiosforthethreetimescales;
thecrossesdenotethebest-estimate,andthebarsdenotethe5th-95thpercentiles.Note
thatthebest-estimateprobabilityratiosforTM1xinScandinaviaandtheIberianPeninsula
wereinfiniteand2000respectively.
16
Figure2:Panelsa-c,probabilitydensityfunctionsofthethreetemporalscalesofevent
statisticfortheBritishIsles,showingHadGEM3-AACT(factual)andNAT(counterfactual)
simulations,andobservationsfromE-OBS;allasanomaliesabovethemodelorobserved
1961:1990meanclimatology.Thickblacklinesshowthe2018eventdefinedprobabilistically
astheHadGEM3-AHistoricaltemperaturethresholdcorrespondingtotheE-OBSreturn
period,dottedblacklinesshowtheeventdefinedintermsofthemagnitudeobserved
directlyfromE-OBS.Paneld,periodogramsofJJAdailymeantemperatureintheBritishIsles
(seasonalityandmeanremoved)calculatedasthemeanofintraseasonalperiodogramsfor
allavailableyears.TheHadGEM3-ApowerspectrumiscalculatedfromtheHistorical
ensemble.