obrved

Globalobrvedchangesindailyclimateextremesoftemperature

andprecipitation

der,1,2,,on,,n,5

ank,k,s,,adeh,9

ur,mar,kar,ths,t,4

nson,,r,,,,15

,ucci,z-Aguirre18

Received31May2005;revid28September2005;accepted8November2005;published15March2006.

[1]Asuiteofclimatechangeindicesderivedfromdailytemperatureandprecipitation

data,withaprimaryfocusonextremeevents,ingan

exactformulaforeachindexandusingspeciallydesignedsoftware,analysdonein

nabledtheprentationof

themostup-to-dateandcomprehensiveglobalpictureoftrendsinextremetemperature

andprecipitationindicesusingresultsfromanumberofworkshopsheldindata-spar

regionsandhigh-qualitystationdatasuppliedbynumerousscientistsworldwide.

Seasonalandannualindicesfortheperiod1951–inthe

sshowed

widespreadsignificantchangesintemperatureextremesassociatedwithwarming,

70%ofthe

globallandareasampledshowedasignificantdecreaintheannualoccurrenceofcold

gions

pliesapositiveshiftinthe

aximum

itation

changesshowedawidespreadandsignificantincrea,butthechangesaremuchless

ilitydistributionsofindices

derivedfromapproximately200temperatureand600precipitationstations,withnear-

completedatafor1901–2003andcoveringaverylargeregionoftheNorthern

Hemispheremidlatitudes(andpartsofAustraliaforprecipitation)wereanalyzedforthe

periods1901–1950,1951–1978and1979–sindicateasignificantwarming

encesintemperatureindicesdistributionsare

particularlypronouncedbetweenthemostrecenttwoperiodsandforthoindicesrelated

ysisofthoindicesforwhichasonaltimeriesare

availableshowsthatthechangesoccurforallasonsalthoughtheyaregenerallyleast

itationindicesshowatendencytoward

wetterconditionsthroughoutthe20thcentury.

JOURNALOFGEOPHYSICALRESEARCH,VOL.111,D05109,doi:10.1029/2005JD006290,2006

1HadleyCentre,MetOffice,Exeter,UK.

2AlsoatBureauofMeteorologyRearchCentre,Melbourne,Victoria,

Australia.

3NowatSchoolofGeographyandEnvironmentalScience,Monash

University,Clayton,Victoria,Australia.

4ClimateRearchBranch,MeteorologicalServiceofCanada,Downs-

view,Ontario,Canada.

5NationalClimaticDataCenter/NOAA,Asheville,NorthCarolina,

USA.

6RoyalNetherlandsMeteorologicalInstitute,DeBilt,Netherlands.

7ClimaticRearchUnit,UniversityofEastAnglia,Norwich,UK.

8BureauofMeteorology,Melbourne,Victoria,Australia.

9AtmosphericScienceandMeteorologicalRearchCenter,Iran

MeteorologicalOrganization,Tehran,Iran.

10IndianInstituteofTropicalMeteorology,Pune,India.

Publishedin2006bytheAmericanGeophysicalUnion.

D05109

11NationalInstituteofWaterandAtmosphericRearch,Auckland,

NewZealand.

12DepartmentofMeteorology,UniversityofReading,Reading,UK.

13ClimateChangeRearchGroup,UniversitatRoviraiVirgili,

Tarragona,Spain.

14PhysicsDepartment,UniversityoftheWestIndies,Kingston,

Jamaica.

15ClimateRearchGroup,OxfordUniversityCentrefortheEnviron-

ment,UniversityofOxford,Oxford,UK.

16ChinaMeteorologicalAdministration,Beijing,China.

17DepartamentodeCienciasdelaAtmo´sferaylosOce´anos,Facultadde

CienciasExactasyNaturales,UniversidaddeBuenosAires,BuenosAires,

Argentina.

18DepartamentodeMeteorologiaGeneral,CentrodeCienciasdela

Atmosfera,UniversidadNacionalAuto´nomadeMe´xico,Coyoaca´n,

Mexico.

1of22

Citation:Alexander,L.V.,etal.(2006),Globalobrvedchangesindailyclimateextremesoftemperatureandprecipitation,

.,111,D05109,doi:10.1029/2005JD006290.

uction

[2]Fordecades,mostanalysoflong-termglobalcli-

matechangeusingobrvationaltemperatureandprecipi-

l

wellrespectedmonthlydatatsprovidereasonablespatial

coverageacrosstheglobe[e.g.,JonesandMoberg,2003;

PetersonandVo,1997].However,analyzingchangesin

extremes,suchaschangesinheatwavedurationorinthe

numberofdaysduringwhichtemperatureexceedsitslong-

term90thpercentile,requiresdailydataindigitalform.

Unfortunately,thedataarenotreadilyavailabletothe

internationalrearchcommunityforlargeportionsofthe

world[Follandetal.,2001].Inearlier‘‘global’’analysisof

extremeindicesbyGroismanetal.[1999]andFrichetal.

[2002],therewerealmostnodataformostofCentraland

SouthAmerica,Africa,uent

studiessuchasKiktevetal.[2003]providedgridded

updatestosomeoftheindicesbutthespatialcoverage

wasstillpoor.

[3]ThejointWorldMeteorologicalOrganizationCom-

missionforClimatology(CCl)/WorldClimateRearch

Programme(WCRP)projectonClimateVariabilityand

Predictability(CLIVAR)ExpertTeamonClimateChange

Detection,MonitoringandIndices(ETCCDMI)coordinated

twocomplimentaryeffortstoenableglobalanalysisof

ortwastheinternationalcoordinationof

thedevelopmentofasuiteofclimatechangeindiceswhich

ndicesarederivedfrom

elopment

oftheindices,includingaur-friendlysoftwarepackage

thatisfreelyavailabletotheinternationalrearchcommu-

nity,involvednotonlyETCCDMImembersbutalso

numerousotherscientists,includingmanyoftheauthors.

Inall,27indicesweredefinedandtwosoftwarepackages,

onewritteninR(RClimDex)andtheotherwrittenin

FORTRAN(FClimDex),te,

ccma//ETCCDMI,dedicatedtothiffort

providescomprehensivedescriptionsofalloftheindices,

detailsofqualitycontrolproceduresandreferencesto

providesafreedownloadofthe

ttinganexactformulaforeachindexandbyusingthe

samesoftwarepackage,analysdoneindifferentcountries

ordifferentregionscanfittogetheramlessly.

[4]ThecondETCCDMIeffortwastocoordinate

regionalworkshopswiththeaimofaddressinggapsindata

availabilityandanalysisinpreviousglobalstudies[e.g.,

Frichetal.,2002].Inmanypartsoftheworldthereare

enoughdailydataavailableindigitalformatthenational

level,althoughaccessingdigitaldailydatacanstillbe

problematicinsomeregions[Pageetal.,2004].Also,

someinstitutionsarereluctanttopartwithdataforvarious

iontothisproblempropodbythe

ETCCDMIwastoholdregionalclimatechangeworkshops

modeledontheAsiaPacificNetwork(APN)workshops

[Mantonetal.,2001;Petersonetal.,2001;Griffithtal.,

2005].TheAPNapproachwastobringtogetherscientists

fromdifferentcountrieswithintheAsia-Pacificregion.

Theparticipantsbroughttheirowndailydatatothe

uidancefrominternationalexperts,

duringtheworkshop,theyconducteddataqualitycontrol

andcomputedindicesusingastandardprocedureand

approachmadeitpossibletoexchange

ghsomeparticipantschonottoshare

theiroriginaldailydatatheymadethederivedindicesries

ional

climatechangeworkshopswereheldin2001inJamaica,to

covertheCaribbeanregion[Petersonetal.,2002]andin

Morocco,tocoverAfrica[Easterlingetal.,2003;Mokssit,

2003].Recognizingthesuccessandproblemsofthe

workshops,theETCCDMIheldfiveadditionalworkshops

in2004andearly2005inSouthAfrica(l.,

Evidenceoftrendsindailyclimateextremesoversouthern

andwestAfrica,submittedtoJournalofGeophysical

Rearch,2005),Brazil[Haylocketal.,2006;Vincentet

al.,2005],Turkey[Sensoyetal.,2006;Zhangetal.,2005a],

Guatemala[Aguilaretal.,2005]andIndia[Peterson,2005;

KleinTanketal.,2006]toprovideadditionalcoveragefor

Africa,SouthAmerica,theMiddleEast,CentralAmerica

andsouth-centralAsia.

[5]Theobjectiveofthispaperistoprovidethemost

comprehensiveanalysisofobrvedglobaltemperatureand

thind,weuhigh-

clude

(1)datathatarefreelyavailabletotheinternationalcom-

munity,(2)datafromalltheETCCDMIworkshopsthat

werenotavailablepreviouslyand(3)datathatonlysomeof

erisorganizedas

ction

includesdetaileddescriptionsofthesourcesofdailydata,

dataqualitycontrolandhomogeneitytestingprocedures,

ion3

weprovideadetailedaccountoftheanalysisoftheindices

data,sare

rsomediscussionofthe

resultsinction5followedbyconclusionsinction6.

ata

[6]Therearethreeinternationaldailydatatsfreely

e(1)theGCOS

SurfaceNetwork(GSN)datat[Petersonetal.,1997],

(2)theEuropeanClimateAsssment(ECA)datat[Klein

Tanketal.,2002]and(3)thedailyGlobalHistoricalClima-

tologyNetwork(GHCN-Daily)datat[Gleasonetal.,

2002].TheECAdataareudtocoverEuropeinthis

analysis,whileGHCN-DailydataareudfortheUnited

dataareudtosupplementthe

sourcesofdata,sdatafromthe

workshopsareudtocovertherespectiveregionswheredata

softheworkshop

dataaredescribedinrelevantworkshopreportsorpapers.

DatafromtheAPNworkshopsarealsoincluded.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

2of22

D05109

[7]Datawerealsoprovidedbytheauthors’institutions

forsomepartsoftheworldwherethedatafortheregions

fromtheabovesourceswerenotavailableorwereofpoorer

thelevelofdevelopmentofhigh-quality

dailystationdatatsdiffersfromonecountrytoanother,

supplied

carefullyhomogenizeddailytemperaturesupto2003for

210stations[Vincentetal.,2002]andahigh-quality

precipitationdatat[MekisandHogg,1999].Australian

temperaturerecordshavebeenadjustedforinhomogeneities

atthedailytimescalebytakingaccountofthemagnitudeof

discontinuitiesfordifferentpartsofthedistribution[Trewin,

2001].Stationsthatwerelikelytobeaffectedbyurbaniza-

tionwereexcluded,althoughrecentstudies[e.g.,Peterson,

2003;Parker,2004;PetersonandOwen,2005]showthat

urbanizationeffectshavehadlittleeffectonlarge-scale

lianprecipitationdataalsocame

fromahigh-qualityprecipitationdatat[Haylockand

Nicholls,2000].TemperaturedatafortheUnitedStates

werechonfromGHCN-Dailystationswherestatistical

homogeneitytestsonbothmaximumandminimumtemper-

aturesdidnotdetectanyinhomogeneities[Menneand

Williams,2005].PrecipitationdatafortheformerUSSR

werehomogeneityadjusted[GroismanandRankova,2001].

Forsomecountriesforwhichprescreenednationaldatats

werenotreadilyavailable,e.g.,Argentina[Rusticucciand

Barrucand,2004],China[Zhaietal.,2005],India,Iran

[RahimzadehandAsgari,2003]andMexico,theauthors

chostationsonthebasisoftheirknowledgeofthebest

stationsintheirowncountryand/

remainingdataweresuppliedprimarilyfromtheGHCN-

Dailydatat,e.g.,Brazil,andtheHadleyCentrearchives.

Inallcasatleastoneoftheauthorshadaccesstotheraw

stationrecordssothatreferencecouldalwaysbemadeto

theoriginaldatashouldqualityissuesariduringthe

analysis.

alityandHomogeneity

[8]Inmostcas,datasuppliedbytheauthorswere

qualitycontrolledandtheindicescalculatedusingstandard

elof

qualitycontroldifferedfromcountrytocountry(eabove)

butinallcasanattemptwasmadetouthebestpossible

asuppliedfromtheworkshopsthe

n

purpoofthisqualitycontrolprocedurewastoidentify

errorsindataprocessing,sucharrorsinmanualkeying.

Negativedailyprecipitationamountsareremovedandboth

dailymaximumandminimumtemperaturesarettoa

missingvalueifdailymaximumtemperatureislessthan

rsindailymaximumand

revalues

study,therange

isdefinedaslyingwithinfourstandarddeviations(std)of

theclimatologicalmeanofthevaluefortheday,thatis,

[mean±4Âstd].Dailytemperaturevaluesoutsidethis

rangearemanuallycheckedandeditedonacabyca

basisbyworkshopparticipantswhoareknowledgeable

abouttheirowndailydata.

[9]Statisticaltestswerenotgenerallyappliedtoprecip-

itationdataanalyzedattheworkshopsbutanyobvious

outliers,identifiedbycarefulexaminationofgraphs,were

lpostworkshopanalysiswas

employedanddataprocesdoutsideoftheworkshops

weresimilarlytestedforoutliersbutmethodsvariedfrom

ticaltests,localknowledge,an

investigationofstationhistoriesorcomparisonwithneigh-

boringstationscanallbeappliedtodeterminewhetheran

rticularly

importanttoidentifymultidayprecipitationaccumulations

thatcanappearerroneouslyinrecordsofdailyprecipitation

[VineyandBates,2004].Theoccurwhenaccumulated

mple,data

extractedfromGHCN-DailyforBrazilwererejectedifa

rainfallvaluegreaterthan1mmfellafteramissing

obrvation[Haylocketal.,2006].Evenafterdatawere

procesdandcollatedforthisstudy,annualtimeriesof

totalprecipitationanddiurnaltemperaturerangeforeach

stationwereassdagaintoidentifyoutliersthatmay

havebeenmisdintheinitialqualitycontrolprocedure.

[10]Dataqualityisarelativelyeasyproblemtoaddress

whencomparedwiththeproblemsassociatedwithdata

ousoutliersandartificialstep

changescaudbychangesinstationlocation,obrving

proceduresandpractice,instrumentationchangetc.

[Aguilaretal.,2003]maketrendanalysisunreliable,

andthereisnotalwaysaconsistentapproachtodeal

withdatainhomogeneity[Petersonetal.,1998].Forthis

reason,RClimDexcanbeudintandemwithasoftware

packagecalledRHtestwhichidentifiesstepchangesin

isbadonatwo-

pharegressionmodelwithalineartrendfortheentire

ries[Wang,2003].Exceptforthedatafromthefirst

fewworkshops,whereaslightlydifferentprogrambad

onasimilartechniquewasud,RHtesthasbeenudto

testforinhomogeneitiesinthetemperaturedatafrom

xceptionsinclude

theuofthemosthomogeneousstationsasdefinedby

Wijngaardetal.[2003]forECAtemperatureandprecip-

itationdataandtheuofhomogeneousstationsidenti-

fiedbyMenneandWilliams[2005]fortheUnitedStates.

[11]Ifstationdatawereidentifiedasbeinginhomoge-

-

geneousdatawerenotadjustedfortwomainreasons

(althoughnotethatsomedatasourceswerealreadyadjust-

ed,e.g.,GroismanandRankova[2001]andVincentetal.

[2002]priortoinclusioninthisstudy).First,therehasbeen

onlylimitedsuccesstodateinadjustingdailytemperatures

[e.g.,Vincentetal.,2002].Second,asthemanystationswe

havecovermanydifferentclimates,adjustmentoftemper-

atureswouldbeanextremelycomplextaskanddifficultto

dowell[Aguilaretal.,2003].Itispossiblethatsomestep

changescouldberealandnotduetoaninhomogeneity

ghlightstheimportanceof

havingaccesstostationmetadatawhichwegenerally

lacked.

[12]Inhomogeneitiesotherthanstepchanges,suchas

gradualchangesintemperature,

aninhomogeneitymightoccurthroughurbanization

althoughPetersonandOwen[2005]suggestthattheeffects

ofurbanizationondataaveragedoveranentirenetworkof

stationsareminimalwhencomparedtothemagnitudeof

blemofsuch

inhomogeneitiesisalsodifficulttoaddressthoughitcould

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

3of22

D05109

potentiallybedealtwithbycomparingdatafromneighbor-

r,ourstationnetworkisnotusually

denenoughforthisapproachtobeadopted.

[13]Figure1showsthelocationsofthe2223temperature

gh

therearemoreprecipitationstations,theyaregenerally

distributedlessuniformlythanthetemperaturestations.

r,

whencalculatingtrendswechoonlytoconsidergrid

boxesforwhichthedataforthetimeperiodunderconsid-

erationwereatleast80%completeandendednoearlier

ationnetworkscontributingdatatothis

studyhavegoodtemporalcoverageforthecondhalfof

the20thcenturysowemostlyfocusonthisperiod.

However,asubtofaround200temperature(depending

ontheindex)and608precipitationstationshadenoughdata

forchangesthroughouttheentire20thcenturytobe

luthe

stationstoputrecentchangesinthecontextofacentury

tionsareprimarilylocatedinNorth

America,EurasiaandAustralia,althoughafewstationsare

locatedinBrazilandSriLanka.

s

[14]Sixteenofthe27indicesrecommendedbythe

ETCCDMIaretemperaturerelatedandelevenareprecipi-

ederivedfromdailymaximumand

e-

scriptivelistoftheindicescanbeobtainedfrom

/ETCCDMI/list_27_

theindiceshasaur-dependentthresholdwhichwehave

iceswerechon

primarilyforasssmentofthemanyaspectsofachanging

globalclimatewhichincludechangesinintensity,frequency

onsof(a)temperatureand(b)ors

rtheUnitedStatesandparts

bersinbracketsindicate

thetotalnumberofstations.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

4of22

D05109

reprenteventsthatoccurveraltimesperasonoryear

givingthemmorerobuststatisticalpropertiesthanmeasures

ofextremeswhicharefarenoughintothetailsofthe

distributionsoasnottobeobrvedduringsomeyears.

Theindicescanbedividedinto5differentcategories:

[15]tile-badindicesincludeoccurrenceof

coldnights(TN10p),occurrenceofwarmnights(TN90p),

occurrenceofcolddays(TX10p),occurrenceofwarmdays

(TX90p),verywetdays(R95p)andextremelywetdays

(R99p).Thetemperaturepercentile-badindicessample

thecoldestandwarmestdecilesforbothmaximumand

minimumtemperatures,enablingustoevaluatetheextentto

cipitationindicesin

thiscategoryreprenttheamountofrainfallfallingabove

the95th(R95p)and99th(R99p)percentilesandinclude,

butarenotbelimitedto,themostextremeprecipitation

eventsinayear.

[16]teindicesreprentmaximumorminimum

cludemaximum

dailymaximumtemperature(TXx),maximumdailymini-

mumtemperature(TNx),minimumdailymaximumtem-

perature(TXn),minimumdailyminimumtemperature

(TNn),maximum1-dayprecipitationamount(RX1day)

andmaximum5-dayprecipitationamount(RX5day).

[17]oldindicesaredefinedasthenumberof

daysonwhichatemperatureorprecipitationvaluefalls

aboveorbelowafixedthreshold,includingannualoccur-

renceoffrostdays(FD),annualoccurrenceoficedays(ID),

annualoccurrenceofsummerdays(SU),annualoccurrence

oftropicalnights(TR),numberofheavyprecipitation

days>10mm(R10)andnumberofveryheavyprecipita-

tiondays>20mm(R20).Theindicesarenotnecessarily

meaningfulforallclimatesbecauthefixedthresholds

udinthedefinitionsmaynotbeapplicableeverywhereon

r,previousstudies[e.g.,Frichetal.,

2002;Kiktevetal.,2003]haveshownthattemperature

indicessuchasFD,thenumberofdaysonwhichminimum

temperaturefallsbelow0°C,haveexhibitedcoherenttrends

overthemidlatitudesduringthecondhalfofthe20th

tion,changesintheindicescanhave

profoundimpactsonparticularctorsofsocietyoreco-

oreweincludedtheindicesinourstudy,

eventhoughsomeofthemmaynotprovidetruly‘‘global’’

spatialcoverageorbetrulyextreme.

[18]onindicesdefineperiodsofexcessive

warmth,cold,wetnessordrynessorinthecaofgrowing

asonlength,cludecoldspell

durationindicator(CSDI),warmspelldurationindicator

(WSDI),growingasonlength(GSL),concutivedry

days(CDD)andconcutivewetdays(CWD).Manyof

theindiceswereudinthenearglobalanalysisofFrich

etal.[2002].Theheatwavedurationindex(HWDI)defined

byFrichetal.[2002]hasbeenfoundnottobestatistically

robustasithadatendencytohavetoomanyzerovalues

[Kiktevetal.,2003].ThisisbecauFrichetal.[2002]

udafixedthresholdof5°Caboveclimatologytocompute

resholdistoohighinmanyregions,such

asthetropics,wherethevariabilityofdailytemperatureis

comethis,theETCCDMIreplacedthisindex

withthewarmspelldurationindex(WSDI)whichis

index

onlysampleddaytimemaximawealsochotoanalyze

spellsofnighttimeminima(CSDI).TheCDDindexisthe

lengthofthelongestdryspellinayearwhiletheCWD

categoryofindicesalsoincludesthelengthofthegrowing

ason(GSL)whichisanindexthatisgenerallyonly

meaningfulintheNorthernHemisphereextratropics.

[19]ndicesincludeindicesofannualprecipita-

tiontotal(PRCPTOT),diurnaltemperaturerange(DTR),

simpledailyintensityindex(SDII),extremetemperature

range(ETR)andannualcontributionfromverywetdays

(R95pT).Theydonotfallintoanyoftheabovecategories

butchangesinthemcouldhavesignificantsocietalimpacts.

ETRandR95pTarenotdirectlycalculatedbyRClimDex

buthavebeendefinedforthisstudyasTXx–TNnand

(R95p/PRCPTOT)*100respectively.

[20]Someoftheindiceshavethesamenameand

definitionasthoudinpreviousstudies[e.g.,Frichet

al.,2002;KleinTanketal.,2002],buttheymaydiffer

icularimpor-

tanceisarecentfindingthatinhomogeneitiexistatthe

boundariesoftheclimatologicalbaperiodudtocom-

putethethresholdsforpercentilebadtemperatureindices,

i.e.,TN10p,TN90p,TX10pandTX90p,becauofsam-

plinguncertainty[Zhangetal.,2005b].Abootstrapping

methodpropodbyZhangetal.[2005b]hasbeenimple-

mentedinRClimDexandisudtocomputeindices

tstrapprocedureremoves

theinhomogeneitiesandthuliminatespossiblebiasinthe

trendestimationoftherelevantindices.

[21]Unlikeprevious‘‘global’’studies,thispaperalso

analysasonalvaluesofsomeindices,suchasthe

percentileindices,theabsolutetemperatureindicesand

sbeenmadepossible

becauRClimDexandFClimDexalsoprovidemonthly

valuesforthoindices.

ology

[22]Weconducttwodifferentanalysoftemporal

minestrendsinstationand

gridpointdata,andtheothercomparempiricalprobabil-

itydistributionfunctions(PDF)forvariousperiodsduring

edescribeourmethodforgridding

thestationdata.

ngtheIndices

[23]Thestationsavailableforthisstudyarenotevenly

evendistri-

butionmakesitdifficulttoaccuratelycomputeglobal

eaverageofdatafromallavailable

stationswouldresultinareprentationbiadtowardareas

tal.[2002]addresdthis

problembythinningthestationnetworksothattherewas

approximatelyonestationevery250,his

approachproducesamoreevenlydistributednetwork,itis

atthecostofdiscardingotherwiufulinformation.

Maximizingthenumberofstationshelpstoreducethe

,thereisno

placeintheworldforwhichwehavetoomuchclimatedata.

Inaddition,thechoiceofstationstoberetainedissomewhat

subjective,especiallyinregionswherethenetworkis

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

5of22

D05109

rtocompareobrvedindiceswith

globalclimatemodelsimulations,Kiktevetal.[2003]

griddedsomeoftheFrichetal.[2002]indicesdataonto

aregularlatitude-longitudegrid,usingamodifiedversionof

Shepard’sangulardistanceweighting(ADW)algorithm

[Shepard,1968].ThisalgorithmisudbecauNewet

al.[2000]foundittobethemostappropriatemethodfor

griddingirregularlyspaceddatawhencomparedtoveral

iedapproachofNewetal.[2000]

andKiktevetal.[2003]hasalsobeenudtogriddaily

maximumandminimumtemperatures[Caesaretal.,2006].

Weadoptedthisapproachbygriddingtheindicesontoa

regularlatitude-longitudegridbyweightingeachstation

accordingtoitsdistanceandanglefromthecentreofa

leddescriptionofthegriddingmeth-

odologyisgiveninAppendixA.

nalysisandFieldSignificance

[24]SomeoftheindicesdatadonothaveaGaussian

distributionandinthecas,asimplelinearleastsquares

oreweuda

nonparametricKendall’staubadslopeestimator[Sen,

1968]tocomputetrendssincethismethoddoesnotassume

adistributionfortheresidualsandisrobusttotheeffectof

r,apositiveautocorrelation

(whichisusuallyprentintimeriesofclimatedata)

wouldmakethistestunreliable[e.g.,vonStorch,1995;

ZhangandZwiers,2004],sowealsoconsideredtherial

correlationintheresidualswhentestingthestatistical

ativeprocedurewasadopted,

originallypropodbyZhangetal.[2000]andlaterrefined

byWangandSwail[2001],tocomputethemagnitudesof

softhis

methodaregivenbyWangandSwail[2001,AppendixA].

Thismethodwasappliedwhencalculatingthetrendsand

idean

overallpicture,globaltimerieswerealsocalculatedby

paper,atrendisconsideredtobestatisticallysignificantifit

issignificantatthe5%level.

[25]Itisalsoimportanttoassstheoverall(field)

yandChen[1983]showed

thatthecollectivesignificanceoftrendsinafinitenumber

ofinterdependenttimeriesneedstobemuchlargerthan

thenominallevel,i.e.,the5%

implementasimilartechniquetoKiktevetal.[2003]in

involvedcreating1000bootstrappedfieldsusingamoving

blockbootstrapresamplingtechnique[Wilks,1997]witha

fixedblocksizeof2foreachindex(eKiktevetal.[2003,

Appendix]foradetaileddescriptionofbootstrapping

method).Thenullhypothesisisthatthepatternoftrends

estimatedfromtheactualstationdataisduetoclimate

eforeestimatethedifferencebetweenthe

1000bootstrappedfieldsandtheactualtrendsderived

fromthestationdatatodefineasuitabletofplausible

trendsthatcouldhavearinthroughnaturalclimate

heresidualstoestimatethe

95%confidenceintervalthateachgridpointhasazerotrend.

Foreachfield,thetotalareareprentedbygridpointswith

calculatethe95th

percentileoftheareasandiftheareafortheactualtrends

exceedsthis,thenourtrendscanbesaidtobefieldsignificant.

Sincethismethodiscomputationallyexpensive,fieldsignif-

icanceisonlycalculatedfortheannualindices.

ilityDistributionFunctions

[26]PDFswereproducedforeachstationwithsufficient

databybinningannualvaluesforvarioustimeperiods.A

asonalanalysiswasconductedbyaveragingthemonthly

indexvaluesforDecember–February,March–May,

June–AugustandSeptember–Novemberifdatafromat

ethenumber

ofsuchstationsvariesovertime,wechotoanalyzethe

samesubtofstationsduringeachofthetimeperiods

lowsfortheasssmentoftemporal

changeswithouthavingtoallowforuncertaintiesarising

alsotestwhethertheprobabilitydistributionsofaparticular

indexfromdifferenttimeperiodsaresignificantlydifferentor

doneusinga2-tailedKolmogorov-Smirnovtest

withanullhypothesisthattwocumulativedistribution

functionscomputedfortwoperiodsareidentical.

s

[27]Inthisction,wefirstprenttrendanalysforthe

temperatureindicesfollowedbytheprecipitationindices.

Attheendofthisction,changesintheprobability

distributionsoftheindicesareprented.

atureIndices,1951–2003

[28]Whenaveragedovertheglobe,almostallofthe

temperatureindicesshowsignificantchangesoverthe

1951–intemperatureindices,asde-

tailedbelow,reflectanincreainbothmaximumand

salsogenerallyamuchlarger

percentageoflandareashowingsignificantchangein

minimumtemperatureextremesthanmaximumtemperature

nitudeofthetrendsisalsogenerally

findingisinagreementwithpreviousstudiesusingmonthly

globaldata,e.g.,Jonetal.[1999]andregionalstudies

usingdailydata,e.g.,Yanetal.[2002].

teandPercentile-BadTemperature

Indices

[29]Figure2showsthedecadaltrendsinextremes

between1951and2003forthepercentile-badtempera-

ghthedefinitionsofthepercentile-

badtemperatureindicesarecalculatedinpercent,the

unitswereconvertedintodaysforeaofunderstanding.

FromTable1weethat74%(73%)ofthelandarea

sampledshowsasignificantdecrea(increa)inthe

annualoccurrenceofcoldnights(warmnights).Table1

alsoshowsthatthechangesarefieldsignificantforallthe

temperatureindicexceptannualmaximumdailymaxi-

mumtemperature(TXx).Annuallythelargestchangein

extremescorrespondingtoanincreainminimumtemper-

atureisoverEurasia(Figures2aand2b).KleinTanketal.

[2006]andYanetal.[2002]showedthatthiswarminghas

onallyvery

largeincreasareenintheannualoccurrenceofwarm

nights(amorethandoublingofthefrequencyofthisindex)

overNorthAfricaandnorthernSouthAmerica(Figure2b).

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

6of22

D05109

Asiaalsoexhibitswidespreadwarminginmaximumtem-

peratureextremesalthoughthetrendpatternsformaximum

temperaturearegenerallymoremixed(Figures2cand2d).

Somepartsoftheglobedoexhibitchangesinextremes

correspondingtodecreasintemperaturealthoughthe

ptiontothisisasmall

partofcentralUnitedStateswhereasignificantincrea

equivalenttoapproximately2daysperdecadeiseninthe

annualnumberofcolddays(Figure2c).

[30]Globallytheannualnumberofwarmnights(cold

nights)incread(decread)byabout25(20)dayssince

1951(Figures2aand2b).Trendsinmaximumtemperature

extremesshowedsimilarpatternsofchange,althoughof

smallermagnitude(Figures2cand2d).Changesinallthe

(indaysperdecade,shownasmaps)andannualtimeriesanomaliesrelativeto

1961–1990meanvalues(shownasplots)forannualriesofpercentiletemperatureindicesfor1951–

2003for(a)coldnights(TN10p),(b)warmnights(TN90p),(c)colddays(TX10p),and(d)warmdays

(TX90p).Trendswerecalculatedonlyforthegridboxeswithsufficientdata(atleast40yearsofdata

duringtheperiodandthelastyearoftheriesisnoearlierthan1999).Blacklinencloregionswhere

trendsaresignificantatthe5%curvesontheplotsarenonlineartrendestimatesobtained

bysmoothingusinga21-termbinomialfilter.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

7of22

D05109

percentile-badindicesemtohaveoccurredaroundthe

mid1970swhichcorrespondswithchangesinmeanglobal

temperature[Follandetal.,2001].Indeedeveryyearsince

1979hasbeenabovethelong-termaveragefortheannual

occurrenceofwarmnightsandeveryyearsince1977has

beenbelowthelong-termaveragefortheannualoccurrence

ofcoldnights.

[31]Theabsolutetemperatureindicexhibitasimilar

pattern(notshown)withlargenegativechangeinthe1950s

forminimumdailyminimumtemperatureandminimum

r,largererrorsare

likelyintheearlierpartoftherecordinpartbecauof

thensitivityoftheparticularindicestospatialcoverage.

Atthestartoftherecord,theglobalanomaliesaredomi-

natedbyextratropicalregionswhereminimumandmaxi-

imum

dailyminimumtemperature(TNn)increadbynearly5°C

between1951and2003,thegreatestchangeintheabsolute

r,thisisdominatedbythevery

largenegativeanomaliesinthe1950swhentherewas

increadsamplinguncertainty.

erTemperatureIndices

[32]Theannualoccurrenceofcoldspellssignificantly

decread(Figure3a)whiletheannualoccurrenceofwarm

spellssignificantlyincread(Figure3b).Bothofthe

resultsarefieldsignificant(eTable1).However,thetrend

inwarmspellsisgreaterinmagnitudeandisrelatedtoa

significantdecreasincoldspellsoccurredpredominantly

incentralandnorthernRussia,partsofChinaandnorthern

Canada,howeveralargepartoftheUnitedStatesshoweda

icantincreasin

warmspellswereenovercentralandeasternUnited

States,neral

increainwarmspellswashighlightedinpreviousstudies

suchasFrichetal.[2002]andKleinTanketal.[2002]

althoughthedefinitionoftheindicatorsvariesbetween

3cshowstherearesignificantdecreas

intheannualoccurrenceoffrostdaysoverpartsofwestern

resultscomparewellwiththeresultsofKiktevetal.[2003]

largestsignificantnegativetrendforfrostdaysappearsin

theTibetanPlateau(Figure3c).Theannualoccurrenceof

frostdayshavedecreadbyapproximately16dayson

average(Figure3c)overthopartsoftheglobewherethis

ualoccurrenceoficedaysis

alsosignificantlydecreasingovercentralRussiaandeastern

andwesternChina(notshown).

[33]ForcomparisonwithFrichetal.[2002],wecreate

theextremetemperaturerangeindex(ETR)fromthe

differencebetweentheannualmaximumdailymaximum

temperatureandminimumdailyminimumtemperature

(Figure3d).Althoughbothstudiesconfirmasignificant

downwardtrendinETR,ourimprovedglobalcoverage

suggeststhatdecreasinETRmightbegreaterthanwere

suggestedbyFrichetal.[2002]whohadnodataovermost

ofthetropics,r,ourstudy

maybeaffectedbysamplinguncertaintiesprimarilyinthe

1950s.

[34]Fortheothertemperature-badindices(notshown)

resignificant

positivetrendsintheannualnumberoftropicalnightsover

centralandsouthwestAsia,NorthAfricaandsouthern

tageofLandAreaSampledShowingSignificantAnnualTrendsforEachIndicatorinTable1a

Indicator

1951–2003

TotalGridPoints+veSignificantTrend,%ÀveSignificantTrend,%

MaximumTmax(TXx)102811.62.7

MaximumTmin(TNx)99724.51.1

MinimumTmax(TXn)132529.32.6

MinimumTmin(TNn)137945.01.9

Coldnights(TN10p)13060.174.0

Colddays(TX10p)13210.546.0

Warmnights(TN90p)140473.10.1

Warmdays(TX90p)127541.00.9

Diurnaltemperaturerange(DTR)10244.239.3

Frostdays(FD)10390.240.6

Summerdays(SU25)95723.43.7

Icedays(ID)10171.327.4

Tropicalnights(TR20)104927.61.2

Growingasonlength(GSL)76116.80.3

Warmspellduration(WSDI)87128.80.6

Coldspellduration(CSDI)101010.226.0

Maximum1-dayprecip(RX1day)8247.02.7

Maximum5-dayprecip(RX5day)5146.02.1

Simpledailyintensity(SDII)65314.62.8

Heavyprecipitationdays(R10)9827.87.4

Veryheavyprecipitationdays(R20)77511.75.0

Concutivedrydays(CDD)8164.06.9

Concutivewetdays(CWD)3233.14.6

Annualtotalprecipitation(PRCPTOT)111510.17.2

Verywetdays(R95p)65311.22.5

Extremelywetdays(R99p)4266.12.8

aThetotalnumbedicatesfieldsignificance

at5%level.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

8of22

D05109

r,therearenegativetrends(althoughnon-

significant)inthisindexandtheannualnumberofsummer

daycountsshow

significantincreasinthisindexoverpartsofnorthern

Canada,westernEurope,theMiddleEast,centralAsia,

AustraliaandsouthernBrazil,althoughithassignificantly

40%ofthelandareasampledshowsasignificantdecrea

inDTR(Table1)fortheperiod1951to2003,consistent

withminimumtemperatureswarmingfasterthanmaximum

reasofsignificancearelocatedover

centralandeasternRussia,muchofChinaandeasternand

centralUnitedStates.

alResults

[35]s4and5

showdecadaltrendsandtimeriesofglobalanomaliesfor

theasonaloccurrenceofcoldandwarmnights,respec-

easonsandforbothindicatorsweagainea

2

showsthatasonallymorelandareaexhibitssignificant

changesinminimumtemperaturethanmaximumtempera-

themostwidespreadsignificantchangein

re2butforindices(a)CSDIindays,(b)WSDIindays,(c)FDindays,and(d)ETR

(i.e.,TXx-TNn)in°C.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

9of22

D05109

thetemperatureindiceswiththeleastchangeusually

stillvalidevenifweanalyze

eachhemisphereparately.

[36]Increasintheasonaloccurrenceofcoldnights

rlydecreasinthe

asonaloccurrenceofwarmnightsaregenerallynonsig-

thesignificantwarminginanyasonis

alresultsshowthatnearlyall

changesintemperatureindicatingwarmingaresignificant

between1951and2003(notshown).Everyyearsince1985

hasbeenabovethelong-termaverageoftheoccurrenceof

warmnightsinbothDJFandMAM(Figures5aand5b).

Similarlyeveryyearsince1988hasbeenbelowthelong-

termaverageoftheoccurrenceofcoldnightsforboth

theasons(Figures4aand4b).

itationIndices,1951–2003

[37]Wefindageneralincreaintheprecipitation

mparedwithtemperaturechanges,we

ealessspatiallycoherentpatternofchangeandalower

tudieshaveasrted

thattherehavebeenincreasinextremeprecipitation.

Mostnotably,Groismanetal.[2005]foundwidespread

increasinveryintenprecipitation(definedastheupper

re2butfortheasonaloccurrenceofcoldnights(TN10p)indaysfor(a)December–

February,(b)March–May,(c)June–August,and(d)September–November.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

10of22

D05109

0.3%ofdailyprecipitationevents)acrossthemidlati-

r,theirindexismuchmore‘‘extreme’’

thananyoftheindicesthatwehavestudiedandtherefore

adirectcomparisonisnotpossiblebetweenthetwo

studies.

Results

[38]Becauofsmallcorrelationscales,thegridded

precipitationindiceshaveamuchsmallerspatialcoverage

computed

onthegridsshowlittlesignificance(Table1).However,

whenaveragedacrosstheglobe,weeasignificant

increainmostoftheprecipitationindicexceptfor

concutivedrydays(CDD)andconcutivewetdays

(CWD),reflectingthefilteringofnoithatmakesiteasier

attrendsintheindices

etal.[2003]and

Frichetal.[2002]alsofoundsignificantincreasinR10

butneitherofthostudiesfoundasignificantincreain

SDII.

[39]Theannualtotalprecipitationhasthebestspatial

r,theanalysisoftotalprecipitation

changeisnotthemainobjectiveofthisstudysince

re2butfortheasonaloccurrenceofwarmnights(TN90p)indaysfor(a)December–

February,(b)March–May,(c)June–August,and(d)September–November.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

11of22

D05109

manyotherstudies[e.g.,Newetal.,2000]haveanalyzed

thisvariableusingmonthlydataandhavebetterspatial

erestwastoudailydatatoanalyzea

rangeofthemoreextremeorintenprecipitationevents

6showsthatwhenaveraged

acrosstheglobethemoreextremeprecipitationeventsin

6ashowsthatthere

havebeensignificantincreasofupto2daysper

decadeinthenumberofdaysinayearwithheavy

precipitationinsouth-centralUnitedStatesandpartsof

centagecontributionfromvery

wetdaystotheannualprecipitationtotalhasslightly

incread(by1%since1951;Figure6b)althoughthe

trendisnotsignificantandthereislittlesignificanceat

thegridboxlevelinthisindicator.

[40]Changesintheannualnumberofconcutivedry

gesttrendsare

decreasinthisindexoverIndiaalthoughsignificant

asbeena

steadydeclineinconcutivedrydayssincethe1960s.

ThisgenerallyagreeswiththeearlierstudiesofFrichet

al.[2002]andKiktevetal.[2003]whoudstationdata

ltrendsinthesimpledaily

intensityindex(Figure6d)alsoagreewellwiththe

resultsfromKiktevetal.[2003]althoughunlikeKiktev

etal.[2003]thedecreasingintensityinthewestern

UnitedStatesisnotidentifiedtobestatisticallysignifi-

cantinthisstudy.

[41]Theotherprecipitationindices,althoughgenerally

coveringsmallerregions,doindicateenhancedprecipitation

r,thereareonlyverysmall

edtotem-

peraturethereismuchlessspatialcoherencebetween

regions,withlargeareasshowingbothincreasingand

decreasingtrends.

alResults

[42]Onlytwooftheprecipitationindiceshavebeen

calculatedasonally:maximum1-dayprecipitationtotals

(RX1day)andmaximum5-dayprecipitationtotals

(RX5day).ThespatialcoverageofRX5dayisbetterthan

RX1daybecauofthefactthatithasalargerdecorre-

lationlengthscale(eAppendixA).Figure7showsthe

lyallasonsthereis

anincreainthisindexalthoughitisnotsoclearin

June–August(Figure7c).Thereareveryfewareasof

significantchangeexceptforsignificantincreasover

south-centralUnitedStatesandnorthernRussiaand

CanadainDecember–January(Figure7a),northernIndia

andsouthernBrazilinMarch–May(Figure7b)and

southeasternUnitedStatesandafewothersmallregions

intheUnitedStatesandEuropeinSeptember–November

(Figure7d).

butionChangesandTrends,1901–2003

[43]Inordertoputtheaboveresultsinahistorical

context,weexaminetemporalchangesintheindicesfor

asubtofstationswithcompletecoveragefor1901–

aretheprobabilitydistributionsofeach

indexfordifferenttimeperiodsinthe20thcenturyand

irsthalf

ofthe20thcenturytherearefarfewer(byaboutafactor

of10)

areabout200temperaturestations(dependingonthe

index)and608precipitationstationsacrosstheglobe.

Wesplitthedataupintoone50-yearperiodandtwo

approximately25-yearperiods,i.e.,1901–1950,1951–

1978and1979–rtyearofthemostrecent

periodwaschonbecauitcoincideswiththestartof

llbeimportantforfuturecom-

parisonwithreanalysisdata.

ilityDistributionFunctionsUsing

StationData:1901–2003

[44]Figures8a–8dshowthePDFsforannualvaluesof

percentile-badtemperatureindicesforeachofthethree

ferencesbetweenthedistributionsforthe

occurrenceofcoldnightsandtheoccurrenceofcolddays

(Figures8aand8c)andtheoccurrenceofwarmnights

andtheoccurrenceofwarmdays(Figures8band8d)

clearlyindicatethatextrememinimumtemperatureoccur-

renceshavebeenincreasingatafasterratethanthatof

8ashowsa

markedreductionintheoccurrenceofcoldnighttime

temperaturesoverthe1901–2003,particularlyforthe

salsoamarked

increaintheoccurrenceofwarmnighttimetemper-

aturesduringthelastcentury,againwithstrongestchange

inthelastfewdecades(Figure8b).Thecoldestminimum

temperature(TNn),thewarmestminimumtemperature

tageofLandAreaShowingSignificantTrendsatthe5%LevelforEachSeasonBetween1951and2003forIndicesThat

CanBeCalculatedSeasonallya

Indicator

DJFMAMJJASON

N+veÀveN+veÀveN+veÀveN+veÀve

TXx129524.24.1124323.02.6125916.23.6127212.32.1

TNx128139.03.2129638.02.4119027.50.8124925.11.3

TXn133034.52.7134433.92.6121113.74.8141420.72.7

TNn149348.62.0137653.81.2131333.51.5144536.21.1

TN10p13280.143.013350.166.912850.540.914210.346.2

TX10p13021.226.313360.538.412453.317.813471.49.4

TN90p129840.40.3136754.30.3123239.00.1125232.40.6

TX90p135224.72.3130223.72.5119418.02.112269.10.2

DTR11833.135.811764.432.611284.922.811683.326.1

RX1day7048.82.76347.72.17564.13.68247.02.7

RX5day96412.15.38099.12.29123.63.58788.92.6

aThenumberofgridboxessampledforeachindexandasonisdenotedbyN.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

12of22

D05109

(TNx),thecoldestmaximumtemperature(TXn)andthe

hottestmaximumtemperature(TXx)havealsoincread

inthelatterhalfofthe20thcenturybutthedifference

betweenthe1951–1978and1979–2003timeperiodsis

lessobvious(notshown).

[45]Similarpatternsofchangeemergefortheother

temperatureindices(notshown).However,theremaining

indicesarehardertoanalyzebecauthedistributionsare

theindices,excepttheannual

occurrenceofsummernightsandwarmspellduration

index,thePDFforthemostrecenttimeperiodissignifi-

cantlydifferentfromthatforthefirsthalfofthe20th

e1979–2003PDFissignificantlydifferent

fromthe1951–1978PDFforallofthetemperatureindices

studied,indicatingashifttowardwarmerconditionsinthe

mostrecentdecades.

[46]Fortheprecipitationindicestherearefewerclear

signsofchange(eFigure9),althoughthemostrecent

timeperiodissignificantlydifferentfromthe1901–1950

ralstatisticaltestsshow

changesintheprecipitationindicesthatareconsistentwith

r,theresultsaredifficultto

re2butforprecipitationindices(a)R10indays,(b)R95pT(i.e.,(R95p/

PRCPTOT)*100)in%,(c)CDDindays,and(d)SDIIinmm/day.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

13of22

D05109

quantifyandtheirsignificancemaybeaffectedbythevery

non-Gaussiannatureoftheprecipitationindices.

ilityDistributionFunctionsUsing

GriddedData:1951–2003

[47]Between1951and2003thereismuchbetter

mparingPDFs

withinthisperiod,d

mize

samplingerrorduetodifferentspatialcoverageindiffer-

entperiods(ection4.1.1),afixedtofgridboxes

whichhavenomissingdataoverthisperiodhavebeen

esultswerealsocomparedwiththePDFs

computedfromgridboxesthathad80%ofcompletedata

sindicatethatthe

varyingsamplemadelittledifferencetotheresultforthe

percentile-badtemperatureindices,buthadamuch

largereffectforsomeoftheotherindices,particularly

becauthepercentile-bad

temperatureindicesarecalculatedwithrespecttothe

localclimate(withanexpectedoutcomeof10%during

thebaperiodregardlessofregionalclimate)whilesome

oftheotherindicescandiffersignificantlybetween

re2butfortheasonalmaximum5-dayprecipitationamount(RX5day)inmmfor

(a)December–February,(b)March–May,(c)June–August,and(d)September–November.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

14of22

D05109

probabilitydistributionfunctionsfor(a)coldnights,(b)warmnights,(c)colddays,

and(d)warmdaysforasubtofstationswithatleast80%completedatabetween1901and2003for

threetimeperiods:1901–1950(black),1951–1978(blue),and1979–2003(red).Thexaxisreprents

thepercentageoftimeduringtheyearwhentheindicatorswerebelowthe10thpercentileorabovethe

berofstationsineachcaisgiveninthetopright-hand

corner.

re8butfor(a)maximum1-dayprecipitationamount,(b)heavyprecipitation

days,(c)annualtotalprecipitation,and(d)verywetdaysandasubtofstationswithatleast80%

res9a,9c,and9dthexaxis

reprentsamount(inmm),andinFigure9bitreprentstheannualnumberofdayswhen

sizesare10forFigures9aand9band50forFigures9c

berofstationsineachcaisgiveninthetopright-handcorner.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

15of22

D05109

ryingspatialcoverage

atthestartandendofthetimeperiod,e.g.,lackofdata

inthetropicsinthe1950s,changesthecompositionof

sreasonwe

onlyshowthepercentile-badtemperatureindicesin

tributionsoftheindicesusingthe

fixedgridsaresignificantlydifferentbetweenthetwo

timeperiods,withverynotableshiftsinthedistribution;

theminimumtemperaturepercentile-badindicesshow

themostmarkedshiftstowardlesscoldnightsandmore

sintheabsolutetemperatureindices

(notshown)aremorecomplextoasssastheydonot

necessarilyappearasasimpleshiftinthedistribution.

However,ingeneralthelatterperioddoesappearwarmer

andwetterthanthe1951–1978period.

[48]Fornineofthetemperatureandtwoofthe

precipitationindiceswearealsoabletoassschanges

gofminimumtemperature

extremesisapparentduringallasonsalthoughchanges

betweenDecemberandMayaregenerallymorepronounced,

withleastchangegenerallyinSeptember–November.

Figure11showstheasonalresultsfortheannual

m

temperaturexhibitasimilarpatternofchangealthough

themagnitudeofwarmingismuchsmallerinallasons.

ThishasledtoasignificantdecreainDTRduringthe

imumdailymax-

imumtemperaturePDFforJJAistheonlyindexwhich

doesnotexhibitasignificantchangebetweenthetime

periods.

inStationData

[49]Foreachindexwecalculatedtrendsfortheperiod

1901–2003forthostationsthathadsufficientdatafor

pleofsuchtrendsisgiveninFigures12

and13fortemperatureandprecipitationindices,respectively.

Extrememinimumtemperatureincreasarerelativelywide-

spreadandcoherent(Figures12aand12b)andwhilewealso

eanoverallwarmingofmaximumtemperaturextremes

since1901,thepatternisnotsocoherent(Figures12c

and12d).77%ofstationsshowedasignificantincreain

theoccurrenceofwarmnights(Figure12b)while51%

showedasignificantdecreaintheoccurrenceofcold

nights(Figure12a).Althoughtheprecipitationindices

showatendencytowardwetterconditionswithsome

regionalsignificance,thereislesslarge-scalesignificance

r,analysisofmaximum1-day

rainfall,maximum5-dayrainfall,verywetdaysand

extremelywetdays(Figure13),showedthat28%,25%,

28%and13%ofstationshaveexhibitedsignificant

gestchangeinanyprecipita-

tionindexisasignificantincreaintheannualprecipita-

tiontotalat37%

resultsindicatethatthewidespreadglobalwarmingand

wettingthatwehaveeninthepast50yearsorsoislikely

tobepartofamuchlongertermtrend.

sion

[50]Oneoftheunansweredquestionspodbythe

climatecommunityiswhetherthedistributionofobrved

probabilitydistributionfunctionsforgridboxesinFigure2withnomissingdata

between1951and2003fortwotimeperiods:1951–1978(blue)and1979–2003(red).Thexaxes

reprentthepercentageoftimeduringtheyearwhen(aandb)minimumtemperatureand(candd)

maximumtemperaturewerebelow(above)the10th(90th)sizesare1.

Thenumberofgridboxesineachcaisgiveninthetopright-handcorner.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

16of22

D05109

globaltemperatureandprecipitationischangingandifso

ngesintemperatureextremesdocumented

herearewhatonewouldgenerallyexpectinawarming

world:decreasincoldextremesandincreasinwarm

ecreasinextrememinimumtemper-

aturesaregreaterthantheincreasinextrememaximum

temperature,theresultsagreewithearlierglobalstudies

[e.g.,Jonetal.,1999]andregionalstudies[e.g.,Klein

TankandKo¨nnen,2003;Mantonetal.,2001;Vincent

andMekis,2006;Yanetal.,2002;etal.,

Spatialandtemporaltemperaturevariabilityandchange

overSpainduring1850–2003,submittedtoJournalof

GeophysicalRearch,2005]whichimplythatratherthan

viewingtheworldasgettinghotteritmightbemore

edwith

theresultsofotherstudies,theasymmetryinthechanges

incoldversuswarmextremesthatareeninthisstudy,

hintsatpotentialchangesintheshapeand/orscaleofthe

re10butforasonalresultsfor(left)coldnightsand(right)warmnights.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

17of22

D05109

tion,the

evidencesuggestscomplexchangesinprecipitation

extremesbutwhichsupportsagenerallywetterworld.

However,toaccuratelyassspotentialchangesinthe

shapeofthedistributionoftemperatureandprecipitation

obrvationsrequiresadditionalrigorousanalysisbeyond

thescopeofthispaper.

sions

[51]Usingglobalstationdatawhichhavebecome

availableforthefirsttimeasaresultofinteninterna-

tionalcollaboration,wehaveprentedapreviously

unenglobalpictureofchangesintemperatureand

thefollowing:

[52]n1951and2003,over70%oftheland

areasampledshowedasignificantincreaintheannual

occurrenceofwarmnightswhiletheoccurrenceofcold

nightsshowedasimilarproportionofsignificantdecrea.

Forthemajorityoftheothertemperatureindices,over20%

ofthelandareasampledexhibitsastatisticallysignificant

changeandallbutoneisfieldsignificant.

[53]fixedtofcompletegridboxes,wefind

thatallindicexhibitasignificantchangebetween1951–

1978and1979–gisapparentinallasons

althoughMarchtoMaygenerallyexhibitsthelargest

changeandSeptembertoNovemberthesmallestchange.

Overnearlyallpartsoftheglobebothtailsoftheminimum

temperaturedistributionhavewarmedatasimilarrate.

Maximumtemperatureextremeshavealsoincreadbut

ecipitationindicesshowa

tendencytowardwetterconditionsbutnotallshowstatis-

ticallysignificantchanges.

[54]tofstationswithnear-completedata

between1901and2003andcoveringaverylargeareaof

theNorthernHemispheremidlatitudesshowsignificant

shiftsassociatedwithwarmingintheprobabilitydistri-

antialriinwarm

nighttimetemperaturesisapparentoverthe25year

periodbetween1979and2003whencomparedtothe

dtailsofminimumtemper-

imumtemperature

indicesshowsimilarchangesbutwithsmallermagni-

tributionsofallbut2ofthetemperature

indicesaresignificantlydifferentwhentheperiodbe-

tween1979and2003iscomparedwith1901–1950.

Precipitationindicesderivedfromasubtofstations

withnear-completedatabetween1901and2003and

coveringtheNorthernHemispheremidlatitudesandparts

100-yeartrendsforthepercentiletemperatureindicesfortheperiod1901–2003fora

subtofstationswithatleast80%completedatabetween1901and2003for(a)coldnights,(b)warm

nights,(c)colddays,and(d)(blue)solid

circlesindicateasignificantincrea(decrea)atthe5%level.

D05109ALEXANDERETAL.:GLOBALEXTREMEINDICES

18of22

D05109

ofAustraliadisplayatendencytowardwetterconditions

withthedistributionsfromthe1979–2003periodsignif-

icantlydifferentfromthe1901–1950periodforevery

index.

[55]Mostoftheindicesudinthisstudyareavailableto

theinternationalrearchcommunityattheET’swebsite

/ETCCDMIthroughtheeffortand

willingnessoftheregionalclimatechangeworkshoporgan-

ddatatsarealsoavailable

from.

AppendixA:GriddingMethodology

[56]Theangulardistanceweighting(ADW)methodof

calculatinggridpointvaluesfromstationdatarequires

knowledgeofthespatialcorrelationstructureofthestation

data,i.e.,afunctionthatrelatesthemagnitudeofcorrelation

inthiswe

correlatetimeriesforeachstationpairingwithindefined

latitudebandsandthenaveragethecorrelationsfalling

mizecomputationonly

pairsofstationswithin2000kmofeachotherareconsid-

methatatzerodistancethecorrelation

ynotnecessarilybethe

bestassumptionfortheprecipitationindicesbecauoftheir

noisynaturebutitdoesprovideagoodcompromitogive

d-orderpolynomialfunc-

tionisthenfittedtothebinaveragesinordertosmooththe

nethedecorrelationlengthscaleorcorrelation

decaydistanceasthedistanceatwhichthemeancorrelation

fromthefittedfunctionfallsbelow1/e[eCaesaretal.,

2006].Usingthedecorrelationlengthscale,L,wecan

defineacorrelationfunction,f[Jonetal.,1997],for

stationias

f

i

¼eÀr

i

=LðA1Þ

wherer

i

isthedistanceoftheithstationtothegridbox

isweareabletodetermineaweighting

functionforeachstation,i,thatdependsonthelocationsof

theothercontributingstations,k,

w

i

¼fm

i

1þ

X

k

fm

k

1Àcosq

k

Àq

i

ðÞ½