JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 106, NO. C3, PAGES 4349-4367, MARCH 15, 2001
Sourcesof global warming in upper oceantemperature during El Nifio
WarrenB. White andDaniel R. Cayan
ScrippsInstitutionof Oceanography,
Universityof Californiaat SanDiego, La Jolla,California
Michael D. Dettinger
UnitedStatesGeologicalSurvey,SanDiego, California
Guillermo
Auad
ScrippsInstitutionof Oceanography,
Universityof Californiaat SanDiego, La Jolla,California
Abstract. Global averageseasurfacetemperature(SST) from 40øSto 60øN fluctuatesñ0.3øC on
interannual
periodscales,with globalwarming(cooling)duringE1Nifio (La Nifia). About 90% of the
globalwarmingduringE1Nifio occursin the tropicalglobaloceanfrom 20øSto 20øN, half becauseof
largeSST anomaliesin the tropicalPacific associated
with E1Nifio andthe otherhalf becauseof warm
SST anomaliesoccurringover-80% of the tropicalglobalocean. From examinationof National Centers
for EnvironmentalPrediction[Kalnayet al., 1996] and Comprehensive
Ocean-Atmosphere
Data Set
[Woodruffetal., 1993]reanalyses,
tropicalglobalwarmingduringE1Nifio is associated
with higher
troposphere
moisturecontentand cloudcover,with reducedtradewind intensityoccurringduringthe onset
phaseof E1Nifio. During this onsetphasethe tropicalglobalaveragediabaticheat storagetendencyin the
layerabovethemainpycnocline
is 1-3 W m-2abovenormal.Itsprincipalsourceis a reduction
in the
poleward
Ekmanheatfluxoutof thetropicaloceanof 2-5 W m'2.Subsequently,
peaktropicalglobal
warmingduringE1Nifio is dissipatedby an increasein the flux of latentheatto the troposphereof 2-5 W
m-2,withreduced
shortwave
andlongwave
radiativefluxesin response
to increased
cloudcovertendingto
canceleachother. In the extratropicalglobaloceanthe reductionin polewardEkmanheat flux out of the
tropicsduringthe onsetof E1Nifio tendsto be balancedby reductionin the flux of latentheatto the
troposphere.Thusglobalwarmingand coolingduringEarth's internalmode of interannualclimate
variabilityarisefrom fluctuationsin the globalhydrologicalbalance,not the globalradiationbalance.
Sinceit occursin the absenceof extraterrestrial
and anthropogenic
forcing,globalwarmingon decadal,
interdecadal,and centennialperiodscalesmay alsooccurin association
with Earth's internalmodesof
climatevariability on thosescales.
1. Introduction
Whiteet al. [1997] found changingglobalaverageseasurface
temperature
(SST) of- 0. IøC occurringin response
to changesin
the Sun'sirradianceat the top of the atmosphere
of 0.5-1.5 W m2 on decadal(8-11 years) and interdecadal(18-23 years) period
scalesoverthe pastcentury. They alsofoundthis globalaverage
SST response
trappedin the upperlayer of the oceanabovethe
mainpycnocline.Whiteet al. [1998] conducteda globalaverage
budgetfor diabatic heat storage(DHS) changesin the upper
ocean,finding corresponding
depthaveragetemperature(DAT)
changesnearly simulatedby the transientform of the StefanBoltzmannradiation law for the Earth's surface [James, 1994].
In both of thesestudies,changesof 0.1øC were also found in
globalaverageSST and DAT on interannualperiodscales(3-7
years), warming (cooling) during E1 Nifio (La Nifia) in the
absenceof significantchangesin the Sun's irradiance. Thus
global warming of the upper oceanduring E1 Nifio appearsto
occurin response
to an internalmodeof interannualvariabilityin
Earth'shydrosphere-atmosphere-cryosphere-terresphere
system.
Copyright
2001bytheAmerican
Geophysical
Union.
Global warminghad beenobservedduringE1Nifio before. In
earlier studieslinking E1Nifio to globalclimatevariability,Pan
and Oort [1983, 1990] found E1 Nifio associated
with positive
SST anomaliesoccurringover most of the globe. Tourre and
White [1995] found the global SST patternassociatedwith E1
Nifio warmer than normal over most of the tropical Indian,
Pacific, and Atlantic Oceans,confirmedrecentlyby Allen et al.
[2000]. Recently,Nerem et al. [1999] found global warming
duringE1Nifio associated
with a stericrise in seasurfaceheight
observedin TOPEX/Poseidonsatellitealtimetry,supportingthe
hypothesisthat global averageDHS increasesduring E1 Nifio.
So the questionis not whetherglobalwarmingof upperocean
temperatureoccursduringE1Nifio but how it occurs.
Rind et al. [1999] constructed
a coupledmodelof the global
ocean-atmosphere-cryosphere-terresphere
systemand placed it
underrealisticsolarforcingprovidedby Lean et al. [1995] over
the past 400 years. They identifiedinternalpositivefeedbacks
that nearly doubledthe global averagetemperatureresponseto
solar forcing expectedfrom the StefanBoltzmannradiation law.
Thesefeedbackswere due to increasedglobal averagemoisture
contentin the troposphere,
decreased
globalaveragecloudcover,
and decreasedseaice extentat high latitudes. During peak solar
forcing, decreasedcloud cover and sea ice extent altered the
global radiation budget, allowing more solar radiation to
penetrateto the sea surface,while increasedhumidity decreased
Papernumber 1999JC000130
0148-0227/01/1999JC000130509.00
4349
4350
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NllqO
• isthemean
depth
of'thetopofthemainpycnocline.
The
the amountof longwaveradiationescapingto spaceat the top of
mean
depth
of
the
top
of
the
main
pycnocline
is
different
at
each
the atmosphere. This modelingexerciseraisesan important
location;it is definedby the depthof an isotherm,
question:do similar radiativeand hydrologicalmechanisms geographical
operateon interannual
periodscalesto produceglobalwarming which is assumedto move with vertical displacementsin the
relativeto it. This
as an intrinsic part of Earth's internal mode of climate mainpycnoclineso that T' canbe measured
isothermis definedas the averageof the coolesttemperatures
to
variability?
mixed layerextendsat each
This globalperspective
on E1Nifio andits globalcounterpart, whichthe winter-springnear-surface
the SouthernOscillation,raisesa numberof questionsthat must grid pointduringthe 40 yearsfrom 1955to 1994. The bottomof
mixed layer is definedto be 0.1øC differentfrom
be answeredin order to understandhow global warming occurs the near-surface
and seasonal
duringE1 Nifio. We first establishthe distribution
of global the SST, accountingfor bothseasonalthermoclines
warmingduringE1 Nifio overthe 41 yearsfrom 1955 to 1995, inversions.
We make extensiveuse of the componentsof NCEP Qt
finding global averageSST anomaliesfrom 40øS to 60øN
dominatedby the tropicalglobalaverage20øSto 20øN, with anomalies(Qt'): incomingQsw', outgoingQlw', outgoingQe',
approximately
half dueto largeSST anomalies
in the tropical andoutgoingQh'. The netheatflux anomalyQt' is computedas
Pacific associated with El Nifio and the other half due to -•80% of
follows:Qt' = Qsw'-Qlw'-Qh'-Qe',so that Qt' is positivewhen
thetropicalglobaloceanwarmingduringE1Nifio. In thetropical directedintothe ocean,with the samesignasQsw' anomaliesbut
globaloceanwe expectwarmingto occurin the absenceof of oppositesignasQlw', Qe',andQh' whicharepositivedirected
significantanomalous
heatexchange
with the deepoceaneither out of the ocean. Anomalieswere constructedby subtracting
through subductionalong isopycnalsand/or cross-isopycnalmonthly mean estimatesfrom long term monthly means.
turbulentheatflux in themainpycnocline.Thusa tropicalglobal Interannualanomalieswere constructedby band-passfiltering
with
averagebudgetof anomalous
DHS abovethe main pycnocline monthlyanomalies,isolatinginterannualsignalsassociated
shouldbe influencedprincipallyby the anomalous
net vertical E1 Nifio from annual and biennial variability at higher
anddecadal,interdecadal,
andseculartrendsat lower
heat exchangewith the overlying troposphereand by the frequencies
anomalousnet meridionalheat exchangewith the extratropical frequencies[White and Cayan, 2000]. Band-passfilteringfor
globalocean.In thepresent
studywe examine
thevalidityof this interannualanomalieswas conductedaccordingto Kaylot [1977]
rather simplisticview of the anomalousDHS budgetin the with half-powerpointsat 2 and 7 year periods. This particular
tropicalglobalocean. In the processwe find the sourcefor admittancewindow was selectedto capture peak variability
with El Nifio-Southern
Oscillation(ENSO) observed
warmingin the tropicalglobaloceanto be the reductionin the associated
(SVD)
net meridionaladvectionof DHS out of the extratropicalglobal overthe pastcenturyin singularvaluedecomposition
byAllenet al. [2000]andin powerspectra
by Whiteand
ocean. When we examinethe extratropical
globalaverageDHS spectra
budget,we find thecorresponding
reductionin thenetmeridional Cayan[2000].
advectionof DHS into the extratropicaloceantending to be
balanced by reduction in the net loss of latent heat to the
3. Error Analyses
troposphere.
A major questionin this study is whetherstandarderrorsin
2. Data
monthlyDHS (DAT) anomalies
of ñ1.0 x 108W s m-2(ñ0.2øC)
[White,1995]canbe reduced
enoughin the globalaverage
to
and Methods
We examine14 differentvariablesextendingover the global
oceanfrom 20øSto 60øN duringthe 41 yearsfrom 1955 to 1995.
We examineDHS andcorresponding
DAT constructed
at Scripps
Institutionof Oceanography
(SIO) (seeAppendix1)' SST, cloud
fraction (CIF), specific humidity (SpH), zonal surface wind
(ZSW), zonal surfacewind stress(zx), meridionalsurfacewind
stress(rv), total or net air-seaheatflux (Qt), and its components
sensible-plus-latent
turbulentheat flux (Qh+Qe) and shortwaveminus-longwaveradiative heat flux (Qsw-Qlw) from the
Comprehensive
Ocean-Atmosphere
Data Set (COADS) [Slutzet
al., 1985' Woodruff et al., 1993' Cayan, 1992], these same
variables from the National
Centers for Environmental
resolve interannual anomalies with an rms of-•2.0 x 107 W s m-2
(-0.05øCin DAT). Errorsin areaaverages
decrease
fromthosein
individualestimates
in proportion
to the inversesquarerootof
the numberof independent
estimates
[e.g.,Beers,1957]. White
[1995]determined
thatindividual
gridpointestimates
of upper
ocean temperaturevariability are not independent,with
anomalousinterannualvariability dominatedby decorrelation
scales
rangingfrom3 to 6 months,5ø latitude,and10ø longitude.
So Whiteet al. [1997] found-•90,--50, and-•30 independent
estimatesoccurringin the Pacific, Atlantic, and Indian Oceans
from 40øSto 60øN, respectively,
every3 months. This result
Prediction
yieldsa conservative
estimate
for standard
errorsin basinaverage
(NCEP) reanalysisby Kalnay et al. [1996], and surfacesolar monthlyDHS anomaliesof ñ0.1-0.2 x 108 W s m-2 (ñ0.02øradiativeflux (S) from a solarirradiancemodel constructedat the 0.04øCin DAT). Theseerrorsare subsequently
reducedby a
Naval ResearchLaboratory (NRL) by Lean et al. [1995]. In factorof 3 (i.e.,ñ0.3-0.6x 107W s m-2(ñ0.01ø-0.02ø
in DAT) in
Appendix2 we compareNCEP and COADS anomaloustropical computing
band-pass
changes
thatdetectinterannual
signals(i.e.,
global average air-sea heat fluxes to the SIO anomalousDHS assuming
four independent
estimates
per year). Standarderrors
tendency, finding NCEP fluxes better associatedwith the in the globalaverage
arereduced
frombasinaverage
errorsby
anomaloustropical global averageDHS tendencythan COADS nearly another factor of 2, reducing standard errors
fluxes (Figure A2). NCEP fluxes are also availableuniformly conservatively
to ñ0.3 x 107 W s m2 in DHS anomalyand
overthe globalocean,while COADS fluxesare absentover large ñ0.01øC in DAT anomalywhen weightedby area. These
parts of the tropical global ocean. Thus we utilize the NCEP standarderrorsare half an orderof magnitudesmallerthan the
reanalysisthroughoutthe presentstudy,takingcarethatbiasesdo rmsof tropicalglobalaverageDHS andDAT anomalies.
not invalidate the results.
The computationof the DHS anomalyis explainedalreadyby
Whiteet al. [1998]. It is T7-/, whereT' is the DAT anomalyand
We find (Figure1) thatdomainaverageNCEP SST anomalies
are largerthandomainaverageSIO DAT anomalies
by factors
rangingfrom1.1to 1.6.Thisbiasoccursprincipally
because
SST
WHITE ET AL.' SOURCESOF GLOBAL WARMING DURING EL NIl•10
4351
Global (40øS- 60øN)
0.30
0.15
œ\
øC 0.00
/-x
/-'%
/q
-0.15
I
-0.30
I
I
I
N/I
Zi•i?,
0.15 _
......
øC 0.00
I
I
I
Extratro
0.30
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
-
'20øN - 60øN
SST-DAT.....
RMS
SST=
0.02
C•
RMS
DAT=0.01
C
_
_
-0.15
_
_
-0.30 -I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
• •
-
Tropical(20øS- 20øN)
0.30
0.15
øC 0.00
-0.15
-0.30
El Nifio Re½
0.30
0.15
Ziiißiii'
.•
180 ø-90øW, 10øS-10øN
RMS
SST=
0.04CC E
RMS
DAT=0.03
_
•
øC 0.00 --.,,._•.,.- -- •
•
•,
•,
_
-,,,,•
,/c%
,.,,,,•/ •
•
•
•_
v
./':X
-.-,,,,• •
•
•
•
•
-
-
_
_
-0.15
_
-0.30 -I
I I I
I
I I
I I
I
I
I
I
I
I
I
Tro }ical Ocean Excludinc
0.30
0.15
I
I I
I
I
I
I I
I
I
I
I
t t I
I
-
El Nifio Rec ion
Ziiii•iiii;
........
....
RMS
0.05
C•Z
RMSSST=
DAT=0.03
C
øC 0.00
_
_
-0.15
_
_
-0.30
Number of Anomalies Greater Than Zero in the TropicalOcean
80 ii:!'.f•ii:
60
% 40
20
0
1955
1960
1965
1970
Global
-vs-Tropical
0.5
0.0
-0.5
1975
1980
Tropical
-vs-ElNiBoRegion
1985
1990
1995
Tropical
-rs-Tropical
Outside
ElNifioRegion
--DAf
-71----_••90%
CI
-1.0
-24
-12
0
12
24 -24
-12
Lag(months)
0
12
Lag(
months
)
24 -24
-12
0
12
24
Lag(months
)
Figure 1. (a) Time sequence
of globalaverageSSTanomalies
from40øSto 60øNandfrom 1955to 1995. (b) Time
sequences
of extratropical
averageSST anomalies
andDAT anomaliesfrom 20øNto 60øN. (c) Time sequences
of
tropicalglobalaverageSSTandDAT anomalies
from20øSto 20øN. (d) Time sequences
of domainaverageSSTand
DAT anomalies
for the E1Nifio regionfrom 10øSto 10øN,180ø to 90øW. (e) Time sequences
of domainaverage
SSTandDAT anomalies
for thetropicalglobalaverageregionoutsidetheE1Nifioregion.(f) Timesequences
of the
percentage
of grid pointsin the tropicalglobaloceanthathavewarmerthannormalSST andDAT anomalies.(g)
Temporallag crosscorrelations
betweentime sequences
of the variousdomainaverageestimatesof SST and DAT
anomalies,
with90%confidence
levelscomputed
for 20 effectivedegrees
of freedom[Shedecor
andCochran,1980].
anomaliesare computedfrom a joint analysisof surfacemarine
weather observationsand satellite observations[Kalnay et al.,
1996], whilst DAT anomaliesare computedfrom a far smaller
density of vertical temperature profiles [White, 1995]. In
Appendix 1, tropical global average COADS, NCEP, and SIO
SST anomalies are compared, finding COADS and NCEP
estimateslargerthan SIO estimatesby factorsrangingfrom 1.3 to
stratify our results,distinguishingthoseover the entire 41-year
recordfrom thoseover the past20 yearsof record.
Standarderrors in monthly NCEP and COADS Qt have
uncertaintiesof ñ30-50 W m-2, while Qt' estimatesare more
accurate with a conservative
standard error of ñ20 W m-2 on a
2.5ø latitude-longitudegrid [Cayan, 1992]. Propagatingthis
standarderror over basinaverages,throughthe band-passfilter,
1.6 from 1955 to 1975 and factors of 1.1 to 1.2 from 1975 to
and over the threebasinaveragereducesit conservativelyto ñ0.4
1995 becauseincreasinglyfewer numbersof verticaltemperature W m-2. This is half an orderof magnitudesmallerthanthe rms
profiles going back in time yield a progressively larger of tropicalglobalaverageQt' of-2.0 W m-2.
underestimationin optimal interpolatedestimatesin the SIO
Computation of the standarderror in anomalousmeridional
reanalysisof upper oceantemperature[White, 1995]. Thus we Ekman heat flux beginswith that in anomalousEkman flow,
4352
WHITE ET AL.: SOURCES OF GLOBAL WARMING DURING EL NIlqO
estimatedto be +0.03 m s-1on a 2ø latitudeby 2ø longitudegrid
by Lagerloefet al. [1999]. Here we computeanomalousEkman
flow on a similar grid following Lagerloef et al. [1999] (see
below), propagatingits standarderror togetherwith that in the
mean gradientof DHS into the standarderror of the anomalous
meridionalEkman heatflux, yielding a conservativeestimateof
+30 W m-2. Propagatingthis standarderror over basinaverages,
throughthe band-pass
filter, andoverthe globalaveragereduces
it conservativelyto +0.6 W m-2. This is half an order of
magnitudesmaller than the rms of tropical global average
meridional Ekman heat flux anomalies of 2-5 W m-2. Moreover,
suppressing
randomerror by domainaveraginganomalousheat
advectionacrossthe tropicalglobal oceanis more effectivethan
line averaginganomalousmeridionalheat transportat 20øS and
20øN [Auad, 1998a, 1998b].
Yet, global averageQt and meridional Ekman heat flux
anomaliesare not limitedby randomerrorsso muchas by biases
due to the mismeasure
of winds and clouds in the NCEP
and
COADS reanalysis. Quantitativemeasuresof these biasesare
largely unknown, whilst those for anomalous DHS are
characterized
principallyby an underestimation
in magnitudeby
factorsrangingfrom 1.1 to 1.6, more duringthe first half of the
recordand lessduringthe secondhalf (see Appendix 1). Thus
we makevaluejudgmentsaboutthe veracityof globalaverageQt
warmerduringE1Nifio. To quantifythesetwo contributions,
we
constructtime sequencesof domain average SST and DAT
anomaliesfor the E1 Nifio region from 180ø to 90øW, 10øSto
10øN (Figure l d) and for the remainderof the tropical global
ocean (Figure l e). For both domains we computedtheir
contributions
by dividingthe domainaverageby the total number
of grid pointsover the tropicalglobaloceanfrom 20øSto 20øN.
The rms of SST (DAT) anomaliesin the E1Nifio regionis 0.04øC
(0.03øC), whilst its contributionfrom the rest of the tropical
global ocean is 0.05øC (0.04øC).
This indicates that
approximately
half of the tropicalglobalwarmingduringE1Nifio
derivesfrom the intensewarming in the E1 Nifio region, whilst
the otherhalf derivesfrom warmingover the restof the tropical
global ocean. This occursbecause75-95% of tropical global
oceanis warmerthan normalduringE1 Nifio (Figure If). These
two contributionsare correlatedwith the tropicalglobal average
at 0.96, with some indication that E1 Nifio SST and DAT
anomalieslead tropical SST and DAT anomaliesby •-3 months
(Figure 1g).
5. RMS of Interannual DHS, SpH, CIF,
and ZSW
Anomalies
Regionalrms maxima of interannualDHS anomalies(Figure
2a) occur in the westernmidlatitudeNorth Pacific and North
ability to correlate with anomalous global average DHS
Atlantic Oceansnear40øN with peak estimatesrangingfrom 1.4
tendency,asfor examplein Appendix2.
x 107 to 2.0 x 107 W s m-2 and in the centralequatorialPacific
Ocean near 150øW with peak estimatesof 1.2 x 107 W s m-2.
Regional rms maxima of DHS anomalies expected in the
4. Time Sequencesof Global Average SST
midlatitude Southern Hemisphere are not available because
and DAT Anomalies
griddedfields of DHS anomaliesdo not extend southof-•20øS.
Global warmingin upperoceantemperatureduringE1Nifio is In the centraltropicalPacific Oceanthe regionalmaximumnear
not uniform over the globe. Earlier, Whiteet al. [1998] found 150øW dominatesrms variability in the tropical Indian and
80%-90% of the global average SST from 40øS to 60øN on AtlanticOceansby a factorof 3 or so.
A regional rms maximum of interannual SpH anomalies
decadal and interdecadalperiod scalesrepresentedby tropical
global averageSST from 20øS to 20øN. Here, on interannual (Figure 2b) in the tropicaloceanis similarto that for interannual
period scaleswe find similar results, with the rms of global DHS anomalies, with maximum variability occurring in the
easternand centralequatorialPacificOcean,reaching0.50 g kg-1
average SST anomalies (Figure la) of 0.10øC, that of
extratropicalglobal averageSST and DAT anomaliesof 0.02ø and extendingpolewardto 10ø-15ø latitudein both hemispheres.
and 0.01øC,respectively(Figure lb), and that of tropicalglobal Weaker regional maxima occur in subtropicalsouth Indian,
average SST and DAT anomalies of 0.8ø and 0.06øC, SouthPacific,and SouthAtlantic Oceans. Regionalmaximain
respectively(Figure l c). Thus tropicalglobal warmingof SST, the tropicalIndian and Atlantic Oceansare approximatelyhalf
ranging over +0.25øC, explains nearly 90% of the global that in the tropicalPacificOcean.
A regional rms maximum of interannual CIF anomalies
warming of SST, whilst extratropicalglobal warming of SST,
rangingover +0.03øCis nearlyan order of magnitudesmaller, (Figure 2c) occursin the centralequatorialPacific Oceannear
approachingthe standarderrorof +0.0 IøC. Peakglobalwarming 140øW, reaching 5.0% and spreadingzonally and poleward
occursduring E1 Nifio years of 1958, 1963, 1966, 1969, 1973, throughout the eastern and central tropical Pacific Ocean.
1977, 1980, 1983, 1987, 1991, and 1995, with global cooling Regional rms maxima in tropical and subtropicalIndian and
occurringduring La Nifia yearsin between. Correlatingglobal Atlantic Oceans, reaching 3.0%, are weaker than that in the
average SST anomalies with tropical global average SST tropicalPacificOcean. Regionalminimaoccurin the midlatitude
anomaliesfinds them correlatedat 0.99 at lag month 0 (Figure North PacificandNorth AtlanticOceans,achieving<1.5%.
1g). CorrelatingglobalaverageSST anomalieswith extratropical
Regionalmaximaof interannual
ZSW anomalies(Figure2d)
global average SST anomalies finds them correlated occur in westerntropical Pacific and easterntropical Indian
insignificantlyat 0.12 at lag month0 (not shown),indicatingthe Oceans,reaching0.8 m s-1. This is associated
with regionalrms
maxima in the subtropicaland midlatitudePacific, Indian, and
latterdoesnot riseup abovethe noiselevel.
Within the global tropicaldomain,upper oceantemperature AtlanticOceans,reaching0.6-0.9 m s-1.
anomaliesin the easternequatorialPacificOceanassociated
with
E1Nifio (that is, from 10øNto 10øS,180ø to 90øW) are sointense
6. Evolution of DHS, SpH, CIF, and ZSW
[Tourre and White, 1995] that they affect the global average anomalies
while occupyingonly about20% of the tropicalglobaloceanand
Extendedempiricalorthogonalfunction(EEOF) analysisof
only about 10% of the global ocean. Yet, Tourre and White
[1995] found some majority of the global tropical oceanalso interannualDHS, SpH, C1F, and ZSW anomalies[Weare and
and meridional Ekman heat flux anomalies, on the basis of their
WHITE ET AL.: SOURCES OF GLOBAL WARMING DURING EL NIlqO
60øNLI::.:•!il''
I ')I" I-r¾,'
I'•-?],rxl
40
oFDHs
Anomalies
4353
1()
ß
(1½w-s
m
-•) • •••,55,;,?;,,;•
•(•
.i/
t/i/•
Eq.
•
,
J/Ill
I
I
,/ / / z/
60øN
40o
20o
Eq.
20o
60øN
40o
200
Eq.
200
60øN
•ZSW
Anomalies
&/)•t//½A.'••i•
(/l
. •
40o
/' ,/// /I
//t/
I / //t//t//
20o
Eq.
200
30øE 60o
90o 120o 150o 180o 150o 120o
90o
60o
30o
0ø
30øE
Figure 2. Distributionsof the rms of (a) DHS', (b) SpH', (c) C1F', and (d) ZSW' overthe globaloceanfrom 30øSto
60øN. Hatchedregionsindicatewhere anomaliesexceed4.0 x 107 W s m-2, 0.25 g kg-1, 2.0%, and 0.5 m s-i,
respectively.Contourintervalsare2 x 107W s m-2,0.05 g kg-1,1.0%, and0.10 m s-i, respectively.
Nasstrom,1982] tells us how thesevariablesevolve as Earth's
1996,with time sequences
of the fourvariablesdisplayingpeaks
internalmodeof interannual
climatevariability
transitions
from during E1 Nifio years of 1958, 1963, 1966, 1969, 1972, 1977,
positive
to negative
phase.Weconduct
EEOFanalysis
onthese 1980, 1983, 1987, 1992, and 1995. This occurswithouthaving
variablesover the globaloceanfrom 30øSto 60øN for the 44
to force the issueby conductinga joint EEOF analysison the
yearsfrom 1955 to 1998(Plate1). Thesefirst EEOF modes four variables,indicatingthattheglobalpatternsandevolutionof
accountfor 20, 30, 18, and 21% of the total interannualvariance
interannualDHS variabilityin the upperoceanare matchedby
in DHS,C1F,SpH,andZWSanomalies,
respectively,
overthe thosein the otherthreevariablesin the lowertroposphere.This
available
record.Thenextmostenergetic
modes
account
for 8, supportsthe hypothesisthat Earth'sprimarymodeof interannual
13, 12,and14%of thetotalinterannual
variance,
respectively.variabilityis characterized
by couplingbetweenupperoceanand
Frominspection
of theamplitude
timesequences
for thesefirst lowertroposphere
on globalspacescales.Thuswe expectthis
EEOFmodes
(Platela) wefindthemoverlapping
from1957to couplingto be associated
with internalprocesses
which maintain
4354
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NllCqO
<DHS>(107
W-sm'2)
8
4
0
-4
_
_
- RMS = 2.0
-
<SpH>
(gkg
'•)
0.8
0.4
0.0
-0.4
- RMS= 0.2
_
<CIF>(%)
1.0
0.5
0.0
-0.5
<ZSW>(m
s'•)
0.2
0.1
0.0
-0.1
<S>(Wm'•)
0.4
0.2:
0.0
{
•
r"/-'\ /'"'k /,--..'"x
/"',,,
x./
•
-RMS=0.1
1955
1960
1965
1970
1975
1980
1985
,--
1990
1995
0.5
0.0
90% Cl
-0.5
:
-1.0
-24
_
,<DHS>-vs-<S>_--
-12
0
12
Lag(months)
24 -24
I
I
•
-12
I
0
I
I
SpH>
I
12
Lag(months)
-
:
<DHS>-vs-<CIF>-:
-
24 -24
•
I
•
-12
I
0
•
I
<DHS>-vs-<ZSW>•
• -
12
Lag(months)
24 -24
-12
0
12
24
Lag(months)
Figure 3. Time sequencesof (a) <DHS'>, (b) <SpH'>, (c) <CIF'>, (d) <ZSW'>, and (e) <S '> averagedover the
tropicalglobal oceanfrom 20øSto 20øNfrom 1955 to 1995. (f) Temporallag crosscorrelationsbetween<DHS'>
and <c3DHS/c3t'>
and <S'>, <SpH'>, <C1F'>, and <ZSW'>, with 90% confidencelevelscomputedfor 20 effective
degreesof freedom[Shedecorand Cochran, 1980].
globalaverageDHS anomalies
againstdissipation
overthe E1 the peak phaseof E1 Nifio in lag month 24 the tropicalglobal
Nifio cycle.
Corresponding
EEOF lag sequences
extendover the one half
ocean
fromapproximately
20øS
to20øNisdominated
bywarm
DHS weights, while the tropical atmosphereis dominatedby
cycle of interannualvariability(Plate lb), displayingthe moist SpH weights of similar pattern (Plate lb (middle left)),
dominanttemporaland spatialevolutionof interannual
DHS, indicatingthat the lower troposphereis anomalouslymoist(dry)
C1F,SpH,andZWS anomalies
occurring
onaverage
overthe40- over warm (cool) DHS anomalies. This occursin conjunction
yearrecord.The firstmodeof eachof thesevariables
is quasi- with warm (cool) air temperatureanomalies in the lower
periodicat the 4-yearperiodscale. Symmetries
occuraboutthe troposphere,
allowingit to carrymore(less)watervapor[Peixoto
equatorin eachvariable,indicating
thatthetropicalglobalocean and Oort, 1992]. Sinceatmosphericwatervapor is a greenhouse
links interannual variability in Northern and Southern gas, increasingit should decreasethe amount of warm DHS
Hemisphere.
TheDHS lagsequence
(Platelb (left))displays
the anomalylost to spacethroughoutgoingQlw anomaly,all else
evolutionfromthetropicalcoolphaseof La Nifia in lagmonth0 beingequal. Moreover,the increasein anomalouswatervaporin
to the tropicalwarmphaseof E1Nifio in lagmonth24. During the lower troposphere
may mitigateoutgoingQe anomaly
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NI]qO
instigatedby warmerDHS anomalies. During E1 Nifio in lag
month24 the spatialpatternof interannualC1F weightsin the
tropical atmosphere(Plate lb (middle right)) bears some
resemblancewith those of interannualDHS and SpH weights,
with positiveC1F weightsin the centraland easternequatorial
Pacific Ocean but negative C1F weights over most of the
remainderof the tropicalglobalocean.We can surmisethat this
particularpattern of C1F weights occurs in responseto an
increasein cumulusconvectionin the equatorialPacific Ocean
associated
with E1Nifio, creatingsubsidence
over the restof the
tropicalglobaloceanthroughanomalous
meridionalHadleycell
activity and extendinginto the Atlantic and Indian Oceans
throughanomalouszonal Walker cell activity [e.g., Webster,
1994]. Since cloud cover both shieldsthe sea surfacefrom
incomingQsw and shieldsspacefrom the outgoingQlw, it
should modulate heating of the upper layer of the ocean
depending
on whichradiativeheatflux termdominates
theother.
Duringthe transitionfrom La Nifia to E1Nifio in lag month12,
C1F weightsare negativeover mostof the globe exceptin the
subtropicalPacific Ocean, indicatingthe possibilityof global
warmingtendencyfrom Qswanomalies.Belowwe establish
the
relativemagnitudeof thesetwo radiativeflux termsandtheirnet
response
to theglobalpattemof anomalous
C1F.
During E1 Nifio in lag month 24 the spatial pattern of
interannualZSW weightsin the tropical atmosphere(Plate lb
(right))displayssomesimilaritieswith thoseof interannual
DHS,
SpH, and C1F weights in the tropical global ocean,with a
tendencyfor westerly(easterly)ZSW weightsto occuroveror to
the west of warm (cool) DHS and positive (negative) C1F
weights. This indicatesthat trade wind intensityis weaker
(stronger)over or to the west of tropicalwarm (cool) DHS
anomalies on interannual timescales.
and <C1F'>, with westerly<ZSW'> reducingtradewind intensity
and leading tropical global warming by ~9 months. From
inspectionof Plate 1 we find thesewesterly<ZSW'> occurring
mostly in the westernand centraltropical Pacific Ocean, with
positive <C1F'> occurring mostly in the central and eastern
equatorialPacific Ocean, dominatingnegative <C1F'> outside
this domain.
Summarizing how Earth's internal mode of interannual
variability evolves from La Nifia to E1 Nifio in Plate 1, we
examinethe transitionfrom cool DHS weightsin lag month0 to
warm DHS weights in lag month 24. This transition occurs
becauseeither <Qt '> or net meridionaladvectiveheat flux (<MF'>) generatesa <DHS'> warming tendencynear lag month
12. In Figure 3, time lag cross correlationstell us that this
<DHS'> warming tendencyis associatedwith negative<C1F'>
andwesterly<ZSW'>. This suggests
that incoming<Qs•v'> and
outgoing<Qlw '> are positive,outgoing<Qh '> and <Qe '> are
negative, and <-MF'> from tropical to extratropicalocean is
negative. The net effect of theseanomalousheat sourceson the
<DHS'> budgetis given below.
7. Tropical Global Average DHS budget
The budgetof <DHS'> from 20øS to 20øN is influencedby
<Qt> and <-MF'> betweentropical and extratropicaloceans,
assumingthe main pycnoclinein the tropical ocean acts as an
impenetrablebarrier to anomalousturbulentheat flux into the
deep ocean. This is representedwith the following linearized
anomalousheatbalanceequation:
<c3DHS/c3t'>
= <Qt'-MF'>,
We can surmise that this
occursin response
to a decrease(increase)in sealevel pressure
(SLP) anomaliesin the tropical oceanover warm (cool) SST
anomalies in association with
4355
anomalous cumulus convection
[Grahamand Barnett, 1987]. Sincewesterly(easterly)ZSW
anomalies
weaken(strengthen)
outgoingQe andQh anomaliesin
the tropicalocean,it shouldprovidea positivefeedbackto warm
(cool) DHS anomalies.Duringthe transitionfrom La Nifia to E1
Nifio in lag month12, ZSW weightsare westerlyover mostof
the globe except in the tropical Indian, easterntropicalNorth
Pacific,andtropicalAtlanticOceans,indicatingthe possibilityof
globaltropicalwarmingtendencyfrom Qe andQh anomalies,
the
magnitudeof the latter dependingon anomalouswind speed
[Cayan, 1992]. Below we establishthe relativemagnitudeof
thesetwo turbulentflux terms and their responseto anomalous
(1)
<-MF'>= <- Vyc3
DHS/Oy'-Vy'c3
DHS/0y>,
ß
whereQt• and DHS havebeendefinedpreviously,the prime
designating anomalies, angle brackets representingthe area
averageover the global tropicaloceanfrom 20øS to 20øN, and
Vy(Vy•) representing
themean(anomalous)
meridional
velocity
componentin the upper ocean. Horizontal eddy heat flux
anomalies
areassumed
to benegligible.
We estimate
Vy with
geostrophic-plus-Ekman
flow in the near-surfacemixed layer
computedby Peterson and White [1998]. We estimatethe
anomalous
Ekmancomponent
of Vy•, computedfrom NCEP
xx'and
xy' following
Lagerloef
et
al.
[1999],
with
Vy'=(-f'cx'+ r'cy')/(H(f 2 + r2)), whereH is the meandepth
of the top of the main pycnoclineover which DHS is computed
[White et al., 1998], f representsthe Coriolis parameter,and r
ZSW.
representsthe Rayliegh friction coefficient in a momentum
Now we constructtime sequences
of tropical global average balanceyielding downwindflow on the equatorwherefgoes to
interannualDHS, SpH, C1F, and ZSW anomalies,includingS zero, with the magnitudefor r scaledasfat 5øN. The anomalous
anomalies(Figure 3), where henceforthanglebracketsrepresent geostrophic
component
of Vy'is unableto be estimated
for lack
the tropical global averageand primes denoteanomalies. As of an adequatearray of continuoushydrographicmeasurements.
expected,
<DHS'> in Figure3 peaknearEl Nifio yearsof 1958, Regardless,
diagnosisof the <DHS'> budgetin an oceangeneral
1963, 1966, 1969, 1973, 1977, 1980, 1983, 1987, 1991, and circulation model (OGCM) below finds the anomalous
1995. The rmsof <DHS'>, <SpH'>, <C1F'>, <ZSW'>, and<S'> meridional
geostrophic
component
of Vy'smallcompared
to the
is 2.0 x 107W s m-2,0.2 g kg-1,0.30%, 0.1 m s-l, and0.1 W m-2, anomalousmeridionalEkman componentin the tropicalglobal
respectively.Temporallag crosscorrelations
between<DHS'> average.
and <S'> (Figure 3f) correlateinsignificantlyat all phaselags,
verifyingthat relativelysmallchangesin the Sun'sirradiancedid
7.1. Tropical Global Average Air-Sea Heat Flux
not influencetropicalglobalwarmingon interannualtimescales. Anomalies
Maximum correlationsbetween<DHS'> and <SpH'>, <C1F'>,
and <ZSW'> are 0.95, 0.50, and 0.70 at lag monthsof 0, +4, and
We plot time sequences
of <c•DHS/c•t'>
togetherwith <Qtß>
-9, respectively(Figure 3f). This indicatesthat tropicalglobal (Figure4). We find themsignificantlycorrelatedat 0.55 over 41
warming during E1 Nifio was associatedwith positive <SpH'> yearsfrom 1955to 1995,with <Qt'> leading<c3DHS/c3tß>
by---12
4356
WHITE ET AL.' SOURCESOF GLOBAL WARMING DURING EL NIIt,10
:',i•• 3.0
Extended
EOFofInterannual
DHS,SpH,ClFandZSWAnomalies
2.0
1.0
0.0
-1.0
-2.0
-:3.0
1960
-
1965
1970
DHS(20%)
1975
1980
SpH(30%)
1985
CIF (18%)
Eq.
,1•1
I
-
1995
ZSW (21%)
3
)'ill
1990
' ...
.
.,• ......•
..
•
, •,'.
,,..,,,, ,
-1.2
• '
':•'• ,
. .,,.. .
-0.9
-0.6
-0.3
0.0
0.3
0.6
0.9
'I
;
' '
,••.. ,:.'
1.2
Plate 1. (a) Amplitudetime sequences
of the dominantEEOF modefor DHS, SpH, CIF, andZSW anomaliesfrom
1957 to 1995. (b) Correspondinglag sequences
of the dominantEEOF mode extendingover 24 lag months. Yellowto-red(blue) color indicatespositive(negative)weightscontouredat intervalsof 0.2 standardized
units.
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NIlRIO
-:?:;a:
<8DHS/Ot>
-
RMS
= 1.1 _-
(W m-2)
_
x.., w ',,,/ k/'-'"
: di::
4357
w
k.,/
<Q•w-Qi•>
RMS
=0.3Z
_
_
_
_
_
_
_
_
1955
1960
1965
1970
<8DHS/St>-vs- <Qt>
1975
1980
1985
1990
1995
<8DHS/Ot> -VS- <-Qh-Qe>
<8DHS/Ot> -vs- <Qsw-Q•w>
<DHS> -vs- <-Qh -Qe>
<DHS> -vs- <Qsw-QIw>
0.5
0.0
-0.5 -
90% CI
_
-1.0
1.0
<DHS> -vs- <Qt>
0.5
0.0
-0.5
-1.0
-24
-12
0
12
24 -24
Lag(months)
-12
0
12
Lag(months)
24 -24
-12
0
12
24
Lag(months)
Figure4. Timesequences
of (a) <c3DHS/&'>,
(b) <Qt'>, (c) <-Qh'-Qe'>,and(d) <Qsw'-Qlw'>from 1955to 1995.
(e) Temporallag crosscorrelations
between
<DHS'> and<c3DHS/&'>
and<Qt'>, <-Qh'-Qe'>,and<Qsw'-Qlw'>,
with 90% confidence
levelscomputed
for 20 effectivedegrees
of freedom[Snedecor
andCochran,1980]
months,
with<DHS'>directly
outof phase
with<Qt'>(Figure respectively. Thus the net loss of heat by the tropical global
4). Thus<Qt'>actsto dissipate
tropical
globalwarming
of the ocean to the overlying atmosphereduring E1 Nifio occurs
upperoceanintotheoverlying
troposphere
duringE1Nifio. The
principallyin response
to <-Qh'-Qe'>. So, positive<SST'> and
rmsof<ODHS/Ot'>
anomalies
is 1.1W m-2,whilethatof<Qt'> <SpH'> duringE1Nifio(Figure3) yieldpositiveoutgoing<-Qh'anomalies
is 2.0 W m-2,largerby a factorof two or so.
Qe'>, theformerdominatingmitigatingaspects
of the latter.
Separating
<Qt'>intocomponents
<-Qh'-Qe'>and<Qsw'Now we separatethe influenceof the two componentsof <Qlw'> (Figure4) findsthermsof <-Qh'-Qe'>of 1.7 w m-2 Qh'-Qe'> on <ODHS/Ot'>(Figure5). The rmsof <-Qe'> is 1.5
dominating
thermsof <Qsw'-Qlw'>
of 0.3 w m-2by halfan W m-2, whilst that for <-Qh'> is 0.3 W m-2, half an order of
orderof magnitude,the latterlessthanthe standarderrorbounds magnitudesmallerandbelowthe noiselevel given by +0.4 W m-
of +0.4 W m-2(seesection
3). Temporal
lagcrosscorrelations2 (seesection3). Thus<-Qh'-Qe'>is dominated
by <-Qe'> and
(Figure4) finds <-Qh'-Qe'>correlating
significantly
with correlatessignificantlywith <ODHS/0t'> and <DHS'> at-0.60
<c3DHS/c3t'>
and<DHS'>at 0.60at lagmonths
of-12 and-0, at lag month-12 and-•0, respectively(Figure5e). So, positive
4358
WHITE ET AL.: SOURCESOF GLOBALWARMING DURING EL NIlqO
b:
<-Qh-Qe>
RMS
=1.7Z
Z
'Z
-I
I
I
I
I
•i•i;
I
I
I
I
I
I
I
J I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
<-Qh
>
I
I
t
-
RMS
=0.3 Z
_
_
_
_
_
_
•
v
_
v
•
•
••
•
v
-
-•
•
v•v
_
_
_
_
_
_
_
-I
I
I
:.e:1.0
I
I
1960
1955
I
I
I
I
I
1965
I
I
I
I
I
1970
I
I
1975
I
I
I
I
1980
I
I
I
I
I
1985
I
I
f
I
I
1990
I
-
1995
<SDHS/St> -vs- <-Qh-Qe>
<SDHS/St> -VS-<-Qe>
<SDHS/St> -vs- <-Qh>
<DHS> -vs- <-Qh-Qe>
<DHS> -vs- <-Qe>
<DHS> -vs- <-Qh>
0.5
0.0
-0.5
-1.0
1.0
0.5
0.0
-0.5
-1.0
-24
-12
0
12
Lag(months)
24 -24
-12
0
12
Lag(months)
24 -24
-12
0
12
24
Lag(months)
Figure 5. Time sequences
of (a) <ODHS/Ot'>,(b) <-Qh'-Qe'>, (c) <-Qe'>, and(d) <-Qh'> averaged
overthetropical
global oceanfrom 20øSto 20øNfrom 1955 to 1995. (e) Temporallag crosscorrelations
between<DHS'> and
<c3DHS/c3t'>
and<-Qh'-Qe'>, <-Qe'>, and<-Qh'>, with 90% confidence
levelscomputed
for 20 effectivedegrees
of
freedom [Snedecorand Cochran, 1980].
<DHS'> duringE1 Nifio (Figure3) is associated
with saturated error boundsof +0.4 W m-2. Temporallag crosscorrelations
specifichumidityat the surfaceof the sea,yieldingpositive (Figure6e) finds <Qsw'> and <-Qlw'> out of phasewith one
outgoing<Qe'> andoverwhelming
themitigatinginfluencefrom another,correlatingsignificantlywith <DHS'> at -0.7 and0.8 at
positive<SpH'> in the air above.
lag month2, respectively.
Thuspositive<C1F'>duringE1Nifio
Now we separate
the two components
of <Qsw'-Qlw'>to (Figure3) reduces
incoming
<Qsw'>andoutgoing<Qlw'>,the
formertendingto coolthe upperoceanandthe lattertendingto
warm it. Thusthe two radiativeheatflux components
interfere
destructively,
effectively
neutralizing
eachanother,to producean
determineif their relativelysmall magnitudeis the differenceof
two significanceradiativeflux components
(Figure6). We find
the rmsof <Qsw'>to be 0.8 W m-2,whilstthermsof <-Qlw'> is
0.5 W m-2, similar in magnitudeand both larger than standard
insignificant
<Qsw'-Qlw'>,with rmsvariableof 0.3 W m-2.
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NIlqO
-?•?•i!•./•ai:;:•,•,•
<aDHS/at>
:
RMS= 1.1-_
(W m-2)
-I
I
I
I
t I
I
I
I
4359
L
I
I
I
I
I
I
I
I
I
I
I
I
l
I
I
I
I
I
I
I
•!.•ii•, <Qsw-Q,w>
I
I
I
-
RMS
=0.3:
_
_
_
_
_
_
_
-I
I,I
I
I
I
I
I
I
I
i
I
I
I
I
I
i
I
I
I
I
I
I
I
I
I
I
I
I
_?:;:!?/•ii?•
<Qsw>
-
-/-%
I
I
t
-
RMS=0.8-
•
1%
/--..
/'%
/%
-
/----.
.•
_
_
_
_
_
_
-I
I
I
I
t
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
2:.'?d?• <-Qiw>
_-••---
I
I
I
-
RMS= 0.52
.....
k,..,/ •-v
_--_...•
._r---.• X..Z
_
_
_
_
_
_
_
_
_
-I
I
I
1955
I
t I
1960
I
I
I
I
I
1965
I
I
I
I
1970
<SDHS/8>t-vs- <Qsw-Qiw>
I
I
I
I
I
1975
I
I
1980
I
I
I
I
1985
I
I
I
I
I
1990
I
-
1995
<SDHS/St> -vs- <Qsw>
<SDHS/St> -vs- <-Qiw>
<DHS> -vs- <Qsw>
<DHS> -vs- <-Qiw>
0.0
-0.5 _
90% CI
--
-1.0
1.0
<DHS> -vs- <-Q•w-QIw>
0.5
0.0
-0.5
-1.0
-24
-
-12
0
12
I
24 -24
I
-12
Lag(
months
)
I
I
0
I
12
Lag(
months
)
I
-
24 -24
-12
0
12
24
Lag(mo
nths
)
Figure6. Timesequences
of (a) <ODHS/Ot'>,
(b) <Qsw'-Qlw'>,
(c) <Qsw'>,and(d) <-Qlw'>averaged
overthe
tropicalglobaloceanfrom20øSto 20øNfrom 1955to 1995. (e) Temporallagcrosscorrelations
between
<DHS'>
and<ODHS/Ot'>
and<Qsw'-Qlw'>,
<Qsw'>,and<-Qlw'>,with90% confidence
levelscomputed
for 20 effective
degrees
of freedom[Snedecor
andCochran,1980]
7.2. Tropical Global Average Meridional
Heat Flux
Anomalies
Vy'ODHS/Dy>component
of <-MF'> in (1) (Figure 7),
computed
fromtheLagerloefetal. [1999]formulation
of Ekman
flowusingNCEP .cx,and'cy'anomaly
estimates
[Kalnayet al.,
1996]. Herewe find rmsvariabilityof 2.0 Wm-2,comparable
beginbyexamining
the<-Vyc3DHS/0y'>
component
of<-MF'> with the 2.0 W m-2 of <Qt'>, bothlargerthan 1.1 W m-2 of
in (1), taking the mean meridional geostrophic-plus-Ekman<ODHS/Ot'>
bya factorof-•2. Thetemporal
lagcross
correlation
flowVyfromPeterson
andWhite[1998].However,
wefindthe between<-MF'> and <ODHS/Ot'>
(Figure7e) is statistically
magnitudeof this componentinsignificant,<0.3 W m-2, well significant
at 0.50-0.70at lag month0, with betteragreement
below the standarderrorboundsof +0.6 W m-2 (seesection3). occurring
duringthe last 20 yearsof record. Thuspositive
So we choosenot to display them. Next we examine the <- <ZSW'> duringthe onsetphaseof E1Nifio (Figure3) yieldsa
We now examine the role of <-MF'>
on <•DHS/Ot'>.
We
4360
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NIlqO
-j•,?i'ii•;•i?iii•
<-MF>
RMS=2.0 -
-vy
v
z
-i
i
i
i
i i
i i
I
I I
I
I
'-'V '"' U
I
I
I
I I I
I
I I
I I
I I
I
I
z
I
I
I
I
-
'z
-:A /\ ,,,,
,,,,
,., ...,-x
Z
/
'-/
Z
Z
-_
-I
4
0
i
I
i
I
I
I
I
I
2i .•
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
<SDHS/St>
-
I
I
i
I
-
RMS
=1.1-
/•
_----w-•'
w v
_
,,,j •'-'
w---
w
L/
•v_,
_
_
-4
_
_
_
_
-I
I
I
I
1955
I
I
I
I
1960
I
<SDHS/St>
-vs-<Qt-MF>
1955-1995--
0.5
-1.0
I
I
-
-
;
/d,-'
%',.
_
1.0
_
•
I
I
I
I
I
-
<DHS>-rs- <Qt-MF>
0.5
I
I
I
I
1970
1 I
1975
<SDHS/8>-vs-<-MF>
I
I
I
1980
I
I
I
I
1985
<SDHS/St>-VS-<Qt>
I
I
I
I
1990
I
I
I
-
1995
< Qt>-vs-<-MF>
-
,9-"-q,
-
' ix
"-5
- I I I
I I I -
<DHS>-rs- <-MF>
- • I I
I I I -
<DHS3-VS-<Qt>
?...,,>--%
0.0
_
-0.5
-1.0
-24
I
-%'?
_
_
-
I
:
•75-1995
-•..,,._
o.o
-0.5
•
I
1965
,••.•"•_• / ,'
_
-12
0
12
Lag(months)
24-24
-12
0
12
Lag(months)
_
24-24
-12
0
12
24
Lag(months)
Figure 7. Time sequences
of (a) <-MF'>, (b) <Qt'>, (c) <Qt'-MF'>, and(d) <c3DHS/c3t'>
averaged
overthetropical
global oceanfrom 20øSto 20øN from 1955 to 1995. (e) Temporallag crosscorrelations
between<DHS'> and
<c3DHS/3t'>
and<-MF'>, <Qt'>, and<Qt' -MF'>, with90% confidence
levelscomputed
for 20 effectivedegrees
of
freedom[Snedecor
and Cochran,1980]
reductionin outgoing<-MF'>, which anomalously
heatsthe <DHS'> (Figure 7e), it actsto dissipate<c3DHS/c3t'>;
because<tropicalglobalocean.
MF'> is in phasewith <c3DHS/c3t'>
(Figure7e), it actsto drive it.
The temporallag crosscorrelationbetween<Qt'-MF'> and
<c3DHS/c3t'>
(Figure 7e) is significant,0.50 at lag month-4 over
7.3 Tropical Global AverageNet Heat Flux
the entirerecordand0.70 at lag month0 overthe last20 yearsof
Anomalies
record. This indicatesthat 25%-50% of the tropical global
Finally,wearein position
to compute
<Qt'-MF'>ontheright warming during E1 Nifio can be explainedstatisticallyby the
handside(1). This producesthe anomalous
net heatflux with anomalousnet heatflux into and out of the tropicalglobalocean.
rms variabilityof 2.9 W m-2 (Figure7c), nearlya factorof 3 Evenso,the rmsof <c3DHS/c3t'>
is nearlyonethirdthatof <Qt"
greater than rms variability of 1.1 W m-2 for observed MF'>; sothe <DHS'> budgetdoesnot close.Sincethe portionof
<ODHS/Ot'>
(Figure7d). Because
<Qt'> is outof phasewith <Qt'-MF'> driving <c3DHS/c3t'>
is <-MF'>, the latter coolsthe
WHITE ET AL.' SOURCESOF GLOBAL WARMING DURING EL NIlqO
•?iiii::J•?
(Wm'2)
RMS:
1.82
',-'"' Uv'-'¾
-I
I
I
I
I
I
I
I
I
-•iii!iiiiii•i!!!.i
i
I
I
I
4361
I
I
I
I
I
I
I
I
I
I
b.,'
I
I
I
I
I
v_-
I
I
<-Qh-Qe>
I
I
I
-
RMS:
1.0Z
_
•
_
_
-I
I
I
I
I
I
I
I
I
2•ii....
i!
i
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
<Qsw-Q•w>
--
•
1955
I
1960
--
1965
v
1975
•
V
1980
I
-
v
1985
1990
1995
<Qsw-Qw>-vs-<-MF>
<'Qh'Qe>'vs- <-MF>
<Qt
> -VS<-MF>
I
RMS
=0.2Z
•
1970
I
0.0
-0.5
-1.0
-24
-12
0
12
24 -24
Lag(months)
-12
0
12
Lag(months)
24 -24
-12
0
12
24
Lag(months)
Figure8. Timesequences
of (a) <-MF'>, (b) <Qt'>,(c) <-Qh'-Qe'>,and(d) <Qsw'-Qlw'>
averaged
overthe
extratropical
globalocean
40øto 20øSandfrom20øto 60øNfrom1955to 1995.(e)Temporal
lagcross
correlations
between
<-MF'>and<Qt'>,<-Qh'-Qe'>,and<Qsw'-Qlw'>,
with90%confidence
levelscomputed
for20 effective
degrees
of freedom[Snedecor
andCochran,1980].
extratropical
globaloceanas it warmsthetropicalglobalocean magnitude. Yet in Figure 1, extratropical<SST'> was found to
since<-MF'> is onlyableto redistribute
DHS',notchange
its rangeover +0.03øC, nearly an order of magnitudesmallerthan
globaloceanaverage.
the tropical<SST'> of +0.25øC and approaching
the standard
error boundsof +0.01øC (see section3). Thus coolingof the
extratropicalglobaloceanthrough<MF'> mustbe balancedby
8. Extratropical Global Average DHS
somemix of extratropical
<Qt'> andsubduction
and/orturbulent
Budget
mixingof extratropical
<DHS'> intothemainpycnocline
below.
Foregoingresults(Figure7) indicatethat the tropicalglobal
Extratropical
<-MF'> (Figure8a), computed
from40ø to 20øS
warmingtendencydue to <-MF> duringthe onsetof E1 Nifio and20ø to 60øN,is nearlyout of phasewith tropical<-MF'> in
correspondsto an extratropicalcooling tendency of similar Figure7, with troughsoccurring
6-18 monthspriorto E1Nifio in
4362
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NllqO
10
-:! ........
I I I
I I I I
I I I I
I I I I
I I I I
I I I I
/*,,
,-,
I I I I
I I I I
•::!;a:::
<-MF> ':•
: .....
(Wm'2) _
,-,
_
.,'
•bserved
•
Model
5
0
•
I I I I
Obs.RMS=2.4
/Model
RMS:
2.9
.
Correlation
=0.6
V,:
"
I I I I
I I I I
-
I I I I
I I I I
I I I I
-
- b?:I
•:i.
I I<Qt
I I> I I I I I I I I I I I I I I I I t I I I I Obs.
RMS=2.6
Model
RMS
=2.2
-A
/\
x,
"
;^
'•
Correlation
=08_
?..-
-5
-I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
-
10
5
::::':C::
<-PF>
Model
RMS:
1.8_
--
I/ x•
I-.
0
i,•
--
,, lit.,
ß
_
_
-5
_
_
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
•
I
I
I
I
I
I
I
I
10
5
0
-5
10
5
0
-5
1955
1960
1965
1970
Observed-vs- Model<SDHS/3t>
1.01955-1995
0.5
0.0
-O.5
I
i
I
•
1975
1980
1985
<Qt>-vs- <-PF>
1990
1995
<-MF> -vs- <-PF>
-
1975-1995---:
",, :
- 90%Cl
'-"",,.•z•_-'•
_
_
-1.0
-24
-12
0
12
Lag(months)
24 -24
-12
0
12
Lag(months)
24 -24
•
I
-12
I
I
0
I
12
I
-
24
Lag(months)
Figure 9. Time sequences
of'obscr¾cd
and OP¾C modelestimatesin the tropicalPacificOceanfrom 20øSto 20øN
of (a) <-MF'>, (b) <Qt'>, (c) <-PF'>, (d) <Qt'-MF'-PF'>, and(e) <c3DHS/c3t'>
from 1955to 1995. (f) Temporallag
crosscorrelations
betweenobservedandmodel<c3DHS/c3t'>,
betweenmodel<Qt'> and<-PF'>, andbetweenmodel
<-MF'> and <-PF'>, with 90% confidencelevels computedfor 20 effective degreesof freedom [Snedecorand
Cochran, 1980].
1958, 1965, 1969, 1973, 1977, 1980, 1983, and 1987. The rms
of extratropical<-MF'> of 1.8 W m-2 is nearly equalto that of
tropical<-MF'> of 2.0 W m-2, but when summed,they do not
cancel as they should in a closed system with perfect data.
Moreover, extratropical <-MF'> is larger than extratropical
<Qt'> of 1.1 W m-2 (Figure 8b). The temporal lag cross
correlation(Figure 8e) finds extratropical<-MF'> correlating
significantlywith extratropical<Qt'> only over the last20 years
of the record, out of phaseat -0.50 at lag month 0. The latter
supportsthe hypothesisthat the extratropicalglobal cooling
tendencyduringthe onsetof E1Nifio by extratropical<-MF'> is
balanced by a warming tendency by extratropical<Qt'>.
Moreover, extratropical<Qt'> is dominatedby <-Qh'-Qe'>
(Figure8c), with <Qsw'-Qlw'> smallerby nearlyhalf an orderof
magnitude (Figure 8d).
Yet other sources and sinks of
extratropical<DHS'> exist to completethis balance,including
subductionand crossisopycnalturbulentmixing into the main
pycnocline[Whiteet al., 1980; Whiteand Bernstein,1981] and
mean westernboundarycurrentcontributionsto extratropical<-
Vyc3DHS/Oy'>
[Roemmich
andMcCallister,
1989].
4363
WHITEET AL.: SOURCES
OF GLOBALWARMINGDURINGEL NIlqO
to the climatologicalair-seafluxes,and a realisticsimulationis
9. Tropical PacificAverageDHS Budgetin
an Ocean General Circulation
run from 1955 to 1996.
Model
We begin by demonstratingthat the OPYC model can
Now we seekto verify conclusions
obtainedfor the tropical simulateobserved<SDHS/St'> in the tropicalPacificOceanfrom
of observedandOPYC
globaloceanby simulating
upperoceantemperature
variability 20øSto 20øN(Figure9). Time sequences
over mostof the 41-yearrecordin a primitiveequationocean <SDHS/Ot'>(Figure 9e) yield significantcorrelationsof 0.70 at
general
circulation
model(OGCM)knownastheoceanisopycnal lag month -3 overthe entirerecord(Figure9f) and0.75 at lag
(OPYC)model[Oberhuber,
1993a,1993b]. TheOPYCmodel month0 duringthe last20 yearsof record. However,the rmsof
hasbeenusedsuccessfully
by Miller et al. [1994] andAuadet al. OPYC <c3DHS/c3t'>
variability (3.3 W m-2) is larger than that
[1998a,1998b]to simulate
interannual,
decadal,
andinterdecadal observed(1.5 W m-2) by a factor of ~2. This occursbecause
upperoceantemperature
variabilityoverthePacificOcean.Here NCEP momentum,kinetic energy,and heat flux anomaliesused
we drive it with NCEP air-sea momentum,kinetic energy, and to drivethe OPYC model(Figure9c) are overestimated
by about
heatflux anomalies
anddiagnose
theanomalous
DHS budgetfor a factorof 2. Thisis soeventhoughthe observed
<SDHS/St'>is
the tropicalPacificOcean(Figure9), comparing
it to that in underestimated
(seeAppendix2) by factorsof 1.2 to 1.6.
Figure 7.
We are gratifiedto find <-MF'> (Figure9a) from the OPYC
The OPYC model extends over the Pacific Ocean from 67.5øS
modelcorrelatingsignificantlywith that constructed
outsidethe
to 66øN and from 119øE to 70øW, with periodic boundary model, similar to that in Figure 7 but for the Pacific Ocean,at
conditionsalong the latitudesof the AntarcticCircumpolar 0.60 at lagmonth0, withrmsvariabilityof 2.9 W m-2and2.4 W
Current. It is constructed
with 10 isopycnallayers(eachwith m-2, respectively.This occursin spiteof the fact that OPYC
nearly constantpotential density but variable thickness, <MF'> includes anomalousgeostrophicadvective heat flux,
temperature,
andsalinity)fully coupled
to a bulksurface
mixed while that constructed outside the model does not. The OPYC
layermodel.Thegridsizeis 1.5ø in theinteriorocean,withthe modelalsogenerates
a significantcross-isopycnal
turbulentheat
zonalgrid scaledecreased
to 0.65ø within 10ø latitudeof the flux anomaly (<-PF'>) in the main pycnocline(Figure 9c),
negligiblein <Qt'-MF'> on the right-handsideof (1).
equator.TheOPYCmodelis noteddyresolving,
butequatorial assumed
instability waves occur along the North Equatorial This model<-PF'> is significant,with rms variabilityof 1.8 W
Countercurrentand mesoscaleeddiesoccur along the subarctic m-2 approaching
the rms of OPYC <Qt'> of 2.2 W m-2.
frontal zone in the midlatitude oceans. The OPYC model is
Moreover,both<Qt'> and<-PF'> are significantly
correlated
at
drivenin a two-stage
process.First,themeanstateis computed lag month0 (Figure9f). This indicatesthat duringpeakglobal
with a 100 yearspin-upintegration
usingclimatological
air-sea warmingthe tropicalupperoceanin the OPYC modelheatsthe
through<Qt'> and warms the main pycnocline
momentum,
kineticenergy,andheatfluxes[Miller et al., 1994]; troposphere
thenNCEP air-seaflux anomalies[Kalnayet al., 1996]areadded through<-PF'> aboutequally.
Table 1. Statistics
of the<DHS'> Budgeton Interannual
PeriodScalesin (1.)
RMS of Anomalous
Heat Flux Components
(1955-1995),W m-2
Correlation
With
< 8DHS/St'
>
< ODHS / Ot' >
1.1
1.0 at 0 ø lead
< Qt'-MF' >
2.9
0.6 at 0 ø lead
< -MF'
2.0
0.7 at 0 ø lead
<Qt'>
2.0
0.5 at 90 ø lead
<-Qh'-Qe' >
1.7
0.6 at 90 ø lead
<-Qe' >
1.5
0.6 at 90 ø lead
<-Qh' >
0.3
0.2 at 90 ø lead
<Qsw -Q!w >
0.3
0.3 at 45 ø lead
0.5
-0.8 at 60 ø lead
< Qsw>
0.8
0.7 at 60 ø lead
-(vg + Ve).V(DHS')
0.3
0.1 at 0 ø lead
-V,• ßV(DHS)
2.0
0.7 at 0 ø lead
>
t
<-Q!w >
t
-V• ßV(DHS)
-(V• + V,•).V(DHS)
found to be negligible
2.0
unknown
0.7 at 0 ø lead
4364
WHITE ET AL.: SOURCES OF GLOBAL WARMING DURING EL NIlqlO
10. Discussion
of 1-4 w m-2. So, positive<SST'> and corresponding
positive
saturatedspecific humidity at the surfaceof the sea during El
Nifio yield positive<Qe'>, overwhelmingmitigatinginfluences
from positive <SpH'> in the overlying air. Positive <C1F'>
during El Nifio decreasesincoming <Qsw'> and outgoing
<Qlw'>, overwhelmingpositive<SST'> influenceswhichtendto
increaseoutgoing<Qlw'>. The net influenceof <CIF'> is small
because<Qsw'> and<Qlw'> tendto canceloneanother.
We find the sourceof tropicalglobal warming during El Nifio
arisingfrom the reductionin the net polewardEkman heat flux
out of the tropical ocean in responseto reduced trade wind
intensityduringthe onsetphaseof E1 Nifio of 2-5 W m-2. Thus
anomalousnet heating(<Qt'-MF'>) correlatessignificantlywith
<ODHS/Ot'> at 0.50 (0.70) at lag month -3 (0) over the entire
record (last 20 years of record). Resultsin Appendix 1 indicate
that attemptsat budgetclosureimproveduringthe secondhalf of
the record becauseof better estimationof <DHS'>. Remaining
sourcesof mismatchbetweenthe rms of <ODHS/Ot'>and <Qt'MF'> arise from the mismeasure(and apparentoverestimation)
of anomalousair-seamomentum,kinetic energy, and heat fluxes
in the NCEP reanalysis and from the neglect of the crossisopycnalturbulentmixing of <DHS'> into the main pycnocline,
the latter revealed in an OGCM simulation diagnosedin this
study.
One questionnow is whetherthe suspected
mismeasureof airseaflux anomaliesby NCEP, andthe neglectof anomalouscrossisopycnalturbulent heat flux in the main pycnocline,is severe
enoughto cast doubt on the basic findings. To check this we
diagnosethe <DHS'> budgetin the OPYC model for the Pacific
basin, driving it with NCEP air-seamomentum,kinetic energy,
and Conclusions
We begin to reveal how the internal mode of the Earth's
global ocean-atmosphere
cryosphereterrespheresystemoperates
on interannualtimescalesto produce tropical global average
warmingin upperoceantemperatureduringE1Nifio overthe 41
years from 1955 to 1995. Here we find global averageSST
anomalies from 40øS to 60øN on interannual period scales
dominatedby tropicalglobal averageSST of +0.25øC from 20øS
to 20øN, nearly an order of magnitude larger than the
extratropicalglobal averageSST of ñ0.03øC. The local E1Nifio
region in the eastern and central equatorial Pacific Ocean
accountsfor approximatelyhalf of the tropical global warming,
while warmingover-80% of the tropical global oceanaccounts
for the other half.
We find a warming<DHS'> tendencyin the upperlayer of the
tropicalglobaloceanduringthe onsetphaseof E1Nifio to range
from 1 to 3 W m-2. Subsequently,peak <DHS'> warming
occurringduring E1 Nifio is associatedwith highertroposphere
moisture content and cloud cover, while the warming <DHS'>
tendencyduring the onsetphaseof E1 Nifio is associatedwith
reducedtrade wind intensity. Examining the tropical <DHS'>
budgetwe find positive<SST'>, <SpH'>, and<C1F'> occurring
duringE1Nifio. We find <-Qh'-Qe'> dominating<Qsw'-Qlw'>
in <Qt'> by nearly half an order of magnitude,summarizedin
Table 1. In the radiativeheatflux both<Qsw'> and<-Qlw'> are
approximatelyonethirdthe magnitudeof <-Qe'>, but fluctuating
out of phase,their sumis half an orderof magnitudesmallerthan
<-Qe'>. In the turbulentheatflux, <-Qe'> dominates<-Qh'> by
half an orderof magnitude.Thus<Qt'> is dominatedby <-Qe'>
a 0.6SIO
I
I
I
I
I
I
I
I
I
NCEP
-
COADS
-
-
0.3
Correlation
SlO -vs- N•EP = 0.80
S I0 - vs- O:)AI::6
= O.75
r,,EEP-vs-o:)AI::6= 0.95
0.0
'
-0.3
_
I
-0.6
R.M.S.
SlO
:0.15
NSEP: 0.24
I
1956
I
1958
I
1960
ß
1962
I
I
1964
1966
I
1968
I
1970
I
1972
I
1974
O:)AI::6
=0.19
1976
0.6NCEP
Correlat ion
COADS
SlO-vs- I'{Y_P
= 0.87
0.3
,o.vs.
_
0.0
.vs.
- R.M.S.
-0.3
-
I
-0.6
976
I
1978
I
1980
I
1982
I
1984
I
1986
I
1988
I
1990
I
1926
I
1994
SlO
= 0.16
•
=0.19
•=0.18
1996
Figure A1. Time sequences
of <SST'> fromNCEP, COADS,andSIO reanalyses
averaged
overthetropicalglobal
oceanfrom 20øSto 20øN extendingfrom (a) 1955 to 1976 and (b) 1975 to 1996. Correlations
amongthe three
anomalousSST analysesare given at right, togetherwith rms of the <SST'>.
WHITE ET AL.' SOURCES OF GLOBAL WARMING DURING EL NI]qO
:a
<aDH
s/at>
_-
--
(W m-•)
-i"'X
_!:1
x'x
I I
•
I I I I
2C
I I I I
-_
&
x"x
f•
I I I I
•,
:
,n
_-V v__
,•
I I I I
•
I I I I
r•
I
1955
I
I
I
t
1960
I
t
', r, A",
I
I
1965
I
I
I
I
CO^D8.....
I
I
I
I
1970
<aDHS/at> -vs- <Qt>
:e 1.0NCEP
-
i I INCEPRMS= 1.7
,, .,.•,,-'•.,-,,_•
-,,V • .,,,
u,,
,,Xj
•,
_
I
-
•
COADS
RMS
= 1.6
z
-I
•
<-Qh-Q•>
b:x
0.5
4365
I
I
1)75
I
I
,A
I
I
1980
I
I
I
I
I
1985
I
I
I
.,:, ,
._
z
_
I
I
1990
I
-
1995
<aDHS/at> -vs- <-Qh-Qe>
<aDHS/at>-vs- <Qsw-Q•w>
<DHS> -vs- <-Qh-Qe>
<DHS> -vs- <Qsw-Q•w>
_
0.0
_
-0.5
_
90% CI
_
-1.0
1.0
<DHS> -vs- <Qt>
0.5
0.0
-0.5
-1.0
-24
-12
0
12
Lag(months)
24 -24
-12
0
12
Lag(months)
24 -24
-12
0
12
24
Lag(months)
Figure A2. Repeatof Figure4 but comparingtropicalglobalaverageNCEP andCOADS air-seaheatflux anomalies
from 20øS to 20øN.
and heat flux anomalies. The OPYC model has internally
consistentheat, salt, momentum,vorticity, and kinetic energy
conservation.Under anomalous
NCEP air-seaflux forcingit
simulates
the observed<ODHS/Ot'>with a significantcorrelation
of 0.70 (0.75) at lagmonth-3 (0) overthe41-yearrecord(last20
year of record). The rms of OPYC <ODHS/Ot'>is nearlytwice
the 1.5 W m-2 observedover the tropical Pacific Ocean.
Diagnosisof the OPYC <DHS'> budgetfinds tropicalPacific
warming occurring for the same reasonsas in the observed
<DHS'> budget,which is <-MF>. This occursin spiteof the
neglect of anomalousgeostrophicadvectiveheat flux in the
computation
of <-MF> outsidethe model,andin the presence
of
significant anomalous turbulent heat flux into the main
pycnocline(<-PF'>). This <-PF'> is nearlyequalto <Qt'> and
correlatessignificantlywith it. This indicatesthat tropicalglobal
warming during E1 Nifio in the OPYC model is not restrictedto
the upper layer of the oceanabove the main pycnoclinebut
penetratesinto the main pycnoclineas well.
Theseforegoinganalysesof the <DHS'> budgetindicatethat
while the tropical global oceanis being warmedby <-MF'>, the
extratropicalglobal oceanshouldbe being cooledby it, yielding
a zero net effect upon global <ODHS/Ot'>. Yet we found
extratropical <SST'> to be negligible compared to tropical
<SST'>.
Thus, the anomalous cooling tendency of the
4366
WHITE ET AL.: SOURCESOF GLOBAL WARMING DURING EL NINO
extratropicalglobal ocean through <-MF'> during the onset
phaseof E1Nifio mustbe balancedby a reductionin the net heat
loss both to the atmosphereabove and to the main pycnocline
below. We testedthis requirementby comparingextratropical<MF'> against extratropical <Qt'>, finding both correlating
significantlyout of phaseduringthe last20 yearsof record,with
<-MF'> dominating<Qt'> by a factorof---1.6.Yet othersources
and sinks of extratropical<ODHS/Ot'>exist to completethis
balance,including some combinationof anomaloussubduction
and turbulentflux of heat into the main pycnocline,as well as
fluctuationsin heattransportedby westernboundarycurrents,all
of which are presently impossibleto measuredirectly. The
situation is confoundedas well by the mismeasureof NCEP
momentum, kinetic energy, and heat flux anomalies,estimates,
which need to be improved.Thus the anomalousextratropical
<DHS'> budgetremainslargelyan openquestion.
This analysisyields one level of understandingconcerning
how Earth's internal mode of interannual climate variability
producesglobal warmingduringE1 Nifio. It certainlydoesnot
explainwhy it happens. The latter will requireexaminationof
the troposphere<DHS'> budgetand its couplingwith the upper
ocean <DHS'> budget. We have established here that
approximatelyhalf of the global tropical warming arisesfrom
intensewarming in the easternequatorialPacific Ocean; so, the
delayed action oscillatormodel of E1 Nifio [e.g., Graham and
White,1988] canbe calleduponto explainthat portionof intense
warming in the easternequatorialPacific Ocean. Warming over
the other ---80% of the tropical global ocean during E1 Nifio
remainsan open question. One approachis to understandhow
zonal and meridional teleconnectionslinking E1 Nifio in the
easternequatorialPacific to the rest of the global oceanplay a
role in this broadscaletropical warming. Yet, the fact remains
that global warming and cooling occurson interannualperiod
scalesin the relative absenceof any known extraterrestrialand
anthropogenicforcing on these scales. This may come as a
surpriseto many readers. It suggeststhat global warming and
coolingon decadal,interdecadal,and centennialperiodscalescan
also occur in the absenceof extraterrestrialand anthropogenic
forcing.
anomaliesin Figure 1 and Figure 3 to comewithin 80% of their
"true"
estimates.
Appendix 2:
COADS <Qt'>
Comparing NCEP
and
We compare time sequencesof <Qt'> from the NCEP
reanalysis[Kalnay et al., 1996] and from the COADS reanalysis
[Slutzet al., 1985; Woodruffet al., 1993; Cayan, 1992] (Figure
A2b). We estimatetheir relativeaccuracyby comparingeachof
them with the time sequenceof observed<0DHS/0t'> (Figure
A2a). Both NCEP and COADS <Qt'> and<-Qh'-Qe'> fluctuate
togetherwith similar magnitudesfrom 1955 to 1975 (Figures
A2b and A2c), but from 1975 to 1995 they are weaker and not
significantly correlated. The reason for this is unknown.
Temporal lag crosscorrelationsbetween<ODHS/Ot'> and both
<Qt'> and <-Qh'-Qe'> (Figure A2e) finds significantcorrelation
with
NCEP
anomalies
but
not
with
COADS
anomalies.
Furthermore <Qsw'-Qlw'> from both NCEP and COADS
reanalysisdo not correlatewith each other over any portion of
the record,with the rms of NCEP anomaliesconsistentlysmaller
than that of COADS anomaliesby more than a factor of two
(Figure A2d).
Temporal lag cross correlations between
<ODHS/Ot'> and <-Qh'-Qe'> anomalies(Figure A2d) finds
significant correlation with NCEP anomalies but not with
COADS
anomalies.
Acknowledgements. This researchwas supportedby the National
Aeronauticsand SpaceAdministration(NASA) under contractsNAG57096 in supportof the World OceanCirculationExperiment(WOCE) and
NAG5-7653 in supportof Solar Influenceson Global Change. Dan
CayanandMike Dettingerare supportedby the U.S. GeologicalSurvey's
GlobalChangeHydrologyProgram.WarrenWhite and Dan Cayanare
supportedby the ScrippsInstitutionof Oceanography.Our thanksextend
to Ted Walker, who provided the computationaland visualization
support,andto AndreaFincham,who preparedthe final figures.
References
Allan, R. J., C. K. Foiland,M. E. Mann, D. E. Parker,I. N. Smith,T. A.
Bassnet,andN. A. Rayner,ENSO and decadal-multidecadal
modesof
climaticvariabilityin global instrumentaldata,Clim. Dyn., in press,
2000.
Appendix 1: Comparing SIO, NCEP, and
COADS
<SST'>
Here we compare<SST'> from SIO, NCEP, and COADS
reanalysis,each averagedover the tropical global ocean from
20øS to 20øN (Figure A1). We make this comparisonfor two
periods;1955-1974 (Figure Ala) and 1975-1995 (Figure Alb).
During the earlier (later) period, the rms of NCEP and COADS
<SST'> are greaterthan that of SIO <SST'> by factorsof 1.31.6 (1.1-1.2). Temporal lag crosscorrelationsbetweenNCEP
and SIO <SST'> (not shown) are significant at 0.8-0.9 at lag
month0. This underestimation
of SIO <SST'> occursin spiteof
the fact that we undertake to minimize it by increasing
decorrelationscalesby a factor of 2 in the objective analysis
methodologyused to interpolatevertical temperatureprofile
estimatesonto a 2ø latitudeby 5ø longitudegrid. This new SIO
upperoceantemperature
reanalysispartiallycompensates
for the
relativelyweak samplingdensityof verticaltemperature
profiles
and yields larger upper oceantemperatureanomaliesthan in the
initial reanalysisby White[1995]. After 1975thisnew reanalysis
is able to bring SIO <SST> anomaliesto within 80% of NCEP
<SST> anomalies;similarly, we expectSIO <DAT> and <DHS>
Auad, G., A. J. Miller, and W. B. White, Heat storagesimulationand
associatedheat budgetsin the Pacific Ocean, 1, ENSO timescale,,/.
Geophys.Res., 103, 27,603-27,620, 1998a.
Auad, G., A. J. Miller and W. B. White, Heat storagesimulationand
associatedheat budgets in the Pacific Ocean, 2, Interdecadal
timescale,,/. Geophys.Res., 103, 27,621-27,635, 1998b.
Beers,Y., Introductionto the Theoryof Error. Addison-Wesley,London,
66pp., 1957.
Cayan,D., Variabilityof latentand sensibleheatfluxesestimated
using
bulk formulae,Atmos.Ocean,30, 1-62, 1992.
Graham,N. E., and T. P. Barnett,Sea surfacetemperature,
surfacewind
divergence,and convectionover tropicaloceans,Science,238, 657659, 1987.
Graham,N. E., and W. B. White, The El Nifio: A natural oscillatorof the
PacificOcean/atmosphere
system,Science,240, 1293-1302, 1988.
James, I. N., Introduction to Circulating Atmospheres,422 pp.,
CambridgeUniv. Press,New York, 1994.
Kalnay,E., et al., The NCEP/NCAR40-yearreanalysisproject,Bull. Am.
Meteorol. Soc., 77, 437-471, 1996.
Kaylor, R. E., Filtering and Decimationof Digital Time Series.Inst.
Phys. Sci. and Tech., Univ. of Maryland, College Park,Tech. Rep.
Note BN 850, 14 pp., 1977.
Lagerloef,G. S. E., G. Mitchurn,R. Lukas,andP. Niiler, TropicalPacific
near surfacecurrentsestimatedfrom altimeter, wind and drifter data,
,I. Geophys.Res., 104, 23,313-23,326, 1999.
Lean, J. L., J. Beer, and R. Bradely, Reconstructionof solar irradiance
WHITE ET AL.: SOURCES OF GLOBAL WARMING DURING EL NIlqO
4367
since1610: Implicationsfor climatechange,Geophys.Res.Lett., 22,
Weare, B., and J. Nasstrom,Examplesof extendedempiricalorthogonal
functionanalysis,Mon. Wea.Rev., 110, 481-485, 1982.
Miller, A. J., D. R. Cayan, T. P. Barnett, N. E. Graham and J. M.
Webster,P. J., The role of hydrologicalprocesses
in ocean-atmosphere
Oberhuber,Interdecadalvariability of the Pacific Ocean: model
interactions,Rev. Geophys.,32, 397-476, 1994.
response
to observedheatflux andwind stressanomalies,Clim.
Dyn., White, W. B., Designof a globalobservingsystemfor basin-scaleupper
9, 830-845, 1994.
oceantemperaturechanges,Progr. Oceanogr.,36, 169-217, 1995.
Nerem, R. S., D. P. Chambers,E. W. Leuliette, G. T. Mitchurn, and B. S. White, W. B., R. L. Bernstein,G. McNally, R. Dickson, and S. Pazan,
Giese,Variationsin global mean sea level associatedwith the 1997The thermoclineresponseto the transientatmosphericforcing in the
98 ENSO event,Geophys.Res.Lett., 26, 30,005-30,008, 1999.
interiormidlatitudeNorth Pacific during 1976-78, •. P:hys.
OberhuberJ. M., Simulationof the Atlantic circulationwith a coupled
Oceanogr., •0,372-384, 1980.
sea ice-mixed layer isopycnalgeneral circulationmodel, I, Model White, W. B., and R. L. Bernstein,Large-scalevertical eddy diffusionin
description,
•. Phys.Oceangr.,23,808-829, 1993a.
the main pycnoclineof the centralNorth Pacific,•. Phys.Oceanogr.,
OberhuberJ. M., Simulationof the Atlantic circulationwith a coupled
11,435-441, 1981.
sea ice-mixed layer isopycnalgeneral circulation model, II, Model White, W. B., J. L. Lean, D. R. Cayan, and M. Dettinger,A responseof
Experiment,•. Phys.Oceanogr.,23, 830-845, 1993b.
global upper ocean temperatureto changing solar irradiance,•.
Pan, Y.-H., and A. H. Oort, Global climate variations connectedwith sea
Geophys.Res., 102,3255-3266, 1997.
surfacetemperatureanomaliesin the easternequatorialPacificOcean White, W. B., and D. R. Cayan,Quasi-periodicityandglobalsymmetries
for the 1958-73period,Mon. Wea.Rev., 111,1244-1258, 1983.
in interdecadalupperoceantemperaturevariability, d. Geophys.Res.,
Pan, Y.-H., and A. H. Oort, Correlation analysesbetween sea surface
103, 21,335-21,354, 1998.
temperatureanomaliesin the easternequatorialPacific and the world White, W. B., D. R. Cayan, and J. L. Lean, Global upperoceanheat
ocean,Clim. Dyn., 4, 191-205, 1990.
storageresponse
to radiativeforcingfrom changingsolarirradiance
Peixoto,J.P., and A. H. Oort, Physicsof Climate, Am. Inst. of Phys.,
and increasinggreenhousegas/aerosolconcentrations,•.Geophys.
CollegePark,MD., 520 pp., 1992.
Res., 103, 21,355-21,366, 1998b.
Peterson,R. G. and W. B. White, Slow oceanicteleconnections
linking White, W. B., and D. R. Cayan, A global ENSO wave in surface
the AntarcticCircumpolarWave with tropical ENSO, •. Geophys.
temperature
andpressure
anditsinterdecadal
modulation
from 1900to
Res., 103, 24,573-24,583, 1998.
1996,J. Geophys.Res.,106, 11,223-11,242,2000.
Rind, D., J. Lean, and R. Healy, Simulatedtime-dependentclimate Woodruff, S. D., S. J. Lubker, K. Wolter, S. J. Worley, and J. D. Elms,
responseto solarradiativeforcingsince1600,•. Geophys.Res., 104,
Comprehensive
Ocean-Atmosphere
Data Set Releasel a: 1980-92,
3195-3198, 1995.
1973-1990, 1999.
Roeromich,D., and T. McCallister,Large scalecirculationof the North
PacificOcean,Prog. Oceanogr.,22, 171-204, 1989.
Slutz, R. J., S. J. Lubker,J. D. Hiscox, S. D. Woodruff, R. L. Jenne,D. H.
Joseph,P.M. Steurer, and J. D. Elms, ComprehensiveOceanAtmosphere
Data Set;Release1,268 pp., NOAA Environ.Res.Lab.,
Boulder,CO., 268 pp., 1985.
Shedecor,
G. W., andW. G. Cochran,StatisticalMethods,507 pp.,Iowa
Earth Syst.Monit., 4, 1-8, 1993.
G. Auad,D. R. Cayan,W. B. White, ScrippsInstitutionof
Oceanography,
UCSD,La Jolla,CA 92093-0230.(wbwhite•ucsd.
edu)
M. Dettinger,UnitedStatesGeological
Survey,SanDiego,CA, 92123
(_m._dde..t_tj_n_•_u_s
gs_.
g_o
y.)
State Univ. Press,Ames, 1980.
Tourre,Y. M., and W. B. White, ENSO signalsin global upperocean (ReceivedNovember8, 1999;revisedJune2, 2000;
temperature,•. Phys.Oceanogr.,25, 1317-1332, 1995.
acceptedSeptember13,2000.)