JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 106, NO. C10, PAGES 22,249-22,265, OCTOBER 15, 2001 Pacific Ocean wind stress and surface heat flux anomalies from NCEP reanalysis and observationsCrossestatistics and ocean model responses GuillermoAuad, Arthur J. Miller, JohnO. Roads,and Daniel Cayan• ScrippsInstitution of Oceanography,La Jolla, California Abstract. Wind stresses and surface heat fluxes over the Pacific Ocean from the National Center for EnvironmentalPrediction (NCEP) reanalysisand the comprehensive Ocean-Atmosphere Data Set (COADS) (blendedwith FSU tropical wind stresses) are comparedover a commontime interval (1958-1997)in their statisticsand in the responses that they inducein seasurfacetemperature(SST) and heat storage when used to force an ocean model. Wind stress anomalies from the two data sets are well correlated in the midlatitude extratropics, especially in the highly sampled North Pacific. In the tropics and subtropics, low correlations were found between the two wind stress data sets. The amplitudes of the stress variations of the two data sets are similar in midlatitudes, but in the tropics NCEP wind stresses are weakerthan the COADS/FSU stresses,especiallyon interannualtimescales. Surface heat flux anomalies from the two data sets are well correlated on interannual and shorter timescalesin the North Pacific Ocean poleward of 20øN, but they are poorly correlated elsewhere and on decadal timescales. In the extratropics the amplitudes of the heat flux variations of the two data sets are comparable, but in the tropics the NCEP heat fluxes are weaker than those of COADS. Ocean model hindcasts driven by both data sets are also compared. The midlatitude SST hindcastswere superior when using the NCEP flux anomalies while tropical SST hindcastswere equally skillful for the two hindcasts when consideringall climatic timescales. The spatial and temporal sampling rates of the COADS observations and their consequentimpacts on constrainingthe NCEP reanalysisappear to be the main factors controlling the results found here. 1. age, 1984, 1987; Barnett, 1984; Wright, 1986; Morris- Introduction Wind stress and surface heat flux fields over the ocean are often usedfor diagnosingocean-atmosphereclimate variations[e.g., Cayan, 1992b; Trenberthand Huttel, 1994; Iwasaka and Wallace, 1995; Tanimoto et al., 1997; Deser et al., 1999; Parrish et al., 2000] and for forcing and testing numerical models of ocean climate sey,1990; Cayan,1992a; Ward andHoskins,1996]that are employed in deriving the fluxes. The portion of the observational error that is random may be reduced by aggregatingover multiple flux estimatesin time and spaceso that well-sampled areas of the ocean, e.g., parts of the North Pacific and North Atlantic Oceans, have more reliable flux estimates than those that are poorly variations[e.g.,Harrisonet al., 1990;Miller et al., 1993, sampled,e.g., much of the tropics and Southern Ocean 1994; Xie et al., 2000; Stockdaleet al., 1998; Fevrier et [Weare,1989; Cayan, 1992a]. More difficultproblems al., 2000]. are those that produce constant or time-varying biases, Surface heat and momentum fluxes derived from di- caused by uncertainties in the bulk formulae and by rect observations, however, are usually sparsely saminstrumental practices such as changesin surfacewind pled in space and time, especially in tropical, subtropestimation proceduresor by increasesin ship anemomeical, and Southern Hemisphereregions. These surface ter height[Taylor, 1984]. flux estimates are also infected by uncertainties in the Atmospheric analysescan provide relatively complete bulk formulae[Blanc,1985; Weare,1989; Taylor,1984] spatial and temporal coveragebut they can be influ- and by errorsin the weatherobservations [e.g., RamThis paper is not subject to U.S. copyright. Published in 2001 by the American GeophysicalUnion. Paper number 2000JC000264. encedby modelerrors[Bettset al., 1996; Weare,1997; Bony et al., 1997; Trenberth and Guillemot, 1998; Scott and Alexander,1999; Yanget al., 1999],by inadequate quantities of assimilateddata [Waliser et al., 1999; Marshall and Harangozo,2000; Putman et al., 2000], 22,249 22,250 AUAD ET AL.: MODEL RESPONSE and by the exclusion of data that are difficult to assim- TO COADS AND NCEP FORCING flux forcing fields for the Pacific for the period 1958- ilate suchas surfaceheat fluxesand rainfall [Janowiak 1997 to determine the similarity of the two data sets et al., 1998.] over monthly, interannual, and interdecadal timescales. In recent years, surfacefluxes from the National Cen- We also use the two data sets to provide anomalous ters for EnvironmentalPrediction/NationalCenterfor forcing for an ocean model of the Pacific and compare Atmospheric Research(NCEP/NCAR) reanalysis have the results with observations of SST and heat content. becomeroutinelyavailable[Kalnayet al. 1996]. Sur- These ocean hindcasts extend the Pacific Ocean decadal face fluxes derived from direct observations associated processes studiesof Miller et al. [1994,1998],Cayanet with the ComprehensiveOcean-AtmosphereData Set al. [1995]and Auad et al. [1998b]to longertimescales (COADS) are also availablefor a similar time period with higher spatial resolution. [Woodruffet al., 1987; Cayan•992a; da Silvaet al., 1994]. The forcingfunctionsderivedfrom thesetwo This comparison provides a general understanding of the similarities and differences of the COADS and sourcesare ubiquitouslybeing usedto forceoceanmod- N CEP flux data sets and the character of oceanic reels for simulating and diagnosingobservedbasin-scale sponseto be expected when using either dataset as forcoceanclimate variations over long timescales. ing. In summary, we find that the COADS and the These ocean hindcasts have revealed numerous asNCEP reanalysis still require improvements, but that pects of the physics of oceanic climate variations on they appear to be adequate in many regionsfor forcing seasonalto interannual and decadaltimescales.Trop- long ocean model hindcasts. ical ocean hindcasting has revealedthe sensitivity of This paper is organized as follows. In section 2 we the tropical ocean circulation to the interannual wind describe both surface flux data sets; in section 3, their stressforcing and the damping effect of surface heat statistics are compared; section 4 compares responses fluxes[e.g.,Harrison et al., 1990;Barnett et al., 1991; of the ocean model to forcing obtained from both data Stockdaleet al., 1998]. In the midlatitudes,heat fluxes appearto play the major role in drivingseasurfacetemperatureanomalies(SST) anomalieson seasonalto interannualtimescales[e.g.,Lukschandyon$torch,1992; Miller et al., 1994; Battisti et al., 1995;Kleeman et al., 1996;Halliwell, 1998; Gieseand Carton,1999;Seagetet al., 2000]. However,horizontalcurrentscan contribute substantiallyto midlatitude SST evolutionduring certain intervalsand on the longertimescales[Miller et al., 1994; Giese and Carton, 1999], and especiallyin westernboundarycurrentregions[Kleemanet al., 1996; Halliwell,1998;Xie et al., 2000; Qiu, 2000].Subsurface midlatitude adjustment processesdue to Rossbywaves sets; section 5 contains a summary and conclusions. 2. Data Sets The time spanof the COADS [Woodruffet al., 1987] studied here is from January 1951 to December 1997 in a Pacific domain from 40øSto 60øN. Figure I shows the global samplingdensity of COADS wind speedand SST measurements. Wind stress(momentumflux) anomalieswerederived from COADS on a 2øX2ø grid followingCayan[1992a]. In the extratropics the COADS wind stress anomalies are derivedfrom monthly meansof the product IVIV [e.g.,Lysneet al. 1997;Miller et al., 1997]andsubduc- from individual wind speedobservations.Drag coeffition by meancurrents[Schneider et al., 1999]havealso cientsweretakenfromIsemerandHasse[1987]andare been identified in these ocean climate hindcasts. These results of driving ocean models with surface heat and momentum flux estimates, however, are impaired by inaccuraciesin the fluxesas well as by inadequaciesin the ocean models. The consistencyof model physicscan be rigorouslytested usingerror statisticsof the atmosphericforcing and of the validating oceanic data sets [e.g., Frankignoulet al., 1989, 1995; Sennechaelet al., 1994;Bennettet al., 1998,2000],but the weakly dependent upon wind speed and air-sea temperature difference, AT. We used this COADS anal- ysis, rather than that of da Silva et al. [1994],becausethe unadjusted COADS observationshave low fre- quencydrifts in primary variables,suchas wind speed, which are thought to be spuriouseffectsof instrumental changes. The COADS analysis we used makes a rough attempt to correct for changesin wind speed, A T, and surfaceheat flux, A Q by removingbasinwide computational requirementsof suchtests of full-physics low-frequency drifts from these variables before fluxes models arecalculatedvia bulkformulae(seeMiller et al. [1994] for additionaldetails).The da Silva et al. [1994]anal- can be ominous. Using observedflux variations to force ocean models clearlyleadsto a better understandingof howthe ocean can interact with the atmosphereto organize climate variations. But it is unclear how the uncertainty in the flux estimates is transferred to errors in the ocean model physical balances. How similar are the COADS and N CEP flux forcing data sets which are commonly used to force ocean models? How differently would a singlemodel respond to forcing by these two data sets? Our goal is to answer these questionsas follows: We first statistically analyze the NCEP and COADS surface ysis doesnot addressthese time-varying problemsthat could have substantial impact upon modeled decadal variability. Since the COADS wind stressobservationsare very lightly sampledin the low latitudes (Figure 1), we use Florida State University (FSU) subjectivelyanalyzed wind stressanomalies[Goldenberg and O'Brien, 1981] in the region +20 ø latitude. In an overlap region of 5 ø latitude at 20øN and at 20øS the COADS and FSU anomaly fields were smoothly merged. Unfortunately, AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP (a) COADS 2-DEG SST,TOTALNO. OF SHIP OBS(1/1955 - 12/1997) CI-1000., FORCING MIN• 0., MAX--- 22,251 251216. 03 xl 2O 19 6DN • ' ""••' • .....•' ß...... ?'•'•-• ....... .. •-:• '•- ••• "' :•'•/:•/•x ........... 5•. .... ' - ' _ •:--•ff .• ...... •:•'•• •. .• •..;•-:.:•'• .... .•-.'3:•:':'"'"'• •:•.:•:<...--,•;;•<•i. •-" <"•'•T'?J •:'" •-:'-..""-""•::•;:..• "•'•••'•' , --'.•,::•-' •:-•.:--•: '::-'-' -•"-'•:• -'•,•:'".•:• "'•:.•:'"• •,•.-. • . ..... ...•..: •.<• •:•.... •,,,•::• ...... 15 14 13 12 •..:4F '•:•:: ::7•:::• .'.... '•'•......... :• ...... :.•..... ;2•?•t• :• 11 •' '........ '.•,•:...: .... "-• •-:•I ' •'•D ........ •, ..:•,.•.•,-.•'•.,,?• .• 10 ---•-:.:--:•-v. '"...•"'•:"•"• -:::: .•,.. 17 16 '•:'• •...••. =•...•..•.. • ,,. ..:.:..,• * . -•'•:' .... ' ;.•.....•:• ...... •:"• 18 •..'"'•7.%, ..,• • - . ..... ........ . .•:•:.•.• •,.•.•.• .....-: ............. •-.• 9 :•.. 8 .• ........ •:•:• •:• 7 6 5 4 3 6•S B 9DE 1 8g •:.'•'• :•, , • 9• (b) COADS 2-DEG WIND SPEED, TOTAL NO.OFSHIPOBS (1/1955- 12/1997) CI-- i •..........•...... !....• : ___•.,,.t,_ ........}..... •. _,....t .... t,..,..,__••_ lOO0.. MIN= {........ : 0., MAX-- 213050, !..... ,.... ...s_.,.__• i .~ 2 1 0 03 xl 20 19 18 17 16 I ½.... . •:;•;:• •.,• ..., .• . ,,. .•' . ."'•'½••• ','-•<•f-v4"...r . ,. • ...... • ..½ ' 15 14 13 12 .-.-.--.--•--•••••••.•••••••, ..... "--'-•"•-•• .....ß<. --•t:•?. 11 ,•,•.•--.•:•:•,:' 10 9 8 7 6 i '•"• 5 4 3 2 1 0 • 9•E 18• 9•W Figure 1. Distributionof the total numberof observations for (top) COADS SSTsand (bottom) winds. Data plotted include observationsfrom 1955 through 1997. FSU winds are only available since 1961. So, we creMissing data values in space and time were treated ated pseudo-FSUwind stressstarting in 1951 as follows. as follows. We first linearly interpolated in time for We computedempiricalorthogonalfunctions(EOFs) of thosemissingvalueswhich had nonmissingvaluesavailFSU wind stressfrom 1961 through 1997. The principal able in the previous and following months. We then componentof the first EOF washighly correlatedto the linearly interpolated in space for any remaining missSouthernOscillationIndex. We regressedthis index for ing values which were surrounded by four grid points 1951-1960 onto spatial loading pattern for EOF-1 from with non-missingvalues. The remaining missing values the period 1961-1997. We then computed the EOFs of were estimated by quadratic interpolation using subthe COADS wind stressesfor the period 1951-1960 in routines "blockmean" and "surface" from Generic Mapthe FSU region. The secondand third COADS EOFs ping Tools(GMT). The last step wasto map all of the for 1951-1960 resemble the second and third FSU EOFs fieldsto a commongrid, which for conveniencewas chofor 1961-1997. So we used the principal componentsof sen to be the ocean model grid (discussedin section the second and third COADS EOF modes as time se- ries for the 2nd and 3rd FSU EOF modes for the period 1951-1960. The three-EOF mode reconstruction 4), a roughly 1.5ø grid coveringthe Pacific Oceanfrom Antarctica 2.1. to the Arctic COADS Wind Stress explained about 60% of the FSU wind stressvariance Heat flux anomalies were likewise computed on a 2ø in the period 1961-1997. We therefore expect that the reconstruction will explain a similar amount of wind by 2ø grid by Cayan [1992a]for the COADS observastress variance for the period 1951-1960. tions poleward of the 10ø latitude. The convention is 22,252 AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING that positive values correspondto heat entering the 2.3. ocean. The turbulent kinetic energy (TKE) flux into the mixedlayeris neededas a forcingfunctionfor the bulk mixedlayerformulationof the oceanmodeldescribedin The COADS latent and sensible fluxes were formed from monthly averagesof productsof individual observationsusing bulk formulae. Exchangecoefficientsweretaken from Isemer and Hasse[1987]and are weakly dependenton wind speedand AT. During the 1950smany observations of SST and wind speed were not accompaniedby humidity measurement,and so heat fluxes were not calculated directly from bulk formulae. Instead, heat flux was determined at those points by a regressionwith SST and wind speed. Two sets of regressioncoefficientswere computed,one for Turbulent Kinetic Energy section 4. The fields of TKE from COADS and NCEP willnotbecompared herebecause itsestimation isvery uncertainwhenusingmonthlymeanobservations. The Weibullparameterization of Pavia and O'Brien[1986] for relatingmeanwind speedcubedto monthlymean wind speedis invokedto estimate its variability as a forcingfunctionfor both flux data sets(seeMiller et al. [1994] fordetails). Theareawiththelargest TKE midlatitudes and another set for the tropics. It is im- anomalies is the central North Pacific which shows an portant to note that in the tropical band (:•10ø) the COADS dataset has many gapsin spaceand time; for increase in its variabilityin the late 1970s. conveniencethesewere filled in usingthe GMT subroutines mentioned previously 3. 2.2. NCEP/NCAR COADS and NCEP Flux Comparisons Reanalysis Our goal is to comparethe two differentflux data The atmosphericreanalysiswas obtained with the setsthat are frequentlyusedin variousoceanmodelNCEP T62 (209km) globalspectralmodelof 28 vertical ing studies. It is important to note that the NCEP levelswith 5 levelsin the boundary layer. The NCEP reanalysisassimilatesCOADS as well as other obserreanalysis modelincludes parameterizations of all ma- vations. For this reason, areas with a high density of jor physicalprocesses,namely, convection,large scale observationswill be more likely to agree. However,the precipitation,shallowconvection,gravity wavedrag, ra- fluxeswerecomputedusingdifferentbulk formulaewith waysfor estimating inputvariables, suchas diation with diurnal cycle and interaction with clouds, different boundary layer physics,an interactive surfacehydrol- SST [Kalnayet al., 1996; Cayan,1992a]. Determinare dueto the differogy,and vertical and horizontaldiffusionprocesses.The ing whetherthe main differences usedin theCOADS/FSUandNCEP details of the model physicsand dynamicsare given by entbulkformulae data sets or to the different input variable estimates Kalnay et al. [1996]. Flux fields are derived from the four times per day employedin computingthe respectiveheat fluxesis benear surfacefields produced by NCEP's global atmo- yond the scopeof this paper. It is important to highspheric analyses. These fields are convertedwith bulk light that eventhoughthe COADS and NCEP heat formulae using a constant drag coefficient. The data fluxes are computedthrough different formulationswe were then monthly averaged for use in comparing to want to comparethem asthey are, without usinga common bulk formulae framework, becausethis is the way COADS and in forcing the ocean model. ANALYZED AREAS 120øE 140øE 160øE 180 ø 160øW140øW120øWlOOøW 80øW 600W Figure 2. Key regionsusedin the correlation and coherence analysisin Tables1 and 2. The "N 20øN"regionin Tables1 and 2 is the entirebasinaveragednorth of 20øN. AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING 22,253 Table 1. Correlations,Standard Deviationsratios, Coherence,Amplitude of Transfer Function and Lag for COADS and NCEP zonal wind stress Coherence Area Corr EEP SD 0.23 WEP (CCS CNP N 20øN K.Ext. WNP 2years 2.20 (0.60) (0.89) (0.89) (0.81) (0.87) (0.75) 2.5years 0.40 (2.83) (0.91) (0.68) (0.78) (0.80) (1.31) (0.65) (0.86) (0.76) (0.77) (0.65) (0.13) 3.3years 0.27 5years 0.28 0.25 (0.84) (0.81) (0.77) (0.61) (0.65) (0.38) 10years 0.15 (085) (062) (076) (085) (080) (0.82) (0.68) (0.86) (0.82) (0.78) (0.62) (0.77) 0.55 0.55 (0.78) (0.69) 0.54 055 CORR and SD were computed using anomalous time series with no filter- ing. Resultsfrom the cross-spectral analysisare shownfor the five lowest(climate) bandsof the computation(2, 2.5, 3.3, 5 and 10 years). All ratios are COADS/NCEP, and a negativelag impliesthat NCEP leadsCOADS. Valuesin parenthesesare significantat the 90% level Amplitude 2years 2.5years 1.30 1.00 3.3years 1.12 Lag (months) 5years 10years 1.10 0.74 2years 2.5years -1.62 3.3years -0.93 -1.06 5years 10years -0.59 4.71 1.92) (0.77) (0.70) (0.97) (0.66) (2.05) (0.64) (0.75) (0.82) (0.81) (1.94) (0.77) (0.78) (0.86) (0.94) (2.00) (0.65) (0.58) (0.83) (0.72) (1.70) 0.40 0.48 (0.90) (0.76) (-0.05) (0.39) (-0.13) (0.01) (-0.25) (0.27) (0.29) (0.09) (-0.23) (0.33) (-0.37) (0.04) (-0.27) (-0.34) (0.04) (-2.48) (-1.78) (0.20) (-0.78) (0.01) (-4.00) -1.56 2.56 (3.33) (4.25) 0.73 (0.53) (0.73) 0.63 0.96 -1.63 (1.63) (3.43) 5.84 4.97 in which they are presently used by the community to force ocean models. With these similarities and differ- there is a clear disagreementin amplitude in spite of significantcoherency.In the tropics, NCEP zonal wind encesin mind, we next analyze their statistics and cross stresses are much weakerthan COADS/FSU [Putman statisticsto provide an a priori, general idea of how the et al., 2000]. This differencedecreases away from the two data sets may differ when used as oceanic forcing tropics, suchas in the western subtropical North Pacific functions. 3.1. area which showsa muchbetter correspondence (Table 1). Wind We next discussthe spatial structure of the zero-lag Stress correlation coefficients between NCEP and Figure 2 showsthe regionsfor which area-averaged COADS/FSU wind stressanomalies. Figure 3 shows These areas were selected that both components of the wind stress field are well becausethey coverkey dynamical regionsof the North correlated north of 20øN, in contrast to low correlaPacific, most of which have been previously studied tions, below 0.3, that occur in the tropical band. In the time series were constructed. [e.g., Stockdaleet al., 1998; Miller et al., 1994; Auad Southern Hemisphere the correlations increase relative et al., 1998a,1998b]). to the tropics, but they reach 0.6 only in a limited area Table I shows some basic statistics between NCEP and COADS/FSU zonal wind stressin thesesevenkey regionsof the equatorial and North Pacific. The meridional stressstatisticsshowa similar level of agreement, although the phase discrepancyis much worse in the tropics. The highest correlationsand the best agreementsin coherenceamplitude and phase occur in the midlati- off the Australian coast. The area displayingthe lowestcorrelationsis the eastern tropical Pacific, where wind stress variables are poorly sampled in space and time. Caution should therefore be used when interpreting these maps because a low correlation does not necessarily mean that the NCEP model is not providing a good representation of the wind field there or that the bulk formulae used to tudes. In the easternequatorialPacific (EEP) region obtain the COADS/FSU data setsare unrealistic.Low the correlation and coherencyare weak and insignifi- correlations could simply mean inadequate sampling in cant. In the westernequatorialPacific (WEP) region, COADS/FSU. 22,254 AUAD ET AL.' MODEL RESPONSE TO COADS AND NCEP FORCING Correlation Coefficients: COADS as the central North Pacific, lags are smaller than 4-1 vs. NCEP month for the ENSO band (a 4% error) and smaller than 4-3 monthsfor the decadalband (a 3% error). It is importantto notethat the squaredcoherences in Figure 5 have a spatial structuresimilar to the correlation maps of Figure 3. Over the basinthe most coherent 6O Heat Flux 4•øN 20øN I band is the ENSO band which has a typical squaredco- herencethat greatlyexceeds the 90%confidence levelof 40 øS ' • ' •"' • :•"'•' '•" •' "•' t 0.35 in the tropicsfrom 120øWwestward,in the western North Pacific north of 20øN, and in the western South Pacific middle latitudes. The eastern tropical Pacific remainsproblematicas insignificantcoherences are ob- "'"'•" • 150øE 160øW 110øW tained for all interannual frequencies. o6ø 40 N •• (b) Similarresults(not shown)areobtainedfor the merid- ional wind stress over the central North Pacific. Un- TAU-xo like the zonal stress coherences,maximum coherences in midlatitudes,for all frequencies,are locatedfurther eastby 200-30ø, whereanomalous windsassociated with 20øS • the Aleutian Low are oriented more meridionally. This alsois a featureof the tropicswherethe easterntropical Pacificshowssomesignificantcoherences for the merid- 40øS 150øE 160øW IlOøW 60 øN STANDARD DEVIATiON RATIOS: COADS vs. NCEP 40W TAU-y 20øN 60øNJ• 40øN'" .... '""• o HeatFluxø 20N 20øS 40øS 150øE 160øW 110øW Figure3. Correlation coefficients between theheat 20 S'. 40os 150øE 160øW 110øW fluxesandwindstresses ofthe COADS/FSUandNCEP data sets. The spatiallyaveraged confidence levels, 60øN : .... based ontheDavis[1976]autocorrelation timescale, are (topto bottom) 0.35,0.33,and0.36.Thefields were smoothedwith a 1000 km radius filter. 40 øN TAU-x Figure 4 showsthat the ratio of the standard deviations of the wind stress components is nearly unity outside the subtropical latitudes. Along the equator, COADS/FSU stressesare larger than NCEP stresses (b) 20øS • 40øS' 150øE 160øW 110øW by a factor of up to 2.5 and 3.5 for the zonal and meridional components, respectively. The largest discrepancies again occur in the eastern tropics, which is an area with low densityof observationsin COADS/FSU and with little data assimilated into the NCEP model. In order to discriminate among different frequency ue•ween the NCEP bands w• •u,•pu•u •u• •1•ucua .... •' ' and COADS/FSU flux time seriesover the entire Pacific basin. We focuson E1 Nifio/SouthernOscillation (ENSO) timescales(2-8 years)and decadaltimescales (>8 years)asthey are the target of manyclimatestudies. Figure 5 showsthe transfer function amplitudesand phases and the squared coherency for the zonal wind (a) TAU-y o '" (c) 2o 20 S 40øS 150øE 160øW 110øW Figure 4. Ratios betweenthe standarddeviationsof the heat fluxesand wind stressesof the COADS/FSU stressbetweenthe COADS/FSU and NCEP fieldsfor (numerator)and NCEP (denominator) data sets.The these two period bands. For the coherent areas, such fields were smoothed with a 1000 km radius filter. AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP 150 øE 160 øWl l OøW (b) FORCING 22,255 150 øE 160 øWl l OøW (c) 150 øE 160 øWl l OøW LAG (months) (e) 150 øE 160 øWl l OøW COHERENCE**2 Figure 5. Crossspectralanalysisbetweenthe COADS/FSU andNCEP zonalwindstresses.The three columnscorrespondto amplitude, phase and squaredcoherenceof the transfer function. The labelson the left denotethe frequency bandin question,decadaland subdecadal (>8 year) and ENSO (2-8 years). The 40 year long time serieswas dividedinto four segments which overlapped50%. The 90% confidence levelis 0.35. The contourintervalfor squaredcoherences is 0.1 starting at 0.15 with contours<0.15 not plotted; the 0.35, 0.25 and 0.15 are includedsince most of them are parallel to the significantcontours> 0.35. The contour intervals for transfer function amplitudesare 1.5, 1.3, 1.1, 0.9, 0.7, 0.5, etc. The contour interval for LAGS are 2 months for the ENSO band and 3 months for the decadal and subdecadal band. The transfer function amplitude is defined as NCEP over COADS. ional component,unlike the zonal wind stresscompo- in the Kuroshio/OyashioExtensionarea, a regionwhich nent. is crucial to decadal midlatitude ocean-atmosphere in- For interseasonal frequencies(3, 6, and 9 month pe- teractions[Nakamuraet al., 1997; Miller et al., 1998; riods) both the zonal and meridionalwind stressesare Deser et al. 1999; Qiu, 2000; Miller and Schneider, significantlycoherent(not shown)only north of 20øN 2000]. and show patterns similar to Figure 5. The transfer Figure 3 (top) showsthe correlation map of heat function amplitude indicates that NCEP wind stresses fluxes from NCEP and COADS. As with the wind are roughly10-30%smallerthan COADS/FSU in the stress correlations, minimum values are observed in interseasonalband in that region. the tropical band and increasetoward higher latitudes. The Northern Hemisphere has larger correlations than 3.2. Heat Fluxes the Southern Hemisphere while the maximum Southern Hemisphere correlations are seen off the Australian Table 2 shows some basic statistics between NCEP and COADS surfaceheatflux in sevenkeyregionsof the coast,mostlikely becauseof a higher sampling rate in this region. equatorial and North Pacific. Midlatitude areas exhibit Figure 4 (top) showsthe standarddeviationratios of a good correspondence in both amplitude and phase. As with the wind stressdata, the major discrepancies heat fluxesfor COADS and NCEP. Unlike wind stresses, are in the eastern and western equatorial Pacific areas the ratios for heat fluxes in the tropical band are much (EEP and WEP areas). The NCEP heat flux in the closerto unity. In the midlatitude Northern Hemisphere EEP area (not shown)has an interdecadaloscillation the ratios which seems to be absent from the COADS ence(not shown)in the tropicsfor Qc -Q•v is 0(5 W record. It is interesting to note that in both data sets the maxi- are between 0.9 and 1.0. The RMS differ- m-2), whilein the subtropical NorthernHemisphere it mum variabilityof heat fluxes(not shown)is observed is 0(-5 to-10 W m-2). In the subtropicalSouthern 22,256 AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING Table 2. Correlations, Standard Deviations ratios, Coherence,Amplitude of Transfer Function and Lag for COADS and NCEP surfaceheat fluxes Coherence Area EEP Corr SD (0.33) (0.96) WEP 0.12 CCS CNP N 20øN K.Sxt. WNP 2years 0.90 (0.86) (0.81) (0.66) (0.79) (0.72) (1.10) (1.26) (1.16) (1.39) (1.18) 2.5years 3.3years 5years 10years 0.18 0.37 0.53 0.52 0.45 0.04 0.01 0.12 0.49 0.40 (0.65) (0.64) 0.52 0.57 0.33 (0.69) (0.77) 0.46 (0.72) (0.39) 0.40 (0.73) 0.08 0.32 0.52 0.02 0.31 0.03 0.12 (0.75) 0.28 0.32 0.10 0.16 (0.77) CORR and SD were computed using anomalous time series with no filtering. Results from the cross-spectralanalysisare shownfor the five lowest (climate) bands of the computation (2, 2.5, 3.3, 5 and 10 years). All ratios are COADS/NCEP, and a negativelag impliesthat NCEP leadsCOADS. Valuesin parenthesesare significantat the 90% level Amplitude 2years 0.34 (0.12) (0.64) (0.81) 0.56 0.57 0.53 2.5years 0.54 0.03 (0.75) (0.83) 0.41 (0.51) 0.61 3.3years 0.79 0.26 0.68 (0.67) 0.25 0.30 0.97 Lag (months) 5years 0.83 (0.32) 0.08 0.31 0.13 0.22 (1.06) 10years 2years 0.79 -0.49 (0.25) 0.27 0.60 2.22 (-1.36) (0.05) 0.19 -3.0 0.30 (0.95) -1.41 -0.63 Hemisphere RMS differences are smaller than in the Northern Hemisphereand tend to zero on average in the band 20øS-40øS. The cross-spectral analysis between N CEP and COADS heat fluxesis shownin Figure 6 for the ENSO (2-8 years)and decadaland subdecadal(>8 years)frequency bands. For regions where significant squared coherencesare seen in the right the ratio of COADS over NCEP is less than unity for all frequencies. As with the wind stresscomparison,errorsin phase(Figure 6, middle) are small (no morethan 4% to 5%). The sparsely sampled tropics and Southern Hemisphere do not show significant coherences.The most coherentareas are the western subtropical North Pacific and some highlatitude areas (Table 2). Similar to the analysisof 2.5years 3.3years -0.52 2.31 11.2 -1.81 (-0.75) -3.34 (-1.23) -1.64 -8.5 0.46 (-1.80) 5years 3.48 (-16.7) 5.16 (-4.52) 10years -1.59 (-43.2) -9.98 10.06 -0.23 -1.14 -41.0 -0.67 0.31 3.88 (0.81) -2.17 (0.69) parable to those of the shortwave fluxes so the shortwave flux errors can degradethe quality of oceanmodel summertime SST hindcasts. In the tropics, shortwave fluxes and the combined effect of sensible and latent heat fluxes are comparable year-round, but they normally act as a damping effect on SST anomaliesin that region. 4. Ocean Model Response to COADS/FSU 4.1. OPYC and NCEP forcing Model The primitive equation ocean model called the Ocean isoPYCnal (OPYC) modelwasdevelopedby Oberhuber wind stresses, the surface heat fluxes in the intersea- [1993]and appliedby Miller et al. [1994,1997, 1998], sonalbands (3 to 9 months)exhibit significantcoher- Cayanet al. [1995],and Auad et al. [1998a,1998b]to ences(not shown)only north of 20øN,with COADS and study monthly through decadal-scaleocean variations N CEP heat fluxeshaving nearly the sameamplitudes. over the Pacific Basin. Here an updated version of the It is important to note that the shortwave fluxes from model was used with higher resolution and a revised the N CEP reanalysis are known to be inaccurate and schemefor forcing with monthly mean fluxes. might contribute to errors in SST anomalies,especially The model is constructed with 10 isopycnal layers duringsummer[Scottand Alexander,1999].In winter, (eachwith nearly constantpotential densitybut varihowever, the combined effect of the latent and sensi- able thickness,temperature, and salinity) that are fully ble heat fluxes is larger than the shortwaveby I order coupledto a bulk surfacemixed-layer model. The grid in midlatitudes[Cayan, 19923]. In summerthe mid- extends from 119øE to 70øW and from 67.5øS to 66øN, latitude latent and sensible fluxes together are corn- with periodic boundary conditions along the latitudes AUAD ET AL.: MODEL RESPONSE of the Antarctic Circumpolar Current. The resolution is 1.5ø in the midlatitude open ocean, with zonal resolution gradually increasedto 0.65ø resolution within a 10ø band around the equator. Although the model is not eddy resolving, equatorial instability waves occur spontaneously,and eddies occur in the west wind drift of the midlatitudes. We only seekto study largescalepatterns in the responseand regard this intrinsic variability as noise. The semi-implicit time step is 0.75 days. Two versions of the surface forcing are employed: (1) monthly mean seasonalcycle fields plus monthly mean COADS/FSU anomaliesand (2) monthly mean TO COADS AND NCEP FORCING 22,257 surface heat flux at each time step with bulk formulae that useevolvingmodel SST with ECMWF-derived atmosphericfields(air temperature,humidity,cloudiness, etc.); the daily mean seasonalcycleof heat flux is then saved(averagedoverthe last 10 yearsof a 99-yearspinup) and subsequentlyused as specifiedforcingduring the anomalously forced hindcasts. Anomalousfields of monthly mean wind stress,total surface heat flux, and TKE are then added to the respective monthly mean seasonal cycles. Because there is no SST feedback to any of the anomalous forcing fields, the model is not constrained to reproduce the observedtemperature variations (as could be achieved by adaptinga Haney[1971]or Barnier et al. [1995]formulation). Becauseof the lack of data in the tropicswe usethe COADS/FSU blendwind stresses(and TKE eset al. [1994]. The monthlymean wind stressclima- timates) as describedin section2. A similar strategy is tology is derived from a combination of monthly mean employedby Putman et al. [2000]to improvethe tropEuropean Centre for medium-RangeWeather Forecasts ical wind stressesof the NCEP fields, but here we use in their originalform. Specifiedsur(ECMWF) midlatitude fields and monthly mean the NCEP stresses Hellerman-Rosensteintropical climatology. The monthly face freshwater flux anomalies are excluded from both mean seasonalcycle climatology of turbulent kinetic hindcasts becauseof inadequate rainfall observations in energy (TKE) input to the mixed layer is estimated COADS and unreliable precipitation estimates from the from the samedata sets [Oberhuber, 1993]. The sur- NCEP reanalysis[e.g.,Janowiaket al., 1998]. The conface freshwaterflux is representedas a combination of tinuous relaxation to annual mean Levitus salinity, howobservedmonthlymeanrainfall [Legatesand Willmort, ever, implies that freshwater flux anomaliesdo occur in 1992],evaporationcomputedby bulk formula,plusa re- the hindcasts; however they have a weak influence on laxation to the annual mean Levitus salinity field over SST and mixed layer depth anomalies. Near the equator the anomalousheat fluxes are poorly 30-day timescales. The monthly mean seasonalcycle climatologyof total surfaceheat flux is computeddur- known because of the many gaps in ship weather reing spin-up(with no anomalousforcing)by determining ports. Moreover, tropical heat fluxes generally serveas seasonalcycle fields plus monthly mean NCEP anomalies. The monthly mean seasonalcycle forcing is derived from various sourcesand is the same as used by Miller 150øE 160øWllOøW (a) 150øE 160øWllOøW 150øE 160øWllOøW (b) 150øE 160øWllOøW T.F. AMPLITUDE (d) 150øE 160øWllOøW LAG (months) (e) (c) 150øE 160øWllOøW COHERENCE**2 (f) Figure 6. As in Figure 5 but for surfaceheat fluxes. 22,258 AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING a dampingmechanism[Liu and Gautier, 1990; Cayan, by computingthe oceanmodel climatologiesoverthree segmentsof the 1951-1997run from 1992b; Barnett et al., 1991], although Seaget[1989] separatesuccessive shows these fluxes can sometimes excite SST anomalies which anomalies are then computed. _ in certain locations. Therefore unless the wind-driven model SST reproducedthe observedSST with nearly perfectfidelity, heat flux anomaliesspecifiedfrom observations would try to damp an SST anomaly that "was not there". This nonphysicalforcing can then lead to an excitation of model SST anomalies with opposite sign as the observed. Hence a physically motivated NewtonJan dampingis employedwithin a 6ø e-foldingscalearound the equator. The SST anomaliesthere are dampedback to the modelSST climatology(averagedfrom the last 10 yearsof the spin-uprun) overtimescalesthat dependon a damping coefficientdivided by the mixed layer depth 4.2. Modeled-Observed SST Comparisons We next examine the performance of the two mod- els runs, forcedwith NCEP and COADS/FSU anomalousforcing,in simulatingSST observedin the COADS. The simple statistical analyses employed here include monthly mean anomalies for all months of the year and give a broad idea of the year-roundmodel performance. A more detailed look at the model performance as a function of season,however, would show that winter month SST anomalies are much better simulated than summermonths(asfoundby Miller el al. [1994]), (followingMiller et al. [1994];seeBarnett et al. [1991] mainly becauseof the strong forcing signalsin winter for a map of the coefficient). The monthlyforcingstrategyof Auad et al. [19984, and the high sensitivityof thin summermixed layersto errors in the specifiedforcing. 1998b]is used to properly weight the monthly mean Figure 7 showsthe correlation coefficientsbetween observed and modeled monthly mean SST anomalies forcing,followingKillworth [1996]. Miller et al. [1994] for both runs. The patterns are fairly similar in that employ simple linear interpolation of the monthly mean forcing, which yields monthly mean model forcing that is typically weaker than the true monthly mean obforcing anomalies,which are added to the seasonalcycle servedforcing.The Killworth[1996]scheme is a method for increasing the input monthly mean forcing fields Correlation to allow linear interpolation in time to yield correct monthly mean forcing. The ocean model is run for two cases. The N CEP run (forcedby anomalousNCEP fluxes)extendsfrom January 1958 to December 1997. The COADS/FSU run (forcedby anomalousCOADS fluxessupplemented with FSU and reconstructedtropical stresses)extends from January 1951 to December 1997. Initial conditions for those runs are from the ninetieth year of the model spin-up with climatologicalforcing. Becausethere is no model SST feedbackto the surfaceheat fluxes(except in a narrow equatorial band) there is a possibilityof Coefficients for SST (a) COADS OBS. vs. COADS RUN 2;oN :0o60 20•0 120øE a drift in the model climatology becausethe model has not reacheda complete equilibrium with mean flux forc- 160øE 160øW 120øW 80øW • (b) COADS OBS. vs. NCEP RUN ing after 90-100yearsof spin-up. The NCEP run (and a separaterun with zero forcinganomalies)indeedex- o60øN• hibits a small and apparently inconsequentialdrift due to this effect. The COADS/FSU run, however,exhibits a larger drift in the model SST climatology. Its origin appears to be mainly associated with COADS heat flux anomalies which have interdecadal variations that are not present in the NCEP heat fluxes. These interdecadal COADS heat flux anomalies are likely erroneouslylarge becausethe "reduced thermal damping" effectsof air-sea coupling should effectively shutoffheatfluxesat very low-frequencies [e.g.,Battisti et al., 1995;Barsugliand Battisti, 1998]. Attempts to o 20øS 40øS• 60ow 20ow soow remove these spurious trends and decadal variations a Figure 7. Correlation coefficientsbetween obpriori by trend removaland high-passfiltering[Kaylot, servedSST (COADS) and modelSST from the (top) 1977] the COADS heat flux forcingfieldsfail to pre- COADS/FSU-forced runandfromthe (bottom)NCEP- serve the true low frequency variability in the forcing forced run. The fields were smoothed with a 1000 km fields. So their effect is instead removed a posterJori radius filter. AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING STANDARD DœVIA TION RATIO$ (a) COAD$ S$T OB$, vs. MODEL RUN (COAD$) 60 øN 40ø17 22,259 than those of Miller et al. [1994] and Cayan et al. [1995]becauseof severalfactors. This time intervalis longerthan the 1970-1988periodmodeledin theseprevious studiesand henceincludestime periodswith less confidentlyobservedforcing and responsefields. For example,the subset1970-1988winter midlatitudeSST, from the previousmodel simulations,matchesthe cor- 20 øN relationsdiscussed by Miller et al. [1994]muchmore o closely. However,the 1970-1988year-roundSST correlationsare smallerthan thoseof Cayanet al. [1995] in the midlatitudes,indicating the summermonth periods are modeled here with less fidelity. This seems to be causedby the present model's previouslymen- 20øS tioned tendency to produce eddies due to instabilities 40øS 120øE 160øE 160øW 120øW 80øW (b) COAD$ $$T OB$. vs. MODEL RUN (NCEP) 60 øN 40W 20øN in the west xvind drift. An ensemble of hindcast runs could drive down this eddy noise to better establish the forcedpart of the response,but that is beyondour computational means at this time. The ratios of standard deviations(observed/model) for SST anomalies are shown in Figure 8. The COADS/FSU run tendsto underestimateSST anomaly variance in the tropical Pacific and in the subpolar region of the North Pacific. The NCEP run overestimates SST anomalyvariancethroughoutthe Pacific exceptin a small region in the central equatorial Pacific. Both o runs seriously overestimate SST anomaly variance in the subtropicsof both hemispheres. 20øS As was mentionedpreviously,errorsin both the forcing and validating data may be large in these regions 40 øS due to limited observations, and errors in the model 120øE 160øE 160øW 120øW 80øW may be large due to the large mean model mixed layer depth in this region. Figure 8. Ratios of the standard deviations between observations of SST anomalies from COADS (m•mera- A feature not evident in Figures 7 and 8 is that, on tor) andthe modelSST anomalies (denominator) ob- interannual timescales,the NCEP run producesE1 Nifio tained from the (top) COADS/FSU-forcedrun and and La Nifia eventsthat are roughly half the observed from the (botttom) NCEP-forced r•m. The fields were amplitude. This occurs in spite of the NCEP fluxes smoothed with a 1000 km radius filter. containing stronger than observed interseasonal variations. The weak interannual they show the best correlations in the central and eastern tropical Pacific and the central and eastern midlat- wind stresses of the NCEP analysismotivated Putman et al. [2000]to blend the FSU tropical stresseswith the NCEP stressesto provide a more useful forcing dataset for future studies. COADS likewise suffers from weak interannual wind itude North Pacificnorth of 30øN. The South Pacific, stress representation, hence motivating the use of a wherethere are not as many observations of SST, wind blended COADS/FSU wind-stressforcing dataset by stresses,and surfaceheat fluxes,exhibitsmuch poorer Miller et al. [1994]and in this study. correlations. The poor correlationsseen in the western North Pacific, centered at 20øN between 160øE and 180ø, also may be due to the lack of observationsin 4.3. Modeled-Observed Heat Storage Comparisons that area. However,the modelbulk mixedlayeris very deepin that region(exceeding 150 m in winter) which We next examine the performance of the two modwasfound,in separatemodelsensitivityexperiments,to elsruns,forcedwith NCEP and COADS/FSU anomabe due to the combinedeffectsof Ekmandownwelling lousforcing,in simulatingheat storage,here definedas andsurfacecoolingduringthe entraining seasons [De ocean temperature integrated over the upper 400 m of Szoeke,1980]. The excessively deepmixedlayermay the watercolumn(scaledby densityandheatcapacity). distributethe surfaceforcingtoo deepinto the water Observedheat storageis takenfrom the White [1995] column yielding poorer results. The correlations between model and observed SST analysis of available subsurfacetemperature observa- tions since1955. Chepv. rin and Carton [1999]discuss (Figure 7) for the COADS/FSU-forcedrun are loxver potential deficitsof this analysis. The simple statisti- 22,260 AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING CORRELATiON COEFFICIENTS FOR HEAT STORAGES JEDAC Obs. vs COADS run 60W 40 øN 20W o 20øS 40øS 120øE 160•E JEDAC 160øW 120øW 80øW Obs. vs NCEP run 60W 20W o 20øS 40øS 120øE 160øE 160øW 120øW 80øW Figure 9. As in Figure 7 but for heat storages. cal analyseshere include monthly mean anomaliesfor much greater area than found by A uad et al. all monthsof the year to give a synopsisof the year- during the 1970-1988 time interval for heat storageon round model performance.Heat storageincludescom- interannual and decadal timescales. Thus, the interponentsof responselocally forcedby diabatic processes seasonalvariations of heat storage, which are inade(evidenced mainlyasSSTanomalies) andadiabaticpro- quately simulated, appear to be dominating the corcesses (seenasthermocline heave)asdiscussed by Auad relation maps in Figure 9. Figure 10 displays the ratios of standard deviations et al. [1998a,1998b].Errorsin the estimationof wind stresscurl forcing, which largely controlsthe thermo- of heat storage(observations/model). Alongthe tropicline heave,may be large and may stronglyinfluence cal strip, where significantcorrelationsoccurredin both the model responseduring periodsof limited observa- runs, heat storage variance is nearly the same as obtions. servedin the western tropical Pacific but greater than The correlations(zero time lag) betweenobserved observedin the eastern tropical Pacific. In the subpoand modeledheat storageare shownin Figure9. Max- lar region, which showedsignificantcorrelationsin the imum zero-lag correlations are obtained in the warm NCEP run, the model seriouslyoverestimatesthe heat pool of the western tropical Pacific and in the eastern storage variability. tropical Pacific. The NCEP run, unlike the The model heat storage variations in the western COADS/FSU run, exhibitssignificantcorrelationsfor tropical Pacific warm pool area are better correlated heat storagein the subpolararea, north of 40øN. The to observations than are SST variations. Since this remidlatitude regionsof insignificantcorrelationscovera gion is dominated by local, rather than remote wind AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING 22,261 forcing[Schneider et al. 1999],a reasonably accurate the time period 1970-1988, which has the highest denrepresentationof the wind stresscurl is responsiblefor the significant model-datacorrelations.However,wind stresscurl from COADS/FSU and NCEP show surprisinglylow correlationsin the warm pool area even sity of subsurfacedata. Figure 9 showslow model-data heat storagecorrelation in midlatitudes occurswhen including all climatic timescalesfrom 1958-1997. The reasonsfor this include the intrinsic variability of the ocean thoughsimulatedheat storages from both runsagree model, sparse subsurfaceoceanic data before 1970, inreasonablywell with observations.It appearsthat a accurate wind stresscurl forcing, and the inclusionof fewenergetic frequencies of the windstress(associatedinterseasonaltimescales. Indeed, Figure 11 showsthat with E1Nifio) in the westerntropicaland subtropical the correlation between COADS and NCEP wind stress Pacificareadrivethemodeled fieldsthat areresponsible curl in the North Pacific is significant but does not exceed 0.7. for this agreement with observed heat storage. In the centralNorthPacific,modelheatstoragevariations are poorly correlatedwith observations,while modelSST anomaliesare significantlycorrelatedwith observations.In this region, COADS and NCEP sur- 5. Summary and Discussion A comparisonof COADS (blendedwith tropical FSU facefluxesare alsowell correlatedand directlyforce wind stresses)and NCEP surfaceflux anomaliesover modelSSTanomalies throughvarious processes [Miller the Pacific Ocean on climatic timescales from 1958et al., 1994]. It is surprisingto find that modelheat 1997 was executed. The two data sets were then tested storageresponseto thesetwo correlatedforcingsdoes as forcing functions in separate simulations of Pacific not matchobservations (exceptin the subpolargyrefor Ocean variability using an ocean model. Wind stress anomalies from the two data sets are the NCEPrun). Auadet al. [1998a,1998b] showed with an earlier version of this model that model thermocline well correlated in the midlatitude extratropics, more dynamicsare adequateto simulateheat storagefluc- so in the highly sampled North Pacific than in the tuations in midlatitudes that are significantly coherent South Pacific. In the tropics and subtropics, low cor- with observations if one focuses on the interannual and relations interdecadal timescales. But that study was limited to sets;somewhat higher correlationsoccurredin the western part of these regions because of denser sampling than in the eastern part. The amplitudes of the stress STANDARD DEVIATION RATIO$ FOR HœA T STORAGES JEDAC Obs./COAD$ eun variations were found between of the two data the two wind stress data sets are similar in midlati- tudes. In the tropics, NCEP wind stressesare weaker than the COADS/FSU stresses,especiallyon interan- øN nual timescales, which is problematic for E1 Nifio simulations. 40 øN Surface heat flux anomalies from the two data sets are well correlated 20W on interannual and shorter timescales in the North Pacific Ocean north of 20øN, but they are poorly correlated elsewhere and on decadal timescales. In the extratropics the amplitudes of the heat flux variations of the two data sets are comparable, but in the tropics the NCEP heat fluxes are weaker than COADS. o 20øS 40øS Errors in the estimationof the COADS/FSU fluxes 120øE 160øE 160øW120øW JEDAC Obs./NCEP 80øW run 60 due to limited data sampling cannot be ruled out as the primary explanation for the low correlations found in this study. Indeed, areas with a high density of observations usually exhibit significant correlation between COADS/FSU and NCEP fluxes. This strongly suggests that data density controls the comparability and usefulnessof the forcing fields. This idea is supported 40W by a frequencydomainanalysis(includingperiodsfrom 3 monthsto 10 years) that showedcoherencyis independent of frequencyin the geographicalareas where 20W o COADS/FSUandNCEP surfacefluxesaresignificantly 20 øS correlated. Ocean simulations were then used to test the sensitiv40øS 120øE 160øE 160øW 120øW 80øW Figure 10. As in Figure8 but for heat storages. ity of the oceanicresponse to the two differentforcing data sets. One simulation used the surface flux anoma- lies specifiedfrom NCEP (1958-1997),and the other usedthoseof COADS,/FSU (1951-1997). 22,262 AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING Wind Stress Curl COADS/NCEP (a) StandardDeviationRatios 60 øN 40 øN 20W o 20øS 40øS 120øE 160øE 160øW 120øW 80øW (b) CorrelationCoefficients 60 øN 40W 20øN o 20øS 40øS 120øE 160øE 160øW 120øW 80øW Figure 11. Wind stresscurlcomparison. (top) standard deviation ratiosof COADS/FSUto NCEP windstresscurlsand (bottom)correlation coefficients. The fieldsweresmoothed with 1000 km radius filter. The midlatitude SST hindcasts were superior when tions of modeled and observed tropical heat storage usingthe N CEP fluxes,suggesting that the N CEP data variationsincludingall climatictimescaleswerefound are eminentlysuitablefor oceanhindcastprocessstud- for both hindcasts.However,only the NCEP simulation correlations with observed midlaties in those regions. Tropical SST anomalieshad sim- exhibitedsignificant ilar skill levels for the two hindcasts when consideritudeheat storage,andthat waslimited to a portionof ing all climatic timescales. However, on interannual the North Pacificsubpolargyre and easternboundary. timescalesthe NCEP tropical wind stressesare weaker SinceCOADS and NCEP fluxeswerecorrelated(r=-0.7) in the two forcing than COADS/FSU stresses,consequently yieldingin- in this sameregion,minordifferences terannual tropical SST anomaliesthat are muchweaker functionssets can lead to substantially different model The maindiscrepancy bethan observed.Adjusting NCEP tropical wind stresses oceanheatstorageresponses. to alleviate this problem is presentlythe most feasible tweenmodeledand observedheat storagecan be traced strategy,as Putnam et al. [2000]showedusingFSU to inadequatecorrelation(r=0.5) betweenNCEP and stresses. COADS wind stress curl anomalies which dominate the variationsin the subpolarNorth Heat storageanomaliesin the two simulationswere drivingof thermocline Miller et al., 1998; also comparedwith observations. Significant correla- Pacific[Auadet al., 1998a,1998b; AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING 22,263 Deseret al., 1999].Apparently,the consistentdynam- tional ScienceFoundation under grants OCE97-11265 and ics of the NCEP reanalysismodelefficientlyorganizes OCE00-82543, and by the National Aeronautics and Space the input observationsto yield a better estimateof wind stress curl than is obtainable from COADS observations alone. Problemsarise in explainingdecadal-scaleCOADS heat flux variationsand in removingtheir potentialeffects on oceanmodel response.Some of the fundamen- Administration under grant NAG5-8292. The views expressedherein are those of the authors and do not necessarily reflect the views of NOAA or any of its sub-agencies. We are grateful to Warren White for providing us with the JEDAC dataset of subsurfacetemperature observationsthat made the modeled-observedheat storage comparisonspossible. We thank Emelia Bainto for her valuable techni- cal support. Josef Oberhuber generously provided modeltal COADS marineweatherobservations (wind speed, ing advice and accessto his latest codes, for which we are SST, etc.) that are usedto estimate the surfacefluxes grateful. Many important suggestionsthat greatly improved are knownto containsuspicious low-frequencyvariabil- the manuscript were provided in detailed reviews by Mike ity due to changesin observationalprocedures,ship Alexander, an anonymous referee, and Mike Spall. characteristics, and instrumentation[Michaudand Lin, 1992; The main discrepancybetween modeled and observed heat storage can be traced to inadequate corre- References Auad, G., A. J. Miller, and W. B. White, Simulation of lation (r=0.5) betweenNCEP and COADS wind stress heat storages and associatedheat budgets in the Pacific Ocean, 1, E1 Nifio-Southern Oscillation timescale, J. Geocurl anomalieswhich dominate the driving of thermophys. Res., 103, 27,603-27,620, 1998a. clinevariationsin the subpolarNorth Pacific [Auad et Auad, G., A. J. Miller, and W. B. White, Simulation of al., 1998a,b;Miller et al., 1998;Deser et al., 1999]. Apheat storages and associatedheat budgets in the Pacific parently the consistent dynamics of the NCEP reanalOcean, 2, Interdecadal timescale, J. Geophys. Res., 103, 27,621-27,635, 1998b. ysis model efficiently organizesthe input observations to yield a better estimate of wind stress curl than is Barnett, T. P., Long-term trends in surface temperature over the oceans, Mon. Weather Rev., 112, 303-312, 1984. Barnett, T. P., M. Latif, E. Kirk, and E. Roeckner, On Problems arise in explaining decadal-scaleCOADS ENSO physics, J. Clim., 4, 487-515, 1991. heat flux variations and in removing their potential Barnier, B., L. Siefridt, and P. Marchesiello, Thermal forceffects on ocean model response. Some of the funing for a global ocean circulation model using a 3-year climatology of ECMWF analysis, J. Mar. Syst., 6, 363damental COADS marine weather observations(wind 380, 1995. speed,SST, etc.) that are usedto estimatethe surface obtainable from COADS observations alone. fluxes are known to contain suspiciouslow frequency variability due to changesin observationalprocedures, Barsugli, J. J., and D. S. Battisti, The basic effects of atmosphere-oceanthermal coupling on midlatitude variability, J. Atmos. Sci., 55, 477-493, 1998. shipcharacteristics, instrumentation[MichaudandLin, Battisti, D. S., U.S. Bhatt, and M. A. Alexander, A modeling study of the interannual variability in the wintertime 1992; Ward and Hoskins,1996]. The NCEP decadal- North Atlantic Ocean, J. Clim., 8, 3067-3083, 1995. Bennett, A. F., B. S Chua, D. E. Harrison, and M. J. McPhaden, Generalized inversion of tropical Atmospherefrom coupledocean-atmosphere modelstudies[Barsugli Ocean data and a coupled model of the tropical Pacific, J. Clim., 11, 1768-1792, 1998. and Battisti, 1998;Bhatt et al., 1998]. The results of this study suggest that the NCEP Bennett, A. F., B. S Chua, D. E. Harrison, and M. J. McPhaden, Generalized inversion of Tropical Atmospheresurfacefluxes overall yielded a better basinwide ocean scaleheat fluxes had much smaller amplitude than those of COADS which is consistentwith what is anticipated hindcastthan the COADS/FSU fluxes,exceptfor the Ocean (TAO) data and a coupled model of the tropical Pacific, part II,: The 1995-96 La Nifia and 1997-98 E1 Nifio, J. Clim., 13, 2770-2785, 2000. Betts, A. K., S. Y. Hong, and H. L. Pan, Comparison of NCEP-NCAR reanalysis with 1987 FIFE data, Mon. Weather Rev., 124, 1480-1498, 1996. seriousproblems with weak interannual variability in the tropics. Denser and more frequent observationsof surfacevariablesfor deriving air-sea fluxes are therefore needed in many key areas of the Pacific Ocean, Bhatt, U.S., M. A. Alexander, D. S. Battisti, D. D. Houghton, especially the eastern tropical Pacific and the western and L. M. Keller, Atmosphere-ocean interaction in the subtropical Pacific. These data can then be used in atNorth Atlantic: Near-surface climate variability, J. Clim., mosphericanalysesto derive consistentlong-term forc11, 1615-1632, 1998. ing functionsfor testing ocean models and diagnosing Blanc, T. V., Variation of bulk-derivedsurfaceflux, stability, andM roughnessresultsdue to the useof differenttransfer climate variations. Sustainedbasin-scalelong-term obcoefficient schemes,M J. Phys. Oceanogr. 15, 650-669, serving programs using new techniquesof remote sens1985. ing and in situ measurements,such as planned under Bony, S., Y. Sud, K. M. Lau, J. Susskind,and S. Saha, the the Pacific BasinwideExtended Climate Study, are Comparisonand satellite assessment of NASA/DAO and NCEP-NCAR reanalysesovertropical ocean: Atmospheric the best strategy for achievingthis goal. Acknowledgments. This research is funded in part by the National Oceanic AtmosphericAdministration, U.S. Department of Commerce, under grants NA77RJ0453 (Experimental Climate Prediction Center), NA47GP0188 (Consortiumfor Ocean'sRole in Climate), NA66GP0274(PaleoclimateProgram),by the Na- hydrology and radiation, J. Clim., 10, 1441-1462, 1997. Cayan, D. R., Variability of latent and sensibleheat fluxes estimated using bulk formulae, Atmos. Ocean, 30, 1-42, 1992a. Cayan, D. R., Latent and sensible heat flux anomalies over the northern oceans: Driving the sea surfacetemperature, J. Phys. Oceanogr., 22, 859-881, 1992b. 22,264 AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING Cayan, D. R., A. J. Miller, T. P. Barnett, N. E. Graham, J. N. Ritchie, and J. M. Oberhuber, Seasonal-interannual fluctuations in surface temperature over the Pacific: Effects of monthly winds and heat fluxes, in Natural Climate Variability on Decadal-to-Century Time Scales, 133-150 pp., edited by Nat. Acad. Press, Washington, D.C., 133150, 1995. 1996. Chepurin, G. A., and J. A. Carton, Comparison of retrospective analyses of the global ocean heat content, Dyn. Atmos. Oceans, 29,, 119-145, 1999. da Silva, A.M., C. C. Young, and S. Levitus, Atlas of Surface Marine Data, Vol. 3, Anomalies of Heat and Momentum Fluxes, NOAA Atlas Set., NESDIS 7, Natl. Oceanic and Atmos. Admin., Washington, D.C., 413 pp., 1994. Davis, R. E., Predictability of sea surface temperature and sea level pressureanomalies over the North Pacific Ocean, J. Phys. Oceanogr., 6, 249-266, 1976. Deser, C., M. A. Alexander, and M. S. Timlin, Evidence for a wind-driven Killworth, P., Time interpolation of forcing fields in ocean models, J. Phys. Oceanogr., •, 136-143, 1996. Kleeman R., R. A. Colman, N. R. Smith, and S. B. Power, A recent change in the mean state of the Pacific basin climate: Observational evidence and atmospheric and oceanic responses, J. Geophys. Res., 101, 20,483-20,499, intensification of the Kuroshio Current Extension from the 1970s to the 1980s, J. Clim., 12, 16971706, 1999. De Szoeke, R. A., On the effects of horizontal variability of wind stresson the dynamics of the ocean mixed layer, J. Phys. Oceano9r., 10, 1439-1454, 1980. Fevrier, S., C. •¾ankignoul, J. Siryen, M. K. Davey, P. Delecluse, S. Ineson, J. Macias, N. Sennechael, and D. B. Stephenson, A multivariate intercomparison between three oceanic GCMs using observedcurrent and thermocline depth anomaliesin the tropical Pacific during 19851992, J. Mar. Syst., 2•, 249-275, 2000. Frankignoul, C., C. Duchene, and M. A. Cane, A statistical approach to testing equatorial ocean models with observed data, J. Phys. Oceanogr., 19, 1191-1207, 1989. Frankignoul, C., S. Fevrier, N. Sennechael,J. Verbeek, and P. Braconnot, An intercomparison between four tropical ocean models- Thermocline variability, Tellus, •7, 351364, 1995. Giese, B. S., and J. A. Carton, Interannual and decadal variability in the tropical and midlatitude Pacific, J. Clim., 12, 3402-3418, 1999. Goldenberg, S. B., and J. J. O'Brien, Time and space variability of tropical Pacific wind stress, Mon. Wea. Rev., 109, 1190-1207, 1981. Halliwell, G. R., Simulation of North Atlantic decadal and multidecadal winter SST anomalies driven by basin-scale atmospheric circulation anomalies, J. Phys. Oceanogr., 28, 5-21, 1998. Haney, R. L., Surface thermal boundary condition for ocean circulation models, J. Phys. Oceanogr., 1, 241-248, 1971. Harrison, D. E., B. S. Giese, and E. S. Sarachick, Mecha- nismsof SST changein the equatorialwave-guideduring the 1982-83 ENSO, J. Clim., 3, 173-188, 1990. Isemer, H., and L. Hasse, The Bunker Climate Atlas of the North Atlantic Ocean, Vol. 2, Air-sea interactions, Springer Verlag, New York, 1987. Iwasaka, N., and J.M. Wallace, Large-scale air-sea interaction in the Northern Hemisphere from a view point of variations of surface heat flux by SVD analysis, J. Meteorol. Soc. Jpn., 11, 781-794, 1995. Janowiak, J. E., A. Gruber, C. R. Kondragunta, R. E. Livezey, and G. J. Huffman, A comparison of the NCEPNCAR reanalysisprecipitationand the GPCP rain gauge- Legates, D. R., and C. J. Willmort, A comparisonof GCMsimulated and observed mean January and July precipitation, Global Planet. Change, 97, 345-363, 1992. Liu, W. T., and C. Gautier, Thermal forcing on the tropical Pacific from satellite data, J. Geophys. Res., 95, 13,20913,217, 1990. Luksch, U. and H. V. Storch, Modeling the low-frequency sea surface temperature variability in the North Pacific, J. Clim., 5, 893-906, 1992. Lysne, J., P. Chang, and B. Giese, Impact of the extratropical Pacific on equatorial variability, Geophys. Res. Left., 2•, 2589-2592, 1997. Marshall, G. J., and S. A. Harangozo An appraisal of NCEP/NCAR reanalysisMSLP data viability for climate studies in the South Pacific, Geophys. Res. Left., 27, 3057-3060, 2000. Michaud, R., and C. A. Lin, Monthly summaries of merchant ship surface marine observations and implications for climate variability studies, Clim. Dyn., 7, 45-55, 1992. Miller, A. J., and N. Schneider,Interdecadal climate regime dynamics in the North Pacific Ocean: Theories, observations and ecosystemimpacts, Progr. Oceanogr., •7, 355379, 2000. Miller, A. J., T. P. Barnett, and N. E. Graham, A comparison of some tropical ocean models - Hindcast skill and E1 Nifio evolution, J. Phys. Oceanogr., 23, 1567-1591, 1993. Miller, A. J., D. R. Cayan, T. P. Barnett, N. E. Graham, and J. M. Oberhuber, Interdecadal variability of the Pacific Ocean: Model responseto observedheat flux and wind stress anomalies, Clim. Dyn., 9, 287-302, 1994. Miller, A. J., W. B. White, and D. R. Cayan, North Pacific thermocline variations on ENSO time scales,J. Phys. Oceanogr., 27, 2023-2039, 1997. Miller, A. J., D. R. Cayan and, W. B. White, A westwardintensified decadal change in the North Pacific thermocline and gyre-scale circulation, J. Clim., 11, 3112-3127, 1998. Morrissey, M. L., An evaluation of ship data in the equatorial western Pacific, J. Clim., 3, 99-112, 1990. Nakamura, H., G. Lin, and T. Yamagata, Decadal climate variability in the North Pacific during the recent decades, Bull. Am. Meteorol Soc., 78, 2215-2225, 1997. Oberhuber, J. M., Simulation of the Atlantic circulation with a coupled sea ice - Mixed layer isopycnal general circulation model, part I, Model description, J. Phys. Oceangr., 23, 808-829, 1993. Parrish, R. H., F. B. Schwing, and R. Mendelsohn, Midlatitude wind stress: The energy sourceof climatic shifts in the North Pacific Ocean Fish. Oceanogr., 9, 224-238, 2000. Pavia, E., and J. J. O'Brien, Weibull statistics of wind speed over the ocean, J. Clim. Appl. Meteorol., 25, 1324-1332, 1986. Putman, W. M., D. M. Leglet, and J. J. O'Brien, Interannual variability of synthesized FSU and NCEP-NCAR satellite combined dataset with observational error considreanalysis pseudostressproducts over the Pacific Ocean, erations, J. Clim., 11, 2960-2979, 1998. J. Clim., 13, 3003-3016, 2000. Kalnay, E., et al., The NMC/NCAR reanalysisproject, Bull. Qiu, B., Interannual variability of the Kuroshio Extension Am. Meteorol. Soc., 77, 437-471, 1996. system and its impact on the wintertime SST field, J. Kaylot R. E., Filtering and decimationof digital time series, Phys. Oceangr., 30, 1486-1502, 2000. Tech. Rep. BN 850, 42 pp., Inst. for Phys. Sc. and Technol., Univ. of Md., 42 pp., 1977. Ramage, C. S., Can shipboard measurementsreveal secular AUAD ET AL.: MODEL RESPONSE TO COADS AND NCEP FORCING 22,265 changesin air-seaheat flux?, J. Clim. Appl. Meteorol., Waliser, D. E., Z. Shi, J. R. Lanzante, and A. H. Oort, 23, 187-193, 1984. The Hadley circulation: AssessingNCEP/NCAR reanalRamage, C. S., Secularchangesin reported surfacewinds ysis and sparsein situ estimates, Clim. Dyn., 15, 719-735, 1999. over the oceans,J. Clim. Appl. Meteorol, 26, 525-528, 1987. Ward, M. N., and B. J. Hoskins, Near-surface wind over the Schneider,N., S. Venzke, A. J. Miller, D. W. Pierce, T. Global Ocean 1949-1988, J. Clim., 9, 1877-1895, 1996. P. Barnett, C. Deser, and M. Latif, Pacific thermocline Weare, B.C., Uncertainties in estimates of surface heat bridge revisited, Geophys. Res. Left., 26, 1329-1332, fluxes derived from marine reports over tropical and sub1999. tropical oceans, Tellus, dl, 357-370, 1989. Scott, J. D., and M. A. Alexander,Net shortwavefluxesover Weare, B.C., Comparison of NCEP-NCAR cloud radiative the ocean,J. Phys. Oceangr.,2g, 3167-3174,1999. forcing reanalyses with observations, J. Clim., 10, 2200Seager,R., Modeling tropical Pacific sea-surfacetempera2209, 1997. ture- 1970-87, J. Phys. Oceanogr.,19, 419-434, 1989. White, W. B., Design of a global observingsystem for gyreSeaget,R., Y. Kushnir, M. Visbeck, N. Naik, J. Miller, G. scaleupper ocean temperature variability, Prog. Oceanogr. Kralmann, and H. Cullen, Causesof Atlantic Ocean cli36, 169-217, 1995. mate variability between1958 and 1998, J. Clim., 13,, Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P.M. Steurer, 2845-2861, 2000. A comprehensiveocean-atmosphere data set, Bull. Am. Meteorol. Soc., 68, 1239-1250, 1987. Wright, P. B., Problems in the use of ship observations for J. Phys. Oceanogr.,2•, 2288-2305, 1994. the study of interdecadal climate changes,Mon. Weather Stockdale,T. N., A. J. Busalacchi,D. E. Harrison,and R. Rev., 11•, 1028-1034, 1986. Seaget,Oceanmodelingfor ENSO, J. Geophys.Res., 103, Xie, S. P., T. Kunitani, A. Kubokawa, M. Nonaka, and 14,325-14,355, 1998. S. Hosoda, Interdecadal thermocline variability in the Tanimoto,Y., N. Iwasaka,and K. Hanawa, Relationships North Pacific for 1958-1997: A GCM simulation, J. Phys. betweensea surfacetemperature,the atmosphericcircuOceanogr., 30, 2798-2813, 2000. lation and air-seafluxeson multiple time scales,J. Mete- Yang, S. K., Y. T. Hou, A. J. Miller, and K. A. Campana, orol. Soc. Jpn., 75, 831-849, 1997. Evaluation of the earth radiation budget in NCEP-NCAR Taylor, P. K., The determination of surface fluxes of heat reanalysis with ERBE, J. Clim., 12, 477-493, 1999. Sennechael, N., C. Frankignoul,and M. A. Cane, An adaptive procedurefor tuning a seasurfacetemperaturemodel, and water by satellitemicrowaveradiomerry and in situ measurements,in Large-scaleOceanographicExperiments and Satellites, edited by C. Gautier and M. Fieux, D. Reidel, NATO ASI SeriesC, Sciences, vol. 128 pp., 223246, 1984. Trenberth, K. E., and C. J. Guillemot, Evaluation of the atmospheric moisture and hydrological cycle in the G. Auad, D. Cayan, A. Miller, and J. Roads, Climate Research Division, Scripps Institution of Oceanography, La Jolla, CA, 92093. (e-mail: guillo@ucsd.edu; dcayan@ucsd.edu; ajmiller@ucsd.edu;jroads@ucsd.edu) NCEP/NCAR reanalyses, Clim. Dyn., 1•, 213-231,1998. Trenberth, K.E., and J.W. Hurrell, Decadal atmosphereocean variations in the Pacific, Clim. Dyn., 9, 303-319. 1994. (ReceivedJanuary 28, 2000; revised January 17, 2001; acceptedJanuary 18, 2001.)