8/14/2019 Maze et al, JGR 2006

1/17

Low-frequency variability in the Southern Ocean region in a

simplified coupled model

Guillaume MazeLaboratoire de Physique des Oceans, Brest, FRANCELaboratoire de Meteorologie Dynamique, Paris, FRANCE

Fabio DAndreaLaboratoire de Meteorologie Dynamique, Paris, FRANCE

Alain Colin de VerdiereLaboratoire de Physique des Oceans, Brest, FRANCE

Received 22 July 2005, accepted 16 January 2006

Abstract. Patterns of interannual variability of the ocean-atmosphere coupled systemin the Southern Hemisphere extratropics are studied with a simple dynamical model, inorder to determine the basic physical processes of interaction independently of tropicalforcing. The model used is an atmospheric quasi-geostrophic model coupled to a slaboceanic mixed layer, which includes mean geostrophic advection by the Antarctic Cir-cumpolar Current (ACC). The ocean-atmosphere coupling occurs through surface heatfluxes and Ekman current heat advection. In a fully coupled simulation, the atmosphericpart of the model, which includes high frequency transient eddies at midlatitudes, ex-hibits a strong Southern Annular Mode (SAM) as the first mode of variability at inter-annual time-scales. The SAM-related wind anomalies induce Ekman currents in the mixedlayer which produce sea surface temperature anomalies. These are then advected alongby the ACC. A forced mechanism where the ocean role is reduced to advect the SSTappears sufficient to reproduce the main features of the variability. Nevertheless, a pos-itive feedback of the ocean was also found. It operates through anomalous Ekman cur-rents heat advection and contributes to the maintenance of the SST anomaly.

1. Introduction

Climate variability of the Southern Ocean (SO) regionhas been comparatively less studied than the regions of theNorthern Hemisphere. Since about twenty years, however,availability of new in situ and satellites observations, and

the consequent improvement of reanalysis datasets, havetriggered new interest. The SO has a peculiar geogra-phy: bounded to the South by the quasi-circular and pole-centered Antarctica continent, it is not interrupted by anymeridional coast from mid to high latitudes. Consequently,it easily redistributes local climatic anomalies among alloceans. Identifying physical processes that drive this re-distribution, as well as the origin of the anomalies is animportant goal in the climate studies.

Using reanalysis data from the ECMWF and sea sur-face height from TOPEX/POSEIDON satellite, White andPeterson [1996, hereafter WP96] and Jacobs and Mitchell[1996] have discovered oceanic and atmospheric anomaliespropagating around Antarctica in 8-10 years. This phe-nomenon, called Antarctic Circumpolar Wave (ACW), hasbeen initially described as a wave phenomenon with a zonal-wavenumber two and a constant phase relationship betweensea surface temperature (SST) and sea level pressure (SLP).The phase relation is similar to the one observed in theNorth Pacific at decadal periods [Nakamura et al., 1997] andin some North Atlantic patterns [Kushnir and Held, 1996]

Copyright 2006 by the American Geophysical Union.0148-0227/06/$9.00

with high SLP located downstream (eastward) of warm SST.In the circular geometry of the SO, this gives SLP and SSTin quadrature.

The ACW has been studied with several analytical mod-els focusing on planetary waves dynamics coupled to sim-ple ocean models [Qiu and Jin, 1997; White et al., 1998;

Goodman and Marshall, 1999; Talley, 1999; Baines and Cai,2000; Colin de Verdiere and Blanc , 2001]. All these modelsreproduce some of the ACWs characteristics as described byWP96. They possess different instability mechanisms whichoften differ also in the value of the growth rates. However,it was also found that an atmospheric stochastic forcing ofthe Southern Ocean is able to reproduce characteristics ofACW-like variability [Weisse et al., 1999; Haarsma et al.,2000; Verdy et al., 2005] emphasising the necessity to in-clude a more realistic atmospheric turbulence.

The ACW has also been studied in oceanic and/or at-mospheric global circulation models [Christoph et al., 1998;Motoi et al., 1998; Weisse et al., 1999; Cai et al., 1999;Bonekamp et al., 1999]. These models allow for a full rep-resentation of the climate system physics and long simula-tions can provide more significant statistics, which is notthe case for short observational datasets. Differences withobservations are significant. General Circulation Models(GCMs) exhibit a zonal-wavenumber three standing wave inthe atmosphere and an eastward propagating wave (some-times quasi-standing) with zonal-wavenumber two or threein the ocean. These results produced a debate concerningthe ACW and its very existence [Christoph et al., 1998; Caiet al., 1999]. Additionally, other studies have shown that cli-matic anomalies in the ACW could be initiated through tele-connections from ENSO [Peterson and White, 1998; Cai and

1

8/14/2019 Maze et al, JGR 2006

2/17

Baines, 2001] and even interact with it [White and Cayan,2000; White et al., 2002; White and Annis, 2004]. Thedifficulty of numerical models to reproduce ENSO-inducedvariabilities in mid-to-high latitudes can be at the origin ofthe differences from observations [Cai and Baines, 2001].

The disagreement between observations and numericalsimulations have been moderated in studies using longerdatasets, that have completed the ACWs description. Itseems that SLP, and other atmospheric variables, are notalways propagating and can take the shape of a zonal-wavenumber three standing wave pattern [Bonekamp et al.,1999; Carril and Navarra, 2001; Venegas , 2003; Park et al.,2004]. The ACW could be the combination of two signalshaving different origins and mechanisms [Venegas , 2003].

The first, eastward propagative and of zonal-wavenumbertwo, may be the result of an extratropical dynamical mech-anism but with ENSO as a tropical trigger. This is thedominant signal for the time period 1985-1994 analysed byWP96. The second, standing in the atmosphere and per-haps propagative in the ocean, of zonal-wavenumber three,is restricted to mid-to-high latitudes and may have dynam-ics related to the Southern Annular Mode [SAM, Thompsonand Wallace, 2000].

The goal of this paper is to study the SO variability inde-pendently of remote forcing from low-latitudes, focussing onintrinsic midlatitudes dynamics. We use a model of interme-diate complexity, in between full blown climate models andthe long wave analytical models reviewed above. It is an at-mospheric quasi-geostrophic tridimensional model coupledto a slab oceanic mixed layer of constant depth, which

includes mean geostrophic advection by the ACC. The at-mospheric part has a good representation of atmospherictransient dynamics, including baroclinic eddies and theirgeneration processes.

This paper studies the interannual variability in theSouthern Ocean, represented by the sea surface tempera-ture. Our goals are first to describe this variability, sec-ond to identify how SST anomalies are created, and last todetermine how important is the coupling between each cli-matic components and how they interact with each other.To do so, the model is integrated in three different configu-rations. First, o cean and atmosphere are fully coupled andcan interact with each other (CPL run). Second, the oceanis passively forced by the atmosphere which in turn onlyfeel climatological boundary conditions at the ocean sur-face. This configuration, called hereafter FR-OC, turns off

ocean feedback on the atmosphere and simply allows to findout if the atmospheric forcing is sufficient to reproduce SOvariability. Third, the atmosphere is passively forced by theocean, i.e. the mean equilibrium atmospheric response toSST anomaly is studied. This configuration, called hereafterFR-AT, identifies how the atmospheric feedback on the SOexplains the differences between CPL and FR-OC simula-tions.

Numerical simulations CPL and FR-OC consist of 1100years long perpetual winter integrations. The first 100 yearsare used to spin-up the ocean and the atmosphere to theirequilibrium regime. We recorded monthly means of eachvariable of the following 1000 years, and then computedmonthly anomalies from the long-term mean state. We alsoused daily atmospheric streamfunction fields from the last100 years of integrations in order to compute stormtrack

statistics of the atmosphere (a 11 days square window wasused). Numerical simulation FR-AT consists of two 50 yearslong integrations. The first one is a control integration, per-formed with a climatological SST field and the second oneis an anomalous integration where a SST anomaly is addedto the climatological SST. The stationary atmospheric re-sponse to the SST anomaly is defined as the 50-years long-term mean difference between the anomalous and the controlintegration.

Section 2 describes the models equations (subsection 2.1)and the climatology of the CPL simulation (subsection 2.2).

Section 3 describes the interannual variability of the CPLsimulation, first focusing on the SST and next extending thedescription to atmospheric covariabilities. Section 4 identi-fies the mechanism responsible for SST anomalies creation.Then, oceanic and atmospheric feedbacks are determined insection 5 where simulations FR-OC and FR-AT are succes-sively analysed in subsections 5.1 and 5.2. Last, results aresummarised and discussed in section 6.

2. Model description

2.1. Equations

The atmospheric part of the model is quasi-geostrophic(QG) with global domain on the sphere and pressure asa vertical coordinate; it was developed by Marshall andMolteni [1993, hereafter MM93]. A version of this modelwas used by Ferreira and Frankignoul [2001] and DAndreaet al. [2005] for studies of the North Atlantic climate vari-ability, while a more complex version was extended to theSouthern Hemisphere by Haarsma et al. [2000]. The quasi-geostrophic potential vorticity (PV) equation with dissipa-tion and forcings:

q

t= J(, q)Dq() + F(, T) + Sq (1)

is discretized at 3 vertical levels: 200, 500 and 800hPa (see

figure 1). is the QG streamfunction. Dq represents lineardissipation terms including Ekman dissipation (orographydependent) and Newtonian thermal relaxation between thelayers. Orographic effects are included in the PV definitionat the lower level. Sq is a time independent source termadded in order to give the model a realistic wintertime cli-matology. Winter was chosen as the season where couplingis strongest between the atmosphere and the ocean mixedlayer. The forcing is computed empirically starting from themean residual of the equation 1 with respect to observations[see DAndrea and Vautard, 2000, and the appendix A formore details]. The third term on the R.H.S. of eq.(1) is theforcing present at 500hPa and 800hPa (of equal amplitudebut of opposite sign) due to heat exchange with the ocean.F represents surface sensible and latent heat fluxes, given

by the bulk aerodynamic formula:

F(, T) = aCDcpa(1 + B1)|Us|(T Ts) (2)

where a is the dry air density, CD = 1.3 103 a constant

drag coefficient, cpa the specific heat at constant volume fordry air, B = 0.5 the Bowen factor (ratio between sensibleand latent heat fluxes), |Us| the surface wind intensity, Tthe sea surface temperature and Ts the surface atmospherictemperature. The expression of can be found in Ferreiraand Frankignoul [2001] and DAndrea et al. [2005]. Poten-tial vorticity is the only atmospheric prognostic variable,surface variables Us and Ts are consequently derived fromit. PV invertion provides streamfunctions at 500hPa and800hPa. These in turn allow to diagnose the temperature

at 650hPa (through thermal wind balance) and the 800hPawind. A simple linear regression (coefficients were estimatedfrom ECMWF analysis) finally provides surface variablesUs,Ts. Hence the dependence ofF on in eqs:1-3. Theocean model is reduced to a single temperature equation fora mixed layer of constant depth Ho with passive advectionby constant geostrophic currents:

T

t= J(g, T) UE.T

F(, T)

CpoDT(T) + ST (3)

8/14/2019 Maze et al, JGR 2006

3/17

where T is the SST, Cpo = ocpoHo the heat capacity ofthe water column (with o the water density and cpo thespecific heat of seawater), DT a linear diffusion term andST a source term computed in a similar way to the termSq in the PV equation. For more details about how thissource term can be computed see Ferreira and Frankig-noul [2005] or DAndrea et al. [2005]. Ho can be verylarge in the SO and subject to a strong seasonal variabil-ity [Kara et al., 2003]. However, the average depth alongthe ACCs core, given 200m, was chosen for winter timeconditions since Ho is relatively uniform along this path,except upstream of Drake passage. Note that taking Hovariable in space should only modify SST anomalies ampli-tude without altering the physical mechanisms we are in-terested in. The geostrophic advection by the ACC is in-troduced via a constant streamfunction g. It is obtainedfrom the dynamic height relative to 1000m, the hydrologybeing that of Reynolds and Smith [1994]. A zonal trans-port of roughly 80Sv was imposed at Drake passage (seehttp://www.mth.uea.ac.uk/ocean/ALBATROSS/ for moredetails about this value). The associated geostrophic cur-rents are shown at the surface in figure 2. The mean zonalcurrent between 40S and 60S is 6.7cm.s1 and 8.3cm.s1

along the ACCs core.The SST is dynamically forced by the atmosphere

through two terms. Surface heat flux F given in eq.(2)above, and advection by Ekman currents, that are also givenby bulk formulae:

UE = aCD|Us

|of0HokUs (4)

where f0 is the Coriolis factor at 45S (k is a vertical unitvector).

The atmospheric (Eq.1) and ocean (Eq.3) models, cou-pled through (Eq.2) and (Eq.4), are integrated in time witha predictor-corrector scheme (first order Adams-Bashforth-Mouton with a time step of 1 hour) and in space on thesphere with a triangular truncation to total wavenumber 21(giving approximately a 5.6o 5.6o resolution for both theatmosphere and the ocean).

2.2. Climatology of the CPL simulation

Despite its simplicity the model has a fairly realistic cli-

matology for both the atmosphere and the SO SST. As il-lustrated in Fig.3, mean 500hPa streamfunction is well re-produced as are the jet maxima in the western part of theIndian basin. Systematic departure of the geopotential fromECMWF JJA climatology (not shown) reaches a maximumof 248m in the Pacific sector of the Antarctic continent butonly 173m at midlatitudes between 20S and 64S. These val-ues are comparable to those of many full fledged AGCMs[DAndrea and coauthors, 1998]. Transient eddy activity, de-fined as the standard deviation (STD) of the 500hPa stream-function high frequency signal is characterised by a maxi-mum of amplitude from eastern Atlantic to eastern Indianbasins (figure 3). The STD of geopotential reaches 100m at500hPa compared with 125m in the observations.

The long-term mean SST difference with respect to JJAclimatological SST from Reynolds and Smith [1994] is less

than 1o

Kall over the SO outside the limit of observed maxi-mum sea ice extent (not shown). Also the strength of merid-ional gradient in the Indian ocean, downstream of the Al-gulhas current is well represented (figure 3).

3. Southern Hemisphere Variability

The Southern Hemisphere interannual variability of thefully coupled integration CPL can be first described by themonthly standard deviation (STD) of SST (figure 4). The

variability is not uniform along a latitude circle with maxi-mum value of 0.3oK occurring in the center of South Indianocean at 40S/73E. This differs from observations where themaximum is located in the Pacific ocean [Cai and Baines ,2001; Park et al., 2004, hereafter YP04].

We use a Fourier method to decompose time/longitudesignal into eastward/westward propagating and stationarycomponents, see the appendix B for details [Park, 1990,YP04]. This decomposition method was preferred to oth-ers (such as Complex EOF) for its simplicity and to allow adirect quantitative comparison with YP04. Figure 5 showsthe time resynthesized eastward propagating and stationarysignal of monthly SST averaged between 41.5S and 47S, fora 30-year long section of the integration. Different choices oflatitude bands in the range 40-60S have been tried, and nosubstantial change of results was found. Three main featurescan be observed.

The stationary component is larger than the east-ward propagating signal and the westward component (notshown) is negligible. The stationary and the eastward prop-agating signals account respectively for 51.2% and 37.7%of the total STD. Qualitatively, this is similar to what wasfound by YP04 in the Comprehensive Ocean-AtmosphereData Set [COADS, Slutz et al., 1985]. However, for the25-year long time series they analysed, the stationary partaccounts for 65% of the total while the eastward propagatingpart accounts 25%.

Note that while the amplitude of the eastward propa-gating component is zonally uniform, the amplitude of the

standing signal is not, being maximum in the Indian basin.In YP04, the SST anomaly amplitude was maximum in thePacific basin and associated to a stationary wave train re-motely forced by ENSO.

When looking at both standing and propagative signals,SST anomalies seem to appear in the Indian ocean sectorand then propagate eastward. A good example is given bythe warm SST anomaly appearing around model-year 212at 60E in the standing component panel. This anomalycan also be followed in the eastward component panel atthe same time/longitude location and then traced for 8-10years before it disappears. A 2D power spectral densityestimate (not shown) did not provide any significative fre-quency peaks for either signals (compare with Fig.9 below).Propagation speed of SST was obtained objectively by us-ing a method based on time-lagged crosscorrelation between

fixed and advected anomalies. Computing longitudinal au-tocorrelation of SST at a given time lag, the longitude ofmaximum correlation gives the longitudinal displacement ofa SST anomaly in that time lag. Taking the mean over lon-gitude and time, a displacement speed can be calculated.Error bars were estimated by subsampling the time series in5 subsections. This yields a propagation speed of 405o.y1.Since the averaged surface speed of the ACC is 8.3cm.s1

(i.e. 35o.y1 at 45S and 44o.y1 at 55S) the propagation ofSST anomalies occurs at about the speed of the ACC.

Geopotential height anomalies at 800 hPa (hereafterZ800) also have a maximum in the Indian sector. The same2D Fourier decomposition was also performed on Z800 alonga midlatitude circle. Stationary wave components dominatestrongly over propagative ones as eastward and westwardpropagative components accounts for 21% of the total STDand the stationary component for 58%.

In order to illustrate how atmospheric variables are re-lated to SST anomalies, we performed a composite analy-sis chained to SST anomalies. Eight p oints were selectedalong the ACC and composites chained to the SST timeseries at those points were obtained. Fields of SST, Z500,Z800, Surface Heat Flux (SHF), Ekman Heat Flux (EKF)and Us, were averaged each time the SST series at the cho-sen location was higher than one STD. Except for the am-plitude, no fundamental differences of pattern were found

8/14/2019 Maze et al, JGR 2006

4/17

between the composites of a given variable at the differentlocations. Consequently, we show the average composite onthe eight longitudinal locations, each map being recenteredat the point selected for the SST scalar series. All maps,normalised to a 1K amplitude of the SST, are shown inFig.6.

The composite SST anomaly is approximately 120o widein longitude and extends meridionaly up to 20o, forminga zonal wavenumber 1 pattern with an aspect ratio of1/6. The atmosphere is equivalent barotropic with positivegeopotential height anomalies at the same longitude as theSST anomaly but shifted southward by 10 degrees. Thereis however a slight westward vertical tilt of 10 degrees from

bottom to top of atmospheric levels. As in baroclinic insta-bilities, such a tilt is the signature of horizontal anomalousheat advection. It has also been found in observations onmonthly time scales [Kidson, 1999]. The meridional shift ofhigh pressure induces an easterly wind anomaly (1m.s1)over the warm SST anomaly. Negative heat fluxes are foundover warm SST anomaly (6W.m2), indicating a tendencyto damp the SST. Easterlies blowing slightly northward ofa positive SST anomaly normally induce poleward Ekmancurrents in the ocean mixed layer which create horizontaladvection of warm water and hence positive Ekman inducedheat fluxes with a maximum amplitude of 14W.m2. Insummary, the transient Ekman currents reinforce a SSTanomaly while the surface heat flux tends to damp it.

4. SST anomalies creation mechanism

In order to understand how SST anomalies are createdby the atmospheric forcing, it is necessary to describe howthe components of the coupled system are connected to eachother in space and time. The method used is the MaximumCovariance Analysis (MCA) which identifies orthogonal pat-terns maximising the covariance between two variables. Itconsists in the Singular Value Decomposition [Brethertonet al., 1992, SVD] of the cross-covariance matrix of the twofields. The method was applied to monthly anomalies, theanomalies have been area weighted but no time filtering hasbeen applied.

The dominant mode of covariability b etween SST andZ800 which accounts for 37.4% of the total covariance is

shown in figure 7. The atmospheric part has the signatureof the Southern Annular Mode [SAM, Thompson and Wal-lace, 2000] with low pressure centered over the Antarcticcontinent and extending to 55S and pattern of opposite signcentered at 45S, extending zonally from the Atlantic to theWestern Pacific sector. The zero pressure contour occursnear latitude 60S 5o. This first mode is consistent withthe observations in the Indian sector but observations ex-hibit a stronger pressure center in the western Pacific andwestern Atlantic.

The oceanic counterpart of the first MCA mode is a dipo-lar SST anomalies with a minimum centered at 55S/120Eand a maximum at 40S/70E. The southern lobe extendsall the way from the Indian to the Western Pacific sectorwhereas the northern lobe extends from the Eastern Atlanticto the Indian sector. The northern lobe is approximately

1.3 times stronger in amplitude than the southern one andis responsible for the maximum of SST variance described insection 3. Both lobes are around 160o wide, explaining thelarge extent of the SST composite shown in figure 6. Notealso how the high atmospheric pressure center appears tobe in quadrature along a meridian with the dipole of SSTanomalies.

Time lagged cross-correlations between principal compo-nents (PC) of this first mode (figure 8) indicate that Z800leads SST by one month with a correlation value of 0.45

(statistically significant to 95%). This very short time-scaleis consistent with the hypothesis of an ocean respondingpassively to the atmospheric forcing [Frankignoul and Has-selmann, 1977; Von Storch, 2000]. Moreover, as shown infigure 9, spectral analysis of PCs reveals no significant peaks.The spectra of atmospheric variables PCs have a white noiseshape with a slight increase of power at low frequencieswhereas the spectrum of SST has a red noise shape. Inagreement with Visbeck and Hall [2004], we found that themodelled Southern Annular Mode is the main mode of in-terannual variability in the Southern Hemisphere, withoutany preferred timescale.

In order to identify how these oceanic and atmosphericpatterns are linked, MCA analysis was also applied to sur-face heat flux (SHF) and Ekman heat flux (EKF).

Figure 10 shows the first MCA mode of Z800 and EKF,accounting for 23.3% of the total covariance. The positivepressure anomaly centered at 45S is covariant with posi-tive EKF at 40S and negative EKF at 50-55S in the In-dian ocean (with maximum amplitude in the western andcenter sector). The first MCA mode of Z800 and SHF,accounts for 21.8% of the total covariance. The positivepressure anomaly is covariant with positive SHF at 35S andnegative at 55S over the whole Indian Ocean. These posi-tive (negative) heat fluxes are due to easterlies (westerlies)generated at 40S/35S (50S/55S) from the Eastern Atlanticto the Western Pacific by the high pressure center. Zonalwinds strongly enhance SHF amplitude and then in conjunc-tion with the positive (negative) atmospheric temperature

induced by the high pressure center (not shown), producethe positive (negative) anomaly. Moreover these easterlies(westerlies) are associated with poleward (equatorward) Ek-man currents at 40S (55S) which in turn have an influencein the Western to center Indian ocean where the meridionalgradient of SST is high. As a result p ositive (negative) EKFdevelop.

Figure 11 shows the first MCA mode of SST and EKF,accounting for 36% of the total covariance. The Indian basinintensified SST is covariant with EKF (once again with max-imum in the western sector). The first MCA mode of SSTand SHF accounts for 27% of the total covariance. The SSTis globally in opposition of phase with SHF. More precisely,the maximum of SST is associated with a minima of SHF,whereas the eastern and western sides of the SST patternare in phase with SHF.

Although patterns of EKF, that are similar in figures 10and 11, patterns of SHF are different. This can be explainedby looking at time lagged cross-correlations between PCs offirst SVDs mode (based upon Z800, Fig.8). Correlationsbetween SHF/EKF and Z800 are in phase (maximum cor-relation occurs at time lag zero) and positive in the Z800-leading side (positive lags), indicating that the atmospheredrives heat fluxes in the way depicted in figure 10. However,we can see on the Z800-lagging side (negative lags), that onemonth after the apparition of the SST anomaly (maximumcorrelation between SST and Z800 at time lag -1 month),Z800 and SHF become anti-correlated whereas Z800 andEKF are still positively correlated. This means that Z800 isdriving positive SHF and EKF until a SST anomaly is cre-ated, then next, Z800 is still driving positive EKF whereasSHF is now driven by the SST and becomes negative as seenin the figure 11 [Von Storch, 2000]. Finally, positive SHFand EKF create warm SST anomaly in the CPL simulation,but with a leading role of EKF in sustaining the SST whereasSHF tends to damp it. Due to the main atmospheric pat-tern of variability (SAM) and it associated midlatitude highpressure center, the SST anomalies creation mechanism islocalised in the Indian ocean. In such a process, the SOmixed layer variability seems to be passively forced by theatmosphere. The main question is now to determine thelevel of the ocean-atmosphere coupling role.

8/14/2019 Maze et al, JGR 2006

5/17

5. Ocean-Atmosphere coupling role

The objective of this section is to identify how active isthe coupling between the ocean and the atmosphere in es-tablishing interannual variability in the CPL simulation. Weanalyse two simulations FR-OC and FR-AT where the oceanand the atmosphere are forced successively by the other.

5.1. Forced ocean

In the FR-OC simulation, the ocean is passively forced bythe atmosphere which in turn only feel climatological bound-ary conditions at the ocean surface. A similar analysis aspreviously integration was performed on the FR-OC sim-

ulation. Results are qualitatively similar between the twosimulations. However, some characteristics are different forSST and surface heat flux (SHF). Monthly anomalies of SSTand SHF are now about 10% smaller in amplitude than inthe CPL experiment. Moreover SST anomalies have longerlifetimes (e-folding time) in the CPL integration. These areestimated from the zonal mean time lagged autocorrelationfunction to be 8.9 0.3 months for the CPL and 7.8 0.3months for the FR-OC integrations. These variations aresmall but significant due to the very long time series.

In summary, SST anomalies are more persistent and ofhigher amplitude in the coupled simulation than in theforced one. Several mechanisms can be hypothesised toexplain these differences. First, diminished SHF dampingdue to low frequency equilibration of ocean and atmospherictemperatures, following Barsugli and Battisti [1998], could

be a possibility since SHF anomalies are indeed 3Wm2

weaker in the CPL integration than in the FR-OC one. Thiscan be seen on PSDs of principal components of the firstMCA mode shown in figure 9, where the FR-OC experimentexhibits smaller variance at low-frequency (for time-scale upto 2 years the cumulative variance is 7% higher in the CPLexperiment). Second, there might be a positive dynamicalfeedback between the ocean and the atmosphere, this possi-bility is explored in the next section.

5.2. Forced atmosphere

In the FR-AT simulation the equilibrium atmospheric re-sponse to a prescribed SST anomaly is calculated and thenthe induced heat fluxes (that can feedback on the SST) arediagnosed. The SST anomaly pattern we choose to prescribe

is the normalised composite computed in subsection 3 fromthe CPL simulation and shown in figure 6 (upper left plot).The equilibrium atmospheric response to this SST anomalyis obtained as the 50-year long-term mean difference betweenthe anomalous and control simulation performed with andwithout the added SST anomaly.

Eight atmospheric responses to the SST anomaly compos-ite centered at eight longitude along 47S latitude have beencomputed. There is a small modulation of the response am-plitude due to the position of the SST anomaly relatively tothe stormtrack but detailed analysis of these responses willbe the subject of future work. Here, we concentrate on theaverage response fields (defined as the average map on theeight longitudinal locations, each map being recentered atthe point of SST anomaly maximum amplitude).

The atmospheric response is summarised in Fig.12. It

can be observed that the atmospheric pattern is meridion-aly roughly in quadrature with the SST anomaly, i.e. highpressure southward and low pressure northward of the warmSST. Response at higher levels (500 and 200hPa) show thesame latitudinal position with an additional shift of about30 downstream (not shown). The response to negative SSTanomalies are qualitatively similar but of opposite signs.

In order to understand how this atmospheric pattern feed-backs on the ocean, EKF and SHF are diagnosed (Fig. 12).The pressure dipole over the SST induce anomalous east-erly winds, which in turn produce poleward Ekman current

and warm advection (positive EKF). EKF being globallypositive with a maximum amplitude of 4W.m2 provides apositive feedback on the SST. On the other hand SHF isnegative all over the SST anomaly but with a much smalleramplitude than EKF, only 0.6W.m2.

In summary there is evidence of a positive feedbackthrough heat flux due to Ekman advection, and a muchsmaller negative one due to surface heat flux. In conjunc-tion with the heat flux damping reduction mentioned above,these two mechanisms tend to reinforce SST anomalies in thecoupled CPL integration.

6. Discussion and Conclusion

Interannual variability of ocean-atmosphere coupled sys-tem was investigated in the Southern Ocean using an in-termediate complexity model, that permits to isolate thepurely extratropical part of the dynamics. The model in-cludes transient atmospheric eddies and an oceanic mixedlayer with the mean geostrophic advection of the ACC; cou-pling occurs via heat fluxes (due to sensible and latent sur-face heat exchanges) and Ekman current advection of heat.

The model has a realistic low frequency variability. SSTanomalies are produced by Ekman advection forced by theSouthern Annular Mode variability, and are then advectedalong by the ACC. A forced mechanism where the sole role ofthe ocean is to advect the SST is sufficient to reproduce themain features of the variability. Nevertheless a positive feed-back of the ocean on the atmosphere was identified whichcontributes to the maintenance of a SST anomaly throughanomalous Ekman currents advection.

Our model being purely extratropical, the principal mech-anism of SST anomaly generation that we identified is theforcing by the SAM. This is demonstrated by the maximumcovariance patterns of Fig.7- 11. Like its northern coun-terpart, the SAM (also known as Antarctic Oscillation orAAO) is the principal mode of variability of the extratropi-cal Southern Hemisphere at least up to interannual periods[Kidson, 1999]. At lower frequencies it is still under debatewhether the SAM still is the principal mode of variability orwhether this role would be taken over by propagative modesof zonal wavenumber 2 or 3 that form part of what is com-monly called the ACW [Hall and Visbeck, 2002; White, 2004;Visbeck and Hall, 2004]. In our work, the SAM dominates

the dynamics at all frequencies.That atmospheric forcing plus mean oceanic advection

by the ACC is sufficient to explain the interannual South-ern Ocean variability patterns is in agreement with previ-ous modelling studies like that of Weisse et al. [1999] andHaarsma et al. [2000], where advection by anomalous oceancurrent was shown to play no role in SST anomalies propa-gation. These studies proposed an advective resonance pro-cess similar that ofSaravanan and McWilliams [1998] as theleading space/time selective mechanism explaining observedSST variability.

This process was also found in the observational study byVerdy et al. [2005], a work of particular interest due to theiruse of an oceanic mixed layer model similar to ours. Theyforced it by NCEP reanalysis data and found both the SAMand remote forcing by ENSO to be the main sources, and ofequivalent importance, of SST variability along the ACCscore, and being in agreement with the advective resonancehypothesis. Because they separated the atmospheric forc-ing into the SAM and ENSO induced components of EKFand SHF, it is attractive to compare Verdy et al. [2005] re-sults to ours. We found a similar SST anomalies amplitudefor the Austral winter (0.3K), a similar propagation speed(8cm.s1) and also a predominant zonal wave number 1 pat-tern. However, they found EKF and SHF playing a equiv-alent role in sustaining and creating SST anomalies, which

8/14/2019 Maze et al, JGR 2006

6/17

differs from our study where a predominance of the formerupon the later was found. This discrepancy may be princi-paly due to the fact that their ocean model cannot feedbackon the atmosphere, which leads to an overestimated surfaceheat flux [Barsugli and Battisti, 1998]. Note also that theirdefinition of the ACCs core localises the 1D ocean mixedlayer model rather south compared to the high meridionalSST gradient in the Indian Ocean, (see their Fig.9). Thisleads to underestimation of both EKF and SHF, but theunderestimation is smaller for SHF. These two points arethought to be responsible for our slightly different conclu-sions about EKF versus SHF roles in the SO variability.

Given the bias induced by limitations of the model physic[such as the absence of ENSO forcing and sea-ice influencethat both move maximum of SST variability in the Pacificsector, Yuan and Martinson, 2000, 2001], our model re-produces much of the interannual variability features thatare found in observations. When decomposing the variabil-ity in a stationary and a propagative component, a domi-nance of the former against the latter was found as in YP04,even if the ratio stationary versus propagative components issmaller in our study. This may be due to the basic modelisedocean dynamic which does not alter the ACC geostrophicadvection as much as in the reality.

An important difference however is that the region ofmaximum SST variability found in our model is the Indianocean, linked to the SAM forcing, while observations [e.g.YP04 or Yuan and Martinson, 2001] place it in the Pacificsector, due to the forcing by ENSO. This is a critical pointof our results; it emphasises the role played by remote forc-ing from low-latitudes in localising a realistic SO interannualvariability, as it was already noticed by Verdy et al. [2005].Note that ENSO teleconnections may enhance the two-waycoupling in the Southern Hemisphere midlatitudes ocean-atmosphere system and then make our study underestimateit compared to the reality. An investigations clearly sepa-rating low from mid-to-high atmospheric forcing of the SOvariability needs to be conducted. Note also that in ourstudy the SST creation location is simply driven by boththe SAM geometry and the meridional SST gradient. A dif-ferent SAM geometry, or other physical mechanisms such asa complete inner ocean dynamic or interactions with sea ice,may localize SST anomalies creation on the Pacific ratherthan in the Indian Ocean; see the low resolution GCM studyof Hall and Visbeck [2002].

The time phase relations between oceanic and atmo-spheric anomalies (see Fig.8) are similar to those found byWP96 in the context of the Antarctic Circumpolar Wavestudies. Some difference were nevertheless found. First, dif-ferent lead-lag times between oceanic and atmospheric vari-ables were found; this is probably due to the severe timefilter (3-7 years admittance) that they used while no timefilter was applied to produce Fig. 8.

Second, our model exhibits somewhat different spatialphases relations between oceanic and atmospheric variables(see Fig.6) with respect to WP96 [and revised by Venegas ,2003]. They observed high pressure downstream, i.e. east-ward, of warm SST. Here, we find high pressure southward ofwarm SST. This is again due to preeminence of the SouthernAnnular Mode in the models variability, which favorises pro-cesses driven by zonal winds. Since temperature gradients

(both in the ocean and in the atmosphere) are mainly merid-ional, two simple mechanisms are present. First, a high pres-sure center induce meridional winds to advect warm/cold airand create a SST anomaly via SHF. This process leads towarm SST anomaly westward of high pressure. Second, ahigh pressure center induces zonal winds and produce merid-ional Ekman currents which in turn advect warm/cold waterand then create a SST anomaly via EKF. This process leadsto warm SST anomaly northward of high pressure. Whiteet al. [1998] revisited WP96 data analysis and found down-stream high pressure to be also shifted southward to SST

anomalies, forming a spiral pattern. This suggests that bothprocesses described above occur in reality, producing a spi-ral pattern with high pressure east-southward of warm SST.Moreover, White et al. [1998] took a f2 dependence of theatmospheric response to SST anomalies which allowed themto model analytically the spiral pattern. They found SSTanomalies to be created and sustained by the ridge-inducedpoleward Ekman transport, as in our study. The importanceof Ekman current heat advection in the coupled experimentCPL have been tested by performing two long coupled sensi-tivity integrations with a doubled constant drag coefficient,first only in the surface heat flux expression, and second onlyin the Ekman heat flux expression. If the SHF amplitudeis increased, SST anomalies are smaller and shorter lived,which confirms the damping role of the turbulent surfaceheat flux. If on the other hand Ekman heat flux amplitudeis increased, the opposite effect is observed.

Spatial phase relations given by figure 6 are very use-ful in attempt to compare our results to those from ana-lytical models of ACW-like variability. Studies of Qiu andJin [1997]; Talley [1999]; Goodman and Marshall [1999] andBaines and Cai [2000] are based on a positive feedback dueto ocean dynamics in response to the overlying equilibratedatmosphere. Since the thermocline is mainly forced by thesurface wind stress curl, a positive feedback needs a down-stream or inphase high pressure at the same latitude thanthe SST anomaly. Given the absence of internal ocean dy-namics in our coupled model, the mechanism for growth ofthe perturbations cannot have such an oceanic origin. Onthe other hand, the vertical phase lags of pressure patternsthat we observe are consistent with an atmospheric energysource as found by Colin de Verdiere and Blanc [2001, CB].Both Goodman and Marshall [1999, GM] and CB modelsproduce equivalent barotropic response in the atmosphere atthe scales for which Rossby waves are stationary (near reso-nance conditions) and we observe this equivalent barotropicresponse here but it is more difficult to check if our dom-inant mode number one lies in the range of scales of sta-tionary Rossby waves with realistic winds. The zonal phaseshift between SST and SLP is absent by construction in GMwhile CB produces high SLP downstream of SST. The latteris consistent with the MCA results (fig. 7) but none of thetwo idealised models however , predict the meridional shiftobserved here. The main surface coupling flux in GM isEKF by construction (since their hypothesis of atmosphereequilibration requires SHF to be exactly zero) and EKF alsoplays the leading role here. On the other hand, SHF is theleading term (with a weaker contribution of EKF) whichallows large growth rates in CB. Extending analytical cou-pled oscillations to more complex models seems to be dif-ficult. This may be due to the presence of synoptic scalehigh frequency activity which make harder to realise in amore complex geometry the parameters set allowing strongresonant conditions.

The relevance of the positive feedback mechanism ofmaintenance of the SST anomalies through EKF for realclimate is difficult to estimate because it is critically depen-dent on the atmospheric response to SST anomaly, itself ahighly studied process but still not completely understoodin the extratropics [Kushnir et al., 2002]. A natural devel-opment to this work is consequently a more detailed studyof atmospheric response to SST anomalies which will be theobject of a forthcoming paper.

Acknowledgments. Among the people that contributed tothe article by discussion and exchange of ideas, the authors wishto thank Frederic Vivier, Arnaud Czaja, David Ferreira, JohnMarshall and Arianne Verdy. Thanks are also due to the ECMWFdata center for making the ECMWF ERA-15 data freely availableover the World Wide Web. We acknowledge support from CNES(Southern Ocean Heat Storage Variability and Climatic Impli-cations, NASA/CNES joint research announcement, the OceanSurface Topography Science Team).

8/14/2019 Maze et al, JGR 2006

7/17

Appendix A: PV equation source terms

In the PV equation (1) we introduced a time-independentsource term Sq that represents all adiabatic and subgrid pro-cesses [DAndrea and Vautard, 2000]. This forcing is com-puted empirically as the mean residual of the equation withrespect to observations following the method introduced byMarshall and Molteni [1993], to which we added a temper-ature correction. The PV equation (1) is satisfied by anychoice of observed fields of , q and T (the hats indicateobservations). Injecting a long observation dataset into thePV equation and taking a long-term mean, the tendency onthe left hand side goes to zero, and we obtain an equation

for Sq:

Sq = [J(, q)] + [D()] [F(, T)] (A1)

The data used were twice daily ECMWF analysis datasetranging from June 1979 to August 1993 for June-July-August, JJA. Brackets stand for the long-term mean.

Using Sq a la Marshall and Molteni [1993] in the case ofthe Southern Hemisphere gave a considerable high pressureerror over the Antarctic continent. Consequently, we addeda zonal mean temperature correction to term Sq. This termwas computed in the following way.

The Marshall and Molteni [1993] forcing (A1) can be in-terpreted as a relaxation temperature in the sense of Heldand Suarez [1994]. Eq.(1) can in fact be obtained eliminat-

ing the vertical velocity in the QG vorticity and temper-ature equations (compare Holton, 1979 p.164):

d2

dt= f0

p(A2)

f0d

dt= + F(, T) +

R

p

1

(T T) (A3)

Which gives

Dq

Dt=

f0

pF(, T)

f0

1

p

R

pT

+

f0

1

p

R

pT

(A4)

Here, q is the potential vorticity, the geostrophic stream-function, f0 the Coriolis factor at 45

oS, the static stability,R the gas constant and a relaxation time scale. The firstterm on the R.H.S. of (A4) corresponds to the heat flux termin (1), and gives the expression of . The second term onthe R.H.S. of (A4) corresponds to the dissipation term of (1)with the dissipative relaxation time = 25days. Finally, thethird term represents the Held and Suarez relaxation tem-perature.

The Marshall and Molteni [1993] forcing method, in otherterms, can be seen as an empirical way to compute a relax-ation temperature T.

In order to correct the forcing term Sq, we added a zon-ally uniform temperature T and recomputed a new termSq

:

Sq =f0

1

p

R

p(T + T)

T is pretty much an ad hoc term, it was computed as afraction of the difference in meridional profile between T

and the observed meridional temperature profile.

Appendix B: 2D Fourier decomposition

We can decompose a signal (x, t) as a sum of harmonicsdefined by the numbers n and m:

(x, t) =

n,mAwnm cos(knx + mt

wnm)

+Aenm cos(knx mt enm) (B1)

where Awnm and Aenm are the westward and eastward waves

amplitude, kn and m the spatial and temporal wavenum-bers, and wnm and

enm the westward and eastward phase

lags. All coefficients can b e computed from Fourier trans-form. The reader is invited to see details in Park [1990] andPark et al. [2004].

References

Baines, P. G., and W. Cai (2000), Analysis of an interactive in-stability mechanism for the Antarctic Circumpolar Wave, J.Clim., 13, 18311844.

Barsugli, J. J., and D. S. Battisti (1998), The basic effects ofatmosphere-ocean thermal coupling on midlatitude variability,J. Atmos. Sci., 55(4), 477493.

Bonekamp, H., A. Sterl, and G. J. Komen (1999), Interannualvariability in the southern ocean from an ocean model forcedby european center for medium-range weather forecasts reanal-ysis fluxes, J. Geophys. Res., 104, 13,31713,331.

Bretherton, C. S., C. Smith, and J. M. Wallace (1992), An inter-comparison of methods for finding coupled patterns in climatedata, J. Clim., 5(6), 541560.

Cai, W., and P. G. Baines (2001), Forcing of the Antarctic Cir-

cumpolar Wave by ENSO teleconnections, J. Geophys. Res.,106, 90199038.

Cai, W., P. G. Baines, and H. B. Gordon (1999), Southern mid-to high-latitude variability, a zonal wavenumber-3 pattern, andthe Antarctic Circumpolar Wave in the CSIRO coupled model,J. Clim., 12, 30873104.

Carril, A. F., and A. Navarra (2001), Low-frequency variability ofthe Antarctic Circumpolar Wave, Geophys. Res. Lett., 28(24),46234626.

Christoph, M., T. Barnet, and E. Roeckner (1998), The Antarc-tic Circumpolar Wave in a coupled ocean-atmosphere GCM,J. Clim., 11, 16591672.

Colin de Verdiere, A., and M. Blanc (2001), Thermal resonance ofthe atmosphere to SST anomalies. Implications for the Antarc-tic Circumpolar Wave, Tellus, 53A, 403424.

DAndrea, F., and coauthors (1998), Northern hemisphere atmo-spheric blocking as simulated by 15 general circulation modelsin the period 1979-1988, Clim. Dyn., 14, 385407.

DAndrea, F., and R. Vautard (2000), Reducing systematic errorsby empirically correcting model errors, Tellus, 52A, 2141.

DAndrea, F., A. Czaja, and J. Marshall (2005), Impact ofanomalous ocean heat transport on the north Atlantic oscilla-tion, J. Clim., 18, 4955-4969.

Ferreira, D., and C. Frankignoul (2001), Coupled ocean-atmosphere dynamics in a simple midlatitude climate model,J. Clim., 14, 37043723.

Ferreira, D., and C. Frankignoul (2005), The transient atmo-spheric response to midlatitude SST anomalies, J. Clim., 18,10491067.

Frankignoul, C. (1985), Sea surface temperature anomalies, plan-etary waves, and air-sea feedback in the middle latitudes, Rev.Geophys., 23(4), 357390.

Frankignoul, C., and K. Hasselmann (1977), Stochastic cli-mate model. part II: Applications to sea-surface temperatureanomalies and thermocline variability, Tellus, 29, 289305.

Goodman, J. C., and J. Marshall (1999), A model of decadalmiddle-latitude atmosphere-ocean coupled modes, J. Clim.,12, 621641.

Haarsma, R., F. Selten, and J. Opsteegh (2000), On the mecha-nism of the Antarctic Circumpolar Wave, J. Clim., 13, 14611480.

Hall, A., and M. Visbeck (2002), Synchronous variability in thesouthern hemisphere atmosphere, sea ice, and ocean resultingfrom the annular mode, J. Clim., 15, 30433057.

Held, I., and M. Suarez (1994), A proposal for the intercompari-son of the dynamical cores of atmospheric general circulationmodels, Bull. Am. Meteorol. Soc., 75, 18251830.

8/14/2019 Maze et al, JGR 2006

8/17

Hoskins, B., and D. Karoly (1981), The steady linear response ofa spherical atmosphere to thermal and orographic forcing, J.Atmos. Sci., 38, 11791196.

Jacobs, G., and J. Mitchell (1996), Ocean circulation variationsassociated with the Antarctic Circumpolar Wave, Geophys.Res. Lett., 23, 29472950.

Kara, A., P. Rochford, and H. Hurlburt (2003), Mixed layer depthvariability over the global ocean, J. Geophys. Res., 108(C3),241.

Kidson, J. W. (1999), Principal modes of southern hemispherelow-frequency variability obtained from NCEP-NCAR reanal-yses, J. Clim., 12, 28082830.

Kushnir, Y., and I. M. Held (1996), Equilibrium atmospheric re-sponse to north Atlantic SST anomalies, J. Clim., 9, 1208

1220.Kushnir, Y., W. A. Robinson, I. Blade, N. M. J. Hall, S. Peng,and R. Sutton (2002), Atmospheric GCM response to extrat-ropical SST anomalies: Synthesis and evaluation, J. Clim., 15,22332256.

Marshall, J., and F. Molteni (1993), Toward a dynamical un-derstanding of planetary-scale flow regimes, J. Atmos. Sci.,50(12), 17921818.

Motoi, T., A. Kitoh, and H. Koide (1998), Antarctic Circumpo-lar Wave in a coupled ocean-atmosphere model, Ann. Glaciol.,27, 483487.

Nakamura, H., G. Lin, and T. Yamagata (1997), Decadal climatevariability in the north Pacific during the recent decades, Bull.Am. Meteorol. Soc., 78, 22152225.

Park, Y.-H. (1990), Mise en evidence dondes planetaires semi-annuelles baroclines au sud de locean Indien par altimetriesatellitaire, C. R. Acad. Sci. Paris, 310(2), 919926.

Park, Y.-H., F. Roquet, and F. Vivier (2004), Quasi-stationaryENSO wave signals versus the Antarctic Circumpolar Wavescenario, Geophys. Res. Lett., 31 (L09315).

Peterson, R. G., and W. B. White (1998), Slow oceanic tele-connections linking the Antarctic Circumpolar Wave withthe tropical El Nino-Southern Oscillation, J. Geophys. Res.,103(C11), 24,57324,583.

Qiu, B., and F.-F. Jin (1997), Antarctic Circumpolar Waves: Anindication of ocean-atmosphere coupling in the extratropics,Geophys. Res. Lett., 24 (21), 25852588.

Reynolds, R., and T. Smith (1994), Improved global sea surfacetemperature analyses using optimum interpolation, J. Clim.,7, 929948.

Saravanan, R., and J. C. McWilliams (1998), Advective ocean-atmosphere interaction: an analytical stochastic model withimplications for decadal variability, J. Clim., 11, 165188.

Slutz, R., S. Lubker, J. Hiscox, S. Woodruff, R. Jenne, P. Steurer,

and J. Elms (1985), Comprehensive Ocean-Atmosphere DataSet; Release 1, 268 pp., NOAA, Environ. Res. Lab., Boulder,Colorado, USA.

Talley, L. D. (1999), Simple coupled midlatitude climate models,J. Phys. Oceanogr., 29, 20162037.

Thompson, D. W., and J. M. Wallace (2000), Annular modes inthe extratropical circulation. Part I: Month-to-month variabil-ity, J. Clim., 13, 10001016.

Venegas, S. A. (2003), The Antarctic Circumpolar Wave: a com-bination of two signals ?, J. Clim., pp. 25092525.

Verdy, A., J. Marshall, and A. Czaja (2005), Sea surface temper-ature variability along the path of the Antarctic CircumpolarCurrent, J. Clim., submitted.

Visbeck, M., and A. Hall (2004), Reply to: Comments on Syn-chronous variability in the southern hemisphere atmosphere,sea ice, and ocean resulting from the annual mode, J. Clim.,17, 22552258.

Von Storch, J.-S. (2000), Signatures of air-sea interactions in acoupled atmosphere-ocean GCM, J. Clim., 13, 33613379.

Weisse, R., U. Mikolajewicz, A. Sterl, and S. Drijfhout (1999),Stochastically forced variability in the Antarctic Circumpolarcurrent, J. Geophys. Res., 104, 11,04911,064.

White, W. B. (2004), Comments on Synchronous variability inthe southern hemisphere atmosphere, sea ice, and ocean re-sulting from the annual mode, J. Clim., 17, 22492254.

White, W. B., and J. Annis (2004), Influence of the AntarcticCircumpolar Wave on El Nino and its multidecadal changesfrom 1950 to 2001, J. Geophys. Res., 109(C06019).

White, W. B., and D. R. Cayan (2000), A global ENSO wave insurface temperature and pressure and its interdecadal modu-lation from 1900 to 1997, J. Geophys. Res., 105(C5), 11,22311,242.

White, W. B., and R. G. Peterson (1996), An Antarctic Circum-polar Wave in surface pressure, wind, temperature and sea-iceextent, Nature, 380, 699702.

White, W. B., S.-C. Chen, and R. G. Peterson (1998), TheAntarctic Circumpolar Wave: A beta effect in ocean- atmo-sphere coupling over the southern ocean, J. Phys. Oceanogr.,28, 23452361.

White, W. B., S.-C. Chen, R. J. Allan, and R. C. Stone (2002),Positive feedbacks between the Antarctic Circumpolar Waveand the global El Nino-southern oscillation wave, J. Geophys.Res., 107, 31653182.

Yuan, X., and D. G. Martinson (2001), Antarctic sea ice extentvariability and its global connectivity, J. Clim., 13, 16971717.

Yuan, X., and D. G. Martinson (2001), The Antarctic dipole andits predictability, Geophys. Res. Lett., 28(18), 36093612.

Laboratoire de Physique des Oceans, Universite de BretagneOccidentale, UFR Sciences, 6, avenue Le Gorgeu C.S. 93837,29238 Brest cedex 3, FRANCE (gmaze@univ-brest.fr)

8/14/2019 Maze et al, JGR 2006

9/17

relationslinear

ACC

T650

q1

q3U3

Ts

Us

q2

800hPa

Ocean surface

200hPa

500hPa

T Mixed layer

Figure 1. Schematic of vertical discretization of the ocean-atmosphere model.

0o

60oE 120

oE 180

oW 120

oW 60

oW

70oS

60oS

50oS

40oS

30oS

Figure 2. Absolute geostrophic surface current of the model. Light, mid and heavy gray contours arefor currents speed higher than 6, 8 and 10 cm.s1.

A Mean 500hPa

0

0

010

10

20

20

30

30

40

40

50

50

60

60

10

1

10

B STD HF 500hPa

4

5

5

5

6

6

6

66

77

7

7

7

8

8

8

8

89

9

9

9

C Mean SST

0

0

0

0

5

5

5

10

10

10

15

15

15

15

20

2020

20

Figure 3. Climatological fields of the CPL simulation. Left: 500 hPa streamfunction long-term mean(contours every 107m.s2). Middle: STD of 500 hPa streamfunction fluctuations having period lowerthan 11 days (contours every 106m.s2). Right: SST long-term mean (contours every 2.5oK).

8/14/2019 Maze et al, JGR 2006

10/17

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1 0

.1

0.1

0.1

0.2

0.2

0.2

0.2

0.2

0.2

0.2

0.3

0o

6

0oE

120oE

180oW

120oW

60

oW

Figure 4. Standard deviation of monthly SST anomalies from the CPL simulation. Contours step 0.05K.

8/14/2019 Maze et al, JGR 2006

11/17

200

205

210

215

220

225

A) TOTAL SIGNAL

Time(year)

0 60E 120E 180 120W 60W0

0.1

0.2

STD

B) STANDING

0 60E 120E 180 120W 60W

C) EASTWARD

0 60E 120E 180 120W 60W

Figure 5. Upper plots: Hovmoller diagram of total (left), stationary (center) and eastward (right) com-ponents of SST anomalies from CPL simulation for a 30 years section of the integration. Anomalies havebeen averaged between 41S and 47S and band-pass filtered with a 2-20 years bandpass filter. Contoursevery 0.05K, zero contours thick, negative contours dashed and positive contours shaded. Lower plots:STD at each longitude for the entire simulation (K).

8/14/2019 Maze et al, JGR 2006

12/17

Figure 6. Normalised zonal-mean composites associated to a 1 STD level SST anomaly centered at 47S.(A) SST, (B) geopotential height at 500 hPa, (C) SHF, (D) geopotential height at 800 hPa, (E) EKF and(F) zonal surface wind. Heat fluxes are positive into the o cean. Negative contours dashed. Longitudesare relative to the SST anomaly center.

8/14/2019 Maze et al, JGR 2006

13/17

12

12

12

4

4

4

4

4

4

4

4

4

4

4

4

12

12

12

12

12

20

0o

60oE

120oE

180oW

120oW

60oW

Z800SVD:1

(37.4%)

0.1

0.1

0.1 0

.1

0

.1

0o

60oE

120oE

180oW

120oW

60oW

SSTSVD:1

(37.4%)

Figure 7. Leading MCA mode as regression of fields onto first principal component time series. Left forZ800 (m) and right for SST (oK). Contours each 4m for Z800 and 0.05oK for SST. Negative contoursare dashed and zero contour omitted. The variance fractions of each modes are indicated on upper leftplots. A Monte Carlo test have shown that they are statiscally significant up to 99%.

12 9 6 3 0 3 6 9 120.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Lags (month)

Crosscorrelation

z800SST

EKF

SHF

Figure 8. Time lagged cross-correlation between first PCs of SVDs between Z800 and SST, EKF andSHF. Z800 leads on the right side and lags on the left one. Only correlations significative to 95% areplotted.

8/14/2019 Maze et al, JGR 2006

14/17

0.02 0.0330.05 0.1 0.2 0.33 0.5 10

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

x 103

Frequency (cycle/year)

NormalizedPSD

A

0.02 0.0330.05 0.1 0.2 0.33 0.5 10

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0.016

Frequency (cycle/year)

NormalizedPSD

B

Figure 9. Normalised power spectral density (PSD) of PCs of first MCA mode for A: Z800 and B: SST(heavy lines) with a red noise fit (light lines) and 99% confidence interval (light lines with vertical shift).In both plots, continuous lines stand for the CPL experiments and dashed ones for the FR-OC one.

8/14/2019 Maze et al, JGR 2006

15/17

1

2

4

44

4

4

44 12

12

12

20

0o

60oE

120oE

180oW

120o

W

60oW

SVD:1

(23.3%) Z800

4

4

4

4

4

8

0o

60oE

120oE

180oW

120o

W

60oW

SVD:1

(23.3%) EKF

12

4

4

4

4

4

44

12

12

12

20

0o

60oE

120oE

180oW

120o

W

60oW

Z800SVD:1

(21.8%)

4

4

4 4

4

4

4

4

8

8

8

8

0o

60oE

120oE

180oW

120o

W

60oW

SHFSVD:1

(21.8%)

Figure 10. Leading MCA mode of the CPL simulation as regression of fields onto first principal com-ponent time series. Upper: between Z800 (m, left) and EKF (W.m2, right). Lower: between Z800(m, left) and SHF (W.m2, right). Contour step 4m for Z800 and 2W.m2 for EKF and SHF, negativecontour dashed and zero contour omitted. The variance fractions of each modes are indicated on upperleft plots. A Monte Carlo test have shown that they are statiscally significant up to 99%.

8/14/2019 Maze et al, JGR 2006

16/17

0.1

0.1

0.2

0o

60oE

120oE

180oW

120o

W

60oW

SSTSVD:1(36%)

4

4

4

8 12

0o

60oE

120oE

180oW

120o

W

60oW

EKFSVD:1(36%)

0.1

0.1

0.2

0o

60oE

120oE

180oW

120o

W

60oW

SSTSVD:1(27%) 44

4

4

4

8

0o

60oE

120oE

180oW

120o

W

60oW

SHFSVD:1(27%)

Figure 11. Leading MCA mode of the CPL simulation as regression of fields onto first principal com-ponent time series. Upper: between SST (oK, left) and EKF (W.m2, right). Lower: between SST (oK,left) and SHF (W.m2, right). Contour step 0.05oK for SST and 2W.m2 for EKF and SHF, negativecontour dashed and zero contour omitted. The variance fractions of each modes are indicated on upperleft plots. A Monte Carlo test have shown that they are statiscally significant up to 99%.

8/14/2019 Maze et al, JGR 2006

17/17

A) SST (oC)

0.2

0.2

0.4

0.6

0.8

80oW 40oW 0o 40oE 80oE

60oS

54oS

48oS

42oS

36oS

30oS

B) Z800 (m)

4

3

21

1

1

2

3

3

4

80oW 40oW 0o 40oE 80oE

60oS

54oS

48oS

42oS

36oS

30oS

C) SHF (W.m2

)

0.6

0.40.2

0.2

80oW 40

oW 0

o40

oE 80

oE

60o

S

54oS

48oS

42oS

36oS

30oS

D) EKF (W.m2

)

0.5

0.5

0.5

1

1

1.5

2

2.53

3.5

80oW 40

oW 0

o40

oE 80

oE

60o

S

54oS

48oS

42oS

36oS

30oS

Figure 12. Atmospheric response to SST anomaly. (A) composite SST anomaly pattern prescribed(contour step 0.2K). (B) geopotential height response at 800 hPa (contour step 1m). (C-D) diagnosedSHF and EKF (contour step 0.2W.m2 for SHF and 0.5W.m2 for EKF). Fluxes are positive in theocean. Negative contours dashed and zero contour omitted. Longitudes are relative to the SST anomalycenter, marked by horizontal and vertical black lines.