observations of coupling between surface wind stress and sea...

1 APRIL 2001 1479C H E L T O N E T A L .

q 2001 American Meteorological Society

Observations of Coupling between Surface Wind Stress and Sea SurfaceTemperature in the Eastern Tropical Pacific

DUDLEY B. CHELTON, STEVEN K. ESBENSEN, MICHAEL G. SCHLAX, NICOLAI THUM, AND

MICHAEL H. FREILICH

College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, Oregon

FRANK J. WENTZ AND CHELLE L. GENTEMANN

Remote Sensing Systems, Santa Rosa, California

MICHAEL J. MCPHADEN

NOAA/Pacific Marine Environmental Laboratory, Seattle, Washington

PAUL S. SCHOPF

Institute for Computational Sciences and Informatics, George Mason University, Fairfax, Virginia, andCenter for Ocean–Land–Atmosphere Studies, Calverton, Maryland

(Manuscript received 27 December 1999, in final form 29 June 2000)

ABSTRACT

Satellite measurements of surface wind stress from the QuikSCAT scatterometer and sea surface temperature(SST) from the Tropical Rainfall Measuring Mission Microwave Imager are analyzed for the three-month period21 July–20 October 1999 to investigate ocean–atmosphere coupling in the eastern tropical Pacific. Oceanictropical instability waves (TIWs) with periods of 20–40 days and wavelengths of 1000–2000 km perturb theSST fronts that bracket both sides of the equatorial cold tongue, which is centered near 18S to the east of 1308W.These perturbations are characterized by cusp-shaped features that propagate systematically westward on bothsides of the equator. The space–time structures of these SST perturbations are reproduced with remarkable detailin the surface wind stress field. The wind stress divergence is shown to be linearly related to the downwindcomponent of the SST gradient with a response on the south side of the cold tongue that is about twice that onthe north side. The wind stress curl is linearly related to the crosswind component of the SST gradient with aresponse that is approximately half that of the wind stress divergence response to the downwind SST gradient.The perturbed SST and wind stress fields propagate synchronously westward with the TIWs. This close couplingbetween SST and wind stress supports the Wallace et al. hypothesis that surface winds vary in response to SSTmodification of atmospheric boundary layer stability.

1. Introduction

A prominent feature of the eastern tropical PacificOcean is a band of cold water near the equator that ex-tends west from South America far into the central equa-torial Pacific. The latitude of the coldest water shiftsnorthward from about 28S at about 858W to the equatorwest of about 1308W (Fig. 1a). The intensity and spatialextent of this ‘‘cold tongue’’ vary seasonally (Mitchelland Wallace 1992) and interannually (Deser and Wallace

Corresponding author address: Dudley B. Chelton, College ofOceanic and Atmospheric Sciences, Oregon State University, 104Ocean Admin. Building, Corvallis, OR 97331-5503.E-mail: [email protected]

1990). During normal years, the cold tongue persists fromJuly to November. The southeasterly trade winds blowacross this band of cold water and into the intertropicalconvergence zone (ITCZ) that is located 58–108 north ofthe equator. The surface wind stress in this southeasterlycross-equatorial flow decreases by more than a factor of4 over the cold tongue and then increases by almost thesame amount to the north of the cold tongue (Fig. 1b).Wallace et al. (1989) hypothesized that the coincidenceof the cold water and light winds is indicative of a cou-pling between the ocean and the atmosphere in the formof oceanic modification of atmospheric boundary layerstability. The objective of this study is to analyze newlyavailable satellite observations of surface wind stress andsea surface temperature (SST) to investigate this hy-pothesized ocean–atmosphere interaction.

1480 VOLUME 14J O U R N A L O F C L I M A T E

FIG. 1. Three-month averages (21 July–20 October 1999) spatially smoothed to remove vari-ability with wavelengths shorter than 28 of latitude by 28 of longitude: (a) TMI measurements ofsea surface temperature; and (b) QuikSCAT measurements of surface wind stress. The locationsof TAO buoys are shown as squares in (a). The solid squares correspond to the buoys used insection 2 to validate the satellite measurements of SST and winds. The white regions in the mapsrepresent missing data owing to island and continental land contamination in the microwavefootprints. The tongue of low wind stress northwest of the Galapagos indicates the presence of awind shadow that extends almost 200 km in the lee of the islands.


The mechanism responsible for the hypothesized cou-pling between SST and surface wind stress is the changeof boundary layer stability that occurs as air from theSouthern Hemisphere crosses the cold tongue and blowsover the warmer water to the north of the equator. Theadvection of relatively warm boundary layer air overthe cold tongue stabilizes the atmospheric boundary lay-er and inhibits downward turbulent mixing of northwardmomentum from aloft. This decoupling of the surfacewinds from the stronger winds at heights of 100 m ormore above the sea surface (Bond 1992) decelerates thenear-surface wind and results in low surface wind stressover the cold water. As the surface winds cross the SSTfront toward warmer water to the north of the coldtongue, the boundary layer is destabilized and surfacesensible heat flux and evaporation increase greatly(Zhang and McPhaden 1995). Convection driven by theenhanced surface heat fluxes increases the downwardflux of northward momentum from aloft and acceleratesthe surface winds toward the north. The downwind ac-celeration results in a divergence of the surface windfield over the northern SST front. This relationship be-tween the surface wind divergence and the SST fieldwas first hypothesized by Wallace et al. (1989) from ananalysis of large-scale climatological average wind andSST fields constructed from historical ship observations.

A unique wavelike feature of the oceanic circulationin the eastern tropical Pacific allows a statistical inves-tigation of the relationship between SST and the surfacewinds. The intense latitudinal shears between the var-ious components of the equatorial current system (Rev-erdin et al. 1994) cause the currents to become unstable(Philander 1978; Cox 1980; Luther and Johnson 1990).This generates ‘‘tropical instability waves’’ (TIWs) thatpropagate westward with periods of 20–40 days, wave-lengths of 1000–2000 km, and phase speeds of ;0.5 ms21 (Qiao and Weisberg 1995). These unstable wavescause large cusp-shaped meridional perturbations of theSST front along the north side of the cold tongue (Le-geckis 1977; Miller et al. 1985; Flament et al. 1996;Kennan and Flament 2000). TIW-induced oceanic eddyheat flux toward the equator has been shown to be com-parable to the Ekman heat flux away from the equatorand the large-scale net air–sea heat flux over the easterntropical Pacific (Hansen and Paul 1984; Bryden andBrady 1989; Baturin and Niiler 1997; Wang andMcPhaden 1999; Swenson and Hansen 1999). TIWs arethus an important component of the large-scale heatbalance of the equatorial cold tongue.

As first noted by Hayes et al. (1989), the space–timevariability of the SST signatures of TIWs provides astringent test of the hypothesized SST-induced atmo-spheric boundary layer modification. If perturbations ofthe surface wind stress field are controlled by pertur-bations of the underlying SST field, then the temporaland spatial variability of the wind stress field shouldvary quantitatively according to the orientation of thewind direction relative to the TIW-induced perturbations

of the SST front. The monthly timescales and ;1500-km wavelength of these perturbations of the SST fieldeffectively yield many realizations from which the re-lationships between winds and SST can be investigatedstatistically.

Time series of wind and SST measured from buoysalong 1108W at 28N, the equator, and 28S were analyzedby Hayes et al. (1989) to investigate the relationshipbetween TIW-induced perturbations of SST and surfacewinds. Variations of SST and near-surface winds werefound to be consistent with the hypothesized atmo-spheric boundary layer modification by the underlyingSST field. The zonal spacing of moored observationsof SST and wind is adequate to investigate the rela-tionship between the meridional components of the gra-dients of the SST and wind fields but is too coarse toresolve the zonal components of the gradient vectors(see Fig. 4 below). In accord with the expected collo-cation of wind divergence and SST gradients, Hayes etal. (1989) found a strong correlation between the me-ridional gradient of SST and the meridional gradient ofthe northward wind component.

The importance of SST-induced modifications of at-mospheric stability has been further investigated fromsatellite observations of the geographical distribution oflow-level cloudiness in relation to the SST signaturesof TIWs (Deser et al. 1993). The highest cloud reflec-tivities were observed over warm SST perturbations,presumably because of increased convective mixing anddeepening of the boundary layer in the unstable regimeover the warmer water. Although surface wind infor-mation was not available, cloud reflectivity was locallyhigh where isotherms were more nearly perpendicularto the climatological average wind direction.

Most recently, TIW-induced modification of the at-mospheric boundary layer has been investigated by Xieet al. (1998) from weekly averages of satellite infraredobservations of SST and Earth Remote Sensing (ERS-1)scatterometer estimates of the 10-m wind field. The de-tailed geographical distribution of wind divergence rel-ative to the SST signatures of TIWs was difficult todiscern, perhaps because of the severe sampling errorsin weekly averaged wind fields constructed from thenarrow 500-km swath width of the ERS-1 scatterometer(Schlax et al. 2001). Westward propagation of wind di-vergence anomalies at apparently the same phase speedas the TIW-induced SST perturbations was nonethelessevident. Xie et al. (1998) interpreted these observationsin the context of a coarse-resolution atmospheric modelforced by an imposed and spatially fixed SST wavepattern. The model results suggest that TIW-inducedperturbations of the low-level winds penetrate wellabove the atmospheric mixed layer, with variability de-tectable to at least 800 mb (about 2000 m above the seasurface). The model also suggests that the TIW-inducedvariability of the surface wind field may force a deepatmospheric response indicated by measurable precip-itation anomalies over the ITCZ. Hashizume et al.


(2001) have recently presented observational evidencein support of this conclusion.

With the exception of the Hayes et al. (1989) analysisof buoy observations at the equator and 28S along1108W, all previous studies of TIW-induced ocean–at-mosphere interaction have focused on the region northof the equator. It has recently been shown that TIW-induced perturbations of the SST field also occur southof the equator (Chelton et al. 2000). As the air flowsacross the perturbed southern SST front, the hypothe-sized SST modification of atmospheric boundary layerstability predicts that the wind stress will adjust to SSTgradients in a manner analogous to the adjustment ofthe flow across the northern SST front. In regions wherethe wind blows from warm to cold water, the wind stressshould decelerate, resulting in a convergence of low-level winds.

The accuracy and spatial resolution limitations ofwind fields available heretofore has long been a rec-ognized problem in the Tropics (e.g., Goldenberg andO’Brien 1981; Reynolds et al. 1989). Wind fields con-structed from satellite scatterometer observations of sur-face winds are likely to contribute greatly to an im-proved understanding of coupled ocean–atmospherevariability in the Tropics. In this study, the hypothesizedcoupling between the surface wind field and SST in theeastern tropical Pacific is investigated from the firstthree months of surface wind stress measured by theQuikSCAT scatterometer. SST fields were constructedfrom near-all-weather satellite observations of SST bythe Microwave Imager on board the Tropical RainfallMeasuring Mission (TRMM). By resolving time- andspace scales that have previously been unaddressable,these newly available satellite datasets reveal ocean–atmosphere interaction with remarkably detailed space–time structure that supports the coupling between thesurface wind stress and SST hypothesized by Wallaceet al. (1989) and Hayes et al. (1989).

The satellite measurements of SST and winds are de-scribed and validated in section 2 from comparisons within situ measurements by the Tropical Atmosphere–Ocean(TAO) buoys (McPhaden et al. 1998) in the eastern trop-ical Pacific. The structure of the relationship between thesurface wind stress field and the underlying SST field ispresented in section 3, with particular emphasis on thederivative wind stress fields (divergence and curl) andtheir relationships to the downwind and crosswind com-ponents of the SST gradient vector. The synchronouswestward propagation of TIW perturbations of the SSTand wind stress fields is described in section 4. Impli-cations of the observed coupling between the wind stressand SST are discussed in section 5.

2. Data description and validation

The SST data for this study were obtained from theTRMM Microwave Imager (TMI) on board the TRMMsatellite (Kummerow et al. 1998). Because the atmo-

sphere is nearly transparent to microwave radiation innonraining conditions, the TMI provides an essentiallyuninterrupted record of the westward propagation ofSST signatures of TIWs (Chelton et al. 2000). Rain-contaminated observations are easily identified andeliminated from further analysis based on vertically andhorizontally polarized brightness temperatures mea-sured by the radiometer itself (Wentz and Spencer 1998).

The TMI operates at microwave frequencies of 10.7,19.3, 21.3, 37.0, and 85.5 GHz. Dual polarization ismeasured at all frequencies except 21.3 GHz, for whichonly vertical polarization is measured. The primarychannel for SST retrievals is 10.7 GHz. At this fre-quency in the absence of rain, the atmospheric contri-bution to the brightness temperatures is less than 3%.The SST retrieval algorithm uses the 19.3-, 21.3-, and37.0-GHz channels to remove this small atmosphericsignal. The more challenging aspect of the SST retrievalalgorithm is the removal of wind speed and wind di-rection contributions to the 10.7-GHz brightness tem-peratures. This is achieved by taking advantage of adistinctive polarization dependence of the wind speedand SST signatures in the 10.7-GHz brightness tem-peratures. The sensitivity of horizontally polarizedbrightness temperature to wind speed is twice that ofvertically polarized brightness temperature. The oppo-site is true for SST; the sensitivity of horizontally po-larized brightness temperature to SST is only half thatof vertically polarized brightness temperature. This ‘‘or-thogonality’’ allows the simultaneous estimation ofwind speed and SST.

There is no distinctive signature of the wind directioneffect on 10.7-GHz brightness temperatures that wouldallow retrievals of wind direction from the TMI passivemicrowave measurements. At wind speeds below 7 ms21, the wind direction signal is very small and doesnot significantly affect the accuracy of SST retrievals.At higher wind speeds, however, the contribution of thewind direction effect can lead to SST retrieval errors inexcess of 18C. To reduce these errors, wind directionestimates are obtained from the operational analyses bythe National Centers for Environmental Prediction.These wind directions are used to apply a correction toremove the small wind directional effect on the hori-zontally and vertically polarized brightness tempera-tures.

A detailed description of the TMI retrieval algorithmis given by Wentz (1998). TMI estimates of SST areobtained from a physically based algorithm that matchesthe brightness temperatures to a radiative transfer modelthat is a function of SST, wind speed, columnar watervapor, and columnar cloud liquid water. The algorithmis a direct extension of the algorithm described by Wentz(1997) for retrieval of wind speed, columnar water va-por, and columnar cloud liquid water from the SpecialSensor Microwave/Imager (SSM/I). SST cannot be ob-tained from SSM/I data because of the lack of mea-surements at 10.7 GHz.


FIG. 2. Binned scatterplot comparison between 3-day averages ofTMI measurements of SST and TAO measurements of temperatureat a depth of 1 m at the locations shown by the solid squares in Fig.1a. The differences between TAO and TMI measurements of tem-perature were binned as functions of (a) the average of the two tem-perature measurements; and (b) TAO measurements of wind speed.Solid circles and vertical bars represent the mean and 61 standarddeviation of the scatter of points within each bin.

The spatial resolution of TMI measurements of SSTis 46 km. Individual measurements have been shown tohave an accuracy of 0.58C in rain-free conditions (Wentz1998). The TMI measurements of SST were validatedfor the purposes of this study by comparing 3-day com-posite averages with collocated 3-day averages of in situmeasurements of 1-m temperature from the TAO buoys.The buoy locations analyzed here are shown by the solidsquares in Fig. 1a. This selection of buoys spans theregion of TIW variability that is the focus of this study.The mean and standard deviation of the differences com-puted from more than 1700 collocated composite av-erages were 0.158C (TMI higher than TAO) and 0.418C,respectively. There was no significant SST dependenceof the relative bias or standard deviation (Fig. 2a). Therewas, however, a small systematic dependence of therelative bias on wind speed; the mean difference in-creased from about 20.48C at 4 m s21 to about zero at9 m s21 (Fig. 2b).

Owing to sensible and evaporative heat fluxes andradiative cooling, the skin temperature in a thin layerat the sea surface is typically a few tenths of a degreeCelsius cooler than the bulk temperature in light winds(Saunders 1967). This difference decreases with in-creasing wind speed and the skin layer is destroyed bywhitecapping at wind speeds higher than about 10 ms21. Estimates of the thickness of the skin layer in light

to moderate winds vary but it is usually considered tobe about 1 mm (Grassl 1976). Since the e-folding ‘‘op-tical depth’’ for microwave radiance at 10.7 GHz isabout 1 mm, TMI estimates of SST are less susceptibleto skin effects than infrared measurements, for whichthe optical depth is less than 10 mm. It therefore seemsunlikely that the differences between skin and bulk tem-peratures would have a significant effect on the cali-bration of TMI estimates of SST. In any case, the smallrelative bias between TMI measurements of SST andTAO measurements of 1-m temperature in Fig. 2 is ofthe opposite sign expected for skin versus bulk tem-perature differences. The weak systematic dependenceof the differences between TAO and TMI temperaturemeasurements on wind speed in Fig. 2b may thereforebe an indication that the wind speed effects on the mi-crowave brightness temperatures have not been com-pletely removed in the TMI SST retrieval algorithm.

For present purposes, the conclusions of this studyare not significantly altered by the possible small sys-tematic error in TMI estimates of SST. We conclude thatthe SST fields constructed from 3-day averages of TMIdata are sufficiently accurate to investigate the couplingbetween SST and the overlying wind field on the time-and space scales of interest in this study.

The surface vector winds analyzed here were obtainedfrom the SeaWinds scatterometer (Freilich et al. 1994)launched on 19 June 1999 on board the QuikSCAT sat-ellite. The geophysical data record began on 15 July1999. The 3-month data record analyzed here extendsfrom 21 July to 20 October 1999. As described in detailby Naderi et al. (1991), scatterometers infer wind stressmagnitude and direction from measurements of micro-wave radar backscatter received from a given locationon the sea surface at multiple antenna look angles. Thewind retrievals are calibrated to the so-called neutral-stability wind at a height of 10 m above the sea surface,that is, the wind that would exist if the atmosphericboundary layer were neutrally stable. The vector windstress t and 10-m neutral stability vector wind areNv10

simply related by t 5 r | |, where r ø 1.2 kgN N NC v vd 10 10

m23 is the air density and is the neutral stability dragNC d

coefficient (Large and Pond 1981).The QuikSCAT instrument measures vector winds

with 25-km resolution over a single 1600-km swath cen-tered on the satellite ground track. The superiority ofQuikSCAT coverage compared with previous scatter-ometers is discussed in detail by Schlax et al. (2001).The 1600-km QuikSCAT swath width is 33% greaterthan its predecessor, the dual-swath National Aeronau-tics and Space Administration scatterometer (NSCAT),and more than 300% greater than the narrow 500-kmsingle-swath scatterometers on board the ERS-1 andERS-2 satellites launched by the European Space Agen-cy. These sampling differences have a major impact onthe accuracies of wind fields constructed from the var-ious scatterometer datasets (Schlax et al. 2001). Sam-pling errors are especially a concern in the derivative


FIG. 3. Binned scatterplot comparisons between 3-day averages ofQuikSCAT and TAO measurements of 10-m neutral-stability windsat the locations shown by the solid squares in Fig. 1a. The differencesbetween TAO and QuikSCAT wind speeds were binned as functionsof (a) TAO measurements of temperature at a depth of 1 m; and (b)the average of the two wind speed measurements. Solid circles andvertical bars represent the mean and 61 standard deviation of thescatter of points within each bin.

wind fields (divergence and curl) that are of interesthere in sections 3 and 4.

Rain-contaminated QuikSCAT observations wereflagged and eliminated from subsequent analysis by com-parison with rain estimates from the three SSM/I satellites(F11, F13, and F14) that are in operation simultaneouswith QuikSCAT. A QuikSCAT observation was flaggedwhen the nearest SSM/I measurement within 50 km and1 h yielded a nonzero rain-rate estimate based on the al-gorithm described by Wentz and Spencer (1998). The clos-est collocations of SSM/I and QuikSCAT measurementswere usually with the F13 satellite.

The prelaunch algorithm used to derive the Quik-SCAT vector winds analyzed in this study was assessedby comparing 3-day composite scalar averages ofQuikSCAT wind speeds with collocated 3-day scalar-average wind speeds measured from the TAO buoys.Daily average TAO winds were converted to 10-m neu-tral stability wind speeds based on TAO measurementsof air–sea temperature difference and surface humidityusing the algorithm described by Fairall et al. (1996).The daily average TAO wind speeds were then scalaraveraged to form 3-day averages for comparison withscalar-average QuikSCAT wind speeds. The mean andstandard deviation of the differences computed frommore than 1700 collocated composite averages were0.74 m s21 (TAO higher than QuikSCAT) and 0.71 ms21, respectively. There were no significant dependen-cies of the relative bias or standard deviation on SST(Fig. 3a) or wind speed (Fig. 3b).

The TAO wind speed measurement accuracy has beenestimated as 0.3 m s21 (Dickinson et al. 2001). Most ofthis measurement uncertainty is due to calibration drifts.The error is therefore not significantly reduced in the3-day averages considered here. A root sum of squarespartitioning of the 0.71 m s21 standard deviation of thedifference between TAO and QuikSCAT wind speedsthus suggests an uncertainty of 0.64 m s21 for the ran-dom component of QuikSCAT wind speed measurementerrors in the 3-day averages considered here.

The systematic 0.74 m s21 mean difference in Fig. 3may be an indication that the wind speeds from theQuikSCAT prelaunch algorithm are biased slightly low.A comprehensive calibration and validation study is un-derway by the QuikSCAT Science Working Team. Forthe purposes of this study, a bias of the QuikSCAT windspeeds has no effect on the derivative wind stress fields(divergence and curl) that are of primary interest insections 3 and 4. The prelaunch algorithm is thereforeadequate to address the coupling between the surfacewind stress field and the underlying SST field; the con-clusions of this study are not altered if the QuikSCATwind retrievals are adjusted to remove the possible 0.74m s21 bias.

The daily average wind observations available fromthe TAO buoys for the time period analyzed here arenot adequate to assess the directional accuracy of theQuikSCAT vector winds. Freilich et al. (2000, unpub-

lished manuscript, hereafter FVD) have compared theQuikSCAT wind directions with over 12 000 buoy ob-servations of winds collected by the National Data BuoyCenter and concluded that the standard deviation of thedirectional difference is 278. The directional accuracyis better characterized in terms of the random componenterror, which FVD find to be about 0.7 m s21 for eachorthogonal component with no significant dependenceon the wind speed. The directional uncertainty of theQuikSCAT winds thus decreases rapidly with increasingwind speed (see also Freilich and Dunbar 1999).

3. The coupling between SST and wind stress

A recent analysis of TMI measurements of the west-ward-propagating SST signatures of TIWs (Chelton etal. 2000) revealed that Pacific TIWs were very ener-getic, both north and south of the equator, during the3-month period considered here. This period coincideswith a La Nina year, during which TIWs are generallyfound to be most energetic. This is in contrast to theNSCAT observational period, September 1996–June1997, during which TIW amplitudes were small (Chel-ton et al. 2001, manuscript submitted to J. Phys. Ocean-ogr.). Contemporaneous QuikSCAT and TMI data thusoffer the first opportunity for high-resolution and densesampling of TIW-induced ocean–atmosphere interactionin the tropical Pacific.


FIG. 4. Three-day average maps over the period 2–4 September 1999: (a) sea surface temperaturewith TAO mooring locations shown as squares; (b) wind stress magnitude; (c) wind stress; (d)wind stress divergence; and (e) wind stress curl. The contours overlaid in (b)–(e) correspond toisotherms at intervals of 18C between 218 and 278C. The SST and wind stress fields were smoothedto remove variability with wavelengths shorter than 28 of latitude by 28 of longitude. To suppressthe amplification of sampling errors, the derivative fields (divergence and curl) were furthersmoothed zonally to remove variability with wavelengths shorter than 48 of longitude.

A 3-day average map of SST is shown in Fig. 4a forthe period 2–4 September 1999, which is a represen-tative time period during the 3-month data record an-alyzed in this study.1 The cusp-shaped features of the

1 Animations of the SST and wind fields can be downloaded fromanonymous ftp by contacting the corresponding author at [email protected].

SST front along the north side of the equatorial coldtongue are characteristic of TIWs (Legeckis 1977). Notethe clockwise rotation of the northern tips of the coldcusps along the northern SST front at about 1608, 1488,1398, and 1288W. Clockwise rotation is especially clearat 1288W. These features are indicative of anticyclonicparticle velocities in the warm regions between succes-sive cusps (Hansen and Paul 1984; Flament et al. 1996;Kennan and Flament 2000; Weidman et al. 1999). Each


of the cold cusps is reproduced with remarkable fidelityin the corresponding 3-day average maps of the windstress shown in Figs. 4b,c. A similar coincidence ofTIW-related perturbations of SST and wind stress mag-nitude is evident along the weaker cusp-shaped southernSST front. The wind stress is higher over warm waterand lower over cold water. As summarized in the in-troduction, the close relationship between SST and windstress magnitude is at least qualitatively consistent withthe SST-induced modification of atmospheric boundarylayer stability hypothesized by Wallace et al. (1989).This relation is quantified below.

The effects of SST-induced changes of atmosphericstability on the overlying wind stress depend on boththe SST gradient and the wind direction relative to theSST gradient vector. There is therefore no general linearrelation between the gradient of the wind stress mag-nitude and the SST gradient. The wind stress responseto geographical variations of the underlying SST fieldcan be better characterized in terms of the wind stressdivergence, = · t , shown in Fig. 4d. In the NorthernHemisphere, a band of high divergence follows the cusppatterns along the northern SST front where the windstress increases as the air flows across isotherms towardwarmer water. The largest wind stress divergences arelocated directly over the strongest SST gradients. It isvisually apparent that the divergence is locally largerwhere the wind stress is aligned perpendicular to iso-therms (i.e., parallel to the SST gradient vector).

The importance of the alignment between the windstress and the SST gradient is also apparent from therelationship between the wind stress divergence andSST south of the equator. The SST gradients are weakerand spatially irregular and the isotherms tend to be morenearly aligned with the wind stress vectors, resulting inrelatively small SST-induced downwind modification ofthe wind stress. Convergence of the wind stress is none-theless apparent in patches where the wind stress de-creases as the air flows across perturbed isotherms to-ward cooler water.

The hypothesized SST modification of the wind stressfield can be independently investigated from the rela-tionship between the crosswind SST gradient and thevertical component of the wind stress curl, = 3 t · k,where k is a unit vector in the vertical direction. Alongthe cuspy northern SST front, the wind stress curl ispositive where the winds blow parallel to isotherms (i.e.,perpendicular to the SST gradient vector; see Fig. 4e).This pattern in the wind stress curl field develops be-cause the winds are stronger over the warmer water tothe right of the wind direction, resulting in a lateralgradient of the wind stress that corresponds to a positivewind stress curl. As the winds over the region of interestblow toward the northwest during the 3-month periodconsidered here (see Figs. 1b and 4c), the largest windstress curl occurs on the southwestern boundaries of thewarm anomalies between successive cold SST cusps(see Fig. 4e).

In the Southern Hemisphere, large negative windstress curl occurs in patches where the wind blows par-allel to perturbed isotherms in regions where water iswarmer to the left of the wind direction. The latitudinalextent of this band of negative wind stress curl is muchwider and the magnitudes are somewhat weaker than inthe northern band of positive wind stress curl becausethe southern SST front is much broader than the northernSST front (see Figs. 1a and 4a).

The band of persistent negative wind stress curl northof about 38N in Fig. 4e is noteworthy. This feature isa manifestation of large-scale eastward curving of thecross-equatorial winds (see Fig. 1b). Lindzen and Nigam(1987), Mitchell and Wallace (1992), and others havenoted that the contrast between the cool atmosphericboundary layer air over the equatorial cold tongue andthe warmer air to the north enhances the northward sealevel pressure gradient in this region. The anticyclonicturning of the cross-equatorial flow north of 38N wherethe Coriolis acceleration becomes important may occurbecause of geostrophic adjustment of the vertically in-tegrated boundary layer flow to the SST-enhanced pole-ward pressure gradient (e.g., Young 1987; Tomas et al.1999). This adjustment process is physically distinctfrom the SST-induced boundary layer modification ofinterest in this study. The statistical analyses of TIW-induced perturbations of the wind stress fields that fol-low are therefore restricted to the latitude range from38N to 58S.

If the relations between the derivative wind stressfields and the underlying SST field described qualita-tively above from the case study shown in Fig. 4 arequantitatively correct, then the wind stress divergencewill vary in proportion to the downwind SST gradient.This component of =T is given by the vector dot product

5 |=T| cosu, where T is SST, is a unit vector=T · t tin the direction of the wind stress, and u is the coun-terclockwise angle from the vector =T to . Similarly,tthe wind stress curl will vary in proportion to the cross-wind component of the SST gradient that can be char-acterized by the vector cross product =T 3 5 |=T|t · ksinu.

Quantifying the relation between the derivative windstress fields and the angle u is complicated by the highlynonuniform (bimodal) distribution of u (see thick solidline in Fig. 5a). As shown by the dashed line in Fig.5a, the peak in the histogram centered near u 5 21008is composed almost entirely of observations within theregion south of the cold tongue (18–58S) where thewinds blow approximately parallel to isotherms (i.e.,=T and are approximately orthogonal). Similarly, ittis apparent from the thin solid line in Fig. 5a that thepeak in the histogram centered near u 5 308 is composedalmost entirely of observations within the region northof the cold tongue (38N–18S) where the winds blowobliquely across isotherms.

The 3-month statistics of the dependencies of themagnitudes of =T and = |t | on u are summarized in


FIG. 5. Binned scatterplots of the angular dependencies of (a) thenumber of observations for the longitude range 1508–1008W and thelatitude range 38N–58S, 38N–18S on the north side of the cold tongue(thin solid line) and 18–58S on the south side of the cold tongue(dashed line); (b) the magnitude of = |t |; (c) the magnitude of =T.The abscissa is the counterclockwise angle u from the SST gradientvector =T to the wind stress vector t . The solid circles and verticalbars in (b) and (c) represent the mean and 61 standard deviation ofthe scatter of points within each bin over the latitude range 38N–58Sand the longitude range 1508–1008W.

FIG. 6. The same as Fig. 5, except for the perturbation SST andwind stress fields, T 9 and t 9, over the longitude range 1508–1008Wand the latitude range 38N–18S on the north side of the cold tongue.The abscissa is the perturbation counterclockwise angle u9 definedin the text.

Figs. 5b,c. The mean values of |=T| were largest nearthe peaks in the histogram of u. Note also that |=T| wasabout a factor of 2 larger on the north side of the coldtongue (positive u) than on the south side of the coldtongue (negative u centered near 1008). In contrast, thepeaks in the histogram of |(= |t |)| had similar amplitudeson opposite sides of the cold tongue. These differencesin the northern versus southern magnitudes of =T and= |t | are visually apparent from the mean SST and windstress fields in Fig. 1. This is discussed further below.

The biases in the distributions of u and |=T| make itdifficult to isolate the predicted sinusoidal dependenciesof = · t and = 3 t · k on u since not all angles u areadequately sampled. These biases are greatly reducedby isolating the TIW-induced perturbations that are ofinterest here by zonally high-pass filtering the down-wind and crosswind SST gradient fields and the windstress fields to remove variability with zonal wave-lengths longer than 208 of longitude (about 2200 km).This filter cutoff is long enough to retain the 1000–2000-km wavelengths of TIWs yet short enough to re-move the large-scale background wind variability thatis unrelated to TIW-induced perturbations of the SSTfield. The perturbation downwind and crosswind SSTgradients will be denoted by ( )9 and (=T 3=T · t

, respectively. The perturbation wind stress willt)9 · kbe denoted by t9. If the hypothesized dependence be-tween the derivative wind stress and SST gradient fieldsis valid on the time- and space scales of the TIW-inducedperturbations, then we would expect = · t9 to be linearlyrelated to ( )9. Likewise, = 3 t9 should be linearly=T · trelated to (=T 3 . These linear relations implyt)9 · kthat the perturbation wind stress divergence and curlshould depend, respectively, on the cosine and sine ofthe perturbation angle defined by

(=T 3 t)9 · k21u9 5 tan .[ ](=T · t)9

In anticipation of the different sensitivities of the de-rivative wind stress fields to SST gradients that are de-duced below, the observations were also segregated into48 latitudinal bands straddling the cold tongue centerednear 18S. As shown in Figs. 6 and 7, the distributionsof u9, |(= |t9|)| and |=T9| within each latitude band aremuch more nearly uniform than in the full SST andwind stress fields shown in Fig. 5.

The 3-month statistics of the dependencies of the zon-ally high-pass filtered wind stress divergence on theangle u9 for the regions north and south of the coldtongue are summarized in Fig. 8. In both latitude bands,there is remarkably good agreement between the binnedmean values of = · t9 and the expected cosine depen-dence on u9. Similarly, the dependencies of the zonallyhigh-pass filtered binned mean values of wind stress curl


FIG. 7. The same as Fig. 6, except for the perturbation SST andwind stress fields, T 9 and t 9, over the longitude range 1508–1008Wand the latitude range 18–58S on the south side of the cold tongue.

FIG. 8. Binned scatterplots of the angular dependencies of zonallyhigh-pass filtered wind stress divergence, = · t9, over the longituderange 1508–1008W for (a) the latitude range 38N–18S on the northside of the cold tongue; and (b) the latitude range 18–58S on the southside of the cold tongue. The solid circles represent the overall meanvalues within each bin over the 3-month data record. The associatedvertical bars represent the 61 standard deviation of the mean valueswithin each bin computed individually for each 3-day period overthe 3-month data record. These standard deviations provide a measureof the uncertainty of the estimate of the overall mean in each bin.The smooth curves represent least squares fits of the binned overallmeans to a cosine.

FIG. 9. The same as Fig. 8, except the angular dependencies ofzonally high-pass filtered wind stress curl, = 3 t9 · k. The smoothcurves represent least squares fits of the binned overall means to asine.

on u9 agree very well with the expected sine dependenceon u9 (Fig. 9). At least on the spatial and temporal scalesof TIWs, these results confirm the hypothesized depen-dencies of the derivative wind stress fields on the ori-entation of the wind stress relative to the SST gradient;the binned mean wind stress divergence or convergenceis largest when the winds blow parallel to the SST gra-dient (across isotherms) and zero when the winds blowperpendicular to the SST gradient. Conversely, thebinned mean wind stress curl is largest when the windsblow perpendicular to the SST gradient (along iso-therms) and zero when the winds blow parallel to theSST gradient.

The statistical relationships between the perturbationderivative wind stress fields and the underlying pertur-bation SST gradient field over the 3-month period con-sidered in this study are quantified for the regions northand south of the cold tongue in Figs. 10 and 11, re-spectively. As expected from the cosine dependenciesin Fig. 8, the binned mean values of the perturbationwind stress divergence, = · t9, vary linearly with thedownwind perturbation SST gradient, , both(=T · t)9north and south of the cold tongue (Figs. 10a and 11a).The different slopes of linear fits to the mean values ineach bin indicate an asymmetric sensitivity of windstress divergence to SST gradients on opposite sides ofthe cold tongue; for a given magnitude of temperaturechange, the change of = · t9 for winds blowing acrossthe southern SST front is about double the correspond-ing change of = · t9 for winds blowing across the north-ern SST front.

The asymmetry of the response of the wind stressdivergence on opposite sides of the cold tongue is alsoqualitatively apparent from Figs. 1a,b, and Figs. 5b,c.As noted previously, the mean SST gradient north ofthe cold tongue is about twice the mean SST gradient


FIG. 10. Binned scatterplots of the relationships between the zonallyhigh-pass filtered SST and wind stress fields in the longitude range1508–1008W and the latitude range 38N–18S on the north side of thecold tongue: (a) the perturbation wind stress divergence, = · t9, plot-ted as a function of the perturbation downwind SST gradient,

; (b) the perturbation wind stress curl, = 3 t 9 · k, plotted(=T · t )9as a function of the perturbation crosswind SST gradient, (=T 3

; and (c) histograms of the number of observations within eacht )9 · kbin for (a) (thick line) and (b) (thin line). The solid circles in (a) and(b) represent the overall mean values within each bin over the 3-monthdata record. The associated vertical bars represent the 61 standarddeviation of the mean values within each bin computed individuallyfor each 3-day period over the 3-month data record. These standarddeviations provide a measure of the uncertainty of the estimate ofthe overall mean in each bin. The smooth lines through the binnedmeans represent least squares fits of the binned overall means tostraight lines.

FIG. 11. The same as Fig. 10, except for the zonally high-passfiltered fields in the longitude range 1508–1008W and latitude range18–58S on the south side of the cold tongue.

south of the cold tongue. In contrast, the mean gradientof the wind stress magnitude is nearly symmetric aboutthe cold tongue. The ratio of |(= |t |)| to |=T| is thusabout twice as large on the south side of the cold tongue.The asymmetry in the mean SST gradient results in anasymmetry in the magnitudes of the SST gradients inthe perturbation SST fields; the mean perturbation SSTgradients |=T9| are about twice as large north of the coldtongue as they are south of the cold tongue (cf. Figs.6c and 7c). The ratio of the amplitude of the cosineangular dependence of = · t9 to the magnitude of theperturbation SST gradient |=T9| is therefore also abouttwice as large south of the cold tongue. This is a qual-itative independent verification of the approximate fac-tor of 2 larger sensitivity of the wind stress divergencesouth of the cold tongue.

The reason for the asymmetric response of the windstress divergence on opposite sides of the cold tongue

is not immediately apparent. One possible explanationis the asymmetric response of atmospheric turbulenceto stability changes in stable versus unstable regimes.The Richardson number and other dimensionless mea-sures of turbulent flow characteristics in the surface lay-er are observed to have a different functional depen-dence on atmospheric stability in stable and unstableregimes (see discussion by Kraus and Businger 1994).All other things being equal, a local increase in stabilityunder preexisting stable conditions will result in a great-er decrease in the magnitude of turbulent vertical trans-fers of horizontal momentum than an equivalent in-crease in stability under preexisting unstable conditions.A similar stability dependence of turbulent and con-vective momentum fluxes is expected above the surfacelayer as well.

Other local boundary layer effects that might con-tribute to asymmetry include a difference between theboundary layer depth north and south of the equator andthe difference in adjustment timescales due to turbu-lence and convection for stable and unstable boundarylayers. Large-scale dynamics may also contribute to theasymmetry of the atmospheric response to TIWs. Thechange in sign of the Coriolis force across the coldtongue introduces a fundamental asymmetry for cross-equatorial boundary layer flow. The large-scale adjust-ment processes in cross-equatorial flow have been ex-amined theoretically by Tomas et al. (1999) and othersassuming neutral static stability in the atmospheric


boundary layer. The strong coupling between surfacewind stress and SST suggests, however, that more re-alistic modeling of air–sea interaction processes is need-ed to model the asymmetric atmospheric response toTIW-induced SST perturbations.

The coupling of the perturbation wind stress curl fieldto the underlying SST perturbations is significantly dif-ferent from that of the perturbation wind stress diver-gence field. In both latitude bands, the binned meanvalues of the perturbation wind stress curl, = 3 t9 · kvary linearly with the perturbation crosswind SST gra-dient, (=T 3 (Figs. 10b and 11b), as expectedt)9 · kfrom the sine dependencies in Fig. 9. The slightly largersensitivity of the wind stress curl field to lateral SSTvariations south of the cold tongue than north of thecold tongue is at most only marginally significant, basedon the error estimates on the binned values in Figs. 10band 11b. However, the sensitivity of the wind stress curlto the crosswind SST gradient is significantly smallerthan the sensitivity of the wind stress divergence to thedownwind SST gradient (by slightly more than a factorof 2 south of the cold tongue and by slightly less thana factor of 2 north of the cold tongue).

The difference between the strength of the couplingof the wind stress divergence to the downwind SSTgradient and the coupling of the wind stress curl to thecrosswind SST gradient in the perturbation fields maybe attributable to the finite timescale for adjustment ofthe vertical atmospheric boundary layer structure as theair encounters changes of SST. For a convectively mixedlayer, the adjustment timescale is on the order of h/(w0

1 we), where h is the boundary layer thickness and w0

and we are velocity scales for thermodynamic fluxes atthe surface and mass entrainment at the top of theboundary layer, respectively (Betts 1983). For a surfacewind speed of 5 m s21, a boundary layer thickness of500 m and a boundary layer divergence of 5 3 1026

s21, the adjustment time is ;14 h, which correspondsto a horizontal displacement of ;250 km.

For the case of flow across an SST front, the boundarylayer must continually adjust to the changing SST inthe downwind direction. The vertical structure of theboundary layer therefore cannot reach equilibrium untilthe flow has passed over the SST front into a region ofrelatively weak downwind SST gradient. For air blow-ing from warm to cool water, the convective and tur-bulent fluxes would be less than their equilibrium valuesas the boundary layer adjusts to the cooler surfaceboundary condition. Assuming that convection trans-ports momentum toward the cool sea surface, the surfacestress over the increasingly cooler water would be lowerthan for the case of an equilibrium boundary layer pro-file. This would result in a horizontal convergence ofthe surface wind stress across the SST front with de-creasing effect downwind. For air blowing from cool towarm water, the turbulent and convective fluxes wouldbe enhanced relative to the equilibrium profile, thus re-

sulting in a stronger horizontal divergence of the surfacestress across the SST front.

The enhanced convergence and divergence of flowacross isotherms can be contrasted with the boundarylayer response for flow along an SST front. In this case,the weak downwind SST gradient along the trajectoryof a given air parcel may allow sufficient time for theboundary layer to come into equilibrium with the SST.As a consequence, the wind stress curl along the SSTfront may be weaker than for the nonequilibrated case.The finite timescale of boundary layer adjustment maytherefore explain the weaker response of wind stresscurl to the crosswind SST gradient compared with thewind stress divergence response to the downwind SSTgradient.

Another factor that could contribute to the differencebetween the coupling strength of wind stress divergenceand curl to the SST gradient field is the effect of oceansurface currents on the wind stress. The observationsof upper-ocean velocity by Kennan and Flament (2000)indicate that the surface velocity to the north of the SSTfront is approximately parallel to isotherms, thus form-ing anticyclonic vortices between successive TIWcusps. From the survey of a single vortex, they observedocean surface velocities of 0.5–1.0 m s21 near the SSTfront. Scatterometers measure wind stress relative to theocean surface velocity. Since the ocean surface velocityis approximately parallel to isotherms, the cross-iso-therm component of the wind stress is unaffected by theocean surface velocity. However, the along-isothermcomponent of the wind stress decreases when the along-wind component of ocean surface velocity is in the samedirection as the wind. The wind stress divergence andcurl respond to the cross-isotherm and along-isothermcomponents of the wind stress, respectively. The dis-tinction between the effects of ocean surface velocityon the cross-isotherm and along-isotherm componentsof the wind stress would therefore be expected to havedifferent effects on the wind stress divergence and curl.

Without direct observations of the ocean surface ve-locity coincident with the wind observations, it is notpossible to quantify the magnitude of the effect of oceansurface velocity on the wind stress. However, sinceTIW-induced ocean surface velocity perturbations of0.5–1.0 m s21 are a substantial fraction of the ;2 m s21

perturbations of the wind speeds in this region, it wouldnot be surprising to find that the wind stress curl re-sponse to the crosswind SST gradient is significantlyaffected by the ocean surface velocity. Because the windis in the same direction as the ocean surface velocityin the regions where TIW-induced wind stress curl per-turbations are large (see Fig. 4 and compare with theocean surface velocity field presented by Kennan andFlament 2000), the ocean surface velocity effect wouldreduce the wind stress curl response, as is observed fromthe QuikSCAT and TMI data.

The couplings between the derivative wind stressfields and the SST gradient field in Figs. 10 and 11 were


FIG. 12. The same as Fig. 10, except for the unfiltered SST andwind stress fields in the longitude range 1508–1008W and the latituderange 38N–18S on the north side of the cold tongue. The dashed linescorrespond to the coupling coefficients between the northern pertur-bation fields shown in Fig. 10.

FIG. 13. The same as Fig. 11, except for the unfiltered SST andwind stress fields in the longitude range 1508–1008W and the latituderange 18–58S on the south side of the cold tongue. The dashed linescorrespond to the coupling coefficients between the southern pertur-bation fields shown in Fig. 11.

deduced from the zonally high-pass filtered wind stressand SST fields. The applicability of these coupling co-efficients (the slopes of the least squares fit lines in Figs.10 and 11) to the unfiltered wind stress and SST fieldsis shown in Figs. 12 and 13. The dashed lines correspondto the least squares fit straight lines from Figs. 10 and11. The solid circles are the binned mean values of theunfiltered wind stress divergence (Figs. 12a and 13a)and unfiltered wind stress curl (Figs. 12b and 13b), plot-ted as functions of the unfiltered downwind and cross-wind SST gradients, respectively.

On the south side of the cold tongue (Fig. 13), it isapparent that the linear relationships deduced from thezonally high-pass filtered fields are closely reproducedin the unfiltered wind stress and SST fields. On the northside of the cold tongue, however, the agreement betweenthe binned mean values of the unfiltered wind stressdivergence and curl and the linear relationships deducedfrom the zonally high-pass filtered fields is not as good(Fig. 12). In the region of positive downwind SST gra-dients near the northern SST front, the coupling coef-ficient for the wind stress divergence deduced from theperturbation fields underestimates the wind stress di-vergence by roughly a constant value of about 0.3 Nm22 (104 km)21. Similarly, the coupling coefficient forthe wind stress curl overestimates the wind stress curlby roughly the same amount near the northern SSTfront.

The misfits of the predicted wind stress divergenceand curl responses to SST gradients in the unfilteredfields north of the cold tongue are likely an indicationof processes other than SST modification of boundarylayer stability that may affect the wind stress field. Forexample, this may be evidence of the importance ofnorthward acceleration from the SST-induced pressuregradient force on the north side of the cold tongue,which has been discussed by Lindzen and Nigam(1987), Mitchell and Wallace (1992), Young (1987), andTomas et al. (1999). These other processes are evidentlyeither of secondary importance in generating TIW-re-lated wind stress perturbations or have large zonal scaleand are therefore eliminated by the zonal high-pass fil-tering applied to highlight TIW-related variability inFigs. 10 and 11.

4. Westward propagation

The case study maps in Fig. 4 and the 3-month sta-tistics in Figs. 8–11 strongly suggest a phase lockingbetween the perturbed derivative wind stress fields (di-vergence and curl) and the perturbed underlying SSTfield as the TIWs propagate westward. This is verifiedin Figs. 14 and 15 from time–longitude plots along 18Nand 38S that bracket the cold tongue in the eastern trop-ical Pacific. Synchronized westward propagation is ev-ident in all of the quantities. Contours of the perturbation


FIG. 14. Time–longitude plots along 18N on the north side of the cold tongue showing zonallyhigh-pass filtered SST and wind stress fields: (a) perturbation SST; (b) perturbation wind stressmagnitude with contours of perturbation SST overlaid; (c) perturbation wind stress divergencewith contours of the perturbation downwind SST gradient, , overlaid; and (d) perturbation(=T · t )9wind stress curl with contours of the perturbation crosswind SST gradient, (=T 3 )9 · k, overlaid.tContour intervals are 0.58C in (b) and 0.48C (100 km)21 in (c) and (d). For clarity, zero contourshave been omitted from (b) to (d).

downwind and crosswind SST gradients are overlaid onthe color plots of perturbation wind stress divergenceand perturbation wind stress curl in Figs. 14c,d and15c,d. The phase relationships between the componentsof the SST gradient and the derivative wind stress fieldsare seen to be consistent across longitude and time withthe couplings deduced statistically from Figs. 10 and11.

Along the northern SST front (Fig. 14), the charac-teristics of the westward-propagating TIWs are verysimilar to those reported by Chelton et al. (2000) forthe 1998–99 TIW season. During the time period con-sidered here, the propagation was quite regular east ofabout 1408W with a westward phase speed of about 0.6m s21. This is somewhat faster than the 0.53 m s21

estimated by Chelton et al. (2000), but the difference


FIG. 15. The same as Fig. 14, except time–longitude plots of zonally high-pass filtered SSTand wind stress fields along 38S on the south side of the cold tongue. In order to reveal the weakersignals south of the equator, the contour intervals are half those used in Fig. 14.

could easily be explained by the larger uncertainty ofthe phase speed estimated from the short 3-month datarecord analyzed here or it could be due to small temporalvariations of the kinematic properties of the TIWs. Inthis regard, an interesting feature of Fig. 14 is the changein the westward phase speed west of about 1408W thatoccurred in late August. This is apparent in the SST,wind stress magnitude, and wind stress divergence fields(Figs. 14a–c), as well as in the contours of the downwindSST gradient overlaid on Fig. 14c. The wind stress curland crosswind SST gradient at these longitudes (Fig.

14d) were too weak for the change of phase speed tobe clearly seen.

Along the southern SST front (Fig. 15), the charac-teristics of westward propagation were very different.During the first month of the data record, the southernTIWs propagated westward at about the same 0.6 m s21

phase speed observed along 18N. In mid-August, thephase speed abruptly decreased to only about 0.25 ms21. The phase speed appears to have increased againin early October. These irregularities are surprising inview of the fact that Chelton et al. (2000) found ;0.5


m s21 westward propagation on both sides of the equa-torial cold tongue throughout the 1998–99 TIW season.The reason for the abrupt changes in the westward phasespeed along 38S during the 3-month period analyzedhere is not yet known. These features of the westwardpropagation are reproduced coherently in the pertur-bation SST, wind stress magnitude, wind stress diver-gence, and wind stress curl fields along the southernSST front. Moreover, as in the case of propagation alongthe northern SST front in Fig. 14, the contours of theperturbation downwind and crosswind SST gradients inFigs. 15c and 15d, respectively, are phase locked withthe perturbation wind stress divergence and curl in themanner deduced statistically from Figs. 10 and 11.

The close coupling between the wind stress field andthe underlying SST field along 38S is further illustratedby the splitting of the westward-propagating SST signalinto two separate warm perturbations from mid-Augustto mid-September between 1208 and 1308W. This split-ting is also apparent in the perturbation wind stress di-vergence field and in the associated perturbation down-wind SST gradient (Fig. 15c). Interestingly, the splittingis not clearly evident in the perturbation wind stress curlfield in Fig. 15d. However, consistent with the statisticalrelation in Fig. 11b, the splitting is also not clearlyevident in the associated perturbation crosswind SSTgradient.

The distinctly different characteristics of the west-ward propagation along 18N and 38S are intriguing. The3-month data record analyzed here is too short to ex-amine this in detail. This can be investigated further asthe QuikSCAT and TMI data records continue to ac-cumulate so that the westward propagation of the SSTsignatures of TIWs can be compared and contrasted forthe 1998–99 and 1999–2000 TIW seasons.

5. Discussion and conclusions

The satellite microwave measurements of surfacewind stress and SST during the 3-month period 21 July–20 October 1999 analyzed here provide definitive ob-servational evidence of ocean–atmosphere coupling inthe eastern tropical Pacific. The southeastern Pacifictrade winds blow northwestward toward the oceanicequatorial cold tongue. The atmospheric boundary layeris stabilized by the cooler water. As first hypothesizedby Wallace et al. (1989) from analysis of climatologicalaverage wind and SST fields, this decouples the surfacewinds from the winds aloft and decreases the surfacewinds, thus resulting in a convergence of the surfacewind stress. As the low-level winds cross isotherms onthe north side of the cold tongue, the atmosphericboundary layer is destabilized over the warmer waterand the enhanced turbulent mixing of momentum fromaloft increases the surface winds, resulting in a diver-gence of the surface wind stress.

TIW-induced perturbations of the SST fronts thatbracket the equatorial cold tongue allow a detailed in-

vestigation of the structure of the SST-induced modi-fication of the surface wind stress and its dependenceon the orientation of the wind stress relative to pertur-bations of the SST gradient vector. The perturbationwind stress divergence varies linearly with the pertur-bation downwind component of the SST gradient. Con-vergence and divergence are thus locally strongestwhere the wind blows parallel to the SST gradient (per-pendicular to isotherms). The perturbation wind stresscurl varies linearly with the perturbation crosswind com-ponent of the SST gradient. In regions where the anglebetween the wind direction and the SST gradient isoblique, a wind stress curl thus develops from lateralvariations of the SST-induced changes of the surfacewind stress. The wind stress curl is locally strongestwhere the winds blow perpendicular to the SST gradient(parallel to isotherms).

The observed couplings of the derivative wind stressfields to the perturbation downwind and crosswind SSTgradients are completely consistent with the SST-in-duced modifications of atmospheric stability hypothe-sized by Wallace et al. (1989). Other dynamical mech-anisms such as pressure gradient forcing as suggestedby Lindzen and Nigam (1987), Mitchell and Wallace(1992), Young (1987), and Tomas et al. (1999) are ev-idently of secondary importance, at least in terms ofSST effects on surface winds on the time- and spacescales of TIWs. On larger scales, the pressure gradientforcing may be more important (see Figs. 12 and 13).

The apparent strong and rapid atmospheric responseto TIW-induced perturbations of the SST field suggeststhat ocean–atmosphere coupling plays an important rolein the thermodynamics of TIWs. The SST perturbationsmodify the atmospheric boundary layer in such a wayas to induce negative thermodynamic feedback on theTIWs. Since the perturbed wind stress and SST fieldspropagate westward in synchronization with the TIWs,the high winds over warm water between successivecold cusps result in continuous upper-ocean heat lossthrough surface evaporation (Zhang and McPhaden1995). In addition, the enhanced vertical mixing overthe warmer water generates increased low-level cloud-iness that reduces the solar insolation that reaches thesea surface (Deser et al. 1993). It has been estimatedthat the cooling by the combined effects of these twocomponents of ocean–atmosphere heat flux may be aslarge as ;0.58C per month (Deser et al. 1993; Mc-Phaden 1996).

The high space–time resolution afforded by Quik-SCAT measurements of surface wind stress also revealsa strong dynamical feedback on the TIWs. The orien-tation of the wind stress relative to the isotherms alongthe cuspy northern SST front generates very strong pos-itive perturbations of the wind stress curl that drives theocean circulation. There is also a broad band of negativewind stress curl south of the equator with local inten-sification in regions where the wind blows parallel toTIW-induced perturbations of the weaker southern SST


FIG. 16. Three-month averages (21 July–20 October 1999) over 48of longitude by 28 of latitude constructed from QuikSCAT measure-ments of wind stress: (a) wind stress divergence; and (b) wind stresscurl. Contours correspond to isotherms from Fig. 1a at intervals of18C between 218 and 278C.

front. A zonal band of strong wind stress divergencejust north of the equator is also noteworthy. Without atheoretical analysis, which is beyond the scope of thisstudy, the nature of the feedback of these wind stresscurl and divergence patterns on TIWs is less clear thanthe simple negative thermodynamic feedback discussedabove. Given the intensity of the wind stress curl anddivergence perturbations and the fact that the locationsof strongest wind stress curl and divergence propagatewest in synchronization with the TIWs, there can be nodoubt that these perturbation wind forcings must be tak-en into consideration in determining the dynamics andenergetics of TIWs, and therefore in assessing the im-pact of TIWs on the large-scale momentum and heatbudgets of the eastern tropical Pacific.

The ocean–atmosphere coupling demonstrated here isalso important on scales much larger than those asso-ciated with TIW variability. Large-scale ocean circu-

lation outside of the equatorial waveguide is driven bythe wind stress curl. The broad band of negative windstress curl on the south side of the cold tongue, as wellas the narrow band of positive wind stress curl just northof the equator in the 3-month average shown in Fig.16b are direct responses of the wind stress to the un-derlying SST field. The southern trade winds blow ap-proximately parallel to the isotherms in the broad time-averaged southern SST front (Fig. 1), resulting in thebroad band of strong wind stress curl south of the coldtongue. Likewise, the wind stress blows at an obliqueangle across the isotherms along the narrow northernSST front, resulting in the narrow bands of positive windstress curl and divergence along the time-averaged SSTfront on the north side of the cold tongue. The zonalband of positive wind stress curl is definitive evidencefor the importance of the Wallace et al. (1989) mech-anism for SST modification of the surface wind field;the pressure gradient mechanism suggested by Lindzenand Nigam (1987) and others could not produce thelateral shear of the wind necessary to create this windstress curl.

To the north of about 38N, the band of convergencein the ITCZ and the bands of negative and positive windstress curl on the south and north sides of the ITCZ,respectively, (see Fig. 16b) are evidently associated withprocesses distinct from the SST modification of atmo-spheric boundary layer stability considered in this study.However, these features of the wind stress curl field areimportant to wind-forced ocean circulation in the Trop-ics. The intensities of the banded structures of the windstress curl and divergence fields are considerably un-derestimated and their narrow meridional scales arepoorly resolved in wind stress curl and divergence fieldsconstructed from all other sources of surface wind in-formation (e.g., Rienecker et al. 1996).

The meteorological significance of the SST-inducedmodifications of the surface wind field on deeper at-mospheric circulations in the vicinities of the coldtongue and ITCZ is less clear without measurements ofthe winds aloft. The analysis of Wallace et al. (1989)and the results of this study suggest that the band ofwind stress divergence north of the cold tongue and theassociated band of wind stress convergence south of thecold tongue in the 3-month average shown in Fig. 16aare direct responses to SST-induced modification of thesurface wind field. Deser and Wallace (1990) and Mitch-ell and Wallace (1992) have noted that the intensitiesof the equatorial cold tongue and cross-equatorial windsin the band between the equator and about 88N are high-ly correlated with variations in the intensity and latitudeof the ITCZ. It is thus possible that the SST-inducedsurface wind stress features are part of the deep merid-ional overturning of the Hadley-type circulation overthe eastern Pacific.

Alternatively, the impacts of the surface divergencefield implied by the QuikSCAT data may be confinedto the atmospheric boundary layer. The modeling results


FIG. 17. Three-month averages (21 July–20 October 1999) over 48of longitude by 28 of latitude constructed from QuikSCAT obser-vations: (a) the zonal component of wind stress; and (b) the meridionalcomponent of wind stress. Contour interval is 0.01 N m22 and neg-ative contours are shown as dashed lines in both panels.

of Xie et al. (1998) and a recent analysis of QuikSCAT,TMI, and SSM/I data by Hashizume et al. (2001) sug-gest that the TIW-induced perturbations may penetrateseveral kilometers above the sea surface and may evenweakly perturb the ITCZ. The intensity of the zonalbands of divergence and convergence in Fig. 16a war-rant further modeling studies and an in situ observa-tional field program to investigate the vertical structureof the wind field in order to quantify the relationshipbetween the SST-induced modification of the atmo-spheric boundary layer and the atmospheric circulationaloft.

The observed geographical structure of the surfacewind stress field (Fig. 1b) raises important questionsabout coupled ocean–atmosphere interaction in the east-ern tropical Pacific. Equatorial upwelling from Ekmandivergence at the equator forced by the zonal componentof wind stress that is generally believed to play a rolein the existence of cold water along the equator cannotexplain the observed equatorial asymmetry and merid-ional scale of the cold tongue. Indeed, between the SouthAmerican coastline and about 1208W, it is evident from

Fig. 17a that the zonal component of wind stress is veryweak within the equatorial waveguide.

Mitchell and Wallace (1992) found that seasonal var-iations of the intensity of the equatorial cold tongue aremuch more highly correlated with the meridional windstress than with the zonal wind stress. They hypothe-sized that this is because of a positive feedback in whichthe northern SST front accelerates the winds north ofthe cold tongue and that these northward winds generateupwelling of cold water just south of the equator, therebysustaining the cold tongue that causes the winds to ac-celerate across the SST front. There is some modelingevidence to support this speculation. Cane (1979) andPhilander (1990, section 4.8.2) showed that the oceanresponse to uniform northward wind generates upwell-ing south of the equator, which may explain why thecold tongue is centered south of the equator. Philanderand Pacanowski (1981) suggested that this upwellingoriginates as coastal upwelling along the eastern bound-ary at low latitudes in the Southern Hemisphere andthen spreads rapidly westward by a combination of ad-vection and Rossby wave propagation.

It is noteworthy that the numerical studies by Cane(1979) and Philander and Pacanowski (1981) both con-sidered the ocean response to the sudden onset of steady,uniform meridional winds. It is evident from Fig. 1bthat this simple pattern of spatially uniform northwardwind stress is a highly idealized representation of theactual wind field. The nonuniformity of the meridionalwind stress in the QuikSCAT observations is more clear-ly seen in Fig. 17b. The SST-induced modification ofthe overlying wind stress results in more than a factorof 2 reduction of the northward component of windstress over the equatorial cold tongue. North of the coldtongue, the northward component of wind stress in-creases by more than a factor of 3. A realistic simulationof the cold tongue response to wind forcing must takeinto consideration this large latitudinal variation of themeridional wind stress field.

The QuikSCAT observations also clearly show thevery inhomogeneous structure of the zonal componentof the wind stress field (Fig. 17a). East of about 1308W,the zonal wind stress is very weak over the cold tongueas noted above and is highly asymmetric about the equa-tor. On the south side of the cold tongue, maximumeasterly winds occur at about 128S. The easterly com-ponent of wind stress decreases to a local minimum overthe cold tongue and then increases slightly just north ofthe cold tongue before decreasing again and then turningwesterly at higher latitudes. The maximum westerlywinds occur at about 88N.

The combined patterns of the zonal and meridionalcomponents of the wind stress in Fig. 17 create thebanded structure of the wind stress curl and divergencefields discussed above (Fig. 16). As noted previously,the observed structures of the wind stress curl and di-vergence fields are poorly represented in all other sourc-es of surface wind information. The intensities of the


wind stress curl and divergence patterns strongly sug-gest that both components of the surface wind stressfield must be taken into consideration when investigat-ing the wind-driven circulation throughout the easterntropical Pacific (even close to the equator). There isclearly a need for additional modeling studies that con-sider the eastern tropical Pacific ocean response to morerealistic wind forcing, including the effects of the windstress curl and divergence.

The ocean–atmosphere interaction demonstrated fromthe QuikSCAT and TMI data also emphasizes the ne-cessity of including stability-dependent boundary layereffects for realistic coupled ocean–atmosphere modelingof the eastern tropical Pacific. This presents challengesfor simplified theoretical models of the coupled system.It is possible that the effects of these boundary layerprocesses are not adequately represented even in com-prehensive coupled general circulation models. Mecho-so et al. (1995) have shown that these models are notable to reproduce the observed climatological structureof eastern Pacific SST patterns. The equatorial coldtongues in these models tend to be too strong, too re-stricted latitudinally, and extend too far to the west. Thewarm pool region to the north of the cold tongue ispoorly simulated and varies significantly between dif-ferent models. In addition, the simulated temperaturessouth of the cold tongue tend to be warmer than isobserved. These systematic errors may be due in partto errors in the wind stress, wind stress divergence, andwind stress curl fields owing to inadequate represen-tation of the strong ocean–atmosphere coupling revealedby the QuikSCAT observations of wind stress and TMIobservations of SST.

Acknowledgments. We thank Greg Johnson andShang-Ping Xie for helpful discussions and commentson the manuscript. The research presented in this paperwas supported by Contracts 957580 and 959351 fromthe Jet Propulsion Laboratory funded under the NSCATAnnouncement of Opportunity, NASA Grant NAS5-32965 for funding of Ocean Vector Winds Science Teamactivities, NOAA Grant NA96GP0368 from the Officeof Global Programs, NASA TRMM Contract NAS5-9919, NASA’s Earth Science Information Partnershipthrough Contract SUB1998-101 from the University ofAlabama in Huntsville and NOAA’s Office of Oceanicand Atmospheric Research.

REFERENCES

Baturin, N. G., and P. P. Niiler, 1997: Effects of instability waves inthe mixed layer of the equatorial Pacific. J. Geophys. Res., 102,27 771–27 793.

Betts, A. K., 1983: Thermodynamics of mixed stratocumulus layers:Saturation point budgets. J. Atmos. Sci., 40, 2655–2670.

Bond, N. A., 1992: Observations of planetary boundary layer struc-ture in the eastern equatorial Pacific. J. Climate, 5, 699–706.

Bryden, H., and E. C. Brady, 1989: Eddy momentum and heat fluxes

and their effects on the circulation of the equatorial PacificOcean. J. Mar. Res., 47, 55–79.

Cane, M. A., 1979: The response of an equatorial ocean to simplewind stress patterns: II. Numerical results. J. Mar. Res., 37, 253–299.

Chelton, D. B., F. J. Wentz, C. L. Gentemann, R. A. de Szoeke, andM. G. Schlax, 2000: Microwave SST observations of transe-quatorial tropical instability waves. Geophys. Res. Lett., 27,1239–1242.

Cox, M., 1980: Generation and propagation of 30-day waves in anumerical model of the Pacific. J. Phys. Oceanogr., 10, 1168–1186.

Deser, C., and J. M. Wallace, 1990: Large-scale atmospheric circu-lation features of warm and cold episodes in the tropical Pacific.J. Climate, 3, 1254–1281., J. J. Bates, and S. Wahl, 1993: The influence of sea surfacetemperature gradients on stratiform cloudiness along the equa-torial front in the Pacific Ocean. J. Climate, 6, 1172–1180.

Dickinson, S., K. A. Kelly, M. J. Caruso, and M. J. McPhaden, 2001:A note on comparisons between the TAO buoy and NASA scat-terometer wind vectors. J. Atmos. Oceanic Technol., in press.

Fairall, C. W., E. F. Bradley, D. P. Rogers, J. B. Edson, and G. S.Young, 1996: Bulk parameterization of air–sea fluxes for Trop-ical Ocean-Global Atmosphere Coupled-Ocean Atmosphere Re-sponse Experiment. J. Geophys. Res., 101, 3747–3764.

Flament, P. J., S. C. Kennan, R. A. Knox, P. P. Niiler, and R. L.Bernstein, 1996: The three-dimensional structure of an upperocean vortex in the tropical Pacific Ocean. Nature, 383, 610–613.

Freilich, M. H., and R. S. Dunbar, 1999: The accuracy of the NSCAT-1 vector winds: Comparisons with National Data Buoy Centerbuoys. J. Geophys. Res., 104, 11 231–11 246., D. G. Long, and M. W. Spencer, 1994: SeaWinds: A scanningscatterometer for ADEOS II—Science overview. Proc. Int.Geosci. Remote Sens. Symp., Vol. II, Pasadena, CA, IEEE No.94CH3378-7, 960–963.

Goldenberg, S. B., and J. J. O’Brien, 1981: Time and space variabilityof tropical Pacific wind stress. Mon. Wea. Rev., 109, 1190–1207.

Grassl, H., 1976: The dependence of the measured cool skin of theocean on wind stress and total heat flux. Bound.-Layer Meteor.,10, 465–474.

Hansen, D. V., and C. A. Paul, 1984: Genesis and effects of longwaves in the equatorial Pacific. J. Geophys. Res., 89, 10 431–10 440.

Hashizume, H., S.-P. Xie, W. T. Liu, and K. Takeuchi, 2001: Localand remote atmospheric response to tropical instability waves:A global view from space. J. Geophys. Res., in press.

Hayes, S. P., M. J. McPhaden, and J. M. Wallace, 1989: The influenceof sea-surface temperature on surface wind in the eastern equa-torial Pacific: Weekly to monthly variability. J. Climate, 2, 1500–1506.

Kennan, S. C., and P. J. Flament, 2000: Observations of a tropicalinstability vortex. J. Phys. Oceanogr., 30, 2277–2301.

Kraus, E. B., and J. A. Businger, 1994: Atmosphere–Ocean Inter-action. Oxford University Press, 362 pp.

Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998:The Tropical Rainfall Measuring Mission (TRMM) sensor pack-age. J. Atmos. Oceanic Technol., 15, 808–816.

Large, W. G., and S. Pond, 1981: Open ocean momentum flux mea-surements in moderate to strong winds. J. Phys. Oceanogr., 11,324–336.

Legeckis, R., 1977: Long waves in the eastern equatorial PacificOcean: A view from a geostationary satellite. Science, 197,1179–1181.

Lindzen, R. S., and S. Nigam, 1987: On the role of sea surfacetemperature gradients in forcing low-level winds and conver-gence in the tropics. J. Atmos. Sci., 44, 2418–2436.

Luther, D. S., and E. S. Johnson, 1990: Eddy energetics in the upperequatorial Pacific during the Hawaii-to-Tahiti Shuttle Experi-ment. J. Phys. Oceanogr., 20, 913–944.


McPhaden, M. J., 1996: Monthly period oscillations in the PacificNorth Equatorial Countercurrent. J. Geophys. Res., 101, 6337–6359., and Coauthors, 1998: The Tropical Ocean–Global Atmosphereobserving system: A decade of progress. J. Geophys. Res., 103,14 169–14 240.

Mechoso, C. R., and Coauthors, 1995: The seasonal cycle over thetropical Pacific in coupled ocean–atmosphere general circulationmodels. Mon. Wea. Rev., 123, 2825–2838.

Miller, L., D. R. Watts, and M. Wimbush, 1985: Oscillations of dy-namic topography in the eastern equatorial Pacific. J. Phys.Oceanogr., 15, 1759–1770.

Mitchell, T. P., and J. M. Wallace, 1992: The annual cycle in equatorialconvection and sea surface temperature. J. Climate, 5, 1140–1156.

Naderi, F. M., M. H. Freilich, and D. G. Long, 1991: Spaceborneradar measurement of wind velocity over the ocean—An over-view of the NSCAT scatterometer system. Proc. IEEE, 97, 850–866.

Philander, S. G. H., 1978: Instabilities of zonal equatorial currents,Part 2. J. Geophys. Res., 83, 3679–3682., 1990: El Nino, La Nina, and the Southern Oscillation. Aca-demic Press, 289 pp., and R. C. Pacanowski, 1981: The ocean response to cross-equatorial winds (with application to coastal upwelling in lowlatitudes). Tellus, 33, 201–210.

Qiao, L., and R. H. Weisberg, 1995: Tropical instability wave ki-nematics: Observations from the Tropical Instability Wave Ex-periment. J. Geophys. Res., 100, 8677–8693.

Reverdin, G., C. Frankignoul, E. Kestenare, and M. J. McPhaden,1994: Seasonal variability in the surface currents of the equa-torial Pacific. J. Geophys. Res., 99, 20 323–20 344.

Reynolds, R. W., K. Arpe, C. Gordon, S. P. Hayes, A. Leetmaa, andM. J. McPhaden, 1989: A comparison of tropical Pacific surfacewind analyses. J. Climate, 2, 105–111.

Rienecker, M. M., R. Atlas, S. D. Schubert, and C. S. Willett, 1996:A comparison of surface wind products over the North PacificOcean. J. Geophys. Res., 101, 1011–1023.

Saunders, P. M., 1967: The temperature at the ocean–air interface. J.Atmos. Sci., 24, 269–273.

Schlax, M. G., D. B. Chelton, and M. H. Freilich, 2001: Samplingerrors in wind fields constructed from single and tandem scat-terometer datasets. J. Atmos. Oceanic Technol., in press.

Swenson, M. S., and D. V. Hansen, 1999: Tropical Pacific Oceanmixed layer heat budget: The Pacific cold tongue. J. Phys.Oceanogr., 29, 69–81.

Tomas, R. A., J. R. Holton, and P. J. Webster, 1999: The influenceof cross-equatorial pressure gradients on the location of near-equatorial convection. Quart. J. Roy. Meteor. Soc., 125, 1107–1127.

Wallace, J. M., T. P. Mitchell, and C. Deser, 1989: The influence ofsea surface temperature on surface wind in the eastern equatorialPacific: Seasonal and interannual variability. J. Climate, 2, 1492–1499.

Wang, W., and M. J. McPhaden, 1999: The surface-layer heat balancein the equatorial Pacific Ocean. Part I: The mean seasonal cycle.J. Phys. Oceanogr., 29, 1812–1831.

Weidman, P. D., D. L. Mickler, B. Dayyani, and G. H. Born, 1999:Analysis of Legeckis eddies in the near-equatorial Pacific. J.Geophys. Res., 104, 7865–7887.

Wentz, F. J., 1997: A well-calibrated ocean algorithm for specialsensor microwave/imager. J. Geophys. Res., 102, 8703–8718., 1998: Algorithm theoretical basis document: AMSR OceanAlgorithm. Tech. Rep. 110398, Remote Sensing Systems, SantaRosa, CA, 65 pp. [Available online at http://www.remss.com.], and R. W. Spencer, 1998: SSM/I rain retrievals within a unifiedall-weather ocean algorithm. J. Atmos. Sci., 55, 1613–1627.

Xie, S.-P., M. Ishiwatari, H. Hashizume, and K. Takeuchi, 1998:Coupled ocean–atmospheric waves on the equatorial front. Geo-phys. Res. Lett., 25, 3863–3866.

Young, J. A., 1987: Boundary layer dynamics of tropical and mon-soonal flows. Monsoon Meteorology, C.-P. Chang and T. N.Krishnamurti, Eds., Oxford University Press, 461–500.

Zhang, G. J., and M. J. McPhaden, 1995: The relationship betweensea surface temperature and latent heat flux in the equatorialPacific. J. Climate, 8, 589–605.

observations of coupling between surface wind stress and sea...

Documents