MARCH 2001 839J O H N S O N E T A L .

q 2001 American Meteorological Society

Equatorial Pacific Ocean Horizontal Velocity, Divergence, and Upwelling*

GREGORY C. JOHNSON AND MICHAEL J. MCPHADEN

NOAA/Pacific Marine Environmental Laboratory, Seattle, Washington

ERIC FIRING

Department of Oceanography, University of Hawaii, Honolulu, Hawaii

(Manuscript received 12 January 2000, in final form 27 June 2000)

ABSTRACT

Upper-ocean horizontal velocity and divergence were estimated from shipboard observations taken from 1991to 1999 in the equatorial Pacific between 1708W and 958W. Mean transports were estimated for the zonal currentsat the mean longitude of the sections, 1368W. Mean meridional currents for the entire longitude range includedpoleward surface flows reaching 20.09 m s21 in the south and 0.13 m s21 in the north as well as equatorwardflow within the thermocline reaching 0.05 m s21 in the south and 20.04 m s21 in the north near 238C (85 m).Vertical velocity was diagnosed by integrating horizontal divergence estimated for the entire region down fromthe surface. Equatorial upwelling velocities peaked at 1.9 (60.9) 3 1025 m s21 at 50 m. The upwelling transportin the area bounded by 3.68S–5.28N, 1708W–958W was 62 (618) 3 106 m3 s21 at 50 m. Strong downwellingwas apparent within the North Equatorial Countercurrent. An asymmetry in the meridional flows suggested thaton the order of 10 3 106 m3 s21 of thermocline water from the Southern Hemisphere was upwelled at the equatorand moved into the Northern Hemisphere as surface water. This interhemispheric exchange path could be partof the route for water from the Southern Hemisphere to supply the Indonesian Throughflow.

1. Introduction

The general circulation of the equatorial PacificOcean is characterized by strong wind-driven zonalflows and a complex meridional circulation involvingshallow overturning cells (Lu et al. 1998). There is along history of describing and analyzing the intensezonal currents, both their means (e.g., Wyrtki and Kil-onsky 1984, hereafter WK) and their spatiotemporalvariability (e.g., Taft and Kessler 1991). The weakermeridional circulation and upwelling within a few de-grees of the equator (e.g., Wyrtki 1981) are not onlyimportant elements of the general circulation, but theyalso exert a strong control on global climate and bio-geochemical cycles (e.g., Philander 1990; Chavez et al.1999). However, being weaker, these elements of thecirculation are subject to aliasing over a wide varietyof time and space scales, and are much more difficultto observe than the zonal currents.

* Pacific Marine Environmental Laboratory Contribution Number2181.

Corresponding author address: Dr. Gregory C. Johnson, NOAA/Pacific Marine Environmental Laboratory, 7600 Sand Point Way N.E.,Bldg. 3, Seattle, WA 98115-0070.E-mail: gjohnson@pmel.noaa.gov.

Equatorial upwelling likely has a magnitude of a me-ter per day (;1 3 1025 m s21). While this value is largecompared with most of the rest of the ocean, it is stillmuch too small to measure directly. This inability tomake direct measurements of upwelling has significantconsequences. For example, in equatorial upper-oceanheat budgets, terms related to upwelling and verticalmixing must be estimated as residuals. These residualscan be as large as or larger than any of the other termsin the budget (cf. Wang and McPhaden 1999).

As a result, equatorial upwelling has been estimatedin a number of ways over a number of spatial and tem-poral scales. Weisberg and Qiao (2000, hereafter WQ)provide a recent review. In the Pacific Ocean chemicaltracers have been used with mixing models to estimateupwelling transports (Quay et al. 1983) and to deduceupwelling sources (Fine et al. 1983). Large-scale cal-culations assuming Ekman and geostrophic dynamicshave been made to diagnose upwelling (Wyrtki 1981;Bryden and Brady 1985). Surface drifter velocities havebeen used to estimate meridional divergence and inferupwelling (Poulain 1993). Local direct estimates of hor-izontal divergence from moored current meter arrayshave been integrated vertically to estimate vertical ve-locity on the equator by continuity (Halpern et al. 1989;WQ). A similar approach has also been applied overlarger scales using 27 repeat shipboard acoustic Doppler

840 VOLUME 31J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y

FIG. 1. Data distribution: (a) CTD/ADCP section longitudes and times; each plus represents a section. (b) Number of sections occupyingeach latitude and depth for ADCP velocities (solid contours) and CTD temperature and salinity (dashed contours); contour intervals 10.

current profiler (ADCP) sections from the Hawaii-to-Tahiti Shuttle experiment (Johnson and Luther 1994).These data also gave direct observations of the off-equatorial meridional velocity field and allowed off-equatorial vertical velocity estimates.

Here shipboard conductivity–temperature–depth(CTD) and ADCP data from 85 meridional sections tak-en across the equator from 1991 to 1999 between 1708Wand 958W were used to make meridionally localizedestimates of the zonal flow, and large-scale estimates ofthe meridional flow and horizontal divergence in theupper tropical Pacific Ocean. This work closely fol-lowed that of Johnson and Luther (1994). However, anindependent dataset was used with three times the num-ber of ADCP sections, much lower instrument error, andgreater latitude, longitude, and depth ranges. The resultswere more certain estimates over a larger area. Up-welling was calculated from continuity to generate anupper-ocean meridional–vertical section of vertical ve-locity for the region. Section 2 describes the data andhow the fields and their errors were estimated. Section3 discusses zonal velocity, temperature, salinity, andzonal currents, all estimated locally near the central lon-gitude of the domain. Section 4 discusses meridionalvelocity, both components of the large-scale horizontaldivergence, and the vertical velocity field (hence theupwelling transport) inferred from integration of thisdivergence. Section 5 concludes the paper by discussingthe results and comparing them to previous work.

2. Data, methods, and errors

The 85 meridional CTD/ADCP sections used weretaken from June 1991 through December 1999 (Fig. 1a).Of these sections, 73 were taken during Tropical At-mosphere–Ocean (TAO) buoy array maintenance cruis-es. The TAO cruises were on the NOAA Ship Discov-erer in the first half of the decade (Johnson and Plimpton1999) and the NOAA Ships Ka’imimoana and RonaldH. Brown for the second half of the decade. A subset

of these sections has been used to study the 1997–98El Nino (Johnson et al. 2000). Another seven sectionswere from World Ocean Circulation Experiment one-time hydrographic survey cruises on several differentships. The remaining five sections were from U.S. JointGlobal Ocean Flux Study Equatorial Pacific ProcessStudy (Murray et al. 1995) cruises on the R/V ThomasG. Thompson.

The individual ADCP and CTD sections were griddedfollowing Johnson et al. (2000). The ADCP velocitieswere objectively mapped assuming a Gaussian covari-ance, meridional correlation length scales of 18, verticalcorrelation length scales of 25 m, and a noise-to-signalenergy ratio of 0.01. The CTD temperature, salinity, andpressure data were splined in latitude on isopycnals.These gridded sections were used in all subsequent anal-ysis.

TAO cruises occupied meridional sections nominallyfrom 88S to 88N at 158 longitude intervals between1708W and 958W. At 1408W the nominal latitude rangewas from 58S to 98N. Two adjacent sections were oftenoccupied during a single cruise, one steaming southwardacross the equator and another steaming northwardacross the equator at least a week later. The resulting7-day temporal and 1660-km spatial separation wereassumed sufficient for each section to be independent,at least in terms of the energetic high-frequency oceanvariability (including internal waves, equatorial Kelvinwaves, Rossby waves, and tropical instability waves) inthe region. This assumption probably holds best for themeridional velocity data, where a very small mean sig-nal is often overwhelmed by large transients such astropical instability waves (Flament et al. 1996). How-ever, even for zonal velocity these separations would besufficient to sample different phases of an equatorialKelvin wave (if the eastern section of the pair wereoccupied first). The data did not always span the entirelatitude and depth range studied (Fig. 1b), with less CTDand ADCP data off the equator and less ADCP databelow 250 m.

MARCH 2001 841J O H N S O N E T A L .

Instrumental errors were thought to be negligiblecompared with the ocean variability mentioned above.The CTD accuracies were near 1% for pressure, 0.0028Cfor temperature, and 0.003 PSS-78 for salinity. Velocityaccuracies for the ADCP were near 0.05 m s21. Theseerrors would have been mostly in the cross-track com-ponent (the zonal velocity for these meridional sec-tions), and generally not systematic over an entire sec-tion. These quasi-random errors were assumed very un-likely to lead to systematic errors in currents or theirdivergences when averaged over the large number ofcruises and sections involved.

The spatiotemporal distribution of sections (Fig. 1a)had some biases. From 11 to 13 sections were occupiedin each of the years 1992 and 1997–99, with 6–9 sec-tions in each of the other 4.5 years. For June 1991–December 1999 the monthly Southern Oscillation index(SOI) mean and standard deviation were 20.5 6 0.9,slightly favoring El Nino conditions. The same statisticsheld for the SOI interpolated to section times. The sam-pling favored the western longitudes. The mean longi-tude of the sections, 1368W, was slightly west of thecentral longitude of the domain, 1328W. Finally, thesampling was somewhat biased toward boreal springand especially biased toward boreal fall. The most un-even 3-month groupings included only 13 sections inDecember–February and 11 in June–August, but 22 inMarch–May and 39 in September–November.

The effects of these sampling biases were exploredin the context of the zonal wind stress. The Florida StateUniversity surface wind analyses (Stricherz et al. 1992)were interpolated to section locations and times. From88S to 108N, generally under the easterly trade windsover the longitude range in question, the zonal meansof the section-interpolated zonal wind stresses wereroughly 0.9 times the 1961–90 zonal means over theentire longitude range. It seems that the effects of sam-pling favoring El Nino, which should tend to decreasethe trade winds, slightly more than counterbalanced thesampling favoring the boreal fall and the west, whichshould tend to increase the trade winds. Weaker tradewinds would generally tend to reduce the strength ofthe South Equatorial Current (SEC) and Equatorial Un-dercurrent (EUC), as well as the surface Ekman diver-gence, thermocline geostrophic convergence, and up-welling.

Estimates of zonal velocity, u; temperature, T; andsalinity, S, were made at the mean longitude of the sec-tions, 1368W. To make these estimates data from allsections were fit to a third-order polynomial function oflongitude at each latitude and depth. This proceduresmoothed the fields, while still allowing sufficient de-grees of freedom to model features such as the EUCmaximum in the center of the region, the eastward shoal-ing of the thermocline, and the like. Additions of anannual harmonic and/or linear SOI response to the re-gression were tested, but were left out to avoid intro-ducing noise into the means. For the most part, these

fits were only used near the center of the region, wherethey were best determined.

Estimates of meridional velocity y , both componentsof horizontal divergence, and vertical velocity w weremade only for the entire region encompassed by thesections. This strategy was necessitated by the fact thatthe small mean y was highly aliased by very energetictransient features. In fact, y was so noisy that attemptsto regress against various combinations of longitude,the annual cycle, and the SOI (with the possible ex-ception of the last) were all judged to have simply in-troduced noise into the resulting mean field. Thus, yand its meridional derivative, y y, had to be estimateddisregarding the longitudinal and temporal variationsanticipated in those fields. At each depth and latitude,all the y data within a 28 lat window were fit linearlyversus latitude. These fits were evaluated at the centersof the window to determine y , and the fit slopes wereused for y y. The horizontal divergence is the sum of y y

and ux, the zonal derivative of the zonal velocity. Sincey y had to be estimated uniformly over the entire lon-gitude range, for consistency ux was also estimated overthe entire longitude range. Slopes of linear fits versuslongitude to all the u data at each depth and latitudewere used for ux. This method gave results very similarto a difference of the mean zonal velocities at 958 and1708W, but used all the data.

Near-surface velocity and divergence estimates merita caveat. The shallowest data from the shipboard ADCPwere at 20 m, but the gridded velocities were objectivelymapped to 5 m. Velocities above 20 m were essentiallylinearly extrapolated by the mapping, mostly using ver-tical shear from 20 to 30 m, but with some influencefrom as deep as 50 m. Where the mixed layer was ab-sent, or deep enough to exceed 30 m, the mapping likelyproduced reasonable results. However, where the ther-mocline was strong and the mixed layer was shallowerthan 30 m the extrapolation may have been problematicbecause of shear changes at the mixed layer base. Thus,the thermocline ridge around 108N between the NorthEquatorial Countercurrent (NECC) and the North Equa-torial Current was a suspect region.

A check on the ADCP velocities mapped to 15 mwas allowed by velocities from drifter data between 1508and 1308W (Baturin and Niiler 1997). Both componentsof the mean mapped ADCP velocities at 15 m agreed(within standard errors) with the drifter means almosteverywhere. The only (barely) significant differencewas, as anticipated, near 108N, where mean mappedADCP velocities at 15 m showed a sharp reduction innorthward velocity that was not so marked in the driftermeans. Hence, the strength of downwelling in the NECCshould be viewed skeptically.

A delete-one jackknife approach (Efron 1982) wasused to estimate sampling errors, which were alwaysreported here as standard errors of the mean. The pro-cedure involved doing each calculation 85 times, omit-ting a section at a time to gain 85 slightly different

842 VOLUME 31J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y

FIG. 2. Vertical–meridional sections at 1368W based on third-order polynomial fits vs longitude of data taken from 1708 to 958W (seetext). (a) Zonal velocity u (1022 m s21); contour intervals (CI) 10, positive (eastward) shaded. (b) Standard error of u, eu (1022 m s21); CI1, |u| . eu shaded. (c) Temperature, T (8C); CI 1. (d) Salinity, S (PSS-78); CI 0.1.

estimates from which standard errors could be inferred.As already argued, each section was assumed to be in-dependent and measurement errors were assumed to benegligible compared with sampling errors. For simplecalculations like the uncertainty of an arithmetic mean,the jackknife estimate gives exactly the same answerfor the standard error of the mean as one would obtainusing more conventional methods. However, when do-ing operations like estimating the zonal velocity-weight-ed temperature of a current, the jackknife approachpropagates the effects of spatial and temporal correla-tions by systematically removing contemporaneous uand T data for each calculation. The same advantageholds for more complex operations like computing thesums of ux and y y at every depth and latitude, and thenintegrating these sums vertically to infer w.

3. Zonal velocity, temperature, salinity, andtransport

The meridional structure of the zonal velocity, tem-perature, and salinity are discussed in this section at themean longitude of all sections used, 1368W, where es-timates were the most reliable. The zonal structure ofthese fields along the equator is also discussed. Zonalcurrent transports and associated properties are pre-sented at 1368W. All of the fields discussed in this sec-

tion were estimated from third-order polynomial fits ver-sus longitude using data from 1708 to 958W.

The u field at 1368W (Fig. 2a) clearly showed themajor current structures. While some of the currentsdiscussed were not delineated by the 0.1 m s21 contour(Fig. 2a), they were discernible where their magnitudesexceed one standard error (shading in Fig. 2b). Eastwardflowing currents included the NECC with a velocitymaximum (core) at 78N, 50 m; the EUC with a core at08, 110 m; and the Northern Subsurface Countercurrent(NSCC) with a core at 48N, 210 m. The two very weakbranches of the Southern Subsurface Countercurrent(SSCC) had cores at 4.58S, 220 m and 78S, 290 m (shad-ing in Fig. 2b). Westward flowing currents included thenorthern branch of the SEC with a core at 28N, 0 m;the southern branch of the SEC with a core at 48S, 0m; and the Equatorial Intermediate Current (EIC) witha core at 2.58N, 330 m (shading in Fig. 2b). There wasalso significant eastward flow with a core at 1.58S, 400m, and westward flow with a core at 3.58S, 330 m (shad-ing in Fig. 2b). These flows were not analyzed becausethey appeared to extend beyond the 400-m depth range.The same held for all westward flow north of the NSCCand under the NECC because it appeared to extend be-yond the 108N latitude range.

The u field along the equator (Fig. 3a) was quite

MARCH 2001 843J O H N S O N E T A L .

FIG. 3. Vertical–zonal sections along the equator based on third-order polynomial fits vs longitude of data taken from 1708 to 958W (seetext). Longitudes displayed denote nominal section locations. (a) Zonal velocity, u (1022 m s21); contour intervals (CI) 10, positive (eastward)shaded. (b) Standard error of u, eu (1022 m s21); CI 1, |u| . eu shaded. (c) Temperature, T (8C); CI 1. (d) Salinity, S (PSS-78); CI 0.1.

similar to the mean zonal velocity from moored currentmeters on the equator (Yu and McPhaden 1999, theirFig. 2d). In fact, the u field along the equator agreed towithin one standard error with means from moored TAOarray ADCP data (Yu and McPhaden 1999) on the equa-tor at three longitudes taken over a similar period, June1991 through late 1999 (Fig. 4). The EUC shoaled fromwest to east, with zonal velocities peaking near the cen-ter of the region. Surface westward flow on the equatorin the SEC was relatively constant across the domain.However, while the SEC on the equator reached nearly50 m at 1708W it was limited to above the upper 20 mby 1108W. Below the EUC, westward flow of the EICwas found on the equator west of 1258W, strengtheningto the west.

The standard errors of u (Fig. 2b) were surface andequatorially intensified but dropped to about 0.02 m s21

outside of these regions. The peak of 0.10 m s21 wascentered near 1.58N, 0 m. High variance extended acrossthe EUC and northern branch of the SEC. A secondarymaximum of 0.06 m s21 was associated with the NECCnear 78N, 0 m. Along the equator (Fig. 3b) the standarderrors of u were again mostly surface intensified. Theyalso increased near the ends of the domain, as expectedwhere the third-order fit was less well constrained. Inaddition, a subsurface maximum existed at 50 m from

1258 to 958W, associated with the upper part of the EUCin the thermocline (Fig. 3c).

The mean temperature and salinity fields were typicalof the equatorial Pacific, and are shown mainly for ref-erence to the other fields. The coolest surface water at1368W was found on the equator (Fig. 2c), there wassome spreading of the thermocline at the core of theEUC, a trough existed between the NECC and the north-ern branch of the SEC, and a deep thermostad was ob-served near 138C between the SSCC and NSCC. Alongthe equator (Fig. 3c) the most prominent features werethe shoaling of the thermocline and the strengtheningof the 138C thermostad to the east. Fresh surface waterwas observed to the north at 1368W under the inter-tropical convergence zone (Fig. 2d), salty water withinthe southern thermocline, a meridional salinity front atthe equator within the thermocline where freshwaterfrom the north and salty water from the south converged,a maximum just south of the equator associated withthe EUC, and another weaker front below the thermo-cline associated with the NSCC (Johnson and McPhaden1999). Along the equator (Fig. 3d), the salinity maxi-mum associated with the EUC weakened to the east,under a fresh surface layer that strengthened to the east.

Direct estimates of cross-sectional areas, mean zonalvelocities, volume transports, temperatures, and salini-

844 VOLUME 31J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y

FIG. 4. Vertical profiles of equatorial zonal velocity, u (1022 m s21) based on third-order polynomial fits vs longitude of data taken from1708 to 958W (thin solid lines show one standard error about the mean, see text) compared to means of moored TAO ADCP records (thicklines dash–dot lines): (a) 1708W (moored data Jun 1991–Oct 1999), (b) 1408W (moored data Jun 1991–Aug 1995, and Sep 1996–Sep 1999),(c) 1108W (moored data Jun 1991–Nov 1999).

TABLE 1. Mean zonal currents in the upper 400 m at 1368W. Estimates are based on third-order polynomial fits vs longitude of data from85 meridional CTD/ADCP sections taken from 1708 to 958W during 1991–99 (see text). Temperatures and salinities are velocity weighted.Uncertainties are one standard error of the mean.

Currentacronym

Area(106 m2)

Speed(m s21)

Transport(106 m3 s21)

Temperature(8C)

Salinity(PSS-78)

NECCSEC*

SEC**

EUCNSCCSSCCEIC

93 6 753 6 6

176 6 10101 6 11

61 6 1042 6 1070 6 11

0.15 6 0.0320.26 6 0.0620.11 6 0.01

0.27 6 0.030.07 6 0.020.03 6 0.01

20.05 6 0.02

14 6 2214 6 3219 6 2

28 6 34 6 11 6 1

23 6 2

24.4 6 0.325.5 6 0.423.3 6 0.418.2 6 0.411.5 6 0.111.9 6 0.411.4 6 0.2

34.46 6 0.0334.72 6 0.0335.33 6 0.0335.07 6 0.0134.74 6 0.0134.87 6 0.0334.79 6 0.01

* 08–58N** 88S–08

ties of these currents were made at 1368W (Table 1).Currents were defined as regions of unidirectional flow.There were three pairs of adjoining currents flowing inthe same directions that required an additional definitionto distinguish them: the NECC and NSCC, the EIC andthe northern branch of the SEC, and the southern branchof the SEC and the deeper westward current centerednear 3.58S, 340 m. The boundaries between these threecurrent pairs were defined by the velocity minima be-tween them. The temperature and salinity means foreach current are velocity weighted.

These estimates can be compared with earlier onesmade between 1588 and 1508W using hydrographic datafrom the 1979–80 Hawaii-to-Tahiti Shuttle (WK),roughly 2000 km to the west. While no error bars weregiven for the earlier estimates, there are many areas ofagreement within the error bars given here. However,the NECC and south branch of the SEC reported herewere weaker than the previous estimates, as might beexpected with Sverdrup transports increasing westward.In addition, the eastward EUC reported here was stron-ger while the westward EIC was weaker than the pre-

vious estimates, consistent with the zonal structure ofzonal velocity along the equator (Fig. 2a).

4. Meridional velocity, horizontal divergence, andvertical velocity

As noted before, noise in y set the strategy for itsestimation as a function of depth and latitude only, with-out regard to longitude. The same held for y y, whichwas estimated over a 28 lat window. To match the zon-ally independent estimate of y y, ux was estimated forthe entire longitude range from the slopes of linear fitsof all u data versus longitude at each depth and latitude.When these two quantities were summed to obtain thehorizontal divergence and then integrated vertically, theresulting w field was estimated over the entire longituderange, meridionally smoothed over about 28 latitude.

At a sufficient distance from the equator, the y field(Fig. 5a) was presumably the sum of surface-intensifiedEkman divergence due to the easterly trade winds andgeostrophic convergence due to the eastward shoalingof the thermocline. The observations showed equatorial

MARCH 2001 845J O H N S O N E T A L .

FIG. 5. Vertical–meridional sections based on centered 28 lat linear fits to data taken from 1708 to 958W, regardless of longitude (seetext). (a) Meridional velocity, y (1022 m s21); CI 2, positive (northward) shaded. (b) Standard error of y , ey (1022 m s21); CI 1, |y |. ey

shaded.

surface divergence and thermocline convergence, andlittle cross-equatorial flow. Meridional velocities weremuch smaller than zonal velocities. Hence contour in-tervals for y were five times finer than for u. Polewardvalues of y were significant to 40–50 m (278–258C).Peak poleward surface speeds were 20.09 6 0.02 ms21 at 4.28S and 0.13 6 0.03 m s21 at 4.68N. Surfacedrifters also show stronger poleward flow in the norththan in the south from 1508 to 1308W (Baturin and Niiler1997), despite the stronger trades in the Southern Hemi-sphere (Hellerman and Rosenstein 1983). Equatorwardsubsurface speeds were significant in the SouthernHemisphere from 60 m (248C) through 160 m (158C)and in the Northern Hemisphere from 60 m (258C)through 110 m (218C). Interior equatorward velocitiespeaked at 0.05 6 0.02 m s21 at 1.48S and 20.04 6 0.03m s21 at 3.68N at 85 m (near 238C).

Meridional geostrophic velocities estimated from ver-tical integrals of zonal density gradients (not shown)were generally equatorward over the entire latituderange. They were also surface intensified and strongerin the south. This hemispheric asymmetry was consis-tent with the direct measurements of equatorward trans-ports. Since the equatorward geostrophic velocities weresurface intensified, the wind-driven poleward Ekmancomponent must have persisted to at least where themaximum in equatorward flow was observed, roughly85 m. This penetration depth was well below the mixedlayer depth (Figs. 2c and 2d).

The standard errors of y (Fig. 5b) were also surfaceintensified with off-equatorial maxima and dropped be-low 0.01 m s21 in places. The maxima were 0.04 m s21

at 4.58N and 0.03 m s21 at 18S, and were likely relatedto tropical instability wave activity on both sides of theequator, but somewhat weaker in the south (Chelton etal. 2000). While the mean meridional currents werenearly an order of magnitude smaller than the meanzonal currents, the y standard errors were still half themagnitude of those for u. This combination of energetictransient variability and a weak mean resulted in a low

signal-to-noise ratio for y . In contrast to the u field, they exceeded their standard errors (shading in Fig. 5b)mainly near the surface and in the thermocline on eitherside of the equator.

Since seawater is nearly incompressible and verticalvelocity, w, is negligible at the surface, the horizontaldivergence, ux 1 y y, can in theory be integrated down-ward from the surface yielded estimates of the verticalvelocity, w. The horizontal divergence was dominatedby the meridional term, y y, in most places. However,the shoaling of the EUC did result in a significant zonalterm, ux, in limited regions.

The largest feature in ux (Fig. 6a) was the verticaldipole centered about the equator with zonal divergenceabove 105 m and convergence from 110 to 270 m. Thispattern was due to the shoaling of the EUC. Equatorialzonal convergence below 270 m was the result of thegrowth of the westward-flowing EIC west of 1408W(Fig. 3a). Zonal convergence in the NECC and SEC wasdue to slight eastward weakening of the eastward-flow-ing NECC and slight westward strengthening of thewestward flowing SEC, as expected from the Sverdrupbalance. Zonal convergences centered near 230 m, 638latitude were tightly linked to divergences 60–80 mshallower and about 28 poleward. This pattern was dueto the shoaling and poleward shift of the NSCC andSSCC as they move eastward (Johnson and Moore1997). A more local estimate of ux from the slope of athird-order polynomial evaluated at 1368W significantlyincreased the magnitude of the dipole on the equator(Fig. 7a, dash–dotted line), and other features off theequator (not shown), but was not appropriate for a large-scale estimate of w.

Since uk y and y y k ux, y y (Fig. 6b) was contouredat five times coarser intervals than ux. On the equatorsurface meridional divergence, y y, was strong (Fig. 7b)at 60 (620) 3 1028 s21. Meridional convergence in thethermocline reached 30 (610) 3 1028 s21 on the equator.The transition from meridional divergence to conver-gence occurred at 50 m. Near-surface meridional con-

846 VOLUME 31J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y

FIG. 6. Vertical–meridional sections based on data taken from 1708–958W (see text). (a) Zonal divergence, ux (1028 s21) from linear fitsof u vs longitude; contour intervals (CI) 2, positive shaded. (b) Meridional divergence, y y (1028 s21); CI 10, positive shaded, estimated asin Fig. 5. (c) Vertical velocity, w (1025 m s21); CI 0.5, positive (upward) shaded. (d) Standard error of w, ew (1025 m s21); CI 0.5, |w| . ew

shaded.

FIG. 7. Mean properties at the equator plotted with standard error bars against depth, based on data taken from 1708 to 958W (see text).(a) Zonal divergence, ux (1028 s21), from the slopes of linear (solid line with error bars) and third-order polynomial (dash–dot line) fits ofu vs longitude. (b) Meridional divergence, y y (1028 s21), estimated as in Fig. 5. (c) Vertical velocity, w (1025 m s21) is the downward verticalintegral of the sum of panels a and b.

vergence was found from 58 to 98N, coincident with theNECC. Finally, subsurface off-equatorial meridional di-vergence reached 40 (610) 3 1028 s21 near 115 m, 38Sand 30 (610) 3 1028 s21 near 105 m, 58N, locationswhere the equatorward flow in the thermocline dimin-ished toward the poles. Equatorward geostrophic ve-

locities (not shown) dropped off much more graduallythan the directly measured ones, and thus suggested thatthe subsurface off-equatorial meridional divergencesmay have been overestimated and too localized.

The vertical velocity was estimated from these twofields by combining them to find the horizontal diver-

MARCH 2001 847J O H N S O N E T A L .

gence and then integrating down from the surface (Fig.6c). Only the regions where w exceeded one standarderror (shading in Fig. 6d) were considered possibly sig-nificant and worthy of discussion. Equatorial upwellingwas found above the EUC core, with a value of 1.9(60.9) 3 1025 m s21 at 50 m (Fig. 7c). Meridionallyintegrating w between 3.68S and 5.28N and applying theresults to the sampled region of 1708–958W resulted innet upwelling of 62 (618) 3 106 m3 s21 across 50 m,the most robust estimate from the upwelling calculation.Significant near-surface downwelling occurred withinthe NECC between 68 and 98N (Figs. 6c and 6d). At 50m, downwelling peaked at 21.5 (60.8) 3 1025 m s21

near 7.88N. However, the apparent strong meridionalconvergence south of 108N may have been exaggeratedabove 20 m because of previously mentioned problemsextrapolating ADCP velocities to the surface around ashallow mixed layer and strong thermocline. Thus thestrength of downwelling within the NECC may havebeen overestimated.

Interpretation of w below around 100 m was prob-lematic, as vertical integration of noise in the horizontaldivergence field started to dominate. By 200 m, w ap-peared to be composed of a series of very strong, fairlydepth-independent upwellings and downwellings,changing sign roughly every 28 of latitude (Fig. 6c).Tellingly, the distance for sign changes was about thesame size as the window used to estimate y y. Whilethese strong vertical velocities stood out from the stan-dard error of the mean over nearly a third of the domain(Fig. 6d), this is just about the proportion expected forrandom noise. Finally, the subthermocline vertical ve-locities remained roughly constant from 200 m downto the bottom of the sampled region, 400 m. This be-havior seems unlikely. For instance, the model of Marinet al. (2000) for the subsurface countercurrents (SCCs)predicts subthermocline downwelling on the equatorand upwelling at the SCCs, but these model w exist ina limited region just below the thermocline, associatedwith the equatorial thermostad.

5. Discussion

The zonal currents reported in this study complementthose of WK. The EUC transport here was larger thanin WK and the EIC transport was smaller. These dif-ferences are consistent with zonal structure of the zonalvelocity (Fig. 3a), which showed a stronger EUC andweaker EIC at 1368W than at 1548W. The NECC andtotal SEC transports reported here were smaller than inWK. These differences in off-equatorial current trans-ports were in the correct sense for zonal differencespredicted by the Sverdrup relation, which would haveboth of these current transports increasing in magnitudeto the west. Of coarse, the choice of reference level andlack of ageostrophic velocities in the WK estimatescould also account for some transport differences. Ve-locities used for this study were directly measured rather

than inferred from geostrophic calculations referencedto 1000 dbar as in WK. Direct velocity measurementsaround the equator have shown that 1000 dbar is notnecessarily an entirely satisfactory level of no motion,even in the mean (Firing et al. 1998), and some near-surface ageostrophy is to be expected. Another big dif-ference was the sparse sampling over an 8-yr period anda wide longitude range for this study compared with thewell-sampled year and small longitude range in WK.Zonal currents tend to change position (depth and lat-itude) and strength (velocity and area) with changes inlongitude, season, phase of the SOI, and under moretransient influences. The simple polynomial fits versuslongitude made here do not account for all of thosevariations. For instance, the NSCC and SSCC transportsand velocities reported here were much weaker thanthose from analysis of the synoptic CTD/ADCP sections(Rowe et al. 2000). Those eastward currents vary inlatitude over time and are embedded in westward flow,so simple Eulerian averaging schemes such as those inthis study reduce their magnitude.

The direct, large-scale estimate of equatorial up-welling transport reported here, 62 (618) 3 106 m3 s21

across 50 m, in the area bounded by 3.68S–5.28N,1708W–958W, is slightly higher than, but agrees withinerror bars with previous indirect large-scale estimates(all between 58S and 58N), but only when the estimatesare scaled by the zonal distances over which they weremade. One estimate from the oceanic 14C distributionwas 39 (611) 3 106 m3 s21 across 50 m between 1708and 908W (Quay et al. 1983). A simple box model basedon geostrophy, Ekman dynamics, and mass and heatbudgets between 1708E and 1008W gave an upwellingof 50 3 106 m3 s21 across 50 m (Wyrtki 1981). A similarbut more detailed diagnostic model yielded an estimateof 22 3 106 m3 s21 across 62.5 m (Bryden and Brady1985). This estimate was made between 1508 and1108W, so, while it is nearly three times smaller thanthat presented here, it covers little more than half thelongitudinal range.

The upwelling velocity reported here, 1.9 (60.9) 31025 m s21 at 50 m, was larger than estimated by Wyrtki,1.15 3 1025 m s21, but smaller than that of Bryden andBrady, 2.9 3 1025 m s21. These differences arise be-cause Wyrtki assumed upwelling within 200 km of theequator, whereas Bryden and Brady confined it to within83 km. The intermediate upwelling velocity found here,with a meridional decay scale near 175 km, was deter-mined directly, although probably broadened by thesmoothing from the objective mapping and estimationof y y over 28 latitude. Drifter data analysis (Poulain1993) suggests that equatorial upwelling may be moremeridionally localized than the resolution of this study,hence stronger. However, the equatorial upwelling trans-port estimated here was not likely to have been influ-enced by this potential resolution problem because itwas made over a latitude range much greater than themeridional smoothing scales employed.

848 VOLUME 31J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y

The large-scale equatorial upwelling velocity esti-mated here from shipboard CTD/ADCP sections wassomewhat larger than, but again agreed within error barswith, a local estimate from a current meter array aroundthe equator and 1408W (WQ). However, there are a fewobvious differences. First, the zonal divergence on theequator reported here (Figs. 6a and 7a) was somewhatless than that of WQ. This difference was probably dueto the much larger zonal extent (758 vs 48) over whichthe zonal gradients were estimated here compared to inWQ. A more local estimate of equatorial zonal diver-gence near 1368W (Fig. 7a, dash–dot line) was in betteragreement with the WQ estimate. Second, the error es-timates for vertical velocity here (Fig. 3f) were muchlarger than those of WQ, probably due to the differencesbetween estimating horizontal divergence over a verylarge area using sporadic snapshots of current, temper-ature, and salinity over 8.5 years versus doing so oversmaller spatial scales with a time series from May 1990through June 1991. Third, WQ reported significant near-equatorial downwelling below the thermocline, whichwas not seen here. However, the error bars for w wereso large here below 100 m on the equator as to rule outcomment on even the sign of w below the thermocline(Figs. 6d and 7c).

Off the equator, the estimates of w made here showeda large and significant near-surface downwelling in theNECC (Figs. 6c and 6d), as has been anticipated bylatitudinal wind stress and thermocline depth variations(WK). This pattern also emerged from a very similarcalculation using a smaller, independent ADCP dataset(Johnson and Luther 1994). In addition, the meridionalconvergence forcing this downwelling is evident, al-though slightly weaker than here, in surface drifter data(Baturin and Niiler 1987).

The temperatures of the surface divergence and thesubsurface convergence are of interest as an indicationof the depth from which the equatorial upwelling wasfed. The poleward surface transports started at 25.58Cnear the equator and warmed to about 27.58C by 688.On the other hand, the mean temperatures of the equa-torward thermocline transports were about 20.58C in theSouthern Hemisphere and 238C in the Northern Hemi-sphere. The dominance of the Southern Hemisphere inthe thermocline convergence implied a mean tempera-ture difference between source water and upwelled wa-ter on the equator of about 48C. This difference was thesame as that suggested for an earlier budget (Wyrtki1981) in which the surface heating was also included.Also, equatorward thermocline velocities reached totemperatures at as low as 138C (about su 5 26.4 kgm23) in the south and 168C (about 25.7 kg m23) in thenorth, somewhat lighter and shallower than the lowerbound of 26.5 kg m23 from chemical tracers (Fine et al.1983; Quay et al. 1983).

The most unusual aspect of this study lies in directmeasurements of the subsurface meridional flow field,which have rarely been reported (but see Johnson and

Luther 1994). An intriguing feature of the meridionalvelocity was greater poleward surface flow in the northcombined with greater equatorward thermocline flow inthe south. This asymmetry in the convergence and di-vergence tended to bring in more Southern Hemispherewater at depth and remove more surface water to thenorth in the amount of roughly 10 3 106 m3 s21. Appliedover the entire longitude range of the survey, the near-surface poleward transport at 58N was 34 (617) 3 106

m3 s21 compared with 223 (69) 3 106 m3 s21 at 58S.The asymmetry in poleward near-surface flow has beenobserved in earlier shipboard ADCP data (Johnson andLuther 1994) and drifter data (Baturin and Niiler 1997).In the thermocline, equatorward transports peaked at 30(617) 3 106 m3 s21 at 1.68S and 220 (621) 3 106 m3

s21 at 3.68N. Interior thermocline convergence on theequator has also been estimated to be roughly 10 3 106

m3 s21 greater in the Southern Hemisphere than theNorthern Hemisphere using geostrophy at slightly high-er latitudes (Johnson and McPhaden 1999). Of coursethe error bars on the direct velocity estimates are solarge as to make this agreement quite likely fortuitous.

The possible presence of stronger poleward surfaceflow in the north was initially puzzling, as the tradewinds were stronger in the south than in the north. How-ever, the strong surface meridional shears in the NECCand the SEC reduced the effective Coriolis parameternorth of the equator (to as little as 75% of nominal at38N) and increased it south of the equator (to as muchas 110% of nominal at 2.58S), allowing slightly strongerEkman transport in the north than in the south despitethe winds, localized to within 100 km of 638 latitude.In addition, the opposition of the NECC to the windsimparted a greater effective stress north of 58N, raisingthe northward Ekman transport there, while the SECrunning with the winds slightly reduced the effectivestress to the south (Kelly et al. 2000). Finally, equa-torward surface geostrophic flows were also about 0.02m s21 larger in the south than in the north, acting toreduce the poleward surface Ekman flow in the southmore than in the north. These three effects may havecombined to resolve the discrepancy between observedmeridional flows and those expected from zonal windstresses, certainly within the errors.

At any rate, the observed 10 3 106 m3 s21 asym-metries in meridional flow, while highly uncertain,could constitute a possible mechanism for interhemi-spheric exchange, where thermocline water primarilyfrom the Southern Hemisphere is upwelled at the equa-tor, with some combination of mixing and surface forc-ing erasing its negative potential vorticity, and then car-ried north at the surface. This process is found in nu-merical models (Blanke and Raynaud 1997; Lu et al.1998; Rodgers et al. 1999). However, a more detailedand better constrained evaluation of the 3D flow fieldin the equatorial Pacific will be required to fully evaluatethis possible interhemispheric pathway using observa-tions. Some such mechanism for interhemispheric trans-

MARCH 2001 849J O H N S O N E T A L .

port is ultimately required to supply the IndonesianThroughflow. The throughflow has a transport (Gordonet al. 1999) of a similar order to the observed asym-metry, and is fed by waters that have been cycledthrough the North Pacific (Ffield and Gordon 1992) aspart of the global thermohaline circulation. A similarconversion of intermediate water to surface water takesplace in the equatorial Atlantic (Roemmich 1983),where a net northward flow of warm water is requiredto balance southward export of North Atlantic DeepWater, another component of the global thermohalinecirculation.

Acknowledgments. This work was funded by theNOAA Office of Oceanic and Atmospheric Research,the NOAA Office of Global Programs, and the NASAPhysical Oceanography Program. It would not havebeen possible without the careful and sustained work ofthe officers, crew, and scientific parties of the NOAAShips Discoverer, Ka’imimoana, and Ronald H. Brown,especially Dennis Sweeny. Kristy McTaggart calibratedand processed much of the CTD data, and gridded theCTD/ADCP sections. Eric Johnson processed the earlierNOAA ADCP data and June Firing processed the laterNOAA ADCP data. Discussions with Ed Harrison, BillyKessler, Jimmy Larsen, Chris Meinen, Dennis Moore,and Peter Niiler were instructive. Comments of twoanonymous reviewers improved the manuscript.

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.

Blanke, B., and S. Raynaud, 1997: Kinematics of the Equatorial Un-dercurrent: An Eulerian and Lagrangian approach from GCMresults. J. Phys. Oceanogr., 27, 1038–1053.

Bryden, H. L., and E. C. Brady, 1985: Diagnostic model of the three-dimensional circulation in the upper equatorial Pacific Ocean. J.Phys. Oceanogr., 15, 1255–1273.

Chavez, F. P., P. G. Strutton, G. E. Friederich, R. A. Feely, G. Feldman,D. Foley, and M. J. McPhaden, 1999: Biological and chemicalresponse of the equatorial Pacific Ocean to the 1997–1998 ElNino. Science, 286, 2126–2131.

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

Efron, B., 1982: The Jackknife, the Bootstrap, and Other ResamplingPlans. CBMS-NSF Regional Conference Series in AppliedMathematics, Vol. 38, SIAM, 92 pp.

Ffield, A., and A. L. Gordon, 1992: Vertical mixing in the Indonesianthermocline. J. Phys. Oceanogr., 22, 184–195.

Fine, R. A., W. G. Peterson, C. G. H. Rooth, and H. G. Ostlund,1983: Cross-equatorial tracer transport in the upper waters ofthe Pacific Ocean. J. Geophys. Res., 88, 763–769.

Firing, E., S. E. Wijffels, and P. Hacker, 1998: Equatorial subther-mocline currents across the Pacific. J. Geophys. Res., 103,21 413–21 423.

Flament, P., S. Kennan, R. Knox, P. Niiler, and R. Bernstein, 1996:The three-dimensional structure of a tropical instability wave.Nature, 383, 610–613.

Gordon, A. L., R. D. Susanto, and A. L. Ffield, 1999: Throughflowwithin Makassar Strait. Geophys. Res. Lett., 26, 3325–3328.

Halpern, D., R. A. Knox, D. S. Luther, and S. G. H. Philander, 1989:Estimates of equatorial upwelling between 1408 and 1108W dur-ing 1984. J. Geophys. Res., 94, 8018–8020.

Hellerman, S., and M. Rosenstein, 1983: Normal monthly wind stressover the world ocean with error estimates. J. Phys. Oceanogr.,13, 1093–1104.

Johnson, E. S., and D. S. Luther, 1994: Mean zonal momentum bal-ance in the upper and central equatorial Pacific Ocean. J. Geo-phys. Res., 99, 7689–7705., and P. E. Plimpton, 1999: TOGA/TAO shipboard ADCP datareport, 1991–1995. NOAA Data Rep. ERL PMEL-67, 23 pp.

Johnson, G. C., and D. W. Moore, 1997: The Pacific subsurface coun-tercurrents and an inertial model. J. Phys. Oceanogr., 27, 2448–2459., and M. J. McPhaden, 1999: Interior pycnocline flow from thesubtropical to the equatorial Pacific Ocean. J. Phys. Oceanogr.,29, 3073–3098., , G. D. Rowe, and K. E. McTaggart, 2000: Upper equatorialPacific Ocean current and salinity variability during the 1996–1998 El Nino–La Nina cycle. J. Geophys. Res., 105, 1037–1053.

Kelly, K. A., S. Dickinson, and G. C. Johnson, 2000: The effect ofocean currents on the winds measured by scatterometers. Eos,Trans. Amer. Geophys. Union, 81 (22), (Suppl.), WP74.

Lu, P., J. P. McCreary Jr., and B. A. Klinger, 1998: Meridional cir-culation cells and the source waters of the Pacific EquatorialUndercurrent. J. Phys. Oceanogr., 28, 62–84.

Marin, F., B. L. Hua, and S. Wacogne, 2000: The equatorial ther-mostad and subsurface countercurrents in light of the dynamicsof atmospheric Hadley cells. J. Mar. Res., 58, 405–437.

Murray, J. W., E. Johnson, and C. Garside, 1995: A U.S. JGOFSprocess study in the equatorial Pacific (EqPac): Introduction.Deep-Sea Res. Part II, 42, 275–293.

Philander, S. G. H., 1990: El Nino, La Nina, and the Southern Os-cillation. Academic Press 289 pp.

Poulain, P.-M., 1993: Estimates of horizontal divergence and verticalvelocity in the equatorial Pacific. J. Phys. Oceanogr., 23, 601–607.

Quay, P. D., M. Stuiver, and W. S. Broecker, 1983: Upwelling ratesfor the equatorial Pacific Ocean derived from the bomb 14Cdistribution. J. Mar. Res., 41, 769–792.

Rodgers, K. B., M. A. Cane, N. H. Naik, and D. P. Schrag, 1999:The role of the Indonesian throughflow in equatorial Pacific ther-mocline ventilation. J. Geophys. Res., 104, 20 551–20 570.

Roemmich, D., 1983: The balance of geostrophic and Ekman trans-ports in the tropical Atlantic Ocean. J. Phys. Oceanogr., 13,1534–1539.

Rowe, G. D., E. Firing, and G. C. Johnson, 2000: Pacific EquatorialSubsurface Countercurrent velocity, transport, and potential vor-ticity. J. Phys. Oceanogr., 30, 1172–1187.

Stricherz, J., J. J. O’Brien, and D. M. Legler. 1992: Atlas of FloridaState University Tropical Pacific Winds for TOGA, 1966–1985.The Florida State University, 275 pp.

Taft, B. A., and W. S. Kessler, 1991: Variations of zonal currents inthe central tropical Pacific during 1970 to 1987: Sea level anddynamic height measurements. J. Geophys. Res., 96, 12 599–12 618.

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

Weisberg, R. H., and L. Qiao, 2000: Equatorial upwelling in thecentral Pacific estimated from moored velocity profilers. J. Phys.Oceanogr., 30, 105–124.

Wyrtki, K., 1981: An estimate of equatorial upwelling in the Pacific.J. Phys. Oceanogr., 11, 1205–1214., and B. Kilonsky, 1984: Mean water and current structure duringthe Hawaii-to-Tahiti shuttle experiment. J. Phys. Oceanogr., 14,242–254.

Yu, X., and M. J. McPhaden, 1999: Seasonal variability in the equa-torial Pacific. J. Phys. Oceanogr., 29, 925–947.

1936 VOLUME 31J O U R N A L O F P H Y S I C A L O C E A N O G R A P H Y

CORRIGENDUM

In the article ‘‘Equatorial Pacific Ocean horizontal velocity, divergence, and upwelling,’’ which appeared inthe March 2001 Journal of Physical Oceanography, Vol. 31, No. 3, 839–849, part of the shaded regions in panels(c) and (d) of Fig. 6 on page 846. The missing shading describes the author’s major finding of significant upwellingon the equator, does not appear in the illustrations due to a production difficulty. The correct figure, showing theshaded regions from about 48S to 68N and at 108N in panel (c) and from about 48S to 28N and 68N to 98N inpanel (d) that are missing in the illustration printed in the March issue, is printed below.

FIG. 6. Vertical–meridional sections based on data taken from 1708–958W (see text). (a) Zonal divergence, ux (1028s21) from linear fitsof u vs longitude; contour intervals (CI) 2, positive shaded. (b) Meridional divergence, yy (1028s21); CI 10, positive shaded, estimated as inFig. 5. (c) Vertical velocity, w (1025s21); CI 0.5, positive (upward) shaded. (d) Standard error of w, ew (1025s21); CI 0.5, | w | . ew shaded.