Progress in Oceanography 130 (2015) 1–18

Contents lists available at ScienceDirect

Progress in Oceanography

journal homepage: www.elsevier .com/ locate /pocean

Zonal jets in the equatorial Atlantic Ocean

http://dx.doi.org/10.1016/j.pocean.2014.08.0080079-6611/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author.1 Present address: Centre National de la Recherche Scientifique (CNRS), Division

Technique de l’Institut de Sciences de l’Univers (DT-INSU), Plouzané, France.

Miquel Rosell-Fieschi, Josep L. Pelegrí ⇑, Jéröme Gourrion 1

Departament d’Oceanografia Física i Tecnològica, Institut de Ciències del Mar, CSIC, Barcelona, Spain

a r t i c l e i n f o a b s t r a c t

Article history:Received 5 April 2012Received in revised form 3 August 2014Accepted 16 August 2014Available online 28 September 2014

We use position data from Argo floats, smoothed out over 400 km � 200 km zonal ellipses and interpo-lated onto a 0.5� grid, to investigate the zonal jet structure of the flow field at the sea surface and on threesubsurface layers (Central Waters, CW, 200 m; Antarctic Intermediate Waters, AAIW, 1000 m; upperNorth Atlantic Deep Waters, uNADW, 1500 m) in the equatorial Atlantic Ocean (15�S to 15�N). Theannual-mean fields exhibit narrow zonal jets, typically 4–5� wide at the sea surface and only 2� at thesubsurface levels, with directions alternating in latitude and maximum speeds about 0.5 m s�1 at the sur-face, 0.1 m s�1 at CW and uNADW, and 0.03 m s�1 at AAIW. The available data also allows us to explorethe seasonal variability of these jets at the surface and AAIW levels. The surface currents are dominatedby an annual cycle between 4�N and 10�N and, to a lesser degree, by a semi-annual contribution close tothe equator. This variability is an outcome of evolving zonal recirculations, with the North EquatorialCountercurrent (NECC) arising from the diversion of the northern branch of the South Equatorial Current(nSEC); the diversion begins in the eastern Atlantic and propagates west between April and August, fol-lowing the Inter-Tropical Convergence Zone (ITCZ). The AAIW current field is largely affected by west-ward propagating anomalies, most visible near 3�S, 0�, 3�N and 7�N, which give rise to currentreversals. Annual averaging produces the illusion of more (5 instead of 3) and slower (peak values about0.03 m s�1 instead of 0.1 m s�1) jets than found on any month.

� 2014 Elsevier Ltd. All rights reserved.

Introduction

The equatorial oceans are characterized by the vertical and lat-itudinal staggering of eastward–westward currents (Table 1). Thepresence of a complex pattern of alternating zonal currents, at sur-face and subsurface depths, was first observed through acousticdropsondes in the Indian (Luyten and Swallow, 1976) and Pacific(Hayes and Milburn, 1980; Eriksen, 1981) Oceans, and later inthe Atlantic Ocean (Ponte et al., 1990; Send et al., 2002). The Atlan-tic equatorial system was initially studied assuming geostrophicbalance (Katz, 1981; Eriksen, 1982; Merle and Arnault, 1985) butnear the equator geostrophy fails and its improved descriptiondemanded direct velocity measures from instrumented moorings(Send et al., 2002; Brandt et al., 2006; Bunge et al., 2006, 2008),ship-borne current profilers (including Acoustic Doppler CurrentProfilers, ADCP, and Lowered-ADCP, LADCP) (Gouriou andReverdin, 1992; Send et al., 2002; Gouriou et al., 1999, 2001;Brandt et al., 2006), acoustically tracked buoys and/or profilingfloats (Richardson and Fratantoni, 1999; Schmid et al., 2001;

Ollitrault et al., 2006; Lankhorst et al., 2009), or a combination ofmultiple measurements (Urbano et al., 2008; Perez et al., 2013).

The first descriptions of zonal jets in the equatorial Atlanticdealt with the near-surface structures. The predominant surfacecurrent is the westward flowing Southern Equatorial Current(SEC), the rather wide northern branch of the South Atlantic sub-tropical gyre which merges with the wind-driven equatorial cur-rents. This current is composed of three, poorly differentiated,branches: central (cSEC, about 3�S to 5�S), equatorial (eSEC, nearthe equator when present) and northern (nSEC, about 2�N to4�N). All these branches merge onto the North Brazil Current, thenorthwestward flowing western boundary current (Stramma andSchott, 1999; Schott et al., 2004; Lumpkin and Garzoli, 2005).

The North Equatorial Countercurrent (NECC), the major surfacezonal jet in the tropical Atlantic, has also received considerableattention (Garzoli and Katz, 1983; Richardson and McKee, 1984;Richardson and Reverdin, 1987; Carton and Katz, 1990; Diddenand Schott, 1992; Richardson et al., 1992; Polonsky andArtamonov, 1997; Bourlès et al. 1999; Fonseca et al., 2004;Artamonov, 2006; Hormann et al., 2012). The NECC is characterizedby an intense annual cycle, its transport ranging between non-sig-nificant late winter values and summer-fall maxima (throughoutthis article we will always refer to astronomical boreal seasons).The NECC is found between the sea surface and depths of about

2 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

350 m, at latitudes between 4�N and 8�N (with a northern branchreaching up to 10�N in fall), north of a zonal band of maximum posi-tive sea surface height values. Urbano et al. (2006, 2008) havereported the NECC to have a double-core structure in the westernAtlantic, which is best defined in the western margin during sum-mer and fall.

The seasonality of the NECC has been related to the annual cyclein the surface winds, specifically to the curl of the wind stress,resulting from the latitudinal displacements of the Inter-TropicalConvergence Zone (ITCZ) (Garzoli et al., 1982; Merle and Arnault,1985; Richardson and Walsh, 1986; Richardson and Reverdin,1987; Arnault, 1987; Richardson et al., 1992; Urbano et al., 2006,2008). Schouten et al. (2005) and Yang and Joyce (2006) have alsoassociated the NECC variability to equatorial wind forcing and itsgeneration of westward propagating waves.

Other major zonal currents are observed at subsurface levels,flowing east under the wind-driven westward SEC. These are theEquatorial Undercurrent (EUC), centered at the equator and 100 mdepth, and the off-equatorial South/North Equatorial Under Cur-rents (SEUC/NEUC), centered at some 150–200 m and 4�N/4�S (e.g.Metcalf et al., 1962; Tsuchiya, 1986; Stramma and Schott, 1999).The EUC feeds from the retroflection of the NBC, typically as a verytight loop near the equator (Flagg et al. 1986; Schott et al. 1998;Hüttle-Kabus and Böning, 2008; Claret et al., 2012). The strengthof the EUC decreases as it flows east, with a maximum transport ofabout 20 Sv, while there are fewer reports on the spatial and tempo-ral variations of the northern and southern branches (Gouriou andReverdin, 1992; Bourlès et al., 1999; Schott et al., 2003, 2004;Brandt et al., 2006; Hüttle-Kabus and Böning, 2008). During springthe EUC surfaces and increases its speed (Brandt et al., 2006;Urbano et al., 2008).

Several authors have shown that zonal jets in the equatorialAtlantic are also found far from the sea surface (Gouriou et al.,1999, 2001; Richardson and Fratantoni, 1999; Bourlès et al.,2003; Schott et al., 2003; Brandt and Eden, 2005; Brandt et al.,2006). The circulation schemes by Stramma and Schott (1999),amended by Schmid et al. (2003) for intermediate waters, indeedemphasized the predominance of zonal jets at deep levels in theequatorial Atlantic region. Equatorial Deep Jets (EDJs), trappedbetween 2�S and 2�N, are found at depths between 300 and2500 m. These jets have relatively short meridional scales, as littleas only 1�, and display alternating directions on vertical distancesof 400–600 m, with maximum velocities about 0.2 m s�1. Theirvertical structure is quite consistent through one same seasonbut changes with season. The EDJs are surrounded by eastward col-umns of Extra Equatorial Jets (EEJs), sometimes named subsurfacecountercurrents (after Tsuchiya, 1986), located at about 3�S/N andextending from as shallow as 200 m down to near the sea floor. TheEEJ velocity cores are found at the depth of the westward EDJs(about 500 m), suggesting the existence of elongated recirculationgyres as observed in the Pacific (Firing et al., 1998). Hüttle-Kabus

Table 1Major characteristics of the zonal jets in the equatorial Atlantic Ocean at surface and sub-

Level Current Core latitude

Sea surface North Equatorial Countercurrent, NECC 5–8�NNorthern South Equatorial Current, nSEC 2–3�NEquatorial Under Current, EUC 0�

Central South Equatorial Current, cSEC 4�S

1000 dbar North Equatorial Intermediate Current, NEIC 3–4�NEquatorial Intermediate Current, EIC 0�South Equatorial Intermediate Current, SEIC 2–4�S

Sub-surface Equatorial Deep Jet, EDJ 2�S to 2�N

Extra-Equatorial Jets, EEJs 3�S and 3�N

and Böning (2008) proposed the eastward EEJs to feed on the sub-tropical cell via tropical instability waves from the EUC; theseauthors found that these flows are dominated by an annual and,to a lesser degree, a semiannual harmonic.

The introduction of Lagrangian buoys has substantially enhancedour skill to observe the horizontal coherence of the equatorial jets atseveral depths. Ollitrault et al. (2006) used acoustic drifters near800 m and profiling floats parked at 1000 m to propose the existenceof a system of rather narrow zonal jets, changing direction aboutevery 2� in latitude: South Equatorial Intermediate Current SEIC(4�S), Southern Intermediate Countercurrent SICC (2�S), EquatorialIntermediate Current EIC (0�), Northern Intermediate Countercur-rent NICC (2�N), and North Equatorial Intermediate Current NEIC(4�N). According to Ollitrault et al. (2006), the SEIC and NEIC flowwest, the SICC and NICC flow east, and only the EIC reverses sign withseason, westward in fall and eastward in winter. In Table 1 weinclude neither SICC nor NICC, as we will argue later that their differ-entiation from the SEIC and NEIC arises only from the seasonality ofthe intermediate currents (section ‘AAIW velocity variability’).

Lankhorst et al. (2009) combined Argo float and acoustic drifterdata within intermediate (600–1050 m) and upper-deep (1200–2050 m) layers to look at the interaction between boundary andzonal flows. The northward flowing North Brazil Under Current(NBUC), which extends from under the surface mixed layer downto 1000 m (Stramma et al., 1995), could possibly be one mainsource for the eastward intermediate flows, although we will laterpresent observations of seasonal changes in direction which pointat the predominance of propagating anomalies. At the upper-deeplevel, the southward-flowing DWBC decreases in intensity as itinteracts with the interior zonal flows.

In November 2007 the Argo program achieved its goal: over3000 simultaneous profiling floats in the Global Ocean that driftat several depths and perform over 100,000 profiles per year, witha mean resolution of about one profile per year in a 60 km � 60 kmgrid. Presently, the dataset is large enough to map mean velocityfields at several water depths and even to examine the seasonalvariability at surface and intermediate layers. In this study weuse the equatorial and tropical Atlantic Argo data to describe theequatorial current system at four different levels (the sea surfaceand three additional drifting levels). The results confirm the exis-tence of the zonal current system and give further insight into itsspatial distribution, as well as its seasonal variation, at the surfaceand intermediate levels.

Data set and methods

Argo-inferred velocities

Only a few studies have obtained and disseminated an Argo-derived velocity data base (Ollitrault et al., 2006; Lebedev et al.,2007). Here we use a simple approach to produce our own data

surface levels.

Flow seasonal cycle (boreal seasons)

East, with fall maximum up to 0.5 m s�1

West, with summer-fall maximum up to 0.6 m s�1

East at subsurface, surfaces in the western basin on summer reaching0.3 m s�1

West with spring–summer maximum up to 0.4 m s�1

West in spring, up to 0.07 m s�1; east in fall, less than 0.05 m s�1

West in summer, up to 0.14 m s�1; east in winter, up to 0.10 m s�1

West in winter–spring, up to 0.07 m s�1; east in summer-fall, up to0.05 m s�1

Between 300 and 2500 m, alternate directions in vertical scales of 400–600 m, direction changes with seasonFrom 200 m down to the sea floor

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 3

set, generating velocity fields at the surface and at several parkingdepths. The procedure infers the surface and parking-depth veloci-ties from the position transmissions while a float remains at thesea surface. Only the first and last surface transmissions are usedto calculate the surface velocities, therefore maximizing the surfacedistance and time, and minimizing the error due to instrumentallimitations. For the deep displacement we use the last positiontransmission before departure from the sea surface and the firstposition transmission after the next sea surfacing. As a result wegenerate one surface and one deep velocity vector per cycle; in con-trast, Ollitrault and Rannou (2013) generate two surface velocityvectors by using an intermediate surface position.

Until 2012 the Argo floats used the Argos tracking system(www.argos-system.com) but since 2013 most of the deployedfloats use the two-way Iridium positioning system (www.irid-ium.com). As a consequence, the floats using the Iridium systemremain at the surface for a relatively short period of time (usuallyless than one hour) as compared with those using the Argos system(typically about 12 h). This does not affect the estimation of thedeep velocities but does increase the error associated to calculatingthe surface velocities; in some instances, the floats using the Irid-ium System have one single position and the surface velocity can-not be estimated.

Fig. 1. Salinity distribution on a meridional section produced using data from all availaintermediate waters, on top we find the central waters and below the deep waters. (Bottoin the top panel, color-coded with (left) pressure and (b) latitude. (For interpretation of thof this article.)

The whole Argo dataset up to September 2013, between 20�Sand 20�N, and 75�W and 15�E is examined. Only data flagged asgood in the trajectory variables (both for position and time) areconsidered and the parking pressure contained in the float’s meta-data file is used as the real drifting depth. Nine different parkingdepths are found in the equatorial Atlantic (200, 250, 300, 400,1000, 1100, 1500, 1900 and 2000 dbar) but here we have ignoredthose depths with a relatively scarce number of floats (250, 300,400 and 1100 dbar) or where floats are localized in some con-strained region (floats at 1900 and 2000 dbar are mostly foundnear the African coast). Therefore, we use all floats to calculatethe surface drift but choose only three parking depths to generatethe sub-surface velocity fields, obtaining the following amount ofvelocity vectors: surface waters, all floats (SW, 57,413 vectors);central waters, floats at 200 dbar (CW; 3314 vectors); intermediatewaters, floats at 1000 dbar (AAIW, 44,308 vectors); and upperNorth Atlantic Deep Water, floats at 1500 dbar (uNADW, 3449vectors).

The Argo data can be used to illustrate the temperature andsalinity values of the above water masses in the tropical Atlantic.Fig. 1 (top panel) shows the salinity distribution on a meridionalsection from 15�S to 15�N, drawn using the data from all availableArgo profiles between 31�W and 29�W. The presence of AAIW inthe tropical Atlantic, defined as corresponding to waters of salinity

ble Argo profiles between 31�W and 29�W. Waters fresher than 34.9 correspond tom) Potential temperature – salinity diagrams obtained using the same Argo data ase references to colour in this figure legend, the reader is referred to the web version

4 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

less than 34.9 (Talley, 1996), clearly separates the upper-thermo-cline CW from the relatively deep uNADW; this water-mass stra-tum is also easily identified in potential temperature – salinitydiagrams as having relatively homogeneous temperatures(between 4 �C and 6 �C) in an extent region (from 50�S to the equa-tor) (Fig. 1, bottom-right panel). AAIW is located between about400 and 1200 m with its core at some 800 m (Talley, 1996;Schmid et al., 2001, 2003) (Fig. 1, top and bottom-left panels);for this reason the floats at 1000 dbar do not sample the core ofAAIW but rather its lower part.

We use several steps in order to obtain a gridded velocity field.The first step is to estimate individual velocity values for all avail-able Argo data in the study region. The next step is to remove thoseextremely large velocity vectors, with a relatively high probabilityof being spurious values. This is not a trivial procedure, as our anal-ysis of the Argo-inferred velocities in the equatorial Atlantic Oceanshows the existence of long tails in the PDFs related to real extremeevents (data not shown). Here we have used a criterion of six stan-dard deviations as a compromise between not removing a signifi-cant number of high-velocity events and the need of eliminatingvery high spurious values. This criterion approximately corre-sponds to velocities greater than 1.5 m s�1 and 0.5 m s�1 at the sur-face and 1000 dbar, respectively. It is a criterion less restrictivethan the threshold values imposed by Ollitrault et al. (2006), 3and 2 m s�1 respectively for the surface and 1000 dbar levels,and yet plenty satisfies the commonly used Chauvenet’s criterion(Taylor, 1997).

Finally, once the individual velocities are obtained, in order tocalculate the time-averaged velocities, we use a 0.5� latitude-longi-tude resolution grid and assign to each grid cell all velocity vectorscontained in an ellipse with a zonal major axis of 400 km and a

Fig. 2. Number of velocity profiles assigned to each 0.5� latitude-longitude cell.From top to bottom: surface, CW (200 dbar), AAIW (1000 dbar), and uNADW(1500 dbar).

meridional minor axis of 200 km. This asymmetry is consistentwith the larger zonal than latitudinal coherence of the flow fieldin the tropical Atlantic (Stramma and Schott, 1999). After removingthe land cells, the 0.5� grid produces a total of 8660 cells at the seasurface and 8178 cells at 1000 dbar. The above procedure renders aspatial smoothed version of the velocity fields, similar to the resultof applying a running filter of 4� in longitude and 2� in latitude,with a total number of surface velocity vectors for most cells rang-ing between about 100 and 300 (Fig. 2).

The available amount of data is excellent for calculating annual-mean values but it is not adequate to examine spatial patterns ofinterannual variability (over 64% of the data has been collectedbetween January 2008 and September 2013). Nevertheless, thecombination of data from all years does allow producing a climato-logical year for each grid point at the surface and AAIW levels. Wehave explored using many different temporal intervals, e.g. a sev-eral-month running window, and have arrived to the conclusionthat the available data is adequate to produce simple monthlyvelocity values. The mean number of velocity vectors per cell andmonth is 11.1 at the sea surface and 8.9 at 1000 dbar. This rela-tively large number of monthly velocity values is the result ofthe long spatial integration. It allows focusing on the seasonal var-iability of basin-scale features but prevents from studying rela-tively fast and small processes, with time scales of the order ofone month or less and spatial scales less than 200 km.

Estimating the velocity errors

The surface velocity estimates are limited by the accuracy in thepositioning system errors and, to a much lesser degree, by poten-

Fig. 3. (Top) Monthly meridional winds, space-averaged in the longitudinal bandbetween 30�W and 20�W, for the three indicated latitudinal bands. (Bottom)Monthly meridional currents calculated for the same domains as in the top panel.

Fig. 4. (Top) Geographic distribution of those profiles containing all variables necessary to estimate the error in the subsurface velocity which is associated to the surfacedrift. (Bottom) Histograms for the relative error in the deep velocity estimate, caused by the combination of the untracked surface drift and the horizontal drift during thefloat’s vertical migration, for the (left) zonal and (right) latitudinal velocity components.

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 5

tial problems in the instrument’s behavior and codification. In par-ticular, Ollitrault and Rannou (2013) have reported differencesbetween the real parking depth and the parking depth in the meta-data file for about 4% of the cases. A different type of limitationarises because Argo floats, cylinders of lengths between 1 and2 m that drift at the sea surface with no additional drogue, maybe affected by both direct wind drag and motions that are not char-acteristic for the bulk of the surface-mixed layer. Wind drag islikely not significant because the floats, except for the topmost sev-eral centimetres and the antenna, are totally immersed in thewater; as a consequence, the wind drag on the emerged portionof the float is very much reduced, and the associated motions mustbe relatively small. The second effect, i.e. the drift associated tomotions that affect only the top several meters of the water col-umn, deserves further consideration.

The bulk of the surface mixed layer is usually considered tohave a slip-like motion, i.e. the surface layer is well-mixed not onlyin heat and salinity but also in momentum, as shown not only bydirect ADCP measurements in the equatorial Atlantic (Perezet al., 2013) but also from the successful application of slab-likemodels to simulate wind-driven motion in the surface mixed layer(e.g. Pollard and Millard, 1970; Price et al., 1978; Schudlich andPrice, 1992; Alford, 2001). The main deviations in the motion ofthe upper meters of the ocean, as compared with the mean motionof the mixed layer, are those related to the surface gravity waves(air-sea momentum transfer through breaking waves and Stokes

drift) and to the surface-most wind-driven shear layer (Csanady,2001; Soloviev and Lukas, 2006). Both effects will depend on thesea-surface wind drag and the way momentum is transferred intothe ocean, both through the waves themselves and through near-surface turbulence. There is no straight-forward relation to com-pute the influence of the surface shear layer but we may easilyassess the relevance of the wave-related drift.

The surface wave drift depends mainly on the locally-generatedwaves, as swell induces almost no net drift because of the compen-sating effects of the Stokes and Stokes–Coriolis effects (Rascle et al.,2006; Rascle and Ardhuin, 2009). In the tropical Atlantic Ocean, thewave field is moderately low all year long, with annual-mean sig-nificant wave heights that arise through similar contributions fromlocal winds and swell and do not exceed 2.5 m (Gulev andGrigorieva, 2006; Gulev et al., 2011). Considering solely the localwind sea, we use monthly maximum wave amplitudes (a = 1 m)together with characteristic wave lengths (k = 100 m) and periods(T = 10 s) for the tropical Atlantic to assess the maximum Stokesdrift velocity at the sea surface, 4p2a2/(kT), of about 0.04 m s�1.

An independent estimate of the velocity anomalies in the uppermeters of the surface ocean, as compared with the mean motion ofthe surface-mixed layer, may be obtained by looking at themonthly-mean velocities in a region near the center of the oceanand very close to the equator, here chosen from 30�W to 20�Wand from 1�S to 1�N. The latitudinal component of the velocityought to be very small, as this is a surface-divergent region, the ori-

Fig. 5. Contour maps for the annual-mean zonal velocities (m s�1) at surface (top), CW (second row), AAIW (third row) and uNADW (bottom) levels. Note the change in scalebetween the left and right panels. Each panel is accompanied by a box that shows the latitudinal distribution of the zonally-averaged (33�W to 20�W) zonal velocity as afunction of latitude (black line). The dashed red lines show one standard deviation as calculated using the cell-mean values (0.5� latitude grid) between 33�W and 20�W. (Forinterpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

6 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

gin of the near-surface tropical cell (Perez et al., 2013). However, itis also a region where the waves flow northwestward all year longand winds do so most of the year, so that any deviation from zeroought to be caused by the combined wind-drag and wave-induceddrift. We find there is an annual cycle with amplitude of 0.05 m s�1

(Fig. 3, bottom panel), with near-zero drift at times when the localwinds have a weak southward component (March and April) and amaximum northward drift of 0.10 m s�1 (September–October)when the northward wind component is largest (Fig. 3, top panel).These values are consistent with the above estimates for the max-imum wave-induced drift: in spring the effects of surface shear(southward motion) and waves (still with a northward predomi-nant motion) likely compensate each other, while in fall they haveand additive effect. It is interesting to note that the difference invelocities, between the latitudinal currents in the northern andsouthern halves of the domain, is a near-constant value of about0.04 m s�1, a result consistent with the idea of a divergent tropicalcell between about 4�S and 4�N (Hastenrath and Merle, 1987;Wang, 2005; Perez et al., 2013). We will see later that these veloc-ity anomaly estimates are one order of magnitude smaller than themonthly-mean maximum surface zonal velocities and a factor of

3–4 times smaller than the largest monthly-mean surface latitudi-nal velocities. This grants us good confidence for using the Argo-inferred surface velocity calculations as good indicators for the pre-dominant currents in the surface-mixed layer.

Finally, it is worth mentioning that Argo floats, as any drifterwith its drag in the surface mixed layer, will experience tidal andinertial motions. Tidal motions in the open ocean are small, com-monly of the order of 0.01 m s�1, but inertial velocities in the sur-face mixed layer may be much greater, typically of the order of0.1 m s�1. Nevertheless, both tidal and inertial velocities have nopreferential direction and their net effect will tend to zero whenaveraging out with a sufficient number of velocity observations.

The subsurface velocity estimates, on the other hand, may con-tain errors arising from (a) the drift experienced by the float duringboth ascend and descend profiling phases and (b) the lag betweenthe time when a float reaches (leaves) the sea surface and the next(last) transmission time. Once we have the surface and parking-depth velocities, we may obtain simple estimates for both the driftand lag-time errors. The drift error is estimated from the knowl-edge of the approximate time the floats take to ascend-descend(including the plunge to 2000 dbar) and by assuming that during

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 7

these vertical motions the floats experience a horizontal velocitywhich is the mean of the (cell-mean) surface and parking-depthvelocities. This last condition is equivalent to the assumption of alinear velocity change with depth; it is a gross idealization becausethe profilers, during their vertical migrations, may cross flow struc-tures independent from those observed at the surface and theparking depth, yet it provides an estimate for the order of magni-tude of the associated velocity error. The drift velocity turns outto be half the vector difference between the surface and parking-depth velocity estimates.

Fig. 6. Contour maps for the zonal component of the monthly-mean velocities at the sea-as calculated by Hastenrath and Lamb (1977), is shown with a dashed line.

The second error, derived from the time lags between the sur-facing/sinking and first/last transmission times, can be estimatedfrom a limited number of instances (about 10% of the cases) whenarrival and departure times are available (Fig. 4, top panel). Thedistribution in the number of velocity vectors as a function of timelag shows that Iridium-positioning floats spend a mean untrackedtime of about 10 min in contrast with a time interval close to twohours for Argos-tracking floats. This information together with thesurface velocity could, in principle, be used to obtain a mean dis-placement correction; this correction would be rather small, lessthan 4% the original speed for the Argos-transmitting floats and

surface (m s�1). The climatological latitudinal location of the ITCZ at each longitude,

8 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

much less for the Iridium ones. Nevertheless, we have chosen notto incorporate this first-order correction as most of the availabletransmission-lag data corresponds to the last few years and wehave no way to know if the Argos-transmitting statistics haveremained unchanged during the last decade.

The total (drift plus time-lag) deep-velocity errors are typicallyjust a fraction of the velocity standard deviation, as calculated fromall velocity vectors used to compute the mean values at each gridpoint, and one order of magnitude smaller than the parking depthmean velocities (Fig. 4, bottom panels). We find that the mostprobable error is less than 3%, with 58.0% of the velocity estimateshaving a relative error less than 10%, and 93.7% of the velocity esti-mates having an absolute error smaller than the velocity valueitself. These results are in good agreement with Lebedev et al.(2007), and grant us high confidence on the procedure used todetermine the velocity fields.

Monthly wind fields

The sea-surface atmospheric circulation in the equatorial Atlan-tic is characterized with the help of monthly WindSat products(www.remss.com/missions/windsat). The winds, taken at a stan-dard height of 10 m, are available on a 2� latitude-longitude grid.A monthly climatology is obtained using data from 2005 to 2012,corresponding roughly to a time period when most of the Argo datawas acquired. Separately, the monthly location of the ITCZ is digi-tized from the plots presented in Hastenrath and Lamb (1977).

Mean velocity fields at four depth levels

Fig. 5 presents annual-mean maps for the zonal velocity at allfour depth levels. Aside each map we show boxes with the latitu-dinal distribution of the zonal velocities, calculated for each 0.5�latitude interval by averaging all cell-mean values between 33�Wand 20�W, as in Ollitrault et al. (2006). The standard deviationturns out to be substantially smaller than the mean values,

Fig. 7. (Top panels) Amplitude (m s�1) and (middle panels) phase (months) distributionscomponent of the sea-surface velocity; a transparency mask (quadratic proportion to aareas with high amplitudes. The bottom panels illustrates the fraction of the variance e

rendering the zonal coherence of the mean fields in the westernAtlantic Ocean.

At all levels we find zonal jets which have zonal continuity overmost of the Atlantic Ocean and alternate direction with latitude,corresponding to what has been reported in the literature (e.g.Stramma and Schott, 1999; Schmid et al., 2003; Schott et al.,2004). The annual-mean meridional velocities (not shown) haveamplitudes substantially smaller than the zonal ones, particularlyat the subsurface levels, and display little spatial coherence. Theonly major exception is the NBC, which flows as a relatively narrownorthwestward surface flow along the northern coast of Brazil,connecting waters from the southern and northern hemispheres.

The annual-mean surface zonal flow is dominated, close to theequator, by the northern and central branches of the westward SECand, further north, by the eastward NECC (Fig. 5, Table 1). The sig-nal of the NBC is also visible in the zonal velocity maps as anintense westward current adjacent to the northern coast of Brazil.These surface currents are relatively wide, about 4� in latitude, asearly reported in the literature (Stramma and Schott, 1999). How-ever, as a consequence of the annual averaging, the NECC appearsas a weakened eastward flow (the NECC disappears in late winterand spring; Garzoli and Katz, 1983; Richardson and Reverdin,1987) and the equatorial region shows up as a region of weakenedwestward flows (the EUC surfaces in spring; Brandt et al., 2006;Urbano et al., 2008).

In contrast to the relatively wide surface currents, the annual-mean and zonally-averaged (33�W to 20�W) CW, AAIW anduNADW flows display a complex pattern of rather narrow cur-rents and counter-currents (Fig. 5). In particular, despite the dif-ferences in the amount of Argo data, the jets at AAIW comparereasonably well with the 33�W to 20�W averaged zonal jet sys-tem reported by Ollitrault et al. (2006). These annual-mean AAIWjets display zonal continuity only in the western Atlantic, as thezonal pattern disappears east of about 23�W. However, we willsee later (section ‘AAIW velocity variability’) that the annual-mean AAIW velocity field is an artefact resulting from the annual

for the (left panels) annual and (right panels) semi-annual contributions to the zonalmplitude) has been applied to the phase distributions in order to emphasize thosexplained by either contribution.

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 9

averaging, as the actual jets change direction throughout the yearand the streams found at any month are substantially wider andswifter than the annual-mean values. We will see there is noclear difference between the SEIC (4�S) and the accompanyingcounter-current SICC (2�S), similarly for the NEIC (4�N) and theNICC (2�N); anticipating this result, we make no distinctionbetween these two adjacent currents, in Table 1 they are simplyreferred as the southern (SEIC) and northern (NEIC) intermediatecurrents.

On the other hand, the mean zonal flow is remarkably continuousacross most of the Atlantic Ocean (where data is available) at the CWand uNADW levels, with no significant changes in direction withlongitude (Fig. 5). It is possible that the CW and uNADW have indeedsuch a complex system of instantaneous zonal jets but, given thelimited amount of available data, we have no way to assure this.

Fig. 8. Time-latitude plots for (left panels) the magnitude of the surface winds (m s�1) anfrom data within 1� of (top panels) 33�W, (middle panels) 25�W, and (bottom panels) 5�Wby Hastenrath and Lamb (1977), is shown as a thick-dashed line; the thin-dashed lineslatitudes (every 2�). In order to emphasize the predominant periodicities, two full clima

Seasonal variability of the surface velocities

Monthly velocity fields

The seasonal evolution of the SEC may be completely character-ized by its northern/central branches (nSEC/cSEC), respectivelylocated north/south of the equator, i.e. there is no need to intro-duce an equatorial branch (Fig. 6). The cSEC remains relatively con-stant, with moderate maxima in spring and summer. The nSECintensifies in summer and early winter, while in spring it weakensand even reverses at the equator because of the surfacing of theEUC. In the northern flank of the nSEC, the NECC displays intenseseasonality, with large eastward velocities between April and Sep-tember. Several authors have endorsed the idea that these changesare largely controlled by the seasonally changing sea-surface trop-ical winds, associated to the latitudinal displacement of the Inter-

d (right panels) the zonal component of the sea-surface velocity (m s�1) as obtained. The climatological latitudinal location of the ITCZ at each longitude, as calculated

illustrate the temporal variation in the amplitude of the anomalies at the differenttological years are shown.

10 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

tropical Convergence Zone (e.g., Philander and Pacanowski, 1986;Hastenrath and Merle, 1987; Stramma and Schott, 1999).

The core of the cSEC is located all year long near 3–4�S, withmean velocities of about 0.2–0.4 m s�1. The maximum westwardvelocities occur in May–June and the minimum ones in October–November. The maps in Fig. 6 show the existence of variations ofabout 0.2 m s�1, associated to a 12-month period. The nSEC flowsnorth of the equator with characteristic speeds of 0.3–0.4 m s�1.Its core is located near 1–2�N most of the year but gets shifted to4–5�N in the western basin on March–April, when the NECC nearlydisappears; this northern shift is also associated to the surfacing ofthe EUC between 2�S and 3�N. In the central part of the equatorialAtlantic (30�W to 20�W), a maximum westward velocity isobserved in June (0.6 m s�1) and a minimum one in March(0.2 m s�1), in phase with the northward displacement of the ITCZ.The peak westward currents progress across the basin, from 5�W inMay to 35�W in July. A second minimum occurs simultaneously inSeptember–October over the whole central and eastern regions.

Another major feature is the quasi-permanent eastward flowfound east of 15�W and between 4�N and 7�N, along the Africancoast: the Guinean Current (GC). The monthly maps show how,between May and September, the origin of the GC gradually extendswest, effectively turning into the NECC to ultimately connect withthe NBC retroflection, in phase with both the meridional shift ofthe ITCZ and the intensification of the westward nSEC flow (Fig. 6).The August encounter of the NECC with the NBC concurs with thereported raise in transport and latitude (Fonseca et al., 2004) andwith the appearance of a double-core structure (Urbano et al.,2006, 2008). With the progressive southward shift of the ITCZ (Octo-ber to February) the NECC gradually weakens until it eventually dis-appears (April) accompanied by a northward shift of the nSEC.

In order to assess the amplitude and phase of the annual andsemi-annual periodicities, we have carried out a classical harmonicanalysis of the zonal velocity component (Appendix). The analysisshows high annual variability in the latitudinal band spanned by

Fig. 9. Surface velocity vectors for February, May, August and November, showing the poline). The velocity-scale vector has magnitude of 1 m s�1. (For interpretation of the referarticle.)

the seasonal journey of the ITCZ, with amplitudes close to 0.25–0.3 m s�1 in the central equatorial Atlantic (30�W to 20�W) andsubstantially higher (as much as 0.5 m s�1) in the western basin;moderate amplitudes are also present close to the equator in theeastern basin (top-left panel of Fig. 7). These high amplitudes ofthe seasonal signal, larger than the mean values (Fig. 5), are consis-tent with the observed seasonal flow reversal (Fig. 6). The contri-bution of the semi-annual signal is smaller than the annual oneeverywhere except in a near-equatorial band, between about 2�Sand 4�N and 0� and 30�W; a secondary band of significant semi-annual amplitudes is centered between 6�N and 8�N, west of30�W (top-right panels of Fig. 7).

The phase distributions illustrate the existence of several latitu-dinal bands, characterized by sharp latitudinal changes as opposedto high longitudinal coherence (middle panels, Fig. 7). Considerfirst the northernmost band, roughly between 4�N and 10�N. Thephase for the annual contribution is uniform across the whole trop-ical North Atlantic Ocean, with maximum eastward velocitiesbetween August (eastern basin) and November (western basin).In contrast, the phase for the semiannual contribution over thesame tropical region is split into two zonal bands, with maximumeastward velocities in February/August at about 8�N and June/December at about 5�N. The next band is roughly centered at2�N, with the maximum annual and semiannual variability takingplace at the western and eastern margins. In the western region thelargest eastward currents occur in April (annual) and in April/Octo-ber (semiannual), while in the eastern region they correspond toNovember (annual) and February/August (semiannual). Finally, inthe central and eastern portions of the southernmost band(roughly along 4�S) the annual contribution is most significant;the phase for the maximum positive values (giving rise to the min-imum westward velocity) correspond to October–November, withminor zonal changes.

The bottom panels of Fig. 7 illustrate the variance explained byeach contribution. The annual signal explains most of the variance

sitions of the ITCZ (blue line) and the boundary between the NECC and the nSEC (redences to color in this figure legend, the reader is referred to the web version of this

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 11

in the region between the southernmost and northernmost posi-tions of the ITCZ (above 75%). In contrast, the variance explainedby the semiannual signal is largest near the equator (typically over50%) and has low relevance in the extra-equatorial dynamics. Thesemiannual signal clearly draws the two branches of the SEC (nSECand cSEC).

Fig. 10. Latitudinal component of the surface velocity, as calculated along theboundary between the NECC and the nSeEC, plotted as a function of longitude andtime. A vector with a distance equivalent to 1 month has speed 0.1 m s�1.

Time-latitude variations at three selected longitudes

The results of the harmonic decomposition are consistent withanalogous analyses by Richardson and Walsh (1986) using shipdrifts and Lumpkin and Garzoli (2005) using surface drifters; thehigh level of agreement between these analyses, using data setsof different origin and spatial resolution, grants us further confi-dence on the high quality of the Argo-inferred velocities. Theessential mechanism behind the seasonal (annual and semi-annual) variability is the movement of the ITCZ, yet the complexannual and semi-annual amplitude and phase patterns point atthe existence of significant ocean feedbacks. In order to betterunderstand the mechanisms that bring out these spatial patterns,we next examine the time-latitude plots of the surface windsand currents at three different longitudes: 5�W, 25�W and 33�W.

Consider first the temporal changes in surface wind and oceanvelocity at 25 ± 1�W, in the central Atlantic (middle panels inFig. 8). The ITCZ shifts from 2�N in February–March to 12�N inAugust. The locus of the ITCZ is characterized by weak winds whilemaximum trades occur at a certain latitudinal distance, so thetropical winds in the northern/southern hemisphere reach maxi-mum values during winter/summer, therefore a predominantannual periodicity. Additionally, the northeastern/southeasterntrade winds alternate during winter/summer between 2�N and10�N, producing two wind-speed maxima per year; this leads tothe existence of a semi-annual wind forcing between 2�N and10�N. However, the response of the surface currents to the zonalband of semi-annual wind forcing (2�N to 10�N) is quite differentdepending on the latitude. The clearest semi-annual responseappears at 2–3�N; further north, up to at least 10�N, the zonal cur-rents are dominated by an annual, somewhat distorted, signal.Between 4�N and 10�N the eastward NECC appears at the time ofthe ITCZ northernmost extension, when the westward winds reachtheir minimum values (Fig. 8, middle panels). The deformation inthe annual wind cycle is the likely reason for the appearance of aweak, yet significant, semi-annual signal in the surface NECC(Fig. 7).

Strikingly, the semi-annual signal in the zonal surface currentsextends until 2�S, well beyond the southernmost limit in the semi-annual forcing (2�N). The semi-annual sea-surface response in theequatorial band, between 2�S and 2�N, is characterized by theintensification of the westward currents in June and December,and its slowing (down to zero) in April and October. The weak Aprilwestward currents are clearly related to the southernmost exten-sion of the ITCZ and the associated decline in the westward winds(Fig. 8, middle panels). The October minimum, however, is harderto explain as it occurs during a period of relatively intense winds: Aprogressive decrease in the intensity of the nSEC coincides with thesummer intensification of the NECC, up to a maximum in Augustand September (Figs. 6 and 8). A similar situation occurs betweenabout 5�S and the equator but advanced by about 1–2 months. ThecSEC decreases at the time of the weakest local westward winds(March) but it also does so at times of very intense winds (Septem-ber). Both factors, i.e. the annual evolution of the nSEC–NECC sys-tem and the existence of a semi-annual surface current near theequator, point at the potential importance of indirect wind forcing,through meridional Ekman transports. We will come back to thisissue in the next subsection.

Consider finally the surface winds and ocean velocities in theeastern (5 ± 1�W) and western (33 ± 1�W) Atlantic. In the westernbasin the situation is similar to the one described for 25�W,although the semi-annual contribution in the equatorial band issubstantially weaker (top panels, Fig. 8); the main difference withthe conditions at 25�W is the clear spring appearance of the EUC,coincident with a northward displacement (from the equator to2–4�N) of the nSEC. In the eastern basin the predominant changesin the intensity of the nSEC are both annual and semi-annual (bot-tom panels, Fig. 8). The GC is present all year long, flowing alongthe zonal coastline of the Gulf of Guinea, with February and, partic-ularly, August–September maxima. The two adjacent currents, theeastward GC and westward nSEC, are totally out of phase; the coin-cidence of the swiftest GC with the weakening of the nSEC, at timesof moderate winds, points at the existence of meridional recircula-tions, the sort of connections bringing upstream inertia that maybe disassociated with the local winds (bottom-right panel, Fig. 8).

Interaction nSEC–NECC–NBC

The seasonality in both atmospheric forcing and ocean responseis noteworthy but there is often no direct relation between them(Fig. 8). As explained above, wind forcing has a predominantannual cycle over most of the tropical and equatorial oceans.Semi-annual forcing also shows up between 2�N and 10�N butdominates only in a relatively narrow zonal band, roughly between2�N and 4�N, where it does explain the concurrent semi-annualzonal currents. Elsewhere there is no direct correspondencebetween atmospheric forcing and ocean response. Between 4�Nand 10�N, the prevailing annual wind forcing is reflected by a pecu-liar annual response in zonal currents: the winds are always west-wards yet the currents turn east in summer-fall, when the windsweaken. Between 6�S and 2�N, the forcing winds have annual peri-odicity but the currents display a semi-annual response: A weak-ening of the westward currents occur at the time of minimumtrade winds (centered in April) but a second reduction occurs cen-tered in October, when the westward winds reach maximumvalues.

The above behaviors point at the existence of indirect feedbackmechanisms in the equatorial and tropical surface ocean, specifi-cally between the nSEC and the NECC. In order to explore this con-nection, we define the boundary between both currents as thelatitude, at any given longitude, where the zonal componentchanges from the nSEC negative values (westwards) to the NECCpositive ones (eastwards). Using this definition, we can calculate

12 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

the location of the boundary every month, with the same 0.5� res-olution of the velocity grid. Finally, we compute the latitudinalcomponent of the velocity vectors along this boundary.

One first remarkable result is the location of the nSEC–NECCboundary: in January and February it is found only in the easternbasin but extends west progressively in time, reaching the coastof South America by July (Fig. 9). The interpretation is now clear:the NECC feeds from the westward nSEC, in regions where thetrade winds power the westward nSEC and its eastward recircula-tion as the NECC. The location of the ITCZ represents a naturalrupture in the westward current and, hence, sets the westernend of the NECC. When this boundary reaches America a connec-

Fig. 11. Contour maps for the latitudinal component of th

tion between the interior zonal currents and the NBC is estab-lished: the NBC feeds from the westward nSEC and theretroflection of the NBC at 7–8�N becomes the last connectionbetween the nSEC and the NECC (Fig. 9). The meridional velocityacross the boundary between the nSEC and the NECC is drawn inFig. 10 as a function of latitude and time of the year. The flow atthe boundary is permanently oriented north, endorsing the feedingof the NECC by the recirculating nSEC.

The idea of a maintenance and intensification of the NECC froma northern diversion of the SEC is further endorsed by the monthlyfields of surface meridional velocity (Fig. 11). The sequence illus-trates how the seasonal cycle in meridional velocities is associated

e monthly-mean velocities at the sea-surface (m s�1).

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 13

to the interaction between the nSEC and the NECC. From January toApril the ITCZ lays south of 6�N and the presence of the northeast-ern trade winds in the central and western basins help inhibit thenorthward nSEC component. Along with the April–May northwardITZC migration, the trade winds weaken at the location of thenSEC–NECC boundary, the nSEC latitudinal component develops,and the NECC gets intensified. The WSW–ENE orientation of theITCZ causes this to happen first at the eastern basin and to progresswestwards in time. If we compare the meridional velocities(Fig. 11) with the zonal velocities (Fig. 6), we observe how thebeginning of the development of the NECC, taking place in Mayin the eastern Atlantic, is anticipated by the growth of the meridi-onal velocities in the boundary between the nSEC and the NECC.

Fig. 12. Contour maps for the zonal component of th

The meridional velocities achieve its maximum around June–July,1–2 months ahead from the maximum development of the NECC,and show a sustained presence until January, precisely when theNECC decays. As a conclusion, the NECC is not controlled by theNBC retroflection, but rather by the westward extension of thenorthward nSEC bifurcation in phase with the seasonal ITCZ dis-placement; further, the NBC retroflection may be interpreted asthe westernmost expression of this recirculating structure.

One relevant consideration is the substantial decay of the NECCfrom August to December (Fig. 6), at times when the meridionaltransfer from the nSEC remains intense (Fig. 11). Such an activemeridional transfer would rather suggest the NECC to remainintense during this period, something that does not happen. It is

e monthly-mean velocities at 1000 dbar (m s�1).

14 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

possible that part of this meridional transport is indeed accommo-dated by the NECC itself. A partial explanation, however, comesfrom the contamination in the meridional velocities caused bythe transfer of momentum to the surface-most ocean metersthrough both direct wind drag and Stokes wave drift, as assessedthrough the floats surface trajectories (section ‘Estimating thevelocity errors’). The results in Fig. 3 do suggest that between Julyand December this momentum transfer may be responsible for ameridional velocity bias as large as 0.10 m s�1. If we subtractedsuch a northward velocity to the monthly meridional fieldsbetween July and December, we would obtain the divergent latitu-dinal fields that characterize the surface tropical cell (Wang, 2005;Perez et al., 2013) and the evolution of latitudinal velocity fieldsbetween 0� and 4�N would be consistent with the observeddecrease of the NECC between 4�N and 10�N (Figs. 6 and 11). Itis important to note that during the first half of the year the con-tamination in the meridional velocities is small (Fig. 3) so thespring and early-summer water transfer from the nSEC to the NECCremains well established.

In the central and western basins and very near the equator(within about 1�) a direct ocean response is at last found, withthe surface currents dominated by the annual signal. The currentsflow west all year long except for a short period near April, approx-imately coincident with the weakest winds: at this time the east-ward pressure force dominates the force balance and the EUCsurfaces (Brandt et al., 2006; Urbano et al., 2008).

Finally, between the equator and as far as 6�S, the predominantwind forcing has annual periodicity but the response has a signif-icant semi-annual signal. Of particular significance is the weaken-ing of the westward currents near October, coincident with thetime of the strongest trade winds. The monthly sequence inFig. 6 illustrates the decline of this westward jet (6�S to the equa-tor) between July and October, coinciding both with an intensifica-tion of the westward currents further west (NBC, one first pulse inAugust and a second one in November) and, to a lesser degree,south (6�S to 12�S, between October and December). This is illus-

Fig. 13. (Top panels) Amplitude (m s�1) and (middle panels) phase (months) distributiozonal component of the 1000-dbar velocity; a transparency mask (quadratic proportion toareas with high amplitudes. The bottom panels illustrates the fraction of the variance e

trative of the transfer of westward momentum via either down-stream fluxes or latitudinal Ekman transports.

AAIW velocity variability

Seasonal variability

The monthly 1000-dbar zonal velocity fields evidence the exis-tence of predominant zonal jets which experience large seasonalchanges (Fig. 12). These jets are substantially thinner than thesurface ones, typically 3–4� for the EIC and about 1–2� for thetwo EEJs (SEIC and NEIC, respectively centered at 4�S and 4�N). Itis remarkable that, despite the 200-km averaging latitudinal win-dow, the jets remain very clear, suggesting they may be even thin-ner and have much larger non-smoothed peak velocities. The AAIWzonal jets reverse direction throughout the year, with the maxi-mum westward flow in the western margin between Septemberand October and the maximum eastward flow in the central andeastern margin in December and January. As a consequence, theannual-mean AAIW fields are clearly inadequate to represent theactual intermediate zonal currents at any time of the year.

A remarkable feature is the alternation of direction betweenadjacent jets, i.e. the off-equatorial jets reverse their zonal velocitysuch that at any time they flow opposite to the EIC. The exact timeof the maximum currents at any latitude, however, changes withlongitude. The data shows the EIC reverses direction seasonally,flowing westward during summer and fall and eastward duringwinter and spring. This eastward flow, contrary to early reportson a westward flowing EIC (Ollitrault et al., 2006; Lankhorstet al., 2009), cannot be related to an unaccounted surface drift, asthe equatorial surface currents almost all year long run towardsthe west. The flow does not switch direction simultaneously acrossthe whole ocean, therefore difficulting the identification of singlejets. Nevertheless, it is clear that the five annual-mean jets foundwithin 4� of the equator (Fig. 5) are an artefact of the long timeaveraging, as most often there are only three zonally-coherent jets.

ns for the (left panels) annual and (right panels) semi-annual contributions to theamplitude) has been applied to the phase distributions in order to emphasize those

xplained by either contribution.

Fig. 14. Time-latitude plots for the zonal component of the 1000-dbar velocity(m s�1) as obtained from data within 1� of (top panel) 33�W, (middle panel) 25�W,and (bottom panel) 5�W. The thin-dashed lines illustrate the temporal variation inthe amplitude of the anomalies at the different latitudes (every 2�). In order toemphasize the predominant periodicities, two full climatological years are shown.

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 15

At AAIW levels, when the EIC is fully developed in the westwarddirection (September), the SEIC and the NEIC flow eastward oneach side of the EIC. About four or five months later (January–Feb-ruary) the EIC reaches its maximum eastward velocity, simulta-neous with the westward flow of the SEIC and followed by theNEIC (March–April). The northern and southern annual-meancountercurrents (NICC and SICC) do not show up in these represen-tations. The seasonal pattern commonly displays three intermedi-ate equatorial currents, with the equatorial waters streaming inone direction and the adjacent belts flowing in the opposite direc-tion. This system of three currents shifts latitudinally throughoutthe year: whilst the EIC stays on the equator, the adjacent jets donot keep their latitudinal position, being closest to the equatorwhen they flow east (summer and fall) and reaching further awaywhen they flow west (winter and spring). One conclusion is thatthere is no clear distinction between NEIC and NICC or between

SEIC and SICC, the appearance of counter flows being more a mat-ter of a seasonal reversal in the flow direction at latitudes between3� and 4� than a change of flow direction with latitude. A briefsummary of the predominant currents and their seasonal characteris presented in Table 1.

The character of the seasonal variability is clarified by the har-monic decomposition, carried out using the methodologyexplained in the Appendix (Fig. 13). The amplitude of the annualsignal reaches values as large as 0.07 m s�1 near the equator and0.04 m s�1 at latitudes near 3–4�, both greater than the annual-mean values (nowhere more than 0.03 m s�1; Fig. 5); even thesemiannual amplitude at many locations along the equatorexceeds the 0.03 m s�1 level. The phases display substantial vari-ability, difficult to interpret, but yet some features are revealed.The annual phases along the equator, between about 10�W andthe coast of Brazil, increase as we move westwards, reflecting awestward propagating disturbance; something similar happensat about 3–4�N, between 15�W and the coast of Brazil.

The variance explained by both the annual and semi-annualcontributions is shown in Fig. 13 (bottom panels). Over a large por-tion of the tropical ocean, the annual signal explains more than 60%of the variance. The semiannual contribution typically explainsmuch less variance, although in some reduced areas it may reachvalues in excess of 80%. Over most of the tropical ocean the twocontributions complement each other very well, explaining morethan 80% of the variance; this contrasts with the surface ocean,where such high levels of explained variance only correspond tothe equatorial band (5�S to 5�N) and the region between about5�N and 9�N, spanned by the seasonal motion of the ITCZ (Fig. 7,bottom panels). The distributions of explained variance, despitebeing very patchy, display some zonal structure that suggests thepossible propagation of anomalies along zonal bands; near theequator, where the currents are stronger, the patchiness decreases.

Finally, we may use time-latitude plots to examine the timing inmagnitude and location of the 1000-dbar zonal currents at 5�W,25�W and 33�W (Fig. 14), i.e. the same longitudes as done for thesurface waters (Fig. 8). The reversing character of the EIC showsup very clearly in the central basin (25�W and 33�W), with someeight months (April–November) of westward currents and onlyfour months (December–March) of eastward flow. At 33�W boththe SEIC and NEIC are out of phase with respect to the EIC, movingfrom about 2–3� to 4–5� as they change direction from eastward towestward. A similar situation, yet with much weakened NEIC andSEIC, shows up at 25�W. At 5�W, the extra-equatorial jets aremostly absent, with any reminiscent signature being in phase withthe EIC. A maximum westward flow takes place between Septem-ber–October at 3�S and 2�N, while the peak eastward velocitiesoccur in November–December close to the equator. These plotsconfirm that the large number of jets in the annual mean is causedby the way the zonal velocity estimates are distributed in the lat-itude-month space.

A comparison between Figs. 8 and 14, for the eastern and cen-tral equatorial band between 2�S and 2�N, shows the maximumeastward EIC to occur at times of peak westward sea-surface cur-rents: November–December (5�W) and January (25�W). This sug-gests the possibility of a baroclinic zonal behavior along theequator, with a westward propagating anomaly that originatesnear the eastern ocean boundary in late fall. This assents withthe idea of baroclinic Rossby waves in the deep equatorial Atlantic(Johnson and Zhang, 2003; Bunge et al., 2008), to be furtherexplored in next section.

Evidence of westward propagating waves at intermediate water levels

Figs. 12–14 confirm previous observations on the large seasonalvariability at AAIW levels (Schmid et al., 2001; Ollitrault et al.,

Fig. 15. Time-longitude diagrams at different latitudes (from left to right: 7�S, 3�S, 0�, 3�N, 7�N): (first row) zonal velocity anomaly, (second row) reconstruction with theannual contribution, (third row) reconstruction with the semiannual contribution, (fourth row) reconstruction with the annual + semiannual contributions. The time axesdisplay 24 months in order to provide a better view of the propagating waves; the color code gives the velocity in m s�1. The lines in the second and third rows illustrate thewave progression, with maximum positive and negative velocity values respectively separated by six and three months (double-arrow lines); the dashed lines at 7�S are onlysuggestive of a possible path, with the same phase velocity as at 7�N. (For interpretation of the references to colour in this figure legend, the reader is referred to the webversion of this article.)

16 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

2006; Bunge et al., 2008). In particular, the phase distributions inFig. 13 raise the possibility that some of the variability is relatedto westward propagating Rossby (planetary) waves, as proposedby several authors (Schmid et al., 2003; Johnson and Zhang,2003; Thierry et al., 2004; Brandt and Eden, 2005; Bunge et al.,2008; Brandt et al., 2011). The underlying hypothesis is that thewaves are generated at selected latitudes in the central or easternequatorial Atlantic Ocean during certain times of the year, possiblyas a result of either atmospheric or boundary forcing, from wherethey propagate westwards.

We may explore this idea looking at the zonal-velocity anoma-lies, calculated by subtracting the annual-mean zonal velocity tothe monthly velocity values, so that a 12 point seasonal time-seriesis obtained per grid point. The zonal velocity anomalies are calcu-lated at different latitudes (between 20�S and 20�N) as a functionof longitude (between 45�W and 0�) and time. The anomalies arecalculated using data from two grid elements (1� of latitude),therefore representing a latitudinal average over 300 km (close to3� of latitude). Here we have chosen to plot the results at the fivelatitudes (7�S, 3�S, equator, 3�N and 7�N) where a propagating sig-nal is best visible (Fig. 15). This possibly indicates there are someselected latitudes where an initial perturbation is more intense;in particular, this seems to occur at the locus of the predominantAAIW zonal currents (3�S, 0� and 3�N).

The annual and semiannual contributions to these anomalyfields have also been calculated (Fig. 15). The annual signal is suf-ficient to reproduce the gross pattern of the velocity anomaly field,in agreement with the results in Fig. 13. The annual pattern prop-agates west, with maximum speeds close to the equator and rap-idly decreasing with latitude, as expected for planetary waves. Atthe equator the annual wavelike propagation is indeed clear, takingsome 4 months to travel between 10 and 40�W, in gross agreementwith simulations by Thierry et al. (2004). This propagating pattern

also shows off clearly at 3�S, 3�N and 7�N; note that at 3� and 7�Nthe wave propagation appears to originate at about 15�W, or thelongitude of a southward extension of the NW African coast. A lin-ear fit to the maxima/minima annual propagating pattern giveszonal phase speeds of 0.32, 0.12 and 0.03 m s�1 at the equator,3�, and 7�, respectively; the values in both hemispheres are, tothe second significant digit, identical. The semiannual contribution,on the other hand, displays a westward motion only in the easternequatorial Atlantic basin (longitudes less than 15�W) and thesignal is less intense than the annual one; at 3�S no propagationis identified. The linear fit to the semiannual propagating valuesprovides phase speeds which are similar, within 20%, to the annualvalues.

Thierry et al. (2004) have shown that the zonal wind stressalong the equatorial band results in Rossby waves propagatingwestward and vertically from the eastern margin, with the firstmeridional mode affecting the AAIW level. The wave energy raysmay start at different depths and reach different longitudes at sub-sequent times, therefore resembling the propagating patternsobserved at any given depth. Thierry et al. (2004) carried out sim-ulations with realistic topography and winds and found that, atdepths of about 1000 m, the annual amplitudes are larger than0.04 m s�1 between 20�W and 35�W, with maximum values of0.06 m s�1, while the maximum semiannual amplitudes exceed0.04 m s�1 only between about 10�W and 20�W. These resultsare in good agreement with our observations, both in locationand magnitude (Figs. 13 and 15).

Conclusions

Our study illustrates, for the tropical Atlantic Ocean, the poten-tial of Argo data to investigate the velocity fields at the surface and

M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18 17

the float parking depths. In this application we have examinedthree parking depths: central waters (200 m), deep intermediatewaters (1000 m) and upper deep waters (1500 m). The Argo dataset is large enough to produce annual-mean fields at all depthsin the equatorial Atlantic Ocean, and even to construct a canonicalyear both at the sea surface and at the deep intermediate levels(AAIW). No attempt is made to search for interannual variabilityas most of the available data for the region (over 64%) has beenacquired since January 2008.

The data confirms the predominance of the zonal jets in theequatorial Atlantic at all levels, as previously reported by severalauthors. At the sea surface, the jets intensify seasonally but donot change direction, except for the summer appearance of theNECC. At the AAIW level, on the contrary, we find three jets thatchange direction throughout the year: one centered at the equator(the 4�-wide EIC) and two adjacent opposite currents (NEIC at 3–4�N and SEIC at 3–5�S). The annual averaging of these seasonalintermediate jets leads to the appearance of additional narrow jets,so the mean field is a poor description of the system at any time. Asimilar artefact may indeed be happening at the central and deeplevels but, from the limited amount of velocity data at these levels,we have no way to confirm it.

The available data allows also exploring the seasonal evolutionof the flow both at the surface and AAIW levels. At the surface, forexample, we find that the April intensification of the NECC starts inthe eastern Atlantic through a northern diversion of the nSECwhich progressively extends westwards until August, followingthe latitudinal displacement of the ITCZ. At latitudes between4�N and 10�N the forcing winds have both annual and semi-annualsignals that are reflected in the surface zonal currents. Between 2�Sand 4�N, the dominant atmospheric forcing is annual yet there is amajor semi-annual response which arises from the latitudinaldiversion of the flow, resulting in the seasonal appearance of theNECC. A novel view of the retroflection of the NBC, as the western-most limit of this recirculation between nSEC and NECC, arises.

At the AAIW level the major jets remain locked within less than5� of latitude from the equator, with both annual and semi-annualcontributions. These seasonal anomalies reach values of about0.1 m s�1, several times larger than the annually-averaged cur-rents, being responsible for the observed flow inversions. Theirzonal evolution is consistent with the speed of westward propagat-ing planetary waves, at latitudes which define the location of theAAIW zonal jets (3�S, 0�, 3�N). The propagation is slow enough tobe detected by our temporally-smoothed velocity data; this con-trasts with the situation observed at the surface level, where anyRossby wave would propagate too fast to be captured by ourmonthly velocity fields. Therefore, at the surface the observed sea-sonal variability does not reflect any transient wave but rather asuccession of states that respond to the seasonally-changing atmo-spheric forcing.

Acknowledgements

Funding for this work comes from the Ministerio de Ciencia eInnovación, Spain, through projects Transmisión de perfiladoresArgo en la Cuenca de Canarias (ARGO-Canarias, Ref. CTM2009-08462-E/MAR), Memoria Oceánica del Clima (MOC2, Ref.CTM2008-06438-C02-01) and Tipping Corners in the MeridionalOverturning Circulation (TIC-MOC, Ref. CTM2011-28867). MiquelRosell-Fieschi would like to acknowledge the Spanish Ministeriode Ciencia e Innovación for funding through a FPU Grant (Ref.AP2008-01879). We are very grateful to Marc Gasser, for his helpwith the harmonic analysis, and to our two reviewers for manyconstructive comments and useful suggestions which have servedto improve an earlier manuscript.

Appendix: Harmonic analysis

The observed zonal velocities u(t) are modeled through a classi-cal linear least-square adjustment to the following expression:

umðtÞ ¼ u0 þmðt � tmÞ þ ua cos½xaðt � taÞ� þ us cos½xsðt � tsÞ� ð1Þ

where u0 is the annual mean and m(t � tm) gives the linear trendwith respect to the center point of the time series, tm; the thirdand fourth terms represent the annual and semi-annual contribu-tions, with angular velocity x = 2p/T, amplitude u, phase u, andperiod T (the annual and semi-annual cycles are indicated by thea and s subscripts, with Ta = 12 months and Ts = 6 months). The timet is given in months, with ta and ts respectively corresponding to thetime of the year with maximum positive (eastward) annual andsemi-annual contributions (these times are related to the phasesthrough ta = ua/xa and ts = us/xs).

The adjustment is done to N repetitions of the 12-point timeseries, with N chosen such that an increase in the number of repe-titions does not lead to any significant change in the linear slope m;a value of N = 5 shows to work properly. The explained variance ateach point is calculated as the variance associated to the corre-sponding contribution divided by the observed variance, calculatedfrom the original time series u(t) � u0 �m(t � tm) and multipliedby 100.

References

Alford, M.H., 2001. Internal swell generation: the spatial distribution of energy fluxfrom the wind to mixed layer inertial motions. Journal of PhysicalOceanography 31, 2359–2368.

Arnault, S., 1987. Tropical Atlantic geostrophic currents and ship drifts. Journal ofGeophysical Research C92, 5076–5088.

Artamonov, Y.V., 2006. Seasonal variability of geostrophic currents in the AtlanticOcean according to altimetry data. Physical Oceanography 16, 177–187.

Bourlès, B., Gouriou, Y., Rémy, V., 1999. On the circulation in the upper layer of thewestern equatorial Atlantic. Journal of Geophysical Research 104, 21151–21170.

Bourlès, B., Andrié, C., Gouriou, Y., Eldin, G., du Penhoat, Y., Freudenthal, S., Dewitte,B., Gallois, F., Chuchla, R., Baurand, F., Aman, A., Kouadio, G., 2003. The deepcurrents in the eastern equatorial Atlantic Ocean. Geophysical Research Letters30, 8002. http://dx.doi.org/10.1029/2002GL015095.

Brandt, P., Eden, C., 2005. Annual cycle and interannual variability of the mid-depthtropical Atlantic Ocean. Deep-Sea Research I 52, 199–219.

Brandt, P., Schott, F.A., Provost, C., Kartavtseff, A., Hormann, V., Bourlès, B., Fischer, J.,2006. Circulation in the central equatorial Atlantic: mean and intraseasonal toseasonal variability. Geophysical Research Letters 33, L07609. http://dx.doi.org/10.1029/2005GL025498.

Brandt, P., Caniaux, G., Bourlès, B., Lazar, A., Dengler, M., Funk, A., Hormann, V.,Giordani, H., Marin, F., 2011. Equatorial upper-ocean dynamics and theirinteraction with the West African monsoon. Atmospheric Science Letters 12,24–30.

Bunge, L., Provost, C., Lilly, J.M., D’Orgeville, M., Kartavtseff, A., Melice, J.-L., 2006.Variability of the horizontal velocity structure in the upper 1600 m of the watercolumn on the equator at 10�W. Journal of Physical Oceanography 36, 1287–1304.

Bunge, L., Provost, C., Hua, B.C., Kartavtseff, A., 2008. Variability at intermediatedepths at the equator in the Atlantic Ocean in 2000–2006: annual cycle,equatorial deep jets, and intraseasonal meridional velocity fluctuations. Journalof Physical Oceanography 38, 1794–1806.

Carton, J.A., Katz, E.J., 1990. Estimates of the zonal slope and seasonal transport ofthe Atlantic north equatorial countercurrent. Journal of Geophysical Research95, 3091–3100.

Claret, M., Rodríguez, R., Pelegrí, J.L., 2012. Salinity intrusion and convective mixingin the Atlantic equatorial undercurrent. Science Marina 76S1, 117–129.

Csanady, G.T., 2001. Air–Sea Interaction, Laws and Mechanisms. CambridgeUniversity Press, Cambridge, 239pp.

Didden, N., Schott, F., 1992. Seasonal variation in the western tropical Atlantic:surface circulation from Geosat altimetry and WOCE model results. Journal ofGeophysical Research 97, 3529–3541.

Eriksen, C.C., 1981. Deep currents and their interpretation as equatorial waves inthe Western Pacific Ocean. Journal of Physical Oceanography 11, 48–70.

Eriksen, C.C., 1982. Geostrophic equatorial deep jets. Journal of Marine Research 40(Suppl.), 143–157.

Firing, E., Wijffels, S.E., Hacker, P., 1998. Equatorial subthermocline currents acrossthe Pacific. Journal of Geophysical Research 103, 21413–21424.

18 M. Rosell-Fieschi et al. / Progress in Oceanography 130 (2015) 1–18

Flagg, C.N., Gordon, R.L., McDowell, S., 1986. Hydrographic and current observationson the continental slope and shelf of the western equatorial Atlantic. Journal ofPhysical Oceanography 16, 1412–1429.

Fonseca, C., Goni, A., Johns, W.E., Campos, E.J.D., 2004. Investigation of the NorthBrazil Current retroflection and north equatorial countercurrent variability.Geophysical Research Letters 31, L21304. http://dx.doi.org/10.1029/2004GL020054.

Garzoli, S.L., Katz, E.J., 1983. The forced annual reversal of the Atlantic northequatorial countercurrent. Journal of Physical Oceanography 13, 2082–2090.

Garzoli, S.L., Katz, E.J., Panitz, H.J., Speth, P., 1982. In situ wind measurements in theequatorial Atlantic during 1979. Oceanologica Acta 5, 281–288.

Gouriou, Y., Reverdin, G., 1992. Isopycnal and diapycnal circulation of the upperequatorial Atlantic Ocean in 1983–1984. Journal of Geophysical Research 97,3543–3572.

Gouriou, Y., Bourlès, B., Mercier, H., Chuchla, R., 1999. Deep jets in the equatorialAtlantic Ocean. Journal of Geophysical Research 104, 21217–21226.

Gouriou, Y., Andrié, C., Bourlès, B., Freudenthal, S., Arnault, S., Aman, S., Eldin, G., duPenhoat, Y., Baurand, F., Gallois, F., Chuchla, R., 2001. Deep circulation in theequatorial Atlantic Ocean. Geophysical Research Letters 28, 819–822.

Gulev, S.K., Grigorieva, V., 2006. Variability of the winter wind waves and swell inthe North Atlantic and North Pacific as revealed by the voluntary observing shipdata. Journal of Climate 19, 5667–5685.

Gulev, S.K., Grigorieva, V., Ster, A., 2011. Global Atlas of Ocean Waves based on VOS.Online Atlas. <http://www.sail.msk.ru/atlas/>.

Hastenrath, S., Lamb, P.J., 1977. Climatic Atlas of the Tropical Atlantic and EasternPacific Oceans. University of Wiscons in Press, 112p.

Hastenrath, S., Merle, J., 1987. Annual cycle of subsurface thermal structure in thetropical Atlantic Ocean. Journal of Physical Oceanography 17, 1518–1538.

Hayes, S.P., Milburn, H.B., 1980. On the vertical structure of velocity in the easternequatorial Pacific. Journal of Physical Oceanography 10, 633–635.

Hormann, V., Lumpkin, R., Foltz, G.R., 2012. Interannual north equatorialcountercurrent variability and its relation to tropical Atlantic climate modes.Journal of Geophysical Research 117, C04035. http://dx.doi.org/10.1029/2011JC007697.

Hüttle-Kabus, S., Böning, C.W., 2008. Pathways and variability of the off-equatorialundercurrents in the Atlantic Ocean. Journal of Geophysical Research 113,C10018.

Johnson, C.G., Zhang, D., 2003. Structure of the Atlantic Ocean equatorial deep jets.Journal of Physical Oceanography 33, 600–609.

Katz, E.J., 1981. Dynamic topography of the sea surface in the Equatorial Atlantic.Journal of Marine Research 39, 53–63.

Lankhorst, M., Frantantoni, D., Ollitrault, M., Richardson, P., Send, W., Zenk, W.,2009. The mid-depth circulation of the northwestern tropical Atlantic observedby floats. Deep-Sea Research 56, 1615–1632.

Lebedev, K.V., Yoshinari, H., Maximenko, N., Hacker, P.W., 2007. YoMaHa’07:velocity data assessed from trajectories of Argo floats at parking level and at thesea surface. IPRC Technical Note 4 (2), 16p.

Lumpkin, R., Garzoli, S.L., 2005. Near-surface circulation in the Tropical AtlanticOcean. Deep-Sea Research I 52, 495–518.

Luyten, J.R., Swallow, J.C., 1976. Equatorial undercurrents. Deep-Sea Research 23,999–1001.

Merle, J., Arnault, S., 1985. Seasonal variability of the surface dynamic topography inthe tropical Atlantic Ocean. Journal of Marine Research 43, 267–288.

Metcalf, W.G., Voorhis, A.D., Stalcup, M.C., 1962. The Atlantic equatorialundercurrent. Journal of Geophysical Research 67, 2499–2508.

Ollitrault, M., Rannou, J.-P., 2013. ANDRO: an argo-based deep displacementdataset. Journal of Atmospheric and Oceanic Technology 30, 759–788.

Ollitrault, M., Lankhorst, M., Frantantoni, D., Richardson, P., Zenk, W., 2006. Zonalintermediate currents in the equatorial Atlantic Ocean. Geophysical ResearchLetters 33, L05605. http://dx.doi.org/10.1029/2005GL025368.

Perez, R.C., Hormann, V., Lumpkin, R., Brandt, P., Johns, W.E., Henandez, F., Schmid,C., Bourlès, B., 2013. Mean meridional currents in the centran and easternequatorial Atlantic. Climate Dynamics 41. http://dx.doi.org/10.1007/s00382-013-1968-5.

Philander, S.G.H., Pacanowski, R.C., 1986. A model of the seasonal cycle in thetropical Atlantic Ocean. Journal of Geophysical Research 91, 14192–14206.

Pollard, R.T., Millard, R.C., 1970. Comparison between observed and simulatedwind-generated inertial oscillations. Deep-Sea Research 17, 813–821.

Polonsky, A.B., Artamonov, Y.R., 1997. North equatorial countercurrent in thetropical Atlantic: multi-jet structure and seasonal variability. DeutscheHydrographische Zeitschrift 49, 477–495.

Ponte, R.M., Luyten, J., Richardson, P.L., 1990. Equatorial deep jets in the AtlanticOcean. Deep-Sea Research 37, 711–713.

Price, J.F., Mooers, C.N.K., Van Leer, J.C., 1978. Observation and simulation of storm-induced mixed-layer deepening. Journal of Physical Oceanography 8, 582–599.

Rascle, N., Ardhuin, F., 2009. Drift and mixing under the ocean surface revisited:stratified conditions and model-data comparisons. Journal of GeophysicalResearch 114, C02016.

Rascle, N., Ardhuin, F., Terray, E.A., 2006. Drift and mixing under the ocean surface: acoherent one-dimensional description with application to unstratifiedconditions. Journal of Geophysical Research 111, C03016.

Richardson, P.L., Fratantoni, D.M., 1999. Float trajectories in the deep westernboundary current and deep equatorial jets of the tropical Atlantic. Deep-SeaResearch II 46, 305–333.

Richardson, P.L., McKee, T.K., 1984. Average seasonal variation of the Atlanticequatorial currents from historical ship drifts. Journal of Physical Oceanography14, 1226–1238.

Richardson, P.L., Reverdin, G., 1987. Seasonal cycle of velocity in the Atlantic northequatorial countercurrent as measured by surface drifters, current meters, andship drifts. Journal of Geophysical Research 92, 3691–3708.

Richardson, P.L., Walsh, D., 1986. Mapping climatological seasonal variations ofsurface currents in the tropical Atlantic using ship drifts. Journal of GeophysicalResearch 91, 10537–10550.

Richardson, P., Arnault, S., Garzoli, S., Bruce, J.G., 1992. Annual cycle of the Atlanticnorth equatorial countercurrent. Deep-Sea Research I 39, 997–1014.

Schmid, C., Molinari, R.L., Garzoli, S.L., 2001. New observations of the intermediatedepth circulation in the tropical Atlantic. Journal of Marine Research 59, 281–312.

Schmid, C., Garaffo, Z., Johns, E., Garzoli, S.L., 2003. Pathways and variability atintermediate depths in the tropical Atlantic. In: Goni, G.J., Malanotte-Rizzoli, P.(Eds.), Interhemispheric Water Exchange in the Atlantic Ocean. Elsevier,Amsterdam, pp. 233–268.

Schott, F.A., Fischer, J., Stramma, L., 1998. Transports and pathways of the upper-layer circulation in the western tropical Atlantic. Journal of PhysicalOceanography 28, 1904–1928.

Schott, F.A., Dengler, M., Brandt, P., Affler, K., Fischer, J., Bourlès, B., Gouriou, Y.,Molinari, R.L., Rhein, M., 2003. The zonal currents and transports at 35�W in thetropical Atlantic. Geophysical Research Letters 30, 1349. http://dx.doi.org/10.1029/2002GL016849.

Schott, F.A., McCreary, J.P., Johnson, G.C., 2004. Shallow overturning circulations ofthe tropical–subtropical oceans. In: Wang, C., Carton, J., Xie, S.-P. (Eds.), Ocean–Atmosphere Interaction and Climate Variability. AGU, Washington, pp. 261–304.

Schouten, M.W., Matano, R.P., Strub, T.P., 2005. A description of the seasonal cycle ofthe equatorial Atlantic from altimeter data. Deep-Sea Research II 52, 477–493.

Schudlich, R.R., Price, J.F., 1992. Diurnal cycles of current, temperature, andturbulent dissipation in a model of the equatorial upper ocean. Journal ofGeophysical Research 97, 5409–5422.

Send, U., Eden, C., Schott, F., 2002. Atlantic equatorial deep jets: space–timestructure and cross-equatorial fluxes. Journal of Physical Oceanography 32,891–902.

Soloviev, A., Lukas, R., 2006. The Near-Surface Layer of the Ocean: Structure,Dynamics, and Applications. Springer, Dordrecht, 579pp.

Stramma, L., Schott, F., 1999. The mean flow field of the tropical Atlantic Ocean.Deep-Sea Research 46, 279–303.

Stramma, L., Fischer, J., Reppin, J., 1995. The North Brazil undercurrent. Deep-SeaResearch I 42, 773–795.

Talley, L.D., 1996. Antarctic intermediate water in the South Atlantic. In: Wefer, G.,Berger, W.H., Siedler, G., Webb, D. (Eds.), The South Atlantic: Present and PastCirculation. Springer, Berlin-New York, pp. 219–238.

Taylor, J., 1997. An Introduction to Error Analysis: The Study of Uncertainties inPhysical Measurements, second ed. University Science Books, Sausalito,California, 328pp.

Thierry, V., Treguier, A.M., Mercier, H., 2004. Numerical study of the annual andsemi-annual fluctuations in the deep equatorial Atlantic Ocean. OceanModelling 6, 1–30.

Tsuchiya, M., 1986. Thermostads and circulation in the upper layer of the AtlanticOcean. Progress in Oceanography 16, 235–267.

Urbano, D., Jochum, M., da Silveira, I., 2006. Rediscovering the second core of theAtlantic NECC. Ocean Modelling 12, 1–15.

Urbano, D.F., De Almeida, R.A.F., Nobre, P., 2008. Equatorial undercurrent and northequatorial countercurrent at 38�W: a new perspective from direct velocity data.Journal of Geophysical Research 113, C04041.

Wang, C., 2005. Subthermocline tropical cells and equatorial subsur face currents.Deep-Sea Research I 52, 123–135.

Yang, J., Joyce, T.M., 2006. Local and equatorial forcing of seasonal variations of thenorth equatorial countercurrent in the Atlantic Ocean. Journal of PhysicalOceanography 36, 238–254.