Anticyclonic and cyclonic eddies of subtropical origin in the subantarctic zone south of Africa

3.1. Core Properties and Origin of Eddy M

[12] The following description of the core properties of eddy M rests on expanded views of vertical property distributions in the region of the two eddies and at depths 0–2000 m. They show the potential temperature (θ), salinity, squared Brunt-Väisälä frequency and dissolved oxygen in Figure 5, and CFC-11, CFC-12, nitrate and silicate in Figure 6. Although the rotational velocity signature of eddy M was sampled at 3 hydrographic stations (stations 46, 47, 48, see 5), the eddy core properties were only observed at station 47. In order to better contrast the core properties with those of surrounding waters, Figure 7 displays the salinity and oxygen profiles, and the neutral density (γ)-salinity and neutral density-dissolved oxygen curves at this eddy core station and at the adjacent stations.

[13] The core of eddy M is characterized by a 500 m-thick homogeneous layer, between ∼80 m and ∼580 m, itself capped by a 30 m surface mixed layer and the intervening pycnocline (Figures 5 and 7). The homogeneous layer has properties θ = 11.8°C, S = 35.15, O₂ = 257.7 μmol kg⁻¹, and γ = 26.82 kg m⁻³ (σ₀ = 26.75 kg m⁻³). With a core temperature lower than 12°C, M is among the coldest of comparable eddies sampled in the region, very similar to Agulhas ring R1 reported by Arhan et al. [1999] (Table 1), which was observed 4 degrees of latitude farther north in March 1995. Eddy E3 described by Gladyshev et al. [2008], also sampled farther north during the 2004 GoodHope cruise, was slightly warmer, with homogeneous core temperatures of 12.5°C. Although all three eddies of Table 1 have similar high core oxygen values around 260 μmol kg⁻¹, the homogeneous core of eddy M is not as distinguishable in vertical sections of oxygen (Figure 5d) as its two counterparts (and as in temperature and salinity sections, Figures 5a and 5b). The reason for this lies in the surrounding oxygenated subantarctic waters. Eddies R1 and E3 of Table 1, being surrounded by less oxygenated subtropical water (∼220 μmol kg⁻¹), appeared as well-defined positive anomalies in their vertical oxygen distributions [Arhan et al., 1999; Gladyshev et al., 2008]. For eddy M, a contrast between the core and neighboring stations is better seen in the dissolved oxygen profiles themselves (Figure 7b), the eddy side profiles being more irregular, and indicative of interleaving, than the central one. The oxygen signature of M is also more visible when using the oxygen saturation, instead of dissolved oxygen concentration. In Figure 8 showing the saturation at 300 m depth at all BGH stations, the core station of M stands out with a 0.98 value, well above neighboring rates around 0.85, and similar to the values computed for R1 and E3 (Table 1). Such oxygen saturations, and the thick homogeneous layer are indications of convection and ventilation of the eddy during the previous winter.

[14] The along-track vertical distributions of CFC-11 and CFC-12 (Figures 6a and 6b) also exhibit the water homogenisation at 100–600 m in eddy M, and pronounced low anomalies at the core stations, relative to values at the surrounding stations. The nitrate and silicate signatures of eddy M (Figures 6c and 6d) are characterized by low core values also indicative of winter convection. In the sharp pycnocline underlying the convected layer (Figure 5c), the core of eddy M is marked by low anomalies of dissolved oxygen, CFC-11, and CFC-12 concentrations at γ ∼ 27.1 (700–800 m depth; Figures 5d, 6a, and 6b). The high salinities of the eddy homogeneous layer (>35.0), and the relative low oxygen (∼225 μmol kg⁻¹) and CFC concentrations at γ = 27.1 kg m⁻³ reflect an eddy origin at subtropical latitudes, the only domain where such values are found at similar densities (Figures 2a and 2b). These properties most certainly make this eddy an Agulhas ring, as was the case for the similar structures mentioned above (Table 1). Still deeper in the core of M, the salinity minimum of the Antarctic Intermediate Water (AAIW) is 34.25, also slightly above the ∼34.2 values of the neighboring stations. Gordon et al. [1992] observed that two AAIW varieties exist in this region, namely, AAIW recently formed in the Atlantic sector (characterized by salinity minima lower than 34.3), and older AAIW from the Indian Ocean sector conveyed south of Africa by the Agulhas Current (with salinity minima higher than 34.3). Although the AAIW salinity minimum in M bears the signature of the Atlantic sector, its high anomaly relative to the neighboring stations is also indicative of an Indian Ocean influence, and of an eddy origin in the Agulhas Current system.

[15] The above water property considerations on the origin of eddy M are best illustrated in the along-track distributions of salinity and dissolved oxygen plotted with neutral density as the vertical coordinate (Figures 9a and 9b). On Figures 9a and 9b the upper boundary provides the surface density and, slightly shifted (by 0.03 kg m⁻³), the pluses show the densities at the base of the surface mixed layer. Most ACC fronts can be detected in Figure 9 from steeper slopes of this upper limit. The STF stands out particularly at ∼38°S and also, though less pronounced, its southern expression at ∼42°S, and the Polar Front at ∼50°S. The SAF, on the other hand, is not marked by any increased slope, and even corresponds to a plateau at 44°S–45°S, in accordance with Rintoul and Trull [2001] (and Gladyshev et al. [2008], along the GoodHope line), who pointed out the near-surface compensating effects of temperature and salinity on density across the SAF.

[16] The homogeneous core of M being represented by one point on such plots, is hardly visible. Its overlying pycnocline, on the other hand, appears as salty and moderately oxygenated relative to the neighboring stations. These contrasts most certainly reflect northward entrainment of fresher and oxygenated SAF water around the anticyclone. Deeper, the above mentioned oxygen minimum at γ = 27.1 kg m⁻³ appears as the southernmost manifestation of low oxygen values at densities 27.0–27.1 kg m⁻³. Farther north this vertical oxygen minimum forms a continuous southward pointing tongue which emanates at the continental slope and protrudes well into the SAZ to ∼41°S. Given the intense mesoscale activity of the region, it is unlikely that this low oxygen tongue at γ = 27.1 kg m⁻³ is associated with a large scale flow. Very likely, it results from eddies akin to M penetrating the SAZ and loosing a part of their intermediate, poorly oxygenated layer there.

[17] The AAIW layer is centered on γ = 27.35 kg m⁻³ (Figure 9a), apparently also subject to mesoscale influences. Eddy M penetrates the upper part of this layer vertically at a short distance north of its subduction latitude. The AAIW salinity minimum at the eddy center (S∼34.25), although reflecting a dominant South Atlantic origin, is higher than at the stations next to the eddy core on its northern side, where values lower than 34.2 are observed. Such values being characteristic of the region south of the SAF indicate that, also at these depths (∼700 m), M is advecting northward (and helping to subduct) AAIW from the Polar Frontal Zone south of the SAF, into the SAZ.

3.2. History of Eddy M

[18] The trajectory of eddy M (Figure 10a) was determined using the same method as employed by Dencausse et al. [2010] (itself drawn from Doglioli et al. [2007]) to track Agulhas rings. Basically, it rests on wavelet analyses of the weekly SSH anomaly fields, a technique which provides the central positions and contours of eddies. If a structure is observed at a given date, a similar anomalous SSH pattern observed on the following week, which has its center within the contour of the former, is assumed to represent the same eddy. Repeating the process until no anomalous SSH pattern is found within the preceding contour provides the eddy trajectory. Having thus determined the eddy trajectory, the time series of its translation velocity could be obtained, and is represented in Figure 10b. The following description of the eddy history relies on Figures 10a and 10b, and on individual SSH maps at the referred dates, not shown here.

[19] Eddy M could be tracked back to 20 May 2007, when an Agulhas ring detached from the Agulhas Current retroflection at ∼40°S-18°E, just east of the northeastern end of the Agulhas Ridge (location 1 in Figure 10). As frequently occurs [Dencausse et al., 2010], the Agulhas ring was rapidly (June 2007) split in two parts by the topography, eddy M being the southern part. Eddy M moved rapidly southwestward, and was further divided under the apparent stirring effect of neighboring cyclones. These two splitting events are certainly a major cause of the small radius of M (∼70 km) as compared with usual values of 150–200 km (illustrated by the newly shed ring at 39°S–14°E in Figure 10a). On 25 July the eddy reached its first southern extreme position (location 2, 43°15′S–13°14′E) where it stayed for 2–3 weeks before resuming motion in a northwestward direction. This southernmost position for eddy M, and change of course, coincided with the eddy encounter with a northward meander of the SAF. The new (northwestward) part of the trajectory involved a very slow movement. Apparently, the eddy was blocked by the Agulhas Ridge, which was crossed only on 19 December 2007 (location 3, 41°36′S-11°34′E). Afterwards, a faster anticyclonic displacement led the eddy back over the bathymetry on 20 February 2008, where it was sampled during the BGH cruise (location 4, 43°19′S-8°14′E). At that period, it was slowly moving eastward, an indication of its entrainment by the neighboring SAF. Eddy M stalled a long time over the ridge, which it only left in mid-May 2008 toward the northwest. The surface signature of the eddy was lost in mid-August 2008 at 40°26′S-4°47′E, close to the STF where it most likely subducted below lighter water.

[20] Overall, eddy M could be tracked 15 months, and was 9.5 months old when it was observed during BGH. It belongs to what Dencausse et al. [2010] called the Agulhas ring southern family. Those eddies, after formation at the Agulhas Current retroflection, enter the region to the southeast of the Agulhas Ridge. Putting the emphasis on rings likely subject to significant property alterations before entering the South Atlantic, Dencausse [2009] paid a particular attention to those rings from the southern family which spend more than 4 months south of the Agulhas Ridge before eventually crossing it northwestward. A total of 14 such rings were detected in the SSH maps from the period October 1992 to January 2007 (Table 2), whose trajectories are drawn in Figure 11a. Their number amounts to about 50% of the whole southern family, the other half either dissipating in the SAZ without crossing the ridge, re-integrating the Agulhas Current retroflection or the Agulhas Return Current, or crossing the ridge rapidly close to its northeastern end. In order to assess the representativeness of eddy M, we present here the outlines of these 14 ring behaviors as inferred from their SSH tracking. About half of them formed from the subdivision of a newly spawned Agulhas ring at the northeastern end of the Agulhas Ridge. The other ones detached from the Agulhas Current Retroflection itself, in situations when it protruded westward in a more southern position than usual. The initial southward displacements of these eddies in the SAZ were rapid, and were only stopped when the eddies hit the SAF, which appears as an impassable barrier for these structures. The westward component of the eddy displacements is always slowed down when approaching the front, and even sometimes reverses into an eastward motion, causing a loop in the trajectories of several eddies (Figure 11a). While the first encounter of eddy M with the SAF does not exhibit any such loop (location 2 in Figure 10a), the second encounter does (location 4).

[21] After interacting with the SAF, the eddies generally move northwestward as did M, also with a lower velocity likely influenced by the vicinity of the Agulhas Ridge. Crossing of the Agulhas Ridge generally occurs at 12°E–13°E (Figure 11b), where a 2000–2500 m deep passage exists. Eddy M itself crossed the ridge at 11°34′E. Most trajectories could not be traced very far beyond the Agulhas Ridge. The longest ones terminate at 37°S–38°S, the very latitude range of the STF, where the eddies might be either dissipated or subducted [Herbette et al., 2004]. Several of them had their trajectories interrupted as they merged with another anticyclone. This is another way of subducting, for, as they align with another (warmer) structure, the cooled eddies are capped by its lighter core. The disappearance of the eddy surface signatures is therefore not necessarily associated with their total dissipation. Observation of Agulhas rings with subducted (subsurface) homogeneous cores in the Cape Basin has been reported by McDonagh and Heywood [1999], Garzoli et al. [1999] and Arhan et al. [1999]. The erosion of the eddies of Figure 11, as measured from the decay of their SSH anomalies, was generally intense during the first stage of their lives, suggesting that it results from the modification of their core properties, rather than from interaction with the SAF, or crossing of the Agulhas Ridge.

[22] The core alterations may be significant, as Agulhas ring core temperatures initially around 17°C [e.g., Gordon et al., 1987; van Aken et al., 2003] may decrease to less than 12°C (as in M). They likely result from a combination of heat loss to the atmosphere and mixing with the surrounding subantarctic water [Arhan et al., 1999]. A model study by Donners et al. [2004] illustrates how convection in the eddy core generates a shallow thermohaline circulation between the core and outer domain, which could enhance the lateral exchanges. The low temperatures of eddies M, R1, and E3 of Table 1 should be related to the fact that they all spent a whole autumn and winter period south of 42°S. Table 2 shows that, while on average the eddies spent about three months in autumn/winter south of the Agulhas Ridge, 4 of them, including R1 and E3, spent more than 6 months south of the ridge. The averaged southernmost latitude reached by the cooled Agulhas rings is 43°12′S, about the latitude reached by eddy M on its two southern excursions (43°15′S and 43°20′S). This latitude varies significantly, however (from ∼42°S to ∼44°40′S), depending on whether the eddies collide with the SAF at developed northern, or southern, meanders.

[23] Except for the fact that eddy M encountered the SAF twice, this structure appears rather typical of the eddy family represented in Figure 11. The eventful lives of these eddies are remarkable. They often stem from the subdivision of an Agulhas ring, interact with the SAF, experience intense air-sea interaction, blocking by the Agulhas Ridge and, after eventually crossing it, often subduct on re-entering the subtropical domain they emanated from.

3.3. Eddy M Kinematics and Water Trapping

[24] Velocities in eddy M were measured during the BGH cruise using the VM-ADCP, the L-ADCP installed on the water sampling rosette frame, and can be inferred from geostrophy. Figure 12, showing the vertical distribution of the velocity component perpendicular to the GoodHope line at latitudes where eddies S and M were observed, also gives the locations of the hydrographic stations relative to the two structures. It so happened that station 47 was very close to the center of eddy M, and stations 46 and 48, located respectively at 55 km and 47 km from station 47, were only slightly on the outer side of the eddy lateral velocity maxima. Using the 5 km averaged VM-ADCP velocities, the azimuthal velocity maxima of eddy M can be located at ∼45 km from the eddy center. Because the eddy is adjacent to the SAF in the south, inferring its outer radius on this side is difficult. On the northern side, a sharp velocity gradient (Figure 4) locates the outer radius at ∼70 km. Vertically the velocities were highest at 190 m depth, with speeds of 1.2 m s⁻¹ westward, and 0.98 m s⁻¹ eastward. The geostrophic velocities computed relative to 2500 dbar for station pairs 46–47 and 47–48 (Figure 13a) suggest that rotational velocities are detectable as deep as ∼2000 m. The geostrophic velocity maxima, around 0.45 m s⁻¹ in both directions are, as expected from the station locations near the eddy center and velocity maxima, about half those inferred from the VM-ADCP. The eddy rotational volume transport, computed from geostrophy, was found equal to 28 Sv.

[25] The L-ADCP profiles at stations 46 and 48 (Figure 14), close to the positions of the lateral velocity maxima, as expected show westward (eastward) highest velocities 1 m s⁻¹ (0.8 m s⁻¹), only slightly lower than the 1.2 m s⁻¹ (0.98 m s⁻¹) neighboring maxima. Figure 14 also confirms the ∼2000 m thickness of the rotational motion, a value comparable to the depth of the Agulhas Ridge at 11°34′E where eddy M first crossed this ridge, and necessarily left its deeper part behind. When observed during its second encounter with the Agulhas Ridge, eddy M was located above a ∼2700 m culminating peak (Figure 4) and was therefore underlaid by a few hundreds of meters of non-rotating water. This corroborates our previous assumption that its southward motion was not stopped by the Agulhas Ridge, but by the SAF. Note that Figure 14 reveals a striking difference in the deep velocities at stations 46 and 48. At station 46, located above the northern flank of the Agulhas Ridge, the velocities nearly vanish at depths greater than ∼2000 m. On the other hand, at station 48 located south of the ridge they increase downward at this depth, then keep values of 0.2–0.35 m s⁻¹ to the vicinity of the bottom at ∼4500 m. These velocity observations recall a similar eastward flow, likely associated with the SAF, present in a schematic of the near-bottom circulation in that region, by Gladyshev et al. [2008]. A passage in the Agulhas Ridge, located slightly upstream at ∼6°E (Figure 4) explains its presence against the southern flank of the ridge in the BGH data.

[26] Flierl [1981] demonstrated that water can be trapped in an eddy when its azimuthal speed exceeds its drift speed. First considering eddy M on 28 February 2008 when observed during BGH, its translation speed was about 0.03 m s⁻¹ (Figure 10b), implying deep water trapping. Assuming, as observed from the VM-ADCP measurements at 0–400 m (Figure 13a), that the maximum azimuthal speed is twice the geostrophic velocity, a trapping depth around 2000 m is inferred at that time. The eddy translation velocity, however, varied significantly since the structure was formed (Figure 10b). During the two periods of southwestward motion (parts 1–2 and 3–4 in Figure 10a) the eddy moved at speeds of around 0.1 m s⁻¹ with weekly peaks at 0.2–0.3 m s⁻¹. The eddy azimuthal velocities themselves likely decreased since it was formed. For a first approximation of trapping depths in M, we here assume that the eddy always had the maximum velocity profile estimated at its observation date. This is a likely conservative hypothesis producing underestimates of the trapping depths. This assumption leads to trapping depths of about 900 m, 1150 m, and 1500 m for drift velocities of 0.3 m s⁻¹, 0.2 m s⁻¹, and 0.1 m s⁻¹, respectively. With these values, and given the 0.3 m s⁻¹ eddy drift velocity shortly after its formation, we can expect trapping of water from the formation region down to ∼900 m. Given the ∼0.2 m s⁻¹ drift speed in February 2008 shortly before the observation date, we may also expect trapping of water from the northern part of the SAZ between ∼900 m and ∼1150 m, and just local water below.

[27] Looking for confirmation of these inferences in water properties, we show in Figure 15 the along-track distributions of potential temperature, salinity and dissolved oxygen on four isopycnals: γ = 27.1 kg m⁻³ which is central to the low oxygen anomaly at 700 m–850 m in M, γ = 27.35 kg m⁻³ which marks the AAIW core layer at ∼1000 m in M, and γ = 27.5 kg m⁻³ (27.7 kg m⁻³) at depth ∼1200 m (∼1450 m) in eddy M. Property extrema indicative of water trapping in the eddy exist at γ = 27.1 kg m⁻³, γ = 27.35 kg m⁻³, and γ = 27.7 kg m⁻³, not at greater densities, suggesting some trapping of non local water down to ∼1500 m. The brief 0.2 m s⁻¹ peak of drift velocity in early February 2008, therefore, apparently did not totally separate the eddy from its deeper-than-1200 m part. The only indication of its possible effect is the absence of property anomalies at 27.5 kg m⁻³ in Figure 15. Between the surface mixed layer and ∼600 m in eddy M, the homogeneous water draws its high temperature and salinity from the eddy formation region, with subsequent modifications resulting from surface cooling and associated lateral exchanges. Below the homogeneous layer, the γ = 27.1 kg m⁻³ curves in Figures 15a–15c show that the 600–850 m oxygen minimum layer also originates in the eddy formation area. The eddy properties at this density are, indeed, only found north of 37°S, close to where the STF criteria were met along the BGH line. Still deeper, the weaker property anomalies associated with eddy M at γ = 27.35 kg m⁻³ and γ = 27.7 kg m⁻³ are observed at 39°30′S–41°S, just north of eddy S, in the northern part of the SAZ. Information on water trapping in eddy M gained from water properties therefore corroborates the inferences drawn from the kinematics above 1200 m. At 1200–2000 m, recent trapping is inferred from water properties, not from the kinematics. An ARGO profiler launched during BGH in the core of eddy M and drifting at 1000 m, itself suggests only a loose trapping at and below 1000 m. Its first temperature and salinity profiles, realized two days after launching at the core station of M, reproduced those of that station. However, the subsequent ones (10 days later and ∼100 km from the launching point) showed no trace of the homogeneous layer, and the next ones (20 days later) had well-recognizable SAF characteristics. This rapid expulsion of the profiler from eddy M might have been caused by the ∼0.1 m s⁻¹ maximum of eddy drift speed on 5 March 2008 (Figure 10b), combined with float descents to 2000 m before each profile.

3.4. Air-Sea Fluxes Over Eddy M

[28] Measuring surface heat fluxes over an Agulhas eddy centered at ∼42°S-20°E in June–July 1993; Rouault and Lutjeharms [2000] found that the structure was a substantial source of heat for the atmosphere, as a result of latent heat fluxes of 500 W m⁻², and sensible heat fluxes of 350 W m⁻². The eddy had somewhat the same origins as eddy M, but was likely younger, and sampled during the opposite season. It had a surface temperature up to ∼17°C and a 250 m-thick mixed layer, whereas eddy M has surface temperatures slightly below 14°C, and a ∼50 m summer mixed layer itself separated from the underlying ∼12°C homogeneous core by a seasonal thermocline. We here present the surface heat flux computations over eddy M, thus allowing for a comparison with the Rouault and Lutjeharms [2000] eddy, and with the cyclonic eddy S (8). The surface turbulent fluxes of momentum, sensible and latent heat fluxes (hereafter SHF and LHF, respectively), were computed using the COARE bulk air-sea flux algorithm [Fairall et al., 2003] applied to the atmospheric measurements performed on board.

[29] The mean in situ Bowen ratio (between sensible and latent heat fluxes) along the summertime BGH cruise (0.041) indicates that a major part of the available energy at the ocean surface is passed to the atmosphere through evaporative processes. The evaporative fraction (0.96 over the whole cruise) is then appropriate for representing the relative contributions of the turbulent energy fluxes to the energy budget. During that summer, the LHF was mainly positive (not shown), indicating that along the cruise track the ocean mainly lost energy through evaporative turbulent processes, which are highly dependent on wind velocity.

[30] The sea surface skin temperature curve (Figure 16) reveals a well-defined high anomaly between 42.9°S (28 February) and 43.8°S (29 February), exactly above the location of eddy M. At these latitudes the air-sea temperature gradient is inverted, as compared to outside the eddy, and induces a positive SHF of 15 W m⁻² on average (reaching up to 23.6 W m⁻²) from the ocean to the atmosphere. In the same latitude interval, the LHF also presents a high average value of 180 Wm⁻² (reaching up to 247 W m⁻²), and strongly contrasts with those outside the eddy.

[31] The ocean heat content between the surface and 100 m (not shown) revealed a positive energy anomaly of 6 10⁵ kJ/m² within eddy M, allowing a long-term and powerful release of energy as long as the structure remained south enough of the STF. The ocean isotherms within the top 60 m revealed a warm anomaly within the eddy core (whose surface imprint is visible in Figure 16), even though a pronounced seasonal thermocline separated the summer surface mixed layer from the ∼12°C homogeneous layer below. Although such thermoclines sometimes blur the surface expressions of underlying mesoscale anomalies, this was not the case for eddy M, a likely consequence of a tight surface water trapping. Provided that the eddy was not intersected too far from its center, its surface temperature imprint in Figure 16 suggests a surface trapping radius of 65 km, very close to the radius (70 km) inferred above from the rotational velocities of Figure 4. The sharp outer cut-off of these velocities, visible on the northern side of the structure, explains the near-coincidence of the two radius estimates.

[32] The wind magnitude (purple in Figure 16) exhibits an abrupt increase above the northern side of eddy M, which might have reflected a wind increase associated with the sea surface temperature anomaly of the structure [Small et al., 2008]. This high wind episode, apparent at 42.6°S-43.4°S, and the other one at the southern border of the eddy (43.7°S-43.8°S), however, are related to the atmospheric synoptic eastward flux rather than to local eddy effects. Indeed, two low pressure systems crossed the cruise track flowing eastward during that period, the first one south of eddy M causing the first above mentioned wind increase, and the second one with a center closer to the structure producing the second wind peak. Such an interpretation was corroborated by an examination of the rotation of the wind direction, and the pressure variations, during the events.

[33] It was expected that the 9.5 month old eddy M, which was sampled during summer, would not be associated with as high heat fluxes to the atmosphere as the younger Agulhas ring sampled during winter by Rouault and Lutjeharms [2000]. The 200 W m⁻² accumulated LHF and SHF computed for eddy M during a period of averaged wind speed ∼15 m s⁻¹, is nevertheless quite significant. Such a summer value is partly due to the atmospheric environment of the structure, but also to the eddy capacity, to keep an important energy reserve in its uppermost tens of meters, despite the previous cooling of its upper core and the presence of a seasonal thermocline.

4.1. Core Properties and Origin of Eddy S

[34] This description of the core properties of eddy S rests on the same diagrams as used for eddy M (Figures 2, 5, and 6), and on Figure 17 showing the salinity and oxygen profiles and the density-salinity and density-oxygen curves at the eddy core station 39 and at the neighboring stations 37 and 40. The two latter stations sampled the eddy rotational velocities (Figure 12), but were outside its hydrographic core.

[35] The upper 200 m of eddy S present water properties which are somewhere between those of the subtropical and subantarctic domains, as illustrated by values of 10.97°C and 34.56 psu at 100 m, which fall in the temperature and salinity ranges defining the STF of Orsi et al. [1995]. Between 200 m and 400 m, eddy S is characterized by vertical maxima in potential temperature (10.28°C at 260 m and γ = 26.85 kg m⁻³), salinity (34.845 at 265 m and γ = 26.89 kg m⁻³) and by pronounced oxygen minimum (165.5 μmol kg⁻¹ at 335 m and γ = 27.07 kg m⁻³) and CFC concentration minima. These extrema coincide with a maximum of the Brunt-Väisälä frequency (Figure 5c) at the transition between an overlying depression and underlying uplift of isopycnals. Still deeper, at 400–900 m in the AAIW layer, the interior of eddy S (minimum salinity of 34.370 at 760 m and γ = 27.52 kg m⁻³) is saltier than the outside. In addition, though the eddy is marked by a vertical oxygen maximum (198 μmol kg⁻¹), it remains less oxygenated than the surrounding waters. In Figures 17c and 17d, the AAIW salinity and oxygen vertical extrema at γ = 26.52 kg m⁻³ (900 m) also mark the deepest point where the eddy core profiles differ from the neighboring ones. The vertical transition at this density is abrupt, suggesting that the non-local water of eddy S had just overlaid the local water at the date of observation. Like the oxygen signature of the eddy (Figure 5d), the CFC and nutrient structures (Figure 6) appear as upward pointing tongues of low values (for the CFCs) and high values (for the nutrients). As the tongue-shaped patterns generally cross the density contours, these eddy anomalies are more than just an isopycnal uplift of properties, and reflect trapping and transport of distant water by the eddy.

[36] The above description of the core anomalies of eddy S reveals a subtropical influence at all levels above ∼900 m (γ = 27.5 kg m⁻³). In Figures 9a and 17c, a major characteristic of the eddy is a high salinity anomaly at 26.75 kg m⁻³ < γ < 27.5 kg m⁻³, whose origin can only be north of the STF. The AAIW salinity minimum of 34.37, well above the 34.3 limit separating the old Indian Ocean and recent Atlantic Ocean varieties in that region, itself undoubtedly points to a subtropical origin. The best marker of the eddy origin, however, is its low oxygen concentration (165.5 μmol kg⁻¹) at γ = 27.07 kg m⁻³, which is only found at similar densities (and combined with similar temperatures and salinities) at the South African continental shelf break (Figures 2b and 9b). This narrow core of low oxygen concentrations at the continental slope, here observed west of Cape Town at 33°58′S, shows a minimum of 150 μmol kg⁻¹ at 290 m and γ = 27.04 kg m⁻³, comparable to a previous nearby (34°18′S) observation of 146 μmol kg⁻¹ at 250 m by Mercier et al. [2003]. In both cases, VM-ADCP measurements showed that the oxygen-depleted water near the continental slope is associated with a poleward alongshore undercurrent at 150–300 m. Farther east, Chapman et al. [1987] also observed low oxygen water (<180 μmol kg⁻¹) at 50–200 m along the continental slope, which was associated with the Agulhas Current, and separating from the coast with it. Although it is uncertain whether both features are connected, water with oxygen as low as 160 μmol kg⁻¹ at 50–300 m appears as characteristic of the alongslope flows around South Africa. The closeness of the oxygen minimum and associated density in eddy S (165.5 μmol kg⁻¹, γ = 27.07 kg m⁻³) to those of the near-slope water at 33°58′S is remarkable. The southward influence of the coastal low oxygen water appears in Figures 2b and 9b through patches of low oxygen (<200 μmol kg⁻¹) at γ = 27.1 kg m⁻³ north of the STF and, as mentioned above, through a large southward tongue with values below 225 μmol kg⁻¹ reaching to 41°S. The core of low oxygen observed at ∼700 m in eddy M also had this density (γ = 27.05 kg m⁻³), yet showing higher concentrations (210 μmol kg⁻¹) indicative of dilution, did not imply an origin right at the continental slope for that eddy.

4.2. History of Eddy S

[37] Because eddy S was smaller in size and less intense than eddy M, its backtracking using the method of Dencausse et al. [2010] proved more difficult. Backtracking could only be performed over two months, to mid-December 2007. Before this date, we had to track eddy S visually, an admittedly less rigorous method which, however, appeared unambiguous, and should provide a correct trajectory, insofar as the SSH maps themselves are correct. Tracking by the wavelet method did not work, either, after the date of the eddy observation during BGH. We also had to rely on visual tracking for that period, and on information from an ARGO profiler launched in S. For lack of a complete detailed trajectory, we show in Figure 18 six weekly SSH maps at dates of outstanding events in the life of eddy S. The eddy trajectory, determined as just described, is shown in Figure 18e.

[38] The formation of eddy S was around 10 October 2007 (Figure 18a), when a cyclone was present at 39°S-20°E south of the Agulhas Bank, separating the Agulhas Current retroflection from an Agulhas ring that detached from it about two weeks before. On that date, the SSH low anomaly of eddy S extends southward beyond 40°S into the SAZ, suggesting that the structure might be formed of subantarctic water entrained northward by the eastern part of the newly spawned ring [Lutjeharms and van Ballegooyen, 1988]. The eddy, however, seems also connected to another cyclonic structure present west of the Agulhas Bank, a connection which was still more apparent in the previous SSH map. This frequently observed cyclone west of the Agulhas Bank was named the “lee eddy” by Penven et al. [2001]. It is fed from the east by cyclonic vorticity cores propagating southwestward along the continental slope [Lutjeharms et al., 2003]. The formation process of eddy S shows that the distinction between subantarctic cyclones (stemming from the SAZ) and Agulhas cyclones (originating in the lee eddy) might not be clear-cut. Furthermore, we noted above that very low oxygen concentrations (<160 μmol kg⁻¹) at γ = 27.1 kg m⁻³ as observed in eddy S might be more representative of the poleward undercurrent at the Atlantic continental shelf break, than of the Agulhas Current, making this undercurrent possibly a third contributing source for eddy S. Whatever its dominant origin (Agulhas Current, Atlantic poleward undercurrent, subantarctic domain), the observation that eddy S was adjacent to the continental slope in early October 2007 would alone explain its marking by along-slope water characteristics, including low oxygen.

[39] On 7 November 2007 (Figure 18b), eddy S was at 42°S–17°E, still close to the Agulhas Current retroflection, yet was stirred and being subdivided by the combined effects of the retroflecting current and nearby vortices. Eddy S, the southern part of the subdivision, moved away from the Agulhas Current retroflection, and, after another subdivision, found itself above the northeastern end of the Agulhas Ridge (41°S–14°30′E) on 21 November 2007 (Figure 18c). After stalling three weeks over the bathymetry, the eddy moved southwestward, then crossed the ridge northwestward on 19 December 2007 (Figure 18d) at 41°15′S–13°20′E. At about this date eddy S became adjacent to eddy M, and it may be that entrainment of S by M helped the former cross the ridge. It is also from this date onward that S could be tracked by the wavelet method. On 27 February, when observed during BGH (Figure 18e), eddy S was northeast of eddy M, still adjacent to the bigger eddy and apparently coupled to it, as its anticyclonic drift suggests. The SSH signal of S was lost about one month later on 26 March 2008 near 42°S–8°E (Figure 18f).

[40] An ARGO profiler launched in eddy S on 26 February 2008 suggests that the eddy lived still longer, to at least 18 June 2008 (Figure 19). The salinity profiles provided at 10 days intervals (not shown) indeed exhibited patterns characteristic of the eddy until that date, namely, a salinity maximum at 250–300 m and a salinity minimum in the AAIW higher than 34.3. As such properties were observed exclusively in eddy S in the SAZ during BGH (Figure 2a), they indicate that the profiler remained for about 3 months in the eddy. After 18 June 2008, the ARGO profiles changed abruptly, as best seen from the potential temperature-salinity diagrams of Figure 19b, indicating that the float left the vortex. The float trajectory until that date (Figure 19a) confirms the anticyclonic drift of eddy S around M, its last detected position being 41°S–8°E.

[41] Overall, eddy S lived for at least 8 months, and was 4.5 months old when observed during BGH. This being significantly more than the average duration of 2–3 months for cyclones inferred by Boebel et al. [2003] from altimetric tracking, suggests a great variability in the life durations of these structures. The net southwestward translation velocity of eddy S was 0.06 m s⁻¹. This is somewhat higher than previous average velocities of 0.036 m s⁻¹, obtained from altimetry by Boebel et al. [2003] and 0.041 m s⁻¹ obtained from subsurface floats by Richardson [2007], yet lower than the 0.096 m s⁻¹ surface drifter estimate by the latter author. The route followed by eddy S is at the southern fringe of the ensembles of cyclone trajectories [Boebel et al., 2003; Morrow et al., 2004; Richardson, 2007] and, in that, are more typical of Agulhas cyclones than Cape Basin cyclones trajectories. Richardson [2007] mentioned cyclones which stalled near the Schmitt-Ott Seamount (see Figure 1) at the eastern end of the Agulhas Ridge, and noted that several cyclones trajectories terminated near the ridge. Eddy S seems also illustrative of such topographic effects, as it experienced subdivision near the tip of the Agulhas Ridge, and the part of the subdivision further tracked under the name of eddy S itself stalled for 3 weeks over the bathymetry. Regarding water properties, Boebel et al. [2003] stressed that while doming isopycnals in cyclones of that region create a surface signal seemingly indicative of an intrusion of water from the SAZ, the true origin of these features can only be revealed by their intermediate depth properties. Eddy S is also illustrative of this point. As noted above, while its SSH signatures tend to connect the eddy to the SAZ when just formed, and to the SAF when observed during BGH, its subtropical origin is only revealed by the 200–900 m water properties.

4.3. Eddy S Kinematics and Water Trapping

[42] Figure 12 shows that the lateral maxima of swirl speeds in eddy S, unlike those of eddy M, fall in the middle of the hydrographic stations intervals. This is a less favorable configuration which did not allow us to get full depth estimates of the azimuthal velocities from the L-ADCP. We had to rely on the geostrophic and VM-ADCP velocities displayed in Figure 13b. A ratio of 1.5 between the maximum azimuthal velocity and the geostrophic velocity, estimated from the top 600 m where VM-ADCP measurements were available (Figure 13b), was assumed to hold downward to ∼2000 m, where velocities vanish.

[43] From the VM-ADCP data, (Figures 4b and 12), the lateral velocity maxima are at ∼15 km only from the eddy center, and the eddy radius, defined as the outer limit of cyclonic swirl velocities, is ∼90 km. Vertically, the velocities show maxima of 0.3 m s⁻¹ at depths of 290 m (on the northward side of the eddy) and 350 m (on the southward side). The eddy rotational volume transport was found equal to 17.3 Sv eastward and 14.4 Sv westward.

[44] The objective tracking of eddy S that could be performed between mid-December and 26 February revealed a weak drift velocity of 0.03 m s⁻¹ the week before the observation. During the 3 preceding weeks, however, the drift speed was ∼0.15 m s⁻¹ westward, a high value likely associated with the anticyclonic entrainment of eddy S by eddy M. Referring to the water trapping criterion of Flierl [1981] and to the profile of maximum swirl velocity obtained by applying the ratio 1.5 to the geostrophic velocities, the trapping depth was found equal to ∼1600 m at the time of observation, and ∼880 m during the three preceding weeks. Estimating the drift speed of the eddy after its observation can only be done from the displacements of the ARGO profiler, which includes a swirl component in addition to the eddy drift itself. Averaging the profiler displacements over the 10 cycles of 10 days during which it stayed in the vortex (Figure 19b), an average drift velocity of 0.1 m s⁻¹ was found, which corresponds to a trapping depth of ∼1100 m. From the above, it is likely that the water trapped in eddy S (down to ∼1600 m) at the date of its observation during BGH was non-local water above ∼880 m, and local water below.

[45] The profiles and density-property curves in the core of S and at neighboring stations (Figure 17) fit in with these inferences. The salinity and oxygen core profiles of S (Figures 17a and 17b) differ significantly from those of the side stations above ∼950 m, and are very similar to the latter below. The γ-salinity and γ-oxygen curves (Figures 17c and 17d) exhibit the two vertical domains still more strikingly, with an abrupt transition at γ = 27.6 located at 900 m in the core of eddy S. This abrupt vertical transition is a sign that vertical mixing has not had time to operate since the non-local water of the eddy overlaid the local waters. It corroborates the conclusion drawn from the kinematics, that the high drift velocities of the preceding weeks split the eddy vertically at ∼880 m. The fact that the ARGO profiler, which drifts at 1000 m, remained in the eddy for 3 months is compatible with the above estimate of a ∼1100 m average trapping depth during that period. This is somewhat surprising, however, as temporary shallower trapping depths must have occurred during the 3 months, and the profiler sinks down to 2000 m every 10 days. A possible initial launching right at the eddy center, and the relatively large eddy radius (90 km) might be explanations for the observed trapping duration. Finally, the isopycnic property distributions along the cruise track (Figure 15) corroborate and bring more precision on the trapped water and its origin. The waters in eddy S at densities γ = 27.1 kg m⁻³ (∼350 m depth in the eddy) and γ = 27.35 kg m⁻³ (510 m) evidently originate from the close vicinity of the continental slope where the eddy was formed. The water at γ = 27.5 kg m⁻³ (800 m) is subtropical water from north of ∼37°S, but not from the eddy formation region. The absence of any signal related to eddy S at γ = 27.7 kg m⁻³ (∼1100 m) confirms the presence of local water at such levels.

4.4. Air-Sea Fluxes Over Eddy S

[46] Eddy S, when intersected, was located under a flat low pressure that was widely spread from 20°S/5°W to south of the Agulhas retroflection area. This flat low pressure was separating a deep perturbation centered at 55°S and moving eastward from a slow low pressure system located over the Agulhas retroflection. This synoptic situation produced the lowest wind observed during the cruise (1.16 m.s⁻¹ at 41.18°S; Figure 16). This was an ideal atmospheric condition to investigate a warm eddy impact on the atmospheric boundary layer. Unfortunately, as mentioned above, eddy S was located just north of the southern STF branch, whose temperature gradient masked any sea surface temperature variation possibly related to the core of eddy S. As a consequence, the surface fluxes do not show any signal that might be related to this structure. The oceanic skin sea surface temperatures present higher values than the atmospheric temperature north of the STF, as expected. The warmer water there produces a positive SHF toward the atmosphere as well as a LHF. The fluxes then decrease steadily southward to beyond the STF, and collapse when the air-sea temperature gradient reverses at 41.65°S, inducing a negative SHF (energy transfer from the atmosphere to the ocean) and a LHF lower than 35 W.m⁻².