## 1. Introduction

It has long been recognized that transient eddies play a key role in the transport of chemical and biological tracers across the main axis of the Antarctic Circumpolar Current (ACC). By extension, they also play a key role in the global meridional overturning circulation [see Marshall and Speer (2012) for a review]. In noneddying models, this role is parameterized. In such models, the circumpolar and across-stream transport of the ACC are related not only to each other, but also to the choice of parameterizations and coefficients. In eddy-resolving models, these linkages are less clear, and it is possible that the circumpolar and meridional transports may be largely decoupled.

The idea that the northward Ekman transport across the main axis of the current is balanced by an eddy mass flux to the south has been coined “eddy compensation” (Viebahn and Eden 2010) and is supported both by numerical studies and observations (e.g., Henning and Vallis 2005; Farneti et al. 2010; Gent and Danabasoglu 2011; Meredith et al. 2011). That this might be the case is particularly easy to grasp in the context of a wind-driven quasigeostrophic (QG) model, where one thinks of the ocean as analogous to a stack of a few immiscible layers. In the absence of diapycnal mixing, it is clear that the equilibrium meridional circulation must be closed in each layer individually. Consider then the transport across a zonally reconnected time-mean streamline of the leading-order flow. Transport across such a streamline can only occur (i) in an Ekman layer, (ii) in association with transient eddy (or bolus) fluxes, or (iii) at higher order in the quasigeostrophic expansion. Thus, any Ekman transport that is not eddy compensated must instead be compensated by an *O*(*ε*) (e.g., Rossby number) correction to the leading-order velocity field. Put another way, either the viscous force or nonlinear advection terms (including time-mean and Reynolds stress contributions) would have to play a prominent role in the time-mean, along-stream *O*(*ε*) momentum balance. For parameter regimes thought to be relevant to the ACC, however, this balance is more typically between wind stress and transient eddy form drag terms. Equivalently, one expects the Ekman transport to be eddy compensated in the upper layer.

Related to eddy compensation is eddy saturation. This idea is that circumpolar transport in eddy-resolving models becomes roughly independent of the strength of the wind stress in strongly forced regimes (e.g., Hallberg and Gnanadesikan 2001, 2006; Munday et al. 2013). For example, if circumpolar transport in the abyssal ocean is blocked by topography, then the thermal wind shear, together with the meridional extent of the current, determines the transport. Although a stronger wind forcing would imply a stronger eddy field, this need not lead to a significant increase in the thermal wind shear or circumpolar transport (e.g., Straub 1993). In a more general context, this net “baroclinic transport” can also be influenced by buoyancy fluxes, (e.g., Gnanadesikan 1999; Munday et al. 2013). Crudely, one might think of the need for eddies (and therefore baroclinic instability) as determining the thermal wind shear and of buoyancy forcing and the global meridional overturning cell as influencing boundary conditions that set the meridional extent over which this circumpolar thermal wind applies. Taken together, these would determine the baroclinic transport of the ACC.

A variety of numerical simulations support the idea that circumpolar transport is only weakly dependent on the amplitude of wind forcing. For example, Hallberg and Gnanadesikan (2006) carried out simulations using an eddy-resolving general circulation model with realistic forcing and geometry and found a 20% increase in the wind stress led to a much smaller increase in the circumpolar transport. Tansley and Marshall (2001) consider balanced “geostrophic vorticity” model simulations in a channel with partial walls and a simplified Scotia Ridge and find transport to be weakly dependent on wind stress. Hutchinson et al. (2010) consider QG simulations in an open channel with complex topography and also observe a saturation of transport. They interpret this as evidence that the model is in an eddy-saturated regime and also point out that using an oceanic velocity–dependent wind stress formulation (e.g., Duhaut and Straub 2006) reduces the strength of the eddy field and leads to an increase in zonal transport.

Consistent with the idea of eddy saturation, Nadeau and Straub (2009, hereafter NS09) and Nadeau and Straub (2012, hereafter NS12) also find zonal (or circumpolar) transport to saturate in wind-driven quasigeostrophic simulations. They point out, however, that such a saturation might also be anticipated by applying ideas borrowed from the theory of midlatitude gyres (e.g., Rhines and Young 1982). This appeal to “basin dynamics” differs qualitatively from older attempts to apply basin dynamics to the ACC [e.g., Stommel 1957; Gill 1968; see LaCasce and Isachsen (2010) for a review] in that the baroclinic structure of the flow plays a central role.

NS09 present a simple analytical model. Buoyancy forcing and connection to the global meridional cell are absent, and circumpolar transport is described as the sum of a basin and channel contribution (*T*_{basin} and *T*_{channel}, respectively). In the saturation regime, these are given as *T* = *T*_{basin} + *T*_{channel} = *c _{R}HL*, where

*c*is the baroclinic long Rossby wave speed,

_{R}*H*is the ocean depth, and

*L*is a meridional extent. For example,

*L*

_{channel}was taken to be the width of the model Drake Passage, and

*L*

_{basin}worked out to be the distance between Drake Passage and the zero wind stress curl line (in their configuration forcing was a zonal wind stress varying only with latitude). NS12 tested the robustness of circumpolar transport to factors absent in the simple theory; specifically, they looked at how circumpolar transport depended on the strength of a bottom drag coefficient and the relative strengths of zero curl and curl-containing contributions to the total wind stress. Considerable sensitivity was found and this was attributed to eddy effects (but not to eddy saturation or compensation ideas per se). Instead, NS12 pointed to the importance of a large-scale recirculation gyre related to a pool of nearly homogeneous potential vorticity. The bottom drag coefficient and the mix of curl-containing and zero-curl contributions to the winds stress affected the abyssal eddy field—and this influenced the strength, zonal extent, and upper-layer expression of the recirculation gyre. All of this affected

*T*

_{basin}. Additionally, they pointed out that transport in their simulations was also influenced by a strong standing eddy near the (single) ridge located near the model Drake Passage.

In this paper, we restrict attention to wind-driven circumpolar flow in a highly idealized quasigeostrophic model of the Southern Ocean. Effects of diapycnal mixing, buoyancy forcing, and the need for the ACC to connect with the global meridional overturning cell are absent. In this configuration, both eddy saturation ideas and the simple theory of NS09 suggest that transport should saturate for sufficiently strong forcing. Note that NS09 makes predictions of both the forcing amplitude above which saturation should occur and the level at which transport is expected to saturate. As far as we are aware, eddy saturation theory does not (i.e., except in situations for which the meridional extent over which these ideas are expected to apply is obvious). Although the predictions of NS09 are crude (e.g., see NS12), we will nonetheless use their analytic model as a benchmark, while recognizing that some of the assumptions made in deriving this simple theory are not applicable to our simulations. In particular, NS09 was derived for a two-layer ocean for which topography was confined to be far west of the model Patagonian Peninsula. Here, we will consider simulations with higher (although still very modest) vertical resolution. We will also consider simulations for which topography is complex and fills the domain (so that the vertically integrated Sverdrup balance need not apply anywhere). Specific goals are

- to discuss how NS09 might generalize for continuous resolution,
- to test the robustness of transport saturation values to horizontal and vertical resolution, and
- to test the robustness of transport values to the choice of bottom topography.

To address the relevance (or lack thereof) of Sverdrup dynamics in our complex topography simulations, we compare simulations for a series of topographies ranging between our single ridge and a complex topography configuration, the latter being similar to that used by Hutchinson et al. (2010). We find circumpolar transport to increase to a value near that predicted by NS09 when topography becomes significantly complex such that the standing eddy ceases to block midocean potential vorticity contours at Drake Passage. We also note clear evidence that the circumpolar transport is related to an interior Sverdrup drift in the single ridge configuration. This interior Sverdurp flow becomes increasingly convoluted as topography becomes more complex. We stress that the increase in transport between the single ridge configuration and our complex topography configuration appears clearly related to the disappearance of the stationary eddy. It does not appear related to a transition between one regime in which some form of basin dynamics does appear to be relevant and another in which it does not. We also stress that we do not view NS09 as necessarily inconsistent with eddy saturation ideas; rather, one can think of their model as providing an estimate of the meridional length scale by which thermal wind shear must be multiplied to obtain the transport (in the absence of buoyancy fluxes).

Section 2 briefly reviews the numerical model we will use and outlines the simulations to be considered. Section 3 presents simulations using our Scotia Ridge topography, comparing resolutions of two and five layers in the vertical, as well as a range of horizontal resolutions. Section 4 compares simulations using more realistic topography (but without realistic coastlines) to simulations using our Scotia Ridge bathymetry. A brief discussion concludes the paper in section 5.

## 2. Numerical model and experiment design

*N*-layer setting. Following Pedlosky (1996), the equations can be written asin whichis the potential vorticity,

*ψ*

_{0}=

*ψ*

_{N}_{+1}= 0,

*f*

_{0}is the mean Coriolis parameter,

*β*is the northward spatial derivative of the Coriolis parameter,

*g*is the gravitational acceleration,

*H*are layer thicknesses,

_{k}*h*is bottom topography,

_{b}*A*is a lateral biharmonic viscosity coefficient,

_{h}*r*is a bottom drag coefficient, and

*τ*is the wind stress. No normal flow and slip conditions appropriate for hyperviscosity are imposed at lateral walls. Additionally, standard techniques similar to those described by McWilliams (1977) are used to specify the zero potential vorticity part of the solution at each time step. The numerical implementation is similar to the two-layer model presented in NS09 and NS12. Briefly, we use a third-order Adams–Bashforth scheme for time derivatives, center differencing in space, Arakawa (1966) for the Jacobian, and a multigrid method for the elliptic inversions.

*τ*

_{0}~ 0.1 N m

^{−2}.

Three topographies are considered. Figures 1a and 1c are two versions of our Scotia Ridge topography and Fig. 1b shows a snapshot of upper-layer kinetic energy for a simulation using the topography in Fig. 1a. The first (Fig. 1a) is considered in section 3 (for which our model domain is smaller) and the second (Fig. 1c) is considered in section 4. They differ in the size of the domain and amplitude of the ridge (2000 m for the simulations in section 3 and 1600 m for those in section 4). Another difference is that a continental rise topography is added for our Scotia Ridge topography experiments in section 4. This was done since the Scotia Ridge topography simulations in this section are compared with our realistic (or complex) topography simulations (Fig. 1d), for which a submerged Patagonia might be thought of as providing an effective continental rise. Our realistic topography is similar to that used by Hutchinson et al. (2010); specifically we use a linear unsmoothed interpolation of the 5-Minute Gridded Global Relief Data Collection (ETOPO5) to a Cartesian grid. Topography data from 33° to 67°S is interpolated on a 20 000 km by 4000 km rectangular domain. Following Hutchinson et al., topography is truncated at ±800 m from the mean depth of the bottom layer. For this reason, the ridge height in our second Scotia Ridge topography is reduced from 2000 to 1600 m (corresponding to the ±800 m used in our realistic topography experiments).

We will also be considering a range of vertical (from two to five layers) and horizontal (7.5–60 km) resolutions as well as several (first baroclinic) Rossby radii (16–48 km). Parameters for the simulations considered are given in Table 1.

Model parameters.

## 3. Scotia Ridge topography

The analytical model of NS09 was formulated in a two-layer QG setting. We will be considering a higher, albeit still modest, vertical resolution and a summary of how the analytical model might be generalized to continuous stratification becomes relevant. A generalization of the analytical model based on homogenization of potential vorticity in the continuously stratified context is discussed in the appendix. Briefly, in this generalization the basin contribution to the total circumpolar transport corresponds to the Sverdrup flux lying to the east of a separation point *x*_{0} defined by the intersection of a level of weak motion with the level of the topography. This idea is illustrated schematically in Fig. 2.

In the two-layer model, the geometry of characteristics (contours of *ψ _{B}* is the barotropic streamfunction) played a key role. Specifically, the longitude where the separatrix Θ contour intersected the Drake Passage latitude band determined the zonal position

*x*

_{0}, east of which the Sverdrup flux fed into the circumpolar transport and west of which it fed into large scale recirculating gyres. Note also that Θ contours coincide with lower-layer potential vorticity contours when the lower layer is at rest. In a multilayer setting, abyssal-layer potential vorticity (east of a pool of nearly homogeneous potential vorticity) has a structure similar to characteristics (see Fig. 3). One possibility, then, is that the analytic model could be generalized by considering

*x*

_{0}to correspond with the position where the separatrix abyssal potential vorticity contour intersects Drake Passage latitudes.

Figure 3 shows the time-averaged streamfunction for a five-layer simulation with 30-km resolution and a first baroclinic Rossby radius of 32 km. This corresponds to an eddy-permitting—but not resolving— regime; for example, the ratio of eddy-to-mean kinetic energy is about 6. Gray shading in the figure indicates the large-scale recirculation gyre. Streamlines are consistent with our conceptual model in that the circumpolar transport extends from a portion of the Sverdrup flow lying to the east of a large-scale recirculating gyre. Potential vorticity is nearly homogeneous over much of the domain in the interior layers (not shown), and the abyssal layer is characterized by a homogenized pool in the western portion of the domain (see bottom panel of Fig. 3), as is familiar from gyre studies in closed box domains (e.g., Rhines and Schopp 1991). From the figure, it appears that the longitude at which the separatrix abyssal potential vorticity contour intersects the Drake Passage latitude band provides at least a crude approximation to the zonal position separating the large-scale recirculation from the circumpolar streamlines. To summarize, all of the circumpolar streamlines feed into Drake Passage latitudes from a Sverdrup-like interior in the eastern part of the domain, and the separation between east and west is roughly determined by the separatrix abyssal PV contour. Unlike in the two-layer case, however, we do not have a simple analytical model to determine the geometry of the abyssal potential vorticity field.

Figure 4 shows the barotropic streamfunction for different model resolutions and for wind forcing corresponding to the saturation regime. Increased resolution, either horizontal or vertical, leads to the development of a strong topographically trapped recirculation lying just east of the model Drake Passage (in the bottom left corner of the figure panels). That this feature is stronger in the five-layer simulations is likely due to the effect of resolving high baroclinic modes; Barnier et al. (1991) showed that resolving these higher modes in double-gyre basin simulations enhances the jet-exit region inertial recirculations and the penetration scale of the midlatitude jet, but leaves the large-scale interior circulation largely unaffected. In our Southern Ocean geometry, the topographically driven recirculation also becomes more energetic.

As discussed in NS12, this recirculation is associated with a westward pressure torque, which increases in strength for stronger forcing. Inspection of time-averaged potential vorticity of the interior layers (e.g., excluding the top and bottom layers) reveals this recirculation to be associated potential vorticity contours blocking Drake Passage (i.e., extending between the model Patagonian and Antarctic Peninsulas; not shown). In other words, the recirculation acts to inhibit circumpolar flow. Associated with this, resolution affects zonal transport in the strongly forced regime. Figure 5 shows transport as a function of forcing strength for our two- and five-layer simulations at various horizontal resolutions. Also included in the figure is the transport predicted using the two-layer analytical model of NS09. The lower-resolution simulations agree qualitatively with the theory; that is, they show a linear increase in transport with forcing strength in the weakly forced regime followed by an approximate saturation of transport for stronger forcing. This saturation occurs at a lower value than predicted by the analytical model, similar to numerical results found in NS09. The higher-resolution simulations show transport to drop off (rather than to saturate) with forcing strength in the strongly forced regime. This effect becomes more pronounced with both vertical and horizontal resolution. It thus appears that increased vertical and/or horizontal resolution allows a strong inertial recirculation to develop over the topography and that this leads to a reduction in transport in the strongly forced regime.

In the Scotia Ridge topography experiments presented in this section, there is only a single ridge against which pressure torques can act to remove eastward momentum input by the winds. Given this, it is perhaps unsurprising that a strong (and arguably unrealistic) recirculation appears over the ridge. In the following section, we consider more complex topography. Our motivation is that, with many topographic features distributed throughout the domain, balancing the zonal momentum budget with topographic form drag will not necessitate the appearance of a strong topographically trapped recirculation, such as the one discussed above.

## 4. Complex topography

In this section, we compare simulations using an idealized Scotia Ridge with simulations using more complex topography. As described in section 2, the Gaussian ridge simulations in this section assume a lower ridge height; additionally, a continental rise is added (Fig. 1c). This addition was to make the two topographies somewhat more similar, that is, in the sense that the complex topography also has a continental rise in the form of a submerged Patagonian Peninsula. Note that the addition of a continental rise topography along the western Patagonian coast makes integration along characteristics from an eastern boundary (used in deriving the NS09 analytical model) problematic. Nonetheless, we will continue to use the transport predicted by NS09 as a benchmark against which to compare model results.

Recall also from section 2 that the complex topography was obtained from an interpolation of the ETOPO5 dataset to a Cartesian grid. Other differences with the simulations in section 3 are that here the domain zonal extent is 20 000 km (roughly twice as long as previously), the width of Drake Passage is increased to 600 km, horizontal resolution is 12.5 km, and a three-layer configuration is used throughout. We consider three Rossby radii: 16, 32, and 48 km. Further details of our model configuration are given in Table 1.

Figure 6 shows time-averaged streamfunctions for the *L _{ρ}* = 32 km simulations and different forcing amplitudes. The ridge experiments show Sverdrup-like gyres in the interior with a complex structure along the western boundary. The small-scale topographically trapped gyre near Drake Passage is also evident for higher levels of forcing. Note that its amplitude is reduced from that seen in the previous section. This could be because the topographic height was also reduced or the larger extent of the domain. The inertial recirculation near the western boundary jet exit region also has a more complex structure than before; this is related to the continental rise topography (NS09). The complex topography simulations do not show a discernible Sverdrup-like circulation. Instead, a topographically steered flow develops and increases in strength with forcing amplitude.

Figure 7 compares upper-layer velocity for the two topographies and three Rossby radii and a realistic forcing amplitude (*τ*_{0} = 0.08 N m^{−2}). In the ridge simulations, zonal jet structures emanate from a region of western intensification and gradually decrease in strength as they head east. The width, strength, and spacing of these jets increases with the Rossby radius, as one would expect. With realistic topography, this picture changes dramatically: the eddy field increases in intensity and the jets are less zonally coherent. Instead they are topographically steered. This is similar to what was observed by Thompson and Sallee (2012), who argued that kinetic energy is stronger downstream of topography where baroclinicity is enhanced owing to topographic steering. Also similar to Thompson and Sallee is the presence of recirculations distributed along the path of the circumpolar current where it crosses major topographic obstacles; the zonal jets seem to be locked between these different recirculations.

Note from Table 2 that abyssal layer kinetic energy is not strongly dependent on the choice of topography. Since energy dissipation is primarily via bottom drag, this implies that the wind energy source is also similar between the experiments. On the other hand, both the spatial distribution of the eddies and the partitioning between mean and eddy abyssal layer kinetic energy vary significantly between the two sets of experiments. Figure 8 shows the mean and eddy kinetic energy for the abyssal layer; also shown is the Eady growth rate parameter, *τ*_{0} = 0.2 N m^{−2}. In the ridge experiments, both kinetic energy and regions of intense baroclinic instability are concentrated in the western boundary region. Additionally, kinetic energy resides mainly in the eddy field. For the realistic topography, eddy activity is distributed more uniformly along the path of the current, and the mean kinetic energy of the lower layer is increased substantially.

Vertical distribution of the domain-averaged kinetic energy at *L _{ρ}* = 32 km.

Figure 9 shows transport as a function of *τ*_{0} for our realistic topography simulations. Somewhat surprisingly given the lack of any obvious Sverdrup-like interior, the realistic topography experiments show better agreement with the analytical model of NS09 than do the ridge experiments. In fact, agreement in the saturation regime is quite good for the *L _{ρ}* = 32 and 48 km experiments. For

*L*= 16 km (corresponding to marginal resolution of

_{ρ}*L*), saturation occurs at a value higher than that predicted by theory. This contrasts the Gaussian ridge topography, for which the

_{ρ}*L*= 32 and 48 km experiments show saturation values considerably lower than the predicted value. The ridge experiments also show evidence of transport decreasing with forcing amplitude, similar to that seen in the previous section. Here, however, this effect is less pronounced—likely due in part to the reduced ridge height (and thus a weaker topographically trapped recirculation near Drake Passage) and in part to the addition of a continental rise topography.

_{ρ}It is natural to ask whether the surprisingly good agreement between transport in our complex topography simulations and the analytical prediction of NS09 is not simply fortuitous. After all, the analytical model assumes Sverdrup dynamics north of Drake Passage latitudes and a Sverdrup-like circulation is not evident in the simulations. Additionally, the realistic topography simulations show eddy kinetic energy (EKE) to behave in a manner that seems consistent with eddy saturation ideas. For example, EKE is more broadly distributed over the domain (Fig. 8) and increases with forcing amplitude (not shown). To address the extent to which agreement with NS09 may be fortuitous, we consider a set of experiments for which topography ranges smoothly between the ridge and complex topography configurations. Specifically, we define bottom topography by *h _{b}* ≡

*αh*

_{realistic}+ (1 −

*α*)

*h*

_{ridge}. As such,

*α*= 0 corresponds to our ridge topography and

*α*= 1 corresponds to our realistic topography. We compare simulations for a range of values of

*α*, with

*L*= 32 km and

_{ρ}*τ*

_{0}= 0.3 N m

^{−2}(corresponding to the saturation regime).

Figure 10 shows the baroclinic circumpolar transport (excluding contributions from the abyssal layer) as a function of *α*. It is relatively constant at high and low values, and undergoes a sharp transition, increasing from around 70–80 Sv (Sv ≡ 10^{6} m^{3} s^{−1}) to around 105 Sv near *α* = 0.4

Figure 11 shows a measure of the Eady growth rate *σ* for various values of *α*. Figures 11a–d correspond to lower transports, Fig. 11e is just after the transition, and Fig. 11f corresponds to our realistic topography (see Fig. 10). For low *α*, large values of *σ* are concentrated mainly in the western boundary layer, whereas for high *α* they are distributed more uniformly along the path of the current. Note that this transition occurs gradually. In particular, an abrupt transition between Figs. 11d and 11e, that is, between the low and high saturation transports, is not observed; both panels show *σ* to be distributed over the domain and both appear unlike Fig. 11a.

Figure 12 shows the time-averaged streamfunction for various values of *α*. A low-pass filter has been applied to emphasize the large-scale structure. Specifically, we use a filter similar to the one described by Nadiga (2008); it performs an inversion of the Helmholtz operator *L _{f}* = 250 km is the filter width and

*ψ*

_{low}=

*ψ*on the boundaries. The resulting streamfunctions show clear evidence in support of a Sverdrup-like circulation feeding into Drake Passage latitudes to form the ACC (ribbonlike region shaded dark blue) for low

*α*.

^{1}Also evident are the large-scale recirculation (light blue) and the topographically trapped inertial recirculation (brown) adjacent to the model Drake Passage. For larger

*α*, the Sverdrup interior becomes increasingly deformed and the large-scale recirculation to the south of recirculating streamlines is replaced with a number of smaller ones lying to the south of the main axis of the current. Both of these transitions occur gradually as

*α*is increased.

The topographically trapped inertial recirculation undergoes a more abrupt change. It is intense and situated so as to “block” Drake Passage (Figs. 12a–d), but is weaker and shifted to the north (Figs. 12e,f). This transition is relatively abrupt at around *α* = 0.4. The northward shift and weakening of the stationary eddy removes both a westward pressure torque on the upper ocean and a potential vorticity “blocking” of circumpolar flow in middepths, both suggestive of increased circumpolar transport.^{2} The increase in transport (between Figs. 12d and 12e) comes mainly from the interior layer (see Table 3).

Transport (Sv) in each layer as a function of *α*.

Note also that, in Figs. 12a–d, the ACC lies adjacent to a boundary over part of its path. This suggests a second explanation for the jump in transport at around *α* = 0.4. That is, it could be related to a transition from a regime in which frictional boundary currents are playing a key role to one in which the ACC remains isolated from boundaries over its entire path. Immediately adjacent to a boundary, one anticipates the lateral friction term to be important in the momentum balance. (Recall that we used biharmonic friction with slip boundary conditions ∇^{2}*ψ* = ∇^{4}*ψ* = 0. Recall also that the fields presented in Fig. 12 are spatially filtered, that is, so that boundary currents may appear more viscous than is in fact the case.) By equating the Coriolis and friction terms in the along-boundary momentum equation, we estimated a frictionally balanced contribution to the order Rossby number correction to the quasigeostrophic velocity. Doing this, we found the frictionally balanced transport in the southern boundary layer evident west of Drake Passage, in Fig. 12d, to be 3% or less of the maximum Ekman transport. Except for the two grid points closest to the wall, it was considerably smaller. Given this—and that the ACC is adjacent to the wall for only a small fraction of its total path—we conclude that friction is not playing a significant role in balancing the Ekman flux. In other words, that the current runs adjacent to the boundary along part of its path, in Figs. 12a–d, does not imply that friction replaces eddies in compensating for the Ekman transport across the path of the ACC.^{3} Instead, we believe that all of our simulations are consistent with eddy compensation; that is, the net Ekman transport across the main axis of the current is balanced by an eddy mass flux in the opposite direction.

None of this, of course, should be taken as proof that ACC transport—even in the wind-only forcing quasigeostrophic context considered here—is determined by the simple theory put forth in NS09. It does, however, seem obvious that Sverdrup dynamics is somehow relevant to the circulation observed in Fig. 12a. Moving from Figs. 12a to 12d, evidence for Sverdrup dynamics (or even some more general form of basin dynamics) is progressively less obvious. That circumpolar transport does not vary much between these panels seems, however, suggestive that whatever determines the transport in Fig. 12a is also at work in Fig. 12d. Between Figs. 12d and 12e, there is a sharp increase in transport. We attribute this to the disappearance of a stationary eddy that appears to partially block circumpolar flow at middepth. It may also be the case that a transition occurs between one regime where basin dynamics are clearly evident (Fig. 12a) to one in which they are not; however, such a transition seems to occur gradually as *α* is increased. As such, it should not be associated with the sharp change in baroclinic circumpolar transport near *α* = 0.4.

## 5. Conclusions

We compared numerical simulations of a wind-driven quasigeostrophic Southern Ocean using different topographies and resolutions paying particular attention to factors determining the zonal transport in a strongly forced regime. Simulations using a Scotia Ridge topography with varying horizontal and vertical resolution show that, when inertial effects are weak, transport is similar to the two-layer analytical model prediction of NS09. When inertial effects were strong, however, transport is much reduced in the saturation regime. This was attributed to additional inertial effects not foreseen in the analytical model, but discussed in NS12. Essentially, a topographically trapped inertial recirculation blocks the outflow of Drake Passage, providing an impediment to circumpolar flow. This blocking effect is much reduced when a more realistic topography is used in our simulations. In this case, recirculations also develop over numerous topographic obstacles. These have the effect of significantly increasing upper-layer kinetic energy, but do not form a “barrier” blocking the circumpolar transport analogous to that seen in our Scotia Ridge experiments.

We also considered a series of experiments with saturation regime forcing and with topographies ranging between our (Gaussian) ridge and realistic configurations. For the ridge topography, time-averaged and spatially filtered streamfunctions show clear evidence that the ACC is formed in association with a Sverdrup-like flow into Drake Passage latitudes. Any connection between the Antarctic Circumpolar Current and a Sverdrup interior becomes progressively less evident as the topography becomes more realistic. We also note, however, that the evolution of the time mean streamfunction between the ridge and realistic topography simulations is gradual. It is difficult to point to a value of *α* (a parameter serving to blend to the two topographies) where the governing dynamics shift from a regime where basin dynamics are clearly relevant to another in which they are not. There is, however, a relatively abrupt transition (around *α* = 0.4) where transport increases significantly and the inertial recirculation adjacent to our Gaussian ridge ceases to effectively block the gap at Drake Passage. In other words, the increase in transport observed around *α* = 0.4 appears to be associated with the removal of this blocking eddy—and not with a transition between one regime governed by “basin dynamics” and the other by “channel dynamics.”

We suspect that a complete understanding of the ACC will ultimately draw on both basin and channel ideas. For example, all of the simulations presented here are consistent with the idea of eddy compensation: the Ekman flux across time mean streamlines appears to be balanced by an eddy flux in the opposite direction. At the same time, some of our simulations are clearly also consistent with basin ideas. Other evidence that some form of basin dynamics might also have a direct influence on the ACC is also supported by the path of the current itself; in places it coincides with western boundary layer regions. For example, the main axis of the current takes a sharp turn to the north just east of Drake Passage (e.g., Hallberg and Gnanadesikan 2006; Mazloff et al. 2010) where it joins the Falkland/Malvinas Current (which can be thought of as dynamically similar to western boundary currents such as the Labrador Current and Oyashio). An additional western boundary current–like region of the ACC is also seen east of the Campbell Plateau southeast of New Zealand. The model of NS09 represents an improvement over previous attempts to apply basin ideas to the ACC, but makes too many assumptions to be applied directly to our complex topography simulations. Perhaps, fortuitously, it nonetheless predicts reasonably well transport in these simulations.

We thank MCRN, FQRNT, and NSERC for their financial support. We thank two anonymous reviewers for helpful suggestions and healthy skepticism.

# APPENDIX

## Analytic Model for Circumpolar Transport

In the two-layer analytic model of NS09, the point at which the separatrix characteristic intersected the northern edge of the Drake Passage latitude band determined the position east of which the Sverdrup flux fed circumpolar transport and west of which it, instead, fed a recirculation gyre. In the lower layer, large-scale basin theory predicts a pool of homogenized potential vorticity in the west and no motion in the east. In a continuously stratified context, potential vorticity homogenization could also be used to predict the baroclinic structure of the circulation. The extent of the homogeneous bowl of potential vorticity defines a level of no motion below the surface. In a similar manner to the two-layer case, this level of no motion might then be used to predict the separation between recirculating and circumpolar streamlines. However, as we will see in the following, the level of no motion is often a poor indicator of the baroclinic structure of the circulation and a level of “weak motion” seems much more appropriate. In the following, we review how to calculate the structure of the homogeneous PV bowl and then apply the idea to a modified version of the analytic model for circumpolar transport considered by NS09.

### a. Continuously stratified gyres

Following Young and Rhines (1982), we consider wind-driven quasigeostrophic flow in a closed box ocean and assume (i) the depth-integrated flow is given by the Sverdrup relation, (ii) the large-scale potential vorticity is homogeneous where the fluid is in motion, and (iii) the fluid is at rest where potential vorticity contours are blocked (i.e., where they extend back to the eastern boundary). This allows us to determine a depth of no motion, *z* = −*D*(*x*, *y*), defining a “bowl” below which there is no circulation. The goal now is to determine *D*.

*βy*

_{0}(

*z*) corresponds to the value of

*q*at a given horizontal level. For simplicity, we consider the classic double gyre problem in a closed rectangular basin with dimensions

*x*∈ (0,

*L*) and

_{x}*y*∈ (0,

*L*), for which symmetry arguments imply

_{y}*y*

_{0}(

*z*) =

*L*/2.

_{y}*ψ*. Assuming the wind stress to be a function of

_{B}*y*alone,

*ψ*is given bywhere the expression applies far from the western boundary region. Integrating (A1) twice from −

_{B}*D*to

*z*giveswhereFurther integrating (A4) over the total water column gives an expression for the barotropic streamfunction:The shape of the bowl,

*D*(

*x*,

*y*), then follows from combining (A5) with (A3). Below, we will be particularly interested in the structure of

*ψ*at the northern latitude of Drake Passage,

*y*=

*y*

_{DP}, that is, the southernmost latitude where one might expect Sverdrup dynamics to apply. We thus restrict discussion to the structure of the bowl intersecting the

*x*–

*z*plane at

*y*=

*y*

_{DP}.

Figure A1 shows the dependence of the shape of the bowl at *y*_{DP} on the decay scale 1/*α* for two values of the Rossby radius. For a large Rossby radius (*L _{ρ}* = 80 km), the level of no motion deepens as

*α*goes to zero (for which a linear stratification is recovered). However, for a small Rossby radius (

*L*= 10 km), curves of

_{ρ}*D*cross one another at the eastern side of the basin, and the maximum depth of the circulation,

*D*

_{max}, shoals as

*α*goes to zero. Note that this crossing would also occur for

*L*= 80 km if the basin were much larger. Put another way, curves in the

_{ρ}*L*= 10 km case would be similar to those in the upper panel were we to zoom into the easternmost 500 km or so of the basin—that is, east of where the curves cross. In the case of a linear stratification (

_{ρ}*α*= 0),

*D*goes like

*x*

^{1/3}, whereas in the exponential case,

*D*becomes essentially linear with

*x*at depths below the decay scale, 1/

*α*. In this latter case, the slope of ∂

*D*/∂

*x*is much larger when

*L*is small so that, if the values of

_{ρ}*D*are representative of the vertical penetration of the gyres in the

*L*= 80 km case, they become absurd for

_{ρ}*L*= 10 km. For example, we find

_{ρ}*D*

_{max}= 62 km on the western side of the basin for 1/

*α*= 250 m and

*L*= 10 km.

_{ρ}*D*is greater than the basin depth

*H*. It turns out, however, that, even in cases where

*D*lies far below the ocean bottom, the flow at

*z*= −

*H*can be so weak as to be essentially negligible. Consider, for example, the curve

*z*=

*l*(

_{μ}*x*) above which lies a given fraction

*μ*of the circulation on the plane

*y*=

*y*

_{DP}. This fraction of the flow lying above

*l*is defined asUsing

_{μ}*ψ*from (A3) and

_{B}*ψ*(

*z*) from (A4) one findsThe transcendental equation (A7) can then be solved numerically for

*l*using

_{μ}*D*from (A5).

Figure A2 shows *ψ*(*x*, *y*_{DP}, *z*) calculated assuming an infinitely deep ocean, but plotted only for the upper 4 km. Two cases are considered: (i) *L _{ρ}* = 10 km with 1/

*α*= 250 m and (ii)

*L*= 30 km with 1/

_{ρ}*α*= 1000 m, while

*N*

_{0}is kept constant. The vertical structure of the flow varies considerably from

*L*= 10 to

_{ρ}*L*= 30 km. For the latter case, it is spread relatively uniformly inside the bowl, while for

_{ρ}*α*= 1/250 m, the flow is much more surface intensified. For example, 99% of the southward transport across

*y*=

*y*

_{DP}lies above 1147 m whereas

*D*

_{max}is 62 km. Moreover, while the maximum value of the velocity at the surface is 1.6 × 10

^{−2}m s

^{−1}, the maximum velocity at 4000 m is only 1.8 × 10

^{−9}m s

^{−1}so that, in this case, the parameter

*D*is clearly a poor indicator of depth of the gyres. Because of this, we anticipate that the actual solutions, that is, taking into account the finite ocean depth, will not differ significantly from that shown in Fig. A2 away from the boundaries.

### b. Analytic model for circumpolar transport

In the two-layer model, the Sverdrup flux was divided into one part that resided solely in the upper layer and another that was distributed over both layers. In the continuously stratified case, typical values of *D* lie below the ocean floor so that the Sverdrup flux is everywhere distributed throughout the water column. That said, velocities at or below the ridge height can be so small as to be negligible.

To adapt the two-layer model to the continuously stratified case, we replace the “level of no motion” with a “level of weak motion” to determine the depth of the circulation. When this level lies everywhere above the topography, the Sverdrup flux into Drake Passage latitudes is considered top trapped (Stommel regime). However, when the level of weak motion descends below the ridge crest, the abyssal recirculation will be stronger, and one cannot rule out the possibility that this recirculation will be felt in the upper part of the water column—that is, creating a recirculation there as well (cf. Fig. 2). In this case, all of the upper-ocean portion of the Sverdrup flux across *y*_{DP} need not add to the ACC.

We thus redefine *x*_{0} as the zonal position east of which the Sverdrup flux can be considered top trapped. Then *T*_{basin} is the zonally integrated Sverdrup transport across *y*_{DP} to the east of *x*_{0}. For weak forcing, it may be that the Sverdrup flux can be considered top trapped at all longitudes and, in this case, *T*_{basin} is the zonal integral from a point, *x*_{west}, just east of the western boundary layer. Note that only a negligible fraction of this would typically lie below the ridge crest.

*D*is quite large, the vast majority of the transport can lie in the upper 1000–2000 m of the water column. Recall (cf. Fig. A2) that

*l*(

_{μ}*x*) is defined such that a fraction

*μ*of the Sverdrup across

*y*

_{DP}is above

*z*=

*l*(

_{μ}*x*). We define

*x*

_{0}as the point at which

*l*drops below the level of the ridge. We thus estimate

_{μ}*T*

_{basin}aswhere

*x*

_{0}is obtained using

*l*(

_{μ}*x*

_{0}) =

*h*

_{ridge}in (A7) and the approximation is that

*ψ*includes a (very) small contribution from below the height of the ridge. Qualitatively (A8) is similar to the two-layered model: it predicts a linearly increasing Stommel regime followed by a saturation regime at stronger forcing. Obviously, since the choice of a threshold value below which motion is considered weak is arbitrary, this model cannot be used to predict quantitatively the saturation of the ACC. However, an

_{B}*l*(

_{μ}*x*) exists for which the continuously stratified model is equivalent to the two-layer model using the first Rossby deformation radius.

## REFERENCES

Arakawa, A., 1966: Computational design for long-term numerical integrations of the equations of atmospheric motion.

,*J. Comput. Phys.***1**, 119–143.Barnier, B., , C. Le Provost, , and B. L. Hua, 1991: On the catalytic role of high baroclinic modes in eddy-driven large-scale circulations.

,*J. Phys. Oceanogr.***21**, 976–997.Duhaut, T., , and D. Straub, 2006: Wind stress dependence on ocean surface velocity: Implications for mechanical energy input to ocean circulation.

,*J. Phys. Oceanogr.***36**, 202–211.Farneti, R., , T. Delworth, , A. Rosati, , F. Griffies, , and S. M. Zeng, 2010: The role of mesoscale eddies in the rectification of the Southern Ocean response to climate change.

,*J. Phys. Oceanogr.***40**, 1539–1557.Gent, P., , and G. Danabasoglu, 2011: Response to increasing Southern Hemisphere winds in CCSM4.

*J. Climate,***24,**4992–4998.Gill, A., 1968: A linear model of the Antarctic Circumpolar Current.

,*J. Fluid Mech.***32**, 465–488.Gnanadesikan, A., 1999: A simple predictive model for the structure of the oceanic pycnocline.

,*Science***283**, 2077–2079.Hallberg, R., , and A. Gnanadesikan, 2001: An exploration of the role of transient eddies in determining the transport of a zonally reentrant current.

,*J. Phys. Oceanogr.***31**, 3312–3330.Hallberg, R., , and A. Gnanadesikan, 2006: The role of eddies in determining the structure and response of the wind-driven Southern Hemisphere overturning: Results from the Modeling Eddies in the Southern Ocean (MESO) project.

,*J. Phys. Oceanogr.***36**, 2232–2252.Henning, C. C., , and G. K. Vallis, 2005: The effects of mesoscale eddies on the stratification and transport of an ocean with a circumpolar channel.

,*J. Phys. Oceanogr.***35**, 880–896.Hutchinson, D., , A. Hogg, , and J. R. Blundell, 2010: Southern Ocean response to relative velocity wind stress forcing.

,*J. Phys. Oceanogr.***40**, 326–339.LaCasce, J., , and P. Isachsen, 2010: The linear models of the acc.

,*Prog. Oceanogr.***84**, 139–157.Marshall, J., , and K. Speer, 2012: Closure of the meridional overturning circulation through southern ocean upwelling.

,*Nat. Geosci.***5**, 171–180.Mazloff, M. R., , P. Heimbach, , and C. Wunsch, 2010: An eddy-permitting Southern Ocean state estimate.

,*J. Phys. Oceanogr.***40**, 880–899.McWilliams, J., 1977: A note on a consistent quasigeostrophic model in a multiply connected domain.

,*Dyn. Atmos. Oceans***1**, 427–441.Meredith, M., and Coauthors, 2011: Sustained monitoring of the southern ocean at Drake Passage: Past achievements and future priorities.

,*Rev. Geophys.***49,**RG4005, doi:10.1029/2010RG000348.Munday, D., , H. Johnson, , and D. Marshall, 2013: Eddy saturation of equilibrated circumpolar currents.

,*J. Phys. Oceanogr.***43,**507–532.Nadeau, L.-P., , and D. Straub, 2009: Basin and channel contributions to a model Antarctic Circumpolar Current.

,*J. Phys. Oceanogr.***39**, 986–1002.Nadeau, L.-P., , and D. Straub, 2012: Influence of wind stress, wind stress curl, and bottom friction on the transport of a model Antarctic Circumpolar Current.

,*J. Phys. Oceanogr.***42**, 207–222.Nadiga, B., 2008: Orientation of eddy fluxes in geostrophic turbulence.

,*Philos. Trans. Roy. Soc. London***A366**, 2491–2510.Pedlosky, J., 1996:

*Ocean Circulation Theory*. Springer, 464 pp.Rhines, P., , and W. Young, 1982: A theory of wind driven circulation. I. Mid-ocean gyres.

,*J. Mar. Res.***40,**559–596.Rhines, P., , and R. Schopp, 1991: The wind-driven circulation: Quasi-geostrophic simulations and theory for nonsymmetric winds.

,*J. Phys. Oceanogr.***21**, 1438–1469.Stommel, H., 1957: A survey of ocean current theory.

,*Deep-Sea Res.***6**, 149–184.Straub, D. N., 1993: On the transport and angular momentum balance of channel models of the Antarctic Circumpolar Current.

,*J. Phys. Oceanogr.***23**, 776–782.Tansley, C. E., , and D. P. Marshall, 2001: On the dynamics of wind-driven circumpolar currents.

,*J. Phys. Oceanogr.***31**, 3258–3273.Thompson, A. F., , and J.-B. Sallee, 2012: Jets and topography: Jet transitions and the impact on transport in the Antarctic Circumpolar Current.

,*J. Phys. Oceanogr.***42,**956–972.Viebahn, J., , and C. Eden, 2010: Towards the impact of eddies on the response of the Southern Ocean to climate change.

,*Ocean Modell.***34**, 150–165.Young, W. R., , and P. B. Rhines, 1982: A theory of the wind-driven circulation. II: Gyres with western boundary layers.

,*J. Mar. Res.***40**, 849–872.

^{1}

Although there is clear evidence of Sverdrup gyres in this simulation, the vertical structure of the circulation differs somewhat from that assumed in the analytic model of NS09. (Recall that a continental rise topography along the eastern boundary calls into question the integration along characteristics from an eastern boundary used in deriving the analytic estimate of transport.)

^{2}

Note that blocked potential vorticity contours do not preclude all circumpolar flow; for example, there could still be (and is) an eddy-driven mean flow that adds to circumpolar transport. One nonetheless expects a removal or reduction in the strength of the block to be associated with increased transport.

^{3}

Values for the frictionally compensated transport across the western boundary layer seen in Fig. 12a showed considerably higher values, up to 25% of the maximum Ekman transport. Even so, this was only in grid points immediately adjacent to the wall. Also, since the ACC path that lies near the wall for only a relatively small fraction of its total length, one again concludes that frictional compensation of across-ACC Ekman transport is minimal.