Intrinsic variability of the Antarctic Circumpolar Current system : low-and high-frequency fluctuations of the Argentine Basin flow

Introduction Conclusions References


Introduction
Ocean model studies of various degrees of complexity forced by steady forcing have suggested in the last two decades that a substantial fraction of the low-frequency variability (LFV) of oceanic frontal structures (ranging from the interannual to the decadal and interdecadal timescales) may be due to highly nonlinear oceanic mechanisms internal to the ocean system: this is the so-called "intrinsic" LFV variability, that can advantageously be analyzed in the conceptual framework of dynamical systems theory (e.g., see Dijkstra, 2005;Dijkstra and Ghil, 2005, for reviews).The mechanisms can involve barotropic and baroclinic instability, eddymean flow interaction, Rossby wave propagation and interaction with topographic and coastal features.In general, identifying the intrinsic part of the ocean variability, and the modality through which it emerges, is necessary to assess the role played by the ocean in the global climate.In addition, studying these ocean changes is fundamental also from a modeling point of view, as they may not be properly represented even in high-resolution general circulation models due to their high sensitivity to the parameterization of unresolved processes.Most studies have been devoted to the major mid-latitude western boundary currents and their extensions, such as the Kuroshio (e.g., Qiu and Miao, 2000;Schmeits and Dijkstra, 2001;Pierini, 2006Pierini, , 2008;;Pierini and Dijkstra, 2009;Pierini et al., 2009), the Gulf Stream (e.g., Schmeits and Dijkstra, 2001;Quattrocchi et al., 2012), and the Agulhas Current (e.g., Dijkstra and De Ruijter, 2001; Le Bars et al., 2012).
A good candidate for considerable intrinsic LFV is also the Antarctic Circumpolar Current (ACC) system (e.g., Rintoul et al., 2001) and its complex frontal structure in the Southern Ocean.The ACC plays an important role in the global climate through local water mass formation, ocean carbon sequestration and heat storage, and because of its ability to connect climate signals at all longitudes.Hogg and Blundell (2006) analyzed the LFV of the ACC by using an idealized multilayer quasi-geostrophic model forced by steady winds.Their simulations display robust intrinsic LFV, which is shown to involve a positive feedback between baroclinic eddies and the mean circulation.Penduff et al. (2011) analyzed the sea level expression of intrinsic LFV in simulations of an eddy-permitting ocean general circulation model (OGCM), and found their results in the Southern Ocean to be basically consistent with those of Hogg and Blundell (2006).O' Kane et al. (2013) used an OGCM (including sea ice) to identify a Southern Ocean southeast Pacific mode of intrinsic LFV through a suite of experiments that include low-frequency (ENSO, SAM) and high-frequency stochastic forcing.Giarolla and Matano (2013) used long time series of sea surface height (SSH), sea surface temperature and wind stress curl to determine the main modes of LFV of the Southern Ocean circulation, but they did not investigate its intrinsic component.
An oceanic region particularly relevant for climate that affects the Southern Ocean dynamics, and where important LFV was documented, is the Argentine Basin (AB), located in the South Atlantic sector near the South American coast.The AB is a crucial region where strongly contrasted water masses meet and mix (Fig. 1), and where the circulation is likely to influence meridional water exchanges between the Southern Ocean and subtropical latitudes (Piola and Gordon, 1989) with associated strong impacts on the global climate system.An intense barotropic anticylonic circulation called Zapiola Anticyclone (ZA) was documented over the Zapiola Rise (ZR), a high sedimentary deposit located in the middle of the AB (Weatherly, 1993;Whitworth et al., 1991;Saunders and King, 1995).The ZA affects considerably the surface exchanges between the ACC and the South Atlantic Current (Smythe-Wright and Boswell, 1998) and is supposed to contribute to determine global deep water mass characteristics (Garzoli et al., 2008).
The ACC dominates the southern part of the basin, splitting into two major fronts, the Polar Front and the Subantarctic Front (Fig. 1).The latter flows northwards after leaving the Drake Passage, forming the Malvinas Current, which transports cold and relatively fresh subantarctic waters equatorward.Its return flow eventually aligns with the southern edge of the ZR and represents the local manifestation of the Subantarctic Front.This front and the Polar Front join at around (49 • S, 45 • W) and diverge further east.From the north, the warm and salty Brazil Current flows southward along the continental shelf, colliding with the Malvinas Current at around 38 • S, where the very energetic and turbulent region known as the Malvinas-Brazil Current Confluence is created.
Significant LFV of the ZA was observed with in situ (Hughes et al., 2007) and altimeter data (Saraceno et al., 2009).Saraceno et al. (2009) documented for the first time that the ZA flow may significantly decrease in strength or even vanish over an interannual timescale, with a cyclonic pattern emerging from time to time.Bigorre and Dewar (2009) developed an idealized quasi-geostrophic ocean process study about the circulation around a large-scale topographic anomaly: the role of bottom friction and eddy diffusivity was found to be consistent with the theory proposed by Dewar (1998) for the mean flow, moreover the modeled LFV was shown to bear important similarities to that observed.Venaille et al. (2011) found intrinsic high-and low-frequency variability of the ZA in their comprehensive ocean model, and explained the internal part as the result of an eddy-driven stochastic process.In the same basin also high-frequency fluctuations were observed (Fu et al., 2001;Tai and Fu, 2005), and were interpreted in terms of topographic Rossby modes (Weijer et al., 2007a, b) and mesoscale variability (Fu, 2007).
In this paper, a model study aimed at identifying and analyzing the intrinsic variability of the ACC system, with a focus on its Subantarctic Front and relative effect on the AB flow, is presented.A primitive equation sigma-coordinate ocean model is implemented in a large portion of the Southern Ocean with an eddy-permitting resolution under steady forcing; the climatological forcing and the stratification are substantially idealized, while the topography (that plays a fundamental role at these high latitudes) is represented in more detail (Sect.2).In Sect. 3 the LFV of the model flows is analyzed, with an emphasis on the AB where the highest level of variance is found.Regime switches occurring on interannual timescales of the AB flow are found to share important similarities with the long-term variations of the ZA as observed by Saraceno et al. (2009).In Sect. 4 the intrinsic high-frequency variability (HFV) in the AB, emerging as a residual from the LFV, is analyzed through a wavelet analysis and interpreted in terms of topographic Rossby waves and mesoscale eddies.In Sect. 5 the relation between the LFV and the HFV is analyzed by introducing dynamical indices, and possible mechanisms of mutual interaction are suggested.Finally, in Sect.6 conclusions are drawn.

The model
The model used in this work (Sgubin, 2012) is the Princeton Ocean Model (POM) developed by Blumberg and Mellor in 1977 and subsequently improved and updated (for general information see www.aos.princeton.edu/WWWPUBLIC/htdocs.pom/).POM is a primitive equation sigma-coordinate model that contains a turbulence submodel for vertical mixing based on the Mellor-Yamada scheme (Mellor and Yamada, 1982).Details of the model equations and numerical solution techniques can be found in the POM user's guide (Mellor, 2003).The quantitative success of the ZA circulation model of de Miranda et al. (1999) was attributed to the use of this kind of vertical discretization (see also Barnier et al., 2006, for relevant modeling issues).
The model domain includes Pacific and Atlantic sectors of the Southern Ocean, extending meridionally from 33.2 to 72 • S and longitudinally from 120 to 0 • W (Fig. 2).Periodic boundary conditions along the eastern and western meridional boundaries are imposed with a 12 • transition region (with 10 • to the west and 2 • to the east; in that region the bathymetry is interpolated so as to match at the two boundaries).On the northern and southern boundaries free-slip boundary conditions are imposed at all depths.
The Mercator grid is adopted with the ETOPO5 bathymetry data (available online at http://www.ngdc.noaa.gov/) on a grid with a horizontal spatial resolution of one-fifth of a degree in latitude (∼ 22 km) and two-fifths of a degree in longitude (∼ 13.75-37.2 km).A limitation of sigma-coordinate models is related to the existence of pressure gradient errors that arise when computing the horizontal pressure gradient near a steep topography (e.g., Haney, 1991;Beckmann and Haidvogel, 1993).This error is caused by large numerical truncations in the transformation from the z coordinate to the sigma coordinate.A reduction of such errors to acceptable values can be achieved by properly smoothing the bathymetry according to a criterion involving the bottom slope and the horizontal and vertical grid resolution.At the same time, such smoothing should retain the main features of the topography in order to consistently reproduce the interactions between the flow and the oceanic bottom, and should avoid hydrostatic inconsistency.This problem was successfully handled by Barnier et al. (1998), Marchesiello et al. (1998) and de Miranda et al. (1999) in setting up a consistent model for the South Atlantic circulation.We have therefore applied the method of Barnier et al. (1998) to obtain a model topography that reduces drastically the pressure gradient error.The method imposes an upper bound on the maximum relative variation r h of the model's ocean depth, defined as where h is the local water depth.By imposing r h < 0.3 the error has been reduced by two orders of magnitude compared to the unsmoothed topography case.Furthermore, in order to avoid hydrostatic inconsistency related to the vertical integration scheme, the maximum relative increase in bottom topography must be less than the distance between two consecutive sigma levels (Barnier et al., 1998).We have thus performed a further smoothing to satisfy this requirement for every vertical level.
The Smagorinsky parameterization has been used for the horizontal eddy viscosity with the dimensionless HORCON parameter (Mellor, 2003) H = 0.12.The model has 12 vertical sigma levels and is integrated for 80 years from motionless initial conditions with external and internal time steps t e = 20 s and t i = 600 s, respectively.approach typical of process studies of the intrinsic LFV, an idealized but relatively realistic steady zonal wind stress field is used: Fig. 3a shows the wind stress profile based on the Trenberth et al. (1989) climatology.The initial stratification, again idealized but relatively realistic, is shown in Fig. 3b for the temperature (the salinity has been held constant): it takes into account the vertical as well as the meridional density gradient, which effectively influences the zonal flow at Drake Passage.In order to sustain such a stratification, idealized steady surface heat fluxes have been imposed: their spatial structure (varying sinusoidally in latitude) has been empirically evaluated so that the initial stratification is not significantly altered during the simulations.
Figure 4 shows the volume transport through the Drake Passage: the average value of ∼ 116 Sv is in reasonable agreement with the real estimated value of ∼ 130 Sv (e.g., Rintoul et al., 2001); an energetic HFV is present in the signal while the LFV in this integrated parameter appears weaker.Figure 5a and b show the temporal means of the SSH η and of the depth-integrated current u, respectively.Figure 5c  The Brazil Current and the Subtropical Front (located at around 39 • S) associated with the Malvinas-Brazil Current Confluence are absent due to the choice of the northern latitude of the domain of integration and of the wind forcing.Thus, the flow along the South American coast is unrealistic north of around 40 • S.However, this limitation does not affect our analysis, which is mainly focused on the intrinsic variability of the ZA directly induced by changes of the ACC frontal system, particularly through its Subantarctic Front (this is in fact an innovative aspect of the present study, in which these sources of intrinsic variability can be isolated).Thus, the branch of the Subantarctic Front indicated by the red arrow in Fig. 5c does not derive from the Malvinas return flow (which is absent for the same reason discussed above), but from a southward current which, mainly because of topographic interactions, acquires nonetheless the correct shape and location.

Intrinsic low-frequency variability
In this section the LFV produced by the model is presented and discussed.The low-frequency signal has been derived by applying a moving average with T = 200 days at each grid point.The decimal logarithm of the resulting rms (root mean square) of the SSH is shown in Fig. 6.Apart from a moderate variability in regions of strong topographic variations between ∼ 230-260 • E and 345-360 • E, a very intense LFV is present all along the Subantarctic Front between ∼ 302 and 345 • E. The particularly intense variability across the southern topographic limits of the AB is consistent with the findings of Saraceno et al. (2009).The analysis will therefore be focused on this region.Figure 7b shows the time series of the SSH taken at point P1 where the variability is maximum (see Fig. 7a), which is ∼ 2 • south of the ZA center.The behavior yields a chaotic vacillation of O(1 m) that has a bimodal character: a detailed analysis will be carried out within the 10-year reference interval delimited by the red lines in Fig. 7b.24 200 days) and after (t = 25 000 days) the anticyclonic circulation is weaker and shifted westward by 2-5 • , while a cyclonic circulation moving from the southeast intensifies and takes its place.Sequence (b) shows an abrupt transition from a collapsed but quite variable ZA (t = 26 200-26 600 days) to an intense ZA (t = 27 000-27 200 days).These transitions from a quasi-climatological state of the ZA (denoted as "active state" heretofore) to a collapsed ZA (and vice versa) are very similar to the variations of the ZA documented by Saraceno et al. (2009).The hypothesis is therefore that the oceanic intrinsic variability plays an important role in this phenomenon.
The preceding analysis has shown that a ZA in an active state is characterized by large values of η at P1 ( η ≈ 1 m) while, for a collapsed ZA, η ≈ 0-0.3 m; to this respect, Fig. 7b may provide qualitative information on the character and statistics of the transitions.However, η gives only very local information and is not necessarily indicative of a circulation regime.A better way to characterize the ZA state is to rely on the (dimensionless) relative vorticity ζ (and of its low-frequency version ζ ) averaged over specific regions of the AB.The two maps in Fig. 10 show ˜ ζ in a collapsed (t = 25 000 days) and active state (t = 27 200 days); it is evident that the two sectors A and B can very efficiently characterize the ZA state in terms of ˜ ζ averaged in each of them.The graph in Fig. 10 shows ζ A (red line) and ζ B (blue line) as a function of time for the reference interval: as could be expected, in a collapsed state ζ A > 0 and ζ B < 0 whereas for an active state both parameters are positive, with a tendency of ζ B to be greater.Figure 11 shows the scatter plot of ζ B versus ζ A for the whole 80 year integration: the active state is represented by the compact cluster with ζ B > 0 while the collapsed state with ζ B < 0 is represented by a more diffuse cluster, which implies that this state cannot be characterized by a well-defined circulation pattern, as is the case for the active state.Moreover, the abrupt character of the transitions is clearly shown by the small number of dots in the intermediate range.Possible mechanisms that govern this LFV will be discussed in Sect. 5.

Intrinsic high-frequency variability
In this section the high-frequency component of the intrinsic variability is analyzed, while its relation with the LFV will be considered in the next section.The high-frequency component is defined here in terms of the SSH as the residual η = η − η, and therefore includes periods shorter than T = 200 days.Figure 12 shows the rms of η in the AB: a region of intense variability crosses the isobaths from the deep sea just off the continental shelf in the southwestern side of the ZR, reaching the peak of the ZR itself, and presents two maxima at P2 (313 • E, 47 • S) and P3 (317 • E, 45.5 • S).
Figure 13a shows the time series of η (red line) and η (blue line) at P3.The HFV has a very intermittent behavior and yields an apparent relation with the LFV, in that the high frequency is more energetic in the collapsed state, corresponding to low values of η.Because of the intermittency of the HFV (also found in altimeter data by Tai and Fu, 2005;Fu, 2007), a spectral analysis can most appropriately be carried out by means of the continuous wavelet transform (e.g., Torrence and Compo, 1998).Figure 13b shows the amplitude w (s, t) of the corresponding wavelet transform of η, where s is the timescale (only the scales s ≤ 200 days are shown).The period band of 0-50 days includes topographic Rossby waves and modes (Fu et al., 2001): here the amplitude is reduced and very intermittent over timescales ranging from 100 to 300 days (in agreement with Fu, 2007).In the period band of 100-150 days the amplitude is higher and less intermittent, but is often clearly related to the higher frequency variability (this is also consistent with the results of Fu, 2007, but see the next section for a more careful comparison).

Ocean
For a qualitative analysis of the flow patterns of the HFV, a sequence of snapshots of the SSH residual η is reported in Fig. 14.The 25-day-long sequence is sampled every 5 days and corresponds to a maximum in the wavelet amplitude at P3 in the period band 0-50 days (see the oval in Fig. 13b).Very complex patterns arise, with length scales ranging from O(1 • ) or less for the mesoscale up to O(5-10 • ) for topographic Rossby waves.The variability is mainly confined over the ZR and shows a clear propagation of features originating from the southwestern side of the ZR itself.A wave train follows the southern limits of the rise and turns counterclockwise along its eastern flank: in doing so the vortices undergo substantial stretching and deformation.These waves can be interpreted as topographic Rossby modes (Fu et al., 2001;Tai and Fu, 2005;Weijer et al., 2007a, b).The patterns appear more complex than those typically shown in this location by altimeter data after high-pass filtering the motions with timescales longer than ∼ 1 month (e.g., Fu et al., 2001).This is because our high-frequency signal contains also the longer-term variability associated with the mesoscale eddy field.
A general feature that deserves to be emphasized is the varying length scale of the vortices, which is smaller in the southwestern side of the ZR and tends to increase as the pattern propagates.This transition from the mesoscale to the Rossby mode scale is compatible with the energy exchange found to be at work in this region by Fu (2007); this aspect will be analyzed in more detail in the next section.From this qualitative analysis it appears that the topographic Rossby modes in the ZR are not necessarily generated directly by the wind forcing, but may also be of intrinsic origin.Presumably this happens through low-frequency fluctuations of the local circulation (e.g., see Pierini, 1996), that in this case are intrinsic.This also supports the hypothesis that the wind-generated topographic Rossby modes may as well be generated through the same mechanism, in which case the current fluctuations that produce them are not intrinsic but directly wind-driven (e.g., Pierini et al., 2002; see Sect.5.3 for further discussion).

Relation between low-and high-frequency variability
In the Sects.3 and 4 we have identified intrinsic LFV on interannual timescales yielding regime switches from an active ZA to a collapsed ZA, but HFV was found as well.A high-frequency range (HFR, 0-50 days) includes topographic Rossby modes; the behavior in an intermediate highfrequency range (IHFR, 100-150 days) appears to be related to that of the HFR.Now, a question arises: is there a relation between these three forms of intrinsic variability in our model results?In general, analyzing this issue is fundamental from a theoretical viewpoint, as it could shed light dynamical mechanisms that involve a wide range of spatial and temporal scales.
A preliminary qualitative analysis of this kind is presented here.First of all we define a Zapiola index as follows: The graph in Fig. 10 shows ζ A and ζ B for the reference interval: the two signals are virtually in counterphase in a ZA collapsed state while they tend to be both positive in an active ZA state, with the second signal being higher.Thus, Z as defined in Eq. (1) (and shown by the green line in Fig. 15 for the reference interval) is a good global, low-frequency indicator of the ZA state, for which a large positive value implies a collapsed state, while a negative or small positive value implies an active state.To construct high-frequency indices we can define the integral of the wavelet amplitude at the reference point P3 within two timescales: (2) Thus, the two indices W 1,50 (s = 1 day is the smallest scale of the wavelet transform) and W 100,150 (red and blue lines in Fig. 15, respectively) are good indicators of the behavior in the HFR and IHFR, respectively.We pass to discuss three possible interactions suggested by Fig. 15 and by other experimental and numerical investigations.

Relation between W 100,150 and Z
Several works (e.g., Dewar, 1998;Bigorre, 2005;Bigorre and Dewar, 2009;Volkov and Fu, 2008;Venaille et al., 2011, Saraceno andProvost, 2012) have suggested that the mesoscale eddy activity provides the main source of energy of the ZA.By studying the correlation between the eddy kinetic energy and the LFV of the ZA with 15 years of altimeter records, Saraceno et al. (2009) found support for that hypothesis, suggesting a rapid adjustment of the ZA to changes in the eddy kinetic energy.By using high-resolution altimeter data produced by the Archiving, Validation and Interpretation of Satellite Oceanographic data (AVISO) project, Fu (2007) found that the variance-preserving spectrum of the mesoscale energy time series in the AB is spread over a wide range of frequencies, with the majority in the seasonal-tointerannual range, but a significant variance is present also in the range of 100-160 days (basically our IHFR); thus, W 100,150 represents a good index for the high frequency mesoscale in the AB.
In our model results, both transitions present in the reference interval from a collapsed to an active ZA (occurring at t ≈ 24 400 and t ≈ 26 700 days, when Z decreases abruptly, Fig. 15) are preceded by a large W 100,150 , which decreases as the transition to the active state is occurring, with a lag of 100-300 days between the two signals.Moreover, the collapse of the ZA at t ≈ 24 700 days is followed by an increase of W 100,150 with a lag of ∼ 200 days.This behavior is compatible with a transfer of energy from the eddy field to the large-scale circulation during the collapsed ZA-active ZA transition, and vice versa, and therefore appears to be in basic agreement with the theoretical and experimental arguments reported above: it is interesting to note that purely intrinsic oceanic mechanisms are able to produce such a dynamical process.

Relation between W 1,50 and W 100,150
Using altimeter observations, Fu (2007) analyzed the possible relationship between the 25-day barotropic Rossby waves and the energetic baroclinic mesoscale eddies in the ZA region.The wave amplitude was found to be coherent with the energy of the high-frequency mesoscale variability, in that when the latter decreases the wave amplitude increases and vice versa, suggesting an exchange of energy between the two scales.An indication of this behavior in our model results (for which changes in the wave amplitude and in the mesoscale eddy field are represented by W 1,50 and W 100,150 , respectively) can be found in the intervals t ≈ 23 800-24 700 days and t ≈ 25 900-26 400 days (Fig. 15), which both precede the transition from a collapsed to an active ZA state.The latter property is particularly interesting, as it suggests a relationship between the energy exchange in the high-frequency ranges and the large-scale LFV of the ZA that deserves to be analyzed in future studies.

Relation between W 1,50 and Z
During the transition from an active to a collapsed ZA in the interval t ≈ 24 600-25 000 days no relationship such as the one described above seems to exist between W 1,50 and W 100,150 ; however, the sudden increase of the energy level in W 1,50 follows the abrupt collapse of the ZA.This suggests that an impulsive generation of topographic Rossby modes can be active, perhaps with the involvement of mesoscale eddies as discussed at the end of Sect. 4. To this respect, laboratory experiments in a rotating tank performed by Pierini et al. (2002) confirmed the hypothesis put forward in a previous numerical study on topographic Rossby modes in the Strait of Sicily (Pierini, 1996), according to which topographic Rossby modes can effectively be excited by rapid changes of the mean flow (provided in the tank by the movement of a large paddle).In a Rossby mode the timescale is set by the bottom topography and coastlines (if present) and not by the spectral content of the wind field, which only indirectly forces the modes through changes in the mean flow.This may explain why Fu et al. (2001) did not find any significant wind variability at periods close to 25 days in the AB, nor could they find any significant correlation between the variability of the wave amplitude with the variability of the wind-stress curl.Thus, the impulsive generation of topographic Rossby modes, which appears to be active in our numerical results, should be considered when explaining the variability in our HFR.
It is clear that understanding the intricate mutual interaction among these three different temporal scales requires a much deeper investigation than that presented in this analysis, which cannot by itself determine any causality between the various scales.Nonetheless, these results can complement those quoted above for future, more advanced investigations in this very peculiar and interesting oceanic site.

Conclusions
In this paper, an eddy-permitting sigma-coordinate ocean model has been applied to a large portion of the Southern Ocean with the aim of identifying and analyzing low-and high-frequency fluctuations of intrinsic oceanic origin -a fundamental task for understanding the role of the ocean in the global climate.The approach is typical of processoriented studies of the intrinsic oceanic variability (idealized but relatively realistic steady winds and stratification) but a realistic topography is used because barotropic motions, and so topographic interactions, are important here.The obtained mean flow is in good agreement with observations  as far as the transport through the Drake Passage and the structure of the main Antarctic Circumpolar Current fronts are concerned.Important variability of both low-and highfrequency nature is found, being particularly intense in the branch of the Subantarctic Front corresponding to the Argentine Basin, which, due to its crucial location plays an active role in determining the circulation in the southwestern Atlantic sector of the Southern Ocean.The variability over interannual timescales shows a bimodal behavior of the Zapiola Anticyclone, connecting an active state to a state in which the anticyclonic circulation collapses and sometimes reverses locally: this is in substantial agreement with the altimeter observations of Saraceno et al. (2009).The highfrequency residual signal shows clear evidence of mesoscale propagating patterns particularly along the southern flanks of the Zapiola Rise, and a counterclockwise rotation of largerscale topographic Rossby modes over the rise, in substantial agreement with observations.Thus, the main conclusion is that these forms of variability are compatible with intrinsic generation mechanisms all internal to the ocean system.A preliminary analysis of the mutual relationship between the low-frequency variability and two components of the high-frequency variability puts in evidence interesting agreements with observations and previous theoretical and modeling studies, and at the same time suggests a deeper analysis of the results.
Future perspectives include new simulations using a domain with a larger latitudinal extension, so as to represent the Brazil Current and the Malvinas-Brazil Current Confluence.This will allow us to directly compare the modeled eddy field in the southwestern Atlantic with that observed from altimeter data (e.g., Saraceno and Provost, 2012) and to assess its contribution to the low-frequency variability of the Zapiola Anticyclone.Other studies will be devoted to analyzing the sensitivity of the model response to changes in the forcing and parameterizations.For example, changing the amplitude of the wind forcing and/or the parameterization of dissipative effects can produce important modifications in the intrinsic variability (e.g., see the analysis of Pierini et al., 2009, in the context of the Kuroshio Extension bimodality).Another fundamental aspect that should be analyzed is the effect of time-dependent forcing on the emergence of the intrinsic oceanic variability.The intrinsic variability is often in the form of relaxation oscillations that are self-sustained beyond a given tipping point (global bifurcation in some state space) associated with a particular control parameter, while they do not emerge below that threshold under steady forcing (e.g., Simonnet et al., 2005).However, the same intrinsic relaxation oscillations can emerge even below the bifurcation point provided an appropriate noise (e.g., Sura et al., 2001;Frankcombe et al., 2009;Pierini, 2010Pierini, , 2012)), a deterministic time-dependent component (e.g., Otterå et al., 2010;Crucifix, 2012;Pierini, 2014) or both (e.g., Pierini, 2011)  wind noise and the main modes of atmospheric variability in the Southern Ocean will have to be carried out.
Fig. 3. (a) Steady zonal wind stress profile used to force the circulation model.(b) Initial stratifications (red/blue line: temperature profile at the northernmost/southernmost latitude).
is a zoom of Fig. 5b limited to the southwestern Atlantic (one arrow out of six is drawn).The position and structure of the local ACC branches, i.e., those associated with the Polar Front and the Malvinas Current are captured quite well by the model (e.g., Rintoul et al., 2001; Saraceno et al., 2004, compare with Fig. 1 from that paper).The ZA has the correct shape and location, being centered at ∼ 315 • E, 45 • S. The Polar and Subantarctic fronts meet correctly at around 50 • S just east of 45 • W, and diverge further east (see Fig. 5b), as in the real ocean.

Fig. 5 .
Fig. 5. (a) Mean SSH (in m).(b) Mean depth-integrated currents.(c) Zoom of the current field delimited by the rectangle in (b).

Fig. 6 .
Fig. 6.Logarithm of the rms of the low-frequency SSH signal.

Figure 8
Figure8shows the SSH (total signal η: black line; lowfrequency signal η: blue line) at P1 during the reference interval; the SSH maps in the AB corresponding to the two sequences of six instants denoted by the green and red dots in Fig.8are shown in Fig.9.In sequence (a) a well-defined ZA centered at 315 • E, 46 • S is present at t = 24 600 days, corresponding to a SSH maximum at P1, but 1 year before (t =

Fig. 8 .
Fig. 8. SSH at P1 for the reference interval (black line: total signal; blue line: low-frequency signal).The SSH snapshots corresponding to the green/red dots are shown in Fig. 9.

Fig. 9 .
Fig. 9. Snapshots of the low-frequency SSH signal (in m) corresponding to the green (a) and red (b) dots in Fig. 8.

Fig. 10 .
Fig. 10.Upper panels: low-frequency relative vorticity ˜ ζ at t = 25 000 and t = 27 200 days.Lower panel: time series of ζ averaged in sectors A and B for the 10-year reference interval.

Fig. 11 .
Fig. 11.Scatter plot of ˜ ζ averaged in sectors A and B.

Fig. 12 .
Fig. 12. Rms of the high-frequency SSH signal (in m) in the Argentine Basin.
Fig. 13.(a) Low-frequency SSH at P3 (blue line) and corresponding high-frequency residual (red line) for the reference interval.(b) Wavelet amplitude of the SSH signal for the same point and time interval (the cyan lines indicate the cone of influence).

Fig. 14 .
Fig. 14.Snapshot sequence of the SSH high-frequency residual (in m) corresponding to a high-frequency maximum of the wavelet amplitude at P3 (see the oval in Fig. 13b).
are added to the forcing.Thus, using steady forcing, as done here, constitutes only the first step toward the identification of the intrinsic variability; further studies that include both www.ocean-sci.net/10