In Part 1 of this study, we showed that the Antarctic Circumpolar Current (ACC) consisted of multiple fronts, each of which was consistently associated with a particular contour of sea surface height (SSH) or approximate streamline. In Part 2 we have used maps of SSH to examine the variability of the ACC fronts between 1992 and 2007. The SSH label associated with each frontal branch is nearly constant around the circumpolar belt. The front labels are also nearly constant in time: the bands of enhanced SSH gradient (i.e., fronts) occur along the same streamlines throughout the 15 year period of observations. Both short- and long-period changes of the SSH frontal labels of the ACC are small. Based on a tight relationship between dynamic height and cumulative baroclinic transport of the ACC, the baroclinic transport variability of the individual branches of the ACC is also expected to be small. The major change in the total ACC baroclinic transport occurs in the Drake Passage. The streamline associated with the northern branch of the SAF (SAF-N) does not pass through Drake Passage and the waters carried by this branch are not observed there. Instead, the transport of the SAF-N turns north in the Pacific to supply the export of water to the Indian Ocean north and south of Australia. Strong eddy activity in the southeast Pacific acts to dissipate the hydrographic signature of the SAF-N there. In the Atlantic, the SAF-N reappears as the same streamline is again associated with enhanced SSH gradients to the east of the Brazil – Malvinas confluence zone. While the large changes in SSH have occurred in the Southern Ocean between 1992 and 2007, there are strong regional differences. Because the ACC fronts are robustly associated with particular SSH contours, the changes in SSH reflect shifts in the position of the ACC fronts. In the circumpolar average, each of the ACC fronts has shifted to the south by about 60 km. The changes in SSH in the Southern Ocean are largely due to changes in ocean circulation, rather than warming and freshening by atmospheric fluxes. Much larger changes in SSH are observed in some locations of the Southern Ocean, particularly where the fronts interact with large-scale topography. The northern branch of the PF (PF-N) near the Kerguelen Plateau is an extreme example, where the PF-N followed a path around the northern end of Kerguelen Plateau between 1992 and 1997, passed through the Fawn Trough after 2003, and oscillated between the two paths between 1997 and 2003.
 In the first part of this two-part study [Sokolov and Rintoul, 2009, hereinafter SR2009], we demonstrated that the Antarctic Circumpolar Current (ACC) consists of multiple jets aligned along particular streamlines throughout the circumpolar path of the current. The ACC frontal structure inferred from maps of sea surface height (SSH) was shown to be consistent with that inferred from high resolution hydrographic data, where particular water mass features are used to identify the fronts. The SSH value associated with each frontal branch was found to be nearly constant, both in time and around the circumpolar path. Sokolov and Rintoul  summarized earlier studies of the frontal structure of the ACC from the perspective of hydrography, high resolution numerical simulations and geophysical turbulence theory.
 The emphasis of Part 1 was on the mean structure of the ACC fronts. Here we investigate the variability of the fronts in space and time. We exploit the fact that the multiple ACC jets are associated with particular contours of SSH to produce a continuous time series of circumpolar maps of the ACC fronts for the period between 1992 and 2007, with a resolution of one week in time and 100 km or less in space. We use the 15 year record to determine the low-frequency variability of the ACC fronts and its relationship to changes in SSH and water mass properties observed in the Southern Ocean.
 A number of recent studies have demonstrated that the state of the Southern Ocean is changing. For example, Gille [2002, 2008] showed that the Southern Ocean is warming more rapidly than the rest of the global ocean. Gille suggested the warming likely reflected a southward shift of the ACC, as well as subduction of warmed surface waters. Böning et al.  analyzed temperature, salinity and density changes along mean streamlines in the Southern Ocean and found that the Southern Ocean was warming and freshening, consistent with a southward migration of isopycnals. Morrow et al.  discussed a trend toward higher SSH south of Australia and showed that the deep ocean made a significant contribution to the steric height trend. With regard to interannual variability of the fronts, Sallée et al.  used the idea that fronts were associated with SSH contours to examine how shifts of the fronts were related to the major climate modes in the southern hemisphere, the Southern Annular Mode (SAM) and El Niño – Southern Oscillation (ENSO).
 We first demonstrate that the SSH values associated with each frontal branch (frontal SSH “labels”) are nearly constant, both in time and with position around the circumpolar belt. The intensity of the fronts (i.e., the SSH gradient across the fronts) varies strongly with location, but does not vary much with time. SSH itself, however, does change significantly during the period between 1992 and 2007 (e.g., SSH south of 30°S increased by 0.048 m between 1992 and 2007). We use the time sequence of maps of the ACC fronts to demonstrate that the regional distribution of trends and variability in SSH can be explained by shifts of the ACC fronts. Then, we use SSH to identify rings pinched off from the fronts. In many cases the distribution of rings is very anisotropic (e.g., formation of warm rings is favored over cold rings, or vice versa), suggesting rings carry significant net transports of mass, heat, salt and other properties across the ACC fronts. Finally, we demonstrate how the intensity and the number of the ACC frontal branches observed circumpolarly match with the global circulation pattern of the World Ocean interbasin exchange facilitated by the ACC.
2. Data and Methods
 We used 15 years (1992–2007) of weekly SSH gradient fields [Le Traon et al., 1998] to map ACC fronts in the entire Southern Ocean. The approach used to identify fronts in SSH data is described in detail by Sokolov and Rintoul  and SR2009. The SSH anomalies were added to a mean surface dynamic height field (relative to 2500 dbar) calculated using the WOCE hydrographic climatology [Gouretski and Koltermann, 2004] (see Sokolov and Rintoul  for a discussion of the sensitivity to the mean field used). We subdivided the circumpolar belt into 12 sectors of 30° of longitude and calculated the spatial gradient of each weekly SSH map. The regions of high ∇SSH tend to be aligned along particular SSH contours (approximate streamlines). A fitting procedure is used to find the SSH contours that most efficiently describe the distribution of ∇SSH maxima. Regions of high gradient were defined as those that exceeded a threshold of 0.25 m per 100 km. Local maxima in SSH gradients were also included to capture weaker southern ACC fronts. The SSH values of the best fit contours are used as “labels” for each frontal branch.
 As with SR2009, we have used 12 SSH contours to map fronts in the Southern Ocean (Table 1). Nine contours are associated with the ACC itself, while three contours correspond to elevated SSH gradients associated with subtropical western boundary currents and their extension along the northern edge of the Southern Ocean. We validated the ACC front positions inferred from satellite SSH maps using independent data from Argo floats and high resolution hydrographic sections. In each case we found remarkable consistency between the front locations inferred using traditional water mass features on synoptic hydrographic sections and the front positions inferred from ∇SSH maps [SR2009].
Table 1. SSH Labels Associated With Each Branch of the ACC Frontsa
Labels are given for two choices of reference level used to derive the mean dynamic height field added to the SSH anomalies to construct absolute SSH. SAF, Subantarctic Front; PF, Polar Front; SACCF, southern ACC front; SB, southern boundary of the ACC; -N, northern branch; -M, middle branch; -S, southern branch. The error bars represent circumpolar changes in SSH frontal labels (±1 standard deviation).
Relative to 2500 dbar.
Relative to 1500 dbar.
1.97 ± 0.02
1.44 ± 0.02
1.71 ± 0.02
1.23 ± 0.03
1.47 ± 0.02
1.03 ± 0.02
1.29 ± 0.01
0.90 ± 0.01
1.15 ± 0.01
0.80 ± 0.01
1.04 ± 0.02
0.72 ± 0.01
0.94 ± 0.02
0.64 ± 0.02
0.84 ± 0.01
0.57 ± 0.01
0.75 ± 0.01
0.51 ± 0.01
 The Argo data reduction is described in detail by SR2009. A total of 70387 profiles reaching 1500 m depth were collected in the Southern Ocean prior to March 2008. Since the most of the Argo profiles do not extend to the depths of 2500 m, we repeat the ACC front mapping using the reference level of 1500 dbar. The correspondence between the frontal labels using the two choices of reference level is provided in Table 1.
 We found that a reference level of 2500 dbar for the mean surface dynamic height field is most suitable for mapping ACC fronts. This level is deep enough to capture most of the baroclinic signal associated with thermohaline changes across the Southern Ocean, while leaving only a few shallow regions where the mean dynamic height is not defined. The ACC fronts are deep reaching [Sokolov and Rintoul, 2002] and rarely cross regions shallower than 2000 m. On encountering shallow topography, the fronts tend to wrap around the bathymetric features and their paths are well approximated by the 2000 m isobath. In regions shallower than 2500 m, the mean surface dynamic heights relative to 2500 dbar were extrapolated according to the changes in density field at shallower depths. The SSH over the shallow bathymetry was used to determine whether the frontal branch passed the shallow feature to the left or the right (i.e., the front path was chosen so that higher SSH was to the left of the front facing downstream).
3.1. Stability of the SSH Labels Associated With Each Front
Figure 1 shows the mean position of the Southern Ocean fronts inferred from SSH. The dashed lines show the front positions based on the circumpolar average label for each branch; the solid lines show the position of the fronts based on the best fit SSH contours in each 30° sector. Throughout the Southern Ocean, the dashed and solid lines are very close to each other, indicating that the circumpolar average SSH values provide a good approximation to the location of each frontal branch. This is consistent with the demonstration by SR2009 that variations in the local front labels around the circumpolar belt are small (less than 2–3 cm, Table 1), much smaller than the change in SSH between adjacent fronts (20–60 cm). Only in a few regions characterized by weak gradients in SSH (and other properties) do the front positions inferred using local or circumpolar mean labels diverge (e.g., the northern branch of the southern ACC front (SACCF) over the Enderby Abyssal Plain at 40–50°E and Amundsen Abyssal Plain at 220–230°E).
 While our main focus in this paper is the fronts of the ACC, we comment briefly here on the SSH gradients observed north of the ACC. The SSH labels associated with enhanced ∇SSH features in the Subantarctic Zone (SAZ) and southern subtropical gyre are more variable than the ACC fronts. The number of SSH contours needed to explain the enhanced ∇SSH bands also varies with location, reflecting the local circulation structure. Typically, three SSH contours are sufficient to represent maxima in ∇SSH north of the ACC. The intense fronts associated with the poleward extension of the subtropical western boundary currents, such as the Agulhas and Brazil Currents, are observed at higher SSH values reflecting the influx of lighter subtropical waters. For example, the largest deviations of local SSH labels from the circumpolar average are found where the Agulhas Current injects Indian Ocean waters into the Southern Ocean (up to 0.28 m [SR2009, Figure 3]). Downstream of the inflow regions, the frontal labels gradually shift to lower values as a result of mixing between subtropical and subantarctic waters and heat loss to the atmosphere. In these mixing regions, the enhanced SSH gradients observed in the SAZ rather reflect perturbations in the depth of the pycnocline due to the presence of eddies and Rossby waves. The enhanced SSH gradients north of the SAF do not correspond to the Subtropical Front (STF), which is shallow and has little signature in SSH (see Sokolov and Rintoul  for further detail).
3.2. Temporal Changes in SSH Labels Associated With Each Front
SR2009 demonstrated that the SSH labels associated with each frontal branch are nearly constant in time within the SSH range spanning the ACC, with variations not exceeding 3 cm. The linear trends in SSH frontal labels estimated from the sequence of weekly maps are also small (Figure 2). The SSH frontal labels over the last 15 years have increased slightly in the Antarctic Zone (AZ) (by less than 15 mm), reduced slightly in the SAF dynamic height range (by 10–20 mm) and risen by a comparatively large amount in the SAZ and subtropical waters (by 20–40 mm). These changes are one to two orders of magnitude smaller than the difference in SSH labels between adjacent frontal branches. We conclude that the circumpolar and time mean SSH labels provide a good approximation for the location of the fronts throughout the period between 1992 and 2007.
 While the SSH labels associated with each frontal branch do not change much around the circumpolar path, the intensity of the SSH gradient across each front varies strongly with longitude (Figure 3). The strongest SSH gradients observed south of 30°S are in the eastward extension of the Agulhas Return Current. East of 50°E, the northern and middle branches of the SAF merge with the Agulhas Return Current (ARC), extending the band of high ∇SSH eastward to the north of the Kerguelen Plateau. Hot spots of ∇SSH within the ACC itself are observed near 30°E, downstream of the Kerguelen Plateau, south of Tasmania and New Zealand, over and downstream of the fracture zones at 220°E, and in the Drake Passage and Scotia Sea. The regions of enhanced SSH gradients tend to span a number of fronts.
 While the intensity of the fronts varies strongly with position, the distribution of elevated gradients of SSH as a function of SSH is relatively constant in time in any one location (Figure 4). Figures 4a–4l corresponds to a sector of 30 degrees of longitude, starting with 0–30°E in the upper left. The percentage of the area along instantaneous SSH contours occupied by high gradients or local maxima in ∇SSH is plotted as a function of time and SSH. Within each sector, enhanced gradients of SSH remain along the same streamlines throughout the 15 year record. However, the distribution of enhanced ∇SSH as a function of streamline varies strongly along the circumpolar path. Where the extensions of the subtropical western boundary currents approach the ACC, large gradients of SSH are observed at high values of SSH (e.g., 0–60°E, Figures 4a and 4b, (Agulhas Current) and 300–330°E, Figure 4k (Brazil Current)). The strong gradients shift to lower values of SSH moving eastward from 0°E (Figures 4a–4c), reflecting the peeling off of streamlines associated with the ARC and subtropical gyre (see Figure 3). The branches of the SAF correspond to SSH values between 1.97 and 1.47 m (Table 1). High gradients are frequently observed in the SAF from the Kerguelen Plateau (70°E) to Drake Passage (Figures 4c–4i), but with some variations with longitude driven by interactions with topography: high gradients are less frequent where the SAF frontal branches diverge from each other (e.g., downstream of the Kerguelen Plateau (Figure 4d) and Campbell Plateau (Figure 4g)); and the fronts are intensified where the flow interacts with the midocean ridge (e.g., over the Southeast Indian Ridge (Figure 4e) and the fracture zones near 220°E (Figure 4h)). The PF (SSH from 1.04 to 1.29 m, Table 1) is frequently associated with elevated ∇SSH upstream of the Conrad Rise (Figure 4b), south of Tasmania and New Zealand (Figure 4f), the Pacific fracture zones (Figure 4h) and in Drake Passage (Figures 4j and 4k). Weak gradients are observed across the entire ACC over abyssal plains in the Pacific where the frontal branches diverge from each other (e.g., 180–210°E (Figure 4g), southwest Pacific; 240–270°E (Figure 4i), southeast Pacific; and 330–360°E (Figure 4l), southwest Atlantic). Figure 4 shows that while the intensity of the SSH gradients varies significantly along the path of the ACC, the variation with time is small. (Note that Figure 4 includes all instances of enhanced ∇SSH, not just those occurrences explained by the fitting of SSH contours. For example, when two frontal branches merge, the gradient of SSH is enhanced on all streamlines between the core streamlines associated with each branch. As a result, the individual frontal branches are not well resolved in Figure 4 (see discussion and auxiliary material of SR2009).
3.3. Circumpolar Changes in Transport
Rintoul et al.  showed there was a tight relationship between dynamic height and cumulative baroclinic transport of the ACC. Here, this relationship combined with Drake Passage estimates (based on the repeat WOCE SR1 section) is shown in Figure 5. The fact that the SSH label associated with each frontal branch does not change with time therefore implies the transport of the fronts also does not change. Using the dynamic height – transport relationship, the small temporal variability of the SSH labels would correspond to transport anomalies of less than 2 Sv. Transport time series derived from repeat hydrographic sections in the Southern Ocean suggest that the baroclinic transport variability of the ACC (above and relative to 2500 dbar) is only 4–5 Sv [e.g., Rintoul et al., 2002; Sokolov et al., 2004], consistent with the 0.04 m standard deviation of SSH labels observed in the SAZ [SR2009, Figure 3]. (The ratio of the baroclinic transport above and relative to 2500 dbar to the full depth transport relative to the deepest common depth at each station pair estimated from repeat sections in Drake Passage and south of Tasmania varies by only a few per cent, suggesting the above conclusion holds for total full depth baroclinic transport as well.)
 Baroclinic transport of the ACC through Drake Passage (137 ± 9 Sv [Cunningham et al., 2003; Sokolov et al., 2004]) is significantly smaller than through the African (153 ± 6 Sv [Legeais et al., 2005], corrected for presence of shallow bathymetry) and Australian chokepoints (162 ± 7 Sv [Rintoul and Sokolov, 2001]). (Note that these transport values are for the ACC contribution alone, not the net interbasin exchange across each chokepoint section.) The 25 Sv reduction in the ACC transport between Australia and the Drake Passage balances the export of water from the Southern Ocean to the Pacific to supply the Indonesian throughflow and the Tasman outflow south of Tasmania [see, e.g., Schmitz, 1996; Sokolov and Rintoul, 2000].
 The streamline associated with the northern branch of the SAF does not pass through Drake Passage and the waters carried by this branch are not observed there [SR2009]. Instead, the transport of the SAF-N turns north in the Pacific to supply the export of water to the Indian Ocean north and south of Australia. Strong eddy activity in the southeast Pacific acts to dissipate the hydrographic signature of the SAF-N and contributes to the extreme properties of the SAMW formed there. Intensive cross-frontal exchange across the SAF-N produced by eddies is highly asymmetrical in this region with cold core rings generated by the front more frequently than the warm core rings. As a result, the large net cross-frontal flux is directed equatorward introducing “southern characteristics” into the SAZ. The southeast Pacific SAMW is the coolest, freshest and the most dense variety of the mode water formed anywhere in the Southern Ocean [e.g., McCartney, 1982; McCartney and Baringer, 1993; Sokolov and Rintoul, 2000]. The core properties of this SAMW are the closest to the properties of the AAIW due to intensive mixing across the SAF-N and ultimate dissipation of the front in this region.
 The dynamic height-cumulative transport curve shown in Figure 5 suggests the SAF-N carries about 25 Sv, quantitatively consistent with the drop in baroclinic transport of the ACC between the Australian and Drake Passage chokepoint sections. In the Atlantic, the SAF-N reappears as the same streamline is again associated with enhanced SSH gradients to the east of the Brazil – Malvinas confluence zone.
3.4. Changes in Thermohaline Frontal Indicators Derived From Argo Profiling Floats
 Since the ACC fronts are associated with particular water mass features, it is traditional to use simple phenomenological criteria based on temperature and salinity to identify the ACC fronts [e.g., Orsi et al., 1995; Sokolov and Rintoul, 2002]. While the circumpolar changes in SSH labels are small, and T, S indicators of the fronts are more variable [SR2009, Table 1], it is important to evaluate how these traditional indicators vary circumpolarly.
 Changes in the traditional phenomenological front indicators have a coherent ocean-wide pattern in both Argo and WOCE climatology data (Figure 6). As a proxy for these changes we used the temperature at a depth of 500 m (T500), which is a good approximation for both the SAF (where the temperature between 400 and 500 m is often used to identify the front) and for the southern ACC fronts (where the temperature in the temperature maximum is used). The largest changes in T500 occur north of the ACC (i.e., for dynamic heights greater than 1.44 m relative to 1500 dbar, Table 1), reflecting the shift from relatively warm and salty SAMW in the Indian Ocean to cool, fresh SAMW in the southeast Pacific [e.g., McCartney, 1982].
 The warm and cold anomalies north of the ACC extend across the SAF with diminished magnitude, resulting in changes of 0.6°C in the T500 values associated with the SAF. South of the PF-N (dynamic heights relative to 1500 dbar = 0.90 m), the CDW lies at 500 m depth. Circumpolar changes in T500 are much smaller south of the PF-N (0.05°C). The T500 values associated with the southern ACC fronts vary by about 0.1–0.5°C. The temperature changes south of the PF-N reflect the pathways of the CDW with newer (warmer and more saline) North Atlantic Deep Water entering the Southern Ocean in the Atlantic sector and spreading circumpolarly by the ACC [see e.g., Sun and Watts, 2001].
 Note that the largest changes in temperature at 500 m take place outside the ACC dynamic range in the SAZ and STZ, while the circumpolar changes within the ACC are smaller and typically do not exceed 1.5°C (Figure 6). Nevertheless, the SSH frontal labels are more robust for the purpose of the ACC front mapping due to smaller circumpolar changes of labels relative to the proxy changes across the ACC. Furthermore, the circumpolar pattern of the SSH frontal labels (Figure 2) depends on the integrated thermohaline structure, rather than on the water property distribution at a particular depth.
3.5. Changes in SSH
Figure 2 demonstrates that the trends in the SSH value associated with each front are small (less than 1–2 cm). In contrast, the SSH field itself has substantially larger trends over the period between 1992 and 2007 (Figure 7). The strongest signal is an increase in SSH in the southern part of the subtropical gyres and associated western boundary current extensions in each basin (as discussed by a number of authors using data from slightly different time periods [e.g., Willis et al., 2004; Roemmich et al., 2007]). SSH has also increased in general in the ACC band, but there are strong regional differences. SSH has increased in the Atlantic sector and in the Australian sector of the Southern Ocean [e.g., Morrow et al., 2008] and declined between 45°S and 55°S near 45°E. In the Pacific, both negative and positive trends in SSH are observed.
 Changes in SSH can be caused by changes in water mass properties or changes in circulation (e.g., movement of fronts). Morrow et al.  showed that changes in the deep ocean make a significant contribution to changes in SSH observed in the Southern Ocean (only about one third of the increase in SSH could be explained by warming of the upper 700 m). Given that the fronts of the ACC extend from the sea surface to near the seafloor [e.g., Orsi et al., 1995; Sokolov and Rintoul, 2002], a possible explanation for the deep-reaching thermosteric changes influencing SSH is a shift of the ACC fronts. Sokolov and Rintoul  used repeat XBT and CTD sections to demonstrate that changes in SST were strongly correlated with subsurface changes in temperature, particularly in the frontal zones, and concluded that temperature changes primarily reflected shifts in the ACC fronts. In this section we explore the regional changes in SSH over the ACC in more detail and test the hypothesis that the SSH changes reflect changes in position of the ACC fronts.
 First, the trend in SSH along the mean location of each frontal branch is shown in Figure 8. Comparing Figure 8 to Figure 2, it is clear that the change in height along the mean front locations is substantially larger than the changes in front labels, and the spatial pattern is also very different. This suggests the SSH trends are associated with a change in front position, rather than a shift of the ACC jets to a different streamline. Figure 9 shows the displacement of SSH contours (in degrees of latitude) as a function of time and longitude, averaged over bands corresponding to the SAZ/STZ, SAF, PF and SACCF (the SSH contours bounding each zone are indicated by the bold lines in Figure 7). All four zones have a general tendency toward a southward displacement of SSH contours with time, with the exception of the region between 25 and 50°E for the PF and SACCF. The strongest trends for the PF/SACCF are observed between 70°E and 150°E over the Australian-Antarctic Basin, corresponding to a net southward shift of about 1 to 1.5 degrees of latitude between 1993 and 2006. Anomalies of the SAZ/STZ and SAF fronts are strong in the southwest Atlantic. (Strong anomalies of alternating sign in Figure 7 occur in regions where the SSH contours bounding each zone have a meridional orientation, such that small zonal displacements in the location of the contour result in large changes in latitude).
 Averaging the SSH contour displacements shown in Figure 9 around the circumpolar belt, the mean displacement of SSH contours in each frontal zone can be determined (Figure 10). The magnitude of the southward shift (about 0.6 degrees) is similar for each frontal system. The time history is also broadly similar for each front: consistently north of the mean position from about 1993–1997, consistently south of the mean position between about 2004–2007, and more variable during the transition period. Translations of the two primary fronts of the ACC, the SAF and PF, are particularly coherent.
3.6. Regional Examples of SSH Change
 While the long-period changes in the SSH labels of the ACC fronts are small, the long-period changes in the SSH field during last two decades reached 6–8 cm in some regions of the Southern Ocean (Figures 7 and 8). This significant rise in SSH within the dynamic range spanning the ACC was particularly pronounced in the region over the Kerguelen Plateau and further east south of Australia and New Zealand. We have shown that the ACC fronts are associated with the same SSH label throughout the 15 year period of observations. Changes in SSH, therefore, must reflect shifts in the location of the fronts.
3.6.1. Kerguelen Plateau
 SSH averaged in the region between 50 and 100°E indicates that SSH across the entire dynamic height range of the ACC increased from about −4 cm in 1992–1997 to +4 cm in 2004–2007 (Figure 11). The trend in SSH in the vicinity of the Kerguelen Plateau between 1992 and 2007 is shown in Figure 12. The contours overlaid on Figure 12 indicate the location of SSH contours corresponding to each frontal branch for three periods: 1992–1997 (thin lines), 1998–2003 (thick dashed lines) and 2004–2007 (thick solid lines). The northern branch of the PF (dark blue lines in Figure 12) is located at about 52°S upstream of the Kerguelen Plateau. In the early period, the PF-N shifted to the north as it approached the topographic obstruction and passed around the northern edge of the plateau (see Methods section for a discussion of how fronts are tracked in waters shallower than 2500 m). The position of the PF-N shifted to the south immediately upstream of the plateau between 1998 and 2003, but still rounded the plateau on the northern side. The front continued to shift to the south upstream of the plateau during the most recent period (2004–2007), but now passed around the southern end of the plateau. In other words, the SSH increase over the northern Kerguelen Plateau can be explained by a shift of the PF-N from the northern to the southern end of the plateau, a meridional shift of nearly 10 degrees of latitude. The separation of the front on the eastern side of the plateau, however, does not change with time: it appears to be locked to small spur extending eastward to the northeast of Kerguelen Island.
 A large-scale flow conserving its potential vorticity will be deflected equatorward upon encountering shallower bathymetry, as occurs at many locations in the ACC [e.g., SR2009]. By this argument, the PF-N would be expected to deflect to the north around the Kerguelen Plateau even if the upstream path of the front shifts to the south, until the upstream path shifts as far south as the latitude of the southern end of the plateau. This is qualitatively what we observe, although the front appears to shift to the southern route before the upstream path has been diverted quite as far south as the end of the plateau. A narrow ridge extending to the southwest near 54°S may deflect the PF-N north or south depending on the latitude at which the front is approaching the Kerguelen Plateau from the west.
 As discussed by SR2009, the frontal structure of the ACC in the vicinity of the Kerguelen Plateau has been a topic of considerable debate in the literature. In particular, authors have disagreed on the position of the PF. The presence of multiple branches of the PF (corresponding to “surface” and “subsurface” indicators, in the terminology of Sparrow et al. ) contributed to the confusion. While there is now consensus that the southern branch of the PF passes through the Fawn Trough, the deep passage between the northern and southern parts of the Kerguelen Plateau, some disagreement still remains on the position of the northern branch. Earlier studies generally placed the PF-N over the continental slope north of Kerguelen Island [e.g., Gamberoni et al., 1982; Orsi et al., 1995; Belkin and Gordon, 1996], based on the position of the northern limit of temperature minimum water less than 2°C near 200 m depth. Some more recent studies have concluded the PF-N crosses the shallow water of the plateau, passing just south of Kerguelen Island [e.g., Park et al., 1998, 2008; Moore et al., 1999]. This flow is shallow and carries only a small transport (5–6 Sv [Park et al., 2008]). Since the top-to-bottom transport in the PF-N upstream of the plateau is estimated to be about 22 Sv (calculated from the WOCE IO6 hydrographic section at 60°E), at least the deeper part of the PF-N, carrying most of the transport, must pass around either the northern or southern tip of the plateau.
 Our results suggest that in addition to the confusion resulting from multiple branches of the front and the use of different criteria to define the front position, the position of the PF-N has also changed over time. The limited hydrographic data available in this region is consistent with our conclusion that the position of the PF-N has changed. The earlier hydrographic sections used by Gamberoni et al. , Orsi et al. , and Belkin and Gordon  consistently show temperature minimum water (Tmin) cooler than 2°C over the continental slope north of the island. A CTD section in 2004 across the northern part of the Fawn Trough shows the Tmin cooler than 2°C only extending as far north as the continental slope on the southern end of the plateau (just south of Heard Island) (E. M. van Wijk et al., Regional circulation around Heard and McDonald Islands and through the Fawn Trough, central Kerguelen Plateau, submitted to Deep Sea Research Part I, 2009). A CTD transect collected by an elephant seal during 2004–2005 also shows the Tmin water cooler than 2°C ending near Heard Island, but with an isolated patch of water with a Tmin cooler than 2°C found further north over the plateau itself [Roquet et al., 2009].
 Changes in the mean position of the southern branch of the SACCF are reminiscent of those observed for the northern branch of the PF. It appears the front changed its position from passing north of the southern Kerguelen Plateau to passing south of the plateau, through the Princess Elizabeth Trough. However, the larger mapping errors associated with this front, owing to poor data coverage due to the presence of sea ice during much of the year, make this case less certain [SR2009].
3.6.2. Australian-Antarctic Basin
 Large changes in SSH are also observed downstream of the Kerguelen Plateau over the Australian-Antarctic Basin (Figure 7). The change in SSH is associated with a coherent southward shift of the PF and SACCF throughout the basin (Figure 13). The largest changes (more than 1 degree of latitude) occur in the position of the southern branch of the PF and the northern branch of the SACCF, whose paths are least influenced by bathymetry. However, coherent southward shifts are also observed where the PF and SACCF intersect the southeast Indian Ridge.
3.6.3. Drake Passage
 SSH also increased across the ACC to the east of the Drake Passage (Figure 7). Since the long-period changes in SSH observed here are smaller (less than 4 cm), than in vicinity of the Kerguelen Plateau and south of Australia and New Zealand, and the ACC frontal structure is strongly bathymetrically controlled by complex bottom topography in the Scotia Sea, the alterations in the ACC front pathways are not as dramatic as in the region of Kerguelen Plateau (Figure 14). The major changes in the ACC front pathways occur just east of South Georgia, where the SACCF and the southern branch of the PF shift to the south. The PF-S swaps from passing north of South Georgia prior to 1997 to passing the island to the south, reminiscent of the behavior of the northern branch near Kerguelen.
3.6.4. SSH Changes Between 35°E and 50°E
 While SSH increased over much of the Southern Ocean, SSH fell in the region between 30 and 50°E and 48–58°S during the 15 year period (Figure 15). This region corresponds to an area of high eddy energy downstream of the southwest Indian Ridge (SWIR). The middle and southern branches of the PF shifted to the north between the SWIR and the Del Caño Rise. The southern branch of the PF also shifted marginally to the north downstream of the SWIR.
3.7. Mesoscale Variability of the ACC Fronts: Eddy Cross-Frontal Exchange
SR2009 demonstrated that the mean pathways and mesoscale variability of the ACC fronts are strongly dependant on underlying large-scale bottom topography. The path of the ACC streamlines is often strongly deflected where the flow interacts with topography. Topographic interactions also affect the spread of the ACC fronts: the fronts tend to converge near large bathymetry (e.g., Crozet Plateau, downstream of the Kerguelen Plateau, near the Southeast Indian Ridge between 145°E and 175°E, and the Eltanin and Udsinov Fracture Zones at 215°E) and become more widely spaced over the abyssal plains (e.g., the Enderby, Australian Antarctic, and Southeast Pacific Basins).
 Evaluation of direct cross-frontal exchange across the ACC fronts is important in terms of both regional changes in the Southern Ocean overturning circulation and formation of water masses with anomalous water properties. In Figure 16 the frequency of occurrence of rings formed on the ACC fronts are shown, where rings are identified as closed contours of the SSH values corresponding to each front. For all ACC fronts the regions of the most intensive eddy shedding are observed downstream of large-scale topographic features. For example, most of the cross-frontal transfer by eddies across the southern ACC fronts (SB and SACCF) occurs in two main regions. One is located downstream of the Southwest Indian Ridge, where the Weddell Front and the southern ACC fronts converge and an intense mesoscale eddy field is observed between 15 and 30°E [see, e.g., Schröder and Fahrbach, 1999]. The other is situated downstream of the Kerguelen Plateau and occupies most of the Australian-Antarctic Basin. Strong eddy flux across the northern branch of the SACCF also occurs downstream of the Pacific–Antarctic Ridge over Amundsen Abyssal Plain, in the Drake Passage and downstream of the Islas Orcadas Rise east of Scotia Sea.
 Eddy water exchange across the three branches of the PF is similar to that observed across the SACCF-N (Figure 16). The major differences are related to the stronger eddy flux in the Pacific sector of the Southern Ocean, more intense eddy formation observed downstream of the SEIR and Macquarie Ridge between 150 and 170°E, and the region of mesoscale eddy generation in the western part of the Atlantic sector is shifted further north, downstream of South Georgia to the South Georgia Basin.
 Eddy-induced cross-frontal exchange in the dynamic height range spanning the SAF is particular strong across the northern branch of the front (Figures 16). Vigorous eddy formation and meandering reach a maximum east of the Greenwich meridian (0–30°E), south of Australia in the vicinity and east of the WOCE SR3 section (130–150°E), downstream of the SAF interaction with the Campbell Plateau and the East Pacific Ridge, and in the southeastern corner of the Pacific sector (Figure 16).
 The eddy-induced cross-frontal exchange is an important component in both the diapycnal mixing in the ocean interior, and in the overall overturning circulation of the Southern Ocean. The eddy formation on the ACC fronts typically occur downstream of the large-scale topographic features, and in some instances is highly asymmetrical (i.e., more cold or warm core rings are observed, and therefore a net eddy flux is directed equator- or poleward). In such regions formation of water masses with anomalous water properties is expected. The cross-frontal transfer of anomalous water masses through the formation of rings likely plays an important role in modifying water mass properties, mixing and sea-air exchange.
4. Summary and Discussion
 In Part 1 of this study, we showed that the ACC consisted of multiple fronts, each of which was consistently associated with a particular contour of SSH or approximate streamline. In Part 2 we have used maps of SSH to examine the variability of the ACC fronts between 1992 and 2007.
 First, we demonstrated that the SSH label associated with each frontal branch is nearly constant around the circumpolar belt. The front labels are also nearly constant in time: the bands of enhanced SSH gradient (i.e., fronts) occur along the same streamlines throughout the 15 year period of observations. Note that this is counter to what might be expected from a uniform warming of the ocean, which would increase SSH but not change the SSH gradient, resulting in a shift of each front to a higher value of SSH. Because there is a tight relationship between dynamic height and cumulative transport of the ACC (Figure 5), the fact that the fronts are observed to remain associated with the same SSH value means the transport is also constant with time. The SAF-N does not pass through Drake Passage, and instead supplies 25 Sv to the Pacific which supplies the inflow to the Indian Ocean via the Indonesian Throughflow and Tasman Outflow.
 Large changes in SSH have occurred in the Southern Ocean between 1992 and 2007 [Willis et al., 2004; Lombard et al., 2005; Morrow et al., 2008]. While the overall trend is an increase in sea level, there are strong regional differences. We showed that because the ACC fronts are robustly associated with particular SSH contours, the changes in SSH reflect shifts in the position of the ACC fronts. In the circumpolar average, each of the ACC fronts has shifted to the south by about 60 km. The value of the southward shift of the ACC is consistent with a simple estimate of by how much the ACC would shift due to observed SSH change:
where δς is observed mean increase in SSH in the Southern Ocean, 〈∂ς/∂ϕ〉ACC is mean SSH gradient across the ACC, and δφ is expected southward shift of the ACC.
 Much larger changes are observed in some locations, particularly where the fronts interact with topography. The northern branch of the PF near the Kerguelen Plateau is an extreme example. The sequence of SSH maps shows that the PF-N upstream of the plateau shifts to the south between 1992 and 2007. Prior to 2003, the front is deflected to the north once it encounters the shallower topography of the plateau, as expected for a flow conserving potential vorticity. After 2003, the PF-N has shifted far enough to the south that it can pass through the Fawn Trough. The time sequence of SSH suggests the PF-N followed a path around the northern end of Kerguelen Plateau between 1992 and 1997, passed through the Fawn Trough after 2003, and oscillated between the two paths between 1997 and 2003. As a result of the dynamical interaction between the front and the bathymetry, a relatively modest southward shift of the front upstream of the plateau results in a nearly 10° latitude shift in the front location near the plateau. SSH over the plateau increased by 8 cm between 1992 and 2007 as a consequence of the southward shift. If this increase in SSH is solely attributed to the warming of the mixed layer, it would require to increase the mixed layer temperature of 120 m depth by approximately 8 degrees. However, such warming over the Kerguelen Plateau is not observed. Given that the PF-N is associated with meridional gradients of temperature, salinity, nutrients and plankton communities, the southward shift of the front would be expected to cause warming and ecosystem changes but there is little data available to test this hypothesis. A similar shift of the southern branch of the PF has taken place near South Georgia.
 The fact that the fronts of the ACC are clearly steered by bathymetry has sometimes been interpreted to mean that the ACC fronts are locked in position and cannot shift. While it is true that sloping topography can inhibit frontal movements [e.g., Gordon et al., 1977; Gille, 1994; Moore et al., 1999; Sokolov and Rintoul, 2007; SR2009], sufficiently strong forcing can change the path of the fronts even in the vicinity of steep bathymetry.
 Our conclusion that the ACC fronts have shifted to the south is consistent with a number of previous regional or circumpolar average analyses [e.g., Aoki et al., 2003; Gille, 2002, 2008; Morrow et al., 2008; Sprintall, 2008; Böning et al., 2008]. In particular, Morrow et al.  attributed the increase in SSH in the Australian sector of the Southern Ocean to a southward shift of the ACC fronts. More generally, they demonstrated that the deep ocean made a significant contribution to thermosteric sea level changes observed in the Southern Ocean: the total sea level signal is at least 2–3 times larger than that calculated from thermal expansion above 700 m. The fronts of the Southern Ocean extend from the sea surface to the seafloor, so translations of the front would result in deep-reaching thermosteric anomalies, as observed.
 Model studies suggest that a southward shift in the westerly winds associated with the positive phase of the Southern Annular Mode (SAM) [Thompson and Wallace, 2000] drives a southward shift of the ACC [Hall and Visbeck, 2002; Sen Gupta and England, 2006]. Dong et al.  found that meridional shifts of the PF were weakly coherent with meridional shifts of the wind field in the zonal mean, with southward wind shifts leading southward front displacements. Sallée et al. , on the other hand, found that the response of the fronts to the major atmospheric modes of variability (SAM and ENSO) varied strongly by region. In the Indian sector, significant (at 95%) covariance between SAM and the latitude of the SAF and PF are limited to a narrow range of longitudes and to motions with frequency of less than 3 months. Likewise in the Pacific, movements of the SAF and PF are only significantly correlated with the SAM between 230°E and 260°E, for frequencies less than 3 months, and the sign of the correlation is opposite to that in the Indian sector (i.e., positive SAM associated with northward shift of the fronts). Sallée et al.  suggest the difference between the two basins can be explained by the position of the fronts relative to the intensifying wind stress. Low frequency (greater than 1 year) variations of both fronts in the Pacific are more highly correlated with ENSO than SAM. In both of these studies, the variance of front positions that can be explained by these modes of variability is relatively small, suggesting that the connection between wind forcing and front position is more complicated than can be captured by correlation with a simple index.
 The Southern Ocean has made a significant contribution to the overall heat storage by the ocean. More than 85% of the total change in heat content of the earth system has gone into warming the oceans [Levitus et al., 2005]. Of this, essentially all of the warming of the southern hemisphere oceans between 1955 and 2005 has occurred south of 30°S [Gille, 2008]. It is therefore important to understand the mechanisms responsible for changes in heat storage by the Southern Ocean. Our results suggest that the warming of the Southern Ocean, and the increase in sea level, is largely caused by a southward shift of the fronts of the ACC. The warming and sea level rise distributions are not uniform around the circumpolar belt as a result of both the spatial distribution of changes in wind forcing relative to the location of the fronts [e.g., Sallée et al., 2008] and to interactions between the fronts and bathymetry. In locations where the southern portion of the ACC approaches the Antarctic continent, like the western Antarctic Peninsula, the southward shift of the fronts may have contributed to the delivery of additional heat to the continental shelf and enhanced basal melt of floating ice [Shepherd et al., 2004; Rignot and Jacobs, 2002; Thoma et al., 2008].