The Institute of Atmospheric Physics (IAP) monthly global ocean temperature and salinity product (Cheng et al., 2017) at a 1°×1° horizontal resolution with 41 levels in 0–2 000 m for 1960–2021 is used to calculate the MLD and investigate its linear trends. The temperature at the surface level (1 m) is used as the sea surface temperature (SST). The ECMWF Ocean Reanalysis 5 (ORA5) dataset of monthly current velocities on a 0.25°×0.25° grid during 1960–2021 is employed to check the variation in ocean surface current. They are regridded onto the grid of 1°×1° to facilitate the comparison. To analyze the flux exchange between the ocean and atmosphere on the surface, the European Centre for Medium-range Weather Forecast (ECMWF) fifth generation atmospheric reanalysis (EAR5; Hersbach et al., 2020) monthly averaged data from 1960–2021 are used, including 10 m wind speed, 10 m horizontal wind, mean total precipitation rate, mean evaporation rate, mean surface net shortwave radiation flux, mean surface net longwave radiation flux, mean surface latent heat flux and mean surface sensible flux with a resolution of 1°×1°.

The MLD is computed based on the monthly mean temperature and salinity. The MLD is defined as the depth at which the potential density referenced to the surface exceeds the density of the water at 10 m by a threshold of 0.03 kg/m³. This criterion has been shown to robustly detect the depth of the mixed layer in various regions of the ocean (de Boyer Montégut et al., 2004; Sallée et al., 2021).

The linear long-term trend in each grid cell is calculated from the MLD anomalies using a least-squares fit. Student’s t test is used to examine the significance for variable trends.

To clarify the position variations of the ACC, the flow axis is analyzed. The ocean current axis is meridional meander where the maximum current velocity occurs, roughly a latitude circle around the Antarctic continent. In addition, the circle of maximum wind speed near the ocean surface can also reflect the change of the flow axis. Moreover, there are obvious reverse shear vortices on both flanks of the maximum wind speed line, so the meridional gradient of wind stress curl could also be a representation for the flow axis position. Therefore, the above three proxies are all selected to analyze the position variation of the flow axis for mutual verification. More specifically, the circumpolar contours for the maximum of wind speed, current speed and meridional gradient of wind stress vorticity from 40°S to 65°S are calculated to represent the climatic mean core position of the ACC. For moving tendency, all positive trends on each longitude from 40°S to 65°S are normalized to obtain the weight coefficients first, and the trends circumpolar of the ACC flow axis is represented by the corresponding weighted averaged position contour.

The annual mean and seasonal MLD time series in the ACC region are displayed in Fig. 1. The time evolution of the MLD exhibits pronounced interannual and decadal variabilities (Figs 1a, c and e). An obvious trend shift of the MLD can be found, with a period with decreasing trend and then a period with almost no trends. The transition between these two time periods occurred around the period from the 1970s to mid-1980s.

To determine the turnaround year of the trend shift more accurately, a quantitative analysis was conducted using the annual mean result. The cutoff point was tested by each year within the period during 1970 to 1985, and then the trend between 1960 and that cutoff year were calculated and evaluated. Results indicate that when cutoff year moves to 1985, the downward trend of the MLD is −0.18 m/a, and this trend exceeds the 99% confidence level for the first time. While in the subsequent period after 1985, the trend is quite weak, with a value of −0.005 m/a.

The MLD trends in austral winter (August–October) and summer (January–March) exhibit different patterns (Figs 1c, e). In austral winter, there is a shallowing trend followed by a weak variation in the later period, which is similar to the annual mean result. On the other hand, the MLD trend in austral summer shows an upward trend of 0.03 m/a in the second phase, which is inconsistent with that in austral winter. The magnitude of MLD variation is much greater in winter than that in summer. The climatic mean MLD in ACC region are around 63.7 m, 18.9 m and 38.3 m for winter, summer and annual mean results, respectively. Moreover, the annual mean MLD variation is almost entirely dominated by the temporal changes of the austral winter MLD, and the time correlation coefficient between them is as high as 0.91.

The variation of SST is gentler than MLD. The climatic mean SST in ACC region are around 4.5℃, 6.5℃ and 5.4℃ for winter, summer and annual mean results. Unlike the MLD trend in the ACC region, SST variation maintained an upward trend whether before or after the mid-1980s (Figs 1b, d and f). The trend difference of SST between the two periods is not significant. For the annual mean result, the SST trends in the two periods are 0.01℃/a and 0.006℃/a. While for the austral winter (summer) results, the trends are 0.009℃/a (0.01℃/a) and 0.005℃/a (0.007℃/a), respectively.

In principle, mixed-layer shoaling on a large spatial scale is well understood under a global warming background due to intensified stratification (Sallée et al., 2021). However, the unexpected MLD trend in the ACC region is more complex, which needs more investigation. Due to the dominating role of winter MLD for the annual mean results, the following analyses are focused on the annual mean and winter MLD, as well as the contributing factors in both two periods.

Since MLD showed different trends before and after the mid-1980s, the spatial patterns of the MLD trend are analyzed in the following two periods: 1960–1985 and 1986–2021. Figures 2a–c show the magnitudes of spatial long-term linear annual mean MLD changes for the ACC region during the above two periods and their differences, with field mean of −0.180 m/a, −0.002 m/a and 0.178 m/a. In the former period, the MLD was shallowing in almost all area of the ACC region, except for the areas south to Australia and in areas of the central Pacific (45°–60°S, 180°–120°W). The maximum shallowing trend is −8.22 m/a around the area off the northernmost point of the Antarctic Peninsula. In the latter period, the shallowing trend almost diminished and deepening MLD emerged in the ACC region (Fig. 2b). The maximum deepening rate in the ACC region is 2.37 m/a around in the area north to the Wilkes Land. The annual mean MLD climatology during the two periods differs little from each other, with a eastward shift for the MLD high value centers (about 150 m) both in South Indian (SI) and South Pacific (SP) sectors. Figure 2c demonstrates that the MLD trends sharply shifted around the mid-1980s, with the most concentrated positive trend differences distributed in the western area in the Drake Passage, the areas south to Australia, and the mid-latitudes in the central parts of the South Indian Ocean (SIO). The maximum deepening trends corresponding to the above centers are 5.86 m/a, 6.49 m/a and 5.65 m/a, respectively. The SST patterns are displayed in Figs 2d–f, and the trend field mean for the two periods and their difference are 0.010℃/a, 0.006℃/a and −0.004℃/a, respectively. During the period of 1960–1985, strong warming trends occupied most areas in the ACC region, with high values concentrated in the central SIO (35°–50°S, 60°–120°E). However, these warming trends decayed, and scattered cooling trends appeared in the Subantarctic region from 1986 to 2021. The trend shift of SST shown in Fig. 2f mainly demonstrates three relatively-weakening warming centers, which corresponds well with the deepening trend centers of MLD in Fig. 2c.

Winter mean MLD and SST trend patterns are shown in Figs 2g−l. The variation patterns of winter MLD in both two periods are quite similar to the annual mean results, but with significantly stronger magnitudes. The climatic MLD peak values are about 300 m, almost twice that of annual mean results. The trend field mean of the two periods and trend shift are of −0.408 m/a, −0.081 m/a and 0.327 m/a. In the three key regions mentioned earlier, the positive winter MLD trends reach 14.83 m/a, 13.31 m/a and 16.56 m/a, respectively. The spatial distribution of winter SST trend patterns also closely resembles the annual mean SST trend patterns. However, unlike the increased amplitudes in winter MLD, the magnitudes variations of SST and its trends between the winter and annual mean results are relatively small. The SST trend field mean in winter are 0.009℃/a, 0.005℃/a and −0.004℃/a, almost same as the corresponding annual mean value. In the three relatively-weakening warming regions, the SST trends difference in winter exhibit extreme values of −0.07℃/a, −0.07℃/a and −0.12℃/a, respectively.

Cause analysis of the evolution mechanism of the MLD trend shift in the ACC region was carried out here to explore the possible drivers, by analyzing wind speed, ocean current, wind stress vorticity, net heat flux and freshwater flux.

The MLD variability may be affected by the wind field. Wind stirring can generate mechanical mixing, wave breaking, internal waves and other submesoscale instabilities in the ocean mixed layer, which can contribute to vertical convection (Huang et al., 2012; Barkan et al., 2017; Buckingham et al., 2019). Moreover, vertical mixing could also be affected by the ocean current and wind stress vorticity, which are both closely associated with winds through Ekman pumping and Ekman transport (Huang and Qiu, 1998).

Figure 3 displays the corresponding trends for the wind field, ocean current and wind stress vorticity. The ACC region is dominated by intense westerly winds almost all year round. Many studies have focused on the association between wind momentum and physical processes in the upper ocean (e.g., Russell et al., 2006; Toggweiler and Russell, 2008; Feng et al., 2020). The wind speed during 1960–1985 showed a significant intensification trend in almost all area of the Southern Ocean, and the strongest trend reached as high as 0.053 m/(s∙a). However, after the mid-1980s, the strengthening winds mainly shrank to the Pacific sector and the Indian sector. Although the wind speed in the ACC region increased throughout the period after 1960 (Yang et al., 2007; Liu and Wu, 2012), the trend in the later period was significantly weaker, especially in the South Atlantic (SA) and SI sector, which can be well reflected in the trend difference shown in Fig. 3c. The strongest value of the wind speed reduction rate was −0.054 m/(s∙a) in the central SI region.

The trends of surface ocean currents are quite different from the wind fields mentioned above (Figs 3d–f). In the former period, the weakening trends of surface ocean currents occupied most regions, while in the latter period, negative trends decayed, and positive trends appeared in more areas. Therefore, the dominant trend shifts for surface ocean currents are positive, especially in the SP sector (Fig. 3f). The trends of surface ocean currents do not keep pace with those of winds, which is consistent with the conclusion that the ACC resists wind-driven acceleration in both observations and model simulations (Munday et al., 2013; Böning et al., 2008).

The wind stress curl trends are displayed in Figs 3g–i. In the ACC region, the trends of the wind stress curl are distributed surrounding Antarctica, roughly along the flanks of the ACC. In both periods, negative values of wind stress vorticity trends were located closer to Antarctica, while the positive values expanded equatorward in all ocean sectors. The differences in wind stress curl trends (Fig. 3i) show trend differences with maximum and minimum trends of 6.4 × 10⁻⁹ N/(m³∙a) and −11.9 × 10⁻⁹ N/(m³∙a) respectively.

To further explore the possible influence of dynamic fields on the MLD trend shift, the flow axis of the ACC is analyzed. The maximum wind speed circumpolar contour in 40°–65°S is selected to be a proxy for the climatic mean position of the ACC. In addition, the weighted average position of the positive trends on each longitude is chosen here to represent the moving trend for the maximum wind speed circle (Figs 3a–c). The meridional location of surface wind did have an interdecadal shift in the two periods, especially in the SI and SP sectors. Moreover, this shift is not symmetrical and does not contract at all longitudes, but it shows a polarward shrinking trend in SI and an equatorial expansion trend in SP. That is, the center of the maximum wind speed circle tends to move toward the Marie Byrd Land direction. The locations of surface ocean current speed (Figs 3d–f) are quite similar to those derived from trends of wind speed. In the analysis of the wind stress curl, the maximum meridional gradient of the wind stress curl trend is used to compare the core meanders of the ACC. In general, there are opposite vorticities on the southern and northern flanks in the ACC region. Therefore, it can be deduced that the location with the maximum meridional gradient of vorticity is just the place where the flow axis exits. Consistent with previous conclusions derived from wind speed and current, the circumpolar contour of the maximum meridional gradient of wind stress vorticity also exhibits a concurrent shift toward the Marie Byrd Land direction.

Figure 4 depicts the winter trend spatial patterns of wind speed, ocean current, and wind stress vorticity. The spatial trend patterns of wind speed in winter are generally similar to the annual mean results (Figs 3a–c). The main differences are as follows. In the former period, the increasing trend of wind speed over the area southeast to New Zealand shifted to a decreasing trend; in the latter period, the increasing trend of wind speed in this region became more significant, which led to a severe trend shift between the two phases during winter (Fig. 4c). There is a band-like area (43°–52°S, 180°–70°W) with wind speed increasing trend with a mean amplitude of 0.06 m/(s∙a). Negative trend shift is also evident in the SA and SI regions. Regarding the maximum wind speed axis, there was a notable moving tendency towards the SP sector direction in the second phase, exhibiting a greater fluctuation compared to the annual mean results (Fig. 3c). The trend variations of surface ocean current are highly stable, with little difference in intensity and spatial distribution between winter and the annual mean results. The changes of the flow axis align with the movement of the annual mean as well (Figs 4d–f). Comparing the trend distribution of wind stress curl during winter with the annual mean results, the patterns are similar, but the magnitudes in winter are significantly stronger in all ocean basins. The most significant changes of winter trends for wind stress curl occurred in the SP sector in both two periods. For the position of the flow axis, although the vorticity trends on both flanks all increase, the position of the maximum curl gradient shows little change, exhibiting a shift to low latitudes in the SP sector (Figs 4g–i). In general, the trend patterns of dynamic variables in winter are consistent with the annual mean results, but with larger variations which primarily occurs in the SP sector.

Figure 5 displays the zonal mean profiles of wind speed, ocean current, wind stress vorticity and MLD in the two periods in both the SP (40°–65°S, 180°–70°W) and SI (40°–65°S, 0°–90°E) sectors. The solid and dashed lines represent the annual mean and winter results respectively. In the SP sector (Figs 5a–d), the annual (winter) mean climatological zonal mean wind speed peaks at approximately 55°S (58°S). From 1960 to 1985, the wind speed increasing trend was most obvious in the region south of 60°S. From 1986 to 2021, the annual (winter) mean strengthening trends of wind speed reached their apex at approximately 51°S (47°S), which apparently resulted in a shift toward the equatorial direction for the wind speed extreme. In terms of ocean currents, the annual mean and winter results have little difference. There was a slight accelerating trend in the region south of 60°S in the former period, and the positive trends moved to latitudes between 45°S and 55°S in the latter period, indicating an equatorial migration. Moreover, the positive to negative transitions of wind stress vorticity also support the same result. The climatological transition of annual mean (winter) wind stress vorticity occurs at 52°S (56°S). The sign of wind stress vorticity trends changed from positive to negative at approximately 60°S during the former period. However, this change arose at lower latitudes of approximately 52°S for annual mean results and 48°S for winter mean in the subsequent period. All these variations indicate that the ACC meanders in the SP sector have shifted toward the equator.

The mitigation in the SI sector is different from that in the SP sector (Figs 5e–h). The climatological zonal mean wind peak is located at approximately 50°S in the SI sector, and the accelerating trend maximum of wind shifts from 52°S to 54°S for annual mean results (54°S to 59°S for winter mean), manifesting a slight polarward shift. In terms of ocean currents, the climatic peak value appeared at 48°S, and there was a slight weakening trend along almost all latitudes from 1960 to 1985. However, strengthening trends replaced the weakening trends south of 46°S from 1986 to 2021, which displayed a poleward mitigation trend for the maximum current contour. Analyzing from the turning point of the wind stress vorticity from positive to negative, the climatological transition is shown at approximately 50°S. In the first period, the transition occurred at 50°S for annual mean result (53°S for winter mean). However, the watershed moved significantly southward to higher latitudes of approximately 54°S (61°S for winter mean) in the second period. The zonal mean profiles of wind speed, ocean current and wind stress curl all indicated a polar shift of the meridional meanders of the ACC in the SI sector.

Taken together and further confirmed by the conclusions drawn in Figs 3 and 4, there was a drift in the flow axis of the ACC, as well as the wind speed and wind stress vorticity field. The SP sector experienced expansion in the equatorial direction, while the SI sector experienced polarward compression in the meridional direction. The center for the axis of the entire circumpolar currents moved toward the Marie Byrd Land direction of the Antarctic.

Up to this point, it is determined that although the trend distributions in the three dynamic fields have their own characteristics, a significant commonality is shared by all of them, which is that all three fields reflect the interdecadal shift of the meridional meanders of the ACC that occurred during the two periods. Therefore, how is this decadal shift related to the variation in MLD in the ACC region? Previous studies have shown that mixing along isopycnals is suppressed by one to two orders of magnitude in the core of the ACC jets, where the flow is sufficiently strong to carry tracers downstream before eddies have a change to mix cross-stream (Ferrari and Nikurashin, 2010). On the other hand, widespread and strong shears exist on the flanks of the ACC, which support the generation and development of mixing processes (Rintoul, 2018). The flow axis of the ACC tended to drift toward the equator in the SP sector; therefore, the original inhibited mixing activities began to develop under the newly formed shear after the flow core moved away (approximately 54°–65°S in Fig. 5d), resulting in a deepening trend for MLD in the ACC region toward the Antarctic. In the SI sector, there was a drifting trend toward the Antarctic for the flow axis of the ACC, and the position of the original core turned into the flank of the ACC, favoring mixing processes there (approximately 42°–56°S in Fig. 5h). As a result, the MLD tended to deepen on the equatorward side of the ACC flow axis. All these findings are in good agreement with the conclusions shown in Figs 2c and i.

To further investigate the influence of flow axis variations on the MLD trend changes, two key regions (green boxes in Fig. 2i) with pronounced flow axis oscillation are selected: the southwestern region of Drake Strait (60°–68°S, 76°–150°W, K1 hereafter) and the central South Indian Ocean region (40°–50°S, 50°–90°E, K2 hereafter). It is the trend variation rather than interannual change that the study focuses on. Therefore, an 11-year running mean are performed to the time series. The standardized time series are shown in Fig. 6. The MLD and related variable changes in the K1 region for annual mean results (Fig. 6a) shows the opposite phases between MLD and the three dynamic variables, along with the shallowing followed by deepening trends of MLD, the wind speed, ocean current and wind stress vorticity gradient all display increasing trends followed by decreasing trends. Their correlation coefficients with MLD are −0.28, −0.60, and −0.56, respectively. This indicates that the deviation of flow axis from K1 region are favorably to MLD deepening here. Similar negative relationship is derived from the results in K2 region (Fig. 6b), where the negative correlations of wind speed, ocean current, and wind stress curl gradients with MLD are −0.79, −0.30, and −0.42, respectively. The winter mean results are shown in Figs 6c and d. The inverse changes of these time series well reveal the inhibition effect of the flow axis on MLD deepening, as well as the promotion effect of the flow axis departure on MLD development.

The contributions of the heat flux and freshwater in the ACC region were also checked (Fig. 7). In climatology, the ocean loses heat to the atmosphere in the ACC region. In the former period, there were significant decreasing trends for the heat flux in the central Pacific sector and the areas south to Australia. This means that the ocean loses more heat in these areas, which can contribute to the weakening stratification and MLD deepening (Fig. 7a). In addition, around the northern region of the Wilkes Land and the Ross Sea, there were small ranges of positive heat flux trends. In the latter period, these trends were both diminished. Compared with the MLD trend shift (Fig. 2c), these significant deepening trend regions could not correspond well with the variations in heat flux, where the trend shifts were almost positive, representing a heating effect that was not conducive to the MLD deepening here. This indicates that the heat flux trend variation is not the dominant factor driving the deepening MLD trends in the ACC region, which is consistent with the conclusion that the upper ocean stability is almost established by the vertical salinity gradient in the high latitudes due to the cold climate conditions there (Sallée et al., 2021). The freshwater flux in 1960–1985 mainly showed increasing trends in the high latitudes around Antarctica, while in 1986–2021, the increasing trends almost disappeared (Figs 7d and e). The saltier trend shift occupied most areas in the ACC region, and a freshening trend shift emerged in the western Pacific and a zonal band between 30°–45°S in the Indian Ocean (Fig. 7f). Compared with the MLD trends in Fig. 2c, the area north to Victoria Land and Wilkes Land (45°–60°S, 120°E–180°) with saltier trend shifts at maximum magnitudes of −7.08×10⁻⁷ kg/(m²∙s∙a) and −4.14×10⁻⁷ kg/(m²∙s∙a) are in good agreement with the deepening MLD there, where the deepening trend of MLD cannot be explained by the dynamic fields due to the absence of axis shift of the ACC.

In summary, the interdecadal shift in MLD trends in the ACC region could be primarily attributed to the meridional shift in the flow axis of the ACC between the two periods, especially for the deepening trends in the SI and SP sectors mentioned above. Meanwhile, the saltier trends of freshwater flux north to Victoria Land and Wilkes Land match well with the strengthening mixing processes there.