The Global Ocean Data Analysis Project (GLODAP, v2.2023) is a synthesis activity for ocean surface to bottom data collected through analysis of water samples, and it contains data from 1 108 scientific cruises, including multiple repeated meridional cruises near the subpolar Southern Ocean (Olsen et al., 2016; Key et al., 2015). The majority of the sections of the dataset have undergone rigorous secondary quality control to ensure reliability and internal consistency, and including the systematic bias assessment (Olsen et al., 2016). Furthermore, in this paper, we reference and discuss Argo data for 2013 to 2022 as a supplement. These data were collected and made freely available by the International Argo Program and the national programs that contribute to it (https://argo.ucsd.edu, https://www.ocean-ops.org). The Argo Program is part of the Global Ocean Observing System. The floating depth of Argo floats can reach up to 2 000 × 10⁴ Pa, and the average observation period is approximately 10 d (Jayne et al., 2017). Given that over 90% of the time, the Argo floats drift in accordance with the ocean currents at the designated depth, they serve as effective tools for Lagrangian observations of oceanic currents (Wong et al., 2020). The expocode of scientific cruises and the float unique identifiers of the Argo floats used in this paper are presented in Table 1.

Ocean Re-Analysis System 5 (ORAS5) is a global eddy-permitting, gridded monthly mean ocean-sea ice ensemble reanalysis system prepared by the European Centre for Medium-Range Weather Forecasts (ECMWF) (Zuo et al., 2019). This dataset, which is based on the NEMO model, incorporates subsurface temperature, salinity, sea ice intensity and sea level anomaly through the Nucleus for European Modelling of the Ocean VARiational (NEMOVAR) assimilation system, and it is forced by ECMWF Re-Analysis-Interim (ERA-Interim) atmospheric reanalysis data. In this study, the monthly temperature, salinity, current and surface flux fields were used to examine the changes in the ASC since 1993 to 2022, with a spatial resolution of 0.25° × 0.25° (equivalent to approximately 9.5 km at 70°S). The three-dimensional variables cover the entire water column from the sea surface to a depth of 5.9 km, with a total of 75 horizontal layers with unequal intervals.

The majority of the reanalyzed atmospheric data effectively captures the near-surface wind speed over the Antarctic margin and mainland. Among these, the fifth-generation ECMWF reanalysis of the global climate ECMWF Re-Analysis version 5 (ERA5) stands out due to its ability to describe the monthly averaged wind speed magnitude and the interannual variability of the near-surface wind speed (Dong et al., 2020). To assess the changes in the wind field over the subpolar Southern Ocean, this study utilizes ERA5’s monthly averaged data on single levels (Hersbach et al., 2020). The variables selected are the eastward and northward components of the 10 m wind, with a spatial resolution of 0.25° × 0.25° covering a duration of 360 months from 1993 to 2022. It is important to highlight that the atmospheric variables utilized in this study differ from the forcing fields (ERA-Interim) of reanalysis datasets. The ERA-Interim data are included as a reference to ensure that these differences do not introduce errors into our analysis. Given that ECMWF has ceased the interactive download service for ERA-Interim, we sourced the monthly mean 10 m zonal wind speed for the period 1993 to 2017 from the Oceanographic Data Center, Chinese Academy of Sciences.

The sea-ice concentration (SIC) data selected in this study were obtained from the National Snow and Ice Data Center (NSIDC), and the monthly mean National Oceanic and Atmospheric Administration (NOAA)/NSIDC sea-ice concentration climate data record (CDR) data are utilized (Comiso and Nishio, 2008). The spatial resolution of the data is 25 km × 25 km. To facilitate comparison, a linear interpolation method was employed to convert the data into an Ocean Reconfigurable Computational Architecture 0.25-degree (ORCA025) mesh (0.25° horizontal resolution), which aligns with the ORAS5 in the horizontal plane. The time span is from January 1993 to December 2022. This dataset exhibits minimal deviation from the observed data for Antarctica when compared to other sea ice intensity datasets (Kern et al., 2019).

While the zonal component of the velocity effectively represents the ASC in most regions, this zonal approximation is more suitable for relatively sparse observations. For the gridded reanalysis data, the meridional and zonal components of the velocity (represented by u, v) are usually transformed into the along-slope and across-slope velocity components and use the along-slope component of the velocity is used to more accurately characterize the flow of the ASC (Stewart et al., 2019; Huneke et al., 2022; Yang et al., 2024). The along-slope component of velocity (${u_{{\mathrm{along}}}} $) is derived from the projection of the tangent to the velocity of climatological state of the ASC within the Cartesian coordinate system. $\mathcal{S}_{ \mathrm{ASC}}$ represents the climate trajectory curve of the ASC.

We assumed that the Argo floats maintain a constant velocity in a straight trajectory along the horizontal plane at their working depths during the intervals between their ascent to the surface for signal transmission, and that the sole driving force behind the horizontal displacement is ocean currents (Park et al., 2005). Based on the Lagrangian measurements, in order to conduct a simple estimation of the flow velocity of the Argo floats drifting (${u_{{\mathrm{argo}}}} $), we assumed that the flow velocity is equal to the drifting displacement (ΔL) is divided by the time interval (Δt):

A narrow belt of easterly winds is set over the Antarctic continental margin in order to emphasize the effect of the wind on the changes of the ASC. In the study area, the local surface wind is defined as the wind at a height of 10 m within a latitude range of 2° north and south of the Antarctic slope. Characterized by the contour representing the minimum wind speed, the wind field in the subpolar Southern Ocean is classified into northern westerlies and southern polar easterlies. However, these contours do not consistently lie on the northern side of the continental slope and intersect land at the Antarctic Peninsula (Neme et al., 2022). Consequently, to concentrate on the effect of the local surface winds on the ASC, the northern boundary for local wind is established at 2° north of the shelf break, which coinciding closely with the contours representing minimum wind speed in the climatological state. Meanwhile, the southern boundary is set at 2° south of the continental shelf break, approximately coincides the coastline. In this paper, the eastward winds are defined as negative values of zonal winds in accordance with established conventions.

The freshwater and heat fluxes are combined to obtain the net surface buoyancy flux:where F_h and F_fw denote the net surface heat flux and surface freshwater flux, g is the acceleration due to gravity (9.8 m/s²), $ \mathrm{\rho } $ is the density of seawater (1.027 × 10³ kg/m³), $ \mathrm{\alpha } $ is the coefficient of thermal expansion (3.529 × 10^–5 K^–1), C_p is the specific heat capacity of seawater [3.89 × 10³ (kg·J)/m³], $ \mathrm{\beta } $ is the shrinkage coefficient of brine (7.846 × 10^–4 kg/g), and S₀ is the sea surface salinity from the ORAS5 reanalysis data.

In this paper, we used multiple linear regression analysis to discuss the quantitative relationship between buoyancy flux, wind, and ASC. Pearson correlation is used to estimate the degree of similarity between the time series. In this study, the multi-year monthly trend and the long-term trends are all classified as the long-term changes.

Figure 1a illustrates the average zonal velocity in the subpolar Southern Ocean during 1993–2022, as well as the climatological state location of the ASC. The solid line delineates the ASC region of concern in this paper. The four green solid lines labeled with Roman numerals indicate the locations of the GLODAP scientific cruise sections, and the three black fan-shaped boxes denote the tracks of Argo floats in Figs 1a and b.

Both the observations and the reanalysis data indicate an acceleration of the ASC. Due to the proximity of the ASC to the continental shelf, the complex terrain and sea ice cause all of the flow axes from the scientific cruise sections to be predominantly located at or slightly beyond the southernmost part of each section (Fig. 2a). These zonal component of the baroclinic geostrophic currents is computed using the thermohaline. The velocity of the ASC increases from east to west, reaching its maximum velocity at 0° (the Weddell Sea) and minimum velocity at 140°E (the Dumont d’Urville Sea) (Table 2). While the velocity variation characteristics near the continental slope differ across the sections, the trend of westward acceleration is universal except for at 140°E (Fig. 2a). The reanalysis data for the aforementioned sections, demonstrate that the acceleration occurred during 1993 to 2022 in all of these sections. The strongest accelerating occurred in the South Atlantic Ocean, and the trend in the southern Indian Ocean is relatively weak (Fig. 2b). In conclusion, the observation and reanalysis data indicate that in most regions of the east Antarctic margin, the ASC exhibited a strengthening trend. However, it should be noted that because the observations are relatively sparse, the analysis of the long-term trends may be disturbed by seasonal changes, and exist errors due to sensitivity to the length of data.

The ASC is accelerating in all seasons, particularly in austral autumn. Figure 3 provides a more detailed representation of the ASC velocity from the ORAS5 reanalysis data. During 1993 to 2022, the average flow velocity in the mixed layer in the ASC is measured at –7 cm/s, with a negative trend of –0.04 (cm·s^–1)/a (passing the 95% confidence test). Except for the Southeast Indian Ocean sector, the velocity trend of the ASC is almost consistent above 2 000 m (Fig. 3a). The trend exhibits strong baroclinicity vertically, and the contribution of the geostrophic component explains about 51% of the overall trend in the mixed layer (Fig. 3b). The average flow velocity in the mixed layer in the ASC accelerate in most months, and the most significant acceleration occurs in austral autumn (March–May) which is also the strongest season in the ASC (Fig. 3c). The average flow velocity in the upper 2 000 m is –4 cm/s, and there is also a weak accelerated trend [–0.02 (cm·s^–1)/a, passing the 95% confidence test]. At the depth of 2 000 m, the seasonal cycle and the multi-year monthly trend of ASC are similar to those in the mixed layer; however, the trend of velocity passing the 95% confidence test is only occurred during autumn (not be shown). The significant accelerating in autumn increases the seasonal range of ASC. The black solid line in Fig. 3d represents the seasonal variations in the ASC flow velocity, which is calculated as the absolute value of the difference between the strongest (autumn) and the weakest (spring) seasons, indicating a strengthening trend [0.03 (cm·s^–1)/a, passing the 95% confidence test]. The event that deviates from the mean by one standard deviation (gray shaded area represents the one standard deviation from the mean seasonal range on left axis) in the flow velocity time-series is classified as an anomaly of the seasonal range (Fig. 3d). The red (blue) solid circles indicate abnormal events that are above (below) the mean, and their corresponding years are referred to as strong (weak) flow years. After 2017, the strong flow years have occurred frequently; while the weak flow years mainly occurred before 2003, with only one occurrence in 2013. Furthermore, compared with the acceleration of the annual, the additional acceleration in autumn is mainly manifested as significant barotropic feature (Fig. 3b).

The occurrence of strong flow is advancing earlier. The maximum monthly mean velocity of the ASC occurs in May, with the flow velocity can reach –8 cm/s in the mixed layer, while the minimum monthly mean velocity occurs in December and is –5 cm/s (Fig. 3c). In this paper, the maximum ASC velocity in the seasonal cycle is called strong flow, while the minimum velocity is called weak flow. The ASC not only exhibits a significant accelerating trend in autumn but the month of strong flow occurrence advances at a rate of 1 month per decade (passing the 95% confidence test). Between 1993 and 2001, the median month for the occurrence of maximum velocity was July. From 2014 to 2022, the median occurrence time of the strong flow has shifted to May, advancing by approximately two months (Fig. 4). Simultaneously, there is virtually no discernible changes in the advance or lag of the month in which the ASC occurs its weak flow velocity (Fig. 4), as well as statistically significant trend.

Furthermore, as a supplement to the scientific cruises, in this study, we utilized data for the Argo floats that measured repeated sections in the subpolar Southern Ocean between 2013 to 2022 (Fig. 1b). According to their drifting tracks, the Lagrange method can be used to calculate their drifting velocities, which are considered to approximate the flow of the ASC during the observation period of the floats. Compared with the other two datasets, the velocities measured by the Argo floats are found to be slower, and the highest velocity observed in Prydz Bay instead of the Weddell Sea (Table 3). In addition, the Argo floats also observed the velocity of ASC exhibited an accelerating towards the west.

The acceleration of ASC could be attributed to the effects of the local surface wind enhancement and the positive trend of net buoyancy flux. Due to the long-term change of ASC flow velocity is most significant in autumn. Consequently, this study focuses on comparing the differences in wind and buoyancy fluxes in autumn between the strong and weak flow years. There is a significant correlation between the velocity of ASC in autumn and the local surface wind speed as well as the heat flux (Fig. 5a). Through composite analysis, it is found that the heat flux (local surface wind) exhibits a significant positive anomaly (negative anomaly) in the years when the seasonal range exceeds one standard deviation from the mean. Conversely, these features will reverse when the seasonal variation falls below the mean (Figs 5b and c). By comparing the local surface wind and heat flux during autumn under these two types of seasonal range anomalies, it is found that the Weddell Sea, the Prydz Bay and the Dumont d’Urville Sea are the key areas where the heat flux is most significant, while the Weddell Sea, the Cosmonaut Sea and the Southern Indian Ocean sector are the key areas of the local surface wind.

Local surface wind is anticipated to exhibit a weak enhancement during austral autumn, with the area of enhancement showing high spatial consistency with key areas identified through the corresponding composite analysis. Figure 6a illustrates the specific spatial range of the local surface wind that effects on the ASC. There are spatially heterogeneous variations in the local surface wind near the shelf break, with the weak enhancement in easterly winds over the Atlantic sector and a decrease over the Indian Ocean sector (Fig. 6b). Focusing on autumn specifically, the local surface wind intensifies from the Weddell Sea to the Cosmonaut Sea west of 40°E, with a weak strengthening trend also observed in the Dumont d’Urville Sea (Fig. 6c). Additionally, the enhancement of the easterly wind over the Weddell Sea may persist from spring to autumn, while the most significant enhancement of the easterly wind on the Prydz Bay mainly occurs in winter (not be shown). The results of ERA-Interim support the ERA5-based findings regarding the easterly wind trends, despite the minor differences in data length (ERA-Interim covers only 1993 to 2017), suggesting that the influence of the local surface winds on the ASC is genuine instead of model errors. Enhanced local surface wind not only accelerate ASC by directly inputting momentum into the ocean, but also contribute to that by facilitating the baroclinic adjustment process in the ocean via Ekman transport.

In addition to changes in the local surface wind, the wind stress curl field along the Antarctic continental margin in the southeastern Indian Ocean sector could generate significant positive anomalies during the positive phase of the SAM. These anomalies also reduce the meridional sea surface height gradient near the slope (Armitage et al., 2018; Auger et al., 2022), to suppress the acceleration of the geostrophic circulation of the ASC, thereby influencing the three-year significant period in the interannual variability of ASC velocity (not be shown).

The buoyancy flux exhibits an overall positive trend in the subpolar Southern Ocean, with the areas displaying the positive trend of heat flux showing a high spatial consistency with key areas identified through the corresponding composite analysis. The previous section discussed the significant baroclinic characteristics of the acceleration in the ASC, and the geostrophic component plays a significant role in it (Fig. 3). Therefore, the change of buoyancy flux is concerned in this study as it significantly influences the density field variations near the slope. In the subpolar Southern Ocean, the buoyancy flux anomaly is generally influenced by the contribution of the freshwater flux and exhibits an overall positive trend (Fig. 7a). In the austral autumn, when the acceleration of the ASC reaches its peak, the contribution of the heat flux becomes particularly pronounced and dominates the spatial variations in the buoyancy flux near the shelf break. Figure 7b shows the trends of the buoyancy flux, freshwater flux, and heat flux in autumn within the same spatial range. The spatial distribution of the trend of the freshwater flux near the shelf break is relatively consistent, whereas the heat flux exhibits a significant positive trend in the eastern Weddell Sea, the Cosmonaut Sea, and the Dumont d’Urville Sea, which are same as the key areas in Fig. 5b. The spatial difference characteristics of the net surface buoyancy flux trend near the continental slope is dominated by the heat flux in autumn, which modulates the spatial characteristics of the ASC acceleration.

The buoyancy flux induces baroclinic adjustments through the introduction of low-density anomalies into the ocean, dominating the acceleration of the ASC and advancing the strong flow. The season variations in the buoyancy flux near the shelf break within a range of 2° to the north and south from 1993 to 2022 are depicted in Fig. 8a. In terms of the seasonal cycle, there are two peaks in the buoyancy flux in December and March. The strongest peak in the buoyancy flux occurs in December. This is closely associated with the melting of the sea ice in summer and primarily is driven by the freshwater flux. In contrast, the peak in March, which is dominated by the heat flux, is a sub-peak within the seasonal cycle and mainly resulting from the expansion of the open ocean following the melting of the sea ice in summer, which promotes the absorption of heat from the atmosphere by the ocean. Figure 8b illustrates the monthly trends of the buoyancy flux. There are two evident peaks in summer (January) and autumn (April), and the peak value in summer almost equal to that in autumn, about 1.3 (W·m^–2)/a (passing the 95% confidence test). The trend of the heat flux is positive in summer and autumn, and negative in winter and spring. This seasonal reversal of the trend of the heat flux further amplifies the positive trend of the buoyancy flux in summer and autumn. Although the net buoyancy flux near the ASC is primarily influenced by the freshwater flux, it is essential to acknowledge the significant contribution of the heat flux in autumn (Figs 7, 8a and b), particularly pronounced in shaping its spatial characteristics. The buoyancy flux in the key region shown in Fig. 5c exhibits the strongest correlation (0.50) with the ASC velocity lagged by one month (synchronous correlation coefficient is 0.44). Additionally, the occurrence of the maximum acceleration of the ASC in the seasonal cycle also lags behind the peak lightening of the density induced by buoyancy flux variations during the seasonal cycle. Dynamically, the positive trend of the buoyancy flux has introduced more low-density anomalies into the ocean, which accumulate onshore under the influence of prevailing easterly winds over the continental slope, ultimately resulting in a significant lightening of the potential density in the mixed layer south of the slope (Fig. 8c). And these processes steepen the isopycnal surface near the slope leading to the baroclinic adjustment, which accelerated the ASC (Fig. 8d), while simultaneously explain the lag of the velocity trend relative to the buoyancy flux trend. Meanwhile, due to the significant low-density trend in the mixed layer south of the slope in summer and autumn, the slope of the isopycnal surface continues to become steeper during the same period, and advancing the strong flow of ASC by promoting the baroclinic adjustment. The low-density anomalies entering the ocean will continue to increase and accumulate across the seasons near the slope, leading to a steeper trend in late winter (October, Fig. 8d), prolonging the period of isopycnal surface is steep.

While the contributions of the local surface wind and the buoyancy flux to ASC are significant, it does not imply that the ASC is solely influenced by these factors (Thompson et al., 2018). To comprehensively assess potential contributions from other factors to the long-term change of the ASC, we established a multiple linear regression model incorporating several standardized factors based on the previous researches (Meijers et al., 2010; Thompson et al., 2018, 2020; Stewart et al., 2019; Moorman et al., 2020; Huneke et al., 2022): the buoyancy flux (B), the local surface wind (W), the topography effect (P), the dense water overflow (D), the sea ice concentration (C), and the southern annual model index (SAM). The terms B, W, S, and SAM represent the timeseries or index of the corresponding factors, respectively. D denotes the maximum potential density within each grid water column where bottom density exceeding 27.8 kg/m³ on the continental slope, while the P corresponds to the timeseries of potential vorticity at the slope. Finally, we obtain a predictive equation that passing the regression testing, and the coefficient of determination (R²) is 0.54:where Z at the left of the Eq. (4) is the prediction of the ASC velocity, while the $ {U}_{{\mathrm{std}}} $ and $ \overline {U} $ represent the variance and mean of the velocity timeseries, respectively. During the regression analysis, we filtered out the seasonal variations of the aforementioned factors and found that both the topography and SAM contributed minimally to the ASC velocity, neither of which passed the confidence test. This may be attributed to the fact that topographic influences on the ASC primarily in morphological and spatial characteristics (Baines, 2009), while the influence of SAM on the ASC are predominantly seasonal (Armitage et al., 2018; Auger et al., 2022) and interannual. The overflow enhances oceanic stratification by restricting the downward transport of the cooling and freshening, generating a meridional density gradient with lower (higher) value in the north (south), thereby suppressing the geostrophic velocity of the westward ASC. However, the overflow is limited by the spatially constrained, occurring in the dense shelf where the DSW exists. Additionally, in areas of anomalously increased winter sea ice along the East Antarctica, the sea ice concentration modulates the ASC by impeding the response of ocean to atmospheric momentum input processes, acting more as a damping effect rather than an active dynamic process. Therefore, we attribute the acceleration of the ASC to the result of the local surface wind and buoyancy flux, especially the contribution of buoyancy flux.

However, the seasonal variation of low-density anomaly at the slope caused by buoyancy flux is not exactly the same as ASC (Figs 3 and 8), especially the low-density anomaly cannot explain the additional acceleration of the ASC in autumn. In Fig. 9a, a depiction is presented for multi-year monthly mean speeds of the local surface winds at 10 m, as well as their trends. The black line denotes the multi-year monthly mean of the local surface winds. Meanwhile, the orange solid line illustrates trends for local surface winds spanning from 1993 to 2022 in different months calculated by ERA5, while the orange dashed line is from the ERA-Interim (1993 to 2017) (Fig. 9a). Months passing the 75% confidence test are marked by black hollow pentagrams. The multi-year monthly trend of the wind speed from ERA-Interim data is highly similar to that from the ERA5 data, and the obtained seasonal variations in the wind trend are nearly identical, particularly during autumn and winter. The results from ERA5 and ERA-Interim strengthen the confidence in the phenomenon of the weak enhancement of the local surface wind in autumn. The multi-year monthly trends of the wind speed are depicted in Fig. 9a, only March, April, June, and September passing the 75% confidence test; none passing the 95% confidence test, which further underscores the complexities associated with the polar easterly alterations. By comparing the meridional gradient of the 1 000 hPa geopotential height between the Antarctic coast and the subpolar trough, we infer that the primary positive trend of the Katabatic wind occurs in autumn (April and May, Fig. 9b). Therefore, the enhancement of local surface wind in autumn could be attributed to the positive anomaly trend of the meridional gradient of the geopotential height over the Antarctic continental margin. Moreover, it is certain that the local zonal wind speed can reach 4.0 m/s in June and 2.8 m/s in January, indicating the easterly wind which driving the ASC is still robust.

Local surface wind accelerates the ASC by increasing the direct momentum input to the ocean, and explains the most significant characteristics of the flow trend of the ASC in autumn from the barotropic perspective. Firstly, Fig. 9c shows the seasonal cycle of SIC and its multi-year monthly trend in the range of 2° north and south of the Antarctic shelf break. The seasonal variations in SIC play an important role in the effect of local surface wind on ASC, as the open water resulting from sea ice melting facilitates the wind momentum transfer to the ocean during strong easterly wind, whereas high SIC hinders the response of ocean to the weakening easterly wind. Slight variations in the long-term change of SIC have kept this period relatively stable. Since polar easterlies are strongest and stronger in autumn (April−May; Fig. 9a), although this does not correspond exactly to the period of the open water, changes in the zonal wind stress near the slope suggest that the stronger easterlies are still able to input more momentum directly into the ocean (Fig. 9d). The synchronous correlation coefficient between autumn wind speed and the time series of the flow velocity average in the mixed layer can reach 0.39 (p = 0.03). Especially in the eastern Weddell Sea and the Cosmonaut Sea, where easterly intensities are most significant (Fig. 6). In the flow field, the average depth flow is usually taken as the barotropic component, while the momentum input of the wind to the ocean greatly affects the barotropic flow on the slope (Núñez-Riboni and Fahrbach, 2009; Auger et al., 2022). The long-term trend of ASC in autumn is the most significant, and its additional acceleration compared to the annual mean is also mainly characterized by barotropic (Fig. 3b). Therefore, while buoyancy flux dominates the annual mean acceleration of the ASC, the additional acceleration of ASC in autumn is attributed to the barotropic adjustment caused by local surface wind.