The Argo gridded products provide a globally distributed and continuous time series of T and S profiles down to 2000 m ocean depth. The present work focused on the AAIW in the South Atlantic Basin (Fig. 2, Region A), which encompasses most of the subtropical gyre and a part of the tropical regimes (Boebel et al., 1997; Talley, 1996). Computed from the Argo gridded data of IPRC, the biennial mean θ-S diagram (Fig. 3a) clearly shows that the AAIW has experienced a process of progressive basin-scale freshening during the period from January 2005 to December 2014. The linear trend of salinity (Fig. 3b) further reveals that the freshening takes up most of the AAIW layer but with a little salinification in the deeper part. Except around the 27.42 γn neutral density surface, the AAIW variation is significant at the 95 % confidence level, using the F-test criteria. In comparison with Fig. 3a, it was found that the cutoff point of transformation from salinity decrease to increase is near the salinity minimum. Above the salinity minimum, the shift of θ-S trends towards cooler and fresher values along density surfaces and seems to be a response to the warming and freshening of surface waters where AAIW ventilates. Such thermohaline change has also been found in the Pacific and Indian oceans over a different time period (Wong et al., 1999). Church et al. (1991) and Bindoff and McDougall (1994) have researched the counterintuitive cooling of AAIW temperature induced by warming of surface water. They showed that a warming parcel in the mixed layer would subduct further equatorward, which would lead the θ-S curve to become cooler and fresher at a given density. The salinity decrease of the AAIW core indicates that such a change can only be induced by freshwater input from the source region, as mixing with more saline surrounding waters cannot give rise to a salt loss in the salinity minimum layer.

To demonstrate the robustness of AAIW variations revealed by the IPRC data, re-plots of Fig. 3a–b using another Argo gridded product from JAMSTEC are also shown for comparison (see Fig. S1 in the Supplement; only the AAIW layer is shown). Not only was the same variation along density surfaces in the AAIW layer found, but so too was a freshening of the salinity minimum. The isoneutral salinity increases in both IPRC and JAMSTEC data below the salinity minimum are quite small. The main discrepancy between them is that the salinity reduction in the JAMSTEC data is somewhat less than IPRC and at a higher 95 % confidence level (a mean of 0.006 between 27.1 γn and 27.6 γn).

The freshwater gain for the basin-scale salinity decrease of AAIW (mean salinity difference of 0.012 between 27.1 γn and 27.6 γn over a mean water mass thickness of 500 m) is estimated at 17 mm yr-1 in its source region. This assumes that the South Atlantic only experienced freshwater input and nothing changed, thus the relationship between the salinity in 2005 and 2014 per unit area was roughly S2005 ⋅ 500=S2014 ⋅ (500+Δd). Here S2005=S2014+0.012 and Δd is the freshwater gain during the covered period. However, the depth-integrated salinity change over the water column (between 26.6 γn and 27.6 γn) was 0.0014, since a salinity increase of thermocline water balances the observed freshening of AAIW. This salinity budget implies contemporary hydrological cycle intensification in the Southern Hemisphere, which is illustrated by the P minus E change from 2000 to 2014, with P - E increasing in the subpolar region and vice versa in the subtropical region (Fig. 4a). In these cases, the thermocline (intermediate) water that ventilates in the high-evaporation (precipitation) subtropical (subpolar) regions gets more saline (freshened), as shown by the hydrographic observations (Fig. 3b).

Against the background of hydrological cycle augmentation, the annual freshwater input in the AAIW ventilation region during the freshening period increased by 0.02 mm day-1, about 17 % of the P - E in 2005 (Fig. 4b). It is considered that the significant P - E increase began around 2003 (Fig. 4b, 5-year running mean line), which means the observed freshened AAIW could be traced back to 2003. Though it was not possible to compute the direct freshwater input to the South Atlantic Basin in this study, the Argo-era freshening of AAIW is qualitatively consistent with the freshwater gain in its source region.

Here, two synoptic transatlantic sections from WOCE hydrographic program were used to explore the decadal freshening signal identified in the above subsection. Similar to Fig. 3a, the sectional mean θ-S diagram (Fig. 5a) displays the same shift of thermohaline values, including freshening of the salinity minimum, salinity reduction in the upper AAIW layer and vice versa in the lower layer. Compared to the θ-S curves of IPRC data (Fig. 3a), the curves of WOCE (Fig. 5a) seem to be, in general, cooler θ and fresher S. It is suggested that this is because the IPRC mean is weighted towards the warmer and saltier waters in the north.

Unlike the Argo gridded product which has a continuous time series of T and S data, there are only two sections in the WOCE observations. Instead of calculating the linear trend of salinity (as was done with the IPRC data), the difference in salinity observed in 2003 and 2011 was estimated (Fig. 5b). The light grey shading denotes the 95 % confidence interval using simple t test criteria and considering the number of degrees of freedom. Above the salinity minimum, the freshening of AAIW revealed by the IPRC and the WOCE data are quite similar, with the maximum appearing near 27.2 γn. Because the last WOCE observation terminated in 2011 and the salinity reduction would continue at least up to 2014, as displayed in Fig. 3a, the magnitude of the freshening in WOCE (Fig. 5b) is smaller than IPRC (Fig. 3b). In the water layer below the salinity minimum (around 27.41 γn), the salinity increase shown in the WOCE data is relatively large (Fig. 5b). This is thought to be because the salinity rise reached its maximum around 2011, which is shown in the time series of basin-wide averaged salinity on 27.45 γn- and 27.55 γn-density surfaces (see Fig. S2).

For the salinification of thermocline water, there is a large discrepancy between IPRC and WOCE data on neutral density surfaces 26.6–26.7 γn (Fig. 5b). It is considered that this would not affect the salinity budget over the water column (Fig. 5b), given that the salt gain of thermocline water would balance the observed freshened AAIW. In conclusion, the general trend and consistency of the detail therein of the salinity change over the last 10-year time period, revealed by the IPRC and the WOCE data, leads us to state that the freshening of AAIW is a robust and valid finding.

In modeling studies, it is widely accepted to use a Lagrangian approach to quantify the leakage (Biastoch et al., 2009; van Sebille et al., 2009). Here, a simplified strategy was employed to compute the leakage by integrating the velocity within AAIW layer (approximately between 610 and 1150 m, according to Fig. 1), which was shown to result in a similar quantification to the Lagrangian one (Le Bars et al., 2014). The depth integration is along the Good Hope section (green line in Fig. 2), using the cross-component velocity. Note that the leakage calculation is from the continent to the zero line of the barotropic streamfunction, which is the separation of the Agulhas regime and the Antarctic Circumpolar Current (Biastoch et al., 2015).

Before showing the transport computed from the SODA velocity data, it is necessary to verify that the SODA hydrographic data show the same freshening of AAIW as other datasets. AAIW in the South Atlantic was also found to have freshened during period 2005–2014, though with relatively small magnitude (Fig. S3). Yearly leakage computation within the AAIW layer was carried out for the period 2000–2015 (Fig. 6). It shows that the leakage in the early 2010s is smaller than that in the middle and late 2000s, forming a decreasing trend in a nearly 10-year period. This estimation of leakage seems to be consistent with the indirect estimate of AL transport given below.

The following calculation is to simply estimate the contribution of the AL transport change to our observed freshening. As shown by Fig. 6, the decreased rate of AL transport could be taken to be 2 Sv in a 10-year time period, assuming that this rate increased year by year in the study period (i.e., 0.2 Sv in the first year, 0.4 Sv in the second year, and so on). Following Sun and Watts (2002), here we take the salinity difference of ΔS=0.1 between the South Indian and the South Atlantic in the AAIW layers. The other parameters, including total seconds in a year, water thickness of the AAIW layer and the area of Region A, are taken to be Δt=365×24×3600 s, Δd=500 m and ΔsA=1.09×1013 m2, respectively. Therefore, the salinity decrease from 2005 to 2014, induced by the change of AL transport, should be (0.2 + 0.4 + … + 2) × 106×Δt×ΔS/(ΔsA × Δd). This results in a salinity reduction of 0.0064, which could account for approximately 53.0 % of the observed freshening revealed by the IPRC data. Though our estimate here is quite rough, we can state that, during 2005–2014, the AL significantly influenced the salinity change in the South Atlantic Ocean within the AAIW layers.

Continuous measurements of the AL transport have never been realized before. An earlier study suggested that an increased AL transport correlates well with a poleward shift of the westerly winds (Beal et al., 2011). However, after using reanalysis and climate models, Swart and Fyfe (2012) argued that the strengthening of Southern Hemisphere surface westerlies has occurred without major transgressions in its latitudinal position over the period 1979–2010, during which period the AL has largely increased (Biastoch et al., 2009). A more recent study from Durgadoo et al. (2013) showed that the increase of AL is concomitant with an equatorward rather than a poleward shift of westerlies in their simulation cases. They also concluded that the intensity of westerlies is predominantly responsible for controlling this Indian–Atlantic transport. Many relevant studies agreed on this relationship, that the enhancement of westerly wind intensity is related to the increase of AL (Goes et al., 2014; Lee et al., 2011; Loveday et al., 2015).

The AL corresponds most significantly to westerly wind strength averaged over the Indian Ocean in contrast to that averaged circumpolarly or locally (Durgadoo et al., 2013). According to the work of Durgadoo et al. (2013), zonally averaged WS was calculated from the ERA-Interim wind product over the Indian Ocean (20–110∘ E) for every 5-year period since 1980 (Fig. 7a and d). Previous studies (Lee et al., 2011; Loveday et al., 2015) have found that the WS has increased considerably from the 1980s to the beginning of the 2000s (Fig. 7d), consistent with the contemporary increase of AL transport. Though there are oscillations during 1990s, the WS reached its peak around the years 2000–2004 (Fig. 7d), then began to decline. It can be concluded that the WS has weakened for period 2000–2014 (Fig. 7d), which implies a concurrent decrease of AL transport.

In addition to the ERA-Interim wind data, we have further checked the zonally averaged WS over the Indian Ocean (20–110∘ E), using another reanalysis product of NCEP-2 (Fig. 7b and e) and the combined QuikSCAT–ASCAT (Fig. 7c and f) satellite-derived wind products. The three zonally averaged WS agree that during the period 2000–2014, the westerlies reached a peak in the years 2000–2004, and then progressively subsided through 2005–2009 to 2010–2014. The process of gradual decline of WS is most pronounced in the NCEP-2 data. It is noteworthy that none of the three products show a significant meridional shift of the latitude of maximum WS from 2000 to 2014, in corroboration with the conclusion of Swart and Fyfe (2012).

Many efforts have been made to estimate AL transport, especially using model simulations (Lübbecke et al., 2015; Loveday et al., 2015). In recent years, Le Bars et al. (2014) provided the time series of AL transport over the satellite altimeter era, computed from absolute dynamic topography data, which can show the decadal variation of AL present. In their result (Fig. 8 in Le Bars et al., 2014), the anomalies of AL from satellite altimetry reached a peak around 2003 (annual average), and then began to subside, apart from a mid-2011 increase. In addition, their negative trend of AL (Fig. 9 in Le Bars et al., 2014) over the period from October 1992 to December 2012 indicates that the transport was reduced during the 2000s in contrast to the 1990s. Another study by Biastoch et al. (2015) should be of help in the present discussion. Though the time series of AL obtained from models did not show a distinct decline of AL transport in the last decade, which seems partly due to the data filter applied and the end of the time series (Fig. 4 in Biastoch et al., 2015), it displays a maximum of salt transport around 2000 (Fig. 5 in Biastoch et al., 2015). This peak and the subsequent decline of salt transport are consistent with the freshening of AAIW over the similar time period considered here.

Thus, in addition to the freshwater input that gave rise to the salt loss of the AAIW in the South Atlantic Ocean, reduced transport of AL or salt would further enhance this signal. Unfortunately, the analyses of the contributions from both the source region and the AL were only quantitative. Future work should be focused on the quantification of each factor based on model simulations.