In the South Atlantic at 34.5∘ S the Deep Western Boundary Current (DWBC) is thought to carry the majority of the cold deep limb of the Meridional Overturning Circulation (MOC) southward toward the Southern Ocean. The MOC system dominates the north–south transport of heat and salt in the Atlantic Ocean (e.g., Trenberth et al., 2001; Ganachaud and Wunsch, 2003; Johns et al., 2011), and studies using numerical climate models suggest significant connections between variations in the MOC and changes in societally relevant quantities such as continental precipitation patterns, hurricane intensification, and regional sea level (e.g., Vellinga and Wood, 2002; Stouffer et al., 2006; Latif et al., 2007; McCarthy et al., 2015; Lopez et al., 2016). The pathways that the DWBC, and the lower limb of the MOC, take as they pass through the South Atlantic are not as well understood as in the North Atlantic. In a recent paper, Garzoli et al. (2015) used all available WOCE and CLIVAR-era hydrographic sections including dissolved oxygen and chlorofluorocarbon (CFC) data, Argo float profile data, and two different analyses of the Ocean General Circulation Model For the Earth Simulator (OFES) to trace the time-mean pathways of the DWBC through the South Atlantic. Together with other historical analyses, their results show that the DWBC crosses 5∘ S as a narrow western boundary current, and it becomes dominated by eddies further south around 8–11∘ S, consistent with earlier mooring observations (Dengler et al., 2004; Schott et al., 2005). Previous work has suggested that when this very energetic eddying flow reaches the Vitória-Trindade Ridge at about 20∘ S, the mean flow appears to follow two different pathways, with a significant fraction (estimates ranging from 3 to 12 Sv) flowing eastward across the Mid-Atlantic Ridge toward the African continent (e.g., Zangenberg and Siedler, 1998; Arhan et al., 2003; Hogg and Thurnherr, 2005; van Sebille et al., 2012) and the remainder flowing southward along the western boundary hugging the South American continental slope. The recent analysis of ship sections of CFC, oxygen, and salinity by Garzoli et al. (2015) clearly demonstrates that the strongest pathway in the South Atlantic south of about 22∘ S for recently ventilated North Atlantic Deep Water (NADW), the primary water mass carried in the DWBC, is along the western boundary in the form of a narrow boundary current similar to what is found in most of the North Atlantic. That study suggests that only a small fraction, about 20 %, of the DWBC volume transport flows eastward near 20∘ S, consistent with Arhan et al. (2003) and van Sebille et al. (2012). The Garzoli et al. (2015) study; however, focused primarily on the time-mean circulation pattern and provides little information about the time variability of the DWBC flow, which is the focus of the present study.

Variations of the MOC and the DWBC have historically been studied mostly in the North Atlantic (e.g., Meinen et al., 2013a; Perez et al., 2015; Srokosz and Bryden, 2015; Frajka-Williams et al., 2016; and citations therein). This has mostly been a matter of convenience and proximity, not a reflection on scientific importance, as theoretical work and numerical models have suggested that variations in the South Atlantic may be critical to the stability and flow of the overall MOC system (e.g., Dijkstra, 2007; Drijfhout et al., 2011; Garzoli and Matano, 2011; Garzoli et al., 2013; Buckley and Marshall, 2016). Only in the past few years have observations been collected to study the MOC and/or the DWBC in the South Atlantic region, beginning with repeated upper ocean expendable bathythermograph (XBT) transects (e.g., Garzoli and Baringer, 2007; Dong et al., 2009, 2014) and full-depth hydrographic sections (e.g., Lumpkin and Speer 2003, 2007; Bryden et al., 2011), and later adding continuous moored observations at a few locations including 11∘ S (Hummels et al., 2015) and 34.5∘ S (Meinen et al., 2012, 2013b). Gridded data sets from Argo float profiles in the upper 2000 m of the water column and satellite altimetry measurements have also been brought to bear on the meridional flows in the South Atlantic (e.g., Schmid, 2014; Dong et al., 2015; Majumder et al., 2016), providing important information about latitudinal variations of the MOC. Note that most of these systems focus only on the volume transport in the upper water column, and do not directly observe the DWBC (e.g. XBT, Argo).

True continuous time series observations of the time varying deep limb of the MOC, the DWBC, are very limited in the South Atlantic. In terms of continuous measurements of absolute

The term “absolute” here refers to transports which include both a `baroclinic', vertically sheared, component and a `barotropic', non-sheared, component. Thus “absolute transport” would include all flow that is operating in a geostrophic manner.

The Southwest Atlantic MOC (SAM) array was first deployed at 34.5∘ S in March 2009 to capture the meridional flow of the western boundary currents, with the primary aim of making long-term measurements of the western boundary flows associated with the MOC (Meinen et al., 2012, 2013b). The ultimate long-term goal was also for the SAM array to be a cornerstone for the South Atlantic MOC Basin-wide Array (SAMBA) at 34.5∘ S, which is coming to fruition with parallel deployments occurring on the eastern boundary in 2013 and 2014 (e.g., Ansorge et al., 2014). The SAM array involves four pressure-equipped inverted echo sounders (PIESs) deployed at depths ranging from about 1300 m down to about 4700 m, and extending roughly 650 km offshore from the outer edge of the continental shelf (see Table 1 and Fig. 1). Note that the Site B PIES (see Fig. 1) malfunctioned in 2010 and was lost during a recovery attempt in July 2011, so there is a roughly 1-year gap at that site in 2010–2011.

The array location was selected to be just north of the northern edge of the meander window of the Brazil–Malvinas Confluence (e.g., Gordon and Greengrove, 1986; Olson et al., 1988; Garzoli 1993; Goni et al., 1996, 2011; Lumpkin and Garzoli, 2011) based on altimeter, sea-surface temperature (SST), and surface drifting buoy measurements. Depending on the precise indicator of the Brazil–Malvinas Confluence selected, the seasonal movement of the Brazil–Malvinas Confluence is characterized either by meridional shifts centered near 38.5∘ S (e.g., Matano, 1993; Lumpkin and Garzoli, 2011) or by pivots around a fixed point located near 39.5∘ S, 53.5∘ W, changing its orientation from north–south in austral winter to northwest–southeast in austral summer (Saraceno et al., 2004). The meridional extremes in the confluence location (denoted by sharp horizontal SST gradients) are typically found in February and August, as can be seen in SST maps (Fig. 1, right panels).

Based on recommendations from the broad South Atlantic Meridional Overturning Circulation (SAMOC) Initiative (see www.aoml.noaa.gov/phod/SAMOC_international/), the PIES array was complemented in December 2012 with two current-and-pressure-equipped inverted echo sounders (CPIESs) as part of the parallel South Atlantic MOC Brazil project (SAMOC-Br). These CPIES instruments were deployed near the midpoints of the two westernmost pairs of existing PIES moorings (Fig. 1) in order to provide better horizontal resolution across the western boundary currents.

The analysis of PIES data has become more commonplace within the scientific community over the past few decades, and their use to study the DWBC and the MOC in both the North and South Atlantic has been well documented (e.g., Meinen et al., 2006, 2012, 2013a, b). Therefore, the PIES analysis methods will only be summarized here briefly, with the remainder of the methodology details left to the references cited.

A PIES makes two measurements every hour: (1) the bottom pressure and (2) the vertical round-trip travel time required for a 12 kHz acoustic pulse to travel from the bottom moored instrument up to the sea surface and back. The bottom pressure measurement is made with a highly precise Paros pressure gauge (e.g., Watts and Kontoyiannis 1990; Donohue et al., 2010), while the round-trip travel time is determined using a transducer and a high-quality crystal clock (e.g., Rossby, 1969; Watts and Rossby, 1977; Tracey and Watts, 1986). The travel-time measurements from each PIES are calibrated into daily, full-water-column profiles of temperature, salinity, and specific volume anomaly via hydrography-derived look-up tables using the Gravest Empirical Mode (GEM) technique (e.g., Meinen and Watts, 2000). The application of the GEM method to the PIES in the SAM array was first done in Meinen et al. (2012)

Note: It was recently discovered that, due to a coding mistake, the time-varying bottom-pressure derived term in the absolute velocity in the 2012 study was added with the incorrect sign. The full time series has been recomputed for the present study. Although the character of the absolute-transport time series changes due to this mistake, the major results of the earlier study were not impacted.

The combined observations made by an array of PIESs are powerful, as when combined with the GEM look-up tables they can provide an estimate of the absolute geostrophic velocity, i.e., the combined baroclinic (vertically sheared) plus barotropic (depth-independent) flow, as follows. Vertically integrating the specific volume anomaly profiles generated from the GEM fields and the PIES travel-time measurements yields daily dynamic height anomaly profiles at each of the four instrument sites. Differences in dynamic height anomaly profiles between neighboring PIES sites provide relative geostrophic velocity profiles orthogonal to the line between the PIESs (the “baroclinic” component of the velocity; e.g., Meinen et al., 2006). Differences in bottom pressure from neighboring PIES sites provide absolute geostrophic velocity variability at the bottom that can be used to reference the relative velocity profiles (the “barotropic” component of the velocity; e.g., Meinen and Watts, 2000). Due to the well-known leveling problem, the time-mean absolute geostrophic velocity at the bottom cannot be determined from the bottom pressure differences (e.g., Donohue et al., 2010). The additional measurement of water velocity made by the CPIESs can characterize the flow 50 m above the seafloor. However, given that the two CPIESs were deployed much further apart than the typical velocity decorrelation length scale (e.g., Donohue et al., 2010), and given that there are only measurements at two locations (Fig. 1), these velocity observations are too sparse to solve the time-mean absolute-velocity reference issue. As such those velocity measurements will not be discussed further in this article, and for the purposes of this study the PIESs and CPIESs will be treated interchangeably. As time variability is the focus of this paper, the time-mean issue is not crucial for this study. However, to provide reasonable time-mean absolute-velocity profiles for discussion, the time-mean velocity from an ocean general circulation model (the model is described in the next section) at 1500 dbar was added to the velocity profiles created using the PIES data

Note that in the earlier Meinen et al. (2012) study the model mean velocity was added near the bottom; however, for this study the model velocity at 1500 dbar was used to avoid the significant ageostrophic velocity components in the model in the nearest-bottom grid cell. The results are not highly sensitive to this choice of reference level. Also the time-mean meridional velocities are quite similar if other numerical models are used in place of OFES, e.g., NEMO (see the description of the NEMO run used in Meinen et al., 2013b). For example, the time-mean meridional velocity for the vertical grid cell nearest 1500 dbar, averaged zonally between 51.5 and 44.5∘ W, from this NEMO run (-2.4 cm s-1) is very close to the same value for the OFES run (-1.6 cm s-1).

Most of the detailed testing of PIES-GEM estimated velocities and transports has been done in the North Atlantic where independent estimates were available at concurrent locations – specifically for the DWBC, this has been done at 26.5∘ N (e.g., Meinen et al., 2004, 2006, 2013a). Meinen et al. (2013a) compared daily PIES-derived transports to those determined from dynamic height mooring data at 26.5∘ N and showed that DWBC transports estimated from the two systems had a correlation of r = 0.96, and a root-mean-square difference of 6 Sv. Furthermore, they showed that correlations for baroclinic transports using an assumed level of no motion at 800 dbar were similarly high. Unfortunately there are no independent absolute-transport continuous time series measurements for a similar comparison at 34.5∘ S as there are at 26.5∘ N. However, it can be noted that the same types of instruments are used for both arrays, and the 34.5∘ S GEM SNR of ∼ 20 in the main thermocline depth range, and of 1–3 below 2000 dbar, are similar to what is observed at 26.5∘ N

Note that, like all bottom pressure gauges, the PIES bottom pressure sensors are subject to exponential and/or linear drift problems. These drifts have been removed via the standard methods (e.g., Donohue et al., 2010) in the analysis presented herein; however, variations with periods longer than the record length of each PIES deployment (∼ 4 years) are probably less reliable than variations at shorter periods.

The GEM look-up tables used herein are based on a data set of 200 CTD profiles and 365 Argo profiles collected before the end of 2008. See Meinen et al. (2012) for more detail and an example GEM look-up table. Since the beginning of the SAM project, quasi-annual CTD sections have been collected along the PIES line, both to provide concurrent calibration for the PIES travel times and to observe finer-scale and better horizontal resolution water-mass changes over time. These CTD data have not been incorporated into the GEM fields, and as such they represent an independent data set. For this study, CTD sections from July 2010, December 2010, July 2011, and December 2012 are averaged in a fairly simple manner solely to provide an overview of the major water masses. The CTDs collected right at the PIES sites were also used to calibrate the PIES-measured travel times into the corresponding travel times that would be observed on a common pressure level (e.g., Meinen and Watts, 1998).

To aid in the interpretation of the observations from the PIES array at 34.5∘ S, output from a high-quality, well-validated, numerical model was also used. The OFES (e.g., Sasaki et al., 2008) was selected for this study. The OFES model is produced by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC), and it is a massively parallelized implementation of the National Oceanic and Atmospheric Administration's Geophysical Fluid Dynamics Laboratory's Modular Ocean Model version 3 (MOM3). The model equations have been discretized on a Mercator B-grid with 54 vertical z levels and a horizontal resolution of 0.1∘. For the analysis presented here, model fields were provided by JAMSTEC on 3-day snapshot intervals with a 0.2∘ horizontal grid (i.e., every other grid point) during the period 1980 through 2006. This roughly 20 km spacing is 5 to 15 times finer than the spacing between the PIES moorings in the real ocean. The OFES model was spun up for 50 years using a monthly climatology derived from National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis atmospheric fluxes (Masumoto et al., 2004), and it then was forced with daily mean NCEP–NCAR reanalysis data from 1950 to 2006 (Sasaki et al., 2008). To avoid remaining spin-up transients, only the output from the final 27 years of the run was used for this analysis. This model run was selected because it has previously been validated against both other models and the limited available observations in the South Atlantic (Dong et al., 2011; Perez et al., 2011; van Sebille et al., 2012; Garzoli et al., 2013, 2015).

The time-mean absolute-velocity section calculated from the PIES data during 2009–2014 via the methods described above shows the Brazil Current flowing southward between PIES sites A and B between the surface and roughly 800 dbar, with the DWBC flowing southward below it (Fig. 2a). These flows appear weak and smooth horizontally; however, keep in mind that because these velocities are calculated via the geostrophic method they represent a horizontal average between each pair of PIES sites – i.e., horizontal averages over 2–3∘ of longitude. Also, along the SAM section west of ∼ 49.5∘ W the entire water column flows southward, so there is no obvious separation in the velocity section between the near-surface Brazil Current and the intermediate and deep-water flows. Immediately offshore of these southward flows, recirculations to the north in both the surface and deep layers are observed. Even further offshore, between PIES sites C and D, the flow turns southward once again.

The basic structure of the mean velocity section from the OFES model (Fig. 2b) compares favorably with the mean section from the data (Fig. 2a), albeit with more finely resolved horizontal structure. Additional horizontal information is available from the PIES/CPIES array during the final 2 years – but before looking at that structure it is instructive to first evaluate the time-mean velocities derived from the original four-PIES array only during the final 2-year period (Fig. 2c). The roughly 2-year average is similar to the 4.5 year average (compare Fig. 2a and c), with the upper layer recirculation being slightly stronger and the deep ocean recirculation being slightly weaker or even slightly southward at some depths during the shorter 2-year average. Averaging the model velocity output between pairs of PIESs to simulate the geostrophic averaging (Fig. 2d) yields a section that is qualitatively similar to the 2-year average from the PIES in terms of horizontal and vertical structure, although there are some differences in intensity (compare Fig. 2c and d).

Including the two CPIES records enhances the horizontal structure of the time-mean section, with a more evident Brazil Current core, a stronger upper ocean recirculation core, and a deep recirculation cell that extends to the bottom (Fig. 2e). The model velocity output averaged between the six sites (Fig. 2f) is quite similar to the PIES/CPIES velocity section, although the northward recirculation in the model is weaker than observed both near the surface and at depth. An important point to remember is that the time-mean model velocity at 1500 dbar was used to set the time-mean PIES flow at that pressure level as mentioned earlier (see dashed black lines in Fig. 2), so there is perfect agreement between the PIES/CPIES time-mean velocity and the model velocity at 1500 dbar by construction. (Apparent differences at 1500 dbar are contouring artifacts only.)

As noted earlier, at 34.5∘ S, the western boundary flows in the upper and deep layers (Brazil Current and DWBC, respectively) overlay one another, such that the meridional velocities near the boundary are generally all southward from surface to bottom. One could attempt to use water-mass properties to identify the level that bounds the base of the Brazil Current and the top of the DWBC; however, as will now be shown, this is not particularly satisfying or successful at this location. Average vertical sections of temperature, salinity, dissolved oxygen, and neutral density (Fig. 3) show obvious water-mass layers, perhaps most clearly in the dissolved oxygen (Fig. 3c). For the purposes of the discussion of deep-water flows in this paper, the following water-mass definitions are used following Preu et al. (2013):

Antarctic Intermediate Water (AAIW): salinity less than 34.25 psu;

The time-mean locations of the main DWBC water-mass interfaces demonstrate some rather surprising results when overlain on the time-mean meridional velocity section (Fig. 4). Very near the continental slope, the NADW is carried southward as one would expect in the DWBC; however, immediately offshore the entire NADW layer is being carried northward, essentially heading back toward the northern formation regions, although the array provides no information on how far to the north these waters are carried beyond 34.5∘ S. Similarly, all of the time-mean flow that can clearly be tagged as AABW at this section is headed southward toward the formation region; this result appears to be robust, as small variations in the time-mean flow added at 1500 dbar from a different model (not shown) would not change these southward values (or the northward sign of the NADW mean recirculating flow). The NADW recirculation is not too surprising, as DWBC recirculations in the NADW layer have been commonly observed at many locations along the DWBC pathway through the Atlantic (e.g., Schott et al., 2005; Johns et al., 2008; Meinen et al., 2012; Hummels et al., 2015). The AABW flow to the south, on the other hand, is somewhat unexpected, as it is counter to both previous hydrographic observations (e.g., Hogg et al., 1999; Mémery et al., 2000) and simple intuition given the location of the formation regions for the AABW.

The historical observations of the flow in this region have primarily been geostrophic estimates relative to an assumed level of no motion, which absolute-velocity observations here and elsewhere in the DWBC have called into question (e.g., Meinen et al., 2012, 2013a). The few absolute-velocity observations that have been obtained previously in the region, Lagrangian float and direct current meter measurements around 28–30∘ S in the Brazil Basin, also found hints of recirculation in both the NADW and AABW layers (e.g., Hogg and Owens, 1999; Hogg et al., 1999). There is no question; however, that AABW is observed further north, reaching at least the subtropical North Atlantic in the western half of the Atlantic basin (e.g., Frajka-Williams et al., 2011). The hydrographic observations reported by Coles et al. (1996) provide a possible answer to this conundrum, suggesting a possible interior pathway that would bring AABW to the Vema Channel along the western flank of the Mid-Atlantic Ridge, which would be offshore of the PIES array presented here. The bottom topography contours at AABW depth levels are nearly parallel to the PIES array, which may also complicate capturing an accurate assessment of the deepest flow layers. Another possible reason is simply that 2 years is insufficient to identify the mean flow; the average velocity over the full 5-year record between sites C and D in the AABW layer is very weakly northward (but not statistically significantly different from zero). A detailed analysis of the water masses and their variability is beyond the scope of the present study, and these issues will be revisited in a future manuscript using hydrographic and lowered acoustic Doppler current profiler data that are being collected on recent and near-future cruises. The results shown here do; however, highlight the importance of collecting and interpreting absolute velocities near the boundary, and not just relative velocities.

As has been found at other locations along the DWBC path through the Atlantic, at 34.5∘ S the time-varying intensity of the DWBC volume transport during 2009–2014 (22.8 Sv standard deviation, and peak-to-peak range of 139.4 Sv) greatly exceeds the time-mean value (-15.2 Sv) – which illustrates the necessity of continuous observation to avoid aliasing. The spectral character of the observed variability has a broad peak within roughly 90–160 days, centered near 145 days, with additional energy found in the 20–50 day band. Composite analysis (integrated between both sites A to D and sites A to B) suggests that the variations near the continental slope west of 49∘ W have some barotropicity, in the sense that they affect the flow at all layers including those near the sea surface. The composite results also show that the strong southward transport anomalies near the slope are partially compensated by increased recirculation to the north immediately offshore of 49∘ W, while weak southward transport anomalies near the slope are also partially compensated by decreased recirculation to the north offshore of 49∘ W.

Coupled with analysis of the time-varying flow along the array and analysis of the broader area in a high-quality, high-resolution, well-validated numerical model, the results suggest that the dominant source of transport variations near the continental slope are westward-propagating coherent vortices that superimpose on top of and modulate the intensity of the DWBC flow to yield large southward or northward anomalies depending on the flow associated with the vortices. This suggests that the observing array might be enhanced or improved through the addition of either or both increased horizontal resolution of observations (to more clearly identify these propagating features) and/or the expansion of the array out toward the Mid-Atlantic Ridge (to more completely capture the offshore recirculations). The results also demonstrate the necessity of directly and independently capturing both the “baroclinic” (vertically sheared) and “barotropic” (vertically constant) flows in order to properly understand the absolute-transport variability of the DWBC at this location.