1. Introduction
The West Antarctic Peninsula (WAP) is bordered by the Antarctic Circumpolar Current (ACC), which flows along the continental slope. Below the surface layer, the ACC advects a warm mass of Circumpolar Deep Water (CDW) with significant heat content relative to the in situ freezing point. The CDW spans a range of properties, though Gordon (1971) distinguishes between an upper (UCDW) and lower (LCDW) variety defined by temperature and salinity maxima, respectively. The southern boundary of the ACC is defined as the southernmost presence of UCDW (Orsi et al. 1995) and, unique compared to the rest of Antarctica, near the WAP the ACC flows immediately adjacent to the shelf break, making UCDW readily available to the shelf (Fig. 1).
The WAP is undergoing rapid climate change and the ocean is a primary heat source, particularly in winter when there is no direct radiative forcing. The marginal seas of West Antarctica are warming (Schmidtko et al. 2014) and the increase in heat content along the WAP margin has been attributed to a warming of the UCDW Tmax (Martinson et al. 2008). Cook et al. (2016) confirm an oceanic role in glacier retreat along the WAP by demonstrating an asymmetry in glacial advance/retreat: southern marine-terminating glaciers that have access to warm subpycnocline waters are retreating whereas northern glaciers under the influence of much colder Bransfield Strait waters are not. More and/or warmer CDW has also left its imprint on the atmosphere. The northern portion of the WAP is undergoing rapid winter warming (Turner et al. 2013), presumably related to lighter sea ice cover venting ocean heat to the atmosphere. The reduced sea ice cover, in turn, may be related to changes in the winds and their effect on UCDW delivery and mixing across the pycnocline (Dinniman et al. 2012).
The myriad consequences of UCDW heat have made the exchange of CDW with the WAP shelf an active area of research (Klinck 1998; Smith et al. 1999; Klinck et al. 2004; Dinniman and Klinck 2004; Moffat et al. 2009; Dinniman et al. 2011, 2012; Martinson and McKee 2012; Spence et al. 2014, 2017; Graham et al. 2016; Couto et al. 2017), which is summarized in a recent review by Moffat and Meredith (2018). Various processes have been implicated in driving the cross-isobath transport of CDW onto the WAP shelf, each of which may dominate on different time and/or space scales. The importance of the mesoscale has been argued for by theoretical means (Stewart and Thompson 2015), demonstrated in high-resolution numerical models (St-Laurent et al. 2013; Graham et al. 2016; Stewart et al. 2018), and observed (Moffat et al. 2009; Martinson and McKee 2012; Couto et al. 2017). Warm-core, subpycnocline, primarily anticyclonic eddy-like features have been found within several tens of kilometers from the shelf break, particularly in the vicinity of Marguerite Trough, and are steered along isobaths. The hydrographic properties of the eddies indicate an injection of UCDW as it is found on the continental slope and their length scale is as large as or slightly larger than the first baroclinic Rossby radius, which near the shelf break is about 5 km. Decorrelation lengths of physical and geochemical scalars on the WAP shelf are also about 5 km (Eveleth et al. 2017), suggesting mesoscale eddies may dominate tracer stirring. The eddies’ large heat content relative to surrounding waters indicates a potentially large onshore heat flux. For example, in numerical models, cumulative onshore heat transport is reduced by as much as 50% when model grid spacing is increased from 1 to 2 km (St-Laurent et al. 2013).
Still, very little is known about the origin of the eddies. Their core hydrographic properties and stratification suggest an origin along the continental slope and their observation along isobaths suggests they are, at least in part, advected within the mean flow. Consistent with the former, an idealized numerical model with a continental slope straddled by a jet similar to that observed along the WAP suggests that the jet soon becomes unstable and a Rossby wave containing alternating warm anticyclones and cold cyclones emerges (St-Laurent et al. 2013). Additionally, it is important to understand the processes that attenuate the mesoscale variability and work toward setting the larger-scale shelf stratification. Once UCDW intrudes onto the shelf it mixes to become modified CDW (mCDW), but attempts to understand UCDW transformation have been based on shelf-integrated budgets and have been process independent (Klinck 1998; Smith et al. 1999; Klinck et al. 2004).
In this study we seek to both understand the origin of mesoscale eddies observed on the WAP shelf and then identify and quantify the major processes responsible for their heat loss. Our focus is on the southern grid region of the Palmer Long Term Ecological Research project (Pal LTER; Smith et al. 1995) and in particular the vicinity of Marguerite Trough as this is a region of frequent eddy activity, is close in proximity to the shelf-break jet that we hypothesize generates the eddies, and contains well-defined bathymetric pathways by which to steer the eddies. We primarily use data from a novel Slocum glider survey designed to sample a known pathway for CDW exchange, documenting gradients along the axis of advection for any eddies encountered along the way, and then identify and track an eddy in real time through data-adaptive sampling. This allows, for the first time, high-spatial and temporal resolution transects directly through some of these eddies and a real-time quantification of the attenuation of their core properties. We supplement this with shipboard CTD and ADCP data used to quantify the mixing processes in the WAP environment around and within the eddies and to diagnose the instability that generates the eddies. Ultimately, we simulate the evolution of the eddy as documented by the glider with a simple diffusion model informed by our parameterized mixing processes.
2. Data and observations
The principal dataset used in this study is a set of temperature and salinity profiles collected by a Teledyne–Webb Slocum glider (Schofield et al. 2007) equipped with an unpumped Sea-Bird CTD. Slocum gliders traverse the water column (to 1000 m) in a sawtooth pattern by changing their buoyancy, traveling with average horizontal speed of 0.35 m s−1. There is a hysteresis effect apparent in the up versus down traces that we correct by applying the thermal lag correction of Garau et al. (2011). We also empirically correct for a salinity bias by regressing glider-recorded salinity against ship-recorded salinity, each averaged in subpycnocline temperature bins, for five stations that were occupied by both platforms (maximum temporal separation of 12 days). To facilitate analysis, all corrected up and down traces are binned into 1-m profiles whose latitude, longitude, and time coordinates are taken as the average of all intraprofile samples.
The glider was deployed at Palmer Station at the head of Palmer Deep Canyon (PD) and traveled downgrid before reaching grid station 400.100 (see Fig. 2 for grid convention and physical setting), at which point it began the first of two phases of sampling. In the first phase, the glider flew against the mean current within the anticyclonic cell extending out of Marguerite Trough in order to sample any eddies being advected across the shelf and to identify lateral gradients in their properties. In the second phase, after reaching station 290.115 within Marguerite Trough, it waited to come across an eddy in order to track it in real time. The sampling strategy and basic observations of each of these stages are described below.
a. Glider survey: Advective Path
1) Sampling strategy
The Advective Path (highlighted in Fig. 2) refers to the anticyclonic cell extending out of Marguerite Trough that carries upwelled UCDW to the northern portion of the grid. This path was confirmed to carry UCDW by Martinson and McKee (2012), and its width and central location were inferred from the depth-averaged currents and CDW dye transport in Dinniman et al. 2011 (their currents superimposed on our Fig. 2). Because the width of the current is about the same as the expected diameter of the eddies we were confident that by flying straight through it we would be able to cross any existing eddies traversing the shelf. Nominal glider profile spacing is 1 km, which should afford several profiles through each eddy.
2) Eddy characteristics
As the glider flew upstream (downgrid), it encountered five eddies imbedded within the subpycnocline Tmax layer. We define a mCDW temperature profile as the along-isopycnal average of all profiles whose Tmax is <1.55°C and then compute heat content per unit area Q of the eddy profiles relative to this mCDW profile by differencing the two and integrating between σshal = 1027.64 kg m−3 and σdeep = 1027.76 kg m−3. We use neighboring mCDW as our reference profile because we are interested in how the eddies mix with their surroundings. We choose these isopycnals since 1) they encompass a positive temperature anomaly within eddies, 2) σshal is relatively stable within the permanent pycnocline, and 3) σdeep is near, but does not intersect, the bottom. Series of temperature, salinity, heat content per unit area, and geostrophic current at 280 dbar relative to the bottom are shown in Fig. 3, with the eddies labeled A–E (geographic locations of eddies A–E shown in Fig. 2). Fundamental statistics for each eddy developed and discussed below are given in Table 1.
Fundamental statistics for eddies sampled by glider along Advective Path, where gl0 and gs0 are the grid line and grid station coordinates, respectively, at χ0.
The cores of the eddies contain as much as 5.5 × 108 J m−2 relative to the reference profile. In T–S space, the properties of these eddies are consistent with UCDW. Particularly so, eddy C contains water disjoint from neighboring water, consistent with UCDW as found upstream on the continental slope and indicating warm-core isolation. The upstream eddies A–C are narrower than their downstream counterparts D–E, having radii R ~3 km (RTD ~5 km) compared to R ~7 km (RTD ~9 km). Eddies D–E are broader than the upstream eddies but do not have a substantially smaller heat content per unit area and in fact tend to have more total heat. The large discrepancy in upstream versus downstream total heat content might be more reflective of a sampling bias (offset trajectory through eddy center and stronger head current encountered through eddies A–C) as opposed to an actual difference in heat content. The injection of UCDW into the shelf water column means that the eddies are associated with a downward deflection of isopycnals and anticyclonic shear (Fig. 3). The deflection of isopycnals is largest in the weakly stratified portion of the water column at and below the Tmax, which leads to a cross-track geostrophic velocity signal. To quantify this, we evaluate the geostrophic current at 280 dbar relative to the bottom. We choose this depth as it is the approximate depth of the moored current meters used by Martinson and McKee (2012) upstream where they found the largest eddy signal. A composite over all of the eddies (Fig. 4) reveals a well-defined signature of anticyclonic rotation centered about a warm core, similar to the composites of Moffat et al. (2009).
We measure temperature gradients at the eddy boundaries via a least squares approach. We define
b. Glider survey: Tracking Stage
1) Sampling strategy
We conducted transects along a “fence” perpendicular to the eastern wall of Marguerite Trough (Fig. 2, blue line) to find an eddy and then track it via real-time adaptive sampling. Given the local Rossby radius (~5 km), the mean flow speed, an assumption that eddies are advected within the mean flow, and the lateral deviation of the mean flow, we determined that the fence should be 15 km wide and that it should be surveyed back to back in a 24 h period, which is within the operational constraints of the glider. During sampling, profiles were inspected for a Tmax ≥ 1.8°C as evidence of a UCDW-core eddy and, if found, trajectories were forecast by integrating along both the vector of 24-h mean glider depth-averaged currents and along a streamline fit to the Dinniman et al. (2011) model time–depth-averaged currents, both of which generally agreed within one Rossby radius. We encountered an eddy shortly after initiating sampling and crossed it five times over 4 days. The trajectory followed the eastern wall of Marguerite Trough, consistent with the eddy-like features observed by Moffat et al. (2009).
2) Eddy characteristics
We define a local reference profile in the same manner as before and subtract it in order to obtain Q. Analogous to the Advective Path, sections of data are shown in Fig. 5. However, unlike in that stage, we are not confident that we consistently crossed the eddy via a chord length. Connecting spatial end-members of threshold Tmax along a line perpendicular to isobaths leads to an estimated mean diameter of 8.5 km, similar to the eddies observed on the Advective Path. There is, in general, a decrease in heat content per unit area over time. Anticyclonic shear is less apparent beyond the first few crossings, and the vertical structure of temperature and salt anomalies is much more complicated. We will quantify the heat loss of this eddy in section 5.
c. Shipboard data
1) Pal LTER CTD and SADCP
Shipboard data are used to supplement the glider data and to apply mixing parameterizations. CTD profiles are collected as part of the standard sampling on the annual cruises to the WAP each austral Summer since 1993 and are currently (since 1999) collected with a dual-pumped Sea-Bird 911+ CTD system (see Martinson et al. 2008 for details). The standard grid locations are indicated as black squares in Fig. 2. The entire grid was occupied until 2008 whereas now only a subset of stations is occupied (nominally one coastal, shelf, and slope station per grid line), though this is generally complemented by various process-study CTD casts at nongrid locations that are not plotted but are utilized here.
Processed, high-resolution (5 min in time, 8 m in vertical) velocity profiles from the ARSV L. M. Gould’s hull-mounted RDI 150 kHz narrowband instrument were obtained from the Joint Archive for Shipboard ADCP (SADCP). The 150 kHz instrument provides velocity profiles good to about 300 m (depending on weather and sea state) when the vessel is on station. For mixing parameterizations we need concurrent shear and stratification profiles, so only the overlapping portion of the database is used (Januaries 2000–15; SADCP installed in mid-1999). For those applications, the hydrographic data are bin-averaged onto the same 8-m grid of the velocity data. Stratification and shear are computed as first differences on the 8-m grid as
2) Casts on Advective Path
We draw special attention to 5 CTD casts collected in January 2014 along the Advective Path that was sampled by the glider the year prior (triangles in Fig. 2). These casts fortuitously sampled one or more eddies and indicate substantial interleaving structure, particularly in the pycnocline (Fig. 6). The layering connects cores of injected slope-type UCDW to the surrounding cooler pycnocline. These data are used to assess the importance and origin of thermohaline intrusions into the eddies.
3) S04P casts across continental slope
We suspect the eddies are generated along the continental slope. To diagnose the structure and stability of the shelf-break current upstream of Marguerite Trough we use CTD data from the WOCE S04P cruise in February 1992 (“x” symbols in Fig. 2). Though it is possible that the hydrographic structure over the continental slope has changed in the long interim period, we use these data as they provide a high spatial resolution transect across the slope, slightly finer than the 2011 reoccupation of the line. The stations are “moved” from their original location to the 200 line by tracing the isobath at the actual station location to the 200 line, a small correction (a few kilometers). Neutral density is computed using the Jackett and McDougall (1997) software and profiles of neutral density, temperature, and salinity are filtered with a third-order Butterworth filter with width 100 dbar to remove noise and filter out Charney-type instabilities. We choose to do this because surface-trapped Charney-type instabilities do not convert significant available potential energy (APE) to eddy kinetic energy (EKE) compared to the pycnocline-level instabilities we are seeking (Smith 2007). Geopotential anomaly is calculated from the smoothed temperature and salinity profiles and is linearly extrapolated to handle bottom triangles. Both neutral density and geopotential anomaly are mapped with a Gaussian weighting function to a two-dimensional grid across the slope using the WOCE global climatology vertical grid (Gouretski and Koltermann 2004) and uniform 5-km horizontal spacing.
3. Origin of mesoscale structure
a. QG model and background state
The background state is defined by the gridded stratification and geostrophic velocity profiles at grid location 200.150. This site is chosen as it is located midway across the slope and the geostrophic shear there agrees well with climatological SADCP shear. The current is assumed to flow parallel to isobaths, specifically at an angle of 40° north of east. The local bottom slope is calculated by fitting a 2D plane to all ETOPO1 bathymetry data (Amante and Eakins 2009) within a radius of 0.25° about the grid location.
One assumption of this model is that the background state varies slowly in the horizontal. Since there is clearly lateral shear in the shelf-break current, we need to justify excluding the potential of barotropic instability. For a generic, mixed baroclinic–barotropic instability, the contribution to the growth rate from baroclinic instability scales as
b. Most unstable mode
The equation is discretized as in Smith (2007) and solved for a range of wavenumbers. For each wavenumber, the most unstable mode is that with the largest imaginary part of ω. We find the overall most unstable mode to have an inverse wavenumber |kmax|−1 = 4.4 km and growth rate 0.4 day−1 (Fig. 7). Over the wavenumber space evaluated, this is both the global and only local maximum. The mode’s vertical structure is characterized by nonzero amplitude below the permanent pycnocline in the CDW depth range with a maximum at about 600 m and a secondary maximum at about 350 m. Specifically, this depth range spans the water column presence of UCDW (T ≥ 1.7°C) and LCDW (S = Smax) within the four profiles across the slope. The instability appears to be qualitatively similar to a Phillips type instability (Phillips 1954) for the following reasons: 1) vertical structure seems to be dictated by interior sign changes in the potential vorticity gradient, 2) ~33° phase shift within the amplitude maximum indicates APE release there, and 3) horizontal scale obeys L ≈ (N/f)hpycnocline ≈ 5 km.
Overall these findings are in very good agreement with the glider observations over the Advective Path (Fig. 3). First, the vertical structure of the mode is concentrated within and below the permanent pycnocline, which is where the eddies exist. Second, the observed diameters of eddies A–C upstream and average crest-to-crest separation are ~8.3 km (averages of 2R and 2RTD) and 22.3 km, respectively, which are comparable to the theoretical diameters 2|kmax|−1 = 8.8 km and crest-to-crest separation 2π|kmax|−1 = 27.8 km for eddies spun off of an unstable Rossby wave. Temperature anomalies relative to mCDW and geostrophic currents provide some evidence for a cold cyclone between eddies B–C, though in general, only warm anticyclones tend to be found on the shelf. The eddies D–E downstream on the path are broader than A–C but have a similar separation. With the caveat that this is a linear model, note that the wavelength implies delivery of (365 days) × u|kmax|/2π = 57–114 eddies per year (for typical shelf currents of 0.05 or 0.10 m s−1), which could account for the totals observed by Moffat et al. (2009) and Martinson and McKee (2012).
c. Roles of bottom slope and current orientation
Exploring the parameter space of bottom slopes and current orientations allows us to simultaneously understand the sensitivity of the most unstable mode’s structure and growth rate to these parameters and to understand how generalizable these results are to other regions around Antarctica. Using the same geostrophic shear and stratification profiles, we repeat the above analysis for all bottom slopes between −0.15 and +0.15 at increments of 0.01 and for all current orientations between 0° (zonal) and 90° (meridional) counterclockwise from east at 10° increments and examine changes in the growth rate, wave vector, and structure of the most unstable modes. Evaluating positive and negative bottom slopes effectively allows consideration of both prograde (isopycnals slope in same sense as bathymetry; e.g., WAP) and retrograde (isopycnals slope in opposite sense as bathymetry; e.g., Ross Sea) jets.
Figure 8 shows the growth rate and inverse wavenumber of the most unstable mode over the entire parameter space. The orientation of the current has essentially no effect on the instability. This is likely because the planetary beta effect is so small at this latitude that potential vorticity gradients are dominated by the stretching term and bottom slope. Indeed the topography plays a large role in determining the strength and properties of the most unstable mode. It is found that negative bottom slopes are stabilizing while increasing positive bottom slopes are destabilizing to a point and then stabilizing. Blumsack and Gierasch (1972) demonstrate that the relevant parameter for the stability problem under QG scaling is not the bottom slope itself but rather the ratio of the bottom slope to the isopycnal slope, δ ≡ α/s (Blumsack and Gierasch 1972; Poulin et al. 2014). In the QG model of Smith (2007) that we use, the bottom slope only enters the problem as a boundary condition in the bottom layer’s potential vorticity gradient. When the bottom slope exceeds the slope of the 1028.0 kg m−3 neutral surface, the potential vorticity gradient vanishes in the lower layer which has the effect of suppressing the growth rate in accordance to the Charney–Stern criteria (Pedlosky 1987). Nevertheless, this does not change the fact that there are still sign changes in the interior potential vorticity gradient and therefore instability persists where δ > 1 or δ < 0 (Isachsen 2011). These regions of weaker instability with inverse wavenumbers of 4–5 km all have similar vertical structure as the most unstable mode obtained in the realistic scenario analyzed earlier.
The fastest growing modes for prograde currents with relatively flat bottoms (0 ≤ δ ≤ 1) are qualitatively consistent with Eady modes for the following reasons: 1) they have amplitude maxima at both ~350 m and near the bottom; 2) they have an inverse wavenumber L ≈ NH/(1.6f) ≈ 9.4 km, probably since the weakly varying N is dynamically similar enough to the uniform N of Eady’s model. These modes have a pronounced spike at ~350 m near the UCDW temperature maximum and, in regions with a flatter bottom slope than the WAP, could also be expected to contribute to exchange of CDW. The region between −2 ≤ δ < 0 represents a sort of transition between the Eady-type modes and the Phillips-type modes. They have smaller length scales (~2 km) and are bottom-boundary trapped. This trend fits the qualities described by Blumsack and Gierasch (1972) for increasingly negative slope parameter, namely decreasing length scale and boundary trapping.
4. Attenuation of mesoscale structure
Having identified a plausible origin for the UCDW eddies, we here consider the processes responsible for their decay. Box inversions of steady heat budgets on the WAP point to diapycnal mixing between overlying remnant winter mixed layer Winter Water (WW) and a constantly replenished CDW layer as the maintenance of the permanent pycnocline. For some context, using a steady advective–diffusive balance, Klinck et al. (2004) place an upper bound on the vertical (lateral) diffusivity of heat at 7.7 × 10−4 m2 s−1 (1600 m2 s−1) in the limit of no lateral (vertical) mixing. In an earlier study, Klinck (1998) used seasonal changes in water properties and a similar integrated budget to find a vertical (lateral) diffusivity of heat as 1 × 10−4 m2 s−1 (37 m2 s−1). Martinson et al. (2008) used interannual variability of WW heat content and assumed a UCDW replenishing time to suggest a vertical diffusivity of 8.5 × 10−5 m2 s−1. Our study, however, focuses on how the properties of the subpycnocline “box” are set by the mixing of advected parcels of UCDW, within which the vertical and lateral gradients are quite different from those used in mean-shelf balances.
a. Shear-driven instability
Shear-driven instability is thought to be important in maintaining the permanent pycnocline and is thought to yield larger heat fluxes on the WAP than double-diffusive instabilities (Howard et al. 2004). Regarding sources of shear, internal tides are likely not important but near-inertial waves may be. Howard et al. (2004) suggest the semidiurnal tide is weak, as is the stratification, making baroclinic conversion unlikely. Further, Beardsley et al. (2004) use rotary spectral analysis of drifter velocity to show power in the semidiurnal band is two orders of magnitude less than that in the near-inertial. In general, we might expect that vertical mixing should be elevated over seamounts or within cross-cutting canyons. At seamounts, internal waves with frequency at the critical slope may be generated whereas in canyons internal waves may become trapped and focused toward the canyon head (Gordon and Marshall 1976).
Because we do not have microstructure measurements within eddies or on the surrounding midshelf, our approach is to use SADCP and CTD data to estimate diffusivities, temperature gradients, and heat fluxes across the permanent pycnocline, which we define globally to be between 98 and 250 m. We exclude grid stations below the 000 grid line as the hydrography there is very different (very deep and cold remnant winter mixed layers). Note that winter mixing by entrainment of the thermocline during brine rejection is not considered here, but is important in the annual heat budget (Martinson and Iannuzzi 1998).
1) General shear-driven instability
2) Internal wave parameterizations
3) Results and interpretation
Fundamental results are presented in the form of profiles of N2, S2, and Kz averaged over the Shelf region (see Fig. 2 for regional boundaries) in Fig. 9. In general, S2 decreases more slowly with depth than does N2, becoming nearly white below the pycnocline and yielding diffusivities that increase with depth. To confirm that the apparently unusual behavior of S2 with depth is real, we compare the SADCP shear at station 200.140 to the shear calculated from a set of six moored current meter records at that site [corrected for finite differencing following Gregg and Sanford (1988) using local GM scaling]. The current meter shear (bold blue line) and SADCP shear (thin black line) at site 200.140 agree remarkably well, suggesting that, at least at this site, the large shear variance (and hence the large diffusivities) at depth is real (Fig. 9b).
While the SADCP shear variance represents an integral across all frequencies, we can bandpass filter the current meter observations between [f, N0] to retain only shear in the internal wave band. Doing so (dashed blue line in Fig. 9b) suggests that the internal wave band shear profile has the same shape as the total shear profile but that SADCP shear might be overestimating internal wave shear by a factor of ~1.6. Because we cannot evaluate this relation at other sites, we do not attempt to make any correction for non-internal wave shear. The histogram of
Both the G89 and PP81 diffusivities increase with depth for reasons discussed above. Diffusivity values in the permanent pycnocline are small (<10−5 m2 s−1), so much so that the asymptotic lower limit of the PP81 method renders it inapplicable there (Fig. 9c). Although the two methods agree better deeper in the water column, all further analyses will exclusively consider the G89 parameterization. The shear (Fig. 9b) and diffusivity (Fig. 9c) profiles are presented as means with the shaded region enclosing two standard errors, where both the means and standard errors are computed on log-transformed data. Owing to the large number of stations sampled, uncertainty in the mean is small.
To get a sense of how diffusivity and vertical heat fluxes vary regionally, Fig. 10 shows composite profiles of diffusivity and heat flux for data partitioned by region alongside histograms of the pycnocline-averaged quantities. The small diffusivities that characterize the permanent pycnocline operating on the background T profile yield vertical heat fluxes across the permanent pycnocline of <1 W m−2 which is smaller than values reported by Howard et al. (2004), who used PP81. Diffusivities increase approximately log-linearly with depth to about 10−4 m2 s−1 at 300 m (Fig. 10), which is as deep as the SADCP yields useable data.
As in Fig. 9, the central values are computed on log-transformed data, however now the spread is indicated by two standard deviations in order to indicate potential values instead of uncertainty in the mean. While the central values are small, the spread at any given depth is large and spans about two orders of magnitude, suggesting that the potential for strong mixing at any depth is high and that mixing is intermittent. Pycnocline diffusivities are significantly different between the Shelf and Slope, the Shelf and Coast, and between the Northern Shelf and Southern Shelf at an α = 10% level. The variance seems to be larger in the Coast. The larger variance there is driven by larger variance in the shear. We also constructed two composites that should be more representative of the environment that most of the eddies are in. The first averages all profiles in the subregion of the Southern Shelf that is bounded by the 250 and 450 lines between stations 030 and 130 (the vicinity of Marguerite Trough) and the second averages only those profiles in that domain with a Tmax ≥ 1.55°C. The heat fluxes in those regions are qualitatively different with a much greater heat flux divergence centered about the Tmax but are statistically indistinguishable from the more comprehensive Southern Shelf in terms of cross-pycnocline metrics. However, compared to the Northern Shelf, the Southern Shelf and both Marguerite Trough composites have significantly lower pycnocline diffusivities, significantly lower shear variance ratios, and insignificantly larger pycnocline heat fluxes. The results suggest that any enhanced vertical heat flux within an eddy compared to its surroundings is due to the altered gradients of the temperature profiles and not the shear.
While the SADCP does not allow for estimates of diffusivity or heat flux below about 300 m, extrapolating the log-linear trend in Kz with depth and using observed temperature gradients suggests that below the Tmax, the vertical temperature gradient is about one order of magnitude smaller than that in the permanent pycnocline, however the diffusivity should be about two orders of magnitude larger. This combines for a heat flux about 10 times greater below the eddy, and the spatial variability in the heat flux profiles is consistent with this scaling. The stratification on the Slope and Southern Shelf is dominated by pure UCDW and a middepth temperature maximum as in the eddies. The middepth maximum in temperature results in an upward heat flux above the Tmax and a downward heat flux below, the former of which is apparent in the observed profiles and the latter of which is implied by the trend toward a ~0 W m−2 heat flux at 300 m (Fig. 10). Because the downward heat flux is about 10 times greater than the upward heat flux, the temperature maximum is mixed downward in the water column as the UCDW is advected northward and shoreward. This is consistent with the positive vertical heat flux increasing with depth on the Northern Shelf and Coast stations (Fig. 10).
b. Thermohaline intrusions
Four of the five CTD casts taken along the Advective Path reveal that within and surrounding the eddies there is substantial along-isopycnal temperature and salinity variance associated with thermohaline intrusions (Fig. 6). Those data are plotted as profiles of temperature in Fig. 11 along with a background profile, which is constructed as a running median in density space with a window size of 0.015 kg m−3. The UCDW eddies are essentially moving fronts. Joyce (1977) derived a model in which medium-scale advection of heat and salt across a density-compensated large-scale front is balanced by small-scale diffusion across thermohaline intrusions, attenuating their T–S characteristics. The medium-scale advection is taken to be in the form of alternating interleaving structures, initiated by velocity perturbations with the energy source coming from thermoclinic energy of the cross-frontal property contrasts. His model makes no distinction as to what small-scale processes conduct the mixing.
1) Double-diffusive growth
Theories for the growth of thermohaline intrusions (e.g., McDougall 1985a,b) in a region of density-compensated thermohaline gradients depend on one of the components of density being unstably stratified in order for an infinitesimal disturbance to grow to finite length. Figure 11 shows that the WAP water column between the WW Tmin and the UCDW Tmax is diffusively unstable and that the layer immediately below the UCDW core can be salt finger unstable. One way to quantify how much the unstably stratified component contributes to the stability of the water column is via the density ratio
2) Internal wave advection
3) Geostrophic turbulence
To unite density-compensated variations into a single variable we consider spice (Flament 2002) which effectively serves as a quantification of distance in T–S space orthogonal to isopycnals. We compute spice profiles C(z) from the glider data along the Advective Path in addition to high-passed spice profiles CHP = C − CLP, which are the total profiles minus a filtered profile (triangular filter with width 25 m), thus retaining only the interleaving layers. Inspection of composites of
Estimation of the mixing length requires knowledge of the mean spice gradient that is presumably stirred. Figure 14c shows time-averaged spice
Importantly, it is not known whether the interleaving layers are generated locally or remotely, so it is not obvious which is the relevant gradient to be stirred. Likewise, it is not immediately clear whether the relevant variance is
4) Discussion
It is not obvious what causes the interleaving. Internal waves cannot provide enough temperature variance unless the lateral temperature gradient is on the larger end of those observed and the internal wave field is strongly biased toward inertial waves (which it somewhat is with Rω = 9). On the other hand, double diffusion, while qualitatively consistent with the vertical structure of the observations, tends to have slow growth rates. A more likely candidate appears to be stirring by the eddies of the cross-shelf temperature gradient within Marguerite Trough. Spice variance is largest at the eddy edges and its magnitude is consistent with the eddy sizes in accordance with mixing length theory. Thermohaline intrusions appear to be important for the decay of the eddies, particularly in their upper hemispheres, as cold neighboring water is brought inwards and warm slope water is ejected.
c. Frictional spindown
5. Validation
Here we use a simple diffusion model informed by the results of the previous section to simulate the total heat loss of the eddy we tracked over crossings 1–5. We would like to emphasize that we do not expect to reproduce the small-scale structure of the eddy’s cooling but instead aim to capture the bulk heat loss. The geometry of the first crossing gives us confidence that we passed directly through the eddy’s center and through its entire diameter, therefore we set the initial condition as the hydrographic structure of the eddy as sampled on crossing 1. We identify the eddy’s geographic center (x1, y1) and radius (R1 = 3.4 km) as the mean and standard deviation, respectively, of a 2D Gaussian fit to the field of heat content per unit area integrated between σshal and σdeep. We then assign each profile a radial distance from that center and grid the T and S on each isopycnal to construct a 3D field for the eddy. Outside of the eddy the domain is populated by the mCDW profile.
a. Diffusion model
b. Fit to data
We use a diffusion-only equation because the eddy’s geographic coordinates [x(t), y(t)] are additional unknowns. Therefore we assume that the eddy is advected by a depth-invariant current and seek an optimal [x(τi), y(τi)] at each crossing i. We do this by performing a grid search by moving the modeled eddy to each location in a grid encompassing the glider track and then subsampling the modeled field along the glider track and calculating the depth-integrated heat content along the track. The position k that minimizes the root-mean-square (RMS) error between
c. Results and interpretation
The best fit locations and modeled heat content are shown in Fig. 15. The predicted eddy track flows counter clockwise about the seamount east of Marguerite Trough (near 300.060), consistent with the mean flow, and the glider’s excursion to the south confirms that this eddy did not go into Marguerite Bay. We are confident that the glider was indeed sampling the same eddy. Simulated particle trajectories (not shown) originating at [x(τ1), y(τ1)] and using all possible 4-day sequences of the time-varying, depth-averaged current fields from Dinniman et al. (2011) confirm that parcels of water can follow the best-fit trajectory and can traverse the proposed distance traveled by the eddy over the total observation time.
In general, there is good agreement between the time series of observed and modeled along-track heat content (Fig. 15b). Particularly, the agreement is good near the edge of the eddy which contributes more to the integral of total heat content. There is a wide variety of intrusions, filimentation, and other submesoscale variability along the track. For example, the fourth crossing appears to skirt the eddy edge but contains a profile of water ejected from the eddy core. The second and third crossings appear to show the eddy core “split” by the cool layer that began penetrating the eddy during the first crossing. Obviously the model does not simulate these processes though it does seem capable of parameterizing their consequences. The total integrated heat contents fall within the range of perturbed model fits (Fig. 15c).
The results can be used to estimate the rate of heat loss of the eddy between crossings 1 and 5. The total heat content of the eddy during crossing 1 is easily obtained by integrating the model initial condition (itself a fit to the data) vertically between σshal and σdeep, and then laterally out to R1. At crossing 5, we obtain Qmax,chord and Rchord by fitting Eq. (2.1) to the data but to calculate the total heat content we need a cross section through the eddy center. Simple trigonometry shows that
6. Discussions and conclusions
The first goal of this study was to understand the origin of UCDW eddies on the shelf. We conducted a linear stability analysis of the shelf-break current to show that its most unstable mode has a CDW-level amplitude maximum and an inverse wavenumber of 4.4 km. Glider observations confirmed that UCDW eddies on the shelf have diameters and crest-to-crest separation consistent with that metric and their anomalies are confined to the subpycnocline layer. The strong shear and weak stratification allow for sign changes in the interior potential vorticity gradient which cause instabilities smaller than the Rossby radius to persist even when Eady modes are suppressed by the bottom slope. This, in combination with the insensitivity to the planetary beta effect, suggests to us that similar UCDW eddies should be common around the Antarctic margins. However, we must be cautious with any broad extensions. While we have shown that either eastward or westward currents can support eddies, westward currents are often associated with the Antarctic Slope Front and very different stratification. The demonstrated sensitivity of eddy fluxes of CDW to other parameters that were not considered here, such as depth of the continental shelf or prominence of the Antarctic Slope Front (Stewart and Thompson 2015), restrict any generalizations and suggest eddy fluxes are still likely highly localized (Stewart et al. 2018) with one region of importance being the WAP.
The second goal of this study was to identify the major mixing processes that disperse eddy heat into surrounding water. Shipboard CTD and SADCP measurements were used to show cross-pycnocline diffusive fluxes should on average be <1 W m−2 while fluxes below the eddy Tmax should be an order of magnitude larger. CTD and glider profiles reveal that interleaving layers and thermohaline intrusions are ubiquitous, particularly in the lower pycnocline/eddy upper hemispheres where they link warm slope-type UCDW to the cooler waters within the adjacent water column’s broad pycnocline. This water is most quickly eroded over the Advective Path while the deeper, core Tmax at ~250 m persists longer. The erosion of the core heat content per unit area is more apparent than that of the temperature maximum, implying erosion at the eddy boundaries and/or a redistribution of heat within the eddy interior. Consistent with this, high-pass-filtered spice variance is largest at the eddy edges.
The origin of the thermohaline intrusions is ambiguous. The magnitude of their thermal variance is larger than would be expected if due solely to internal wave shear, although this conclusion is highly dependent on the magnitude of the lateral temperature gradient across the eddy. In addition, thermal variance is largest above and just below the eddy core, both regions susceptible to double-diffusive instabilities. Instead, given the observed along-isopycnal spice variance, the mean spice gradient, and the size of the first-mode instability, a more plausible explanation for the origin of the interleaving layers may be stirring by the eddies themselves. A recent modeling study (Stewart et al. 2018) suggests that eddies may transfer heat across the Antarctic slope primarily by along-isopycnal stirring as opposed to advection by their overturning streamfunction. This also supports stirring as an explanation for the thermohaline variance.
The flexibility afforded by Slocum gliders allows for real-time, data-adaptive sampling. One of the novel aspects of this study is the collection of cross sections through an eddy as it traversed the shelf, providing a first estimate of the eddies’ rate of cooling, which is consistent with that of a 3D diffusion model applied to the initial crossing. A lower limit on the time required to eliminate all heat relative to the mCDW profile within one radius from the eddy center is given by Qtot(τ1)/(∂Q/∂t) = 13.1 days which assumes that the gradients are constant in time. Because lateral mixing dominates, an alternate cooling time is given by the solution to the two-dimensional diffusion equation with homogeneous lateral boundary conditions at infinity and a Gaussian initial condition of radius R. That solution implies that the heat content per unit area at the eddy center decreases as R2/(R2 + 4Kht). For the tracked eddy a 50% decrease is achieved after 10.3 days. For a typical eddy (R = |kmax|−1), a 50% decrease takes 17.5 days.
Importantly, a preference for the eddies to mix laterally and downward suggests that the eddies are good at redistributing heat rather than immediately venting it to the atmosphere. This sheltering has implications for the ability of intra and subpycnocline heat to persist and make its way shoreward toward marine-terminating glaciers even as the eddies themselves cease to remain coherent structures.
Acknowledgments
McKee and Martinson were supported by NSF Grants OPP-08-23101 and PLR-1440435; Schofield was supported by NSF Grant ANT-08-23101 and NASA Grant NNX14AL86G. Richard Iannuzzi processed the CTD and glider data, made Fig. 1, and provided valuable comments. We thank David Aragon and the Rutgers COOL group for deploying the glider and providing logistical support. We also thank Pat Caldwell at the JASADCP for culling and providing the SADCP data and Mike Dinniman for providing model current fields. We also acknowledge the captains and crews of the L.M. Gould and the N.B. Palmer for their efforts in acquiring many years of LTER data. Insightful comments from the editor and two anonymous reviewers greatly improved this manuscript. This is LDEO contribution 8310 and Palmer LTER contribution 620.
APPENDIX
The Garrett–Munk Spectrum on the WAP Continental Shelf
REFERENCES
Amante, C., and B. W. Eakins, 2009: ETOPO1 1 Arc-Minute Global Relief Model: Procedures, data sources and analysis. NOAA Tech. Memo. NESDIS NGDC-24, National Geophysical Data Center, 25 pp., https://www.ngdc.noaa.gov/mgg/global/relief/ETOPO1/docs/ETOPO1.pdf.
Beardsley, R. C., R. Limeburner, and B. W. Owens, 2004: Drifter measurements of surface currents near Marguerite Bay on the western Antarctic Peninsula shelf during austral summer and fall, 2001 and 2002. Deep-Sea Res. II, 51, 1947–1964, https://doi.org/10.1016/j.dsr2.2004.07.031.
Blumsack, S. L., and P. J. Gierasch, 1972: Mars: The effects of topography on baroclinic instability. J. Atmos. Sci., 29, 1081–1089, https://doi.org/10.1175/1520-0469(1972)029<1081:MTEOTO>2.0.CO;2.
Bormans, M., 1992a: Effect of density ratio on double diffusive interleaving. Deep-Sea Res., 39, 871–884, https://doi.org/10.1016/0198-0149(92)90126-E.
Bormans, M., 1992b: An experimental study on the formation and survival of stratified subsurface eddies. J. Geophys. Res., 97, 20 155–20 167, https://doi.org/10.1029/92JC01973.
Cook, A. J., P. R. Holland, M. P. Meredith, T. Murray, A. Luckman, and D. G. Vaughan, 2016: Ocean forcing of glacier retreat in the western Antarctic Peninsula. Science, 353, 283–286, https://doi.org/10.1126/science.aae0017.
Couto, N., D.G. Martinson, J. Kohut, and O. Schofield, 2017: Distribution of Upper Circumpolar Deep Water on the warming continental shelf of the West Antarctic Peninsula. J. Geophys. Res. Oceans, 122, 5306–5315, https://doi.org/10.1002/2017JC012840.
Dewey, R., R. Muench, and J. Gunn, 1999: Mixing and vertical heat flux estimates in the Arctic Eurasian Basin. J. Mar. Syst., 21, 199–205, https://doi.org/10.1016/S0924-7963(99)00014-7.
Dinniman, M. S., and J. M. Klinck, 2004: A model study of circulation and cross-shelf exchange on the west Antarctic Peninsula continental shelf. Deep-Sea Res. II, 51, 2003–2022, https://doi.org/10.1016/j.dsr2.2004.07.030.
Dinniman, M. S., J. M. Klinck, and W. O. Smith Jr., 2011: A model study of Circumpolar Deep Water on the West Antarctic Peninsula and Ross Sea continental shelves. Deep-Sea Res. II, 58, 1508–1523, https://doi.org/10.1016/j.dsr2.2010.11.013.
Dinniman, M. S., J. M. Klinck, and E. E. Hofmann, 2012: Sensitivity of circumpolar deep water transport and ice shelf basal melt along the West Antarctic Peninsula to changes in the winds. J. Climate, 25, 4799–4816, https://doi.org/10.1175/JCLI-D-11-00307.1.
Eveleth, R., N. Cassar, S. C. Doney, D. R. Munro, and C. Sweeney, 2017: Biological and physical controls on O2/Ar, Ar, and pCO2 variability at the Western Antarctic Peninsula and in the Drake Passage. Deep-Sea Res. II, 139, 77–88, https://doi.org/10.1016/j.dsr2.2016.05.002.
Flament, P., 2002: A state variable for characterizing water masses and their diffusive stability: Spiciness. Prog. Oceanogr., 54, 493–501, https://doi.org/10.1016/S0079-6611(02)00065-4.
Garau, B., S. Ruiz, W. G. Zhang, A. Pascual, E. Heslop, J. Kerfoot, and J. Tintoré, 2011: Thermal lag correction on Slocum CTD glider data. J. Atmos. Oceanic Technol., 28, 1065–1071, https://doi.org/10.1175/JTECH-D-10-05030.1.
Garrett, C., 1982: On spindown in the ocean interior. J. Phys. Oceanogr., 12, 989–993, https://doi.org/10.1175/1520-0485(1982)012<0989:OSITOI>2.0.CO;2.
Garrett, C., and W. Munk, 1975: Space-time scales of internal waves: A progress report. J. Geophys. Res., 80, 291–297, https://doi.org/10.1029/JC080i003p00291.
Gordon, A. L., 1971: Oceanography of Antarctic waters. Antarct. Res. Ser., 15, 169–203, https://doi.org/10.1029/AR015p0169.
Gordon, R. L., and N. F. Marshall, 1976: Submarine canyons: Internal wave traps? Geophys. Res. Lett., 3, 622–624, https://doi.org/10.1029/GL003i010p00622.
Gouretski, V. V., and K. P. Koltermann, 2004: WOCE global hydrographic climatology: A technical report. Bundesamt für Seeschifffahrt und Hydrographie Rep. 35, 52 pp.
Graham, J. A., M. S. Dinniman, and J. M. Klinck, 2016: Impact of model resolution for on-shelf heat transport along the West Antarctic Peninsula. J. Geophys. Res., 121, 7880–7897, https://doi.org/10.1002/2016JC011875.
Gregg, M. C., 1989: Scaling turbulent dissipation in the thermocline. J. Geophys. Res., 94, 9686–9698, https://doi.org/10.1029/JC094iC07p09686.
Gregg, M. C., and T. B. Sanford, 1988: The dependence of turbulent dissipation on stratification in a diffusively stable thermocline. J. Geophys. Res., 93, 12 381–12 392, https://doi.org/10.1029/JC093iC10p12381.
Gregg, M. C., T. B. Sanford, and D. P. Winkel, 2003: Reduced mixing from the breaking of internal waves in equatorial waters. Nature, 422, 513–515, https://doi.org/10.1038/nature01507.
Hebert, D., N. Oakey, and B. R. Ruddick, 1990: Evolution of a Mediterranean salt lens: Scalar properties. J. Phys. Oceanogr., 20, 1468–1483, https://doi.org/10.1175/1520-0485(1990)020<1468:EOAMSL>2.0.CO;2.
Howard, S. L., J. Hyatt, and L. Padman, 2004: Mixing in the pycnocline over the western Antarctic Peninsula Shelf during Southern Ocean GLOBEC. Deep-Sea Res. II, 51, 1965, https://doi.org/10.1016/j.dsr2.2004.08.002.
Isachsen, P. E., 2011: Baroclinic instability and eddy tracer transport across sloping bottom topography: How well does a modified Eady model do in primitive equation simulations? Ocean Modell., 39, 183–199, https://doi.org/10.1016/j.ocemod.2010.09.007.
Jackett, D. R., and T. J. McDougall, 1997: A neutral density variable for the world’s oceans. J. Phys. Oceanogr., 27, 237–263, https://doi.org/10.1175/1520-0485(1997)027<0237:ANDVFT>2.0.CO;2.
Joyce, T. M., 1977: A note on the lateral mixing of water masses. J. Phys. Oceanogr., 7, 626–629, https://doi.org/10.1175/1520-0485(1977)007<0626:ANOTLM>2.0.CO;2.
Klinck, J. M., 1998: Heat and salt changes on the continental shelf west of the Antarctic Peninsula between January 1993 and January 1994. J. Geophys. Res., 103, 7617–7636, https://doi.org/10.1029/98JC00369.
Klinck, J. M., E. E. Hofmann, R. C. Beardsley, B. Salihoglu, and S. Howard, 2004: Water-mass properties and circulation on the West Antarctic Peninsula continental shelf in austral fall and winter 2001. Deep-Sea Res. II, 51, 1925–1946, https://doi.org/10.1016/j.dsr2.2004.08.001.
Kunze, E., E. Firing, J. M. Hummon, T. K. Chereskin, and A. M. Thurnherr, 2006: Global abyssal mixing inferred from lowered ADCP shear and CTD strain profiles. J. Phys. Oceanogr., 36, 1553–1576, https://doi.org/10.1175/JPO2926.1.
Levine, M. D., 2002: A modification of the Garrett–Munk internal wave spectrum. J. Phys. Oceanogr., 32, 3166–3181, https://doi.org/10.1175/1520-0485(2002)032<3166:AMOTGM>2.0.CO;2.
Martinson, D. G., and R. A. Iannuzzi, 1998: Antarctic ocean-ice interaction: Implications from ocean bulk property distributions in the Weddell Gyre. Antarctic Sea Ice: Physical Processes, Interactions and Variability, M. Jeffries, Ed., Antarctic Research Series, Vol. 74, Amer. Geophys. Union, 243–271.
Martinson, D. G., and D. C. McKee, 2012: Transport of warm Upper Circumpolar Deep Water onto the western Antarctic Peninsula continental shelf. Ocean Sci., 8, 433–442, https://doi.org/10.5194/os-8-433-2012.
Martinson, D. G., S. E. Stammerjohn, R. A. Iannuzzi, R. C. Smith, and M. Vernet, 2008: Western Antarctic Peninsula physical oceanography and spatio-temporal variability. Deep-Sea Res. II, 55, 1964–1987, https://doi.org/10.1016/j.dsr2.2008.04.038.
McDougall, T. J., 1985a: Double-diffusive interleaving. Part I: Linear stability analysis. J. Phys. Oceanogr., 15, 1532–1541, https://doi.org/10.1175/1520-0485(1985)015<1532:DDIPIL>2.0.CO;2.
McDougall, T. J., 1985b: Double-diffusive interleaving. Part II: Finite amplitude steady state interleaving. J. Phys. Oceanogr., 15, 1542–1556, https://doi.org/10.1175/1520-0485(1985)015<1542:DDIPIF>2.0.CO;2.
Moffat, C., and M. Meredith, 2018: Shelf-ocean exchange and hydrography west of the Antarctic Peninsula: A review. Philos. Trans. Roy. Soc. A., 376, 20170164, https://doi.org/10.1098/rsta.2017.0164.
Moffat, C., B. Owens, and R. C. Beardsley, 2009: On the characteristics of Circumpolar Deep Water intrusions to the west Antarctic Peninsula Continental Shelf. J. Geophys. Res., 114, C05017, https://doi.org/10.1029/2008JC004955.
Naveira Garabato, A. C., K. L. Polzin, B. A. King, K. J. Heywood, and M. Visbeck, 2004: Widespread intense turbulent mixing in the Southern Ocean. Science, 303, 210–213, https://doi.org/10.1126/science.1090929.
Orsi, A. H., T. Whitworth, and W. D. Nowlin, 1995: On the meridional extent and fronts of the Antarctic Circumpolar Current. Deep-Sea Res. I, 42, 641–673, https://doi.org/10.1016/0967-0637(95)00021-W.
Pacanowski, R. C., and S. G. H. Philander, 1981: Parameterization of vertical mixing in numerical models of tropical oceans. J. Phys. Oceanogr., 11, 1443–1451, https://doi.org/10.1175/1520-0485(1981)011<1443:POVMIN>2.0.CO;2.
Pedlosky, J., 1987: Geophysical Fluid Dynamics. 2nd ed. Springer-Verlag, 710 pp.
Pelland, N. A., C. C. Eriksen, and C. M. Lee, 2013: Subthermocline eddies over the Washington continental slope as observed by seagliders, 2003–09. J. Phys. Oceanogr., 43, 2025–2053, https://doi.org/10.1175/JPO-D-12-086.1.
Phillips, N. A., 1954: Energy transformations and meridional circulations associated with simple baroclinic waves in a two-level, quasi-geostrophic model. Tellus, 6, 273–286, https://doi.org/10.1111/j.2153-3490.1954.tb01123.x.
Poulin, F. J., A. Stegner, M. Hernandez-Arencibia, A. Marrero-Diaz, and P. Sangra, 2014: Steep shelf stabilization of the Bransfield coastal current: linear stability analysis. J. Phys. Oceanogr., 44, 714–732, https://doi.org/10.1175/JPO-D-13-0158.1.
Schmidtko, S., K. J. Heywood, A. F. Thompson, and S. Aoki, 2014: Multidecadal warming of Antarctic waters. Science, 346, 1227–1231, https://doi.org/10.1126/science.1256117.
Schofield, O., and Coauthors, 2007: Slocum gliders: Robust and ready. J. Field Robot., 24, 473–485, https://doi.org/10.1002/rob.20200.
Smith, D. A., E. E. Hofmann, J. M. Klinck, and C. M. Lascara, 1999: Hydrography and circulation of the West Antarctic Peninsula Continental Shelf. Deep-Sea Res. I, 46, 925–949, https://doi.org/10.1016/S0967-0637(98)00103-4.
Smith, K. S., 2007: The geography of linear baroclinic instability in Earth’s oceans. J. Mar. Res., 65, 655–683, https://doi.org/10.1357/002224007783649484.
Smith, K. S., and R. Ferrari, 2009: The production and dissipation of compensated thermohaline variance by mesoscale stirring. J. Phys. Oceanogr., 39, 2477–2501, https://doi.org/10.1175/2009JPO4103.1.
Smith, R. C., and Coauthors, 1995: The Palmer LTER: A long-term ecological research program at Palmer Station, Antarctica. Oceanography, 8, 77–86, https://doi.org/10.5670/oceanog.1995.01.
Spence, P., S. M. Griffies, M. H. England, A. M. C. Hogg, O. A. Saenko, and N. C. Jourdain, 2014: Rapid subsurface warming and circulation changes of Antarctic coastal waters by poleward shifting winds. Geophys. Res. Lett., 41, 4601–4610, https://doi.org/10.1002/2014GL060613.
Spence, P., R. M. Holmes, A. M. Hogg, S. M. Griffies, K. D. Stewart, and M. H. England, 2017: Localized rapid warming of West Antarctic subsurface waters by remote winds. Nat. Climate Change, 7, 595–603, https://doi.org/10.1038/nclimate3335.
Stern, A. A., D. M. Holland, and L. P. Nadeau, 2015: Instability of ocean jets along the Antarctic continental slope. J. Phys. Oceanogr., 45, 2315–2338, https://doi.org/10.1175/JPO-D-14-0213.1.
Stewart, A. L., and A. F. Thompson, 2015: Eddy-mediated transport of warm Circumpolar Deep Water across the Antarctic Shelf Break. Geophys. Res. Lett., 42, 432–440, https://doi.org/10.1002/2014GL062281.
Stewart, A. L., A. Klocker, and D. Menemenlis, 2018: Circum-Antarctic shoreward heat transport derived from an eddy- and tide-resolving simulation. Geophys. Res. Lett., 45, 834–845, https://doi.org/10.1002/2017GL075677.
St-Laurent, P., J. M. Klinck, and M. S. Dinniman, 2013: On the role of coastal troughs in the circulation of warm Circumpolar Deep Water on Antarctic shelves. J. Phys. Oceanogr., 43, 51–64, https://doi.org/10.1175/JPO-D-11-0237.1.
Tennekes, H., and J. L. Lumley, 1972: A First Course in Turbulence. 1st ed. MIT Press, 300 pp.
Thompson, A. F., S. T. Gille, J. A. MacKinnon, and J. Sprintall, 2007: Spatial and temporal patterns of small-scale mixing in Drake Passage. J. Phys. Oceanogr., 37, 572–592, https://doi.org/10.1175/JPO3021.1.
Turner, J., T. Maksym, T. Phillips, G. J. Marshall, and M. P. Meredith, 2013: The impact of changes in sea ice advance on the large winter warming on the western Antarctic Peninsula. Int. J. Climatol., 33, 852–861, https://doi.org/10.1002/joc.3474.