Freshwater distributions and water mass structure in the Amundsen Sea Polynya region, Antarctica

We present the first densely-sampled hydrographic survey of the Amundsen Sea Polynya (ASP) region, including a detailed characterization of its freshwater distributions. Multiple components contribute to the freshwater budget, including precipitation, sea ice melt, basal ice shelf melt, and iceberg melt, from local and non-local sources. We used stable oxygen isotope ratios in seawater (δ18O) to distinguish quantitatively the contributions from sea ice and meteoric-derived sources. Meteoric fractions were high throughout the winter mixed layer (WML), with maximum values of 2–3% (±0.5%). Because the ASP region is characterized by deep WMLs, column inventories of total meteoric water were also high, ranging from 10–13 m (±2 m) adjacent to the Dotson Ice Shelf (DIS) and in the deep trough to 7–9 m (±2 m) in shallower areas. These inventories are at least twice those reported for continental shelf waters near the western Antarctic Peninsula. Sea ice melt fractions were mostly negative, indicating net (annual) sea ice formation, consistent with this area being an active polynya. Independently determined fractions of subsurface glacial meltwater (as one component of the total meteoric inventory) had maximum values of 1–2% (±0.5%), with highest and shallowest maximum values at the DIS outflow (80–90 m) and in iceberg-stirred waters (150–200 m). In addition to these upwelling sites, contributions of subsurface glacial meltwater could be traced at depth along the ~ 27.6 isopycnal, from which it mixes into the WML through various processes. Our results suggest a quasi-continuous supply of melt-laden iron-enriched seawater to the euphotic zone of the ASP and help to explain why the ASP is Antarctica’s most biologically productive polynya per unit area.


Introduction
The Amundsen Sea Polynya (ASP; Figure 1) is the most biologically productive polynya (per unit area) in Antarctica (Arrigo and van Dijken, 2003). At the same time, many of the ice shelves of the Amundsen Sea embayment are experiencing basal melt rates that are high enough to drive ice shelf thinning (Paolo et al.,

Figure 1
ASPIRE station locations and regional bathymetry.
Locations of CTD observations acquired during ASPIRE (NBP10-05), overlain on regional bathymetry (blue shading and dark gray contours) (Nitsche et al., 2007) with coastlines and grounding lines from MOA2009 (Haran et al., 2014). The light gray line shows the NBP10-05 cruise track between stations. Freshwater distributions and water mass structure in the ASP region 3 Elementa: Science of the Anthropocene • 3: 000065 • doi: 10.12952/journal.elementa.000065 tides) (Robertson, 2013). We note that the ASPIRE's sampling program began when near bottom temperatures in front of the DIS were near the average reported for the 1.2-year mooring deployment ( Figure 2).
The DIS is estimated to be melting at an average rate of 7.8 m yr −1 , while the Pine Island, Thwaites and Crosson ice shelves to the east are melting at 16.2, 17.7 and 11.9 m yr −1 , respectively . This more rapid melting to the east is partly attributed to a less-modified (i.e. warmer) CDW entering the eastern trough system leading to these other ice shelves , as well as to differences in subice shelf circulation ( Jenkins et al., 2010;Joughin and Alley, 2011). The bathymetry of the Dotson trough may also limit the amount of CDW accessing the DIS ice shelf cavity, given the narrowing of the trough north-northeast of Martin Peninsula and the relatively shallower trough depths (Nitsche et al., 2007).
The ASP area is bounded to the east by the Thwaites Fast ice and Iceberg Tongue (TFIT), to the south by the coast (including DIS and GIS), and to the north and west by sea ice (Figure 1 inset). Similar to the Mertz Glacier Polynya (Massom et al., 2001), and other coastal polynyas with fast/land ice barriers to their east (Fraser et al., 2012;Nihashi and Ohshima, 2015), the ASP typifies a barrier coastal polynya. Prevailing southeasterly winds blow over the TFIT to keep the ASP ice-free. The ASP is thus a perennially occurring latent heat polynya (Arrigo et al., 2012), though there may be a sensible component to the polynya heat budget as well . A relatively small portion appears to remain ice-free in winter (< 10,000 km 2 ), but in November it begins to expand, reaching a mean maximum opening in February (> 60,000 km 2 ), after which it rapidly and consistently closes in March. At its spring-summer extent, the open waters of the ASP cover a wide range of depths, including the shallow bank area north of Bear Peninsula (~ 150-300 m) and the deep trough area (~ 600-1200 m) underlying the central ASP.
Until recently, there was a paucity of data from the ASP area, with attention instead given to the coastal area and/or the outer shelf area (Wåhlin et al., 2010;Arneborg et al., 2012;Jacobs et al., 2012;Wåhlin et al., 2012Wåhlin et al., , 2013. Here, we summarize the first densely-sampled hydrographic survey of the ASP (Randall-Goodwin, 2012), including a detailed assessment of freshwater distributions. We first examine the properties, spatial extent and circulation of water masses in the ASP as observed during ASPIRE. We then analyze the vertical and spatial distributions of meteoric and sea ice melt fractions determined from stable oxygen isotope (d 18 O) measurements. These distributions are compared to fractions of subsurface (below the mixed layer) glacial meltwater estimated from profiles of temperature, salinity and dissolved oxygen. We then discuss the relative contributions and delivery of freshwater inputs to the ASP, including the importance of the DIS outflow, its circulation in the ASP, and the role of drifting icebergs ubiquitous in the ASP region in mixing melt-laden CDW up into the mixed layer.  (Ducklow et al., 2015). Trap was located at 73.82°S and 113.07°W (bottom depth 785 m). Also shown are the upper and lower temperature ranges observed by the Jacobs mooring (deployed at 74.06°S and 112.42°W during 2006-07, sensor depth 851 m, ranges based on Figure 7 in Jacobs et al., 2012). This time series shows considerable temporal variability, from weekly to seasonal. The latter differs somewhat from the seasonal variability observed at 851 m by the Jacobs mooring in 2006Jacobs mooring in -2007

Collection and sources of data for ASPIRE
An intensive hydrographic, biological and biogeochemical survey of the ASP area was conducted during a 27-day cruise in mid-December 2010 to early January 2011 aboard the RVIB Nathaniel B. Palmer (NBP10-05). The cruise track crossed bathymetric gradients and included a high-resolution survey of the DIS front ( Figure 1). The ASP hydrographic data comprised 86 conventional Conductivity-Temperature-Depth (CTD) casts, 31 Trace Metal Clean (TMC) CTD casts, and 57 Expendable Bathythermograph (XBT) profiles. At many stations (Figure 1), multiple CTD and TMC casts were acquired, but at every station there was at least one full-depth profile taken within 10 m of the bottom (determined with a Benthos PSA-916 altimeter and a Sea-Bird Electronics (SBE) bottom contact switch with a 10 m line).
The CTD and TMC packages included a SBE 911plus system on 24-and 12-bottle SBE32 rosettes, respectively, each equipped with dual temperature/conductivity sensors and 1 (TMC) or 2 (CTD) SBE43 oxygen sensors. Sensors were calibrated before and after the cruise and post-processed following SBE protocols. All post-cruise corrections were negligible and within the reported accuracy of the SBE temperature and conductivity sensors (0.001°C and 0.0003 S m −1 , respectively). During the cruise, CTD salinity measurements were also validated against discrete salinity samples measured on a Guildline Autosal salinometer, while the oxygen sensor data were independently calibrated against discrete samples taken for oxygen titration using the Winkler method (Knap et al., 1996).
Ocean currents were measured with a Teledyne RD Instruments 'Ocean Surveyor' 38-kHz ship (hull-mounted) Acoustic Doppler Current Profiler (SADCP). The SADCP data were calibrated and corrected post-cruise by Eric Firing and Jules Hummon at University of Hawaii. Bathymetry data were acquired by a Simrad EM120 Multibeam system aboard the NBP. This system was also used to estimate the DIS draft during an alongshelf survey. On December 17 from 0900 to 1400 GMT the NBP performed a transect along the DIS front, with the ship maintaining a distance of ~ 200 m from the ice shelf front to record the approximate draft of the DIS, corresponding bathymetry and ocean current and thermal structure (the latter by deploying XBTs every 2-4 km). Underway measurements continuously recorded surface properties, and ice shelf freeboard was determined by several sextant observations. We also acquired 4 TMC and 1 CTD casts at select locations, including the eastern-and western-most locations and over the deepest portion of the central trough.
The Polar Geospatial Center at University of Minnesota and the National Snow and Ice Data Center provided visible, infrared and microwave satellite data. The image data used in this study are from the MODIS (Moderate Resolution Imaging Spectroradiometer) 250 m-resolution instrument on the NASA Terra and Aqua satellites. Those data, when cloud-free, provided excellent sea ice coverage, polynya boundaries, and locations of icebergs.

Collection and analysis of oxygen isotope samples
Samples for d 18 O analysis were drawn from CTD Niskin bottles into 20-mL glass scintillation vials with poly-seal cone liner caps that were sealed further with vinyl electrical tape. A total of 384 samples were drawn from 44 CTD casts acquired at 35 stations. (Sample depths and station locations are presented as part of the analyses under Results.) Typically 6-12 depths were sampled from each cast. The samples were boxed and kept cool during transport from Antarctica to the U.S., where they were analyzed at the Department of Earth and Planetary Sciences, Rutgers University.
At Rutgers University, the H 2 18 O/H 2

16
O ratio measurements were made on a FISONS OPTIMA Mass Spectrometer equipped with a MicroMass Multiprep automatic sample processing system after water sample equilibration with CO 2 using standard methods (Epstein and Mayeda, 1953;Fairbanks, 1982). Oxygen isotope ratios are reported as the per mil (‰) deviation of the 18 O/ 16 O ratio of the sample from that of Vienna Standard Mean Ocean Water (VSMOW) (Coplen, 1994). Precision is estimated to be ±0.034‰ (1 SD) as determined by multiple (n = 12) daily analyses of a laboratory standard. All samples were run in duplicate and replicates were required to measure within 0.068‰ to be included in the finalized data set. Instrument linearity and accuracy were determined by comparison of the laboratory standard to NIST reference standards: VSMOW, GISP, and SLAP. The d 18 O depth-discrete data were then plotted to visually inspect for obvious outliers (i.e., any values lying well outside the range described by all of the d 18 O data), which resulted in the identification of 6 individual outliers from 6 different depths in 5 different CTD casts. There were no anomalous departures in the sensor or bottle data associated with these 6 anomalous d 18 O values, thus we suspected that theses d 18 O samples must have been contaminated. These six values were removed from further analysis.

Determination of meteoric and sea ice melt fractions
The primary aim of this paper is to characterize the freshwater inputs that create the stratified water column found within the ASP region. In this context, it is possible to quantify the overall freshwater prevalence at any point by simple comparison of its measured salinity with that of the oceanic source water, i.e., the Freshwater distributions and water mass structure in the ASP region 5 Elementa: Science of the Anthropocene • 3: 000065 • doi: 10.12952/journal.elementa.000065 regional CDW. However, this approach does not provide information on the source of the freshwater added; to elucidate sources, additional tracers are required.
For the ASP region, the freshwater sources include meteoric water, i.e., freshwater inputs derived from the atmosphere. These inputs can be delivered as direct precipitation to the ocean, as snow falling onto sea ice that later melts, or as snow that accumulates to form ice sheets and ice shelves, ultimately to be injected into the ocean as glacial discharge. Additionally, within the active ASP, sea ice growth and decay seasonally redistribute freshwater vertically (either by mixing down surface sources of freshwater via brine rejection during sea ice growth in autumn-winter or by retaining freshwater sources in shallow surface layers during spring-summer sea ice melt). The advection of sea ice into or out of an area can also lead to net increases or decreases of sea ice melt, respectively. Finally, the presence of seasonal sea ice modulates the injection of meteoric water to the ocean when sea ice melts each spring. These are the processes that we quantify herein.
We focused initially on partitioning the net freshwater balance between the total inputs from all meteoric sources (precipitation and glacial discharge in the form of melting ice shelves and icebergs) and the input/ extraction due to sea ice melt/formation. Following Meredith et al. (2008Meredith et al. ( , 2010Meredith et al. ( , 2013, we used a threeendmember mass balance approach (Equation 1) (Östlund and Hut, 1984), employing salinity and d 18 O as tracers. This approach assumes that the observed salinity and d 18 O values in the water column have resulted from a mixture of sea ice melt, meteoric inputs and CDW, all three of which have well-separated salinity and d 18 O values in pure form.
In this set of three equations and three unknowns, f is the fraction of the freshwater components in pure form, denoted by subscripts for sea ice melt (sim), meteoric water (met), and cdw, while subscripted S and d are the salinity and d 18 O values of the same three components in pure form. The quantities S obs and d obs on the right-hand side are the observed values of salinity and d 18 O in each water sample.
We assigned values of 34.62 and −0.059‰ for the CDW salinity and d 18 O endmember, respectively, based on our observations of CDW in the Dotson trough ( Figure 3A). For sea ice, we used values of 7 and 2.1‰ for salinity and d 18 O, respectively. This choice was based on our own unpublished observations of salinity and d 18 O measurements in sea ice from a previous expedition to the Amundsen Sea ( January-February 2009); these values are also consistent with Fransson et al. (2011) and Meredith et al. (2013). Given known variability in CDW properties at depth in the Dotson trough  and in age/thickness of sea ice, these endmember values will vary seasonally and yearly. However, as discussed below, the uncertainty in the meteoric endmember is much greater, and will thus dominate the uncertainty in our freshwater fractions.
For the meteoric endmember, we assigned values of 0 and −25‰ for salinity and d 18 O, respectively. This choice was based on the distribution of our data in d 18 O-salinity space (shown and discussed below). However, as discussed by Meredith et al. (2008Meredith et al. ( , 2010Meredith et al. ( , 2013, the greatest uncertainty in this estimate lies in the d 18 O meteoric endmember (whereas the endmembers for CDW and sea ice are relatively well-constrained by observation). We therefore bracketed our results using d 18 O endmembers of −20‰ and −30‰ (i.e., ±5‰) based on values previously reported for the southern Bellingshausen Sea  and eastern Amundsen Sea (Hellmer et al., 1998;Jenkins, 1999), respectively. Given this range in the d 18 O endmember, the uncertainty in the derived meteoric and sea ice melt percentages (which includes uncertainty in measurement error), is no more than ±0.5%.
Once meteoric and sea ice melt fractions were determined, column inventories were computed based on vertical interpolations of the d 18 O depth-discrete profiles. Various interpolations were tested, including a shape-preserving piecewise cubic spline, but we elected to use a simple linear interpolation as it yielded the most conservative estimate. Before performing the interpolation, some d 18 O profiles had to be amended as follows. For those few casts where samples were not drawn from the near surface bottle, a surface value was assigned that matched the next sampled depth (i.e., within the surface mixed layer). There also were a few casts where the bottom bottle was not sampled; in those cases, a very conservative value of zero was assigned for d 18 O of the bottom depth. Uncertainty in the column inventories resulting from the uncertainty in the freshwater percentages was approximately ±2 m.
We note that the choice of representative "regional" properties for the CDW and other endmembers in our analyses is a subjective one that influences our results. For example, choosing on-shelf CDW properties for the endmember means that saltier waters found on the continental slope will have a negative freshwater fraction. However, our main interest is in the ASP region, so we focused on the freshwater inputs that create the observed stratification in that region, noting that there will be contributions both from local freshwater inputs and from the transport of waters with a non-zero freshwater fraction into the region as well.
Freshwater distributions and water mass structure in the ASP region

Determination of subsurface glacial meltwater fractions
Having quantified the freshwater inputs due to meteoric water and sea ice processes, we next sought to distinguish meteoric inputs from subsurface melting of glacial ice, as these deep inputs to CDW result in buoyancy gain and can drive upwelling or formation of a melt-laden CDW layer available to deep mixing. Because subsurface glacial melt is a meteoric input, d 18 O cannot be used to distinguish it from surface-derived meteoric inputs. For example, isotopic values in local precipitation and ice shelf melt are similar, unless the interior parts of the catchment basin for the ice shelf are remote from the coast and at high elevation, which they are not in the case of DIS. However, subsurface inputs of glacial meltwater can be distinguished from other freshwater sources by their low potential temperature (because the latent heat of fusion is extracted from the ocean) and high dissolved oxygen concentration (because all the air trapped in the ice goes into solution).
To calculate glacial meltwater fractions below the wind-mixed layer, Jenkins (1999) used a similar approach to that in the previous section, but here we are using potential temperature, salinity and dissolved oxygen as tracers, and CDW, Winter Water (WW) and glacial meltwater as endmembers. Using WW as an endmember effectively excludes those surface inputs of freshwater that create the WW properties, so the results of these calculations should be seen as the component of the total meteoric water input that has come from deep melting and upwelling versus that derived from surface input and convective mixing beneath growing sea ice. The use of three tracers with three endmembers can be solved algebraically in a number of ways. Jenkins (1999) used three separate estimates of meltwater fraction based on the three possible tracer pairs, and formulated the algebraic solution to the equivalent of Equation (1) as a conservation equation for a composite tracer defined by: The χ denotes a CTD sensor observation at each depth, and superscripts 1 and 2 represent the conservative tracers (potential temperature, salinity, or dissolved oxygen); the subscripts, as before, denote water masses WW and CDW. The subsurface glacial meltwater fraction is then calculated by The numerator in Equation (3) is the composite tracer observed at each depth of the CTD cast, while the denominator is the value for pure glacial meltwater (gmw), where we used −89°C, 0 and 25.9 mL L −1 for potential temperature, salinity and dissolved oxygen, respectively, based on previous studies in the Amundsen Sea ( Jenkins, 1999;Jacobs et al., 2011) and on the extrapolation of mixing lines in property-property space. The composite tracer values for CDW (0.75°C, 34.62, 4.52 mL L −1 ) and WW (−1.81°C, 34.13 and 6.49 mL L −1 ) for potential temperature, salinity and dissolved oxygen, respectively, were identified from observation as "core" properties of the warmest, saltiest water mass in the deep trough on the shelf and the coldest, saltiest water mass in the winter mixed layer, respectively ( Figure 3A). For additional details on the fundamentals of mixing and endmember choices within the context of ocean-ice shelf interactions in the Amundsen Sea, see Jenkins (1999) and references therein. In our profile plots of glacial meltwater fraction, we considered results from all three composite tracers (i.e., O 2 -θ, O 2 -S and θ-S; see Results). Below the wind-mixed layer (distinct from and shallower than the WML), the three composite tracers agreed well, but in the wind-mixed layer they became non-conservative. There the curves for O 2 -θ and θ-S diverged rapidly from each other due to the impact of interactions with the atmosphere on all tracers. For this study, the depth of surface influence was identified as the level where O 2 -θ and θ-S deviated more than 8‰ in glacial meltwater fraction, and values shallower than that level were ignored.
As discussed by Jenkins and Jacobs (2008), the uncertainty in the glacial meltwater fractions derives from the uncertainty in defining linear trends associated with mixing in the main pycnocline and between CDW and a meltwater source that is assumed to be uniform. The assumption of linearity completely breaks down in the surface mixed layer, where temperature and dissolved oxygen signals are influenced by equilibration with the overlying atmosphere, so these values were ignored. Below this level the main source of error in the calculation is the deviation from the assumed linear trend between CDW and WW properties ( Figure 3A), which introduces a bias of up to 5‰ (i.e., 0.5%). It is possible that the curvature is a result of sub-surface meltwater addition, but it is also possible that some is an artifact introduced by the spatial and temporal variability in the properties of the WW that is incorporated into the thermocline. Therefore, precise error estimates are difficult to provide, but we expect them to be of order 0.5% or less.

Regional hydrography and ocean circulation
Water mass structure In the ASP, there are two primary water masses: warm, saline CDW at depth and an overlying deep layer of cold, relatively less saline winter water (WW) (Figures 3A and 4). During spring-summer, there is also a relatively thin surface layer of Antarctic Surface Water (AASW). This summer surface layer is essentially WW that has warmed and freshened to varying degrees by insolation and seasonal ice melt. Hydrographic sections of temperature (shading) and density (contoured), south-north along the Dotson trough ( Figure 4A) and crossing the trough west-to-east (Figure 4 B-D), revealed the spatial distributions of AASW, WW and CDW in the ASP region observed during ASPIRE. Relative to offshore ( Figure 4A, right) CDW on the shelf was cooler and fresher ( Figure 4A, left). Nonetheless, within the trough at depth, CDW was still more than 2 degrees above freezing when it reached the DIS ( Figure 3B).
The base of the WML, here defined by the ~ 27.5 kg m −3 isopycnal, was deep, ranging from ~ 300 m in shallow waters to greater than 400 m over the deep trough. In the southern trough area (south of ~ 73.5°S), the WML attenuated (shoaled and warmed) (Figure 4 A, C, left) due to strong vertical mixing, detected both in front of the DIS and in the vicinity of a drifting iceberg (detailed below). South of ~ 73.25°S a thin layer of AASW corresponded to the open waters of the ASP (Figure 1, inset).
Mixing lines between water masses were identified in potential temperature-salinity (θ-S) space ( Figure 3A). Mixing between CDW and WW and between CDW and glacial meltwater ( Jenkins, 1999) illustrate the degree of CDW modification observed in the ASP region. For the ASPIRE data, the meltwater-CDW mixing line was an extrapolation from the observed θ-S value for CDW (as observed at depth in the Dotson trough) to the θ-S value for glacial meltwater (−89°C and 0, respectively), taking into account the latent heat of fusion of the ice ( Jenkins, 1999). This extrapolation yielded a slope of 2.59, similar to other slope values reported for the Amundsen Sea (e.g., Jacobs et al., 2011). The meltwater-CDW mixing line is used here to provide a visual reference for detecting subsurface glacial meltwater presence.
When viewed in θ-S space ( Figure 3A), all of the ASPIRE profiles approached the CDW-meltwater mixing line (upper dotted line) at depth. In contrast, there is a general lack of data on the CDW-WW mixing line, which is due to the fact that few ASPIRE observations represent mixtures of core CDW and WW properties (as defined earlier in the previous section). Instead, the range of observations to the left of the CDW-WW mixing line reveals varying mixtures of modified versions of CDW and WW (e.g., cooler, less salty CDW mixing with warmer, less salty WW) in addition to some amount of subsurface glacial meltwater. The profile highlighted in dark blue, acquired at the shelf break (Station 4), smoothly deviated from the CDW-meltwater mixing line at depth to the CDW-WW mixing line at mid-depths. The profile acquired in front of the GIS (Station 5, cyan) was similar, though with less bending towards the CDW-meltwater mixing line at mid-depths. Most of the other ASPIRE profiles showed layered-like deviations extending towards the CDW-meltwater mixing line at mid-depths, similar to the profile highlighted in green (discussed below).
The two profiles highlighted in red (in Figure 3) are Station 60 (upper red line) and 61 (lower red line). The profile from Station 60 (red circle in Figure 1) closely follows the CDW-meltwater mixing line and was located at the western end of the DIS (just west of Station 10) in 635 m of water and in an area of strong outflow (detailed below). Station 61 (second red circle just north of Station 60 in Figure 1) was north of the outflow area in somewhat shallower waters (520 m); it too showed melt-laden CDW, more so at depth, with less upwelling near the surface (as otherwise indicated by the linear θ-S profile for Station 60). The other profiles in front of the DIS, east of the outflow, also showed melt-laden CDW at depth, as well as deviations  Figure 1; Figure 5A).
Station 57 was designated the 'iceberg station' and was located approximately 30 km north of the DIS within the 700 m isobath of the deep trough. The trajectory of the iceberg was from the southeast to the northwest over the locations occupied by Station 57 ( Figure 5A). The estimated drift speed was ~ 0.3 knots under light southerly winds. The iceberg was quasi-tabular, ~ 3 km long by ~ 1 km wide, with an irregular freeboard range of 37-50 m (determined by sextant) and an estimated keel depth range of 340-450 m (based on isostasy).
At Station 57, several CTD casts were taken in the vicinity of the drifting iceberg over a period of ~ 2.5 days, both downstream and upstream of its drift trajectory, i.e., in waters perturbed and undisturbed by the drift of the deep-keeled iceberg. Profiles taken in the perturbed waters downstream and closest to the iceberg were similar (e.g., profiles from s57.26 and s57.30 in green and dark green, respectively; Figure 5B), while profiles taken in the undisturbed waters upstream or away from the iceberg (e.g., profiles from s57.32 and s57.35 in blue and black, respectively) showed no deviations in meltwater loading.

Ocean currents and thermal anomalies
Depth-averaged (31-423 m) ocean currents acquired underway ( Figure 6) revealed the dominant flow direction as observed during our occupation of the ASP region. We later show a subset of velocity depth profiles, but here our intent is to show the dominant flow directions over the upper ~ 400 m where variability in the vertical was relatively low (< 3 cm s −1 SD). In particular, there was a strong northward flow (> 20 cm s −1 )

Figure 5
Iceberg drift track and mixing of temperature and salinity. north of the DIS outflow and extending north along the western bank of the deep trough. There were other areas of strong northward flow further north and to the east, as well as a few areas of strong southward flow. The SADCP data were not de-tided, and we only show those depth-averaged currents (tidal or non-tidal) greater than 5 cm s −1 for clarity. It is certain that some of the flows shown here ( Figure 6) contained shear, were tidally amplified and/or generated by transient eddies, all of which are important factors in terms of mixing and melting (e.g., Robertson, 2013). In terms of current direction, however, many locations had overlapping cruise tracks separated in time, and several (though not all) of those revisited locations showed similar directions, most notably the strong outflow area in front of the DIS and the northward flows observed during the crossings onto/off the shallow bank to the northeast of the DIS.
A more detailed examination of ocean currents just north of the DIS ( Figure 7A) showed strong northward and southward flows (> 10 cm s −1 ) concentrated at the western and eastern boundaries of the ice shelf front, respectively, while weaker north-south currents were observed elsewhere along the front. (Here we did not exclude the weaker currents, but we did smooth with a 3-point running mean.) The strong southward (inflow) current of 10-20 cm s −1 observed at the far eastern boundary was concentrated at depths below ~ 400 m (and likely extended beyond the SADCP depths of 500-600 m). In contrast, the strong northward (outflow) current observed at the far western boundary was considerably shallower and stronger (> 20 cm s −1 ) between ~ 100 and 400 m. We estimated the magnitude of the outflow by multiplying its mean (northward) flow (~ 0.20 m s −1 ) by its estimated depth/width (300 m deep by 2760 m wide) (e.g., based on Figure 7A) and arrived at ~ 0.17 Sv (or 0.17 x 10 6 m 3 s −1 ). The width of the outflow may actually be greater towards the west, but unchartered (and shallow) waters prevented us from venturing further in that direction.
The current structure in the outflow area of the DIS compared well with its thermal structure ( Figure 7B), the latter acquired from 17 XBTs spaced every 2-4 km. There was a distinct shoaling of the thermocline (black contour representing the −1°C isotherm) in the outflow area, corresponding also to a thinning of the ice shelf towards the western end (as estimated from multibeam data). The observed temperature anomalies (temperature at a given depth minus a depth-mean temperature) showed a warm thermal anomaly in the same depth interval where the strongest outflow was observed. The temperature of the outflow waters was ~ 1.0-1.5°C warmer than the in-situ freezing point and consisted of melt-laden CDW as indicated by its profile in θ-S space ( Figure 3B) and its meltwater content (shown and discussed below).

Regional oxygen isotope values
Reflecting the contribution from the meteoric endmember (−25‰) relative to the endmembers for CDW (−0.059‰) and sea ice (2.1‰), the lowest d 18 O values (< −0.6‰) were found in thin surface layers over the deep trough and south towards the DIS (Figure 8), suggesting local surface freshwater inputs from the TFIT and DIS and the import of surface freshwater from further east, upstream of the DIS (e.g., the Pine Island Bay area). The surface freshwater inputs come from solar-warmed waters inducing surface melt of icebergs, ice shelves and other sources of precipitation (snow on sea ice).
Low values of d 18 O (−0.4 to −0.6‰) were also observed throughout the deep winter mixed layers, indicating redistribution of meteoric surface inputs and/or contributions from deep melting and upwelling     (Figure 8D), the former caused by the upwelling of melt-laden CDW. This profile stands in contrast to the cast acquired at the mid-ice-shelf location (Station 11; Figure 9A, red plus symbol), which showed much lighter isotopic values throughout its mixed layer. The profile near the GIS (Station 5; Figure 9A, cyan) also showed lighter isotopic values throughout its mixed layer. The cast from the iceberg station (Station 57.26; Figure 9A, green) showed a deviation towards heavier isotopic water at ~ 200 m, where mixing by the iceberg caused entrainment of CDW, consistent with the deviations detected in θ-S space ( Figures 3A and 5B).
When plotted in d 18 O-S space ( Figure 9B), many of the surface (light gray) and mixed layer (dark gray) values deviated from the meteoric-CDW mixing line (bold diagonal). Deviations left or right of the meteoric-CDW mixing line are due to net sea ice melt (freshening) or formation (salinification), respectively. Surface values showing net sea ice melt tended to be from locations near or in the sea ice bordering the polynya to the north, i.e., not to the south where sea ice production is most active due to persistent cold southerly winds that create conditions favorable for new ice formation and northward advection. Consequently, in the northern area of the polynya there is an influx of sea ice from the south, which contributed to either less net freezing or net sea ice melt (as indicated by the surface values to the left of the meteoric-CDW mixing line). In contrast, surface samples from the southern polynya indicated net sea ice formation, as do most of the winter mixed layer values, consistent with this area being an active latent heat polynya under the influence of persistent southeasterly winds throughout the year. Details of these regional distributions are discussed below.  observed elsewhere (i.e., > 1.5%), with column inventories averaging around 8 m. In contrast, where upwelling or mixing of isotopically-heavier CDW into the WML occurred, as at the DIS outflow (Station 60; Figure 10A) and at the iceberg 'mixed' stations ( Figures 10 E-G), meteoric fractions decreased and water column inventories of total meteoric water were slightly lower.

Fractions of meteoric, sea ice melt and subsurface glacial meltwater
Consistent with the distribution of surface values in d 18 O-S space discussed above ( Figure 9B), a few locations exhibited positive sea ice melt fractions at the surface (Figure 10 J-K and O-P), indicating net sea ice melt. Otherwise throughout most of the ASP area, sea ice melt fractions were mostly negative throughout  the water column, indicating net salinification due to net (annual) sea ice production. Geographic distributions of meteoric water column inventories ( Figure 11A) showed that inventories were highest over the deep trough area, particularly in locations not experiencing upwelling or deep mixing. Inventories of sea ice melt were all negative and were highest in the mid-shelf trough area ( Figure 11B). Vertical profiles of the fractions of subsurface glacial meltwater (Figure 10, in black) also showed vertical and regional variability across the four representative areas. The effects of upwelling or mixing are clearly seen in the subsurface glacial meltwater profiles at all the DIS and iceberg-affected stations (Figure 10 A-G). In contrast, where there was little upwelling or mixing, a well-defined subsurface glacial meltwater maximum resided well below the surface layer along the 27.55-27.59 isopycnal (Figure 10 H, I-P; Figure 12B), which ranged in depth from ~ 600 m to 350 m ( Figure 12C). The highest glacial meltwater maxima were at the DIS outflow (2.1%; Figures 10A and 12A) and where there was upwelling or mixing (1.0-1.5%; Figures 10 B-D, E-H and 12A). Elsewhere, the glacial meltwater maxima were of moderate magnitudes (0.6-0.8%) ( Figure 12A).
The column inventories of subsurface glacial meltwater (Figures 10 and 12D) were considerably higher where there was active upwelling, e.g., in front of the DIS and at the iceberg affected area (Figure 10 A-G). Nonetheless, the lower and deeper values are also important, as they indicate a distinct subsurface melt-laden layer throughout the ASP region that is available to deep mixing. However, and by default, the column inventories of subsurface glacial meltwater content were of lesser magnitude than the column inventories of total meteoric content ( Figures 11A and 12 D) due to the fact that the subsurface column inventories were computed over a much smaller depth interval (i.e., from the subsurface black horizontal line in Figure 10 to the bottom of the cast).
We also note that the subsurface glacial meltwater estimates do not distinguish how much of the subsurface glacial meltwater originated from basal ice shelf melt or was generated locally by deep iceberg melt. Further, the portion of the subsurface glacial meltwater content originating in the DIS outflow that was detectable north of the DIS (by this method) was only that portion remaining below the mixed layer. Thus, we are unable to track upwelled glacial meltwater inputs (beyond the upwelled site) or the shallower inputs of glacial meltwater at the DIS outflow, except through their contributions to the total meteoric content. This issue is well illustrated by Figure 10, which shows the fractions of subsurface glacial meltwater in relation to the meteoric fractions (black and blue lines, respectively). Indeed, the portion of locally derived subsurface glacial

Figure 12
Regional and vertical distributions of subsurface glacial meltwater.
The subsurface glacial meltwater (A) maximum (‰), (B) σ θ isopycnal at the maximum, (C) pressure (dbar) at the maximum, and (D) column inventories (m) of the total glacial meltwater below the surface layer. Bathymetry is from Nitsche et al. (2007); coastlines are from MOA2004 (Haran et al., 2013). The four dotted lines in (A) are the station transects plotted in Figure 10. meltwater (e.g., from the DIS) that was mixed into the winter mixed layer is unknown by these approaches, but it is an inferred significant contributor to the relatively high meteoric content observed in the deep winter mixed layers. These interpretations, and the caveats of showing fractions of subsurface glacial meltwater along side meteoric fractions (Figure 10), are discussed in the next section. Finally, we note that the fractions of meteoric, sea ice melt and subsurface glacial meltwater are relative to the endmembers used in our calculations. Specifically, these estimates are relative to CDW observed in the deep trough on the shelf and relative to the region-wide values used for sea ice melt and meteoric inputs. One effect of this relativity is that positive (non-zero) meteoric and glacial meltwater values were found at depth for the profiles located away from the deep trough area (e.g., Figure 10 A, B, D, M-P), where a more modified CDW is isotopically lighter relative to its less modified value deep in the trough. We also note that the accumulation of freshwater in the deep WMLs has likely occurred over time scales longer than a single season and that this freshwater originated from both local and non-local inputs.

Assessing freshwater sources
We found that substantial meteoric water (7-13 m) accumulates in the deep WMLs of the ASP region. The largest amounts (11-13 m), found in front of the DIS and over the deep trough area, are approximately twice the amount reported for the rapidly-warming western Antarctic Peninsula region, where values of 4-6 m of meteoric water were found in the Marguerite Bay area (Meredith et al., 2010). The greater accumulation of meteoric water in the ASP region is made possible by WMLs being approximately twice as deep in the ASP area (as compared to Marguerite Bay) and by the large supply of meteoric water to this region.
The high accumulation of meteoric water in the deep WMLs is the net impact of both local and remote freshwater sources. Key to interpreting our results is the acknowledgement that the estimated freshwater content reflects contributions made over large spatial and temporal scales. That the sea ice meltwater fractions are largely negative, for example, indicates that this region is a net sea ice producer, even if some local sea ice melt had occurred during the few weeks prior to our sampling. Further, some of the accumulated meteoric water comes not only from basal ice shelf melt upwelled in front of the DIS, but also from basal ice shelf melt upwelled in front of ice shelves upstream (east) of the DIS (e.g., Crosson, Thwaites, Pine Island) (e.g., Jacobs et al., 2011;Mankoff et al., 2012), or from icebergs within the TFIT and elsewhere upstream. The glacial meltwater inputs upstream of the DIS are carried westward into the ASP region by the surface circulation (Assmann et al., 2005). Additional glacier meltwater comes from numerous icebergs drifting through the ASP, as well as from seasonal glacier melt at the air-ice-ocean boundary by open waters warmed by solar insolation. Added to these sources is the accumulation of precipitation inputs to the WML, both locally and remotely, the latter carried into the ASP by the surface circulation. In the overall freshwater budget for the ASP region, these local and non-local inputs are then partially compensated by the salinification due to net sea ice production, such as typifies this polynya area.
The subsurface glacial meltwater estimates as quantified in Figures 10 and 12 are those contributions from deep melting and upwelling to the total meteoric estimates as quantified in Figures 10 and 11. However, the glacial meltwater estimates from basal melt of ice shelves and icebergs can only be detected subsurface (with the approach used here). Thus, how much of the upwelled basal ice shelf and iceberg meltwater contributed to the total meteoric content is unknown; we infer that the contribution is significant based on the following.
The highest maximum meteoric concentrations, reaching 2.6%, were observed near the surface in front of the DIS outflow (Station 60) and at DIS Station 11 (mid-ice-shelf ) (Figure 10 A, C). Because these observations were coincident with supersaturated pCO 2 and low dissolved oxygen at the sea surface (Mu et al., 2014), they likely represent upwelled deep waters made buoyant by glacial meltwater. The highest subsurface glacier meltwater concentration was 2.1% and was observed at the DIS outflow (Station 60) at a depth of 82 m, i.e., just before the signal was lost to surface contamination. Although we cannot explicitly determine how much of the higher meteoric content at the surface was derived from upwelled basal ice shelf melt versus surface meltwater inputs, the dissolved gases corroborate that the major contributor is basal ice shelf melt. Rignot et al. (2013) estimated that the DIS discharges approximately 45.2 Gt of basal meltwater per year (as observed over 2003 to 2008). We estimated the western DIS outflow in December 2010 to be approximately 0.17 Sv (or 5.4 x 10 12 m 3 yr −1 ). Over the depth interval of this outflow (~ 100-400 m), the average glacial meltwater fraction is approximately 1.5%. This fraction equates to 8.1 x 10 10 m 3 yr −1 of pure meltwater in the DIS outflow, or 81 Gt yr −1 . This instantaneous estimate is nearly twice as high as the 5-year average from Rignot et al. (2013) and likely reflects seasonal and yearly variability inferred from CDW temperature observations ( Figure 2). However, the CDW temperature in December 2010 was near average for the year of deployment; thus, this higher estimate could also reflect an overall increase in discharge. Volume losses from Antarctic ice shelves appear to be accelerating (Paolo et al., 2015), with losses from West Antarctic ice shelves increasing by 70% in just the last decade. Thus, the DIS may be contributing upwards of 81 Gt yr −1 of basal meltwater to the ASP region since ~ 2010. Elementa: Science of the Anthropocene • 3: 000065 • doi: 10.12952/journal.elementa.000065

Variability in freshwater inputs to the ASP and biological productivity
A key question is whether spatial or temporal variability in the supply of the glacial meltwater impacts biological productivity. The ASP is not only the most productive polynya (per square meter), but its productivity is also the most variable (Arrigo and van Dijken, 2003), raising the issue of whether this biological variability could be linked to variable meltwater inputs. Previous studies have shown considerable temporal variability in basal ice shelf melt in the Amundsen Sea embayment region linked to variability in CDW properties accessing the ice shelf cavities ( Jacobs et al., 2011Dutrieux et al., 2014). In turn, temporal variability in CDW properties on the continental shelf has been linked to variability in atmospheric forcing.
Atmospheric forcing in this region is largely driven by the Amundsen Sea Low (ASL), which migrates seasonally. In summer, with the ASL north of the continental shelf, the dominant winds at the shelf break are southeasterly winds. In late winter to early spring, the ASL shifts to the southwest, causing the dominant winds near the shelf break to become northwesterly (Steig et al., 2012;Turner et al., 2013). Thus, winterspring conditions tend to enhance CDW inflow onto the continental shelf (Thoma et al., 2008), through the strengthening of an undercurrent associated with the Antarctic Slope Front that transports CDW into the trough areas at the shelf break (Walker et al., 2013). The annual migration of the ASL, however, is highly variable and sensitive to atmospheric perturbations caused by the Southern Annular Mode (SAM) (e.g., Turner et al., 2013) and by tropical teleconnections such as the El Niño Southern Oscillation (ENSO) (e.g., Steig et al., 2012;Turner et al., 2013) and the Atlantic Multidecadal Oscillation (Li et al., 2014). These various factors contribute to the high seasonal and interannual variability of CDW moving onto the continental shelf Dutrieux et al., 2014). Time series of near-bottom temperatures from moorings in front of the DIS   (Figure 2) are also consistent with these observations.
Seasonal variability also occurs in other mechanisms that contribute freshwater to the deep WMLs of the ASP region. In fall-winter in the central ASP (i.e., distant from the variable but near continuous upwelling occurring at the DIS), inputs of subsurface glacial meltwater occur through entrainment from brine-induced mixing during sea ice production, whereas in spring-summer, they occur through solar surface warming. However, strong wind events during the spring-summer opening of the ASP could also potentially mix melt-laden CDW vertically, particularly where the subsurface glacial meltwater maximum shoals in the shallower areas, e.g., north of the DIS outflow and east over the shallow bank area ( Figure 12C). Coastalderived meltwater may also be continually mixed by tides (Robertson, 2013), as well as advected into the ASP by eddies generated in the coastal current, that in turn transport and diffuse the glacial meltwater offshore (Spall et al., 2008;St-Laurent et al., 2014).
The DIS outflow was enriched in dissolved iron (dFe) compared to the central polynya (Sherrell et al., 2015). The dFe enrichment is hypothesized to be derived from continental material released during basal ice shelf melt, but may also have been derived from sediments near the grounding line or where the outflow interacts with sediments along the western margin of the ice cavity (Sherrell et al., 2015). Elevated dFe at the DIS outflow and at specific depths in the WML throughout the ASP is consistent with the idea that the DIS outflow is a major contributor of Fe-enriched meltwater. A key question related to potential biological impacts is how much of the Fe-enriched meltwater exiting the DIS is delivered to the euphotic zone of the ASP, particularly to the shallow bank area, and by what mechanism.
The eastern bank area, comprising the central ASP, was characterized by high biological activity, with observed chlorophyll a values > 15 mg m −3 (versus < 15 mg m −3 over the deep trough areas) (Alderkamp et al., 2015). The areas with the highest primary productivity were accompanied by low surface CO 2 concentrations (< 170 ppm) and high surface oxygen (10.8 mL L −1 ), indicative of enhanced biological drawdown and production (Mu et al., 2015). To support such high productivity, the ASP must be fueled by a quasi-continuous supply of bioavailable Fe. As described above, this supply is likely accomplished by winter entrainment of subsurface Fe-enriched glacial meltwater, and more continuously by deep wind mixing and iceberg stirring, bringing subsurface meltwater to the euphotic zone over the course of the bloom period. Finally, eddy activity may transport additional Fe-enriched melt-laden waters from the coastal current northward into the central ASP. All of these delivery mechanisms are sensitive to climate change (as discussed next), with potentially contrasting influences on Fe availability and thus biological productivity.

A rapidly changing environment
The Amundsen Sea is changing rapidly in terms of losses of glacial ice mass along its coast (Pritchard et al., 2012) and of sea ice north of its coast ( Jacobs and Comiso, 1997;Stammerjohn et al., 2012Stammerjohn et al., , 2015. Sea ice changes over the period of 1979-2013 have resulted in a lengthening of the open water season by ~ 2.6 months in the ASP area, due primarily to an earlier spring opening by ~ 2.2 months . These changes allow more solar radiation and potentially more surface melt to enter the ASP, but also more wind mixing as well; these are competing factors that alter water column stratification and the delivery of Fe-enriched meltwaters to the euphotic zone, with consequent impacts on primary production.
In the Amundsen Sea, ice shelf thinning and grounding line retreat have been linked to an increasing presence of warmer CDW on the continental shelf, driving a more intensified circulation under the ice Elementa: Science of the Anthropocene • 3: 000065 • doi: 10.12952/journal.elementa.000065 shelves ( Jenkins et al., 2010;Jacobs et al., 2011;Dutrieux et al., 2014). Currently, estimates of basal ice shelf melt and calving fluxes in the Amundsen Sea embayment are some of the highest observed anywhere in coastal Antarctica Paolo et al., 2015). If the strengthened westerly winds (Bracegirdle, 2013) persist, as well as the increases in deep ocean heat content on the continental shelf (Martinson et al., 2008;Jacobs et al., 2011;Martinson, 2012), then there will likely be sufficient ocean heat delivered to the ice shelf cavities in the Amundsen Sea embayment to drive continued or even increased basal ice shelf melt and grounding line retreat. The Thwaites Glacier in particular is experiencing a rapid grounding line retreat , along with rapid changes to its glacier tongue (MacGregor et al., 2012). Given these changes, icebergs may become a more significant factor in distributing glacial meltwaters in the ASP.
These climate-sensitive meltwater changes in the ASP will continue to have a strong influence on the biogeochemistry and biology of the ASP, yet there are many unanswered questions. For example, how will the vertical distribution of freshwater change in the future, with increased meltwater inputs, both at depth and at the surface? What will be the response to increased winds versus increased surface freshwater inputs in regard to stratification and the availability of Fe? How will Fe-enriched meltwater sources and the delivery of Fe to the euphotic zone change? Progress in answering these questions is underway with ongoing studies focused on ocean-ice shelf interactions.

Summary
Here we present results from a densely-sampled hydrographic survey of the Amundsen Sea Polynya (ASP) region, including a detailed characterization of its freshwater distribution. Our survey extended from the Dotson Ice Shelf (DIS) north to the continental shelf break and included both the deep Dotson trough (~ 600-1200 m) that bisects the continental shelf and the shallow bank (150-300 m) east of the trough that lies adjacent to, and west of, the Thwaites Fast ice and Iceberg Tongue (TFIT). To spatially and vertically resolve different contributions to the freshwater budget, we first quantified inputs from sea ice melt and from meteoric-derived sources using stable oxygen isotope ratios in seawater (d 18 O). Meteoric sources include direct precipitation and glacial melt from both local and non-local sources. We then separately quantified glacial meltwater inputs entrained into Circumpolar Deep Water (CDW) at depth either from basal ice shelf melt or from iceberg melt. We summarize the results as follows.

•
A hydrographic survey along the length of the DIS front showed a melt-laden outflow of CDW emanating from under the DIS, with strongest flows concentrated at its western end.

•
Observed ocean currents showed northward flow of the melt-laden CDW along the deep trough region, as well as into the shallow ASP area (the former is consistent with previous observations of topographically steered meltwater outflow along troughs) (St-Laurent et al., 2013;Wåhlin et al., 2013). • Vertical distributions of meteoric fractions (following Meredith et al., 2013) showed maximum values ranging from 2 to 3% in the deep (300-400 m) winter mixed layers (WMLs), with highest fractions in the top 100-200 m. • Given the deep WMLs, column inventories of meteoric water were high, ranging from 10-13 m adjacent to the DIS and in the deep trough, and from 7-9 m in shallower areas; the former were twice those reported for continental shelf waters near the western Antarctic Peninsula. • Sea ice melt fractions were mostly negative, indicating net freshwater extraction (i.e., salinification) due to net (annual) sea ice production, consistent with this area being an active polynya.

•
Observations near a drifting iceberg showed vertical mixing of CDW, laden with meltwater, thus indicating that drifting icebergs, which frequent the ASP region (typically breaking free of the TFIT and then drifting westward across the ASP), are important mechanisms for vertically mixing melt-laden waters. • Away from the DIS and drifting iceberg, the fractions of subsurface glacial meltwater indicated a meltladen layer along the ~ 27.6 isopycnal; from this depth subsurface glacial meltwater can be entrained into the mixed layer each winter from brine-induced mixing during sea ice production, and each spring-summer through wind mixing, tidal mixing, iceberg stirring and eddies, the latter 'diffusing' the freshwater offshore (Spall et al., 2008;St-Laurent et al., 2014).

•
Although we cannot determine explicitly how much of the relatively high meteoric content in the ASP region was derived from DIS basal ice shelf melt, given the strong melt-laden outflow at the western end of the DIS (estimated to be 0.17 Sv discharging 81 Gt yr −1 of pure meltwater), we infer that DIS basal meltwater is a major contributor to the freshwater budget in the ASP region.

•
Our results suggest a quasi-continuous supply of melt-laden iron-enriched seawater to the euphotic zone of the ASP, which helps to explain why the ASP is Antarctica's most productive polynya per unit area.