Volume 125, Issue 7 e2019JC015842
Research Article
Free Access

Gallium: A New Tracer of Pacific Water in the Arctic Ocean

Laura M. Whitmore

Corresponding Author

Laura M. Whitmore

School of Ocean Science and Engineering, University of Southern Mississippi, Stennis Space Center, MS, USA

Correspondence to:

L. M. Whitmore,

[email protected]

Search for more papers by this author
Angelica Pasqualini

Angelica Pasqualini

Department of Earth and Environmental Engineering, Columbia University, New York, NY, USA

Search for more papers by this author
Robert Newton

Robert Newton

Lamont-Doherty Earth Observatory, Palisades, NY, USA

Search for more papers by this author
Alan M. Shiller

Alan M. Shiller

School of Ocean Science and Engineering, University of Southern Mississippi, Stennis Space Center, MS, USA

Search for more papers by this author
First published: 19 May 2020
Citations: 17

Abstract

Determining the proportions of Atlantic and Pacific Ocean seawater entering the Arctic Ocean is important both for understanding the mass balance of this basin as well as its contribution to formation of North Atlantic deep water. To quantify the distribution and amount of Pacific and Atlantic origin seawater in the western Arctic Ocean, we used dissolved Ga in a four-component linear endmember mixing model. Previously, nutrients, combined in their Redfield ratios, have been used to separate Pacific- and Atlantic-derived waters. These nutrient tracers are not conservative in practice, and there is a need to find quantities that are conserved. Dissolved Ga concentrations show measurable contrast between Atlantic and Pacific source waters, shelf-influenced waters show little impact of shelf processes on the dissolved Ga distribution, and dissolved Ga in the Arctic basins is conserved along isopycnal surfaces. Thus, we explored the potential of Ga as a new parameter in Arctic source water deconvolution. The Ga-informed deconvolution was compared to that generated with the NO3:PO4 relationship. While distributions of the water masses were qualitatively similar, the Ga-based deconvolution predicted higher amounts of Pacific water at depths between 150 and 300 m. The Ga-based decomposition yields a smoother transition between the halocline and Atlantic layers, while nutrient-based solutions have sharper transitions. A 1-D advection-diffusion model was used to constrain estimates of vertical diffusivity (Kz). The Ga-based Kz estimates agreed better with those from salinity and temperature than the nutrient method. The Ga-based approach implies greater vertical mixing between the Pacific and Atlantic waters.

Key Points

  • Dissolved gallium can be used as a tracer to determine the percentage of Pacific- and Atlantic-derived seawater in the Arctic Ocean
  • The deconvolution using gallium predicts greater amounts of Pacific water between roughly 150 and 300 m than a nutrient-based deconvolution
  • Distribution of Pacific waters in the western Arctic is more vertically mixed than previously predicted with nutrient tracers

Plain Language Summary

The Arctic Ocean is a key player in global ocean circulation and climate. The waters of the Arctic Ocean are influenced by a number of sources including seawater from both the Atlantic and Pacific oceans, sea-ice formation and melting, and river input. Unraveling these source waters is important both for deciphering the inflow of nutrients and for other materials to the Arctic Ocean as well as understanding the basin's contribution to oceanic overturning circulation and the impact of ocean heat input to this basin on Arctic sea ice. Finding methods for unraveling these sources and how they may be altered by ongoing environmental change is thus an important research objective. Previous methods for distinguishing Pacific from Atlantic water in the Arctic Ocean have relied on nutrient-based tracers that could be biased by poorly quantified biological processes. Herein, the authors demonstrate that gallium (Ga), a metal that primarily enters the ocean from dust, can be used in place of nutrient-based tracers to yield a potentially more realistic distinction between the two ocean water sources.

1 Introduction

Buoyancy in the Arctic Ocean is determined primarily by salinity, which is derived from inputs of river discharge, local precipitation, sea-ice melt, and Pacific- versus Atlantic-derived seawater (since the shallow North Pacific Ocean is less saline than the shallow North Atlantic Ocean). The distribution of low salinity waters in the Arctic Ocean is important locally within the Arctic Ocean and more broadly for its impacts on the regional and global circulation through export to the North Atlantic. Within the Arctic Ocean, a freshwater-derived buoyancy anomaly helps stabilize all of the major boundary currents of the basin including the coastal currents along the Eurasian and North American boundaries, the shelf-slope currents of the Atlantic layer, and the topographically guided “loop” currents (e.g., Rudels, 2018). In addition, freshwater content in the halocline insulates sea ice from the warmer waters below and allows perennial sea-ice cover in the central Arctic Ocean (Steele & Boyd, 1998). As freshwater is exported from the Arctic Ocean, it enters the North Atlantic, where it has important regional impacts on the physical and ecological conditions of the Nordic Seas (Torres-Valdés et al., 2013). By controlling the vertical stratification, Arctic Ocean freshwater export modulates the formation of deep water in the Nordic and Labrador seas where the northern descending limb of the meridional overturning circulation is formed.

Over the last several decades, a great deal of progress has been made in tracking sea ice and river runoff through the Arctic marine system using a combination of in situ marine chemical sampling and satellite observations (Bauch et al., 2011; Guay & Falkner, 1997; Jones & Anderson, 1986; Newton et al., 20132017; Serreze et al., 2006). However, the contribution of Pacific-sourced waters to buoyancy anomalies has been difficult to precisely identify, largely because Pacific water undergoes a range of biogeochemical transformations as and after it enters the Arctic Ocean.

While freshwater inputs and outputs to the Arctic Ocean must be equal in the long run, recent studies have observed a pattern of freshwater accumulation followed by major freshwater export events occurring on the order of decades (Belkin, 2004; Dickson et al., 1988; Häkkinen, 1993; Hunkins & Whitehead, 1992). Currently, the Arctic Ocean is in a period of freshwater accumulation (Haine et al., 2015). This buildup can influence circulation patterns and, therefore, the distribution of water mass components in the Arctic Ocean.

The polar mixed layer overlays Pacific-derived waters with high nutrient concentrations and Atlantic-derived waters containing a substantial heat signal. The flux of heat and nutrients to the surface is regulated by the strength of stratification, which in the Arctic Ocean is primarily controlled by the vertical salinity gradient. The strength and distribution of the halocline can influence the spatiotemporal extent of ice cover (e.g., Shaw et al., 2009) and euphotic zone nutrient concentrations, thereby influencing primary productivity and carbon export. Thus, making sensible projections of the impact of climate change on carbon cycling, and freshwater exports from the Arctic Ocean requires accurate assignment of water mass sources.

To distinguish individual sources of salinity (and other geochemical constituents), samples in the Arctic Ocean are typically decomposed into four components: (1) meteoric water, (2) sea-ice melt/formation, (3) Pacific-derived seawater, and (4) Atlantic-derived seawater. Ideally, conservative tracers with characteristics unique to each water type are applied in a linear mixing model in order to calculate the relative influence of individual sources given a sample's composition. Oxygen isotopes (or sometimes total alkalinity) and salinity have been applied and are useful in distinguishing between meteoric, sea ice, and seawater components. However, a definitively conservative tracer for Atlantic- and Pacific-derived seawater has not yet been described.

Up to now, researchers have used nutrient-based tracers since nutrient concentrations are high in the Pacific inflow to the Arctic Ocean and low in Atlantic inflow (e.g., Ekwurzel et al., 2001; Jones et al., 1998; Newton et al., 2013; Wilson & Wallace, 1990). To overcome the nonconservative nature of nutrient concentrations, nitrogen and phosphorus can be used simultaneously by combining them in their respective NO3:PO4 ratios (Figure 1). One commonly used nutrient tracer, the Arctic nitrate-phosphate (ANP) tracer exploits an offset in the Pacific and Atlantic NO3:PO4 trends observed in the central Arctic Ocean (Figure 1) (Jones et al., 1998; Newton et al., 2013). The distance a sample falls between the two trendlines characterizes the amount of Pacific- or Atlantic-derived water in the sample (see also, section 2.5 below). The Pacific-derived seawater has high PO4 relative to Atlantic-derived seawater due to nutrient regeneration in the deep Pacific Ocean and upwelling of those nutrient-rich waters in the Bering Sea. Furthermore, the Pacific inflow transits the broad continental shelves before entering the Arctic Ocean basin; during this transit, shelf processes cause it to accumulate PO4 relative to NO3 (Cooper et al., 1999; Yamamoto-Kawai et al., 2008). This shelf processing leads us to identify the waters detrained from the shelf (rather than the Bering Strait inflow) as the endmember, sometimes referred to as “shelf-modified Pacific inflow” (Newton et al., 2013).

Details are in the caption following the image
The 2015 Arctic GEOTRACES samples plotted in N-P space; data points are from the surface 500 m of the western Arctic Ocean water column (this study). The trendlines are drawn for Atlantic-derived seawater (orange and light brown) and Pacific-derived seawater (black and gray). The trend lines from this study (solid lines) are compared to those of Jones et al. (1998), the study that first defined and applied the nutrient water mass deconvolution described herein (dashed lines). Our trends are shifted from theirs, which demonstrates variability in the slopes over time and indicates a source of uncertainty in the method.

The results of nutrient-based tracers have been useful qualitatively in identifying areas of Pacific, Atlantic, and shelf-sea influence; however, they are of questionable quantitative value as significant errors appear as a result of community composition and other processes that impart non-Redfieldian ratios on the nutrients in the dissolved phase, including nitrification, denitrification, phosphate regeneration, and exchange of oxygen at the ocean surface (e.g., Alkire et al., 20152019; Bauch et al., 2011; Ekwurzel et al., 2001). In addition, the method requires knowledge of the endmembers, which are highly varied in space and may change with time (e.g., Alkire et al., 2019; Cooper et al., 1999). Indeed, the nutrient endmember trendlines from our 2015 study (Figure 1, solid lines) differ from those of Jones et al. (1998), the study that first defined and applied this method (Figure 1, dashed lines).

Since the slopes of “Atlantic” and “Pacific” waters differ (Figure 1), the fraction of Pacific-derived water (fPac) in a sample depends on how one draws the mixing line between the two endmember nutrient trends. Also, it should be noted that results from using different tracers (e.g., alkalinity instead of δ18O-H2O or the PO4:oxygen ratio instead of NO3:PO4) have yielded differences in fPac greater than 10% (Ekwurzel et al., 2001 vs. Jones et al., 1998, reporting on the 1994 Arctic Ocean survey). Alkire et al. (2015) compared several methods and endmember definitions and documented differences of up to 60% in fPac between nutrient tracer methods. Even when the same parameters are used, choosing different (but plausible) endmembers led to a 13% median standard deviation for fPac. Generally, authors report errors between 10% and 14%, derived from endmember variation tests and perturbations in the slopes of the NO3:PO4 trend (Alkire et al., 2015; Bauch et al., 2011; Yamamoto-Kawai et al., 2008) even without including the uncertainty generated by biologically mediated modification of nutrient ratios. Uncertainty in the meteoric and sea-ice fractions is consistently much less (i.e., <1% median standard deviation; Alkire et al., 2015).

Because of the uncertainty in nutrient-based Arctic deconvolution methods, there is a need to identify a conservative deconvolution tracer that can yield a more quantitative understanding of Arctic buoyancy and the distribution of Pacific- and Atlantic-sourced waters.

Dissolved gallium (Ga) is a less reactive analog of Al (e.g., Orians & Bruland, 1988; Shiller, 19881998). This lower reactivity (i.e., longer residence time) suggests the possibility of using Ga as a water mass tracer over moderate spatial/temporal scales. Additionally, as pointed out by McAlister and Orians (2015), the contrast between incoming low-Ga Pacific waters versus high-Ga Atlantic waters suggests the possibility of using dissolved Ga as a water mass deconvolution parameter in the Arctic Ocean, potentially replacing nutrient-based tracers. For Ga to be a useful parameter in the Arctic, it must behave sufficiently conservatively over the mixing scale of Atlantic and Pacific waters. This is likely to be the case in the central Arctic basins since previous work has indicated nonconservative Ga behavior is most commonly associated with atmospheric input (Orians & Bruland, 1988; Shiller, 19881998), bottom input (Ho et al., 2019), or scavenging in particle-rich waters (McAlister & Orians, 2012): factors that are unlikely to be important in the central Arctic Ocean.

Here, we present the first trans-Arctic Ocean Ga sections describing data in the Makarov and Canada basins of the western Arctic Ocean. Our data agree with McAlister and Orians’ (2015) supposition that Ga could be a useful tracer in the Arctic Ocean. Pacific-derived waters have low Ga concentrations (<10 pmol kg−1) versus Atlantic-derived waters (28 pmol kg−1). This contrast, in addition to Ga's low reactivity relative to the residence times of shallow waters in the Arctic, is what drives Ga's potential use as a tracer (Schlosser et al., 1999; Shiller, 1998). We assessed the utility of Ga as a tracer for Pacific and Atlantic water by employing a linear endmember mixing model and comparing these results to the nutrient tracer method described in Newton et al. (2013). We applied a vertical advection-diffusion model to assess the resulting vertical distributions of the Pacific fraction and compare them to expectations based on vertical gradients in temperature and salinity.

2 Materials and Methods

2.1 Section Description

We present and discuss freshwater components calculated from measurements of stable isotopes of water, salinity, gallium, and nutrients along the 2015 U.S. Arctic Ocean GEOTRACES section (GN01) in the upper 500 m of the water column. Samples were collected aboard the USCGC Healy at 66 hydrocast stations (22 GEOTRACES and 44 CLIVAR repeat hydrography program) along two transects extending from the continental shelf to the North Pole roughly along longitudes 180°W and 150°W (Figure 2a). We define our Pacific endmembers using shelf data from the Bering Strait, as all Pacific waters are constricted through this location.

Details are in the caption following the image
Regional and hydrographic description of the study area. (a) Map of sampled stations with hydrographic data (CLIVAR-only stations) as black triangles or hydrographic and trace metal data (GEOTRACES stations) as blue circles. Geographic features are numbered and major surface circulation features are dictated on the map. (b) Cross section of the Arctic Ocean (indicated on the map to the right) detailed with the rough distribution of major water masses. Figure modified from Whitmore et al. (2019).

The western Arctic Ocean water column is characterized by the polar mixed layer (PML), the Pacific halocline (PH), and Atlantic halocline (AH) and Atlantic water (AW) (Figure 2b). The PML is composed largely of sea-ice meltwater and meteoric water. Below the PML, cold, relatively fresh, and nutrient-rich Pacific water (S ~ 32.5) sits above warm, saline AW (S ~ 34.8). Brine rejection can influence the water column, although the exact influence is elusive (e.g., Bauch et al., 2011). Traditional brine rejection is the formation and subsequent sinking of hypersaline waters that are derived from ice formation. Salt rejection during sea-ice formation can result in hypersaline brine channels in sea ice, which may be seasonally rejected to the surface ocean. Previous literature demonstrates the highest influence of brines above and below the PH (Bauch et al., 2011; Yamamoto-Kawai, 2005); although there is discussion that brines may drive deep water geochemical signals (e.g., Jones et al., 1995).

2.2 Ancillary Hydrographic Data

Ancillary hydrographic data for GN01 were produced by the Scripps Oceanographic Data Facility (ODF), Shipboard Technical Support group. Salinity samples collected from the rosette were analyzed onboard using a Guildline 8400 salinometer (accuracy approximately ±0.002). Nutrient concentrations were measured aboard the icebreaker within 1–4 hr from collection. Nutrient analyses were performed on a seal analytical continuous-flow auto-analyzer 3 (AA3) following the World Ocean Circulation Experiment (WOCE) standard techniques (Gordon et al., 1993; Hager et al., 1972). Analytical methods used in 2015 were in accordance with the GO-SHIP repeat hydrography manual (Hood et al., 2010). All hydrographic data are publicly available online through CLIVAR and Carbon Hydrographic Data Office (CCHDO; Kadko et al., 2015) and through the Biological and Chemical Oceanographic Data Management Office (BCO-DMO; Cutter et al., 2019a2019b).

2.3 δ18O-H2O Analytical Methods

Oxygen isotope ratios (H218O/H216O) were measured at Columbia University's Lamont Doherty Earth Observatory using a Picarro L2130-i Cavity Ring-Down Spectroscopy (C.R.D.S.) analyzer following Walker et al. (2016). Oxygen isotope ratios are reported as δ18O, that is, the per mil deviation of the H218O/H216O ratio from that of Vienna standard mean ocean water (VSMOW-2) (Craig, 1961; Gat & Gonfiantini, 1981). Analytical precision is approximately ±0.027 per mil (Pasqualini et al., 2017). Stable isotopic analyses were performed on samples covering the entire water column at GEOTRACES stations and the upper 500 m at U.S. Repeat Hydrography stations. This study only assesses the upper 500 m at all stations.

2.4 Ga Sample Collection and Analytical Methods

Following GEOTRACES standards (Cutter et al., 2014), Ga samples were filtered (0.2 μm) into acid-cleaned high-density polyethylene bottles from GO-FLO bottles mounted on a trace metal clean rosette and subsequently acidified to 0.024 M HCl. Acidified seawater samples were prepared for analysis on an inductively coupled plasma mass spectrometer (ICP-MS) at the University of Southern Mississippi (Center for Trace Analysis, ThermoFisher Element-XR) by preconcentrating with magnesium hydroxide coprecipitation coupled with an isotope dilution approach (Ho et al., 2019; Shiller & Bairamadgi, 2006). Briefly, samples (7-ml seawater) were spiked (spike volume variable with seawater Ga concentration) with an enriched isotope spike of 99.8% 71Ga (Oak Ridge National Laboratories). Aqueous ammonium was added to the solution, and the precipitate was collected and rinsed three times with a solution of ~0.1% NH4OH to remove barium, which can analytically interfere with 69Ga as doubly charged 138Ba. The precipitate was dissolved in 3% ultrapure nitric acid (0.47 M HNO3; Seastar Chemicals, Baseline) and analyzed on the ICP-MS for 69Ga, 71Ga, and 138Ba. Samples were introduced to the ICP-MS through a PC3 Spray chamber (Elemental Scientific, Inc.) and analyzed in low resolution.

Data are quality controlled with multiple approaches. Atlantic Ocean bulk surface and deep water samples (GS and GD, respectively) distributed from the 2008 GEOTRACES Intercalibration Cruise were analyzed during each run, and the average concentration and precision from these repeat measurements is presented in Table 1. Additionally, recovery was estimated for each run by determination of Ga-spiked and unspiked aliquots of a large-volume seawater sample used in our lab for tracking analytical consistency (Table 1). Intercalibration efforts were made in accordance to GEOTRACES protocols. Ga data are available via the Biological and Chemical Oceanography Data Management Office (BCO-DMO; Shiller, 2019).

Table 1. Gallium Reproducibility Assessment
Ga, pmol/kg SD N
GS 41.9 1.1 9
GD 32.8 1.4 13
% Recovery 96.8 5.3 11

2.5 Linear Mixing Model

Water from the upper 500 m of the Arctic Ocean can be divided into four primary components: (1) meteoric water (fmet), (2) sea-ice melt (fSIM), (3) Pacific-derived waters (fPac), and (4) Atlantic-derived waters (fAtl). A linear mixing model approach was used to calculate the fraction of each water type in the upper 500 m of the water column (Equations 26). We applied salinity (S), δ18O, and an N:P-based tracer (which we call the Arctic nitrate-phosphate tracer, ANP) or Ga concentrations to calculate the fractions of each sample. The ANP value was calculated following Newton et al. (2013) whereby ANP is determined by the distance a sample falls between the Pacific and Atlantic trendlines (Figure 1; Equation 1). The Atlantic and Pacific trendlines are determined using samples within the Atlantic layer (characterized by salinity and temperature) and by assessing samples in a nutrient-rich tongue characterized with high silica above the Atlantic-derived waters. The best linear fit is established for each sample subset, which is referred to herein as the Atlantic and Pacific trendlines.
urn:x-wiley:21699275:media:jgrc23997:jgrc23997-math-0001(1)
The ANP tracer is a value between 0 and 1, where 0 is Atlantic-derived seawater and 1 is the Pacific-derived seawater; these values are set in NO3:PO4 space and defined by the trends observed in the surface 500 m of the central Arctic Ocean (Figure 1; Equation 1). It is important to note that there are several irreducible errors in this method, beyond analytical (measurement) errors, including non-Redfield nutrient processing and local nitrate or phosphate inputs. Additionally, there is no a priori mixing line between the Atlantic and Pacific NO3:PO4 regressions. We have used the minimal distance from the sample to each trendline.
urn:x-wiley:21699275:media:jgrc23997:jgrc23997-math-0002(2)
urn:x-wiley:21699275:media:jgrc23997:jgrc23997-math-0003(3)
urn:x-wiley:21699275:media:jgrc23997:jgrc23997-math-0004(4)
urn:x-wiley:21699275:media:jgrc23997:jgrc23997-math-0005(5)
urn:x-wiley:21699275:media:jgrc23997:jgrc23997-math-0006(6)

If all source waters had exactly their endmember concentrations, then Atlantic (fAtl), Pacific (fPac), and meteoric water (fmet) fractions would lie strictly between 0 and 1. In reality, there is scatter in the endmember concentrations, which leads to small excursions in the calculated fractions above 1 or below 0. Some authors cap their reported fractions to avoid showing “nonphysical” concentrations. We find that practice arbitrary and report the calculated values without constraining the model. The sea-ice fraction (fSIM) can, in principle, range from negative 1 to positive 1. Positive fSIM values indicate sea-ice melt and negative fSIM values represent brines: sea water from which freshwater has been extracted by the formation of sea ice.

When available, endmember values were determined from our data set; however, several values were taken from the literature (Table 2).

Table 2. Endmember Parameter Values
Water mass Salinity [g kg−1]b δ18O [‰]b Ga [pmol kg−1] Arctic N:Pa
Atlantic water 34.92 0.3 28 0
Pacific water 32.50 −1.1 5 1
Meteoric water 0 −20 25 0
Sea-ice meltwater 4 Surf. + 2.6 ‰ 2.3 Surface
  • a Pacific water: slope = 14, intercept = −11; Atlantic water: slope = 17, intercept = −2.
  • b Newton et al., 2013 and references therein.

Our study was not able to directly measure the meteoric endmember of Ga. However, we chose a value of 25 pmol kg−1 based on an analysis of available fluvial Ga data, an examination of our halocline data, and a sensitivity test of our selection. Specifically, Colombo et al. (2019) made observations in rivers outside of the central Arctic in the Canadian Arctic Archipelago (CAA) with dissolved Ga in North American continental rivers ranging from 18–35 pmol kg−1 while rivers on islands in the CAA had higher dissolved Ga, most were below 50 pmol kg−1. This is in general agreement with the results of Shiller and Frilot (1996) for a suite of small streams in California where dissolved Ga ranged from 1–79 pmol kg−1. In contrast, Gaillardet et al. (2014) compilation of global river Ga yields a much higher mean value, but those data may be biased with results from studies where analytical interferences were not carefully evaluated. As an alternative approach to constraining the meteoric Ga, we examined our near-surface Arctic Ocean samples where the meteoric fraction was comparatively high. For samples in the upper 20 m of the water column, dissolved Ga was generally low (<14 pmol kg−1) and had compositions of predominantly Pacific and meteoric sources (2–25% fmet and 87–100% fPac by the ANP method). By solving Equation 5 using ANP-derived water mass estimates for surface waters with meteoric compositions > 10%, we estimate a meteoric endmember composition of approximately 25 pmol kg−1 (range: 0–83 pmol kg−1). Finally, to test for the sensitivity of our deconvolution to the to our meteoric Ga endmember, we perturbed the meteoric Ga endmember between 0 and 100 pmol kg−1 in our calculations (additional sensitivity analysis is discussed in section 3.2). For river endmember concentrations greater than 50 pmol kg−1, the fractions of Atlantic and Pacific water in the surface deviate from realistic compositions. Considering a riverine Ga range between 0 and 50 with an ideal endmember of 25 pmol kg−1, the sensitivity analysis resulted in a variance in fmet of <0.01%; fSIM varied within 1% and fPac and fAtl within 13%. This variability is similar to sensitivity tests conducted on nutrient-method endmember choices (Alkire et al., 2015). Any uncertainty associated with the meteoric Ga endmember may be decreased with targeted studies assessing endmember geochemical properties.

3 Results and Discussion

3.1 Gallium Distribution

The gallium distribution at isopycnal densities associated with the Pacific-influenced halocline (sometimes referred to as the “lower halocline,” hereafter referred to as the “Pacific halocline”) indicates little modification of Ga within this water mass (Figures 3a and 3b). The Pacific halocline (Ga = 7.4 ± 1.2 pmol kg−1; n = 41) is equivalent in concentration to the incoming Bering Strait waters (Ga = 7.7 ± 1.1 pmol kg−1; n = 4; Station 4). Notably, surface waters (<100 m) at the Bering Sea slope (i.e., Pacific-derived waters that have not had exposure to the shelf; Station 1) are 4.4 ± 1.4 pmol kg−1; this indicates some transformation of Pacific water as it transits northward (Figure 3). Other deviations from the mean Bering Strait waters occur at shelf Station 66, where the concentration was 2.6–3.9 pmol kg−1 (Figure 3). Additionally, shelf Station 3 had low surface Ga (Ga = 4.4 pmol kg−1; z = 1 m) and higher concentrations at depth (Ga = 9.7 ± 0.5 pmol kg−1; z > 15 m). Aside from Station 3, shelf stations were homogeneous with depth (within the error of our measurement) (Figure 3). The relative stability in Ga concentrations on the shelves and in the Pacific halocline contrasts with nutrient concentrations, which are substantially modified during transit over the shelves. The exchange of nutrients, both near the surface and at pore-water interfaces, as waters pass over the shelves, implies that even Atlantic-sourced waters may develop “Pacific-like” nutrient signals as they transit over the Eurasian Arctic shelves.

Details are in the caption following the image
Shelf profiles of dissolved Ga (pmol kg−1); station locations and names are indicated on the map and by the legend. Dotted lines are indicative of the assigned Pacific endmember (5 pmol kg−1) and Atlantic endmember (28 pmol kg−1).

In the basin below the Pacific halocline, Ga increases into the Atlantic water layer (Ga = 28–40 pmol kg−1; Figure 4a). Gallium concentration in the western Arctic Ocean below 500 m increases to roughly 30 pmol kg−1; however, the one station sampled in the Eurasian basin has much higher Ga below the depth of the Lomonosov Ridge (Ga ~ 40 pmol kg−1; Figure 4a). Although there are small deviations in Ga concentrations at our shelf stations relative to the Bering Sea (Station 1), this is minor relative to the much higher Ga in Atlantic waters of our section.

Details are in the caption following the image
Literature comparison of dissolved Ga profiles (pmol kg−1). (a) GN01 section (this study), (b) Data from the GIPY14 cruise conducted during the international polar year and published in McAlister and Orians (2015). Colored data points indicate regions of the sections as indicated in the legend to the right and the map to the left.

Our vertical Ga distributions, low in the surface and higher at depth, agree with the supposition made by McAlister and Orians (2015) that the distribution of Ga in the Arctic Ocean basins lends itself well as a tracer of Pacific- and Atlantic-derived waters. We compared our Canada and Makarov basin Ga profiles to McAlister and Orians (2015) and note that there is consistency between basins within the error of our measurements (Figure 4). Data from the northernmost stations (>85°N) have a different vertical distribution (Ga starts increasing at a shallower depth) than observations made by McAlister and Orians (2015) (Figure 4); this latitudinal demarcation is roughly the location of the Pacific-Atlantic front in the western Arctic during the GN01 sections.

Prior work suggests that over modest basin-wide spatial-temporal scales, dissolved Ga is relatively unreactive (Hayes et al., 2018; Shiller, 1998; Shiller & Bairamadgi, 2006). Where Ga shows clear nonconservative behavior, there is an influence of particles by atmospheric input (Orians & Bruland, 1988; Shiller, 1988) or by scavenging in particle-rich river plumes (McAlister & Orians, 2012). Sediment resuspension is also a likely contributor to Ga input (Ho et al., 2019). Scavenging and/or sediment input may indeed be a factor on the Arctic shelves as evidenced by the limited Ga variability in our shelf-influenced Pacific waters as well as by similar observations by McAlister and Orians (2015). However, in general, these input and removal processes should be of minimal importance in the particle-poor Arctic basins (Honjo et al., 2010).

Compared to Ga distributions, nitrate and phosphate concentrations increase in the Pacific-halocline relative to the Bering Strait endmember because the shelves substantially modify nutrients. Furthermore, due to modifications over the shelf, the N:P slope is different between shelf-influenced Pacific-derived waters and Pacific waters before they interact with the shelves (e.g., Cooper et al., 1999). Because of shelf modifications, the Pacific halocline is often discussed as “shelf influenced” waters rather than truly Pacific-origin waters. Our Ga distribution is less variable than nutrient concentrations, which supports its application as a conservative tracer.

3.2 Linear Mixing Model Results

In comparing the Ga-derived and ANP-derived water mass fractions, fmet and fSIM are roughly equivalent (Figures 5c and 5d) and have discrepancies of less than 0.05 and 0.02, respectively. The discrepancy is determined by subtracting the fraction determined using the Ga method from the ANP method (fmet median < 0.01 and fSIM median < 0.01). This is not surprising since these two fractions are largely determined by the S and δ18O distributions.

Details are in the caption following the image
Comparison of Ga-derived water mass fractions and ANP-derived water mass fractions. (a) Fraction of Atlantic (fAtl), (b) fraction of Pacific (fPac), (c) fraction of meteoric (fmet), and (d) fraction of sea-ice melt or formation (fSIM).

Both methods produce qualitatively similar spatial distributions of fPac and fAtl (Figures 6e6h). However, quantitatively, there can be a discrepancy of up to 0.66 in fPac or fAtl between the two methods (Figures 5a and 5b). Between the nutrient and gallium method, the median difference of the fPac and fAtl are 0.11 and 0.10, respectively (Figures 6i and 6j).

Details are in the caption following the image
Sections of ANP-based and Ga-based water mass deconvolutions. The left column is the Makarov Basin transect (180°W), and the right column is the Canada Basin transect (150°W); the bathymetry for each section is indicated in the bottom panel. (a, b) ANP distribution. (c, d) Dissolved Ga concentration (pmol kg−1). (e, f) Pacific fraction determined by applying the ANP tracer. (g, h) Pacific fraction determined by the Ga tracer. (i, j) Distribution of the difference between the ANP- and Ga-derived Pacific fractions. Contour lines are isopycnal surfaces (σθ) at 26.46 kg m−3 (core of the Pacific halocline; shallowest), 27.51 kg m−3 (core of the lower halocline; middle), and 27.93 kg m−3 (core of the Fram Strait branch water; deepest). Section panels made using Ocean Data View 5.1 software.

The fractions of Pacific and Atlantic waters are roughly the inverse of each other, so in this discussion, we focus on fPac (replicated figures for all other fractions are presented in the Supporting Information Figures S1S3). In terms of fPac, the ANP-derived water mass fraction underpredicted fPac relative to the Ga method. The greatest differences between the two methods arise at depths between the core of the Pacific halocline and the core of the Atlantic layer for stations north of 85°N (i.e., at stations north of the Pacific-Atlantic front).

The methods carry uncertainties in both the assumed endmember concentrations and the sample data. Sample analyses generally have lower error than the endmember uncertainties; however, Ga sample errors (RSDs) are often higher than those for nitrate or phosphate. To determine the sensitivity of the method, the endmembers for ANP and Ga were perturbed to their maximum and minimum expectation. The sea-ice and meteoric fractions consistently have low sensitivity to the endmembers. The ANP method varies by about 10% and 5% for the Atlantic and Pacific fractions, respectively. Comparatively, the Ga method varies by <5% for both the Atlantic and Pacific fractions. The sensitivity in the Ga method can be reduced by refining the endmember estimate. Importantly, despite the uncertainties in the approach, changing the endmember values does not change the shapes of the resultant profiles nor does it resolve the differences between the ANP and Ga methods.

The profiles of the tracers (i.e., Ga or ANP), and the resulting fractions of Pacific and Atlantic water, suggest two contrasting mixing modes (Figure 7). For stations north of 85°N, the ANP-derived fPac transitions from high Pacific water to nearly zero Pacific water across a very narrow depth range. This feature indicates minimal mixing between the water masses. The Ga-derived Pacific fractions (fPac) transition between high and low Pacific content more gradually with depth, suggesting greater mixing between Atlantic and Pacific water types than the ANP profile exhibits. Thus, understanding which tracer more reliably illustrates the relative distributions of the Atlantic and Pacific waters has implications about Arctic circulation that relate closely to the exchange of materials and heat between the water types.

Details are in the caption following the image
Profiles of the Pacific fraction from the (a) ANP and (b) Ga method. (c) Bottle salinity data. (d) CTD temperature profile (at bottle depths). Solid lines are mean profiles determined using Java Ocean Atlas.

Previous literature has presented multiple approaches to determining the “nutrient-tracer” and has outlined the variability between approaches (e.g., Alkire et al., 2015). We chose not to replicate this in depth, as the uncertainties between different nutrient methods are well described. To discuss how different nutrient approaches may compare to the Ga-tracer method, we reproduced water mass deconvolution using the approach described in Yamamoto-Kawai et al. (2008) and compared it to the results using the ANP approach described above (originally detailed in Newton et al., 2013). The Yamamoto-Kawai et al. (2008) approach differs from that of Newton et al. (2013) in the choice of slopes used to define the “Atlantic” and “Pacific” endmembers in N:P space. The choice of slopes does make a quantitative difference in the division between Atlantic and Pacific water masses, but the difference is small relative to other sources of uncertainty and is not important to the current discussion. We note that in comparing the nutrient approaches (Yamamoto-Kawai et al., 2008 vs. Newton et al., 2013), distributions are qualitatively similar—fPac is highest in the same regions but still there are deviations of up to 0.6 in fPac between methods. Again, these differences have been described elsewhere, and the important comparison in our study is that, regardless of nutrient methodology, the Ga approach systematically predicts a greater fPac in the transition zone than the nutrient approaches (Figures S4–S5).

3.3 Inventories

Geochemists and physical oceanographers have visualized the distribution of freshwater in the Arctic Ocean by summing the freshwater anomaly vertically in the water column and showing a total “height” of freshwater at each location (e.g., Ekwurzel et al., 2001; Serreze et al., 2006). The freshwater distribution is sufficient as a snapshot of the current state of buoyancy forcing. However, to understand the mechanisms underlying this state, and to make reasonable forward projections as the Arctic Ocean evolves in time, it is necessary to estimate the sources and sinks of buoyancy—which starts with knowing the spatial distribution of the main sources of freshwater: meteoric water, sea-ice melt, and Pacific inflow. To visualize the differences between the ANP and Ga approaches, we determined the Pacific-sourced freshwater inventory, as estimated from each method, and plotted the totals as a function of latitude (Figure 8).

Details are in the caption following the image
Pacific freshwater inventory (m). (a) Ga (blue) and ANP (black) inventories at stations in the western Arctic basin. (b) Difference in the Pacific water inventory as calculated from each method (ANP or Ga).

As we stated above, the freshwater contribution from the Pacific fraction is greater in the Ga-derived approach. This increase in freshwater from fPac (relative to the ANP method) is balanced by a decrease in the meteoric freshwater component (Figure 5c). Near the continental slope (~72°N), the two methods nearly agree, while the difference between them increases approximately linearly to a maximum of about 7 m, at the northern end of the sections (Figure 8b). To understand this pattern, recall that Ga is carried by the Atlantic fraction and that the ANP value of Pacific water is set, by definition, by the values at the southern end of the transect, where shelf-modified Pacific inflow detrains from the Chukchi Shelf.

To try to understand whether the Ga method or ANP method is producing a better estimate of the fPac distribution, we start first with a logical argument and then, in the next section, proceed to a modeling argument. In our logical approach, we will consider the implications of the ANP tracer being the more conservative parameter and then the implication of the Ga distribution being more conservative parameter.

First, let us assume that the ANP tracer is relatively conservative and accurately estimates the Pacific water fraction throughout the transect. Recall that the difference between the methods is that the Ga method estimates more Pacific water and less Atlantic water. To obtain this result, the observed Ga concentrations must be less than our assumed endmember concentrations. Throughout the sections, the Ga tracer and ANP tracer have the closest agreement in the south (Figure 8). Thus, the observed Ga concentrations are closest to the assumed Atlantic endmember furthest from the Atlantic source and most inaccurate at the northern end, where Atlantic waters are closest to their source. One explanation for this may be that Ga experiences extreme scavenging between Fram Strait and the Lomonosov Ridge and is then added back into the water column between the ridge and the Chukchi Shelf. Given the degree of scavenging required (a removal of 10–15 pmol kg−1), we find this unlikely.

On the other hand, we can assume that the Ga tracer is conservative and consider whether biological processing of nutrients accounts for an inaccuracy in the ANP results. Again, the tracers have the best agreement at the southern end (near the Chukchi continental slope). As Pacific waters spread northward, there is an increasing difference between the results of the Ga and ANP tracers. This scenario implies that nutrients in the Pacific-derived water masses are modified by nutrient addition or removal at non-Redfieldian ratios as they transit further from their source. Indeed, nutrient addition at N:P ratios greater than 16:1 could explain the observed changes. One possible mechanism for this is nitrogen fixation. There is a growing body of literature providing evidence of diazotrophy in the Arctic Ocean (Blais et al., 2012; Fernández-Méndez et al., 2016; Harding et al., 2018; Sipler et al., 2017). Early work with Ga in the Arctic posited that differences between N* and Ga distributions may provide evidence of N-fixation at the surface, where N* is locally defined as 13.6 × PO4 + 11 (McAlister & Orians, 2015). However, the greatest difference in the ANP- and Ga-derived water mass fractions occurs at depths between 150 and 300 m where the N* and Ga profiles are similar. A review of nutrient processes in the upper Arctic is beyond our scope, but we consider nutrient dynamics (nitrogen fixation, for example) to be a more plausible mechanism than open-ocean Ga scavenging, followed by subsequent redissolution along the same isopycnals.

Indeed, Ga scavenging would be unusual away from sedimentary interfaces, and there is no independent evidence of significant Ga addition over the Amerasian Basin. Thus, through this analysis, we are inclined to accept the Ga-supported fPac estimates. In section 3.4, we turn to a vertical advection-diffusion model to assess the diffusivity implied based on the shapes of the fPac vertical profiles.

3.4 Vertical Advection-Diffusion Model

Since ANP- and Ga-based deconvolutions imply different degrees of diapycnal mixing, it is instructive to use each to constrain the vertical mixing coefficient in a one-dimensional, vertical mixing model. The model is a two-parameter advection-diffusion model with a small, background upwelling, constant vertical diffusion, and fixed boundary conditions in above (in the core of the Pacific halocline, ~100 m) and below (in the core of the Atlantic layer, ~300 m). Few studies detail upwelling/downwelling rates in the central Arctic Ocean, but the indications are that they are low (Fer, 2009; Wallace et al., 1987). The essential limitation of the model (its main divergence from the real Arctic Ocean situation) is that the vertical gradients are assumed to be a result of purely vertical processes. We know that lateral spreading along isopycnals is important in ventilating the upper and intermediate layers of the Arctic Ocean (e.g., Steele & Boyd, 1998). Therefore, we present these results with the understanding that the model's vertical eddy diffusion subsumes the results of a combination of three-dimensional processes.

Away from the continental shelves and slopes, over the deep Arctic Ocean, the shapes of property profiles fall into two types, those north and those south of 85°N. We modeled the northern and southern stations along the GN01 track separately. Gallium data were too sparse between the Pacific halocline and Atlantic layer to capture inflection points at each station, so we gridded the data vertically and created mean profiles for those two groups of stations (Figure 7). The model ran with a simple forward Euler numerical integration for 10 years with a 12-hr time step, with upper and lower boundary values held fixed at the mean observed values throughout the run. The upwelling rate for all runs was fixed at 0.7 × 10−7 m s−1, but we adjusted the model so that the median of the simulated values was always at the same depth as the median of the observations. The model was run iteratively with diffusivities (kz) of 1 × 10−5, 5 × 10−5, 1 × 10−6, 5 × 10−6,and 1 × 10−7 m2 s−1. These diffusivities are low in the context of global oceanic diffusivities (on the order of 1 × 10−4 m2 s−1; Munk, 1966; Talley et al., 2011). However, Arctic Ocean diffusivities tend to be far more estuarine like (i.e., lower) due to freshwater inputs to the surface ocean and minimal surface turbulence due to sea-ice cover (Fer, 2009; Rainville & Winsor, 2008; Shaw & Stanton, 2014; Wallace et al., 1987). We compared the model results to the mean profile of the observations (Figures S6S9), assessing the goodness of fit by applying a linear regression and then determining the R-squared (RSQ), the squared residuals (SSR), and the root mean squared error (RMSE) (Table S1). We applied this process to temperature and salinity from the CTD traces and to the Pacific water fractions as determined by the Ga- and ANP-based decomposition techniques. For temperature and salinity, the model results indicate a vertical diffusivity coefficient of roughly 5 × 10−6 m2 s−1 for stations both north and south of 85°N. Previous studies have indicated diffusivities of roughly 1 × 10−6 m2 s−1 (e.g., Nguyen et al., 2009; Wallace et al., 1987; Zhang & Steele, 2007); we therefore consider our results reasonable for the purposes of our study.

South of 85°N, both fPac_ANP and fPac_Ga profiles yielded eddy diffusivity coefficients equivalent to those determined from salinity and temperature (kz = 5 × 10−6 m2 s−1; Table S1; Figures S6S9). Because of the sharp vertical gradient in fPac_ANP between 100–400 m compared to fPac_Ga at stations north of 85°N (Figure 7), one would expect lower diffusivities for the fPac_ANP profiles relative to the fPac_Ga profiles, which is indeed what we find (for fPac_ANP, kz = 1 × 10−6 m2 s−1; for fPac_Ga, kz = 5 × 10−6 m2 s−1; Table S1). Importantly, the best-fit diffusivity coefficient to the fPac_ANP profiles north of 85°N was also lower than the best fits for temperature or salinity, which aligned closely with the Ga-derived coefficients (for S and T, kz = 5 × 10−6 m2 s−1). Thus, while the Pacific water profiles from both methods are within the range of plausible diapycnal rates, the Ga method is a better match, overall, to those rates implied from the basic physical parameters, salinity and heat.

We note, in addition, that one of the differences between the two methods is that the gallium-based decomposition extends the Pacific-influenced upper halocline farther toward the Eurasian side of the Arctic Ocean than the ANP-based analysis does. This method's contrast is even more severe when comparing with a descriptive analysis of the extent of the Pacific water influence using the NO parameter (a semiconservative tracer where NO = 9 × NO3 + O2); Alkire et al. (2019) determined NO using data from the same cruise as our study (in addition to other cruises) and describe a sharp front as the extent of the Pacific halocline. Even along isopycnal surfaces, Ga-based distributions imply a more gradual transition from Pacific to Atlantic source waters. That is, overall Ga implies a more dispersive mixing regime over the central Arctic Ocean basins than the nutrient-based deconvolution.

4 Conclusions

The use of a nutrient tracer to deconvolve Pacific- and Atlantic-derived waters in the Arctic Ocean has limited quantitative utility. Different nutrient tracer methods disagree significantly when applied to the same data sets. Even when the same tracer approach is applied, slight changes in the Redfield relationship between years and cruise tracks increase uncertainty in the determined Pacific inventories. Indeed, Alkire et al. (2019) recently illustrated this point by comparing the nutrient tracer approach to the distribution of NO, noting clear differences in the extent of the Pacific fraction and the extent of the NO maximum.

The application of Ga as a tracer further exposes uncertainty associated with the assumed quasi-conservative behavior of nutrient tracers in an Arctic Ocean linear endmember mixing model for Atlantic and Pacific waters. The Ga distribution indicates minor variability resulting from shelf exposure, especially when compared to nutrients (e.g., Cooper et al., 1999). In applying Ga as a conservative-type tracer in place of nutrients, the resulting water-source distributions are effectively equivalent for fraction of meteoric water and fraction of sea-ice melt. However, the Ga method often predicts a greater fPac (and lower fraction of Atlantic waters) relative to historically used nutrient methods.

The spatial distribution of fPac_Ga − fPac_ANP reveals that the greatest difference of fPac occurs in the northernmost region of the western Arctic Ocean (near the North Pole). This regional inconsistency is also observed in our simplified vertical advection-diffusion model. In assessing vertical mixing processes, we find that the Ga method is more reflective of the salinity and temperature distributions for regions north of the Pacific-Atlantic front. For regions south of the Pacific-Atlantic front, the two methods have slightly better agreement.

Our study has demonstrated that relative to the nutrient approach, Ga-derived water mass deconvolution implies more dispersive mixing between the Atlantic and Pacific water types. The Ga method thus implicates the potential for greater heat transfer and nutrients between the water masses than the ANP method. In assuming that the Ga tracer is accurate, this implies an addition of nitrate or removal of phosphate independent of Redfieldian processes.

Although the collection and analysis of a trace element sample may seem more involved than nutrient determination, Ga is not a particularly contamination-prone element. Reducing the analytical uncertainty in Ga determination while increasing sample throughput will also increase the utility of this new Arctic Ocean tracer. Indeed, ongoing work in our lab suggests an automated extraction method should be able to reduce uncertainty in Ga determination to less than 3%.

Future studies should include more samples in the Eurasian basin, which would enable the assessment of the transpolar drift influence and of Eurasian shelf influences, since the GN01 transect only had one station in the Eurasian basin (and that station was heavily influenced by TPD surface waters). Additionally, a multiple tracer technique (i.e., using an overdetermined system) integrating information from the suite of tracers that projects like GEOTRACES has generated may be a useful tool.

Acknowledgments

This research was supported by the National Science Foundation (OCE-1435312 [AMS], OCE-1436666 [RN]). Thank you to the USCGC Healy crew and the scientific leadership—Bill Landing, Dave Kadko, and Greg Cutter—for their support in a successful expedition. Further thanks are extended to Melissa Gilbert for ICP-MS support at the University of Southern Mississippi’s Center for Trace Analysis.

    Data Availability Statement

    Data used in this study are available at the Biological and Chemical Oceanography Data Management Office (DOI: 10.1575/1912/bco-dmo.772645.1) and the EarthChem Library (DOI: 10.1594/IEDA/100633).