Spatiotemporal Variability of Barium in Arctic Sea-Ice and Seawater Journal of Geophysical Research: Oceans

Freshwater export from the Arctic is critical in determining the density of water at sites of North Atlantic deep water formation, which in turn inﬂuences the global ﬂux of oceanic heat and nutrients. We need geochemical tracers and high-resolution observations to reﬁne our freshwater budgets and constrain models for future change. The use of seawater barium concentrations in the Arctic Ocean as a freshwater tracer relies on the conservative behavior of barium in seawater; while this has been shown to be an unreli-able assumption in Arctic summers, there are a lack of studies observing seasonal progressions. Here, we present barium concentrations from seawater and sea-ice collected during the Norwegian Young Sea ICE expedition from boreal winter into summer. We use other tracers (salinity, oxygen isotopes, and alkalinity) to reconstruct freshwater inputs and calculate a barium ‘‘deﬁcit’’ that can be attributed to nonconservative processes. We locate a deﬁcit in winter when biological production is low, which we attribute to uptake by barite formation associated with old organic matter or by internal sea-ice processes. We also ﬁnd a signiﬁcant barium deﬁcit during the early spring bloom, consistent with uptake into organic-matter associated microenvironments. However, in summer, there no strong barium deﬁcit near the surface, despite high biological production and organic carbon standing stocks, perhaps reﬂecting phytoplankton assemblage changes, and/or rapid internal cycling. Our ﬁndings challenge the assumptions surrounding the use of barium as an Arctic freshwater tracer, and highlight the need to improve our understanding of barium in sea-ice environments.


Introduction
The balance of fresh and marine waters in the Arctic is critical in determining the mixing of nutrients, highlatitude heat fluxes, and the density of seawater exported to the North Atlantic. There has been an increasing concern about changes in the characteristics of water masses in the Arctic in recent decades, in particular subsurface temperatures, and how these changes influence regional heat fluxes, biological production, and carbon export (Alexeev et al., 2013;Smedsrud, 2005). Understanding such processes and their influence on these climatically sensitive regions requires a robust quantification of freshwater inputs, and the subsequent physical, chemical, and biological responses of the Arctic system.
Total barium concentration, [Ba], has been used extensively in the Arctic Ocean-in conjunction with seawater salinity, oxygen isotope ratios (d 18 O w ), and total alkalinity (A T )-as a tracer of freshwater inputs (Cooper et al., 2008;Taylor et al., 2003). Salinity, d 18 O w , and A T are used together to calculate the total amount of meteoric water contributing to the freshwater flux (as opposed to sea-ice melt) (Chierici et al., 2011;Fransson et al., 2001); [Ba] is then used as a further nuanced constraint to determine the input of rivers (Dodd et al., 2009;Macdonald et al., 1999;Thomas et al., 2011) or oceanic basins (Abrahamsen et al., 2009). generally shows a ''nutrient-like'' profile, exhibiting a strong relationship in seawater with dissolved silicon and alkalinity on regional and global scales (Jacquet et al., 2005;Jeandel et al., 1996;Pyle et al., 2017). Furthermore, particulate barium-largely in the form of barium sulfate or barite-shows strong links with particulate organic carbon. Specifically, particulate ''excess Ba'' (Ba xs i.e., any Ba that is present in particles that is unsupported by lithogenic material) correlates with the flux of particulate organic carbon (POC) measured in suspended particulates Thomas et al., 2011), sediment traps , and seafloor sediments (Eagle et al., 2003). More recent studies, including those utilizing marine barium isotopes, also highlight the nonconservative behavior of barium in near surface waters (Bates et al., 2017;Horner et al., 2015;Hsieh & Henderson, 2017;Jullion et al., 2017).
These two critical assumptions may also be questioned in the specific conditions of the Arctic, in particular as a result of biological uptake in the Arctic summer, which leads to barium drawdown associated with organic matter (Abrahamsen et al., 2009;Thomas et al., 2011). Other studies suggest that nonconservative behavior may originate in high-latitude sea-ice zones that are independent of silicon cycling and other indicators of biological activity . Here, we examine these assumptions using a seasonal data set of sea-ice and underlying seawater samples collected in the region north of Svalbard, the Nansen Basin and shelf slopes, during the Norwegian young sea ICE expedition (N-ICE2015) (Granskog et al., 2016). Our analyses highlight the nonconservative behavior of barium in seawater under both true Arctic winter conditions in addition to during spring and summer algal blooms, which has important implications for the use of [Ba] as a tracer in Arctic waters.

Sample Collection
Seawater and sea-ice samples were collected on board the R/V Lance as part of the Norwegian Young Sea ICE expedition (N-ICE2015) project (January-June 2015; Figure 1a). The N-ICE drift expedition comprised four Floes from January to June 2015. Full details of the Floes are given in the supporting information. All seawater samples were collected using Niskin bottles deployed through holes in the sea-ice from the CTD rosette on the ship, or from casts taken through the sea-ice in a tent away from the ship, and stored in acidcleaned HDPE bottles for barium. Samples for salinity were stored in sealed glass bottles until onboard-ship analysis and for A T in borosilicate glass bottles until analysis on shore (Fransson et al., 2017). Ice cores were drilled with an ice corer (Mark II coring system, KOVACS enterprise, USA), subsectioned and allowed to melt in a sealed container in darkness overnight Fransson et al., 2015;Kauko et al., 2017) before being sampled cleanly into acid-washed HDPE bottles for barium and in borosilicate bottles for A T (Fransson et al., 2015).

Oceanographic Setting
The hydrography of the N-ICE2015 sections used in this study (Figure 1b, see supporting information Figure  S5 for T-S plot) was characterized by typical Arctic Ocean water column structure, with cold, fresh Polar Surface Water (PSW) separated by a strong halocline from underlying warm Atlantic Water (AW). The Mixed Layer Depth (MLD) during the N-ICE2015 expedition ranged from approximately 100-200 m in winter, shoaling to less than 20 m after 26 May concurrently with a shallowing of the pycnocline (Meyer et al., 2017b). The average depth of AW before 26 May was approximately 150 m and decreased to approximately 50 m after the 26 May 2015, over the Yermak Plateau (Meyer et al., 2017a). Vertical mixing in the Arctic Ocean results largely from the action of winds and internal tides (Rippeth et al., 2015). It should be noted that the N-ICE2015 ''winter'' and ''summer'' sampling does not strictly define a true seasonal time series: the winter samples are collected in a region that is likely to represent Arctic Ocean conditions; however, the summer samples are collected in a region of continual AW incursion near the slope environment where tidal forcing is stronger and the MLD is likely to be permanently shallow. A complete description of oceanographic conditions is presented in Meyer et al. (2017b). Heat fluxes from the warmer southern-sourced AW contribute significantly in this region north of Svalbard to the melting of Arctic sea ice, especially in regions of rough, shallow topography (Rippeth et al., 2015;Sirevaag & Fer, 2009;Smedsrud, 2005). During this study, the ocean-ice heat fluxes were highly variable and were driven strongly by wind forcing, internal tides, and water mass incursions (Koenig et al., 2016). Each drift started in pack ice, with distance to open water declining through time for each Floe (Meyer et al., 2017b). There was no thermodynamic ice formation during the period of study, although there were flooding and snow-ice transformations throughout the sampling period (Granskog et al., 2016;Provost et al., 2017). In January and February, there was relatively slow sea-ice drift with respect to wind speed, the sea-ice drift sped up later in the winter and into the spring concurrent with irreversible damage caused by a large storm events .
During the winter months, biological activity in the water column was very low (Chlorophyll a [Chl a] below $0.1 mg/m 3 ). After 26 May 2015, a phytoplankton bloom dominated by Phaeocystis developed under the ice due to strong lead formation (Chl a 1-8 mg/m 3 ), POC, and decline in total dissolved inorganic carbon (DIC; Assmy et al., 2016Assmy et al., , 2017Fransson et al., 2017). Diatoms were not the most abundant phytoplankton, although sea-ice diatoms (e.g., Nitzschia) were found predominantly in first-year ice with multiyear ice forming an important winter ''repository'' of sea-ice algae that supplied cells to younger ice in the spring and summer .

Analytical Methods
All samples were unfiltered and equilibrated over several months, such that the [Ba] measurements represent total reactive barium (Guay & Falkner, 1997). Seawater and sea-ice samples were prepared for isotope dilution (ID) under clean laboratory conditions (see Pyle et al., 2017, for full details). Aliquots of acidified seawater (samples and reference standards) were spiked gravimetrically with a 135 Ba-enriched solution (10 mg/ mL 135 Ba, Inorganic Ventures, Christiansburg, VA, USA, gravimetrically diluted to a working spike solution with a concentration of approximately 100 nM Ba using 3% Romil UpA HNO 3 ) to achieve a 138 Ba/ 135 Ba ratio of 0.65 to 1. The volumes of sample (or standard) and spike solution were determined gravimetrically, and were varied according to the estimated seawater [Ba] in order to keep the spike-to-sample ratio at approximate unity to minimize error magnification.
Spiked samples were left to equilibrate overnight, and were vortexed prior to analysis. All samples and standards were analyzed using inductively coupled plasma mass spectrometry (Bristol Isotope Group, Thermo Element 2). Barium isotopes 135 Ba, 137 Ba, and 138 Ba were measured in low resolution using an SEM detector in counting mode. 137 Ba was measured to monitor for interferences on 135 Ba and 38 Ba. The precise isotopic composition of the spike solution was determined by calibration, utilizing a reverse-ID of blended solutions containing a mixture of the spike solution and a Ba-natural standard from High Purity Standards. The spiked samples and standards were diluted typically 20-fold with 3% Romil UpA HNO 3 , although the dilution factor depended on the original salinity. Matrix effects as a result of the seawater salts were assessed by routinely analyzing two solutions of dilute seawater standard NASS-6 alongside a 1 ppb Banatural standard solution (High Purity Standards): the first NASS-6 solution was diluted 20 times in 3% HNO 3 , and was used to correct the intensity signal of the second NASS-6 solution, which was diluted 20 times in the 1 ppb Ba-natural standard. The depression or amplification of the ''true'' signal as a result of the dilute seawater matrix was calculated using the difference in signal intensity between the corrected second NASS-6 solution and the 1ppb Ba-natural standard. This correction varied between runs, but typically resulted in a 20% modification of the blank on both 135 Ba and 138 Ba, and did not impact the isotopic ratio significantly. A mass bias correction coefficient (typically 2%) was calculated using the deviation of the second NASS-6 solution 138 Ba/ 135 Ba ratio from the average natural ratio of 138 Ba/ 135 Ba (10.88). The final concentration was calculated using equation (1) (Dickin, 1995): Where Conc sample is the sample [Ba]; Conc spike is the spike [Ba]; m is the mass of sample, spike, or natural standard; R is the ratio of 138 Ba/ 135 Ba in sample, spike, or natural standard; f is the abundance of 135 Ba in spike or natural standard; and K is mass bias correction coefficient for dilute seawater (ratio of literature to measured ratios in dilute seawater). Uncertainties were fully propagated and were typically less than 0.1%. Duplicates were run for every sample, and agreed typically within 0.5%. Reference seawater standards were run routinely to assess precision and accuracy: NASS-6 ([Ba] 5 46.9 nmol/kg or 48.1 nM assuming a density of 1.025 kg/L, 60.9%) and SO ([Ba] 5 71.6 nmol/kg or 73.4 nM assuming a density of 1.025 kg/L, 60.7%), agreeing well with previous studies and other laboratories .
Seawater samples were analyzed for oxygen isotope ratios at the British Geological Survey. Oxygen isotope (d 18 O) measurements were made using the CO 2 equilibration method with an Isoprime 100 mass spectrometer plus Aquaprep device. Isotope measurements used internal standards calibrated against the international standards VSMOW2 and VSLAP2. Errors are typically <0.05& for d 18 O. Results were reported in standard d 18 O notation (equation (2)), with reference to a standard (Vienna Standard Mean Ocean Water [VSMOW]). Salinity measurements were carried out on board using a Guildline Portasal salinometer, with a typical uncertainty of 60.003 (Dodd et al., 2017;Meyer et al., 2017b).
Details of nutrient and Chl a analyses and uncertainties are detailed in . Briefly, nutrient analysis was carried out spectrophotometrically using an adapted Scalar autoanalyzer, and pigments were analyzed on board fluorometrically. Seawater A T was measured by potentiometric titration with 0.1 N hydrochloric acid using a Versatile Instrument for the Determination of Titration Alkalinity (VINDTA 3S, Marianda, Germany), and accuracy was assessed using analyses of Certified Reference Materials (see Fransson et al., 2017 for details). A T in melted sea ice samples were analyzed using potentiometric titration with 0.05 N hydrochloric acid at the Institute of Marine Research (IMR), Tromsoe, Norway, using method described in detail by Fransson et al. (2015).

Seawater Barium Measurements
Seawater [Ba] values range from 38.4 to 51.3 nmol/kg, and exhibit relatively low variability and absolute values compared to previous Arctic studies (Abrahamsen et al., 2009;Dodd et al., 2009;Thomas et al., 2011). Despite this low variability, the [Ba] profiles still show characteristic enrichment at depth, with some subsurface peaks above the halocline (Figure 2).
[Ba] showed significant correlations with salinity, A T , d 18 O w , nitrate, and-most significantly-silicic acid (Table 1). Separate regression analysis for the winter and spring sampling sites may provide further insight, given the very different environmental conditions relating to the water depth, depth of AW, MLD, and biological activity of the two regimes. Multivariate linear analysis of the seawater samples prior to the late-May algal bloom (before 26 May 2015) indicates that [Ba] can be predicted using silicic acid alone (p < 0.001, n 5 124), with little significance added from A T and d 18 O w . However, a similar statistical analysis of the samples after to the late-May algal bloom (post 26 May 2015) reveals that [Ba] can be predicted instead using salinity (p 5 0.008, n 5 21) and nitrate (p 5 0.044) rather than silicic acid (p > 0.05).
Alternatively, we can use Principal Component Analysis (PCA) to transform the complex, multiparameter data set to a smaller number of axes, in order to investigate the links between the different parameters ( Figure 3). PCA was conducted using a covariance matrix after data normalization for ''shallow'' (<50m) and ''deep'' (>50m) waters. Chi-squared testing showed that the eigenvalues were significantly different in each case.
For the shallow samples, the eigenvalues for PC1, PC2, and PC3 represent a cumulative 73% of the covariance matrix. PC1 has a strong loading with parameters such as date, d 18 O w , nitrate, and silicic acid (which show high intercorrelation), with significant loading with pressure, temperature, and salinity, suggesting this is a strong seasonal signal. PC2 has a strong loading with A T and-again-with significant loading with pressure, temperature, and salinity. However, [Ba] shows no significant loading with A T , or nutrients on these axes. PC3 has a strong loading with [Ba], a significant loading with salinity, but no other significant loadings. There are significantly different component scores between the winter (January-March) and summer (April-June) samples for PC1 (Mann-Whitney U tests: PC1 p < 0.001) but not for either PC2 or PC3.
For the deep waters, the eigenvalues for PC1 and PC2 represent a cumulative 72% of the covariance matrix ( Figure 3). PC1 and PC2 both have significant loadings from all the key environmental parameters, with potential intercorrelation between [Ba], silicic acid, and pressure. Furthermore, there are significantly different component scores between the winter (January-March) and summer (April-June) samples (Mann-Whitney U tests: PC1 p 5 0.029; PC2 p < 0.001).

Barite Saturation Indices in Seawater
Barium saturation indices were estimated following Monnin (1999) and Thomas et al. (2011) using the following equation: Where m denotes the molality of the species in aqueous form, K sp is the solubility product, c is the total (or stoichiometric) mean activity coefficient of aqueous barium sulfate (for the general conditions here the following assumptions were made: log K sp 5 210.482, c 5 0.128 kg/mol given water temperatures of 1 6 28C, Monnin, 1999), and Q is the ion activity product of aqueous barium sulfate. The molarity of sulfate was calculated assuming seawater concentrations of 28.23 mM at a standard salinity of 35. The molality values for sulfate were converted from molarity using seawater density calculated from temperature, salinity, and pressure measurements from the CTD. The calculations suggest the whole water column is always  undersaturated with respect to barite (saturation index 0.5-0.7); any barite formed in the water column must be doing so associated with organic matter and biological production (e.g., Bertram & Cowen, 1997;Collier & Edmond, 1984;Ganeshram et al., 2003;Gonz alez-Munoz et al., 2003;Paytan & Kastner, 1996).

Sea-Ice Barium Measurements
Sea-ice section [Ba] values were highly variable and ranged from 5.8 to 79.3 nmol/kg. The relationships between [Ba], salinity, d 18 O w , and nutrients in the sea-ice were also highly variable, and likely related to the structure and of the sea-ice, and whether rafting had occurred (Figure 4). Sea-ice [Ba] did not show significant correlations with salinity, A T , d 18 O w , or nitrate, and a weakly significant relationship with silicic acid (Table 2). Multivariate linear analysis also indicates only a weakly significant correlation exists between [Ba] and silicic acid (p $ 0.02), and salinity (p $ 0.01).

Freshwater Decomposition Analysis and Barium Deficiency
The proportion of freshwater fluxes can be calculated from the measured seawater parameters (salinity, A T , and d 18 O w ) using mass balance equations assuming characteristic end-member values for sea-ice, meteoric water, and AW. We calculated the input of the different freshwater sources using two different approaches, using the salinity measurements and either the A T or d 18 O w measurements and end-member values in Table  3 to solve equations (4), (5) and either (6) or (7), respectively (Abrahamsen et al., 2009;Cooper et al., 2008;Dodd et al., 2009;Macdonald et al., 1999;Meredith et al., 2001Meredith et al., , 2008Taylor et al., 2003;Thomas et al., 2011). A where f SI , f MW , and f AW are the fractions of sea-ice melt, meteoric water, and seawater end-members, respectively, that we seek to determine; S SI , S MW , and S SW are the respective salinities of the end members; d SI , d MW , and d SW are their corresponding d 18 O w values; and A TSI , A TMW , and A TSW are their corresponding alkalinities (Table 3). We assessed the meteoric end-member using data from the upper 400 m of the water column. In the case of the seawater end-member, we plotted A T versus S to obtain a linear fit (equation (8)). Given that the highest salinity was 35.06 this results in a seawater endmember A T of 2312 mmol/kg (Fransson et al., 2001). Mean sea-ice salinity was calculated (range salinity of 0-15), truncating values at 15. Seaice A T (370 mmol/kg) at mean sea-ice salinity (5.8) was calculated obtained from sea-ice A T versus sea-ice salinity linear relation A T 5 63.84 *S (r2 5 0.97; N 5 69) based on N-ICE2015 A T data, where outliers from the regression line were removed due to nonlinear processes such as calcium carbonate precipitation/dissolution, affecting A T (Fransson et al., 2013(Fransson et al., , 2017. For comparison, assuming a simple mixing line with seawater without nonlinear sea-ice processes, A T of 382 mmol/ kg was obtained (Chierici et al., 2011;Thomas et al., 2011); using this calculated value in the mass balance calculations yields the same results within uncertainty as the observed value (supporting information). Sea-ice d 18 O was taken as the maximum value in the data set; seawater d 18 O was taken as the value the water column data converges to at depth. The meteoric water inputs represent the greatest uncertainties in these mass balance calculations, as there are no direct measurements and a wide range of seasonality (leading to a range in d 18 O w ) for the nonmarine freshwaters. Sensitivity testing, using a range of possible A T and d 18 O values for the end-member compositions, shows that the uncertainties in our calculated freshwater contributions are unlikely to be significant relative to the water column variability (supporting information). Furthermore, these end-member uncertainties are likely to be systematic and do not necessarily impact the patterns and trends observed in the data .
Last, we converted the [Ba] Def into a carbon flux, assuming a C org :Ba xs value of 225 g C (g Ba) 21 (Thomas et al., 2011). However, assuming that there is a consistent link between the ''barium deficit'' and carbon uptake is problematic; the reconstructed carbon fluxes are very similar for the winter and summer samples, for example using the A T -based reconstruction results in all profiles yielding approximately 0.4-2.3 mol C m 22 . While these values are the same order of magnitude as the measured integrated POC levels in the water column (Figure 2), they do not reflect the temporal variation in biological standing stocks.

Discussion
The water column [Ba] measurements, statistical analysis, and freshwater calculations point toward an uptake of barium via nonconservative processes that are active in seawater and sea-ice in winter as well   Thomas et al. (2011) andChierici et al. (2011). as during the biologically productive summer months. There are a series of questions that arise from these findings: first, are the freshwater component mass-balance calculations reliable? And, second, if so, what are the processes that result in a strong [Ba] Def in the water column in winter and early summer, before the spring bloom? Here, we will assess the assumptions made when calculating [Ba] Def , and the possible processes that can explain the ''seasonal'' patterns that are observed.

Assessing the Freshwater Reconstruction
The method of separating freshwater inputs using a combination of salinity measurements and either d 18 O w or A T is well-established, with-in each case-the resulting derived freshwater fractions reported here as percentages with typical errors of <1% on point values that can be attributed to variations in endmembers (i.e., the sea-ice, meteoric, and oceanic compositions) (e.g., Meredith et al., 2013, and sensitivity testing-see supporting information; Pyle et al., 2017). If these end-members are invariable in time and space, then these uncertainties will be systematic across the whole data set. The generally good agreement when carrying out the calculations using either d 18 O w or A T suggests that the reconstructed freshwater components are reliable, at least for determining relative spatial and temporal changes. However, if these end-members are temporally or spatially variable, then there could be some bias introduced into the calculations. A further problem arises when subsequently calculating the [Ba] Def from this method; the calculation assumes that there is one, uniform riverine composition, and does not take into consideration seasonal or spatial changes in river input.

The Behavior of [Ba] in the Water Column
The statistical analysis of the N-ICE2015 water column data set reveals a complex relationship between [Ba] and other environmental parameters, which may impact on the utility of dissolved barium as a tracer of water masses. In deeper waters, there is likely a link between [Ba] and the supply of underlying AW: for samples >50 m water depth there is some clustering with [Ba] and other parameters in the Principal Component Analysis (PCA) plots (Figure 3), suggesting some intercorrelation with silicic acid and pressure (i.e., depth). However, the results of our multivariate analyses seem to suggest that [Ba] is influenced by different factors at different times of year. This likely reflects the different oceanographic conditions rather than temporal changes per se, with ''true'' Arctic water masses in the winter, compared to the location over the Yermak Plateau occupied in the summer, where there is permanent AW incursion at depth.
In contrast to deeper waters, our PCA analysis showed that surface water [Ba] (<50 m) has little relationship to any parameter except salinity. PC1 revealed a strong seasonal signal, which likely reflects variations in the mixed layer depth (MLD) and the strong variability in meltwater in surface waters. Although 50 m was selected as an approximate average MLD for the cut-off between ''shallow'' and ''deep'' for the purposes of the PCA analysis, MLDs were much shallower in the late summer, with AW found up to the surface during some periods later in the drift expedition (Meyer et al., 2017b). However, this change in MLD does not drive the variability in [Ba] observed in the surface waters, as the PCA plots do not reveal any clustering between [Ba] and other parameters or any significant loading of [Ba] on PC1 (Figure 3). Instead, [Ba] only significantly contributes to the loading of PC3, which also has some contribution from temperature and salinity andweakly-from A T and pressure.
Our results suggest that it is challenging to associate [Ba] straightforwardly with another environmental parameter in the near surface ocean-except potentially salinity most likely as a result of riverine inputsand that there is no clear difference in the behavior of surface [Ba] between the winter and summer months. It is possible that there is a further environmental parameter that impacts [Ba], which was not included in the PCA, or that the variability in [Ba] was too low to detect intercorrelations with other parameters effectively.

Mechanisms Behind the Seasonal Changes in Ba Deficiency in Seawater
The seawater profiles illustrate that there are potential barium ''deficits'' in the water column throughout the winter and spring that are not observed strongly in the summer, regardless of how the freshwater input is calculated. The spring barium deficit occurred in late April-May, and is likely a result of Ba uptake associated with biological material produced during the early under-ice production when moderate Chl a concentrations were detected at approximately 40 m . However, Chl a and POC levels in winter were low and biological production is unlikely to have formed an efficient sink of barium.
Assuming the [Ba] Def calculations are reflecting the water column conditions, there are several possible explanations for this variability in barium uptake that relate to conditions external to the water column system. First, processes within sea-ice-either during its earlier formation or during its later development and snow-ice transitions-could cause nonconservative behavior of barium that leaves a depleted surface water signal behind (see section 4.4). Evidence for a sea-ice influence comes from a comparison between the calculated [Ba] Def and the meltwater mass balance calculations in winter. The strong barium deficit in winter seems to correspond to a higher meteoric water input, which is unexpected if riverine inputs are assumed to be the major supplier of dissolved Ba. However, there is also a negative contribution to freshwater from sea-ice, indicative that freshwater had been lost from that particular water mass as a result of sea-ice formation prior to sampling ( Figure 6).
Second, barium could be taken up by association with slowly remineralizing, old organic matter that remains in the water column from the previous summer or autumn, rather than new production (Reigstad et al., 2002). The N-ICE2015 observations reveal that there was POC present in the water column during winter, despite very low Chl a concentrations in the near surface water (Figure 2), suggesting that there could have been old POC remaining from the previous summer Olsen et al., 2017). The older, more recalcitrant organic matter could remain in the water column for a sufficient amount of time to develop barite saturation within microenvironments.
Third, there could be variable depletion of [Ba] by association with organic matter, as a result of spatial and temporal differences in phytoplankton community structure. Only a very small proportion of particulate barium in seawater is directly incorporated into biological material (Horner et al., 2015), but is most likely adsorbed or precipitated as barite in microenvironments within the organic matter. Given that there is no fixed stoichiometry between dissolved barium and inorganic nutrients (Collier & Edmond, 1984), it is possible that the capacity for barium uptake is dependent on the properties such as the chemical composition of both skeletal elements and soft parts, cell size, and density. Barium uptake may be greater in diatomdominated communities compared to the Phaeocystis-dominated communities observed during the summer months of N-ICE2015 . Furthermore, the dominance of Phaeocystis after 26 May could potentially explain why there is a weaker coupling between [Ba] and silicic acid, and a stronger relationship with nitrate, in the later part of the study.
Last, there could be rapid remineralization of barium in near surface waters during the summer due to an enhanced microbial loop resulting from warmer temperatures, changes in sea-ice conditions or proximity to the shelf. Rapid and shallow turnover, with a strong halocline, may result in barium becoming effectively trapped near the surface. This process, again, could also explain why there appears to be a stronger relationship between [Ba] and nitrate in summer, rather than silicic acid, given the former has the shallower remineralization depth. Such changes in remineralization could also be driven by phytoplankton community structure.
The lack of a strong barium deficit in summer may be driven by a number of nonmutually exclusive processes. There could be additional barium inputs near the shelf, as a result of season-dependent meteoric water compositions (Colbert & McManus, 2005;Cooper et al., 2008;Guay & Falkner, 1997; different AW concentrations due to shelf sediment remobilization or diffusion, especially in suboxic sediments below highly productive regions (McManus et al., 1994); or changes in water mass age or ventilation.

Sea-Ice Barium: Non-Conservative Behavior or Multiple End-Member Mixing?
Barium in the sea-ice samples did not fall on a simple linear mixing line that would represent conservative dilution by low-[Ba] freshwater (Figure 7), which has been observed for trace elements-including barium-in Antarctic sea-ice (Hendry et al., 2009;Lannuzel et al., 2011). This observation means that either nonconservative processes are active within sea-ice on or that the system cannot be described as simple mixing between two end-members, and sea-ice could be a source or sink of barium that would invalidate the assumptions that form the basis of equation (8).
For nonconservative behavior within sea ice, barium must be removed from or added to solution independently of salinity changes. Sea-salt ions are rejected from the ice crystal lattice as sea ice forms, forming pockets of brine that become trapped in a network within the ice; the trapped, dense brine drains downwards through brine channels under the influence of gravity, typically leaving upper layers of ice fresher (Weeks & Ackley, 1982). The extent of this desalination is determined by the permeability of the sea ice, which in turn is controlled by the structure of the ice itself and its thermodynamic properties, with freezing in brine channels decreasing porosity (Backstrom & Eicken, 2006;Golden et al., 1998). Within these brine channels, barium could be removed from solution into solid phases either by adsorption onto particulates, or through the precipitation of barite. The formation of barite may be promoted in microenvironments associated with the concentration of Ba 21 and SO 2-4 ions in brine channels released from the decay of organic matter, promoted by a dynamic sea-ice microbial community (Lannuzel et al., 2011;Palmisano & Garrison, 1993;Stroobants et al., 1991). Provided that such barite crystals did not redissolve, this mechanism would remove barium from solution independently of salinity fluctuations. However, if this were the primary mechanism driving the dissociation between [Ba] and salinity (supporting information) that is observed in the Arctic sea ice core samples, a correlation could be expected between [Ba] and the concentrations of dissolved inorganic nutrients (as indicators of productivity), which is not typically observed in these samples. Barite precipitation may also be able to occur within sea ice without the aid of microbial microenvironments. SO 2-4 ions make up approximately 8% of the mass of salt ions in sea water, and it is possible that brine pockets may reach high enough concentrations of barium and sulfate to cause abiotic precipitation of barite.
An alternative explanation is that there is an additional input, i.e., a third or higher-order end-member-of the barium within the sea-ice (Figure 7), which could derive from various possible sources. Whilst it appears that seawater quantitatively supplies the [Ba] in Antarctic sea-ice (Lannuzel et al., 2011), is it possible that the higher atmospheric deposition of dust in the northern high latitudes results in a nontrivial contribution of atmospheric barium to Arctic sea-ice? This hypothesis is supported by the relatively high normalized [Ba] content of the snow pack (Figure 7), and could be particularly the case in this study as samples were not filtered on collection.

Conclusions
The high-latitude Arctic is a critical region in determining global distributions of heat, carbon, and nutrients through its role in controlling the density of water exported to the North Atlantic and, so, deep-water formation in the Labrador Sea. Freshwater contributions are commonly calculated using geochemical tracers, for use in models that are critical for assessing future oceanographic change. We present new observations that challenge the assumptions surrounding the use of barium as a freshwater tracer of riverine input in the Arctic, showing nonconservative behavior even during periods of low biological productivity, most likely as a result of processes occurring in sea-ice microenvironments. In contrast, in the high productivity summer months, we find surplus near surface barium, despite high standing stocks of organic carbon, perhaps reflecting phytoplankton assemblage changes, or/and rapid internal cycling. Our findings highlight the need to better understand the behavior of barium in high-latitude regions, and further insight will be gained from detailed examination of particulate barium in sea-ice samples, and through barium isotope process studies.