Water samples were collected during three summer cruises (August–September) on the USCGC Healy in 2010, 2011, and 2012 (HLY1002, HLY1102, HLY1202; Figure 1; Robbins et al., 2013a,b,c, 2014). Surface waters were collected at a depth of ~8 m using an underway flow-through system. Profile samples were collected at various depths (2–3800 m) at 17 stations using a rosette sampler equipped with 24 Niskin bottles and a Conductivity-Temperature-Depth (CTD) probe (Sea-Bird SBE45) used to measure water salinity, temperature, and pressure. A total of 770 water samples were collected for analyses of DOC, CDOM, and δ¹⁸O, with about 700 matching samples for inorganic nutrient analysis. Samples for measurements of DOC and CDOM were collected in pre-cleaned 60-mL amber glass bottles and frozen immediately at −20°C until analysis in the laboratory. Samples for measurement of nutrient concentrations (phosphate, nitrate + nitrite, ammonium, and silicate) were frozen immediately at −80°C in scintillation vials prior to analysis. Samples for measurement of oxygen isotopic composition (δ¹⁸O values) were collected in glass vials and stored at 4°C prior to analysis. Nutrient analyses were completed using a SEAL AA3 four-channel segmented flow analyzer at the Nutrient Analytical Facility at the Woods Hole Oceanographic Institution, Woods Hole, MA. Additional total silicon concentrations were measured on separate samples using an Agilent Inductively-coupled Plasma Mass Spectrometry at the University of South Florida, St. Petersburg, FL. δ¹⁸O values are reported on the Standard Mean Ocean Water (SMOW) international reference scale and were measured by Isotope Ratio Mass Spectrometry (Thermo Finnigan Delta V) at the University of South Florida, Tampa, FL.

The broad spatial scale of sampling in this study covered a wide range of environments that varied substantially in sea-ice cover (Figure 1). Surface waters in the Bering Sea, Chukchi Sea, and southern Canada Basin were mostly ice-free, whereas areas north of 80°N were largely covered by sea ice during the sampling periods. Sea-ice area in the Arctic Ocean reached a low of 3.0 × 10⁶ km² during the HLY1202 cruise (Parkinson and Comiso, 2013). Comparisons in summer sea ice cover among 2010, 2011, and 2012 revealed a notable ice reduction in 2012 in the northern Chukchi Borderland, southern Mendeleev Plain, and the central and southern areas of the Canada Basin (Figure 1). In this study, the geographic names and boundaries of oceanic regions were defined according to Marine Regions (http://www.marineregions.org) with modifications.

Water samples for measurements of DOC and CDOM were thawed and filtered in the laboratory through pre-combusted (450°C, 4 h) Whatman GF/F filters (0.7-μm pore size) prior to analysis. Concentrations of DOC were measured by high temperature combustion using a Shimadzu total organic carbon (TOC) TOC-V_CSN analyzer equipped with an autosampler. The Shimadzu platinum catalyst was conditioned by repeated injections of acidified Milli Q UV-Plus water (pH 2–3) prior to sample analysis (Benner and Strom, 1993). Before injection, samples were acidified with 2 mol L⁻¹ hydrochloric acid (HCl) to pH 2–3 and were sparged with ultra-high purity oxygen for 2 min to remove the inorganic carbon as CO₂. The injection volume was 100-μL. Milli Q water and deep-sea reference standards (University of Miami) were injected every 6th sample. Values of blanks (Milli Q UV-Plus water) were negligible throughout the measurements of seawater samples. The coefficient of variation (CV) of four repeated injections of a given sample was typically ±0.9%. The final concentrations of DOC were quantified using an external calibration curve generated with five concentrations of standards (glucosamine; Sigma Chemical CO.) that bracketed the range of values observed in the samples.

Absorbance spectra (250–800 nm) of filtered water samples were measured using a Shimadzu ultraviolet-visible 1601 dual-beam spectrophotometer and 5-cm cylindrical quartz cuvettes. Water samples were warmed to room temperature prior to analysis. Absorbance was blank corrected (the average absorbance between 690 and 700 nm) and then converted to Napierian absorption coefficient, a_λ (m⁻¹), using Equation (1) (1)aλ=2.303Aλr where A_λ is the absorbance measured across path length r at a wavelength of λ. The spectral slope coefficient (S) was calculated using a linear regression of log-linearized a_λ1 as in Equation (2) (2)aλ1=aλ2exp[−S(λ1−λ2)] where, a_λ1 and a_λ2 are absorption coefficients at wavelengths λ₁ and λ₂ (λ₁ > λ₂). S in the 275–295 nm spectral range is reported as S_275–295 with units of μm⁻¹.

Concentrations of dissolved lignin phenols were retrieved from CDOM absorption coefficients as in Fichot et al. (2016), using Equations (3, 4). When a₂₅₀ is <4 m⁻¹, (3)ln(TDLP9)=0.7672×a263−0.3987 when a₂₅₀ is ≥4 m⁻¹, (4)ln(TDLP9)=−2.282×ln (a350)−8.209×ln (a275)                              +11.365×ln (a295)+2.909 Concentrations of total dissolved lignin phenols (TDLP₉) are the sum of nine dissolved lignin phenols (p-hydroxybenzaldehyde, p-hydroxyacetophenone, p-hydroxybenzoic acid, vanillin, acetovanillone, vanillic acid, syringaldehyde, acetosyringone, and syringic acid) and are reported in units of nmol L⁻¹. These two models were constructed from paired measurements of dissolved lignin and CDOM in oceanic samples that were collected mostly in the Arctic Ocean (Fichot et al., 2016) and are therefore applicable to this study. Herein, the models were applied only to the upper 300 m of the water column (i.e., polar mixed layer and halocline waters) where riverine inputs were evident from the measured oxygen isotopic composition (δ¹⁸O) and calculated meteoric water fraction.

The effects of filtration and freeze/thaw on CDOM properties of seawater were determined using waters collected in the surface (45 m) and deep (1200 m) Atlantic Ocean. Six replicates of seawater samples were collected from each depth in pre-cleaned (450°C, 4 h) glass vials. The samples divided into three treatments: (1) filtered and refrigerated at 4°C; (2) filtered and frozen at −20°C; (3) frozen without filtration and filtered prior to analysis. For each treatment, one set of replicates (n = 3) was filtered through GF/F filters (0.7-μm pore size; Whatman®) and the other set (n = 3) was filtered through Supor membranes (0.2-μm pore size; Supor®-200, Life Sciences). The refrigerated samples were analyzed immediately upon return to the home laboratory (within 6 days of collection), and the frozen samples were stored at −20°C for 28 days before analysis. The analysis of CDOM absorbance and calculations of absorption coefficient (a_λ) and spectral slope coefficient (S_275–295) followed procedures described in the previous section.

Seawater in the Arctic Ocean can be characterized as a mixture of freshwater sources, such as meteoric water (MW; precipitation and river runoff) and sea-ice melt (SIM), with seawater sources such as Pacific water (PW), and Atlantic water (ATW). Each water source can be characterized by a distinct combination of chemical tracers that range from conservative (e.g., salinity and oxygen isotopic composition, i.e., δ¹⁸O values) to semi-conservative (e.g., nutrient concentrations or alkalinity). For example, Arctic MW, including river runoff and direct precipitation, has a distinct combination of very low salinity and nutrient concentrations combined with ¹⁸O-depleted δ¹⁸O values (≈−20%0). SIM has low salinity (slightly above zero) and nutrients due to brine rejection, combined with δ¹⁸O values similar to that of its source water due to a relatively small ¹⁸O/¹⁶O fractionation factor (about +2.6%0, i.e., sea ice is more enriched in ¹⁸O than the surface seawater; MacDonald et al., 1995; Melling and Moore, 1995). Thus, two tracer parameters, salinity and δ¹⁸O, provide a robust estimate of MW and SIM additions to seawater, the latter of which is characterized by high salinity and δ¹⁸O values near 0 (Robbins et al., 2013a). In order to further separate seawater sources, a third tracer is necessary, and various semi-conservative parameters have been used (e.g., silicate, phosphate, nutrient ratios, etc.) with variable results (Alkire et al., 2015). In general, these are based on the observation that PW has higher concentrations of nutrients compared to ATW as well as somewhat lower salinity and δ¹⁸O values.

Using this mass-balance method, fractions of n water sources can be calculated by solving a set of n linear equations, which require (n − 1) tracers (i.e., salinity and δ¹⁸O separate water into 3 sources, one additional tracer is required to separate into four sources). In this study, we utilize only the freshwater source estimates of SIM and MW. Thus, variations between ATW and PW sources that depends on the tracer method chosen do not affect the results here. We use the full three-parameter tracers as follows: fMW+fSIM+fPM+fATW=1fMW(SMW)+fSIM(SSIM)+fPW(SPW)+fATW(SATW)=SobservedfMW(δ18OMW)+fSIM(δ18OSIM)+fPW(δ18OPW)                                +fATW(δ18OATW)=δ18OobservedfMW(SiMW)+fSIM(SiSIM)+fPW(SiPW)                                +fATW(SiATW)=Siobserved where, f, S, and Si refer to fraction, salinity, and silicate concentration, respectively. The subscripted abbreviations (MW, SIM, PW, and ATW) denote the four water sources. End-member values of the tracers used follow Bauch et al. (1995), Yamamoto-Kawai et al. (2008), and Robbins et al. (2013a): MW: S = 0, δ¹⁸O = −20%0, Si = 10 μmol kg⁻¹; SIM: S = 4, δ¹⁸O = δ¹⁸O_surface +2.6%0, Si = 1 μmol kg⁻¹; PW: S = 32.5, δ¹⁸O = −0.8%0, Si = 40 μmol kg⁻¹; ATW: S = 34.87, δ¹⁸O = 0.3%0, Si = 6 μmol kg⁻¹. We follow Bauch et al. (1995) in using dissolved silicate concentration (Si) as the third semi-conservative tracer, but f _PW and f _ATW are not used in this study.

Statistical differences between variables that were normally or not normally distributed were assessed using independent Samples T-test (two-tailed, α = 0.05) or Mann-Whitney U-test (two-tailed, α = 0.05), respectively. The normality of data distribution and equality of variances were tested using the Kolmogorov–Smirnov test (two-tailed, a = 0.05) and Levene's test (two-tailed, a = 0.05), respectively. The statistical analyses were performed in SPSS 20.0 (IBM Statistical Package for the Social Sciences Inc.). The significance of least-squares linear regression was determined using the enter approach in SPSS.

A total of 132 DOC reference standards (DOC_ref) were measured along with the Arctic samples to evaluate the performance of the DOC analyses (Supplementary Table 1). Concentrations of DOC_ref ranged from 42.2 to 45.9 μmol L⁻¹ (avg.: 44.1 ± 0.9 μmol L⁻¹) and were well within the reported range (42–45 μmol L⁻¹). Concentrations of DOC_ref determined during sample analysis for the three cruises showed similar variability and the means were not significantly different (T-test, p > 0.05). Precision of the analysis, represented by coefficient of variation (%) or standard deviation (μmol L⁻¹) of DOC_ref concentrations, was better than 2.0% or 1.0 μmol L⁻¹. These results demonstrate high consistency and reproducibility of the DOC analysis, which facilitates evaluation of the natural variability in DOC concentrations.

Water samples for optical measurements are preferably stored at 4°C, but logistical issues often require the storage and shipping of frozen samples. Some studies observed substantial changes in CDOM absorbance and fluorescence intensity in freshwater samples following freeze/thaw (Spencer et al., 2007; Fellman et al., 2008; Hudson et al., 2009), while others did not (Spencer et al., 2010; Yamashita et al., 2010). Very few studies have determined the influence of freeze/thaw on CDOM properties in seawater. In this study, comparisons between refrigerated (4°C) and frozen (−20°C) seawater samples showed relatively minor differences (<10%) in CDOM absorption coefficients at 250–400 nm and in S_275–295, regardless of sampling locations (Supplementary Figure 1). Freezing appeared to result in some loss of CDOM at lower wavelengths, which was less apparent in samples that were stored unfiltered (Supplementary Figures 1C,D). Paired analyses of 0.7- and 0.2-μm filtered samples in each experiment showed similar (mostly <5% difference) CDOM absorption coefficients and S_275–295. These results indicated a minor impact of freezing on CDOM properties in seawater, consistent with previous assessments using seawater samples (Del Castillo and Coble, 2000; Conmy, 2008). In support of this, CDOM absorption coefficients and S_275–295 measured in the HLY1102 samples in this study were very comparable to those measured in refrigerated (4°C) samples collected along a similar cruise track during the same period of time (Guéguen et al., 2015).

This study presents detailed distributions of dissolved lignin phenols in the upper Arctic Ocean. To evaluate the validity of the CDOM-derived lignin values, we compared the calculated TDLP₉ concentrations to those measured previously using the CuO oxidation method in similar regions of the Arctic Ocean. Surface TDLP₉ concentrations estimated here in the Beaufort Sea, Canada Basin, and Makarov Basin were 2–22, 2–4, and 9–20 nmol L⁻¹, respectively. These values are similar to those measured in the Mackenzie River Plume (8–30 nmol L⁻¹), Beaufort shelf (3–24 nmol L⁻¹), southern Canada Basin (mostly 2.2–3.0 nmol L⁻¹), and eastern Makarov Basin (10–20 nmol L⁻¹) in samples collected between 2002 and 2009 (Fichot et al., 2016; Benner, unpublished; Kaiser et al., in review). These comparisons from similar locations but in different years indicate the CDOM–derived estimates of lignin phenol concentrations are within the ranges of measured values.

Surface distributions of salinity, meteoric water, and sea-ice melt were highly variable in the western Arctic Ocean (Figure 2 and Table 1). The most saline surface waters were observed in the Makarov Basin (30.50 ± 0.78), whereas the least saline waters were found in the Beaufort Sea (24.81 ± 3.32). Surface salinities in the Canada Basin (26.54 ± 2.06) were among the lowest measured in the deep basins. A notable change in surface salinity (by ~3) occurred across the ice front at ~78°N in the Canada Basin and near the boundary between the Chukchi Borderland and Mendeleev Plain (Figure 2A).

Fractions of meteoric water (f _MW) in coastal regions were low in the Bering-Chukchi Seas (0.00–0.15) and high in the Beaufort Sea (0.12–0.32) (Figure 2B and Supplementary Table 2). The f _MW values in the deep basins were relatively constant, averaging 0.18 ± 0.01, 0.17 ± 0.01, and 0.15 ± 0.02 in the Canada Basin, Mendeleev Plain, and Makarov Basin, respectively (Supplementary Table 2). Fractions of sea-ice melt (f _SIM) were highly variable in coastal regions, varying from −0.04 to 0.15 in the Bering-Chukchi Seas and from −0.02 to 0.20 in the Beaufort Sea (Figure 2C and Supplementary Table 2). Negative f _SIM indicate brine rejection from sea-ice formation. The f _SIM in the deep basins showed a reverse trend to that of salinity, with elevated values in the southern Canada Basin (0.10–0.20) and negative values in the Makarov Basin (−0.04–0.02). A rapid northward decrease in f _SIM also occurred across the ice front at ~78°N (Figure 2C).

The DOC and TDLP₉ concentrations and values of a₃₂₅ and S_275–295 in surface waters of the western Arctic Ocean varied ~3.5-fold (42–148 μmol L⁻¹), ~22-fold (1–22 nmol L⁻¹), ~13-fold (0.16–2.06 m⁻¹), and ~1.8-fold (21.5–38.9 μm⁻¹), respectively (Figure 3). The highest DOC, a₃₂₅, and TDLP₉, and lowest S_275–295 were observed in the Beaufort Sea (DOC: 104 ± 23 μmol L⁻¹; a₃₂₅: 1.06 ± 0.52 m⁻¹; S_275–295: 25.8 ± 2.2 μm⁻¹; TDLP₉: 10 ± 6 nmol L⁻¹) and the Makarov Basin (DOC: 108 ± 9 μmol L⁻¹; a₃₂₅: 1.36 ± 0.18 m⁻¹; S_275–295: 22.8 ± 0.8 μm⁻¹; TDLP₉: 15 ± 3 nmol L⁻¹) (Table 1). Much lower DOC and TDLP₉ concentrations and a₃₂₅ values and higher S_275–295 values were found in the Canada Basin (DOC: 69 ± 6 μmol L⁻¹; a₃₂₅: 0.32 ± 0.07 m⁻¹; S_275–295: 31.7 ± 2.3 μm⁻¹; TDLP₉: 3 ± 0.4 nmol L⁻¹) and its adjacent areas (Chukchi Borderland, Sever Spur, and Mendeleev Plain). Surface distributions of DOM across the Canada Basin, Chukchi Borderland, and Mendeleev Plain showed no apparent latitudinal trends (Figure 3).

The different DOM signature in the Makarov Basin was due to the influence of the Transpolar Drift. A subset of the dataset was extracted to characterize concentrations and compositions of DOM in the Transpolar Drift. Data (n = 52; 8 m) collected in regions north of ~83°N, including the Makarov Basin, Mendeleev Ridge, and Alpha Ridge, were ranked by each of four DOM parameters (DOC, a₃₂₅, TDLP₉, S_275–295), and the data within the upper 50% of every parameter (lower 50% in the case of S_275–295) were averaged (n = 19) to represent the concentration and composition of DOM in the Transpolar Drift. Values for DOC (114 ± 5 μmol L⁻¹), a₃₂₅ (1.48 ± 0.09 m⁻¹), S_275–295 (22.4 ± 0.3 μm⁻¹), and TDLP₉ (17 ± 1 nmol L⁻¹) in the Transpolar Drift indicated the highest concentrations of terrigenous DOM observed in this study (Supplementary Table 2).

Regression analyses were performed to investigate the influence of meteoric water and sea-ice melt on salinity, DOC, CDOM, and dissolved lignin phenols in surface waters of the western Arctic Ocean (Figures 4–6 and Supplementary Table 3). Data from the Canada Basin, Chukchi Borderland, and Sever Spur were combined (denoted as Canada Basin^*) in the regression analysis given their geographic proximity and similar distributions of DOM. It is interesting to note that the geographic connections among the deep basins (Canada Basin^*, Mendeleev Plain, and Makarov Basin) were revealed by the continuum of salinity and DOM distributions (Figures 4–6).

Salinity in surface waters was influenced to varying extents by meteoric water and sea-ice melt inputs (Figure 4 and Supplementary Table 3). Meteoric water showed a moderate dilution effect on the surface salinity in the Beaufort Sea (R² = 0.630, p < 0.001), Mendeleev Plain (R² = 0.461, p < 0.001), and Makarov Basin (R² = 0.613, p < 0.001; Figure 4A). This dilution effect was much weaker in Canada Basin^* (R² = 0.287) and was not apparent in the Bering-Chukchi Seas (R² = 0.005). In comparison, sea-ice melt showed a moderate to strong dilution influence on the surface salinity in all regions (R² = 0.420–0.920, p < 0.001) except the Makarov Basin (Figure 4B). The influence of sea-ice melt was most pronounced in the Canada Basin^* (R² = 0.920, p < 0.001) and it gradually diminished in the Mendeleev Plain (R² = 0.420, p < 0.001) and Makarov Basin (R² = 0.273, p < 0.001).

Relationships between meteoric water and DOM parameters were spatially variable (Figure 5 and Supplementary Table 3), with moderate to strong relationships in coastal regions (Bering-Chukchi Seas and Beaufort Sea: R² = 0.595–0.764, p < 0.001) and weak or no relationships in the deep basins (Canada Basin^*, Mendeleev Plain, and Makarov Basin: R² = 0.01–0.25). Increasing meteoric water in the Bering-Chukchi and Beaufort Seas resulted in higher concentrations of DOC, a₃₂₅, and TDLP₉, and lower S_275–295. One exception occurred in the Bering-Chukchi Seas, where S_275–295 values showed little variation with meteoric water (Figure 5C).

The influence of sea-ice melt on DOM was also variable among regions (Figure 6 and Supplementary Table 3). In the Bering-Chukchi Seas, increasing sea-ice melt resulted in decreasing DOC concentrations (R² = 0.501, p < 0.001), a₃₂₅ values (R² = 0.507, p < 0.001), TDLP₉ concentrations (R² = 0.332, p < 0.05), and increasing S_275–295 values (R² = 0.721, p < 0.001). By contrast, increasing sea-ice melt in the Beaufort Sea led to higher DOC concentrations (R² = 0.316, p < 0.001), a₃₂₅ values (R² = 0.400, p < 0.001), TDLP₉ concentrations (R² = 0.364, p < 0.001), and lower S_275–295 values (R² = 0.347, p < 0.001). Sea-ice melt in the deep basins showed weak or no relationships with DOM parameters (R² = 0–0.12). One exception was found in S_275–295, which increased significantly with sea-ice melt in the Canada Basins^* (R² = 0.237, p < 0.001; Figure 6C).

Water masses in the Arctic Ocean are distinguishable by their physical and chemical properties (Jones et al., 1995; Rudels et al., 1996; Shimada et al., 2005; Rudels, 2009). In this study, six water masses were identified in the western Arctic Ocean based primarily on salinity and potential temperature (Table 1). The polar mixed layer (PML) occupied the upper 30–80 m of the water column (shallower in the Makarov Basin) and had relatively low salinity (<32.0). The upper halocline (UH) resided below the PML at depths of 80–200 m and centered at a salinity of 33.1. The UH waters had higher concentrations of brine (i.e., negative f _SIM values) and nutrients (e.g., silicate) and lower temperatures than ambient waters (Figure 7 and Table 1). These characteristics were prominent in the Beaufort Sea, Canada Basin, and Mendeleev Plain but were absent in the Makarov Basin. The lower halocline (LH) was found at depths of 200–300 m with higher salinities ranging from 33.9 to 34.7. Upper and lower haloclines were not distinguishable in the Makarov Basin, where a cold halocline (θ < 0°C) was present at depths of 50–150 m with salinities ranging from 32.5 to 34.5 (Steele and Boyd, 1998; Rudels et al., 2004; Bauch et al., 2016). Below the halocline waters, the denser (34.7 < S < 34.9) and warmer (θ > 0°C) Atlantic waters were identified at depths down to ~1000 m. Below the Atlantic waters are the cold (θ < 0°C) Arctic deep (1000–2500 m) and bottom (>2500 m) waters. Salinities increased and temperatures decreased with depth in Arctic deep waters and they became relatively constant in Arctic bottom waters (Table 1).

Concentrations of DOC varied ~2 to 3-fold with depth in the basins, ranging from 44 to 88 μmol L⁻¹ and from 42 to 100 μmol L⁻¹ in the Canada Basin and adjacent areas (Sever Spur and Mendeleev Plain) and from 43 to 129 μmol L⁻¹ in the Makarov Basin (Figure 8 and Table 1). Depth distributions of DOC concentrations in the Canada Basin and adjacent areas were very similar and showed no apparent latitudinal trends (Figure 8A). Comparisons between the Canada and Makarov Basins revealed different distributions of DOC concentrations in the upper 300 m of the water column (Figure 8C). In the Canada Basin, DOC concentrations were slightly elevated in the PML (69 ± 6 μmol L⁻¹) and gradually decreased with increasing depth from the halocline (UH: 65 ± 3 μmol L⁻¹; LH: 61 ± 3 μmol L⁻¹) to the Atlantic layer (53 ± 3 μmol L⁻¹) and to Arctic deep (49 ± 3 μmol L⁻¹) and bottom waters (46 ± 1 μmol L⁻¹). In comparison, concentrations of DOC in the Makarov Basin were greatly elevated in the PML (108 ± 9 μmol L⁻¹) and decreased rapidly and conservatively with salinity (R² = 0.78, p < 0.001, n = 10) in the halocline layer (70 ± 14 μmol L⁻¹; Figure 8B). The DOC concentrations in the Atlantic layer (54 ± 4 μmol L⁻¹) and the Arctic deep (49 ± 4 μmol L⁻¹) and bottom waters (46 ± 2 μmol L⁻¹) in the Makarov Basin were very similar to those in the Canada Basin (Mann-Whitney U-Test, p > 0.05; Figure 8C and Table 1).

A striking gradient in DOC concentrations was observed in deep waters of the Canada and Makarov Basins (Figures 8C, 9). Concentrations of DOC decreased by 6–8 μmol L⁻¹ with increasing water mass age at depths of ~1000–2500 m (Figure 9 and Table 1). Declines in DOC concentrations were significantly related to increases in salinity and apparent oxygen utilization (AOU) (DOC = 3434 − 96.85 × salinity − 0.05 × AOU: R² = 0.62, p < 0.001) and inorganic nitrogen and silicate concentrations (R² = 0.50–0.70, p < 0.001). Concentrations of DOC in Arctic bottom water (>2500 m) were less variable with depth, consistent with the relatively constant physical properties in these waters (Table 1).

The a₃₂₅ values in the deep basins were much more variable than DOC concentrations, varying ~4-fold (0.13–0.55 m⁻¹) in the Canada Basin and ~13-fold (0.13–1.70 m⁻¹) in the Makarov Basin (Figure 10 and Table 1). As observed with DOC, distributions of a₃₂₅ were quite different between the two basins in the upper 300 m of the water column (Figure 10C). The a₃₂₅ in the Canada Basin was relatively low in the PML (0.32 ± 0.07 m⁻¹) and showed a maximum of 0.48 ± 0.03 m⁻¹ in the UH, below which a₃₂₅ decreased conservatively with salinity in the LH and upper Atlantic layer (200–300 m: R² = 0.933, p < 0.001, n = 18; Figure 10A). In comparison, a₃₂₅ in the Makarov Basin reached a maximum of 1.36 ± 0.18 m⁻¹ in the PML, and the values decreased rapidly and conservatively with salinity in the halocline layer (0.58 ± 0.29 m⁻¹; R² = 0.92, p < 0.001, n = 10; Figure 10B). The a₃₂₅ below 300 m showed similar values and distributions between the two basins (Mann-Whitney U-Test, p > 0.1). Deep-sea gradients in a₃₂₅ were observed and they showed opposite trends between the Arctic deep water and Arctic bottom water (Figure 10C). The a₃₂₅ in Arctic deep water decreased substantially with depth in the Canada Basin (from 0.20 to 0.15 m⁻¹) and Makarov Basin (from 0.19 to 0.13 m⁻¹), a feature similar to the decline in DOC concentrations. By contrast, a₃₂₅ values in Arctic bottom water at all eight stations were significantly elevated (by 10–20%) toward the seafloor (T-Test, p < 0.05; Figure 10C).

The S_275–295 ranged from 21.3 to 38.1 μm⁻¹ with depth in the Canada Basin and from 21.5 to 31.3 μm⁻¹ in the Makarov Basin (Figure 11 and Table 1). The S_275–295 in the Canada Basin was elevated in the PML (31.7 ± 2.3 μm⁻¹) and decreased rapidly to a minimum of 22.9 ± 0.7 μm⁻¹ in the UH, below which S_275–295 increased with depth and the values varied mostly between 25 and 28 μm⁻¹ (Figure 11A). In the Makarov Basin, S_275–295 values were low in the PML (22.8 ± 0.8 μm⁻¹) and increased with depth in the halocline (23.7 ± 1.5 μm⁻¹; Figure 11B). The S_275–295 values below 300 m were comparable to those at matching depths in the Canada Basin (Mann-Whitney U-Test, p > 0.05; Figure 11C). Lower S_275–295 values were observed near the seafloor of the Canada and Makarov Basins.

These data provided an opportunity to evaluate relationships between DOC concentrations and CDOM absorption in the Arctic Ocean. A total of 755 paired measurements in this study revealed a strong relationship between DOC concentrations and CDOM absorption (e.g., a₃₂₅: R² = 0.77–0.87, p < 0.001; Supplementary Figure 2). This feature is absent in the Pacific, Atlantic, and Indian Oceans (Nelson et al., 2010), and it allows the developments of empirical models to retrieve DOC concentrations from CDOM measurements in the Arctic Ocean. By re-parameterizing a previously developed multi-linear regression model (Fichot and Benner, 2011) using the present data set, two sub-models separated by a cutoff value of a₂₇₅ = 1.42 m⁻¹ (i.e., the median) were established.

When a₂₇₅ ≤ 1.42 m⁻¹, (5)ln[DOC]=3.881±0.036+0.763±0.057×ln(a275)                                 −0.375±0.061×ln(a295) When a₂₇₅ > 1.42 m⁻¹, (6)ln[DOC]=3.643±0.032+1.223±0.055×ln(a275)                             −0.701±0.049×ln(a295) The standard error of each regression coefficient is given in the equations. The estimated concentrations of DOC exhibited a fairly strong relationship with the measured values (R² = 0.951, p < 0.001; Figure 12). The error associated with the retrieval of DOC concentration using this approach was generally within 15%. The broad range of DOC and CDOM values included in the data suggests the models could be applicable to other regions of the Arctic Ocean.

This study demonstrated that the abundance and distribution of DOM in polar surface waters were very heterogeneous and strongly linked to hydrological conditions. High concentrations of DOC and lignin phenols, high values of a₃₂₅, and low values of S_275–295 were observed in surface waters of the Beaufort Sea. These concentrations and values were significantly correlated to f _MW, reflecting the influence of the Mackenzie River. Elevated DOC (>100 μmol L⁻¹) and lignin phenol concentrations (>10 nmol L⁻¹) were present in the northwest region of the Beaufort Sea, indicating the transport of riverine DOM to surface waters in the Canada Basin. Previous investigations (2002–2004) do not report such high concentrations of DOC in similar regions (mostly 65–80 μmol L⁻¹; Mathis et al., 2005; Shen et al., 2012). This large inter-annual variability in terrigenous DOM concentrations is consistent with a recent change in the routing of Mackenzie River outflow, which has shifted from a predominantly eastward path to a northwestward path since 2006 in response to climatic variability (Fichot et al., 2013). The routing switch enhanced the input of terrigenous DOM to the Canada Basin.

Contrasting DOM characteristics were present in surface waters of the Canada and Makarov Basins, reflecting distinct sources and processing of DOM. Surface waters of the Canada Basin contained relatively high f _MW (0.18 ± 0.01) but had lower DOC concentrations (Table 2) and CDOM absorption (Stedmon et al., 2011; Guéguen et al., 2015) compared to other Arctic Basins. This feature appears to result from a substantial contribution of precipitation in the inflowing waters from the Pacific Ocean and a lesser contribution of freshwater from continental runoff (Woodgate and Aagaard, 2005; Yamamoto-Kawai et al., 2008). Although high f _SIM was observed in surface waters, as discussed below, this had a negligible influence on the surface DOC concentrations. Relatively low concentrations of lignin phenols (2–4 nmol L⁻¹) were observed over large areas of the Canada Basin, due in part to minor contributions of continental runoff. Surface waters in the Beaufort Gyre have a decadal residence time that allows substantial removal of DOM (Rutgers van der Loeff et al., 1995; Hansell et al., 2004; Cooper et al., 2005). The S_275–295 values were high (>30 μm⁻¹) and elevated in ice-free waters, indicating photobleaching and potential removal of DOC due to coupled photochemical and biological degradation (Fichot and Benner, 2014).

By contrast, surface waters in the Makarov Basin had a lower f _MW (0.15 ± 0.02) but were characterized by much higher DOC concentrations and CDOM absorption and lower S_275–295values compared to Canada Basin surface waters (Table 2). The high concentrations of lignin phenols (15 ± 3 nmol L⁻¹) in surface waters were indicative of a strong Siberian river influence associated with the Transpolar Drift (Opsahl et al., 1999; Benner et al., 2005). The Siberian rivers (e.g., Lena River) have higher concentrations of DOC than the Mackenzie River (Anderson and Amon, 2015; and reference therein). In addition, unlike the long-term processing in the Canada Basin, DOM entering the Makarov Basin is rapidly transported (<1–3 yr) by the Transpolar Drift (Rutgers van der Loeff et al., 1995). Concentrations of DOC (114 ± 5 μmol L⁻¹) and lignin phenols (17 ± 1 nmol L⁻¹) determined here for the Transpolar Drift (>83°N) were comparable to values measured at similar salinities (~30) in waters off the Siberian shelves (Guay et al., 1999; Kattner et al., 1999; Kaiser et al., in review), suggesting minor losses of DOM occurred during transport to the central Arctic.

The axis of the Transpolar Drift is variable and shifts between the Mendeleev and Lomonosov Ridges, depending on the phase of the Arctic Oscillation (AO) (Guay et al., 2001; Mysak, 2001; Steele et al., 2004). A negative AO phase dominated the summers of 2010–2011 and resulted in a deflection of the Transpolar Drift toward the Lomonosov Ridge and a strengthened Beaufort Gyre. Under these conditions the Eurasian runoff and terrigenous DOM entrained in the Transpolar Drift would be confined mainly to regions between the Mendeleev and Lomonosov Ridges (Mysak, 2001). Such routing of the Transpolar Drift was substantiated by observations of relatively low concentrations of DOC, lignin phenols and a₃₂₅ values, and high values of S_275–295 over the Mendeleev Plain. This route would bypass the Canada Basin and thereby favors export of DOM to the North Atlantic Ocean. These results revealed contrasting distributions and fates of DOM between the Canada and Makarov Basins and further highlighted strong hydrographic controls on the DOC cycle in the Arctic Ocean.

The effect of sea-ice melt on DOC concentrations in underlying waters depends on the biota and organic compounds within the ice. Sea-ice formation excludes DOM as well as salts (Giannelli et al., 2001; Amon, 2004), whereas ice algae, other microorganisms and entrained sediments with associated organic matter can contribute high concentrations of organic carbon to the ice (Smith et al., 1997; Eicken, 2004; Gradinger, 2009). The spatial and temporal variability of biota and organic matter in Arctic sea ice are high (Thomas et al., 1995; Smith et al., 1997; Song et al., 2011), thereby complicating the net impact of sea-ice melt on DOC concentrations. Previous studies report similarly variable observations (Anderson, 2002; Eicken, 2004; Mathis et al., 2005, 2007; Anderson and Amon, 2015; Logvinova et al., 2016).

Results of this study demonstrated that the influence of sea-ice melt on concentrations and compositions of DOM in polar surface waters was region-dependent. Contrasting effects of sea-ice melt were observed along the margin between the Beaufort Sea and the Bering-Chukchi Seas, and it appeared to be related to the influence of continental runoff. Sea ice formed on the river-influenced Beaufort Shelf includes suspended riverine materials (i.e., “dirty” ice; Eicken et al., 2005). Melting of this “dirty” ice can potentially release high concentrations of terrigenous organic matter and increase DOM concentrations in the surface layer. The significance of these processes was apparent in the Beaufort Sea from the observation of elevated lignin concentrations in waters containing higher f _MW.

In the Bering-Chukchi Seas negative relationships between DOC, lignin phenols, CDOM and f _MW indicated a low abundance of organic matter in the ice. Terrigenous inputs inferred from f _MW (0.07 ± 0.04) and lignin phenol concentrations (4 ± 3 nmol L⁻¹) were relatively low in the Bering-Chukchi Seas. Ice algae production in this region is high (Gosselin et al., 1997; Gradinger, 2009), but algal contributions of DOM are often confined to the bottom of the ice (Thomas et al., 1995; Smith et al., 1997; Song et al., 2011). Previous analyses of ice cores in the Chukchi Sea indicate variable but generally lower DOC concentrations and CDOM absorption than found in adjacent surface waters (Krembs et al., 2002; Mathis et al., 2007; Logvinova et al., 2016). Our results add to these observations indicating a dilution effect of sea-ice melt on DOM in the Bering-Chukchi Seas during August-September.

The influence of sea-ice melt on DOM in surface waters of the deep basins is minor. Extensive sea-ice melt was observed in surface waters of the Canada Basin and southern Mendeleev Plain. However, the melting of sea ice did not result in significant changes in surface concentrations of DOC, lignin phenols, and a₃₂₅ values in the region. The lack of significant response to sea-ice melt indicates concentrations of DOM in the melted ice were comparable to those in the underlying waters. Considering the large difference in terrigenous influence, sea ice in the Canada Basin and Mendeleev Plain likely contained much less DOM than sea ice found in the central Arctic and Fram Strait (e.g., <100–300 μmol L⁻¹; Melnikov, 1997; Opsahl et al., 1999).

Halocline waters of the Canada and Makarov Basins displayed distinct DOM distributions related to their sources and formation. A prominent feature in the Canada Basin was the maximal a₃₂₅ (0.48 ± 0.03 m⁻¹) and minimal S_275–295 (22.9 ± 0.7 μm⁻¹) in the upper halocline, which indicated higher molecular weight chromophores and minimal photobleaching. Upper halocline waters are derived from brine rejection during sea-ice formation over the Chukchi and East Siberian shelves, where nutrients and organic matter released from sediments are added to the dense brines and advected into the Canada Basin (Jones and Anderson, 1986; Cooper et al., 1997, 2005). These processes are an important source of CDOM in the upper halocline (Cooper et al., 2005; Guéguen et al., 2007; Stedmon et al., 2011) and they were apparent from our observations of negative f _SIM and elevated inorganic nutrients. Concentrations of dissolved lignin phenols in the upper halocline (3 ± 0.2 nmol L⁻¹) were similar to values (3 ± 0.4 nmol L⁻¹) in surface waters, indicating a terrigenous signature associated with brine waters. These results are consistent with previous observations of relatively high concentrations of lignin phenols and Δ¹⁴C in DOC of upper halocline waters (Benner et al., 2004), demonstrating a contemporary riverine DOM source. Previous observations of elevated CDOM in the upper halocline were mainly from the southern Canada Basin (Cooper et al., 2005; Guéguen et al., 2007; Nakayama et al., 2011). Our results indicate these CDOM characteristics can be traced throughout the central and northern Canada Basin.

In contrast with the Canada Basin, DOC concentrations and CDOM absorption in the Makarov Basin were highest in surface waters and decreased rapidly with depth in the upper 200 m of the water column. The average values of DOC concentrations (70 ± 14 μmol L⁻¹), a₃₂₅ values (0.58 ± 0.29 m⁻¹), S_275–295 values (23.7 ± 1.5 μm⁻¹), and lignin phenol concentrations (4 ± 4 nmol L⁻¹) in the halocline were comparable to those in upper halocline of the Canada Basin. Atlantic water that flows through Fram Strait and is modified over the Eurasian shelves supplies the halocline in the Makarov Basin (Rudels et al., 1996, 2004; Bauch et al., 2016). The Atlantic inflow waters in the Fram Strait have relatively low concentrations of DOC (~61 μmol L⁻¹) and CDOM (a₃₅₀ < 0.2 m⁻¹) (Opsahl et al., 1999; Amon et al., 2003; Granskog et al., 2012). Sea-ice formation and brine rejection over the Siberian shelves can transport elevated concentrations of DOC and CDOM from Siberian rivers to the halocline. The relatively high concentrations of lignin phenols indicate an important riverine contribution to the DOC and CDOM in halocline waters. Distributions of DOC, a₃₂₅, and lignin phenols along the salinity gradient in the halocline were largely conservative (R² = 0.78–0.92), indicating minimal loss of DOM components during transport within halocline waters of the Makarov Basin.

The DOC concentrations and CDOM absorption in Atlantic waters below the halocline were similar between the Canada Basin and Makarov Basin. Comparisons with previous DOC measurements revealed two particular features: little inter-annual variability within the Canada and Makarov Basins and a concentration gradient between the Canadian and Eurasian Basins. Average concentrations of DOC in Atlantic waters decreased from 60 ± 5 μmol L⁻¹ in the Nansen Basin to 56 ± 2 μmol L⁻¹ in the Amundsen Basin, and to 52–55 μmol L⁻¹ in the Canada Basin (Table 2). These DOC values were also lower than those in the inflowing Atlantic water (<600 m; 61 μmol L⁻¹; Opsahl et al., 1999; Amon et al., 2003), indicating DOC removal from Atlantic waters during transport in the Arctic Ocean.

Concentrations of DOC in the deep Arctic Ocean exhibited a strong vertical gradient at depths of 1000–2500 m and a relatively homogenous distribution at depths >2500 m. The variability in deep DOC concentrations observed in this study was considerably lower than that reported by Bussmann and Kattner (2000) (Table 2). On average, concentrations of DOC in deep waters (1000–2500 m) of the Canada Basin (49 ± 3 μmol L⁻¹) and Makarov Basin (49 ± 4 μmol L⁻¹) were similar to but slightly lower than those in deep waters of the Amundsen Basin (54 μmol L⁻¹), Nansen Basin (50 μmol L⁻¹), outflowing deep waters from the Eurasian Basin (1800–2500 m; 50 ± 2 μmol L⁻¹), and Nordic Seas deep waters (1000–4000 m; 48–50 μmol L⁻¹; Table 2). Concentrations of DOC in bottom waters of the Canada and Makarov Basins (>2500 m; 46 ± 2 μmol L⁻¹) were lower than those in any Arctic water masses mentioned above, but they were higher than concentrations in deep waters of other major ocean basins (35–45 μmol L⁻¹; Druffel et al., 1992; Hansell and Carlson, 1998).

There are few CDOM measurements in Arctic deep waters. The a₃₂₅ measured herein at depths >1000 m in the Canada Basin (0.17 ± 0.02 m⁻¹) and Makarov Basin (0.16 ± 0.02 m⁻¹) were within the range in deep waters (>2000 m) of the South Atlantic and North Indian Oceans, but were slightly elevated relative to those in deep waters (>2000 m; 0.10–0.15 m⁻¹) of the Pacific, North Atlantic, and South Indian Oceans (Nelson et al., 2010).

An outstanding feature of the deep-Arctic data was that concentrations of DOC decreased (by 6–8 μmol L⁻¹) with increasing water mass ages from depths of 1000 m to 2500 m in the Canada and Makarov Basins. Such a loss of DOC resembled the gradient (6 ± 1 μmol L⁻¹) observed in North Atlantic Deep Water from 65°N to 19°N (Carlson et al., 2010). The decline in deep Arctic DOC concentrations was significantly related to an increase in salinity, apparent oxygen utilization and nutrient concentrations, indicating that physical mixing and microbial degradation are responsible for the decline in DOC concentrations. The radiocarbon content of DOC in Canada Basin deep water decreases from −380%0 at 1000 m, a value close to that in North Atlantic deep water, to −466%0 at 2500 m (Griffith et al., 2012). The difference in Δ¹⁴C (−86%0) indicates the DOC ages vary by over 700 yr between 1000 and 2500 m, whereas the CFC-based ventilation age of water increases by ~200 yr. These differences between DOC and ventilation ages indicate the DOC that is removed is ¹⁴C-enriched relative to the bulk DOC.

Assuming the decrease in deep-water DOC concentrations was all due to microbial degradation, estimates of maximal degradation rates can be derived from the regression slope between DOC concentrations and ventilation ages at depths of 1000–2500 m (R² = 0.51–0.83, p < 0.001). The estimated rates ranged from 0.036 μmol C L⁻¹ yr⁻¹ in the Canada Basin to 0.039 μmol C L⁻¹ yr⁻¹ in the Makarov Basin, and they represent the upper limits for DOC biodegradation in these deep waters. These values were 2–3 orders of magnitude lower than those (0.2–5 μmol C L⁻¹ yr⁻¹) estimated in the mesopelagic zone of the North Pacific and North Atlantic (Carlson et al., 2010; Kaiser and Benner, 2012), and they were also lower than those (0.04–0.20 μmol C L⁻¹ yr⁻¹) in deep waters (>1000 m) of the Greenland Gyre, North Atlantic Ocean, Mediterranean Sea, and East Japan Sea (Amon et al., 2003; Carlson et al., 2010; Kim et al., 2015). These comparisons indicate slow microbial degradation of DOC in deep waters of the Arctic Ocean. In addition to the relatively low abundance of bioavailable DOM in deep waters (Amon and Benner, 2003; Davis and Benner, 2005), low water temperatures (<0°C) could also influence rates of microbial degradation of DOM in the deep ocean (e.g., Kim et al., 2015).

An unexpected elevation of a₃₂₅ (by 10–20%) with minor changes in DOC concentrations was observed near the seafloor (mostly <50 m above the bottom) in the Canada and Makarov Basins. Our observations differ from those of Bussmann and Kattner (2000) who reported losses of DOC near the seafloor around the Lomonosov Ridge. The elevated a₃₂₅ we observed near the seafloor was associated with lower S_275–295 values and higher nutrient concentrations, suggesting that the elevated CDOM was largely derived from benthic remineralization of organic matter in sediments. This observation is in agreement with growing data on benthic abundance, diversity, and carbon demand indicating active biological processing in sediments of deep Arctic Ocean basins (Clough et al., 1997; Klages et al., 2004; Bluhm et al., 2011). Overall, these data indicate DOC and CDOM in Arctic Ocean deep and bottom waters are more dynamic than previously recognized.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.