Absorption coefficients of colored dissolved organic matter (CDOM) were measured together with salinity, δ18O, and inorganic nutrients across the Fram Strait. A pronounced CDOM absorption maximum between 30 and 120 m depth was associated with river and sea ice brine enriched water, characteristic of the Arctic mixed layer and upper halocline waters in the East Greenland Current (EGC). The lowest CDOM concentrations were found in the Atlantic inflow. We show that the salinity-CDOM relationship is not suitable for evaluating conservative mixing of CDOM. The strong correlation between meteoric water and CDOM is indicative of the riverine/terrigenous origin of CDOM in the EGC. Based on CDOM absorption in Polar Water and comparison with an Arctic river discharge weighted mean, we estimate that a 49–59% integrated loss of CDOM absorption across 250–600 nm has occurred. A preferential removal of absorption at longer wavelengths reflects the loss of high molecular weight material. In contrast, CDOM fluxes through the Fram Strait using September velocity fields from a high-resolution ocean–sea ice model indicate that the net southward transport of terrigenous CDOM through the Fram Strait equals up to 50% of the total riverine CDOM input; this suggests that the Fram Strait export is a major sink of CDOM. These contrasting results indicate that we have to constrain the (C)DOM budgets for the Arctic Ocean much better and examine uncertainties related to using tracers to assess conservative mixing in polar waters.
 Here we use the ultraviolet and visible light absorption properties of CDOM to characterize and trace the passage of CDOM supplied by Arctic rivers, recently reported by Stedmon et al. , through the Arctic Ocean and on to the North Atlantic via the Fram Strait with the EGC. Although the export of freshwater from the Arctic Ocean to the North Atlantic is thought to be approximately equally split between passage through the Canadian Arctic Archipelago (CAA) and the Fram Strait [Rudels, 1987], the amount and characteristics of CDOM exported via two pathways may differ in respect to DOM origin and export, due different freshwater sources [Jahn et al., 2010] and due to the fact that the surface waters of the Eurasian Basin are more influenced by terrigenous river CDOM loading and consequently has higher DOM concentrations [Amon et al., 2003; Cooper et al., 2005; Benner et al., 2005; Stedmon et al., 2011]. In this study we apply existing source water fractionation techniques to assess the amount of CDOM exported to the North Atlantic through the Fram Strait in September 2009. We compare our results with recent observations of Arctic riverine CDOM loading [Stedmon et al., 2011], estimate the magnitude and characteristics of changes in terrigenous CDOM absorption and estimate the flux of CDOM through the Fram Strait.
2. Data and Methods
 Samples for salinity, stable oxygen isotope ratio (δ18O), colored dissolved organic matter (CDOM; 0.2 μm) and inorganic nutrients (PO4, NO2+NO3) were collected along a section across the Fram Strait just south of 79°N, from the east Greenland shelf to the Spitsbergen shelf (Figure 1) in September 2009. Sampling focused on the top 400 m and was a quasi-synoptic survey of the water masses in the strait.
 Nutrient samples were collected in acid-washed polyethylene bottles, which were frozen immediately, and kept at −20°C until analysis at National Environment Research Institute (Denmark) using an autoanalyzer (Skalar) following the approach ofHansen and Koroleff . Precision for nutrient analyses were 0.03 μmol L−1 for phosphate and 0.05 μmol L−1 for nitrate. Salinity samples were collected in glass bottles that were rinsed three times before being filled and analyzed on board using a Guildline 8410A salinometer with precision better than ±0.003. The δ18O samples were collected in 30 ml glass vials, caps closed tightly and sealed with Parafilm. The δ18O samples were measured ashore at the National Oceanographic Centre, Southampton, by standard equilibration with carbon dioxide following the procedure described by Epstein and Mayeda . Accuracy was estimated to be better than ±0.04‰ relative to Vienna standard mean ocean water (VSMOW). CDOM samples were syringe filtered (Pall Acrodisc® PF; 0.8/0.2 μm pore size, with Supor® membrane) into precombusted amber glass vials immediately after collection, and stored in dark at +4°C until analysis, within a month. CDOM was quantified as the absorption coefficient in the spectral range from 250 to 650 nm using a Shimadzu UV-2401PC spectrophotometer and 100 mm quartz cells with ultrapure water as reference, with a precision of 0.025 m−1 [Stedmon and Markager, 2001]. Here the absorption at selected wavelengths is used as an indicator of CDOM concentration (see below). CDOM characteristics are evaluated using the exponential slope of the absorption spectrum between 300 and 650 nm (S) [Stedmon and Markager, 2001] and the slope ratio (SR), which is the ratio of the slopes determined between 275 and 295 nm, and 350 nm and 400 nm [Helms et al., 2008]. It has been demonstrated that both these parameters can be used to distinguish between terrigenous and marine sources of CDOM, the latter from biological productivity in seawater. All slopes were calculated with a nonlinear regression using the nlinfit routine in MATLAB 7.11.0 R2010b software (MathWorks).
 The water mass composition of discrete water samples from the Fram Strait was determined using a four-end-member mass balance following the approach ofDodd et al. . The composition was derived using a combination of the procedures established by Östlund and Hut  and Jones et al. [2008a]defined by the following four-end-member mass balance equations:
Here fmw, fsim, fpw, faw are the derived fractions of meteoric water, sea ice meltwater, Pacific seawater and Atlantic Seawater respectively; SALmw, SALsim, SALpw, SALaw are the assigned salinities of meteoric water, sea ice meltwater, Pacific seawater and Atlantic Seawater (0, 4, 32.0, and 34.9 respectively); δmw, δsim, δpw, δaw are the corresponding δ18O values (−18.4, 0.5, −1.3, and 0.3‰ respectively); SAL, δ, P and N are the measured salinity, δ18O, dissolved phosphate concentration and dissolved nitrate concentration, respectively. Where the fraction of Pacific seawater determined for a sample is <1% the following three-end-member mass balance is used in preference to the four-end-member balance:
 The slope and intercept used to calculate Ppw follow Jones et al. [2008a]. The slope and intercept used to calculate Paw were determined by regression of nitrate (N) and phosphate (P) samples with a potential density >1027.6 kg m−3 collected along the Fram Strait section. The potential density threshold is used to exclude any Pacific water present [Dodd et al., 2012]. Atlantic seawater salinity (34.9) is prescribed based on Östlund and Hut , while the δ18O value is based on a regression of δ18O and salinity (n = 125) to data from the eastern Fram Strait, with δ18O = 0.3‰ at salinity 34.9 [Dodd et al., 2012]. Pacific seawater salinity is based on data from Bering Strait [Woodgate and Aagaard, 2005] with a flow weighted average of 32.0, and data from Cooper et al.  is used to estimate δ18O = −1.3‰ at salinity 32.0 [Dodd et al., 2012]. The sea ice end-member salinity (4) is based onÖstlund and Hut  and the δ18O value (+0.5‰) is calculated using the mean surface δ18O value in the Arctic Ocean of −1.67‰ (estimated using data collated by G. A. Schmidt et al., Global Seawater Oxygen-18 Database, v1.21, 1999,http://data.giss.nasa.gov/o18data) and a fractionation factor of 1.0021 [Melling and Moore, 1995] to account for the fractionation that occurs during sea ice formation. Meteoric water has a salinity of 0, and the δ18O value is a flow weighted estimate based on data from the ten largest Arctic rivers, with recent data from Cooper et al.  (Yenisei, Ob, Lena, Kolyma, and Mackenzie rivers) and earlier data for Pechora, Severnaya Dvina, Indirka, Yana, and Olenek rivers [Dodd et al., 2012], this results in a δ18O value of −18.4‰. Precipitation is assumed to have a similar value. We refer the reader to Dodd et al. for the details behind the choice of end-member values and for the sensitivity of the estimates of freshwater fractions to variations in the end-member composition. These uncertainties were found to not significantly influence the estimates of CDOM removal presented below.
 In order to evaluate the changes and removal of terrigenous CDOM during its passage through the Arctic Ocean, data from a recent study on Arctic river CDOM loading is used [Stedmon et al., 2011]. The river data consists of two years of measurements from six major Arctic rivers (Kolyma, Lena, Mackenzie, Ob, Yenisei, and Yukon). These rivers have an annual discharge of about 2200 km3 [Stedmon et al., 2011], which is roughly two thirds of the estimated total river runoff (about 3300 km3) to the Arctic Ocean [Serreze et al., 2006]. CDOM loading was calculated for the whole absorption spectrum for each river using a nonparametric regression to discharge data [Stedmon et al., 2011], then normalized to each rivers annual discharge to provide a discharge weighted mean CDOM absorption spectrum for each river (Figure 2a). To calculate the combined discharge weighted riverine CDOM spectrum the loadings from each river were summed and then normalized to the combined discharge (Figure 2b). We use these values, shown in Figure 2b, as an end-member concentration of CDOM in Arctic rivers. From this data we derive a discharge weighted average CDOM absorption value for all the rivers combined of 15.5 m−1 (at 350 nm). If we exclude the influence of the Yukon River the discharge weighted value is 16.2 m−1. If both the Yukon and Mackenzie are omitted from the calculation, the discharge weighted value increases further to 18.3 m−1. For the rest of this manuscript we will use the estimate based on all rivers, which represents a conservative estimate for the actual CDOM riverine end-member to the Arctic Ocean. For the discharge weighted average CDOM spectra of all Arctic rivers (Figure 2b), the S and SR values are 16.4 μm−1 and 0.85, respectively. Holmes et al.  indicated that 20–40% of the DOM exported by these rivers during the spring freshet is bioavailable and therefore possibly utilized during its passage through the estuarine and shelf waters of the Arctic Ocean. In order to examine how this would affect the riverine loadings of CDOM, the data in Figure 2 were recalculated with the assumption that 30% of the riverine CDOM from the months of May, June and July was removed by microbes. This results in a lower discharge weighted average CDOM spectra and a350 for the combination of all six rivers was reduced from 15.5 to 12.3 m−1 (Figure 2b). The S and SR values did not change notably (<0.5%). Since the riverine freshwater flux to the Arctic Ocean from these six rivers represents approximately two thirds the total riverine flux (excluding Hudson Bay) [Serreze et al., 2006], the combined estimate from the six rivers studied [cf. Stedmon et al., 2011, Table 2] was scaled up to derive the total riverine CDOM flux to the Arctic Ocean (Table 1), assuming the unmeasured discharge has the same CDOM absorption as all the six rivers used here.
Table 1. Estimates of Annual CDOM (Based on Absorption at 350 nm a350) and Volume Fluxes Through the Fram Strait Compared to Riverine Loadinga
Data for volume flow are taken from the ECCO2 model (annual and September mean in 2009). Riverine discharge is from Serreze et al. , and CDOM loading to Arctic Ocean is calculated from data in Stedmon et al. . The Fram Strait is divided into five segments, the North East Greenland Coastal Current (NEGCC) (18° to 15°W), the shelf (15° to 11°W), the slope (11° to 5°W), the core of the East Greenland Current (EGC) at 5°W to 2°E, and Atlantic water inflow with the West Spitsbergen Current (WSC) east of 2°E. Negative values indicate flux out of the Arctic Ocean (i.e., southward).
CDOM flux (1012 m2 yr−1)
Volume flux (Sv)
CDOM flux (1012 m2 yr−1)
Volume flux (Sv)
 To obtain a velocity field covering the whole Fram Strait and the east Greenland shelf we use results from high-resolution, coupled ocean–sea ice modeling results of Estimating the Circulation and Climate of the Ocean, phase 2 (ECCO2). The model has typically a horizontal resolution in the Arctic of 4.5 km and in the vertical there are 50 levels ranging from 10 m near the surface to 450 m at maximum model depths. For the atmospheric boundary conditions ECMWF ERA-40 reanalysis, and Japanese 25 year Reanalysis Project (JRA25) covering 1979–2004, were used. More details on the model parameters and initialization for a simulation at a coarser resolution are described inNguyen et al. . The model output is used here to estimate the volume fluxes on the east Greenland shelf and across the Fram Strait in 2009. The model results agree well with observed velocities of the EGC for which moored current meter measurements are available (i.e., east of 6.5°W) [Fahrbach et al., 2001; de Steur et al., 2009].
3. Results and Discussion
3.1. Distribution and Origin of CDOM in the Fram Strait
 The salinity distribution across Fram Strait (Figure 3a) includes surface waters with salinity as low as 29 in the EGC, but mostly on the shelf (the core of maximum velocity of the EGC is approximately between 3° and 4°W at 79°N [de Steur et al., 2009]), in the western Fram Strait. The fraction of meteoric water (i.e., combination of river runoff, glacier or ice sheet meltwater and precipitation), fmw, largely follows the salinity distribution (Figure 3d), and contributes significantly to the liquid freshwater transport in the EGC, in agreement with the fact that river water is the largest apparent source of freshwater in the Eurasian sector of Arctic Ocean and along the East Greenland coast [Jones et al., 2008a, 2008b]. The fmw values of about 0.10–0.15 agree with earlier estimates of meteoric water in the EGC [Meredith et al., 2001; Jones et al., 2008a; Dodd et al., 2009; Sutherland et al., 2009; Rabe et al., 2009]. The highest a350 values are in the subsurface of the EGC and coincide with the cold upper halocline layer with excess brine from sea ice formation, close to the fsim minimum (Figures 3b and 3c). This is likely because sea ice melt at the very surface has diluted the rather high CDOM concentrations in the EGC. The location of the CDOM maximum is in agreement with earlier observations of CDOM fluorescence in the area [Amon, 2004; Amon et al., 2003] and with recent observations of subsurface CDOM fluorescence maxima in the western Arctic Ocean linked to sea ice formation and brine expulsion [Guéguen et al., 2007; Nakayama et al., 2011]. Maximum CDOM levels are found at salinities between 32 and 33 and this agrees well with earlier findings by Amon et al.  in the Fram Strait and Guéguen et al.  in the western Arctic halocline waters. This also coincides with the salinity range for the upper halocline waters formed on the Siberian shelves, and it is apparently a fingerprint of waters from the Siberian shelves that fmw and fsim are tightly coupled [Bauch et al., 2011], as observed in the Fram Strait [Dodd et al., 2012]. This indicates that CDOM maximum waters are sourced from the Siberian shelves with high riverine CDOM input [Stedmon et al., 2011], where river water (fmw) is entrained together with brine [Bauch et al., 2011] and CDOM into the upper halocline. This brine, CDOM and meteoric water enriched water can be found at depths of 30 to 70 m [Bauch et al., 2009; Stedmon et al., 2011], which corresponds well to our observations in the Fram Strait.
 The highest CDOM concentrations in the Fram Strait are found in Polar Water (temperature less than 0°C and salinity less than 34.8). The lowest CDOM concentrations are found in Atlantic water flowing northward in the eastern Fram Strait below 50 m depth (Figure 3b). The distinct subsurface maximum of CDOM in the EGC indicates that the growth of sea ice and subsequent formation of halocline waters in the Arctic Ocean plays an important role in the cycling of (C)DOM, highlighting the fact that changes in sea ice formation (rates, timing and location) may have significant impacts on cycling of DOM and other constituents, such as iron [Nakayama et al., 2011], in the Arctic Ocean.
 CDOM concentrations closely follow the distribution of fmw (Figure 3) indicating that river waters are a major source of CDOM in the EGC, as was found for the surface waters of the Eurasian Basin [Stedmon et al., 2011]. This is also apparent from CDOM optical characteristics that indicate a terrigenous origin of CDOM in the EGC (Figure 4) [cf. Stedmon and Markager, 2001]. All samples in Polar Water with high meteoric water fractions are characterized as terrigenous CDOM as they are outside the proposed model limits by Stedmon and Markager  for marine produced CDOM. In contrast, CDOM in Atlantic water inflow to the Arctic (with low fmw) has properties that indicate that it is predominantly of marine origin (Figure 4). The Fram Strait data set presented here provides an independent test of the model proposed by Stedmon and Markager  with data from the Greenland Sea. Samples with high fmw, and therefore likely higher terrigenous CDOM, are indeed clearly outside the model limits for marine CDOM (Figure 4), and shows that simple absorption measurements can in this case be used to indicate the origin and mixing of CDOM. Amon et al.  found that DOM in the EGC is largely of terrigenous origin using fluorescent properties of DOM, our results support this conclusion.
 In Polar Water, the samples with fmw from 0.07 to 0.13 are used to represent the characteristics of the riverine CDOM exported with the EGC (samples with fmw > 0.13 are excluded, see below, although including these samples would not change the S and SR values significantly). For this subset of samples (n = 31) S and SR values are 20.1 ± 1.2 μm−1 and 1.22 ± 0.13, respectively. Samples in the Atlantic water inflow (n = 97), S and SR values are 19.0 ± 2.7 μm−1 and 1.88 ± 0.47, respectively. The difference in SR values between EGC and Atlantic waters is largely due to a much higher spectral slope in the 275–295 nm range in Atlantic waters, which is indicative of more photochemically altered material or of another CDOM source. SR values in Polar Water of about 1.2 are relatively low, and comparable to coastal waters reported by Helms et al. , further highlighting the terrigenous origin of CDOM in the EGC.
 In polar waters it is more appropriate to regress fmw rather than salinity against a350 in order to evaluate the conservative behavior of terrigenous material, as there are more than two major sources of saline or fresh water [Dittmar and Kattner, 2003a; Benner et al., 2005; Granskog et al., 2009]. Indeed there is a strong and significant correlation between fmw and a350 in the Fram Strait as indicated in Figure 3 and confirmed in Figures 5c and 5d. The salinity–a350 relationship is affected by interference from sea ice melt [cf. Letscher et al., 2011]. Samples with high positive sea ice melt are displaced from the Atlantic water to halocline water mixing line (Figure 5a). It is evident that sea ice formation and melt can leave an imprint in the salinity-CDOM relationship, and one has to be cautious in using salinity-property relationships alone in Arctic waters to deduce conservative behavior especially if only surface samples are used [Letscher et al., 2011]. The fmw and a350 show a strong correlation for fmw < 0.13 in Polar Water (Figure 5d), R2 = 0.89, implying that the meteoric water fraction can to a large degree explain the variability in CDOM. However, a number of samples with fmw > 0.13 deviate from the main mixing line show in Figure 5d. These samples (west of 7°W and shallower than 25 m) contain excess sea ice meltwater, relative to the saline subsurface waters with high-brine and meteoric water (although they still have a net negativefsim fraction), diluting the concentration of CDOM. The CDOM in this subset of samples has similar S and SRvalues as all the other samples in Polar Water, which suggests that they have not been affected by photobleaching, but rather dilution, reducing the intensity but not the characteristics of CDOM absorption. The subset of points can be extrapolated toward a low-CDOM, low-salinity source (shown as open circles inFigure 5b), which also has a high δ18O value (not shown). The high δ18O value of the source reveals it to be sea ice meltwater rather than precipitation or meltwater from Greenland.
3.2. Changes in CDOM Optical Properties
 The regression of fmw and a350 (regression shown in Figure 5d) can be carried out for all wavelengths (here 250 to 600 nm), to estimate the apparent river water spectra in Polar Water (for this study taken to be fmw= 0.1, in order not to extrapolate) and the marine end-member spectra (i.e.,fmw = 0) in the Fram Strait (Figure 6a). The spectra differ in shape (S and SR), which indicates the preferential loss of a specific fraction of CDOM. S and SR values are 16.4 μm−1and 0.85 respectively, for the observed river end-member reflecting the terrigenous nature of CDOM [Stedmon et al., 2011]. The Fram Strait derived riverine CDOM spectrum (i.e., fmw = 0.1) from the regression model has higher S and SR values (20.2 μm−1 and 1.10, respectively). This change suggests that during its passage from river waters, across the shelf seas and the Arctic Ocean, riverine CDOM has been altered with a preferential loss of absorption at longer wavelengths, which can qualitatively be seen in the different shapes of the spectra in Figure 6b. From the model, the marine end-member (i.e., atfmw = 0) has values of 20.0 μm−1 and 1.86 for S and SR, respectively, which agrees with the measured data. Absorption at the longer wavelengths is associated with more aromatic and larger molecular weight material [Blough and Del Vecchio, 2002; Helms et al., 2008]. Thus an increase in S and SR indicates a decrease in average molecular weight and aromaticity of CDOM. This is consistent with what is expected during the exposure of terrigenous organic matter in the marine environment. Here physical, biological and photochemical processes act to either remove the high molecular weight fraction or transform it to lower molecular weight material [Amon and Benner, 1996; Søndergaard et al., 2003; Dittmar and Kattner, 2003b; Helms et al., 2008].
3.3. Tracer-Based Estimates of CDOM Conservation or Removal
 The regression of a350 with fmw gives a zero fmw intercept of 0.14 m−1 (Figure 5d), which agrees with the a350 values measured in Atlantic water on the western part of the transect (Figure 3b), and an intercept for fmw at unity of 4.0 ± 0.3 m−1, indicative of the apparent riverine end-membera350in the Fram Strait. Compared to the discharge-weighed Arctic riverinea350 of 15.5 m−1 (Figure 2), or 12.3 m−1 if we take into account removal of the labile terrigenous CDOM fraction, this implies a a350 loss of 67–74%, assuming all meteoric water is of riverine origin. However, Serreze et al.  has shown that as much as 25% of the meteoric water in the surface Arctic Ocean could originate from direct precipitation and this would influence this loss calculation. Assuming no CDOM in precipitation, the combined effects of microbial removal of the bioavailable fraction of the riverine CDOM loading during the spring freshet and thereafter the dilution by the contribution from precipitation (25% to the total meteoric water fraction in surface waters), results in an expected a350end-member atfmw = 1 of 9.2 m−1 which implicates a loss of 57% in a350during its transit to the Fram Strait. In simple terms, if the meteoric end-member at the Fram Strait has been diluted to a fraction of about 0.125,a350 values of about 1.2 m−1 would be expected in the Fram Strait, while values that are observed are about 0.6 m−1.
 This loss can be calculated for all wavelengths (regression of absorption at each wavelength against fmw is used, as is done for 350 nm in Figure 5d) as an integrated absorption loss between 250 and 600 nm and is estimated to be 59%, using the river spectrum averaged for all six rivers as reference (Figure 6a). If the bioavailable portion (30%) of riverine CDOM inputs during spring freshet is removed (see above) from the river inputs, the spectrally integrated loss is about 49%. These estimates agree with the above estimates calculated for a350 only.
 Photochemical degradation can result in a preferential removal of CDOM relative to DOC [Vähätalo and Wetzel, 2004; Osburn et al., 2009]. For example, in a study on the Mackenzie shelf in the western Arctic, Osburn et al.  report losses of <10% of CDOM compared to 1% for DOC. However, there is little evidence that photochemistry is currently a major sink for terrigenous DOM in the Arctic Ocean [Benner et al., 2005; Bélanger et al., 2006; Osburn et al., 2009]. Other likely removal processes that may be influencing riverine CDOM in the Arctic Ocean are microbial degradation and sea ice fractionation. The effects of microbial degradation on CDOM absorption spectra are unclear and likely the net result will vary depending on the relative proportion of terrigenous and marine CDOM.
 The fractionation of CDOM during sea ice formation is poorly understood although initial studies indicate there is a preferential retention of up to 34–39% CDOM within sea ice relative to salts [Müller et al., 2011]. It is likely that this process is occurring but it is difficult to derive an estimate of its quantitative importance. As the volume of ice formed relative to the volume of the underlying seawater on the shelves can be considered to be very small, it is unlikely that sea ice retention is a dominant process in controlling CDOM concentrations. In fact field data suggest that the net effect of sea ice formation is to increase CDOM concentrations in brine influenced waters such as those of the halocline layer [e.g., Amon et al., 2003; Guéguen et al., 2007; Nakayama et al., 2011] and dilute the CDOM concentrations in Arctic surface waters upon melt.
 Although the above estimates of CDOM removal are crude, they do fall within the ranges reported by Alling et al.  and Letscher et al. for DOC, which were also based on property-property plots to assess conservative behavior. What remains is to consider these estimates in light of a mass balance for CDOM in the region similar to those carried out for DOC in earlier studies [Amon et al., 2003; Benner et al., 2005].
3.4. Assessing the Export of CDOM in the Fram Strait
 The estimates of CDOM removal from the end-member extrapolations based on the mixing curves (Figure 5) can be put into context by comparing with the actual fluxes of CDOM across the Fram Strait with observed riverine loading based on data from Stedmon et al. . Here we use the observed CDOM concentration in the Fram Strait from September 2009 (Figure 3b) jointly with modeled currents, September and annual mean flows in 2009 from the ECCO2 model run (annual mean velocity field is shown in Figure 7a) to calculate the CDOM fluxes in the Fram Strait (Figure 7c). To further assess the importance of the northward transport with the North East Greenland Coastal Current (NEGCC), which is found both on a monthly and annual basis in the ECCO2 model (Figure 7a), and also has been observed along the Greenland coast [see Rabe et al., 2009, and references therein], we make the assumption that CDOM decreases by 20% going from the westernmost station at 15°W to the Greenland coast (Figure 7b). This agrees with the relative decline of in situ CDOM fluorescence apparent in an earlier study [Amon et al., 2003].
 The majority of CDOM resides on the Greenland shelf (Figures 3b and 7b), a part of the strait that makes a comparatively minor contribution to the net volume transport estimates, according to the ECCO2 model results (Table 1). This is mainly due to the northward flow of the NEGCC, but nevertheless these waters contribute significantly to meteoric water transport southward [e.g., Rabe et al., 2009]. Further the CDOM flux estimates are very sensitive to the match or mismatch between the volume transport field and CDOM distribution. Most of the volume transport is offshore of the shelf slope, east of about 4°W in ECCO2, and the tongue of high CDOM observed in 2009 at around 5°W, is associated with the greatest flux of CDOM out of the Arctic in the ECCO2 based estimates (Figure 7c). This phenomenon may be a transient feature of the CDOM field, such as an eddy. Therefore, we surmise that the CDOM flux estimates given here for EGC (Table 1) represent a slight overestimate of CDOM transport out of the Arctic Ocean through the Fram Strait in the core of the EGC. The largest uncertainty arises from the fact that observations only represent the CDOM distribution in September, as similar data from other months are not available. However, preliminary data from CDOM fluorometers installed on moorings from September 2011 to September 2012 show that there is minimal seasonal variability in CDOM at depths of 50 and 100 m, where the maximum CDOM absorptions are found (M. A. Granskog and C. A. Stedmon, unpublished data, 2012). Estimates of annual CDOM transport flux using the September and annual mean velocity fields agree fairly well (Table 1) and we therefore believe that the September conditions do not deviate considerably from annual averages neither in terms of the velocity field or CDOM. Despite these uncertainties this approach still offers a more refined CDOM flux estimate compared to previous estimates in the literature, which were based on bulk volume fluxes in the EGC and mean terrigenous DOM concentrations alone [e.g., Opsahl et al., 1999].
 Transport through the Fram Strait is characterized by a northward flux of marine CDOM from the North Atlantic via the WSC (Table 1). Further to the west CDOM fluxes are dominated by terrigenous CDOM in the EGC, shelf and slope waters flowing southward. Near the Greenland coast there is a recirculation of a fraction of this flux back northward with the NEGCC. Assuming that the CDOM transported in the shelf, slope, EGC and NEGCC waters is all of terrigenous origin, we obtain a net flux of terrigenous CDOM of 24.9 1012 m2 yr−1 (Figure 8). Comparing this flux with the estimated riverine loading of CDOM to the Arctic Ocean (Table 1 and Figure 8) it appears that almost half the terrigenous CDOM input is exported to the North Atlantic through the Fram Strait. This is in the upper range of the values of terrigenous DOM export through the Fram Strait as estimated by Opsahl et al.  and Amon et al. , but is higher than the estimate by Benner et al. . Since CDOM is strongly related to meteoric freshwater in the Fram Strait and the flux of freshwater through CAA and the Fram Strait are roughly equally split [Rudels, 1987], this would indicate that CDOM export through CAA could equal that through the Fram Strait, indicating that terrigenous CDOM supplied to the Arctic Ocean is basically conserved and exported to the North Atlantic. In order to test this hypothesis further a better understanding of the export of terrigenous CDOM from the Eurasian to the Canada basin is required as well as measurements of the fluxes of terrigenous CDOM from the Arctic Ocean through the CAA (Figure 8). However, seasonality in the CDOM exports needs to be examined closer in the Fram Strait since there is both seasonality in the freshwater fluxes [de Steur et al., 2009] and seasonal and interannual variability in meteoric water fluxes [e.g., Meredith et al., 2001; Dodd et al., 2012].
3.5. Uncertainties in the Arctic Ocean Terrigenous CDOM Budget
 The results from the end-member extrapolation using mixing curves and the flux estimates (see above) are conflicting, with the former indicating significant removal of terrigenous CDOM and the latter suggesting nearly conservative behavior as export fluxes (through the Fram Strait and CAA combined, based on current knowledge) could equal terrigenous CDOM inputs. The question remains how to resolve this contradiction arising from these two independent approaches.
 Obviously there is a need to improve estimates of the fluxes of terrigenous CDOM through the CAA and the Fram Strait. Transport in deeper waters (below 500 m in the Fram Strait) need to be estimated, however, this is unlikely to be significant due to limited meteoric water at greater depths. Further, although we assumed that all CDOM exported by the EGC is of terrigenous origin, based on the fact that the material has a terrigenous signal (spectral slope value), a fraction of the export is nevertheless likely composed of a “background” component imported with the marine water inflow to the Arctic Ocean, and our export estimates therefore constitute an upper limit of the terrigenous CDOM export through the Fram Strait.
 Whether the relationship with meteoric water is similar in the CAA to that observed in the Fram Strait needs to be examined. This might not be the case since the predominant freshwater source in CAA is Pacific and North American runoff (latter with low CDOM), in contrast to Eurasian runoff and Pacific waters in the Fram Strait [Jahn et al., 2010]. Further the transport of CDOM through CAA could be significantly lower than through the Fram Strait due to increased remineralization of DOM in the Beaufort Gyre as a result of its longer residence time [Hansell et al., 2004].The recent accumulation of freshwater in the BG [e.g., Morison et al., 2012] has not yet resulted in any significant changes in the freshwater fluxes through the Fram Strait [Dodd et al., 2012]. Atmosphere-ice-ocean interactions largely control the circulation of surface waters in the Arctic, and hence the circulation patterns of Siberian runoff [Timmermans et al., 2011], and therefore indirectly will affect the residence times and time available for removal of CDOM before exported to the North Atlantic. However, it is evident that largest CDOM export fluxes in the Fram Strait are actually in the subsurface waters, and hence sea ice formation may play an equally important role as surface circulation patterns, in the dynamics of CDOM in the Arctic Ocean [cf. Guéguen et al., 2007]. It is also evident that fractions of meteoric water and brine are coupled in the Fram Strait [Dodd et al., 2012], with more brine associated with higher meteoric fractions. This suggests that CDOM is introduced to the halocline water with river water drawn to depth by brine expulsion during sea ice formation [Guéguen et al., 2007]. The source of this water mass in the Fram Strait is suggested to be the Siberian shelves, which is supported by the fact that little evidence of Pacific water was found in the Fram Strait in 2009 [Dodd et al., 2012].
 To restrain the export fluxes better there is a need for CDOM data in the CAA, especially at the important gateways (Smith and Lancaster sounds) not to mention coincident volume and freshwater flux data. Further the methods used to assess conservative behavior based on synoptic observations of conservative tracers need also to be scrutinized. In particular an assessment of how contributions from different rivers with slightly different end-member characteristics vary spatially along the shelves and subsequently alter the apparent combined river signal. Additionally the effects of repeated ice melt and formation cycles on shelf water properties, which until now are assumed conservative (i.e., would not affect the net (C)DOM budget) needs to be assessed if accurate estimates of DOM removal rates and export are to be obtained.
 Estimates for terrigenous inputs of (C)DOM to the Arctic Ocean could also be improved. A better assessment of the smaller rivers draining to the Arctic Ocean is needed, both from the North American and Eurasian drainage basins, as they contribute a substantial portion of the total river freshwater efflux, and only the largest rivers have been sampled for CDOM [Stedmon et al., 2011]. An assessment of the possible contribution through Bering Strait of terrigenous CDOM from the Pacific is also lacking. Contributions from coastal erosion cannot to be completely neglected [e.g., Stein and Macdonald, 2004], although little is known how effectively the largely particulate contribution from erosion is transferred to the dissolved phase and would contribute to CDOM inputs to the Arctic Ocean. Similarly the contribution to CDOM characteristics and distribution from the degradation of particulate terrigenous organic matter in surface sediments on the shallow shelves, is unknown. Before estimates based on both approaches converge, the fate (export or remineralization) of terrigenous CDOM in the Arctic Ocean remains elusive.
 Concurrent observations of CDOM with water mass tracers in the Fram Strait, one of the major pathways of Arctic Ocean exchange with the world ocean, have shown that rather simple measurements can be used to examine the dynamics and fate of terrigenous (C)DOM in the Arctic. CDOM absorption could aid to distinguish low-CDOM freshwater from riverine derived meteoric water, and on the other hand also be used as a parameter to quickly estimate the fraction of meteoric water (or river water) in the Fram Strait, although more data are required to see if the relationship is robust. Using the fluorescence of CDOM could even be a more sensitive and suitable tool [cf.Amon et al., 2003], as then certain terrigenous fractions of the CDOM pool could be traced independently.
 The East Greenland Current carries a strong terrigenous signal, which is modified by both riverine inputs but also sea ice formation. The strong coupling of sea ice brine and meteoric water found in the Fram Strait in 2009 [Dodd et al., 2012], with high CDOM imprint, is indicative of a possible source from shallow shelves of Laptev and Kara seas [Bauch et al., 2009, 2011], where upper halocline waters are formed in coastal polynyas and entrain a river water signature into the Arctic halocline.
 Based on conservative mixing estimates it appears that by the time river water exits the Arctic Ocean in the East Greenland Current a third to a half of the CDOM has been removed, which agrees with recent studies on Siberian shelf seas [Alling et al., 2010; Letscher et al., 2011]. First-order flux estimates of CDOM indicate that up to a half the riverine supply of CDOM to the Arctic Ocean could be exported through the Fram Strait, in fair agreement with earlier estimates [Opsahl et al., 1999; Amon et al., 2003]. Since freshwater transport through the Canadian Arctic Archipelago equals or is larger than through the Fram Strait [Serreze et al., 2006], but the sources of freshwater likely differ from that in the Fram Strait [Jahn et al., 2010], it is critical to assess the exchange of terrigenous CDOM between the Eurasian and Canada basins and subsequent transport of (C)DOM through CAA, before an overall assessment of the DOM exports from the Arctic Ocean. Seasonality through important gateways is also important to quantify better.
 It is apparent that both sea ice formation and the pathways of freshwater, especially Siberian river water, play an integral role in the fate of terrigenous (C)DOM in the Arctic Ocean. Atmospheric circulation patterns to a large degree control the pathways of Siberian runoff and brine-rich waters formed on the shelves [e.g.,Timmermans et al., 2011; Bauch et al., 2009], and this is likely to have an effect on the export pathways and the residence time of river waters in the Arctic basin. Both sea ice formation and pathways of river water play an important role when assessing the present-day and future fate of terrigenous CDOM in the Arctic Ocean. Pathways and residence times of river water and halocline waters may therefore control whether CDOM is exported from or removed within the system. The data presented here stem from a “normal year”; the amount of freshwater/meteoric water present in the Fram Strait during this period was comparable to earlier years [de Steur et al., 2009; Dodd et al., 2012], in line with the observed atmospheric circulation patterns [Timmermans et al., 2011].
 Many hurdles remain with respect to tracing the flux and fate of terrigenous DOM in the Arctic Ocean. In addition to a large terrigenous loading of DOM there is also considerable autochthonous production, which is often not taken into account. Primary production is on the order of 500 Tg C per year [Arrigo et al., 2008] and assuming 10% is channeled to DOC suggests that the riverine and autochthonous DOC supply to the Arctic Ocean are of equal magnitude. However, simple CDOM measurements present a promising approach to differentiate between these two DOM sources.
 The 4.5 km Arctic Ocean simulation was provided by G. Spreen and was carried out as part of the NASA Modeling Analysis and Prediction (MAP) ECCO2 project. M.A.G, P.A.D., and E.H. were supported by the Centre for Ice, Climate and Ecosystems (ICE) at the Norwegian Polar Institute. R.M.W.A. was funded by grants from the National Science Foundation (ARC 0425582, ARC 0713991, and OPP 0229302 subcontract). C.A.S was funded by the Carlsberg Foundation, Greenland Climate Research Centre, and Danish Strategic Research Council (NAACOS project). A.K.P. was supported by the Fram Arctic Climate Laboratory. Comments from three anonymous reviewers aided in improving the manuscript.