Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 The optical properties of chromophoric dissolved organic matter (CDOM) were investigated in the Canadian Archipelago and coastal Beaufort Sea surface waters using fluorescence spectroscopy coupled with parallel factor analysis (PARAFAC). Environmental dynamics of individual components were evaluated and compared to salinity, in situ fluorescence, absorption at 312 nm (a312), dissolved organic carbon, and lignin phenol concentrations. A positive linear relationship between four fluorescent components and lignin phenols suggests a terrestrial origin, whereas two components were unrelated to a river source, suggesting an autochthonous source. Elevated concentrations of terrestrial components were observed in the Mackenzie River plume near the coast of Alaska and decreased as water was transported to the Canadian Archipelago. The two nonterrestrial components exhibited only background levels in concentrations along the transect, suggesting minimal productivity within plume and archipelago surface waters. The relative abundance of terrestrial components in relation to nonterrestrial components allowed us to distinguish water masses including Atlantic, Archipelago, and Mackenzie River plume, respectively. This study illustrates the usefulness of PARAFAC to fingerprint water masses based on the optical characteristics of CDOM and shows promise to improve our understanding of upper Arctic Ocean ventilation.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Dissolved organic matter (DOM) is a heterogeneous mixture of compounds which remains largely uncharacterized, yet it plays an active role in the biogeochemistry of the carbon cycle on a global scale [Hansell and Carlson, 2002]. It results from a range of processes including algal exudation, viral lysis, grazing and fluvial inputs and its composition varies depending on proximity to source and exposure to degradation processes [Hansell and Carlson, 2002]. Colored DOM (CDOM) represents the light absorbing constituent of the DOM pool in natural waters and absorbs light in the ultraviolet and visible wavelength range [Hansell and Carlson, 2002]. The optical properties of CDOM have been previously used to predict dissolved organic carbon (DOC) concentrations, distinguish compositional characteristics and discriminate between terrestrial and marine DOM sources [Ferrari and Dowell, 1998; Stedmon and Markager, 2005a; Spencer et al., 2008]. More recently, the combination of spectroscopic fluorescence Excitation and Emission Matrices (EEMs) with Parallel Factor Analysis (PARAFAC) has enabled researchers to decompose the combined CDOM fluorescence signal into components corresponding to a chemical analyte, or group of strongly covarying analytes [Stedmon et al., 2003; Murphy et al., 2007; Stedmon and Bro, 2008]. The ability to differentiate and trace sources of CDOM and to determine the underlying factors controlling speciation during its transport and mixing represents a major progress in the field of CDOM biogeochemistry and opens new possibilities for the use of CDOM as a more specific tracer in oceanography.
 In this study, we apply parallel factor analyses of EEMs from a set of samples to characterize CDOM throughout the Canadian Archipelago and coastal Beaufort Sea surface waters and to determine its usefulness to trace river water in the Arctic. The Arctic Ocean is an ideal region to study the fate of terrestrial DOM because it receives ∼11% of the global river runoff, yet it constitutes ∼1% of the global ocean volume [Opsahl et al., 1999]. Additionally, the concentration of organic carbon in Arctic rivers is relatively high [Lobbes et al., 2000]. Therefore, the flux of terrestrial DOM to the Arctic is much greater than corresponding fluxes to other oceans in the world, but its fate is not entirely clear. Determining the fate of terrestrial DOM in the Arctic is of global importance because a large fraction of terrestrial organic carbon is stored in soils, mostly in the vast permafrost regions around the Arctic, which are susceptible to climate change. As the climate warms, freshwater discharge by large Arctic rivers is predicted to increase, with a ∼7% increase already observed [Peterson et al., 2002]. Therefore, increased freshwater discharge combined with permafrost erosion and extended ice-free periods have the potential to increase the supply of soil organic carbon to the Arctic Ocean.
 The dominant source of terrestrial DOM to the coastal western Arctic is the Mackenzie River in northern Canada. The Mackenzie is the fourth largest river in the Arctic in terms of discharge [Gordeev, 2006] and supplies the coastal Beaufort Sea with approximately 1.4 × 109 kgs of DOC per year [Raymond et al., 2007]. A major portion of its fresh water is thought to be exported through the Fram Strait and the Canadian Archipelago [Cuny et al., 2005].
 Terrestrial DOM in the Arctic Ocean has been previously traced by determining the concentration of lignin phenols, an important component of vascular plants [Opsahl et al., 1999]. Unfortunately, using lignin phenol analysis to characterize terrestrial DOM is an expensive and arduous process involving large sample volumes and many time consuming steps of sample preparation and analyses. As the chemical nature of CDOM defines its optical properties, the absorbance and fluorescence properties of DOM have become a much more convenient proxy of DOM in the Arctic [Guay et al., 1999; Amon et al., 2003]. The goal of this study is to determine how the behavior of PARAFAC components compares to lignin phenol trends as a circulation tracer in the Arctic Ocean and to analyze what additional chemical information can be deduced from their spectral character. To investigate various sources contributing to the fluorescence signal, PARAFAC components were related to salinity, absorbance at 312 nm (a312), DOC, and lignin phenol concentrations. This study illustrates how fluorescence paired with PARAFAC can be used to describe the nature of DOM and demonstrates its potential to characterize surface waters in the Arctic Ocean.
2.1. Sample Collection
 Samples were collected during 8–28 July 2005 aboard the Swedish Icebreaker Oden. The transect began at 61°05.159′N, 013°32.750′W in the North Atlantic and ended at 70°58.75′N, 145°13.31′W near Barrow Alaska, via the Canadian Archipelago (Figure 1). In situ fluorescence was recorded and discrete surface samples were collected through a stainless steel intake approximately 12 m below the surface. In addition, we collected one surface ice melt sample (0 m) at station 26 (73°25.54′N, −96°16.15′W). Water samples were analyzed for salinity, DOM absorption and fluorescence, DOC and lignin phenol concentrations.
 In situ fluorescence was measured using a backscatter fluorescence probe (Haardt, Optic and Mikroelektonic, Germany) from the seawater intake line. The fluorometer emits light over a broad range of wavelengths and uses a band pass filter to obtain a fixed excitation ranging from 350 to 460 nm and collects emission at a fixed wavelength of 550 nm +/− 20 nm. These wavelengths were chosen based on empirical calibrations with terrigenous humic substances in order to obtain an optimal signal-to-noise ratio [Amon et al., 2003].
 Samples collected for optical properties and DOC were immediately filtered through precombusted 0.7 μm GF/F filters (Whatman) and stored in sealed precombusted glass ampoules at −20°C until analysis in the lab. For lignin samples, 10–15 L of seawater was filtered using a 0.2 μm pore size NuclePore™ filter cartridge, acidified to pH 2.5 using concentrated HCl (reagent grade), followed by solid phase extraction (SPE) using 60 CC/10 g C18 columns (Varian [Louchouarn et al., 2000]). Cartridges were stored at −20°C until analysis in the lab. Samples for optics, DOC and lignin phenols were analyzed within one year of sample collection.
Spencer et al. [2007a] investigated the effects of freeze/thaw on spectroscopic parameters for freshwater DOM samples with relatively high organic carbon concentrations. They suggest frozen storage of samples for optical properties and DOC concentrations may result in the fading of fluorescence and loss of DOC due to flocculation [Spencer et al., 2007a]. Freshwater DOM samples dominated by somewhat highly colored organic rich samples are typically more prone to flocculation loss during freezing than oceanic samples and therefore, freezing is a common storage method for oceanographic DOM studies [Amon et al., 2003]. Currently, we do not know the long-term effects of freezing on fluorescence fading during this study as the recovery sensitivities were not determined but, the loss of DOC during longer time storage at −20°C does not seem to be problematic in the Arctic based on our experience with sample comparisons and mass balances during ultrafiltration (R. Amon et al., unpublished manuscript, 2009).
2.2. DOC and Lignin Phenol Analysis
 DOC concentrations were determined using a modified MQ-1001 TOC Analyzer [Qian and Mopper, 1996; Peterson et al., 2003]. Potassium hydrogen phthalate was used as a standard to create a daily calibration curve and deep sea standards supplied by D. Hansell (University of Miami) were run daily to assure quality control. DOC concentrations were calculated using a calibration curve bracketing sample concentrations, followed by subtraction of a Milli-Q blank. The typical coefficient of variation was 3% for samples below 60 μM DOC and the international deep sea standard averaged 47 ± 1.4 μM DOC.
 Lignin-derived CuO oxidation products were determined using the method developed by Hedges and Ertel  with modifications by Louchouarn et al.  and Kuo et al. . Briefly, SPE columns were eluted with 35 mL HPLC grade methanol into a 250 mL flask, and then dried in a Savant SpeedVac (SC210A). The dried SPE extracts (2–5 mg OC [Louchouarn et al., 2000]) were mixed with CuO (≈300 mg) and Fe(NH4)2(SO4)2·6H2O (≈50 mg) in a stainless steel minireaction vessel, to which 8 %wt nitrogen-sparged NaOH solution was added (3 mL). Trans-cinnamic acid (3-phenyl-2-propenoic acid) and ethyl vanillin (3-ethoxy-4-hydroxy-benzaldehyde) were used as surrogate standards and were added directly (∼3 μg) to each minivessel after cooling. The aqueous solution was then acidified with 6N HCl and extracted (×3) with ethyl acetate. Extracts were dried with Na2SO4 and evaporated to dryness using a LabConco™ solvent concentrator. The CuO reaction products were redissolved in a small volume of pyridine (300 μL), and a subsample was derivatized with N,O-bis(trimethylsilyl)trifluoroacetamide (BSTFA) containing 1% trimethylchlorosilane (TMCS) at 75°C in a heating block (1 h). Separation and quantification of trimethylsilyl (TMS) derivatives of CuO oxidation by-products were performed using gas chromatography–mass spectrometry (GC/MS) with a Varian Ion Trap 3800/4000 system fitted with a fused silica column (VF 5MS, 30 m × 0.25 mm i.d. or 60 m × 0.25 mm i.d.; Varian Inc.). Each sample was injected, under splitless mode, into a deactivated glass liner inserted into the GC injection port and using He as the carrier gas (∼1.0 mL min−1). The GC oven was programmed from 65°C (with a 2 min initial delay) to 300°C (held 10 min) using a 4°C min−1 temperature ramp. The GC injector and GC/MS interface were maintained at 280°C and 270°C, respectively. The mass spectrometer was operated in the electron ionization mode (EI, 70 eV) using multiple reaction monitoring (MRM). Compound identification was performed using GC retention times and by comparing precursor/product spectra with those of commercially available standards. The analytical precision of the major CuO oxidation products and related parameters was derived from replicate analyses of standard estuarine sediments (i.e., NIST SRM 1944 [Kuo et al., 2008]), as well as standard fulvic acid extracts (IHSS 1S101F), and averaged 10% or lower.
 Lignin parameters characterized in this study are found in Table 1 and include Sigma6 (Σ6; defined by the sum of six major CuO oxidation products of lignin: vanillin, acetovanillone, vanillic acid, syringaldehyde, acetosyringone, and syringic acid) and Lambda6 (Λ6, carbon normalized yields of lignin (mg (100 mg OC)−1)). Λ6 was used to calculate the percent of terrestrial DOM within a water mass using the formula %tDOM = [(Λ6 Ocean/Λ6 River) *100%] according to Opsahl et al. .
Table 1. DOC, Lignin Phenol, and Optical Signatures of Different Water Masses in the Study Areaa
Here n.m., not measured; n.a., not available; n.d., not detected.
Fluorescence and lignin data used to represent the Mackenzie River end-member were obtained from the PARTNERS project, which includes seasonal data from the six largest Arctic Rivers and represents peak flow conditions during 2004 and 2005 from six of the largest Arctic rivers including the Mackenzie (R. Amon et al., unpublished manuscript, 2009). Stations used to represent other end-members are as follows: Atlantic 1–8; Archipelago 10–49; eastern plume edge 50–52; MR plume 54–59; western plume edge 60–61; high-salinity Archipelago stations 18, 19, 30, and 34; low-salinity Archipelago stations 37–39, 43, and 44; ice melt (0 m) 26.
Average of terrestrial (Group I) and nonterrestrial (Group II) fluorescent components.
 Samples were warmed to room temperature prior to optical analysis. Absorbance measurements were recorded on a Shimadzu UV-2401PC/2501PC using a 5 cm quartz cuvette. Absorbance spectra were measured from 200 to 800 nm at 0.5 nm increments and were blank corrected using Milli-Q water as a reference. Absorption coefficients (a, m−1) were calculated by the formula (2.303*A)/L, where A is absorbance at a specific excitation wavelength and L is the cuvette path length in meters.
 Fluorescence measurements were made on a Photon Technologies International Fluorometer (Quanta Master-4 SE) using a 1 cm quartz cuvette with excitation and emission slit widths set to 5 nm. Excitation emission matrix scans (EEMs) for each sample were obtained by collecting a series of emission wavelengths ranging from 230 to 600 nm (2 nm increments) at excitation wavelengths ranging from 220 to 450 nm (5 nm increments). Emission and excitation correction files generated by the manufacturer were applied to each sample EEM to correct for instrument specific biases; the signal was normalized to that from a reference detector during measurement to account for fluctuations in light source characteristics. Due to deteriorating signal-to-noise ratios, excitation wavelengths below 240 nm and emission wavelengths below 300 nm were removed from the data [Stedmon et al., 2003].
 For many samples, absorption coefficients were high enough to influence the excitation and emission light in the cuvette and therefore an inner filter effect correction was applied [Lakowicz, 2006]. An inner filter factor (I) was calculated by I = 10−[(a_ex) (0.005)) + ((a_em) (0.005))], where a is the absorption coefficient. I was then divided into each excitation emission pair within each EEM. Next, the fluorescence spectra were Raman calibrated by normalizing the data to the area under the Raman scatter peak at excitation wavelength 350 nm of a Milli-Q water sample ran daily [Lawaetz and Stedmon, 2009]. To remove the Raman signal, the Raman normalized Milli-Q EEM was then subtracted from each sample EEM, resulting in Raman units (R.U., nm−1). Rayleigh scatter effects were removed by deleting emission measurements at wavelengths less than or equal to excitation wavelengths + 20 nm.
2.4. PARAFAC Modeling
 This study used the “DOMFluorToolbox” in MATLAB according to the recommendations for data pretreatment and validation given by Stedmon and Bro . A total of 62 samples were used to generate the model. To organize the data for the modeling process, regions of no fluorescence or scatter were removed. This resulted in EEMs that ranged from 240 to 400 nm along the excitation axis and 320 to 580 nm along the emission axis. Next, a region of zeros were inserted in areas of no fluorescence (where excitation ≫ emission). To aid in correcting second-order scatter, emission wavelengths of 550 nm and greater were excluded. Cutting off these higher emission wavelengths greatly reduced the size of the region of missing values, aiding in the PARAFAC modeling process. Once the data were organized, outlier identification was performed and a six component PARAFAC model was validated using split-half validation and residual analysis.
 To identify potential outliers, the initial model was run with nonnegativity constraints with simple two to four-component models and sample weights for each model were compared. No outlier EEMs were identified during the analysis. A series of PARAFAC models were then run with 2 to 9 components fitted to the data, narrowing down that five to seven components may be adequate to describe the fluorescent variation within the data. Next, split-half validation was used to divide the data into two random halves of equal size and the model was run independently on the two halves. A strong overlapping of component loadings from the two halves provided evidence that a six component model best suited this data set because it did not reflect random noise, rather it reflected the intrinsic variation of the two independent data sets. Further, a seven component model could not be split-half validated. Next, residual analysis was used to help assess how much fluorescence was not being explained by the PARAFAC model. Examples of measured, modeled and residual EEMs for the six component model can be seen in Figure 2. Subtraction of the modeled from the measured spectra yielded a residual fluorescence an order of magnitude lower than the measured EEMs. Also, the six component model was capable of characterizing the fluorescence of DOM from two contrasting environments, plume waters versus ice melt, further confirming a six component model is adequate. Because both split-half validation and residual analysis of the EEMs validated a six component model and because the components identified by the model have unimodal emission maxima [Stedmon and Bro, 2008], we assume the six component model successfully grouped the fluorophores present in the study area.
3.1. Fluorescence Characterization by PARAFAC
 The six components identified by the PARAFAC model are referred to as BERC1–BERC6 (Figure 3) and their spectral characteristics are compared with those from earlier studies in Table 2. Results from previous studies have identified BERC1–BERC3 as being terrestrially derived material. Stedmon and Markager [2005a] found a similar component to BERC1 and showed this component to be humic and ubiquitous to a wide range of environments. In a study on surface ocean DOM, Murphy et al.  identified a comparable component to BERC2 and defined it as being humic-like and associated it with terrestrially derived DOM. Additionally, Stedmon et al. [2007a] found a similar component to BERC2 that was produced after exposing Baltic Seawater to UVA light and suggests it is a common product of terrestrially derived DOM, removed by either microbial or physical processes. Most noteworthy is the wide range of studies that have identified a fluorescent component identical to BERC3, suggesting it is ubiquitous to a wide range of environments [Stedmon et al., 2003; Stedmon and Markager, 2005a; Murphy et al., 2006; Stedmon et al., 2007a]. In earlier studies, BERC3 has been shown to represent terrestrial humic-like DOM and its production is dependent on the presence and quantity of other humic-like components [Stedmon and Markager, 2005a; Stedmon et al., 2007a]. BERC4 has not been reported in previous studies and is grouped with the terrestrial humic-like components based on its positive relation to terrestrial biomarkers used in this study.
Table 2. Spectral Characteristics of the Six Components Identified by PARAFAC Compared to Previously Identified Components
 The fluorescence spectrum of BERC5 is similar to the excitation and emission wavelength range of free or protein bound amino acids and therefore may be a mixture of tryptophan and tyrosine. Previous studies have associated this component with microbial activity, indicating autochthonous derived DOM [Coble, 1996; Stedmon and Markager, 2005b; Murphy et al., 2006]. BERC6 is similar to that of the classic M peak defined by Coble  and is defined as a marine humic-like component.
3.2. Spatial Variation of Salinity, DOC, DOM Absorption, Fluorescence, and Lignin Phenols
Figure 4 (and Table 1) illustrates the distribution of salinity, in situ fluorescence, the absorbance coefficient a312, DOC, and lignin phenols along the cruise transect at 12 m water depth. All parameters clearly indicate the Mackenzie River plume between station 54 and 59 at around 140°W. Salinity values ranged from 12.2 to 35.2 and averaged 27.9. Higher salinity values were observed in the North Atlantic (salinity of 35.2; station 1), at the southern edge of Greenland within the tip of the East Greenland Current (EGC; average salinity of 34.9; stations 2) and the southeastern side of the Davis Straight in the West Greenland Current (WGC; average salinity of 32.8; stations 3–8). Rudels  suggests Atlantic return water flows south within the EGC, which has been shown to mix with Polar surface waters enriched in terrigenous DOM [Opsahl et al., 1999; Amon et al., 2003]. Additionally, Cuny et al.  reported waters remnant of the Gulf Stream flow north within the WGC which can explain the higher salinities observed at these stations. Within the Archipelago, salinity slightly decreased as we approached the Mackenzie River plume (average salinity of 28.1; stations 10–49). Average salinities are lower than that defined by Rudels  for the outflow through the upper Canadian Archipelago (salinity 32.5–33), which likely results from the influence of ice melt at 12 m water depth. Salinity values began to decrease slightly between stations 50–52 (average salinity 25.4), defining the eastern edge of the plume located near the entrance of the Archipelago. A dramatic decrease in salinity occurred between stations 54 and 59, defining the plume region (average salinity 17.1). Between stations 60 and 61 salinity values began to increase slightly, indicating the western edge of the plume located in the coastal Beaufort Sea (average salinity 26.4) and a region more likely influenced by the inflow of Pacific water through the Bering Strait [Rudels, 2001; Yamamoto-Kawai et al., 2005].
 The optical properties, DOC and lignin phenol concentrations all exhibit an inverse relationship to salinity along the transect, experiencing higher values inside the Mackenzie River plume and lower values within the Archipelago and regions influenced by the Atlantic (Figures 4a and 4b and Table 1). All parameters showed minor variations between stations 1 and 49 with an obvious increase between stations 54 and 59. A slight decrease occurred at station 60 to 61 for in situ fluorescence, a312, DOC and lignin phenols, once again indicating the western edge of the plume.
 The fluorescent components identified can be grouped into two patterns (Figures 5a and 5b) based on their relation to other parameters in this study. Group I components (BERC1–4) follow the same trend as in situ fluorescence, a312, lignin phenols and DOC concentrations, showing higher concentrations in the Mackenzie River plume region and lower concentrations in the Archipelago and regions influenced by the Atlantic (Figure 5a). Group II components (BERC5–6) were highly variable along the transect and were less prominent in plume waters relative to Group I components (Figure 5b).
3.3. Salinity Relationships
 The relationship between DOM parameters and salinity were used to identify different end-members present in this study (Figure 6 and Table 1). It is clear there are two low-salinity water masses, including the Mackenzie River Plume (low salinity 17.1, high DOM) and Archipelago surface waters heavily influenced by ice melt (low salinity 28.1, low DOM). Lower-salinity samples were identified within the Archipelago region to represent locations mainly influenced by ice melt (salinity 23.0) and higher-salinity samples were also identified (salinity 29.9). The one high-salinity marine end-member includes waters derived from the Atlantic within the North Atlantic and within the EGC and WGC (high salinity 33.5, low DOM), which are at least partially mixed with Polar Surface Water (PSW). Two areas have been classified as mixing regions due to their moderately higher salinities and lower DOM concentrations relative to plume waters and represent the eastern (entrance to Archipelago; salinity 25.4) and western edge of the plume (coastal Beaufort Sea; salinity 26.4), respectively. The higher salinities observed in Beaufort Sea coastal waters at the western edge of the plume likely results from the influence of Pacific inflow waters which enter through the Bering Strait [Rudels, 2001; Yamamoto-Kawai et al., 2005]. The slight decrease in salinity observed in the Beaufort Sea relative to the Atlantic end-member is reasonable as Pacific water has a lower salinity (32.5) than Atlantic waters (35) [Rudels, 2001] and because it is likely Beaufort Sea coastal waters are influenced by plume waters. The samples from the Mackenzie plume were easily distinguishable from the rest of the data and fell along its own mixing line (Figure 6).
 Linear regressions were tested for their significance on the Mackenzie plume data and the remaining data (Table 3). The Mackenzie plume data had a significant (P < 0.001) relation to salinity for DOC, a312, and BERC1–4. The regressions for the remaining data (excluding the western plume edge in the costal Beaufort Sea) were not significant (P > 0.01) for most parameters.
Table 3. Results of the Regression Analysis Carried Out Between the Parameters and Salinitya
The data were split into two groups; Mackenzie River (MR) plume region (stations 49–59 inclusive) and the remainder (Archipelago; stations 1–48) of the data excluding the Beaufort Sea. Only regressions which were significant (P < 0.01) are shown; n.s., not significant.
3.4. PARAFAC: Terrestrial Biomarker
 Optical parameters used to detect terrestrial DOM in this study show a significant relation to salinity and to the concentration of lignin phenols (Figures 6 and 7) in the river plume region. The absorbance coefficient a312 (Figure 7a) and fluorescence component BERC1 (Figure 7b) are significantly correlated to the terrestrial biomarker within the plume region, with a312, BERC1 and BERC3 (data not shown) showing the strongest relationship. Lignin phenols, a312 and terrestrial component BERC1 show a significant relation to the DOC concentration in the plume region (Figures 7c and 7d) and explain much of the variability in the DOC data set, with lignin explaining 79% (data not shown), a312 88% (Figure 7c) and BERC1 78% (Figure 7d) of the variability, respectively. The relationship of the optical parameters to DOC is much weaker outside the plume (Figures 7c and 7d), indicating the presence of other DOM sources or the preferential removal due to photobleaching of chromophoric DOM during coastal mixing [Osburn et al., 2009].
4.1. Role of OM Sources for PARAFAC Model Components
 Terrestrially derived DOM is mainly responsible for total DOC concentrations within plume surface waters and is less abundant in surface waters found in the Archipelago (heavily influenced by ice melt) and surface waters influenced by the Atlantic (Figure 7). Group I components clearly follow the trend of salinity and terrestrial biomarkers (Figures 4 and 5) and exhibit a significant positive relation to lignin phenols (Figure 7b), indicating their link to riverine DOM. Although more data points are required to fully validate the relationship of different PARAFAC model components to lignin phenols, initial results suggest that Group I components have the potential to trace terrestrial organic matter in the Arctic Ocean.
 BERC3 showed the strongest relationship to the in situ fluorescence signal (data not shown; R2 0.8) whereas BERC1 was best explained by lignin phenol concentrations (Figure 7d) and was found to be more relevant in all waters in comparison to the other terrestrial components (Figure 5a). In addition, a component very similar to BERC1 was identified by a PARAFAC model using data from the 6 largest Arctic rivers, which includes seasonal data and represents peak flow conditions during 2004 and 2005 (Tables 1 and 2) (PARTNERS Project, R. Amon et al., unpublished manuscript, 2009). Collectively, these results suggest that while all components could be used as a potential proxy of river water in Arctic surface waters, BERC1 may be best suited to track river water being exported from the Mackenzie.
 Group II components showed little to no relation to salinity or lignin phenols and were highly variable along the transect (Figure 5b). Both BERC5 and BERC6 are indicative of autochthonous production of DOM [Stedmon and Markager, 2005b; Stedmon et al., 2007b] and are not associated with terrestrial biomarkers. Because both components exhibited low fluorescence along the transect, there is likely minimal microbial production within the upper 12 m of plume and Archipelago surface waters. Holmes et al.  describes a substantial seasonal variability in the lability of DOC transported by Alaskan rivers to the Arctic Ocean, with DOC being more labile during peak discharge in late May to early June when freshet occurs and the majority of DOC is exported, and more refractory during mid to late summer months. While these components are nonconservative and cannot be used as a tracer, the low production observed in plume waters likely results from sampling past peak flow conditions.
4.2. Relative Contribution of Terrestrially Derived DOM
 Carbon normalized yields of lignin oxidation products (Λ6) have been used in end-member mixing models to estimate the relative contribution of terrestrial DOM (% tDOM) in marine systems, assuming dissolved lignin is removed from seawater at similar rates as riverine DOC [Opsahl et al., 1999]. In our data set, Λ6 indicates that 37% of DOM is of terrestrial origin in the plume area, 17% of DOM in the surface waters of the Archipelago is of terrestrial origin and 19% is terrestrially derived in waters within the EGC and the WGC. This estimate falls within the range of that reported by Opsahl et al. , who estimated 14–24% of the total DOC in polar surface water is terrigenous.
 Two avenues have been proposed as possible export routes of polar surface waters with relatively high concentrations of terrigenous DOM, including the EGC and the Canadian Archipelago [Opsahl et al., 1999]. Amon et al.  found a fluorescence maximum within the EGC and used the relationship between in situ fluorescence and lignin to show the terrestrial fraction of DOC ranged between 1.5 and 25% in near surface waters in the EGC. Further, Opsahl et al.  used lignin oxidation products to show that 9–27% of surface waters in the EGC were of terrigenous origin. While our sampling focused on the upper 12 m of Archipelago waters, meaning we likely missed a major portion of the river derived DOM leaving the Arctic Ocean through the Canadian Archipelago, a substantial fraction of DOM in these waters is of terrigenous origin and is comparable to values found in the EGC.
4.3. PARAFAC Components as a Fingerprint for Water Masses in the Arctic
 In a property/property plot of the two fluorescent component groups different water masses could be clearly distinguished, where Group I is used to represent terrestrially derived DOM and Group II represents marine production in the different water masses (Figure 8). The Atlantic end-member is mainly characterized by low production with minimal fluvial input. The two stations which stand out are 5–6 along the coast of Greenland, showing slightly elevated levels of Group II components, indicating a slight increase in production. Results from Hood and Scott  suggest that reductions in glacial extent have the potential to deliver increased amounts of labile DOM, which would increase production in nearshore coastal ecosystems and may explain the slight increase in production observed at these stations.
 Archipelago surface waters are characterized by minimal autochthonous DOM production and low levels of terrestrial DOM which likely results from the influence of ice melt at 12 m water depth. Within the Archipelago, ice melt at 0 m has a similar Group I/Group II signature as the Atlantic end-member, characterized by little in situ production and minimal fluvial influence and was lower than all other samples within the Archipelago. On the other hand, stations with elevated salinities sporadically identified throughout the Archipelago at 12 m are characterized by low levels of terrestrial DOM with slightly elevated levels of Group II components relative to plume and Atlantic waters. We have no information on the microbial community in those samples but, it is possible that primary production and microbial processing of organic matter within sea ice plays a role in the increased levels of Group II components [Amon et al., 2001].
 The eastern plume edge mixing region lies between plume and Archipelago end-members, showing a slight increase in both Group I and II components relative to the Archipelago end-member. Plume waters are characterized by a slight increase in production and reflect the abundance of terrestrial DOM being transported from the Mackenzie River to the coastal Arctic Ocean. The western edge plume mixing region, located in the coastal Beaufort Sea, has a similar GroupI/GroupII signature as the Archipelago end-member, which likely reflects the influence of Pacific inflow waters through the Bering Strait. Overall, results indicate a clear distinction of water types based on the relationship between modeled fluorescence components and suggest the usefulness of parallel factor analyses of EEMs to fingerprint water masses in the Arctic Ocean where DOM is characterized by a variety of sources and alterations [Amon and Meon, 2004].
5. Conclusion and Implications
 This study illustrates the potential of parallel factor analysis of EEMs to deliver model fluorescence components that can be related to biomarkers, in particular to lignin phenols. To date, there are limited studies which relate CDOM and lignin phenols concentrations [Hernes and Benner, 2003; Spencer et al., 2007b, 2008, 2009] and this is the first known attempt to relate PARAFAC components to this unambiguous tracer for terrestrial organic matter. Because the analysis of lignin is expensive, requires large water samples and is extremely labor intensive, the use of EEMs combined with PARAFAC can deliver a valid proxy for lignin phenols and most importantly can greatly increase sample resolution. Results from this study provide evidence that, in conjunction with a suitable calibration to lignin phenol measurements, PARAFAC components are a valuable asset for tracing the distribution of terrestrial organic matter in the Arctic Ocean, even when environmental concentrations are low.
 Components derived in this study were grouped into two classes based on sources, allowing for the distinction between terrestrial and marine derived DOM. Lignin parameters agree well with 4 of the derived fluorescent components tracing the influence of the Mackenzie River along the coastal Beaufort Sea into the Canadian Archipelago. Two components seem to be related to autochthonous DOM and suggest minimal production within surface waters at 12 m water depth. The relationship between the two groups of fluorescent components allowed us to distinguish different water masses encountered in this study and indicates the potential of this approach to trace water masses on a larger scale in the Arctic Ocean.
 This study was funded by grants from the National Science Foundation to R.M.W.A. (ARC 0425582 and ARC 0713991) and the Carlsberg Foundation to C.S. Additional funding was received through Texas A&M University including the Mooney Travel Grant, Texas Institute of Oceanography fellowship, and the Graduate Student Office. We thank the Captain, scientists, and crew of the Oden during Beringia 2005 Leg 1 for their assistance during the cruise and Amanda Rinehart for her help with sampling and assistance with fluorescence analysis.