Ultrafiltered and low molecular weight dissolved organic matter (UDOM and LMW-DOM, respectively) fluorescence was studied under simulated estuarine mixing using samples collected from Delaware, Chesapeake, and San Francisco Bays (USA) transects. UDOM was concentrated by tangential flow ultrafiltration (TFF) from the marine (>33 PSU), mid-estuarine (∼16 PSU), and freshwater (<1 PSU) members. TFF permeates (<1 kDa) from the three members were used to create artificial salinity transects ranging from ∼0 to ∼36, with 4 PSU increments. UDOM from the end- or mid-members was added in equal amounts to each salinity-mix. Three-dimensional fluorescence excitation-emission matrix (EEMs) spectra were generated for each end-member permeate and UDOM through the full estuarine mixing transect. Fluorescence components such as proteinaceous, terrigenous, and marine derived humic peaks, and certain fluorescent ratios were noticeably altered by simulated estuarine mixing, suggesting that LMW DOM and UDOM undergo physicochemical alteration as they move to or from the freshwater, mid-estuarine, or coastal ocean members. LMW fluorescence components fit a decreasing linear mixing model from mid salinities to the ocean end-member, but were more highly fluorescent than mixing alone would predict in lower salinities (<8). Significant shifts were also seen in UDOM peak emission wavelengths with blue-shifting toward the ocean end-member. Humic-type components in UDOM generally showed lower fluorescent intensities at low salinities, higher at mid-salinities, and lower again toward the ocean end-member. T (believed to be proteinaceous) and N (labile organic matter) peaks behaved similarly to each other, but not to B peak fluorescence, which showed virtually no variation in permeate or UDOM mixes with salinity. PCA and PARAFAC models showed similar results suggesting trends could be modeled for DOM end- and mid-member sources. Changes in fluorescence properties due to estuarine mixing may be important when using CDOM as a proxy for DOM cycling in coastal systems.
 For several decades, intensive research efforts have taken place to better understand dissolved organic matter (DOM) cycling, natural concentrations, chemical structures and biogeochemistry in coastal systems [Deuser, 1988; Hedges et al., 1997; Ittekkot, 2003; Ittekkot and Haake, 1990]. A large fraction of the total reduced organic carbon pool in coastal margins is represented by terrestrially derived organic matter (T-DOM) of which ∼1015g C is delivered to global estuaries each year [Hedges et al., 1997]. Due to the nearly absent terrestrial signal in open ocean DOM, there must be considerable T-DOM alteration from biological, photochemical and physicochemical processes during estuarine transport to or within the coastal ocean. Colored or chromophoric DOM (CDOM) is the light absorbing fraction and a major constituent of terrestrial and freshwater DOM, thought to be largely refractory and primarily diluted (∼90%) from estuarine waters during transport to the sea [De Souza Sierra et al., 1994, 1997; Vodacek et al., 1997]. Photodegradation and flocculation have been suggested as CDOM removal mechanisms during estuarine mixing [De Souza Sierra et al., 1994, 1997; Kieber et al., 1989; Sholkovitz, 1976; Vodacek et al., 1997], however, the nearly complete loss of this material in ∼2000 years, roughly the oceanic turnover time, necessitates a better understanding of DOM lability within estuarine systems [Deuser, 1988]. Estuaries are dynamic environments with multiple inputs, outputs and modes of constituent flux. The properties which make T-DOM “refractory” under a set of static conditions may be less evident under the constantly changing and heterogeneous physicochemical conditions found in estuaries.
 Numerous sources and largely uncharacterized biogeochemical DOM processing in estuaries and coastal ocean have made separating and identifying the exact nature of CDOM difficult, even under reasonably static conditions. Fluorescence is one property which is relatively easy to measure and provides a useful means of characterizing DOM functional unit (e.g., proteinaceous, humic, etc.) composition [Coble et al., 1990; Coble, 1996; Mopper and Schultz, 1993]. CDOM is readily found throughout estuaries and coastal regions where it controls the penetration of UV solar radiation, and is involved in the photochemical reactions of organic substances and biologically important metals like iron and copper [Zepp, 2002]. CDOM has been studied in various aquatic environments with a number of fluorescence spectroscopy techniques including emission scanning at fixed excitation wavelengths, synchronous scanning at a constant offset [Cabaniss and Shuman, 1987; Coble et al., 1990; Coble, 1996; Matthews et al., 1996; Senesi et al., 1991], and excitation-emission matrix spectroscopy or EEMs [Boyd and Osburn, 2004; Coble et al., 1990; Coble, 1996; Moran et al., 2000]. A distinct advantage of EEMs measurement is that data can be reprocessed as excitation spectra, emission spectra, or synchronous scan spectra [Coble et al., 1993]. Unlike single excitation scans, EEMs allows the determination of the peak position of the maximum emission at the maximum excitation as well as the fluorescence peak intensity. Peak position provides information that may be used to distinguish the origin of the CDOM [Coble, 1996; Coble et al., 1998; Del Castillo et al., 1999], study the changes that occur to CDOM as a result of various degradation processes [Moran et al., 2000; Parlanti et al., 2000] and to track water masses [Coble et al., 1998; Del Castillo et al., 1999]. The majority of CDOM fluorescence is attributed to humic-like compounds, which may account up to 50% of the total DOC of natural waters [Thurman, 1985]. Proteinacious compounds are another key functional group contributor to CDOM fluorescence [Coble et al., 1990; Mopper and Schultz, 1993]. The specific mechanisms underlying natural organic material functional group fluorescence have not been fully determined, as the molecular structure of this material is not universal or well characterized [Hedges et al., 2000]. However, EEMs analysis allows differentiation between proteinaceous compounds, terrestrial- or marine-derived humic substances, and microbially derived humic substances [McKnight et al., 2001].
 Recent investigations into biological and photochemical reactivity suggest that changes in salinity may transform CDOM thereby changing its lability during estuarine transport to the coastal ocean [Boyd and Osburn, 2004; Langenheder et al., 2003; Osburn et al., 2009; Osburn and Morris, 2003]. Mixing studies using isotopes to observe total DOC changes showed that significant transformations occur to DOM and possibly CDOM as well under certain estuarine conditions [Raymond and Bauer, 2000, 2001]. Simulated estuarine mixing along with corresponding microbial incubations on samples from the Chesapeake and San Francisco Bays showed that mixing alone and biological degradation of freshwater CDOM may alter optical properties [Boyd and Osburn, 2004]. With discrete waters collected during freshwater to marine transects, exposure to solar radiation caused a reduction in bulk fluorescence and shifts toward shorter excitation and emission maxima consistent with humic compound degradation [Del Vecchio and Blough, 2002; Moran et al., 2000]. Willey  showed that the addition of magnesium to estuarine systems caused an increase in natural fluorescence due to hydrogen ion loss or displacement of fluorescence depleting metals. A number of factors alter the optical properties of CDOM in natural waters; therefore a better understanding of the processes driving CDOM change is needed. For instance if CDOM undergoes significant structural changes during estuarine mixing perhaps optical properties such as fluorescence will show predictable changes within given estuarine regions (where mixing is prevalent). In this study, the effects of estuarine mixing were investigated, using CDOM from the Delaware, Chesapeake, and San Francisco Bay estuaries (USA). Data were analyzed in ways commonly reported in the literature: peak regions (Table 1) and peak ratios with more recent ordination techniques (PCA and PARAFAC) in order to provide historical context.
Table 1. CDOM Components Found in Representative EEMs
 Samples were obtained during cruises through the San Francisco Bay (2003) aboard the R/V Point Sur and in the Delaware and Chesapeake Bays from (2003, 2004) aboard the R/V Cape Henlopen. Three stations were sampled in the San Francisco, Delaware and Chesapeake Bays representing coastal marine, mid-estuary, and freshwater (Figures 1a–1c). While annual variations exist, residence times for these estuaries are generally on the order of months for San Francisco and Delaware Bays [Church, 1986; Smith and Mackenzie, 1987], and around one year for the Chesapeake Bay [Boynton et al., 1995]. Using the shipboard flow-through salinity measurements on every cruise, marine end-member (>30 PSU), estuarine mid-member (∼15 PSU) and freshwater end-member (<1) stations were identified. Sixty to eighty L water samples were collected at each station at a depth below 1% incident light penetration (420 nm; PUV-501, Biospherical Instruments, San Diego, CA) using the ship's CTD rosette. A Teflonlined tube was affixed to the CTD (conductivity, temperature and density sensor array) at the level of the salinometer. The proximal end was affixed to a pneumatic pump (Wilden, PVDF housing) and after ∼30 s of pumping to clear unwanted water from the hose, each large-volume sample was successively filtered through 3 μm and 0.2 μm pore size filter cartridges (Nucleopore Type QR) and transferred to acid-washed sterile 20 L polycarbonate carboys. DOM > 1 kDa (ultrafiltered DOM; UDOM) was concentrated by tangential flow ultrafiltration (TFF) using a custom made system consisting of a polyethersulfone membrane with a nominal surface area of 2.3 m2 [Benner, 1991]. A concentration factor of greater was calculated using absorbance at 280 nm (rather than DOC concentration) by the standard concentration factor equation
[Guo and Santschi, 1996]. Absorbance was used because all subsequent data in this manuscript are based on optical properties. All concentrations factors were greater than 85% (Table 2). The system was cleaned before initial and after each sample using 0.05 N HCl and 0.05 N NaOH and flushed with 40 L of >18 MΩ Milli-Q water, until background DOC concentrations were achieved. The UDOM and TFF permeates were collected, re-filtered and stored refrigerated in the dark using 0.2 μm polyethersulfone collection filter units (Nalgene) from collection until experiments were conducted in 2004–2006. Periodic UV-vis scans from the permeate and UDOM collection show no changes in optical signal (to this day).
Table 2. Salinity, Fluorescence Ratios, and TFF Concentration Factors for Initial Shipboard Samplesa
TFF CF (%)
M UDOM had very low fluorescence intensities (nearly at instrumental limit of detection, see Figure 2), such that the fluorescence ratio values were not useful in observing any trends. N.D., not determined.
May03, SF Bay
Mar03, DB Bay
Sep03, Ch Bay
Mar04, Ch Bay
Aug04, Ch Bay
2.2. Mixing Experiments
 The ultrafiltration permeates were used to create an artificial salinity mixing gradient (every 4 PSU) by mixing proportions of the freshwater (F) and mid-estuarine (E) (to create ∼0–16 PSU) and mid-estuarine (E) to marine (M) (to create ∼16–36 PSU). Initial salinities were measured using the ship's CTD system at the point of collection (Table 2). Each TFF-permeate salinity mixture was re-filtered and allowed to mix thoroughly. There was never any evidence of flocculation in any experiment conducted for this manuscript, either with UDOM concentrates or permeates. End- (freshwater) or mid-(estuarine) member UDOM was then added at 1:10 dilution to each permeate salinity mix. Marine UDOM was also added to the salinity transect, but resultant signal was too low for comparative results (therefore not presented here). For permeate “blanks,” a 1:10 dilution of freshwater (for F UDOM samples) or estuarine (for E UDOM samples) permeate was added to the salinity mix. In this way, any freshwater or estuarine permeate material in UDOM concentrates would be accounted for across the salinity gradient (see Figure 2). Some inherent absorbance was present in the permeates (ranging 10–20% more than Milli-Q blanks at 250 nm for the collection of samples). Each UDOM and permeate sample (tubes in Figure 2) was scanned from 250 to 600 for absorbance (referenced to MilliQ water) and on the fluorometer (with a MilliQ blank run for every 10 samples). For example, the 4 PSU sample with 1:10 dilution of freshwater UDOM was referenced to the 4 PSU sample with a 1:10 dilution of freshwater TFF permeate. The 4 PSU sample with 1:10 dilution of estuarine UDOM was referenced to the 4 PSU sample with a 1:10 dilution of estuarine TFF permeate. A 1:10 dilution was used to minimize the change in salinity while still providing ample optical signal.
2.3. Optical Measurements
 Every sample was analyzed by UV-vis absorbance spectroscopy on a Shimadzu UV-1601 spectrophotometer and excitation-emission matrix (EEM) fluorometry on a Shimadzu RF-5301-PC spectrofluorometer as previously described [Boyd and Osburn, 2004]. Daily instrument corrections were performed with fresh Milli-Q water. Absorbance was measured using a 5 cm cell from 250 to 600 nm with Milli-Q in the reference cell. For the EEMs collection, 5 nm steps were used to scan the excitation wavelength range from 250 to 450 nm. The emission intensity was measured from 250 to 600 nm at 1 nm intervals. The emission wavelength was scanned at 300 nm min−1. For sufficient resolution, slit widths were set for 5 nm excitation and 1.5 nm for emission. The highest instrument sensitivity settings (in software) were used to obtain the scans. Error associated with fluorescence measurements was estimated by running two separate samples 3 times (replicates) using a synchronous fluorescence scan with a 10 nm offset (250 to 600 nm). An average standard error was calculated for uncorrected spectral responses. The average error for each spectral wavelength was estimated as 1.7% of the instrument measurement. Error for peak selection (see section 2.4) routines was evaluated by running triplicate EEMs for each of the 10 UDOM samples (DB Mar03). Standard errors were compiled for each peak maximum value (in Raman units). Standard errors ranged from 0.80 to 17.9% with an overall average of 5.4%. UV-vis spectra had standard errors ranging from 0 to 0.45% with an average of 0.16%. Highest errors were at longer wavelengths (where signal was lowest). Shorter wavelengths were most important in inner filter corrections, so UV-vis spectral errors were considered negligible during mathematical processing. Manufacturer recommended excitation and emission corrections using barium sulfate and rhodamine B dye were applied to all data.
2.4. Data Analysis
 Optical data were exported to ASCII files and processed and corrected using a MATLAB™ script [Boyd and Osburn, 2004]. In the case of fluorescence measurements, scattering peaks (first and second order) were removed and smoothed over using a linear fill method and maximum excitation and emission wavelengths and intensities were determined for each peak region (Table 1). Fluorescence data were blank-corrected and normalized to the water Raman spectrum in order to obtain Raman units (R.U. nm−1) [Stedmon et al., 2003]. The fluorescence measurements were inner-filter corrected using absorbance measurements to allow for consistency when comparing between sample sites having different CDOM concentrations [McKnight et al., 2001]. For each experimental sample (see tubes in Figure 2), a UV-vis spectrum (referenced to MilliQ) and an inner-filter corrected excitation matrix was obtained. Permeate samples had very low UV-vis absorbance, but extremely high fluorescence at low salinities. This indicated highly fluorescent, low absorbing materials were present in freshwater (and perhaps estuarine permeates). This may have been due to lower molecular weight fulvic acids known to exhibit such properties [Mopper et al., 1996]. We originally intended to directly subtract the permeate EEMs from the UDOM+permeate EEMs, but the extremely high fluorescence at lower wavelength excitation produced permeate-corrected UDOM EEMs with negative values. To better estimate the permeate fluorescence contribution to the UDOM+permeate samples, we assumed that photons originating from permeate moieties would be subjected to quenching not only from permeate moieties (very low in absorbance at low wavelengths) but also to UDOM moieties (high absorbance at low wavelengths). For this reason, to subtract permeate EEMs from UDOM+permeate EEMs, the permeate inner filter corrections were re-calculated from permeate fluorescence spectra using the respective (seasonal sampling) UDOM+permeate UV-vis absorbance values. For instance, the 4 salinity DB Mar03 permeate (used to subtract permeate fluorescence from UDOM fluorescence) was inner-filter corrected with the DB Mar03 UDOM UV-vis spectrum. Permeate EEMs were reduced by the inner-filter correction factor (rather than increased), to account for quenching of photons emitted by permeate fluorophores within the UDOM+permeate matrix. The resultant permeate EEMs were then subtracted from corresponding UDOM+permeate EEMs to estimate the UDOM contribution to EEMs.
 Permeate-corrected UDOM EEMs (Figure 3c) and permeate EEMs (Figure 3d) for each seasonal sampling were then screened for corresponding excitation and emission wavelengths representing literature value “regions” identified in Table 1 [Blough and Del Vecchio, 2002; Coble, 1996; Parlanti et al., 2000]. The maximum peak excitation-emission pair was selected as the one giving the highest fluorescence signal within the defined peak region [Boyd and Osburn, 2004]. Using the excitation of 370 nm, a fluorescence index (370F450/500) was obtained from the ratio of the emission intensity at 450 nm divided by the emission intensity at 500 nm [McKnight et al., 2001]. Similarly, fluorescence ratios were calculated using 307 nm excitation (305 nm was used for the 5 nm spaced parameters) and 375 nm emission intensity divided by the 400 nm emission intensity (307F375/400) and similarly the fluorescence ratio of intensities at 430 nm emission divided by 540 nm emission and excited at 265 nm (265F430/540). These ratios were used to estimate the “microbial,” “terrestrial,” or “marine” character of the UDOM [Del Castillo et al., 2001; McKnight et al., 2001]. For plotting against salinity, the freshwater end-member was given a nominal value of 0 and the oceanic end-member was given a value of 36 salinity. This allowed samples with different end-member salinities to be more easily compared on the same graphs.
 In order to reduce dimensionality in the large data sets, Principal Component Analysis (PCA) and Parallel Factor Analysis (PARAFAC) were performed on peak data (e.g., Table 1) and inner-filter corrected spectral data, respectively. Statistical processing routines were performed in Matlab™ using PLS_Toolbox (Eigenvector Research, Wenatchee, WA). PCA decomposes the input array returning principal components that capture inherent variability and may assist in finding trends difficult to visualize within the original array. PCA was conducted by seasonal Bay sampling with salinity as the independent variable and peak excitation (nm), emission (nm), intensities (raman units) and fluorescence ratios (unit-less) were dependent variables. Each sampling event (e.g., DB, SF, CBS03, CBA04, CBM04) was identified as a class in order to be separately identified within the PCA model. Four PCA models were developed: permeate mixes with freshwater permeate (1:10), permeate mixes with estuarine permeate (1:10), freshwater UDOM + permeate and estuarine UDOM + permeate. Values were mean centered and scaled to unit variance prior to analysis. This analysis was conducted to determine if fluorescence property co-variation occurred within and between peaks and between estuaries. A single matrix including all peak excitation, emission, and intensity maximums and fluorescence ratios was analyzed by PCA. For graphing, score-based biplots were created; then quadrants were labeled to identify major loadings responsible for the sample groups. For instance, peak intensity loadings all fell within the positive PC1/positive PC2 quadrant (see Figure 9).
 PARAFAC decomposes arrays of N order (where N > = 3) into the summation over the outer product of N vectors (a low-rank model). PARAFAC was employed to determine if spectral regions (i.e., the 7 “standard” peak types) could be statistically identified as useful regions in tracking salinity-induced changes in UDOM during estuarine mixing. PARAFAC analysis was constrained by setting excitation and emission wavelength values and loadings along the salinity axis to nonnegative numbers and the sample mode was set to observe salinity. Four compiled data sets were made from all spectra from each experiment; freshwater permeate mixes (all seasons), estuarine permeate mixes (all seasons), freshwater UDOM mixes (all seasons), and estuarine UDOM mixes (all seasons). All models were checked for convergence using random initialization. Seven components were chosen for the initial model (e.g., Table 1). Random subset cross-validation was used for all model runs and selected components were sequentially lowered until core consistency for each component was at least 50%. Convergence was tested using four runs (with random initialization) to confirm identical fits.
 There were no systematic shifts in emission wavelengths for the standard peaks (identified in Table 1) with permeate mixing. This implies that there is little blue or red LMW DOM shifting from the sampled estuaries and their associated coastal ocean environments. Due to the number of identified peaks (7) and treatments (5 sites, permeate and UDOM), only a small subset of peak parameters are presented here (Figure 3). There was a general attenuation in peak intensity for all compiled freshwater permeate peaks (Figure 3a), however, the degree was minimal for the DB Mar03 sampling and for the B peak (Figures 3a and 3c). Mixes with estuarine permeate added exhibited similar behavior; showing intensity decrease with salinity increase (e.g., Figure 3d).
 A modeled salinity mix for each peak was created by mathematically mixing the freshwater permeate fluorescence intensity appropriately with the estuarine permeate fluorescence intensity. For instance a freshwater permeate 4 salinity contained 8 mL freshwater permeate, 4 mL estuarine permeate plus 1 mL freshwater permeate (see Figure 2). One mL of permeate was added to permeate mixes as we assume 1 mL UDOM contains similar amounts of permeate moieties (which are not selectively retained by TFF). See Figure 2, bottom right. Thus the total volume would be 11 mL. The freshwater permeate intensity was multiplied by and the estuarine permeate intensity was multiplied by , then added. A “general mixing model” was created by using the average 0 and 16 salinity values and slope calculated above. The modeled data were then compared to the actual fluorescence intensity from 0 to 16 salinity. The actual freshwater permeate peak intensities from Table 1 were statistically the same (P < 0.01) when compared to modeled data for all seasonal samplings besides DB Mar03 for all peaks except B. The B peak had greater intensity for all samplings (P > 0.05). All freshwater permeate peaks from the DB Mar03 were different from the modeled data (P > 0.05). This effect was more pronounced with estuarine permeate mixes. For estuarine permeates mixed across the salinity gradient, all but the C peak for CB Sep03 and the A peak for CB Aug04 were significantly different than the modeled data (P > 0.05). The actual data was non-conservative, greater than simple mixing would predict (Figure 4). This indicates that LMW DOM residing in- or created in mid-estuarine regions non-conservatively increases in fluorescence when mixing toward the freshwater end-member. Despite the non-conservative mixing behavior, linear correlations with a goodness of fit (r2) greater than 0.75 were obtained between salinity and fluorescence intensities for the C, M, and C-M intermediate peaks with freshwater permeates (except for SF May03). The C, M and C-M intermediate peaks also showed reasonable correlation with estuarine permeates. In this instance, the DB Mar03 sampling had the goodness of fit less than 0.75. The B and T peaks were most poorly correlated.
 Fluorescent ratios often showed considerable changes related to salinity for both freshwater and estuarine permeate mixes. Under the assumption that the freshwater end-member contains a higher proportion of aromaticity, the McKnight fluorescence index should be lower at lower salinities [McKnight et al., 2001]. Interestingly, the permeates demonstrated a higher index at low salinities and decreased toward the ocean end-member (Figure 5a). This held true for both freshwater and estuarine LMW DOM. Also opposite of expected results is the gradual decrease in 265F375/400 with increasing salinity; true for both freshwater and estuarine permeates (Figure 5b). Typically, values for this ratio below 1.0 are indicative of terrestrial-derived fluorescent organic matter and range from ∼0.50 to 1.10 in natural samples [Del Castillo et al., 2001]. DB Mar03 permeates showed the least variation with salinity (Figure 5b). Consistent with riverine input, most 305F430/540 ratios decreased from lower salinities to higher salinities (Figure 5c). All ratios were higher than 2.0, up to ∼8.0 for SF Bay May03 permeates. These ratios normally range from 2 to 2.50 in natural samples [Del Castillo et al., 2001]. Finally, CDOM fluorescence (ex, 355; em, 450) showed a linear decrease for late summer/early fall CB samples (Sep03 and Aug04), but increased slightly toward mid-salinity then decreased toward the ocean end-member in the wet season CB (Mar 04) and faster flushing DB and SF (Figure 5d).
3.2. UDOM Emission Maxima, Peak Intensity Maxima, and Fluorescence Ratios
 For freshwater and estuarine UDOM in all bays there was a decrease in λEMmax (or blue-shift) with increasing salinity for the C and A peaks (Figures 6a–6d). The exception was DB with freshwater UDOM, which showed very little change with salinity (Figure 6a). The effect appeared more pronounced with estuarine UDOM (Figures 6b and 6d). Permeate mixes did not show any systematic emission wavelength maxima variation with salinity (section 3.1), but the impact of spectra subtraction might have been partially responsible for the observed shifts. Even with re-processing (using the UDOM + permeate UV-vis spectra), there was extremely high fluorescence at lower salinities in the permeate samples. Peak shapes were visibly different before and after permeate subtraction. However, the blue shift observation was also made with UDOM plus permeate spectra (before permeate subtraction), therefore, we believe the observed shifts reflect salinity effects on freshwater and estuarine UDOM.
 For the most part, UDOM maximum peak intensities were relatively low at the freshwater end-member (Figure 3). This may be partially due to the considerable fluorescence intensity observed for permeates in the low salinity ranges. Permeate fluorescence was extremely high in this range and even with adjusted inner-filter corrections (using the UDOM plus permeate UV-vis spectra), mathematical subtraction may have introduced some bias in this region. Focusing on the mid-to oceanic range of the salinity gradient where data showed less extreme variation, two general observations can be made. For shorter residence time estuaries (SF and Delaware Bays) and the “wet season” sampling in CB (Mar04), there was very little variation in peak intensity for freshwater UDOM (e.g., Figure 3a). For CB in summer/fall conditions, there was a slight decrease in humic-type peak signals (C, A, M, and C-M intermediate) and the N peak from mid salinities toward the ocean end-member (Figures 3a–3d). The protein-like B and T peaks showed more variability with estuarine UDOM than with freshwater UDOM, but showed no “general” trend. To evaluate the general trend toward the ocean end-member, a linear regression analysis was applied to peak intensities from 16 to 36 salinity (Table 3). Goodness of fit statistics (r2) were generally poor, however, for all above 0.50, calculated slopes were negative, indicating a gradual decrease in fluorescence intensity going ocean-ward. If CDOM fluorescence were conservative, one would expect a “zero” slope.
Table 3. Fluorescence Peak Trends Toward Ocean End-Member From 16 to 36 Salinitya
Intermediate C-M Slope
Peaks represent those reported in Table 1. Italics indicate significant correlations.
CB Aug04 F UDOM
CB Mar04 F UDOM
CB Sep03 F UDOM
DB Mar03 F UDOM
SF May03 F UDOM
CB Aug04 E UDOM
CB Mar04 E UDOM
CB Sep03 E UDOM
DB Mar03 E UDOM
SF May03 E UDOM
 Peak ratios for freshwater and estuarine UDOM showed little variation except in the case of 305F430/540. This ratio increased from the freshwater end-member to the ocean end-member for SF May03, CB Sep03, and CB Mar04 with freshwater UDOM (Figure 7a) and in all samplings except DB Mar03 with estuarine UDOM (Figure 7b). This indicates that freshwater and estuarine UDOM may appear more “riverine” in nature as they travel down-estuary and into the coastal ocean [Del Castillo et al., 2001]. The 265F375/400 ratio exhibited a hyper-variable region at low salinities and gradually increased toward the ocean end-member in all but the DB Mar03 sampling for both freshwater and estuarine UDOM (Figures 7c and 7d). The ratio was below 1.0, indicative of riverine signature, but showed transition toward coastal character as salinity increased. The McKnight fluorescence index did not show significant variation over the salinity gradient. However, the values were 1.1 to 1.2 for freshwater UDOM and 1.05 to 1.15 for estuarine UDOM, significantly lower than those observed with corresponding permeates (see Figure 5a) which is consistent with results previously reported [Belzile and Guo, 2006] wherein permeate ratios were significantly higher than UDOM ratios.
3.3. LMW DOM PCA Analysis
 The reduced data matrix, containing the emission, excitation and maximum intensity values for each of the 7 peaks (Table 1), along with the 370F450/500, 265F375/400, and the 305F430/540 ratios, were subjected to PCA analysis. The intent was to reduce any co-varying trends into new but fewer vectors. Each sampling event was identified as a class within the PCA model. PC1, which accounted for 42% of the variability showed a marked and linear decrease with salinity for freshwater permeates (Figure 8a). A conservative mixing model, calculated according to section 2, was used to model “ideal” mixing for permeates. The slope and mid-salinity scores were used to calculate score “ideal” mixing. The same was evident for estuarine permeate PC1 (Figure 8c) which accounted for 34% of the variability. PCA scores demonstrated similar non-conservative increase with decreasing salinity observed for humic-type peaks below 12 salinity (Figures 4 and 8c). PC2, which accounted for 14% of the variability in freshwater permeate mixes and 17% in estuarine permeate mixes, showed no linear variation with salinity, however, a generally bell-shaped curve was observed, with the freshwater and ocean end-members either higher or lower than mid-estuarine member.
 A PC1 vs PC2 bi-plot showed some site separation based on score clusters (Figures 9a and 9b). With freshwater permeates, DB Mar03 clustered “away” from other sites/seasons, but others were not well separated (Figure 9a). Sample scores fell out along peak intensities and fluorescence ratio loadings at lower salinities and along peak emission loading at higher salinities for CB Mar04, CB Sep03, and SF May03. DB Mar03 and to some extent CB Aug04 scores plotted along the positive PC2/negative PC1 to negative PC2/negative PC1 axis, driven primarily by peak emission loadings. As shown in the mathematical mixing model, estuarine permeate scores showed considerable variability at extremely low salinities (Figure 9b). Zero salinity scores plotted outside the 95% confidence limit for the PCA model with SF May03, DB Mar03, and CB Sep03 samples. Sample site/season scores did not generally separate for the estuarine permeate PCA model.
3.4. UDOM PCA Analysis
 Freshwater UDOM PC1 showed variation with salinity and accounted for 42% of the peak and ratio data matrix variability (Figure 8c). PC1 either increased or decreased from the freshwater to marine end-member (SF May04 or CB Mar04 and CB Aug04) or had a bell-shaped curve, changing between low and high salinities. PC1 for estuarine UDOM, accounting for 32% of the sample matrix variability, showed a marked increase in low salinities with either gradual increase toward the ocean end-member (DB Mar03, CB Mar04, and to some extent CB Aug04) or slight decrease toward the ocean end-member (CBSep03, SF May04) (Figure 8d). PC2 for both freshwater and estuarine UDOM showed no model-able trend when plotted against salinity. Loadings fell out into the labeled quadrants in Figure 9.
 For the freshwater UDOM PCA models, PC1 scores plotted against PC2 scores showed linear distributions for SF May03, CB Mar04, and CB Sep03 with low salinities in negative PC1 and positive PC2 data space. Higher salinities were represented in the negative PC2 and positive PC1 direction (Figure 9c). DB Mar03 and CB Aug04 where more closely clustered, although when plotted individually, showed a similar (but opposite) salinity trend along the peak emission to fluorescence ratio loadings axis. Peak emission maxima loadings were largely negative with respect to PC1, but positive with respect to PC2, while peak intensities loaded positively with respect to both PC1 and PC2. Fluorescence ratios were positively loaded with respect to PC1, but negatively with respect to PC2 (Figure 9c). This indicates that peak emission maxima were most important in defining low salinity fluorescence characteristics while peak ratios were more important in defining higher salinity spectra (DB Mar03 was the exception with peak ratios more important in defining low salinities). Most peak intensities loadings defined where estuaries plotted relative to one another. Data were mean-centered and scaled, so raw fluorescence intensity should not have defined scoring along this variable. The estuarine UDOM combined PCA model was similar in general to the freshwater UDOM PCA model. However, there was tighter score clustering with several of the lower salinity scores outside the 95% confidence level for the model (Figure 9d). As with PC1 values for individual site models, the zero salinity treatments produced scores which plotted near or outside the 95% confidence interval for all sites (see scores in Figure 9d).
3.5. Permeate PARAFAC Analysis
 Five components were validated for freshwater permeate and four components were validated for estuarine permeate mixes. PARAFAC loadings were plotted to visualize which components scores which corresponded to known fluorescence peaks (Table 1). Excitation and emission maxima were used to define peak regions (Table 4). With freshwater permeates added, scores corresponding to C and A peaks showed a generally linear trend, decreasing from the freshwater to the marine end-member for slow flushing periods (Sep03 and Aug04 in CB). For the faster flushing estuary and CB Mar04, there was a gradual increase in C and A peak scores toward mid-estuary, then gradual decrease toward the ocean end-member (Figure 10a). For scores with loadings in the M peak region, there was a consistent decline with salinity. Scores with loadings in the B peak region showed no decreasing trend (Figure 10b), while the T peak (also potentially representing protein material) behaved more like humic-like material (similar to Figure 10a). C-M peak scores and N peak scores also decreased with salinity similar to humic-type peaks.
Table 4. CDOM Components Found in PARAFAC Loadings EEMs
 With estuarine permeates added, several scores corresponding to C and A peaks showed a generally linear trend, decreasing from the freshwater to the marine end-member for slow flushing periods (Sep03 and Aug04 in CB). These were consistent with additive mixing models. For the faster flushing estuary and CB Mar04, there was a gradual increase in C and A peak scores toward mid-estuary, then gradual decrease toward the ocean end-member (Figure 10c). As with freshwater permeates, peak scores other than B showed a decreasing trend from low-to higher salinities (Figure 10d).
3.6. UDOM PARAFAC Analysis
 Four components were validated for both freshwater and estuarine UDOM mixes (Table 4). PARAFAC analysis with freshwater UDOM EEMs stacked by salinity provided scores which showed substantial variability in the low salinity ranges. The C, Int and A peaks showed an increasing trend toward mid-salinities, then decreasing toward the ocean end-member (Figure 11a), although the trend was not universal. The T and N peak loadings overlapped and were low at lower salinities (except DB) and increased toward the ocean end-member (Figure 11c). The B peak showed variability at low salinities (<8), then perhaps very slight decrease toward the ocean end-member (Figure 11e).
 With estuarine UDOM, PARAFAC loadings also peaked together within the C, Int and A peak regions. As with freshwater UDOM, the scores in this region varied at low salinity with general increase or decrease toward mid-salinities (Figure 11b). Scores for all loadings in the N peak area were very low at the 0 salinity extreme and increased toward the ocean end-member, aside from the CB Sep03 sample which spiked at 4 salinity, decreased, then gradually decreased ocean-ward (Figure 11d). Estuarine UDOM PARAFAC scores overlapping both the N and T peak also increased from freshwater (0 salinity) to the ocean end-member (Figure 11f).
 In addition to loadings falling within “known” peak regions, there were components which corresponded to recent literature PARAFAC components. Components were observed corresponding to C3/P3, C4, C7, and C9 [Murphy et al., 2008]. Because freshwater and estuarine UDOM were used within the experimental design, these components are apparently represented in UDOM from all regions within the terrestrially influenced watersheds sampled in this study, even though spectra may be implied to be of marine origin (i.e., the M peak).
4.1. Peak Emission Maxima, Peak Intensities, and Peak Ratios
 Initially, we hypothesized that the residence time would likely be a major driving factor in determining UDOM and permeate spectral characteristics. While peak intensities varied between permeates and UDOM from different seasonal site samplings, there were no consistent trends identified among estuaries with short (e.g., Delaware and San Francisco Bay at ∼1–2 month) [Church, 1986; Smith, 1987], to longer residence times, (e.g., Chesapeake Bay at ∼1 year) [Boynton et al., 1995]. The DB Mar03 peak intensities and peak ratios for both permeates and UDOMs showed the least variation with salinity. Additionally, this sample site showed the greatest “clustering” of scores for PCA analysis (Figure 9). The only notable observation in which faster flushing estuaries (DB and SF) and wet season CB showed similar patterns was for permeate 355F450 intensity (Figure 5d).
 As with a previous study using TFF permeates [Belzile and Guo, 2006], we noted a higher relative wavelength specific quantum yield (calculated as fluorescence intensity divided by emission wavelength absorbance) for permeates relative to UDOM. Permeate absorbance spectra were only 10–20% of UDOM spectra in lower wavelengths (<320 nm). Their inner-filter corrected fluorescence at low excitation wavelengths was often higher than UDOM fluorescence at comparable wavelengths. Thus permeate wavelength-specific quantum yields were higher relative to UDOM samples. Originally, permeate spectra were directly subtracted from UDOM spectra in an attempt to quantify UDOM-specific fluorescence changes during estuarine mixing. However, permeate fluorescence intensities were so high in some cases the resultant subtracted spectra were had negative intensities. This necessitated recalculating “permeate+UDOM” inner filter corrections for permeates with added UDOM (see section 2.4). A mixing model calculation for permeates showed that in some cases for freshwater permeates and in most cases for estuarine permeates, there was disproportionally high fluorescence at lower salinities (<12) (Figure 4). Peak intensities were modeled using linear regression over the entire salinity range (0:36) and from 16 to 36 salinity. The goodness of fit statistic (r2) increased above 0.75 for the C (4 samples in 0–36 salinity transect to 5 samples in the 16–36 salinity transect), M (2 samples to 4 samples), T (1 sample to 4 samples), C-M Intermediate (5 samples to 6 samples) and N (1 sample to 4 samples) peaks as well as 355F450 ratio (2 samples to 5 samples) when estuarine permeate was modeled. This observation argues that some component of estuarine permeate is either created in situ or perhaps altered at mid-salinities such that its optical properties are no longer conservative when mixing toward the freshwater end-member. Again, there was no evidence of any flocculation in any samples used in this study.
 Humic and fulvic materials have been the focus of a number of studies relating to estuarine mixing and are found in spectra and models generated during these experiments. Flocculation of these chemical classes was initially identified as a possible removal means in estuarine systems [Sholkovitz, 1976]. However, subsequent experiments showed no net removal of DOC even though humic substances may be lost through an estuarine mixing transect [Fox, 1983]. It is unclear exactly what factors within an estuary lead to alteration of humic substances. Mobed et al. , similar to the studies of Willey , used International Humic Substances Society (IHSS) standards as model compounds, and KCl to show little fluorescence property changes due solely to ionic strength. However, at the molecular level, ionic strength has been shown to impact the titratability [Ephraim et al., 1995] and humic material electrophoritic mobility [Duval et al., 2005]. At present it is difficult to ascribe the changes in humic-type fluorescent properties observed in this study to ionic strength or presence of divalent ions in ocean end-member permeates. However, because the experiments described here most closely resemble those of Willey (natural organic matter with contemporaneously collected seawater end-member added) [Willey, 1984], we conclude that changes in peak emission wavelength and intensities observed here using the “natural” seawater milieu are more a factor of divalent seawater ions than a simple function of ionic strength. A study using artificial seawater with the addition of selected individual ions might prove a worthwhile follow-on study to investigate which ions are most responsible for shifts in fluorescence properties of our UDOM isolates.
 In terms of changes in the proteinaceous peak fluorescence properties, less estuary-relevant natural organic matter work has been published. It has long been known that intracellular and extracellular changes in pH and ion strength alter the higher-order conformation of enzymes and thus change their activity [cf. Sillen et al., 2003] as an example of a marine organismal enzyme impacted by ionic strength). In physiological terms, ionic strength within an estuarine transect may go from hypotonic (∼0–9 PSU) through isotonic (∼9–11 PSU) to hypertonic (∼ > 11 PSU). The exact chemical makeup and structure of the proteinaceous component of bulk DOM is unknown, but most likely contains amino acids and partially degraded proteins. In addition to releases from traditional means (cell death and lysis, sloppy feeding, viral rupture, etc.), aquatic bacteria may release considerable amounts of extracellular hydrolytic enzymes into the water column in order to degrade particle-associated organic material [Keith and Arnosti, 2001]. The latter is an interesting hypothesis for changes in fluorescence properties in the B and T peak regions as the conformational structure of extracellular enzymes might be impacted by ionic strength in an ever-changing environment such as an estuary. Interestingly, the B and T peak fluorescence behaved differently with the B peak being much less variable with salinity in both the permeate and UDOM fractions (Figures 3c, 10d, and 11d). B peak intensities did not fit a linear mixing model for either freshwater or estuarine permeate fractions. This suggests that B peak components are likely low in molecular weight thus distributed across the mixing gradient and perhaps of more universal origin and distribution. They may also represent smaller peptide moieties less susceptible to 3° conformational changes under different ionic strength regimes. While the B and T peaks are both considered proteinaceous [Blough and Del Vecchio, 2002], it appears from this study that they have different behaviors when separated into different size fractions.
4.2. PCA Analyses
 The combined PCA analysis provides interesting results that suggest changes in fluorescence peak properties and ratios may be both watershed and season dependent although no single peak or ratio property was a driving force in placing scores along the sample site direction in both directions (Figures 9c and 9d). Peak intensities were positively loaded in both PC1 and PC2 for the freshwater and estuarine UDOM models, thus differences in UDOM properties across sample sites appear to be largely driven by relative fluorescence intensity. It is unknown if this more a function of relative fluorescence intensity between peaks or between sample sites. Interestingly, the freshwater and estuarine UDOM fluorescence peak and ratio properties appear somewhat universal and fall into the broad literature peak classifications (i.e., humic, proteinaceous, etc., Table 1). However, spanning of components across the salinity gradient suggested at least two major factors dictating fluorescence behavior during estuarine mixing: (1) source, season or residence time (partially loaded on peak intensities, and (2) salinity and associated ionic mixing (loaded by peak emission maxima at low salinities and by fluorescence ratios at higher salinities).
 The first factor should be somewhat expected as EEMs analysis have been used for some time to differentiate various sources of natural organic matter [e.g., Chen et al., 2003; Komada et al., 2002; Yan et al., 2000]. However, mixing-induced variability in peak intensity, emission maximum and peak ratios suggest that the use of certain peaks or peak ratios within an EEM might not allow definitively linking fluorescent properties to a particular source of organic matter (i.e., terrestrial, freshwater, marine, etc.). The observation that emission wavelength, fluorescence intensity, and the fluorescent ratios loaded in different quadrants in bi-plots indicates considerable variability in these properties as a particular source of organic matter mixed through an estuarine gradient. Results from this study suggest that in all but the case of DB, fluorescent ratios showed considerable variability when end-member UDOM was exposed to simulated estuarine mixing. Interestingly, freshwater and estuarine UDOM fluorescence ratios showed the greatest variability at lower salinities (see Figure 5), thus they may be more appropriate for tracing UDOM from mid-salinities ocean-ward. Even though data were mean centered before PCA analysis, moderate variations in fluorescence intensities likely resulted in estuary separation on scores plots as peak intensity loaded along the bottom left to top right axes of the graphs (Figure 9).
4.3. PARAFAC Analyses
 EEMs represent a considerable amount of data and determining the differences between samples let alone what chemical moieties to attribute them to have been problematic. Defining peak types and their properties provides comparative data across sites, but may “miss” spectral information which is yet to be described or understood. PCA in this study relies on the compiled peak and peak ratio data obtained across the salinity gradient to reduce variables and allow a more straightforward means of describing “collective” spectral changes across an estuarine salinity gradient. Other data reduction techniques such as regional integration [Chen et al., 2003] and partial least squares discrimination [Hall et al., 2005] have been applied to EEMs as a means of describing content. PARAFAC analysis has emerged as a robust means to pull out factors which vary across ‘stacks’ of spectra [Hall et al., 2005; Stedmon et al., 2003]. In our case, EEMs were stacked by salinity to observe changes over the gradient. One PARAFAC advantage is its ability to retain spectral information from the original data set while potentially identifying features within the data not easily observed in the raw spectra. PARAFAC analysis was visualized in each sampling event by producing a multidimensional matrix of salinity and spectral response (both excitation and emission wavelength). We considered that variability in spectral regions might show mixing-induced trends not observed in the reduced data set used for PCA analysis.
 PARAFAC analysis identified two components which fell into the C and A peak range for permeates and UDOM samples. For permeates there were two general trends for humic material fluorescence across the salinity gradient: one in which there was a linear or exponential (some estuarine permeates) decrease in signal moving from the freshwater to the marine end-member, the other in which signal gradually increased to mid-salinities, then decreased toward the ocean end-member (Figure 10). This trend was similar with freshwater UDOM, but far more variable with estuarine UDOM (Figure 11). This indicates that there are may be two “humic-like” moeities within the LMW DOM fraction that are representative in all environments and one which is more prevalent in freshwater or mid-estuarine environments, which in turn may fluoresce to a degree greater than simple mixing would suggest. The higher fluorescence for some components at mid-salinities may also indicate a fraction of “humic-like” material which has higher relative fluorescence intensity at mid-salinities. Humic material has been shown to decrease binding efficiency for divalent metals between low and mid-salinities which might explain differential fluorescence [Lores and Pennock, 1998]. For UDOM, components corresponding to C and A peaks showed extreme variability at low salinities with more gradual trending toward the ocean end-member (usually a decrease). The A peak is perhaps believed to be terrestrial in origin [Murphy et al., 2008] and may be very poorly represented in marine water samples. Because UDOM was from freshwater and mid estuaries in this study, components representing the A peak should be reasonably well represented. The observation, however, was that this component (and that overlapping the C peak region) did not produce conservative fluorescence signal over the salinity gradient perhaps indicating differential divalent metal binding or conformational changes, for instance “coiling” causing apparent molecular size decrease in the “humic-type” material during estuarine mixing [Conte and Piccolo, 1999; De Haan et al., 1987; Myneni et al., 1999].
 The “marine-humic” M, and C-M Intermediate peaks behaved similarly to the A and C peaks in regard to salinity except there was no mid-salinity maximum noted. For freshwater and estuarine permeates, there was a linear or exponential-like decrease in intensity going from the freshwater to ocean end-member salinity, respectively. Unlike partial components representing the C and A “humic-type” peaks, these permeate peaks behaved conservatively with salinity and, other than the variable region below 8 salinity, their abundance appeared to be dilution-dictated. In the case of freshwater and estuarine UDOM, these peaks all increased substantially from very low signal at the freshwater end-member to high signal toward the ocean end-member. This would indicate that as with A and C moieties, there is possibly quenching of UDOM fluorescence at lower salinities or conformational changes driving fluorescence intensity behavior.
 The N peak has been described as having a non-humic character [Coble et al., 1998] and in this study its PARAFAC behavior was similar to that of the T peak: decreasing with increasing salinity for both freshwater and estuarine permeates and increasing with salinity for freshwater and estuarine UDOM. The T peak is presumed to be proteinaceous and would thus be susceptible to ionic strength conformational changes which might explain intensity variations over the salinity gradient (Figure 11). In a previous study, PARAFAC component corresponding to the T peak showed little correlation with salinity toward the freshwater end-member (below salinity ∼20) followed by linear decrease [Yamashita et al., 2008]. While these are natural samples with associated concentration differences, one could hypothesize that at least some of the variability may have been due to physicochemical-mediated alterations in fluorescence properties as well as in situ synthesis and removal processes. A recent report suggests the T peak consists of two entities one a low molecular weight component (in this case above 1 kD) and a high molecular weight component (>5 × 104 D) [Maie et al., 2007].
 The component corresponding to the B peak showed little variation with salinity especially with the permeate addition samples (Figures 10b and 10d). There was variability in low salinities with freshwater and estuarine UDOM with some small systematic decrease going ocean-ward along the salinity gradient (Figure 11e). In all cases, regression analysis had goodness of fit well below 0.75 and P values greater than 0.05 (indicating no significant deviation from 0 slope). These results fit well with the study by Yamashita et al.  in which the component abundance corresponding to the B peak showed considerable scatter when plotted against salinity. The B component, while presumed to be proteinaceous, showed little alteration from changes in ionic strength and was the only peak relatively unaffected by salinity either in the LMW or HMW pool. The same was true for the actual peak intensities from the processed EEMs spectra (Figure 3).
5. Conclusions and Carbon Cycle Implications
 The net movement of terrestrial-derived organic matter into and through estuaries is toward the ocean end-member. CDOM has traditionally been thought to undergo conservative mixing in estuaries as its quantity tends to correlate with salinity (a conservative tracer). However, this view has recently come under challenge as ancillary data and more definitive optical measurements are applied to estuarine and coastal waters [e.g., Rochelle-Newall and Fisher, 2002; Yamashita and Tanoue, 2004]. Kowalczuk et al.  reported changes in peak region CDOM fluorescence from black water river to ocean transects in North Carolina (USA). In their work, they noted a decrease in A, C, and M peak intensity concomitant with a relative increase in protein peak fluorescence as riverine water mixed toward the ocean end-member. Interestingly, our results on freshwater CDOM mixing alone showed changes in intensity for the A, C M, C-M intermediate, T and N peaks and a relative unchanged B peak abundance form either the freshwater end- or estuarine mid-member to the ocean. Although our sampling locations are not black water estuarine systems, if results presented here represent trends in CDOM optical properties during estuarine mixing, it may be necessary to consider the alteration of fluorescence due to mixing alone before using CDOM fluorescence as a conservative index in estuarine mixing. CDOM, in the form of LMW components also showed non-conservative mixing behavior at the proximal end of estuaries, with either processed freshwater or in situ-created mid-estuarine DOM. This implies, as studies showing inconsistencies between CDOM and DOM concentration do, that CDOM dynamics in estuaries are likely driven by both inputs and degradation, which may “mask” tracking an individual CDOM source by fluorescence. We hypothesize that changes in CDOM end-member optical properties during mixing through estuarine waters coupled with in situ production of CDOM, with overlapping and perhaps complementary optical properties, might collectively “appear” as conservative mixing when related to a tracer such as salinity. This observation has considerable implications when using a conservative mixing model to calculate carbon export. If fluorescence properties change as a result of ionic strength, divalent ion associations or other as yet described processes, calculations which fail to take into effect systematic (or more commonly non-systematic) changes may be error prone, particularly if scaled to regional or global biogeochemical cycles.
 This research was performed while Bethany Barham held a National Research Council Research Associateship Award at the Naval Research Laboratory. The Office of Naval Research supported this work. PCA and PARAFAC modeling was greatly assisted by Kevin J. Johnson. The authors also wish to thank the captain, crew, and support personnel of the R/V Cape Henlopen and R/V Pt. Sur and N. V. Blough for shared ship-time.