2.1. Study Sites
 The YRB covers 854,700 km2 of northwest Canada and Alaska, and primarily composes remote wilderness (Figure 1). Mean annual air temperature throughout the YRB ranges from −6 to −1°C (1971 to 2000, from Alaska Climate Research Center (ACRC), http://climate.gi.alaska.edu). Precipitation in interior Alaska is spatially variable, averaging 235–380 mm yr−1 in the continental climate of interior Alaska, 400–500 mm yr−1 in the Yukon Delta area (1971–2000, ACRC), and greater than 1000 mm yr−1 in the headwaters in the St Elias Mountains [Striegl et al., 2007]. Much of the YRB is underlain by permafrost, which is actively warming and thawing throughout the basin [Osterkamp and Romanovsky, 1999; Osterkamp, 2007]. Approximately 30% of the YRB is covered by low-lying wetlands [Brabets et al., 2000]. Watersheds of the YRB vary with respect to hydrology, parent material and source water contributions to stream flow. Blackwater streams (e.g., Porcupine River) typically drainwatersheds underlain by permafrost and have extensive wetland coverage. In clearwater streams (e.g., Clearwater Creek), groundwater discharge dominates stream flow. In glacially fed streams (e.g., Tanana River), flow originates as meltwater from alpine glaciers and snowfields. Despite being glacially fed, groundwater discharge to the Tanana River accounts for more than 25% of annual flow [Walvoord and Striegl, 2007].
 Parent material underlying watersheds of the YRB is spatially variable. Nearly 75% of the Yukon –Tanana upland area is underlain by rocky colluvium, which is composed primarily of quartz-mica schists beneath a relatively thin loess mantle [Ping et al., 2006]. Thick loess deposits (>20–30 m thick) are also present, accumulating in unglaciated regions of the YRB during the late-Pleistocene and early Holocene [Muhs et al., 2003]. Alluvial deposits are common along floodplains and terraces of large glacial rivers (e.g., Tanana River), and are typically composed of a loamy cap above sandy and gravelly substrate [Ping et al., 2006]. Glacial deposits (e.g., outwash, till, moraine) are also common, particularly along the northern foothills of the Alaska Range [Ping et al., 2006].
 Permafrost extent in the YRB is largely discontinuous (50–90% of area) with localized regions of sporadic (10–50%) and isolated permafrost (<10%) in lowlying areas such as Tanana and Koyukuk Flats [Jorgenson et al., 2008]. Continuous permafrost (>90%) also exists in northern parts of the basin, such as in the Porcupine River basin. Thermokarst features have been documented throughout the YRB, with collapse-scar fens, bogs, and thermokarst lakes identified as the most common modes of thaw [Jorgenson et al., 2007]. Wildfire frequency and severity have increased in recent decades in Alaska's boreal region [Turetsky et al., 2011] and may exacerbate rates of permafrost thaw relative to climate warming alone [O'Donnell et al., 2011a].
2.2. Study Overview
 We describe the chemical composition of DOM during winter flow from 60 streams draining subcatchments of YRB and points along the main stem of the YRB (Figure 1). We also describe samples from two representative groundwater sources (Fox Spring, Tape Well, both located in or near Fairbanks). For this study, we define winter flow as the period spanning November 1 through April 30, following the convention used by Striegl et al. . For samples collected and processed between 2001 and 2010, data were available through the U.S. Geological Service National Water Information System (NWIS, Web interface: Water Data for the Nation; http://waterdata.usgs.gov/nwis). In March 2011, we collected winter streamflow samples from 25 additional sites to better characterize spatial variability in river DOM character throughout the YRB. To examine the unique chemical composition of DOM during winter, we compared and contrasted DOM properties during winter versus summer-autumn (July 1–October 31) for several representative sites (Porcupine, Salcha, and Tanana Rivers; Hess Creek; Yukon River at Eagle, Stevens Village and Pilot Station). Summer-autumn data were also obtained through the NWIS website. Under-ice samples were obtained using an ice-auger powered either by hand or by a Stihl 8D powerhead. Samples were characterized for DOC (n = 173 samples) and dissolved organic nitrogen (DON) concentrations (n = 57 samples), UV-visible absorbance (n = 173 samples at 254 nm; n = 47 samples for all other wavelengths), fluorescence (n = 72 samples), and XAD8/XAD4 fractionation (n = 104 samples). We also developed a simple mixing model to evaluate river DOM composition in response to projected changes in permafrost thaw and groundwater discharge.
2.3. DOC Concentration and DOM Chemical Fractionation
 DOC concentrations were determined using an O.I. Analytical Model 700 TOC Analyzer via the platinum catalyzed persulfate wet oxidation method [Aiken et al., 1992]. DON was calculated as the difference between total dissolved nitrogen (TDN) and inorganic nitrogen species (nitrate (NO3−) + ammonium (NH4+)). TDN was determined using a Skalar Model Formacs Total Nitrogen Analyzer and NH4+was determined using Dionex Model DX-300 and DX-500 Ion Chromatographs with conductivity detectors following the methods ofSmith et al. . NO3− concentrations were either determined using ion chromatography or using colorimetric methods following Antweiler et al. .
 Stream water samples were chromatographically separated into different fractions: hydrophobic acids (HPOA), hydrophobic neutrals (HPON), hydrophilic organic matter (HPI), and transphilic acids (TPIA) using Amberlite XAD-8 and XAD-4 resins [Aiken et al., 1992]. The resins preferentially sorb different classes of organic acids based on aqueous solubility of the solute, chemical composition of the resin, resin surface area, and resin pore size. The amount of organic matter within each fraction, expressed as a percentage of the total DOC concentration, was calculated using the DOC concentration and the sample mass of each fraction. UV-Vis absorbance was run on each of the major DOC fractions. The standard deviation for the mass percentages of the fractionation was ±2%.
2.4. Optical Properties of Chromophoric DOM
 For our analyses, we used several approaches for reporting and analyzing the optical properties of chromophoric DOM (CDOM). UV-Vis absorbance was measured at room temperature using a quartz cell with a path length of 1 cm on an Agilent Model 8453 photo-diode array spectrophotometer. Based on convention and ease of comparison with literature data, measured decadal absorbance values (A) were converted to absorption coefficients which can be expressed in both decadal units (a), as is common in the freshwater and wastewater literature, and Naperian units (α), which is the convention among marine chemists. The following equation shows the conversion to Naperian absorption coefficient.
where A(λ) is the decadal absorbance at a specified wavelength (λ) and l is the cell path length in meters [Green and Blough, 1994]. CDOM absorption coefficients in Naperian units are reported at wavelengths of 254, 350, and 440 nm. Absorption coefficients at these wavelengths have been shown to correlate strongly with riverine DOC concentration, particularly where allochthonous organic matter inputs dominate [Spencer et al., 2012]. The absorption coefficient at 350 nm (α350), in particular, has been shown to be strongly correlated with lignin phenol concentrations [Spencer et al., 2008], which are derived from vascular plant material.
 Spectral slope (S) was calculated by fitting an exponential equation to the absorption spectra between 275 and 295, 290–350, and 350–400 nm using
where ag(λ) is the absorption coefficient of CDOM at a specified wavelength, λref is a reference wavelength, and S is the slope fitting parameter [Helms et al., 2008; Spencer et al., 2008]. The spectral slope of the 275–295 nm region (S275–295) has been shown to be negatively correlated with the molecular weight of DOM [Helms et al., 2008]. Prior studies have shown the spectral slope between 275 and 295 nm (S275–295) to be sensitive to changes in DOM source (e.g., riverine versus estuarine versus open ocean and composition) [Spencer et al., 2007]. Slope ratio (SR; 275–295-nm slope:350–400-nm slope) was also calculated, and has also been shown to be strongly correlated with DOM molecular weight [Helms et al., 2008].
 Specific UV absorbance (SUVA254) was determined by dividing the decadal UV-Vis absorption coefficient per meter atλ = 254 nm by DOC concentration. SUVA254, which is typically used as an index of DOC aromaticity, provides an “average” absorptivity at λ = 254 nm for the DOC [Weishaar et al., 2003]. SUVA254 is reported in units of L mgC−1 m−1. Weishaar et al.  also showed that UV absorbance values can be influenced by the presence of iron. To account for potential effects of iron on UV absorbance, we measured total iron concentrations on a subset of winter flow samples. We applied correction factors based on experimental work by B. Poulin and G. Aiken (unpublished data, 2012), who observed a significant positive correlation between α254 and the concentration of Fe3+, reflected by the equation A254-corrected = A254-measured – 0.0687*[Fe3+] (R2 = 0.98; P < 0.0001). Using this relationship, we corrected UV absorbance values at 254 nm for samples where we measured total iron concentration. pH was also measured in the field on most samples, averaging 7.25 ± 0.60 (standard deviation, n = 126, range = 5.60–8.83).
 Fluorescence excitation-emission matrices (EEMs) were measured on select samples and a subset of major chemical fractions (HPOA, TPIA, HPI) at room temperature using a Jobin-Yvon Horiba Fluoromax-3 fluorometer. Samples were diluted to minimize inner filter effects with deionized water, when necessary, to a UV absorbance atλ= 254 nm of 0.2 absorbance units (1 cm cell). Using a 5-nm slit width, EEMs were collected over an excitation range of 240–450 nm every 5 nm, and an emission range of 300–600 nm every 2 nm. Scans were corrected for instrument optics, inner filter corrected, Raman area normalized, Raman normalized blank subtracted, and multiplied by the dilution factor if necessary [Murphy et al., 2010]. Fluorescence index (FI) was determined as the ratio of the intensities at excitation (ex) and emission (em) wavelengths ex370/em470 and ex370/em520 [Cory et al., 2010].
 EEMs were fit to the previously validated 13 component parallel factor analysis (PARAFAC) model presented in Cory and McKnight to identify the relative abundance of fluorescing components of the EEMs collected for this study. Two of the components (C8 and C13) were proposed by Cory and McKnight to be associated with protein-like fluorophores, whereas components C2 and C4 are associated with aquatic humic substances [McKnight et al., 2001; Cory and McKnight, 2005].
2.5. Modeling the Effects of Increasing Groundwater Discharge on DOM Composition
 We developed a simple mixing model to test the effects of increasing proportion of groundwater discharge to total streamflow on DOC concentration and DOM composition. We selected three rivers (Porcupine River, Salcha River, Tanana River) that differ with respect to DOM composition and the proportion of groundwater discharge to streams at present [Walvoord and Striegl, 2007; Brabets and Walvoord, 2009]. Our overall objective was to track DOM across a range of conditions reflecting projected increases in groundwater discharge. For the end-member mixing model, we solved the following equations
where [DOM] represents DOC concentration or a DOM chemical property in stream water (STREAM), groundwater (GW), or shallow soil water or runoff (SW), and f is the fraction of stream water derived from either GW or SW. We prescribed values of 0.05, 0.25, 0.50, 0.75 and 1.00 for fGW to capture the full range of potential groundwater fractions, though recognizing the unlikelihood of achieving close to 1.00 fGW, even for a totally permafrost-free condition. We focused our analysis on changes in chemical composition during the summer-autumn period (July through October), because the relative influence of groundwater on flow during the snowmelt period (May–June) is minimal.
 Findings from this exercise are intended to gain basic understanding of system behavior and place bounds on potential changes in DOM with increased groundwater discharge. Absolute values of predicted change have limitations due to the influence of assumptions required for analysis. First, we assume no change in summer-autumn flow, which is in agreement with historical records [Brabets and Walvoord, 2009] and supported by some climate projections that predict only modest increases for the YRB [Aerts et al., 2006]. Second, we assume that DOC and DOM properties respond linearly to increased groundwater discharge. Recent literature, however, suggests that a nonlinear response to warming-driven changes in DOC production, transformation, mineralization, and permafrost-carbon release may be expected [Christ and David, 1996; Neff and Hooper, 2002; Dutta et al., 2006; Waldrop et al., 2010]. Third, we assume that two end-members are sufficient to constrain summer-autumn conditions. In actuality, post-thaw changes in hydrology are likely very complex [e.g.,Carey and Woo, 2000], particularly in the discontinuous permafrost zone, and cannot be accounted for with this data set.