Carbon dioxide (CO2) and methane (CH4) emissions are important, but poorly quantified, components of riverine carbon (C) budgets. This is largely because the data needed for gas flux calculations are sparse and are spatially and temporally variable. Additionally, the importance of C gas emissions relative to lateral C exports is not well known because gaseous and aqueous fluxes are not commonly measured on the same rivers. We couple measurements of aqueous CO2 and CH4 partial pressures (pCO2, pCH4) and flux across the water-air interface with gas transfer models to calculate subbasin distributions of gas flux density. We then combine those flux densities with remote and direct observations of stream and river water surface area and ice duration, to calculate C gas emissions from flowing waters throughout the Yukon River basin. CO2 emissions were 7.68 Tg C yr−1 (95% CI: 5.84 −10.46), averaging 750 g C m−2 yr−1 normalized to water surface area, and 9.0 g C m−2 yr−1 normalized to river basin area. River CH4 emissions totaled 55 Gg C yr−1 or 0.7% of the total mass of C emitted as CO2 plus CH4 and ∼6.4% of their combined radiative forcing. When combined with lateral inorganic plus organic C exports to below head of tide, C gas emissions comprised 50% of total C exported by the Yukon River and its tributaries. River CO2 and CH4 derive from multiple sources, including groundwater, surface water runoff, carbonate equilibrium reactions, and benthic and water column microbial processing of organic C. The exact role of each of these processes is not yet quantified in the overall river C budget.
 Streams and rivers are primary routes for delivery of dissolved and particulate inorganic and organic carbon (DIC, PIC, DOC, POC) from terrestrial landscapes and inland waters to coastal areas and oceans [Meybeck, 1982; Gaillardet et al., 1999; Cole et al., 2007]. They also act as reactors for biogeochemical transformations among aqueous carbon species, including the production and consumption of carbon gases and the subsequent exchange of carbon dioxide and methane (CO2, CH4) with the atmosphere [Richey et al., 2002; Billett et al., 2004; Mayorga et al., 2005; Guerin et al., 2006, 2007; Cole et al., 2007]. Because most inland waters are supersaturated with C gases, they are net sources of CO2 and CH4 to the atmosphere and are quantitatively important in ecosystem carbon budgets at local to global scales [Cole et al., 2007; Tranvik et al., 2009; Aufdenkampe et al., 2011; Butman and Raymond, 2011]. However, C gas emissions by rivers remain poorly quantified. This is because data are generally lacking for: 1) seasonal and spatial distribution of river pCO2 and pCH4, 2) gas exchange rates across the range of river flow conditions, and 3) river water surface areas. From a river system carbon balance perspective, whole-river accounts of particulate, dissolved, and gaseous carbon exports are almost entirely lacking in the literature. Therefore, linkages among watershed sources and sinks of C, including net ecosystem C exchange, and riverine exports are not accurately quantifiable.
 Recent warming of northern high latitudes has resulted in substantial research of potential changes in C cycling of terrestrial arctic and boreal ecosystems [Chapin et al., 2005; Grosse et al., 2011]; biogeochemical characterization and quantification of lateral exports of C from watersheds and to northern seas and the Arctic Ocean [Carey, 2003; Frey and Smith, 2005; Striegl et al., 2005, 2007; Guo and Macdonald, 2006; Guo et al., 2007; Walvoord and Striegl, 2007; Spencer et al., 2008, 2009; Frey and McClelland, 2009; Holmes et al., 2012]; and permafrost thaw effects on greenhouse gas exchange between land ecosystems and the atmosphere [Wickland et al., 2006; Zhuang et al., 2004]. Although it is well known that arctic rivers tend to be supersaturated with CO2 relative to the atmosphere [Kling et al., 1991], quantitative basin-wide comparisons of arctic and boreal river aqueous C exports and gaseous C emissions have not been made. We hypothesized that CO2emissions are quantitatively important in high latitude river C budgets even though within-river biogeochemical cycling is likely small due to cold water temperatures, fast river velocities, and long winter ice cover. River CH4 emissions, although potentially important from a greenhouse gas perspective, were hypothesized to be much smaller or negligible because of the general lack of impoundments or other sites conducive to CH4 production in most arctic rivers. Consequently, in this paper we: 1) evaluate the seasonal and spatial distribution of river pCO2 and pCH4, 2) characterize water to air gas exchange rates, and 3) estimate the total surface area for rivers and streams throughout the Yukon River basin (YRB) of USA and Canada (Figure 1). Those factors are then used calculate annual CO2 and CH4 losses from the Yukon River system to the atmosphere. The losses are compared with annual lateral exports of DIC, PIC, DOC, and POC by the Yukon River [Striegl et al., 2007] to quantify the relative importance of C gas loss and quantify total river C exports for the basin. Sources of Yukon River gas emissions are discussed, as are potential climate warming effects on C gas emissions from other boreal and arctic rivers in the Arctic Ocean watershed.
2. Study Area and Methods
2.1. Yukon River Basin
 The 3340 km long Yukon River is the largest free-flowing river in North America, draining 854,700 km2 of northwest Canada and central Alaska, USA and discharging approximately 205 km3 of water per year to Norton Sound on the Bering Sea (Figure 1). The majority of the basin is boreal forest, with the exception of mountain ranges, including the highest mountains in North America; ice and snowfields in headwater areas of the Alaska Range and the Wrangell, St. Elias, and Coast Mountains; and coastal tundra in the western portion of the YRB. Unlike other large rivers that discharge freshwater to the Arctic Ocean, the generally east to west course of the Yukon River remains subarctic; does not flow through arctic tundra; and has continuous, discontinuous or sporadic permafrost throughout the basin. Because of this, the YRB is particularly well suited for evaluation of permafrost thaw effects on hydrology and carbon cycling in boreal forest. Basin lowlands include black spruce forest, extensive areas of wetlands and lakes known as “flats,” shrublands, and tundra. Vegetated uplands include spruce, pine, aspen, and birch forest, shrublands, and tundra [Brabets et al., 2000; Striegl et al., 2007].
2.2. Background: Flux Calculations
 Gas flux across the water – air interface can be calculated by modification of Fick's First Law [Fick, 1855] where river gas flux (Jg) in moles or grams of C per m2 per day is the product of the gas concentration gradient between the river and the atmosphere (Cw − Ca) (moles or grams C m−3), and the gas transfer velocity for water - air gas exchange (k) (m d−1):
 Jg varies across large river basins because Cw and k vary with geography, hydrology, and other environmental factors. This necessitates partitioning large river basins by subbasin, region, or some other physical, geochemical, or ecological characteristic(s) in order to calculate the total annual riverine gas flux. Details regarding generalization of the terms of Jg and the partitioning of YRB into subbasins are explained in following sections.
2.3. Gas Concentrations
 CO2 and CH4 concentrations were determined on >900 open water samples collected from >240 locations during 2001–2011, as part of field campaigns conducted during 2001–2005 that focused on water and materials exports by the Yukon River and its tributaries [Schuster, 2003, 2005a, 2005b, 2006; Dornblaser and Halm, 2007; Halm and Dornblaser, 2007; Striegl et al., 2005, 2007; Dornblaser and Striegl, 2007, 2009; Walvoord and Striegl, 2007], as part of studies conducted during spring high flows at USGS streamgaging stations and other sites during 2005–2011, and on samples collected by native Alaskan community volunteers from throughout the YRB during 2006–2011. Bubble-free water samples were collected from the upper 10 cm of flowing water using a 60 mL polypro syringe fitted with a 3-way stopcock. Fifteen mL of sample were injected from the syringe through a 0.45-micron syringe filter into 30 mL glass serum bottles that were previously flushed with N2 at atmospheric pressure and sealed with gas inert stoppers at our Boulder, Colorado laboratory. Serum bottles also contained 2 g KCl to inhibit microbial activity [Striegl et al., 2001]. Surface water temperature, air temperature, specific conductance and pH were recorded when the samples were collected. CO2concentration (volume fraction) in serum bottle headspace was determined at our laboratory using a Li-Cor 6252 infrared CO2 analyzer fitted with a sample injection port and N2 carrier gas. CH4 concentration was determined using a HP 5890 gas chromatograph having a flame ionization detector [Striegl and Michmerhuizen, 1998]. Aqueous CO2 and CH4 concentrations at field conditions were determined from measured headspace gas volume fractions, average total barometric pressure at the sample collection site, field water temperature, laboratory equilibration temperature, and the appropriate Henry's law constants [Yamamoto et al., 1976; Plummer and Busenberg, 1982].
 When serum bottles were collected without CO2 gas flux measurements, the CO2concentration above the air-water interface was assumed to be 390 ppm by volume (ppmv) for the purpose of calculating the gas concentration gradient (Cw − Ca). When CO2 flux measurements were made, the CO2concentration above the water-air interface was measured in the field using a PP Systems EGM Infrared Gas Analyzer (EGM-IRGA). For the occasions when CH4flux measurements were made in the field, a sample of the air above the air-water interface was collected and injected into evacuated serum bottles for GC analysis in the laboratory. For the majority of instances, direct CH4 flux measurements were not made and the CH4concentration above the water-air interface was assumed to be 1.90 ppmv, which was representative of field measurements.
2.4. Gas Flux Measurements and Gas Transfer Velocities
 Besides being very large, most of the Yukon River basin is very remote and difficult to access, precluding application of tracer techniques or other methods for determination of water - air gas exchange rates that require cumbersome field equipment. We empirically derived gas transfer velocities across the water - air interface from floating chamber measurements. On 365 occasions and at 62 locations, CO2 flux was directly measured by recording the ppmv CO2of air circulating through a 290 mm tall by 300 mm diameter floating chamber using an EGM-IRGA. Flux chambers disrupt natural gas concentration gradients with time [Healy et al., 1996] and can also cause surface disturbance when deployed on water [Cole et al., 2010, and references therein]. Our field strategy for accurate measurements was therefore to minimize deployment times and surface disturbances. Our deployment times were typically 5 min for CO2 measurements. CH4 measurements required additional time for sufficient concentration change to calculate CH4 flux and were typically 20–30 min. We minimized surface disturbance by using lightweight chambers that penetrated the water surface ∼25 mm. Whenever possible, fluxes were measured with the chamber and observation boat floating together with the river current [Guerin et al., 2007]. Chambers were stationary on smaller streams, recognizing that the deployment caused some surface disturbance [McMahon and Dennehy, 1999]. Chamber fluxes were calculated from the rate of measured change in chamber CO2 concentration and chamber height according to:
where is the CO2 concentration change in the chamber air (mol or mg C m−3 air) over the deployment time and h is chamber height (m). Combining equations (1) and (2) with the measured CO2 concentration gradient (Cw − Ca) at the time of chamber deployment, kCO2 was empirically derived as:
 We simultaneously measured CH4 and CO2 emissions on 25 occasions to field verify the relationship between kCO2 and kCH4 for our measurement system. This was done in order to calculate kCH4 for locations where CH4 flux was not directly measured by using our much larger (n = 365) kCO2 database, having greater spatial and temporal coverage. This comparison was based on the assumption that kCH4 can be related to kCO2 according to:
where Sc is the Schmidt number, defined as the ratio of kinematic viscosity of water (ν) to mass diffusivity (D) of the gas, Sc = ν/D. Since both gases were measured simultaneously under identical flow conditions, we hypothesized that the ratios of their diffusivities should hold fairly constant across the range of measurement conditions. The exponent, n, varies from 1 to 0.5 based on the dominant process of gas transfer (film model versus surface renewal, respectively) [Jähne et al., 1987]. A value of 0.67 is commonly used to denote moderately turbulent conditions, but n can be derived precisely using simultaneous gas exchange measurements for two gases as:
 For the field comparison, CO2 and CH4 measurements were made as previously described, and a syringe port was added to extract 4 – 6 chamber air samples into evacuated serum bottles for CH4 analysis in the lab. Linear regression of empirical kCH4 versus kCO2 produced a slope of 1.27 (r2 0.88; n = 25). At standard temperature and pressure, DCH4/DCO2 = 1.41, and so n = log(1.27)/log(1.41) = 0.69. This result suggests that kCH4 and kCO2 are related using equation (4) and assuming moderately turbulent conditions such as those common in rivers. Therefore, we calculated kCH4 using equation (4) and assuming n = 0.69. For the purpose of comparing our results to other studies, we will also discuss CO2 exchange velocity as k600, which is the empirical gas exchange velocity of CO2 (kCO2) normalized to 20°C. Our empirical kCO2 values were typically much lower than k600 because of the low average water temperature (9.4°C).
2.5. Water Surface Area of Streams and Rivers and Ice-Free Days
 Stream and river water surface areas were derived for the YRB using two approaches. For the Yukon River and major tributaries, we derived water surface areas using Landsat data. Because the moderate resolution of Landsat data (30 m × 30 m pixel) limits the detection of small streams and rivers, small features (sub-pixel streams) were estimated from a stream network generated from a digital elevation model (DEM) [Jenson, 1991]. Total stream area was calculated as the sum of sub-pixel stream areas and the Landsat estimated areas for larger streams and rivers.
 A supervised classification approach was used to estimate larger stream and river surface water areas from Landsat data. Sixty-six scenes of Landsat Enhanced Thematic Mapper Plus (ETM+) data acquired during the growing season between 1999 and 2002 were processed to create a seamless mosaic of the Yukon River basin [Bouchard et al., 2009]. The data were collected prior to the ETM+ Scan Line Corrector failure in 2003. ETM+ bands 7, 4, and 2 (shortwave infrared, near-infrared, and green, respectively) were radiometrically normalized during the mosaic process [Bouchard et al., 2009] and training pixels representing water and nonwater features were manually selected across the basin. The training data were then used to create a decision tree model, which in turn, was applied to the 3-band image mosaic (www.rulequest.com). The final classified image contained two classes, water and nonwater. The water features were extracted by converting the water pixels to polygons. The polygons were manually edited to reduce misclassifications and then categorized as the main stem Yukon River, a tributary, or a lake. Area was calculated for each non-lake polygon in the final map. A large subset of the final map (93,412 km2) was used to produce an estimate of model classification error (4.6%; n = 1,381).
 Sub-pixel stream and river areas were estimated from a stream network [Jenson, 1991]. Digital elevation data for the USA (http://ned.usgs.gov) and Canada (http://www.geobase.ca) were combined to create a 50-m DEM for the basin. From the DEM, we used a flow accumulation model to compute the accumulated flow to each pixel. A minimum flow accumulation threshold of 3000 pixels (2.7 km2) was selected to represent a stream or river and each stream and river was ordered using the Strahler stream order method. From this stream network, total stream length was calculated. The area of sub-pixel streams was then estimated by combining the calculated lengths with a mean stream width based on order. Confidence intervals were constructed from the standard deviation of the data [Downing et al., 2012].
 Ice-free days vary throughout the YRB, depending on stream size, proportion of groundwater contributing to flow, stream channel slope and aspect, and local weather conditions. Detailed records of ice-on and ice-off dates are lacking for the Yukon River basin and extensive cloud cover combined with the Landsat repeat overpass cycle of 16 days during ice-on and ice-off periods preclude precise assessment of ice duration. Open water occurs at some locations throughout winter, ice cover persists at other locations into June, and the Yukon River is generally ice covered from late November to mid-May. For the purposes of this study, we estimate 200 ice-free days per year and assume that all aqueous CO2 and CH4 transported by the Yukon River during ice cover is discharged to Norton Sound and not directly emitted to the atmosphere.
2.6. Calculation of Annual River Gas Emission
 We assumed that k and (Cw − Ca) values would differ between the main stem Yukon and its tributaries because these properties are known to change with stream order [Butman and Raymond, 2011]. We also assumed that k and (Cw − Ca) would vary spatially, reflecting the transition from prevalent steep terrain, mountainous areas, glaciers and ice fields in the upper part of the watershed to a greater influence of low-relief flats areas marked by numerous lakes, ponds, and wetlands in the downstream parts of the watershed. Based on this, we divided the Yukon River basin into three geographic subbasins from east to west, the Upper Yukon, Middle Yukon, and Lower Yukon (Figure 2), and into main stem Yukon River and tributaries, for a total of six areal categories. The Upper Yukon includes the Yukon headwaters, upper Yukon River, and the Teslin, Pelly, Stewart, and White River watersheds. The Middle Yukon includes the basin area downstream to just below the Tanana River and including the Porcupine, Chandalar, and Tanana River watersheds. The Lower Yukon includes all YRB area downstream of the Tanana River.
 Gas fluxes were calculated using empirical kCO2, pCO2, and pCH4 values and the derived kCH4. Annual emissions were determined separately for the six areal categories, multiplying by the area of stream and river water and the number of ice-free days within each category, and then summing to produce a total basin estimate. Mean values of (Cw − Ca) and k were used in equation (1). Non-parametric 95% confidence intervals on the mean were computed for both gas concentration andk using the adjusted bootstrap percentile method [Efron, 1987] on 100,000 replicates generated using ordinary bootstrapping.
3. Results and Discussion
3.1. Gas Concentrations and Gas Transfer Velocities
 The partial pressure of CO2 averaged >1,500 μatmos (n = 981) for all samples, with approximately 95% of samples having pCO2 in excess of atmospheric equilibrium. Samples having near or less than atmospheric pCO2 were mostly located in the Upper Yukon region, in streams receiving meltwater directly from ice and snowfields (Figure 2a). Excess CO2 (Cw − Ca) averaged 68 μmol L−1 across the YRB and was highly positively skewed (γ = 5.6), meaning that there were relatively few very high values. Excess CO2 was much greater in the tributaries, averaging 100 μmol L−1, as compared with the main stem Yukon River, 40 μmol L−1 (Figures 3a and 3b) (two-tailed t-test on log-transformed data,df = 979, t = 15.8, P < 0.0001).
 River pCH4 averaged 8.4 μatmos (n = 947) and all samples were supersaturated with CH4 respective to the atmosphere (Figure 2b). Excess CH4 averaged 0.37 μmol L−1 and was also positively skewed (γ = 9.5). Similar to CO2, excess CH4 was significantly greater in the tributaries, 0.6 μmol L−1, than in the main stem Yukon River, 0.2 μmol L−1 (Figures 3c and 3d) (two-tailed t-test on log-transformed data,df = 945, t = 6.4, P < 0.0001).
 Mean kCO2 was greater in the tributaries, 5.2 m d−1, than in the main stem Yukon River, 3.1 m d−1 (Figures 2c, 3e, and 3f) (two-tailed t-test on log-transformed values,t = 3.4, P < 0.001). The data had slightly positive skew (γ = 2.1 and 1.2 in the main stem and tributaries, respectively). Other studies from the boreal region report widely ranging k600 values for streams and rivers, from 1–3 m d−1 [Aufdenkampe et al., 2011] to 6–15 m d−1 [Humborg et al., 2010]. Our gas exchange velocities, expressed as k600, fall within this range, 7.6 m d−1 in the tributaries and 4.9 m d−1 in the main stem Yukon River and were similar to the average k estimated for streams of like order in the contiguous USA [Butman and Raymond, 2011]. Humborg et al.  calculated k600 based on water discharge and stream slope [O'Conner and Dobbins, 1958] resulting in lower k values at high stream discharge for their observation locations. Our sample locations were generally on larger rivers having greater average discharge and smaller measured k600 than the Swedish streams, which is consistent with the model used by Humborg et al. .
 One-way ANOVAs were performed on log-transformed river concentration data, defined as log C = ln(100 + C). Calculations of excess CO2 confirmed the validity of our a priori decision to divide the basin into three subbasins, the Upper Yukon, Middle Yukon, and Lower Yukon (Figure 3). The three subbasins differed significantly from each other in both the main stem (F = 68, P < 0.0001, LSD P < 0.05 for all levels) and the tributaries (F = 32, P < 0.0001, LSD P < 0.05 for all levels). Excess CO2 generally increased in the main stem Yukon River from upstream to downstream (Figure 2a). Excess CO2 in the tributaries was greatest in the Middle Yukon region (Figure 2b).
 The data for excess CH4 and kCH4were not normally distributed after being parsed into subbasins, even after several different attempts at transformation. So, one-way ANOVA statistics would not necessarily be instructive to test for differences among the regions for those values. Nevertheless, we used the six a priori categories originally proposed for basin-wide extrapolation because differences were found for excess CO2 among the regions and because differences were apparent between the Yukon River and its tributaries.
3.2. Water Surface Area of Streams and Rivers
 The total stream and river surface area for all of the YRB is approximately 10,269 km2, or 1.2% of the total basin area (Table 1). Of this, 3666 km2 (36%) was attributed to the main stem Yukon River and 6603 km2(64%) was attributed to tributaries. The width of the Yukon River is commonly >1 km in many reaches and many major Yukon River tributaries are also very wide, making Landsat estimates of river surface area very reliable. While accounting for over 80% of the total stream length in the YRB, the sub-pixel tributaries only comprised 5.1% of the total computed surface area. Additionally, the Yukon does not have extensive flood plains that remain flooded for long periods like the Amazon River [Richey et al., 2002] or some other large rivers, so the area of the river remains fairly constant throughout the season.
Table 1. Comparison of CO2 Emissions (Gas Fluxes) and Lateral Carbon Exports From Other Large River Basins and Regionsa
k600 (m d−1)
Areal CO2 Flux (mmol C m−2 d−1)
CO2 Emission (Tg C yr−1)
CO2 Yield (g C m−2 yr−1)
Lateral C Flux (Tg C yr−1)
Lateral C Yield (g C m−2 yr−1)
Stream coverage is the percent of stream and river water surface area relative to the total area accounted for. Areal CO2 flux is per unit of water surface area; annual CO2 emission and lateral C flux are per basin or region; CO2 yield and lateral C yield are per unit of watershed area.
 A river water surface area of 1.2% of basin area is high relative to the conterminous USA and worldwide, but much lower than the Amazon basin (Table 1). We speculate that the relatively high river water surface area for the YRB is because the river system is unregulated, having extensive river meanders and braided channels, especially in the upstream main stem Yukon River and in tributaries like the Tanana River. This also suggests that current global estimates of river and stream areas for boreal and arctic regions are low. Aufdenkampe et al.  tabulate a range in total stream plus river surface area for the boreal plus arctic of 10,000 −185,000 km2. The YRB has >10,000 km2, yet it comprises only 7.5% of the basin area of the 6 largest rivers discharging to the Arctic Ocean [Holmes et al., 2012], and a much smaller fraction of all northern high latitudes.
3.3. YRB Carbon Dioxide and Methane Emissions
 We used the product of the bootstrapped mean estimates of the excess CO2 or CH4concentration and gas transfer velocities to scale the results of field studies and estimate C gas emissions for the tributaries and main stem Yukon River in the three geographic regions of the YRB. Despite the high positive skewness in our data, we used the mean values for the basin-level extrapolation. The principal assumption in this approach is that the data are spatially representative of actualpCO2 and pCH4. The implication is that exceptionally high pCO2 and pCH4 are rare but quantitatively important to overall basin fluxes. Therefore, the mean is the best estimator of overall areal flux. Average CO2 excess ranged from 16.3 μmol L−1 in the Upper Yukon main stem river to 110.6 μmol L−1 in the Middle Yukon tributaries and kCO2 ranged from 0.3 m d−1 in the Upper Yukon main stem to 5.6 m d−1 in the Lower Yukon tributaries (Table 2). CH4 excess ranged from 0.15 μmol L−1 in the Upper Yukon main stem to 0.69 μmol L−1 in the Lower Yukon tributaries and kCH4 ranged from 0.4 m d−1 in the Upper Yukon main stem to 7.2 m d−1 in the Lower Yukon tributaries (Table 3).
Table 2. Summary of Bootstrapped Values Used for the CO2 Emission Estimate for the Entire Yukon River Basina
Excess CO2 (μ mol L−1)
Transfer Velocity (m d−1)
Areal CO2 Flux (mmol m−2 d−1)
Stream Surface Area (km2)
Mass Flux (Tg C yr−1)
Mean values are the mean outcome of bootstrapping estimation for the given parameter, with C.I. representing the 95 percent confidence interval of the mean. Areal CO2 flux is per unit of water surface area; sums are for the entire Yukon River basin; and areal C yield is per unit of Yukon basin surface area.
Areal C yield (g C m−2 yr−1)
Table 3. Summary of Bootstrapped Values Used for the CH4 Emission Estimate for the Entire Yukon River Basina
Excess CH4 (μ mol L−1)
Transfer Velocity (m d−1)
Areal CH4 Flux (mmol m−2 d−1)
Stream Surface Area (km2)
Mass Flux (Gg C yr−1)
Mean values are the mean outcome of bootstrapping estimation for the given parameter, with C.I. representing the 95 percent confidence interval of the mean. Areal CH4 flux is per unit of water surface area; sums are for the entire Yukon River basin; and areal C yield is per unit of Yukon basin surface area.
Areal C yield (g C m−2 yr−1)
 River CO2 emissions totaled 7.68 Tg C yr−1 and CH4 emissions totaled 55 Gg C yr−1for the YRB, assuming 200 ice-free days per year (Tables 2 and 3). Although quantitatively small when compared with CO2 emissions, CH4 emissions are important because CH4 is much more radiatively active than CO2, and CH4 emissions are expected to increase considerably with climate warming and permafrost thaw in high latitudes [Wickland et al., 2006]. Although our calculated CH4 emissions represent only 0.7% of the C mass of gas emitted by YRB rivers, they account for about 6.4% of the radiative forcing [Forster et al., 2007]. Normalized to water surface area, the CO2 emissions averaged 750 g C m−2 water yr−1 for all stream and river surfaces, ranging from 14.4 g m−2 yr−1 for the Upper Yukon main stem river to 1320 g m−2 yr−1 for the Lower Yukon tributaries.
 Annual YRB river C gas emissions equaled the 7.8 Tg C yr−1 of dissolved and particulate C exported by the Yukon River to below head of tide during 2001–2005 [Striegl et al., 2007]. This result was somewhat surprising considering the low average water temperature throughout the basin (<10°C) and the fact that soil respiration rates are lower in high latitude ecosystems than in temperate systems [Raich and Schlesinger, 1992], both of which should limit mineralization of organic matter and CO2 production. A first assumption would be that the ratio of Yukon River C gas flux: C lateral flux might be low compared to temperate and tropical systems. Comparison with other river basins having similar data (Table 1) indicates that although this assumption holds true for comparison with the Amazon basin (gas flux: lateral flux = 6.6) [Richey et al., 2002], the 1:1 ratio for the YRB is essentially identical to the ratio for the temperate Mississippi basin (0.9:1.0) [Dubois et al., 2010]. The ratio for Sweden (0.2:1.0) [Humborg et al., 2010] is closer to what we initially expected, and it is likely that many of the small and high elevation watersheds in the YRB behave similarly to the watersheds included in the Swedish C flux estimates.
 Ultimately, CO2 emission from river basins depends on complex interactions among factors that vary geographically, temporally, and perhaps independently of features such as latitude and mean annual temperature. For the Yukon and the Mississippi Rivers, similarity of the ratio of gas flux: lateral flux can be related to the mix of abiotic and biotic processes controlling CO2 emission. Both the Mississippi [Raymond et al., 2008; Dubois et al., 2010] and the Yukon [Striegl et al., 2007; Tank et al., 2012] derive the majority of their DIC from terrestrial carbonate weathering, utilizing soil respiration as a CO2 source for bicarbonate production. The Yukon has additional within stream weathering of glacial carbonate sediments that consume CO2 while also contributing to DIC production [Eberl, 2004; Striegl et al., 2007; Dornblaser and Striegl, 2009]. Weathering sources are sufficient to maintain high DIC concentrations in both rivers and strong upward pCO2 gradients to the atmosphere. However, the percent river water surface area and gas transfer velocities (k) are much greater in the Yukon leading to greater potential for flux. In contrast, the Yukon has ice cover >5 months a year, leaving less time for emissions to occur.
 From a biological perspective, Wickland et al.  report that Yukon River biodegradable DOC is in the range 12–18% of total DOC export, which likely supports <6% of the total CO2emission. Other respiration sources are also likely to be small in the YRB, at least in the river main stem, because of fast currents, low temperatures, lack of pools, and negligible light penetration. Consequently, in-stream biotic processes controlling CO2emission in the Yukon River are small. Conversely, agricultural nutrient loading, slower velocities, warmer temperatures and light penetration in river pools all contribute to in-stream production in the Mississippi, with resultant decreased emission or downward CO2flux into the river during the photoperiod and within-river storage of a fraction of the photosynthate.
 Warming conditions across high latitudes point toward future increases in C gas emissions from the Yukon River system and other high latitude river systems. Bicarbonate yields increase with decreased permafrost coverage across the circumboreal [Tank et al., 2012] and permafrost degradation has been linked to increased base flow and DIC contributions to rivers in the YRB [Striegl et al., 2005; Walvoord and Striegl, 2007]. Additionally, regional hydrologic modeling suggests increased groundwater flow, and potential river migration and redistribution of lakes and other surface water bodies with permafrost degradation [Walvoord et al., 2012]. Any of these changes can directly influence the carbon cycle and CO2 and CH4 emissions from inland waters in boreal and arctic regions.
4. Summary and Conclusions
 We analyzed pCO2, pCH4, gas transfer velocity, and stream and river water surface area for the boreal aquatic network of the Yukon River basin and conclude that annual gaseous C emissions are essentially equal to the lateral C export by the Yukon River system. This was unexpected given the northern high latitude of the Yukon basin. Comparison with the temperate Mississippi River system indicates that the ratio of riverine C gas emission: lateral C export is similar for the two basins. Moreover, it is clear that all sources of CO2 emission for the Yukon River and for other river systems [e.g., Butman and Raymond, 2011] are not fully accounted for and that more detailed C balance data for rivers are needed. Our results utilize directly measured gas concentration and flux data to build on recent assessments of the importance of inland waters in regional and global C cycles [Cole et al., 2007; Battin et al., 2009; Tranvik et al., 2009; Alin et al., 2011] and quantify the importance of CO2 and CH4 emissions in the C balance of a large boreal river system. From the perspective of climate warming, permafrost thaw, and the changing C biogeochemistry of high latitude basins [Frey and Smith, 2005; Striegl et al., 2005; Guo et al., 2007; Frey and McClelland, 2009; Holmes et al., 2012, Tank et al., 2012] our results also establish a baseline that more completely characterizes current C biogeochemistry and hydrology of the Yukon River system [Striegl et al., 2007; Walvoord and Striegl, 2007; Dornblaser and Striegl, 2009; Spencer et al., 2008, 2009] so that future comparisons can be made.
 The National Research Program and the Earth Resources Observation and Science Center of the Water and the Climate and Land Use Change Mission Areas of USGS supported this research. Native Alaskan community volunteers and the Yukon River Intertribal Watershed Council assisted with sample collection throughout the Yukon basin. D. Halm and J. Crawford conducted many of the flux measurements, N. Bliss generated the DEM, and B. Granneman provided water-mapping support. NSF AON, Arctic Great Rivers Observatory Project (OPP-1107774), contributed to collection of samples at Yukon River at Pilot Station during spring in 2009–2011. We thank David Butman, the associate editor, and two anonymous reviewers for their constructive comments that helped to improve the final manuscript. Any use of trade or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.