Emission of Greenhouse Gases From Water Tracks Draining Arctic Hillslopes

Abstract

Experimental and ambient warming of Arctic tundra results in emissions of greenhouse gases to the atmosphere, contributing to a positive feedback to climate warming. Estimates of gas emissions from lakes and terrestrial tundra confirm the significance of aquatic fluxes in greenhouse gas budgets, whereas few estimates describe emissions from fluvial networks. We measured dissolved gas concentrations and estimated emissions of carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O) from water tracks, vegetated depressions that hydrologically connect hillslope soils to lakes and streams. Concentrations of trace gases generally increased as ground thaw deepened through the growing season, indicating active production of greenhouse gases in thawed soils. Wet antecedent conditions were correlated with a decline in CO₂ and CH₄ concentrations. Dissolved N₂O in excess of atmospheric equilibrium occurred in drier water tracks, but on average water tracks took up N₂O from the atmosphere at low rates. Estimated CO₂ emission rates for water tracks were among the highest observed for Arctic aquatic ecosystems, whereas CH₄ emissions were of similar magnitude to streams. Despite occupying less than 1% of total catchment area, surface waters within water tracks were an estimated source of up to 53–85% of total CH₄ emissions from their catchments and offset the terrestrial C sink by 5–9% during the growing season. Water tracks are abundant features of tundra landscapes that contain warmer soils and incur deeper thaw than adjacent terrestrial ecosystems and as such might contribute to ongoing and accelerating release of greenhouse gases from permafrost soils to the atmosphere.

Plain Language Summary

The Arctic is warming at least twice as rapidly as the rest of the Earth. Warming thaws permafrost, defined as ground that remains frozen for more than 2 years, and results in increased production and release of greenhouse gases by microbes. These gases, including carbon dioxide, methane, and nitrous oxide, trap heat in the atmosphere and contribute to further warming. Emissions of these gases have been extensively observed from soils and lakes in the Arctic, but fewer observations describe greenhouse gas emissions from stream networks. We estimated emissions of greenhouse gases from water flowing in vegetated depressions on Arctic hillslopes, called water tracks, which are tributaries of lakes and rivers. Water tracks may be conduits for gases from soils to the atmosphere. Compared to terrestrial tundra, water tracks emitted more carbon dioxide and methane than expected from their small areal coverage. Water tracks typically withdrew nitrous oxide from the atmosphere, though smaller water tracks supported higher concentrations and net emission of nitrous oxide. The abundance of water tracks combined with high rates of carbon dioxide and methane emission relative to adjacent ecosystems suggests that these features might contribute to increased release of greenhouse gases from Arctic landscapes as permafrost thaws.

Key Points

• Water tracks draining Arctic hillslopes are estimated to emit CO₂ and CH₄ to the atmosphere at greater rates than terrestrial tundra
• Water tracks produced and consumed N₂O, resulting in low net emissions to the atmosphere
• Dissolved gas concentrations varied among water tracks and over time and were related to precipitation, redox, and primary productivity

1 Introduction

Climate warming in the Arctic can result in increased emissions of greenhouse gases that offset gains of carbon (C) by primary production, decreasing C storage in organic matter, and contributing to a positive feedback to climate warming (Schuur et al., 2015). Potential for increased release of greenhouse gases as permafrost warms is evidenced by observations of experimental and ambient disturbance to permafrost. Elevated fluxes of carbon dioxide (CO₂) follow ambient and experimental warming of terrestrial tundra (Dorrepaal et al., 2009; Euskirchen et al., 2017; Schuur et al., 2009). Disturbance of permafrost soils, including cryoturbation and thaw, causes emission of nitrous oxide (N₂O; Repo et al., 2009; Voigt et al., 2017). Lakes formed or expanded by rapid thaw of ice-rich permafrost (i.e., thermokarst) release methane (CH₄; Walter Anthony et al., 2016). However, few estimates describe greenhouse gas emissions from high-latitude fluvial networks, and riverine fluxes are therefore omitted from trace gas and carbon (C) budgets for the Arctic (e.g., USGCRP, 2018). Moreover, continental- and global-scale syntheses report evasion of gaseous C from fluvial networks that balances or exceeds terrestrial net ecosystem production (Butman et al., 2016; Drake et al., 2018; Raymond et al., 2013), and fluvial networks of high-latitude regions might similarly offset terrestrial gains of C (Stackpoole et al., 2017).

Accumulating evidence indicates significant, though uncertain, evasion of gaseous C from high-latitude rivers. The Yukon River and its major tributaries emit nearly 8 Tg C year⁻¹ as CO₂ and 55 Gg C year⁻¹ as CH₄ (Striegl et al., 2012), and if fluxes from the Yukon basin are representative of all large river basins in the region, riverine fluxes could account for a significant fraction of the total gaseous C emissions from Arctic and boreal Alaska (25 Tg year⁻¹; Commane et al., 2017). However, fluxes measured in large rivers likely underestimate total emissions of greenhouse gases from fluvial networks, because smaller streams typically emit gases at greater rates (Denfeld et al., 2018; Lundin et al., 2013), drain a larger fraction of the landscape, and typically account for most stream length (Downing et al., 2012). High rates of gas emissions in small streams might result from inputs of dissolved gases produced in soils, as well as delivery of DOC that can be transformed to CO₂ by biodegradation and photodegradation in surface waters (Ward & Cory, 2016). Further, thawing permafrost and increased precipitation cause expansion of low-order channel networks (Liljedahl et al., 2016), which could result in increased cumulative emissions of greenhouse gases from fluvial features in the Arctic.

Fluvial networks draining upland permafrost include abundant preferential, zero-order flow paths known as “water tracks” that flow through hillslope soils and vegetation, transporting the majority of stormflow to receiving streams and rivers (McNamara et al., 1998; Rushlow & Godsey, 2017; Stieglitz et al., 2003). As depression features, water tracks accumulate deeper snow that warms soils (Rushlow et al., 2020), contain perennially saturated soils, and support more rapid throughput and cycling of nutrients (Chapin et al., 1988; Yano et al., 2010), all of which could foster elevated rates of reactions that produce greenhouse gases. For example, significant retention of nitrate observed within saturated organic soils of water tracks (Harms & Ludwig, 2016) could be due to denitrification, which produces N₂O as an intermediate. Carbon dioxide emissions from water track soils can exceed those of adjacent tussock tundra (Jones et al., 1999; Oberbauer et al., 1991), though enhanced primary production in water tracks can also contribute to net uptake of CO₂ (Curasi et al., 2016). In addition to gas production within water tracks, gases produced in adjacent hillslope soils may be hydrologically transported to water tracks, where advective flow at the surface forms an efficient conduit for dissolved gases to the atmosphere. However, emissions of dissolved gases to the atmosphere have not yet been quantified from water tracks.

To assess the potential contribution of low-order portions of fluvial networks to Arctic greenhouse gas budgets, we measured dissolved CH₄, CO₂, and N₂O concentrations in water tracks and estimated emissions using modeled rates of gas transfer velocity. We also assessed responses of dissolved gas concentrations to attributes of climate (i.e., precipitation), hydrologic flow paths (i.e., depth and redox potential), and terrestrial production to test hypothesized mechanisms driving current and potential changes in greenhouse gas emissions. Significant loss of C is hypothesized to occur via lateral transport in terrestrial ecosystems undergoing permafrost thaw (Plaza et al., 2019). Therefore, we expected that water tracks are sources of greenhouse gases within Arctic catchments, emitting gases at rates greater than previously observed from adjacent terrestrial and aquatic ecosystems.

2 Methods

2.1 Site Description

Water tracks are preferential flow paths that develop on hillslopes of upland permafrost terrain. They are vegetated depressions that contain ephemeral to perennial surface flow and perennially saturated soils. Water tracks are ubiquitous features throughout northern Alaska, as well as elsewhere in the Arctic and high Arctic (Curasi et al., 2016; Jorgenson et al., 2008; Paquette et al., 2017). In the study area within the Upper Kuparuk River Basin (~150 km²), located in the northern foothills of the Brooks Range near Toolik Field Station in Alaska (Figure 1), the contributing areas of water tracks account for up to a third of the study basin area (McNamara et al., 1998; Trochim et al., 2016). We sampled six water tracks draining independent catchments that ranged from 0.013 to 0.085 km² at the point of monitoring (Evans et al., 2020; Rushlow & Godsey, 2017).

Figure 1

Left: locations of focal water tracks (WT) and their contributing areas (yellow) within the upper Kuparuk River basin on the North Slope of Alaska. Right: ground and aerial views of WT5. Map credit: C. Rushlow.

The Upper Kuparuk Basin was last glaciated ~150,000 years ago and is characterized by ground moraine landforms and loess deposition derived from sandstone (Hamilton, 2003; Keller et al., 2007). Though carbonate is depleted in soils of the Upper Kuparuk relative to more recently glaciated surfaces in the region, carbonate weathering is the dominant weathering regime (Keller et al., 2007). The study basin encompasses moist acid tundra, with organic soils 10–50 cm thick (Hinzman et al., 1991) and pH ~4 in soils of water tracks (Harms et al., 2014). Vegetation cover in water tracks is continuous, dominated by either dwarf shrubs (Betula nana, Salix pulchra) or sedges (Carex aquatilus, C. bigelowii) and also includes moss (Sphagnum spp.) and forbs (Rubus chamaemorus, Ledum palustre, and Vaccinium vitis-ideaea). Hillslopes adjacent to water tracks included in this study are covered by tussock-forming sedges (Eriophorum vaginatum).

Mean annual air temperature measured at the nearby Toolik Field Station is −10°C, with lowest mean monthly temperature typically in January (−25°C) and highest in July (11.5°C). During the study period (2012–2014 and 2018), temperatures recorded in the Upper Kuparuk basin (2012–2014, http://ine.uaf.edu/werc/projects/NorthSlope/upper_kuparuk/uk_met/current.html) or an adjacent catchment (2018; NEON, 2019) were similar to the long-term average, with exceptions of winter low temperature ~5°C colder than average in 2012 and 2013, and summer high temperature ~2°C colder than average in 2014. The catchment receives 320 mm precipitation annually on average, with 200 mm falling between June and August. Summer precipitation was average in 2012 (199 mm), low in 2013 and 2018 (169 mm), and greater than average in 2014 (233 mm).

2.2 Field Data Collection

Discharge was measured sporadically in 2012 and continuously in 2013–2014. A weir was installed to the depth of maximum annual thaw at each site, and thus, flow through the weirs captured total downslope discharge, including both surface and subsurface flow. In 2012, discharge measurements were made by salt slug dilution at the weir. In 2013–2014, a flume with a stilling well was installed on the downstream side of each weir, and a pressure transducer was installed in the well. A rating curve for each site and year was developed from discharge measured by salt slug dilution. Salt slug dilution was also used to measure discharge at several points along one water track in 2018, but without a weir present at each of these locations, discharge measurements captured only surface flow.

Dissolved gas concentrations in water tracks were measured approximately every 2 weeks from June through August, 2012–2014 (total of 21 sample dates). Triplicate water samples were collected on the upstream side of the weir, and as such, the water sampled was a mixture of surface and subsurface water. Surface water was drawn into a syringe, and bubbles incurred from the atmosphere were removed before filling the syringe with water to a fixed volume (40 ml). Syringes were stored under ice water until headspace equilibration in the laboratory within 12 hr of collection, a protocol similar to other Arctic studies (e.g., Lundin et al., 2013; Walter et al., 2006), but that presents opportunity for biological activity affecting dissolved gas concentrations. Estimates of dissolved CO₂ concentration measured by headspace equilibration (2012–2014) bracketed values measured in situ by a probe (2018; described below) for the same site and month of year, suggesting little effect of this holding time on dissolved gas concentration. We also assessed the potential production of CO₂ in water samples and found that CO₂ concentration in these samples was unlikely to change by more than 10% during holding, due to relatively high dissolved CO₂ concentration, low lability of DOC, and holding temperature near freezing (supporting information). Biocide treatment or immediate equilibration of samples upon collection would eliminate this potential source of uncertainty in estimates of dissolved CO₂ concentration.

Surface water was collected simultaneously with dissolved gas sampling for analysis of solutes, filtered in the field (0.7 μM, Whatman GF/F), and frozen on the day of collection. In 2013, we additionally sampled both surface and soil water in a spatially intensive grid at two sites (Water Tracks 1 and 4), once each in June and August. Sampling grids consisted of 25 sampling points at each site, arrayed along five transects oriented perpendicular to flow direction, with 5 m between adjacent points, on average. Soil water was collected by inserting a needle (perforated over the bottom 5 cm) to the depth of soil thaw and withdrawing samples using a syringe connected by silicone tubing and a stopcock. Surface water was not present in Water Track 4 at the time of sampling in August.

To assess potential input of CO₂ from soils to water tracks, we measured dissolved CO₂ concentration and collected samples of surface water longitudinally along 250 m of Water Track 1 at 15 locations spaced every 10–30 m in July 2018. During this survey, dissolved CO₂ concentration was measured by a handheld sensor (Vaisala DM70, Helsinki, Finland), following methods described in Johnson et al. (2010). Water samples were filtered at the time of collection (0.45 μm, Filtropur Sarstedt, Nümbrecht Germany) and frozen or acidified within 6 hr of collection. Samples were analyzed for bicarbonate (HCO₃⁻), DOC, major anions and cations, and δ¹³C-dissolved inorganic C (DIC) as indicators of the sources of C (mineral [~0‰] vs. organic matter [−27‰]) and depth of contributing flow paths.

Thaw depth was measured four to eight times total during May through August in 2012–2014 within and adjacent to each water track. A thaw probe was used to measure depth of refusal at >30 points along two transects perpendicular to flow direction. Measurements of thaw were made every 2 m outside of the water track and every 20–50 cm within each water track. Mean thaw depth inside and outside of each water track was linearly interpolated through time to estimate thaw depth on the dates of gas sampling.

The gas transfer velocity between the air-water interface (k) was modeled using published relationships with geomorphic and hydrological parameters developed for first- to fifth-order stream channels of low slope (Raymond et al., 2012). Specifically, we used equation 5 in Raymond et al. (2012), which includes slope (S) and water velocity (V) as predictors of k. The slope of each water track was obtained from surface elevation collected using ground-based LiDAR (Rushlow & Godsey, 2017). Water velocity was estimated as time to peak conductivity following release of a salt slug, divided by distance between the release and monitoring locations. We measured water velocity a total of 57 times, though the number of velocity measurements was unbalanced among sites; we therefore pooled all measurements of velocity to estimate k at each site using a bootstrap approach (see section 2.4).

2.3 Laboratory Analyses

Dissolved gas concentrations were measured by headspace equilibration. Water samples were equilibrated with a headspace of either ultrahigh-purity nitrogen (biweekly samples 2012–2014) or with atmospheric air (spatially intensive sampling 2013) to determine dissolved gas concentrations. Atmospheric air used in headspace equilibration was collected outdoors, upwind of potential sources of the analyzed gases, and the equilibration gas was analyzed as described for headspace samples. Samples were shaken for 5 min after introduction of the equilibration gas, and 14 ml of the resulting headspace was transferred to an evacuated 12 ml vial fitted with a butyl septum. All gas transfers were completed under water. Concentrations of CO₂, CH₄, and N₂O were measured within 3 months of collection on a Varian CP-3800 gas chromatograph configured with an electron capture detector, a flame-ionization detector, and a methanizer.

Solute concentrations were measured as proxies of substrate availability, redox conditions, and depth of dominant flow paths through catchments. DOC was analyzed on a Shimadzu Total Carbon analyzer (TOC-L) by combustion followed by nondispersive infrared gas analysis (limit of quantitation [LOQ] = 8 μM). Sulfate (LOQ = 0.10 μM), nitrate (LOQ = 0.04 μM), ammonium (LOQ = 0.7 μM), and calcium (LOQ = 2.9 μM) were analyzed on a Dionex IC25 or Thermo ICS-2100 with AS18 columns for anions and CS16 columns for cations. Nitrate concentration was below the LOQ in >90% of samples and was not included in further data analysis. Samples from the longitudinal transect in 2018 at WT1 were additionally analyzed for ¹³C of DIC. Samples for δ¹³C_DIC determination were acidified with H₃PO₄ as a preservative (Taipale & Sonninen, 2009) and to transform all HCO₃⁻ and CO₃²⁻ to CO₂(g). Carbon isotopic composition of DIC was determined using a Gasbench II extraction line coupled to a Finnigan MAT 252 mass spectrometer. Results are given as per mil (‰) deviations from the standard (PDB), δ¹³C_DIC = (R_sample/R_standard−1) × 10³, where R is the ratio of ¹³C/¹²C. From repeated measurements of standards, the reproducibility was calculated to be better than 0.1% for δ¹³C. Alkalinity was measured on unfiltered water samples from the longitudinal survey that were collected without headspace by titrating to pH 4.5 with HCl (equivalence point between H⁺ and HCO₃⁻) using a Metrohm automated titration system and Metrohm Aquatrode Plus pH electrode (Metrohm AG, Switzerland). Alkalinity was calculated from the difference in the volume of HCl used in titration and the sample volume. Bicarbonate (HCO₃⁻) concentration was calculated from measured alkalinity and pCO₂ values using PHREEQCI (Parkhurst & Appelo, 1999).

2.4 Data Analysis

Concentrations of dissolved gases were calculated from local pressure, temperature at equilibration and of water at collection, and temperature-specific Henry's coefficients. Concentrations resulting from spatial sampling in 2013 were corrected based on measured gas concentrations in the air used for equilibration. Gas fluxes to the atmosphere were calculated using gas-specific Schmidt numbers based on measured temperature and gas-specific k calculated following Raymond et al. (2012). The gas fluxes were then estimated as the product of k and the difference in concentration between water and the atmosphere for a given gas. Atmospheric concentrations were taken as the mean during the growing season measured at the Barrow Atmospheric Baseline Observatory (NOAA; ~400 km northwest of the study sites). Positive fluxes indicate release of the gas from the water surface to the atmosphere, whereas negative fluxes indicate uptake of the gas from the atmosphere.

We applied a Monte Carlo approach to quantify uncertainty in the calculated gas fluxes. To estimate flux for each site, gas, and sampled day, we propagated uncertainties as follows. First we randomly selected one velocity estimate from all measured values and used it to calculate k₆₀₀ using: k₆₀₀ = V × S^{0.76 ± 0.027} × 951.5 ± 144, where V = water velocity (m/s), S = channel slope (m/m), and the fixed coefficients are indicated as mean ± 1 SD from Raymond et al. (2012). Here we also randomly selected the fixed coefficients of the k₆₀₀ model with the parameters described above. k₆₀₀ was then converted to a gas-specific k at field temperature (following Raymond et al., 2012). To account for analytical uncertainty in the gas concentrations, we used one of the triplicate gas concentration values selected at random from the samples collected at the particular site and day to calculate the flux. This process was repeated 10,000 times for each site, gas, and sample day to obtain a mean value and 0.05–0.95 quantiles.

2.5 Statistical Analysis

All statistical analyses were performed in R v. 3.5.3. We quantified correlations between dissolved gas concentrations and hypothesized environmental drivers using mixed effects models, with water track and year as random effects. Potential predictors in the models included measures or proxies for hypothesized contributions of accumulated precipitation in the previous day or week; soil strata contributing to trace gas dynamics (thaw depth); autotrophic respiration, plant-derived organic substrates, and photo-degradation (GPP and PAR measured by eddy flux and associated sensors in the adjacent Imnavait Creek catchment; Euskirchen et al., 2017); C substrate availability (DOC); limiting nutrients (ammonium; Chapin et al., 1995); depth of dominant flow paths (calcium; Keller et al., 2007); and redox (sulfate; Pester et al., 2012). Though temperature is a potential driver of biogenic production of trace gases, its direct influence on solubility is included in calculation of dissolved gas concentration, and temperature was therefore not additionally included as a candidate predictor of gas concentration. We assessed potential covariation of water temperature with all candidate predictor variables and found limited association between them (marginal R² < 0.2). Predictor variables were centered and standardized such that reported coefficients from mixed effects models represent the change in dissolved gas concentration resulting from a 1 SD change in the predictor variable. Correlated predictors (|r| > 0.5) were not allowed in the same model. We identified the five strongest predictors by constructing all possible subsets of linear regressions containing a maximum of five predictors using the leaps package. We then included the random effects and fit all combinations of four or fewer predictors, using Akaike's information criterion (AIC) as a metric to select the most parsimonious model that was best supported by the data, considering models with a difference of AIC values equal to 2 as equivalent. Random effects for all three gases included group-level intercepts for water track and year nested within water track. Models predicting CH₄ concentration additionally included group-level slopes. Mixed effects models were fit initially using the lme4 package. Assumptions of normality and homogeneous variance were assessed visually on residuals. CH₄ was log-transformed to meet the assumption of normality. Heterogeneous variances for CO₂ were modeled as an exponential function of calcium concentration using the nlme package. Variance explained by fixed effects and the combination of fixed and random effects were quantified by marginal and conditional R² values, respectively, using the MuMIn package.

We examined horizontal and vertical variation in dissolved gas concentrations using samples collected in 2013 from spatially explicit grids. To quantify spatial patterns, we fit spatially correlated error structures to the data using the nlme package. However, these models did not perform better than models lacking spatial structure. Therefore, we applied mixed effects models with sampling point nested within site (water track) as a random effect to account for pairing of surface and soil water samples. Season (June and August), water source (surface and soil water), and their interaction were included as fixed effects. CH₄ concentration was log-transformed and models for both CO₂ and CH₄ concentrations included terms accounting for heterogeneous variance. Post hoc Tukey's pairwise comparisons were applied to the interaction term using the emmeans package with α = 0.05. To examine a potential effect of depth on concentrations of gases dissolved in soil water, we applied mixed effects models with water track site as a random effect and depth, season, and their interaction as fixed effects. We compared these models to a null model lacking these fixed effects and used AIC values to evaluate support for effects of depth and season.

We examined common trends in the longitudinal data (collected in 2018) using dynamic factor analysis (DFA) with the MARSS package (Holmes et al., 2012). Similar to principal components, DFA identifies common trends within autocorrelated data sets. We applied model selection to determine an appropriate error structure and the optimal number of trends explaining co-occurring patterns in concentrations of CO₂, DOC, bicarbonate (HCO₃⁻), calcium (Ca²⁺), and ¹³C-DIC. Data were centered and standardized prior to analysis.

3 Results

3.1 Greenhouse Gas Concentrations and Fluxes

Concentrations of dissolved CH₄, CO₂, and N₂O varied across sites and between years (Figure S1). Whereas CH₄ and CO₂ were always supersaturated relative to the atmosphere, N₂O was typically observed in equilibrium, though both supersaturated and undersaturated concentrations also occurred. Undersaturation of N₂O was most common in the wet year (2014), and supersaturation occurred more frequently during the year of average precipitation (2012).

All water tracks were estimated as net sources of CH₄ and CO₂ to the atmosphere, while generally being a weak sink of N₂O (Figure 2; Table 1). Large variation in estimated CH₄ fluxes among water tracks was contributed by tenfold greater fluxes from WT1 and WT5 compared to the other water tracks (Table 1). Cross-site variation in estimated fluxes of CO₂ was only 2.5-fold, with the highest flux to the atmosphere in WT5 and lowest in WT1. On average, all water tracks except WT3 were a weak sink of N₂O over the growing season, though each water track was occasionally a source to the atmosphere. Fluxes were calculated using modeled estimates of k that varied among sites only due to differences in slope (Table 1), and therefore, temporal patterns in fluxes followed the same patterns as gas concentrations (Figure S1).

Figure 2

Emissions of greenhouse gases from six water tracks. Points represent bootstrapped means with 95% confidence intervals reflecting uncertainty in estimated k and analytical error in gas concentrations. Gray horizontal lines represent equilibrium with the atmosphere for CH₄ and N₂O (i.e., no emission or uptake of the gas).

Table 1. Mean Greenhouse Gas Fluxes From Water Tracks During the Growing Season (June–August) 2012–2014

Site	CH4 flux (mmol CH4 m^−2 day^−1)	CO2 flux (mmol CO2 m^−2 day^−1)	N2O flux (mmol N2O m^−2 day^−1)	k600 (m day^−1)
WT1	16.70 (3.22–36.90)	719 (152–1,530)	−0.009 (−0.030–0.010)	8.6 (2.0–19.1)
WT2	1.01 (0.22–2.08)	1,020 (215–2,178)	−0.018 (−0.050–0.006)	15.9 (3.7–34.8)
WT3	3.33 (0.21–10.4)	1,392 (297–2,178)	0.007 (−0.040–0.059)	18.1 (4.3–39.7)
WT4	0.91 (0.19–1.91)	1,040 (223–2,196)	−0.012 (−0.050–0.020)	15.1 (3.6–33.0)
WT5	12.20 (2.65–25.10)	1,855 (406–3,861)	−0.020 (−0.080–0.030)	19.6 (4.6–42.7)
WT6	1.76 (0.38–3.72)	1,451 (311–3,038)	−0.010 (−0.050–0.020)	14.3 (3.4–31.5)

• Note. Values in parentheses designate uncertainty of the estimates (0.05–0.95 quantiles of the bootstrap distribution). Positive values of gas fluxes designate emission to the atmosphere and negative values indicate uptake.

3.2 Seasonal Patterns

Most variation in CH₄ concentration was attributed to contrasts among water tracks and years, whereas CO₂ and N₂O concentrations varied in concert with measured environmental attributes. In mixed effects models, random effects of water track and year explained ~70% of observed variance in CH₄ concentration. Variation contributed by site was largely due to CH₄ concentration that was typically 2–10 times greater in Water Track 1, the largest catchment in the study (Figure S1), compared to the other water tracks. Less variation (~10%) in CO₂ and N₂O concentrations was attributed to water track and year. CO₂ concentration was positively correlated with terrestrial GPP, DOC, and calcium concentrations and was negatively related to precipitation falling in the previous week (Figures 3, S2, and S3). N₂O concentration was negatively correlated with DOC and calcium concentrations, weakly negatively related to sulfate concentration, and positively correlated with PAR (Figure 3). Gas concentrations were unrelated to thaw depth.

Figure 3

Results of linear mixed effects models predicting dissolved gas concentrations. Plots show coefficients (±95% confidence interval) for fixed effects in models explaining ln (CH₄), CO₂, and N₂O, respectively. Coefficients reflect the change in gas concentration resulting from a 1 SD change in each predictor. R²m represents variance explained by the marginal (fixed) effects only, and R²c represents the conditional (fixed + random) effects.

3.3 Spatial Patterns

Though dissolved gas concentrations were spatially heterogeneous in both surface and soil water at two water tracks during June and August (Figure 4), statistical models incorporating spatial structure did not improve model fit, and we conclude that spatial patterns were not discernable at the grain (~5 m) and extent (~25 m) sampled. Concentration of CO₂ tended to increase in August compared to June, and this pattern was significant for both surface and soil water (Figure 4). Tests of paired differences between surface and soil water at each sampled location indicated significantly greater concentrations of CO₂ and lower concentration of N₂O in soil compared to surface water during August (Figure 4). Within seasons, spatial variation in depth of soil water collection did not explain variation in gas concentrations.

Figure 4

Dissolved gas concentrations in surface and subsurface water of two water tracks. Center line represents the median, extent of boxes the 25th and 75th percentiles, and whiskers 1.5 times these values for n = 22–25 points sampled on a regular grid spanning the full width of each water tracks, along 25 m longitudinal distance and 5 m average grain. Note the log scale of the concentration axes. Surface water was not present within the sampling grid at WT4 during the time of sampling in August 2013. Horizontal lines represent the average atmospheric concentration of each gas during summer (measured at Barrow Atmospheric Baseline Observatory).

Discharge, dissolved C concentrations, and isotopic composition of DIC varied systematically along a 250 m reach of Water Track 1 (Figure 5). Discharge, ¹³C-DIC, calcium, and HCO₃⁻ concentration increased downstream, whereas DOC and dissolved CO₂ concentrations declined (Figure 5). The best fit DFA model of these longitudinal patterns included three statistically distinct longitudinal trends (i.e., states), as well as equal variances of all variables and covariances equal to zero. For each of the three trends, DOC and CO₂ loaded with opposite sign from the remaining geochemical variables, reflecting opposing downstream trends (Figure 5).

Figure 5

Longitudinal patterns in flow and solute chemistry of WT1, July 2018. Left column: observed solute concentrations, δ¹³C-DIC, and discharge. Middle: longitudinal trends extracted by dynamic factor analysis. Right: factor loadings on the trends for each constituent.

4 Discussion

Fluxes of greenhouse gases from Arctic fluvial networks remain an uncertain contributor to the high-latitude C budget and to climate forcing. Water tracks, ubiquitous zero-order flow paths draining upland tundra, were generally sources of CH₄ and CO₂ to the atmosphere, but weak sinks of N₂O. Concentrations of dissolved gases were related to precipitation, dominant hydrologic flow paths, and terrestrial productivity, all features of catchments that are expected to change under a warming climate. Together with observations of expanding fluvial networks where permafrost is thawing (e.g., Liljedahl et al., 2016), these patterns suggest that water tracks currently contribute significantly to greenhouse gas budgets and their contribution might increase during periods of permafrost thaw.

4.1 Potential Drivers of Variation in Dissolved Gas Concentrations

Supersaturation of CH₄ is typical of high-latitude freshwaters, though average concentration of CH₄ observed in water tracks (340 ppm) exceeded that previously observed in Arctic streams (~25 ppm), and was similar to concentrations in flows draining thermokarst features (~83–1,000 ppm; summarized in Zolkos et al., 2019). Undersaturation of CH₄ relative to the atmosphere also occurred in some sites earlier in the thaw season (Figures 2 and 4), which would contribute a small, ephemeral sink for CH₄. Significant variation across sites was largely due to consistently elevated CH₄ concentration in Water Track 1, the site with largest contributing area and greatest discharge, during wetter years (2012 and 2014). Perennially inundated soils at this site likely lead to low redox potential, which was indicated by gleyed soils at depth (T.K. Harms, unpublished data). Street et al. (2016) also documented increased dissolved CH₄ concentration in Arctic soils and streams when anoxia persisted. The negative correlation of dissolved CH₄ concentration with sulfate (Figure 3), a pattern also observed in ponds of the High Arctic (Emmerton et al., 2016), further suggests that redox potential influenced seasonal variation in CH₄ dynamics (Figure 3), as sulfate is a redox-sensitive species.

The average concentration of CO₂ in water tracks (2,480 ppm) was similar to or greater than the concentration observed in undisturbed streams across the Arctic (~1,690 ppm; summarized in Zolkos et al., 2019), exceeded that of streams and rivers in the study region, including the Upper Kuparuk River (610 ppm), and was similar or greater than concentration in Arctic lakes (1,160 ppm; Kling et al., 1991). Seasonal variation in CO₂ concentration was greater than variation among sites or years and was explained by positive correlations with DOC concentration and GPP, indicating the roles of substrate availability and a potential contribution of autotrophic respiration. A positive correlation between dissolved CO₂ and calcium concentration (Figure 3) might indicate increasing capacity for production of CO₂ from a larger volume of thawed soil as thaw progresses seasonally (Hicks Pries et al., 2012), or release of calcium during weathering reactions. Calcium is sourced from deeper soil horizons (Keller et al., 2007), and concentration typically increases in water tracks as the thaw season progresses (Figure S3). Further, hydrologic models confirm increasing contribution of groundwater to flow in water tracks as the thaw season progresses (Evans et al., 2020). The observed negative correlation of CO₂ with precipitation could indicate that supply of CO₂ to water tracks is limited, if soil CO₂ is flushed from soils and depleted during storms. Suppression of ecosystem respiration by precipitation was also observed in wet sedge tundra (Euskirchen et al., 2012), which is characterized by saturated soils similar to water tracks.

N₂O in water collected from the surface was typically at equilibrium with the atmosphere, indicating either balanced production and consumption of N₂O or low rates of N cycling. However, drier sites and time periods resulted in net emission of N₂O to the atmosphere, providing evidence of active transformations of nitrate in tundra soils. A positive correlation of N₂O concentration with calcium concentration (Figure 3) indicated increased production or declining consumption of N₂O as the thaw season progressed and flow paths deepened. Rate of denitrification (i.e., reduction of nitrate to N₂) declines with depth in saturated tundra soils (Harms & Jones, 2012), which might suggest that less consumption of N₂O by denitrifiers occurs along deeper flow paths that are activated later in the thaw season. The observed negative relationship of N₂O concentration with DOC (Figure 3) further corroborates the importance of denitrification as a sink for N₂O, because heterotrophic denitrifiers produce less N₂O as a final product when organic C is more available (Firestone & Davidson, 1989). A strong positive correlation of N₂O concentration with mean daily PAR (Figure 3) is surprising, particularly in the absence of an association with GPP, but perhaps reflects a microbial response to supply of labile C by root exudates during peak growing season (Henry et al., 2008). Another possible explanation is that GPP within or near water tracks contrasts with the footprints of the nearby eddy flux towers used to quantify GPP, which do not include water tracks.

4.2 Vertical and Longitudinal Patterns

Greater concentrations of CH₄ and CO₂ in soil water than in surface water reflect production of these gases in soils and significant evasion from surface waters (Figure 4). Further, elevated concentration of CO₂ in August compared to June indicates greater production due to warmer soils, a larger volume of thawed soil supporting microbial activity, or increased availability of labile substrates. Laboratory experiments confirm temperature sensitivity of CH₄ and CO₂ production in tundra soils, as well as the influence of organic C composition and availability of alternative electron acceptors, including iron (Chowdhury et al., 2015; Herndon et al., 2019; Treat et al., 2014).

Spatially intensive observations of surface and soil water revealed N₂O concentrations that ranged from atmospheric equilibrium to undersaturation relative to the atmosphere, indicating active cycling of nitrate despite undetectable concentration of nitrate in surface and soil water. Undersaturation of N₂O at Water Track 1 suggested consumption of N₂O in soils (Figure 4), which might have resulted from lower redox potential at Water Track 1 than at Water Track 4, fostering complete denitrification (i.e., reduction of N₂O). These findings align with previous observations of active transformations of nitrate in arctic tundra soils, including denitrification in shallow, saturated soils (Harms & Jones, 2012); uptake of nitrate by upland plants (Liu et al., 2018); and net retention of nitrate in soil water of water tracks (Harms & Ludwig, 2016). Further, presence of microbial communities capable of transforming nitrate in undisturbed tundra might support large fluxes of N₂O upon disturbance of permafrost, such as when inorganic N becomes more available following thermoerosion of water tracks to form thermokarst gullies (Harms et al., 2014; Repo et al., 2009; Voigt et al., 2017).

Longitudinal patterns in C chemistry coincided with a downslope increase in discharge (Figure 5) and suggested that dissolved CO₂ is primarily derived from shallow soils. DOC yields from tundra soils are typically greatest from shallow, organic-rich horizons (organic soil depth ~20 cm in water tracks; Evans et al., 2020) and decline with depth due to declining organic content of soils, increased time for DOC degradation, and sorption of organic matter to mineral particles (Kawahigashi et al., 2006; Striegl et al., 2005; Uhlirova et al., 2007). The observed longitudinal decline in DOC concentration therefore suggests decreased contribution of shallow relative to deeper flow paths (Figure 5). An opposing, increasing downslope trend in calcium concentration further indicates that deeper flow paths interacting with mineral soils contribute disproportionately to downslope gains in discharge, because calcium concentration increases in deeper, less weathered soils of the study basin (Keller et al., 2007). Positive downslope trends in concentration of HCO₃⁻, also derived from mineral soils, and increasing ẟ¹³C-DIC, caused by mineral weathering or gas evasion (Giesler et al., 2013), indicate a corresponding increase in contributions of mineral, rather than plant-derived C. The downslope decline in dissolved CO₂ concentration thus suggests that CO₂ in water tracks is largely derived from organic soils and rapidly evades upon reaching the water track. The lower CO₂ concentration downstream might also indicate dilution due to inputs from mineral-derived flow paths. Overall, longitudinal patterns indicate that the routing of hydrologic flow paths through catchments influences emissions of CO₂ at the hillslope scale, consistent with findings in larger stream networks (Lupon et al., 2019; Rocher-Ros et al., 2019).

4.3 Water Track Contributions to Greenhouse Gas Budgets

We estimated that areal rates of CO₂ and CH₄ emissions from water tracks were up to 20 times greater than those of adjacent terrestrial tundra, rivers, and lakes. Areal rates of CO₂ emission from water tracks, based on modeled gas transfer velocities, were among the largest from aquatic ecosystems in permafrost-influenced ecosystems, and average fluxes of CH₄ from two water tracks were within the upper quartile of fluxes previously observed from Arctic aquatic ecosystems (Figure 6). Subsurface transport of gases from soils in the contributing area, where CO₂ concentrations are usually >1,000 ppm (Figure 4) and exposure to turbulence when water emerges as surface flow over steep slopes likely drives CO₂ emissions from water tracks. CH₄ fluxes were heterogeneous among water tracks (Figure 6), with emissions possibly reflecting the intersection of favorable redox conditions for methanogens with input of soil CO₂. Methods for estimating gas emissions varied among the data sets compiled from the literature (Figure 6), including contrasts in sample incubation times and equilibration methods, and approaches to modeling or measuring gas transfer velocities. Optimization of methods for accuracy and applicability in Arctic ecosystems would reduce uncertainty and increase relevance of dissolved gas emissions for regional and global greenhouse gas budgets.

Figure 6

CH₄ (top) and CO₂ (bottom) fluxes from permafrost-influenced aquatic ecosystems of the Arctic and sub-Arctic. Symbol shape denotes permafrost type. Values for water tracks represent the mean and bootstrapped 95% confidence intervals of daily flux estimates across 3 years of observation (n = 18–20). Data from the literature represent varying time periods or seasons, and observations of single sites or summaries of multiple sites as reported in the original publications.

Despite large rates of estimated gas emissions relative to other aquatic ecosystems, the small spatial extent of water tracks could constrain their contributions to the Arctic C budget, as observed for other aquatic features in the High Arctic (Emmerton et al., 2016). We scaled the estimated emissions of CO₂ and CH₄ from surface water within water tracks and compared them to terrestrial emissions to the atmosphere within the contributing area (Table 2). Calculated emissions of CO₂ and CH₄ were disproportionately large relative to the spatial coverage of water tracks. Surface flows in water tracks contributed ~4–6% of gross CO₂ emissions (i.e., ecosystem respiration) from the contributing area and 53–85% of the total emissions of CH₄, despite occupying 0.4–0.6% of catchment area (Table 2). Compared to the rate of net ecosystem exchange during the growing season estimated during the study period (2012–2014) for the contributing areas of WT1 and WT5 (3,833–8,417 mmol C day⁻¹; Euskirchen et al., 2017), estimated gaseous emissions of C from surface water of water tracks during the growing season (Table 2) thus decreased the estimated terrestrial C sink by 5–9%. Soils within water tracks can remain thawed until midwinter (Rushlow et al., 2020) when they might continue to convey water (Evans et al., 2020). Water tracks could therefore continue to emit greenhouse gases in autumn and winter, as documented in both upland and lowland tundra (Arndt et al., 2019; Euskirchen et al., 2017), enhancing the net flux of C to the atmosphere that occurs outside of the growing season.

Table 2. Scaled Estimates of CH₄ and Gross CO₂ Emissions Within the Contributing Areas of Two Water Tracks

	Catchment component	Area (m^2)	CH4 flux (mol day^−1)	CO2 flux (mol day^−1)
WT1 (wet sedge)	WT surface water	557 (0.6%)	9.35 ± 11.19 (53.4%)	403 ± 268 (3.7%)
	dry WT	5,443 (6.1%)	7.37 (42.4%)	791 (7.3%)
	non-WT hillslope	83,000 (93.3%)	0.73 (4.2%)	9,658 (89.0%)
WT5 (shrub)	WT surface water	195 (0.4%)	0.19 ± 0.32 (84.7%)	201 ± 138 (5.9%)
	dry WT	2,430 (5.0%)	0.02 (0.8%)	283 (4.7%)
	non-WT hillslope	46,375 (94.6%)	0.41 (14.6%)	5,396 (89.4%)

• Note. Percentages are relative to total area or mean flux within each contributing area. Uncertainties in fluxes for surface water are shown as 1 SD of mean daily estimates across 3 years (2012–2014). Surface area of moist acidic tussock tundra comprising the contributing area to the water track, dry water track (wet sedge in WT1 and tussock tundra in WT5), and inundated water track (i.e., surface water) were calculated from channel width and length, contributing area (Rushlow & Godsey, 2017), and average wetted width of the water tracks (Harms et al., 2019), the latter of which were available for two water tracks. Estimates of CO₂ emission from ecosystem respiration (i.e., gross flux of CO₂ to the atmosphere) in moist acidic tussock tundra (non-WT hillslope, dry WT of WT5) and wet sedge tundra (dry WT of WT1), as well as CH₄ from wet sedge tundra, were made by the eddy covariance technique in the adjacent Imnavait basin during the same time period as observations from the water tracks (June–August 2012–2014; Euskirchen et al., 2017). Estimates of CH₄ flux from moist acidic tussock tundra were made in static chambers at nearby Toolik Lake (June–August 2012; Blanc-Betes et al., 2016). Note that plant uptake of CO₂ exceeds ecosystem respiration from these ecosystems during summer.

Use of modeled gas transfer velocities (k) contributes uncertainty to our estimates of gas flux. The estimated gas transfer velocities applied here (Table 1) are within the range of estimates derived from tracer studies in small, steeply sloped channels. A recent compilation yielded a median gas transfer velocity of 11.6 m day⁻¹ for discharge <10 L s⁻¹ and channel slope 0.05–0.23 m m⁻¹ (Ulseth et al., 2019) and small, steep boreal streams produced a median rate of 17.6 m day⁻¹ for discharge <14 L s⁻¹ and slope 0.02–0.07 m m⁻¹ (Wallin et al., 2011). Median rate of gas transfer velocity of 15.5 m day⁻¹ for discharge <10 L s⁻¹ and slope 0.06–0.19 m m⁻¹ in water tracks is within the ranges of these previously measured rates. We further assessed the reliability of estimated fluxes by comparing them to the deviation between expected and observed concentrations of CO₂ along the longitudinal transect at Water Track 1. We estimated the dissolved CO₂ concentration expected from upstream and lateral inputs and estimated CO₂ loss by difference from the observed concentration (Figure S5). Observed CO₂ concentrations were consistently less than expected, which indicates that losses of CO₂ in addition to downstream export contributed to the observed downstream decline in CO₂. Indeed, the average areal rate of this CO₂ loss (645 mmol C m⁻² day⁻¹) compares closely with the mean estimated CO₂ flux to the atmosphere from Water Track 1 (719 mmol C m⁻² day⁻¹; Table 1). There could be other CO₂ losses and sources that are not accounted for in this exercise, such as CO₂ consumption by methanogenesis, or production of CO₂ within soils of the water track. However, estimates of gas transfer velocity from hydraulic relationships do not account for turbulence contributed by emergent vegetation, and thus, estimates of gas transfer velocity and resulting gas emissions from water tracks are likely conservative.

We conclude that relative to their small spatial extent, flowing waters within water tracks contribute disproportionately to emission of greenhouse gases from Arctic landscapes during the growing season. Large areal rates of gas emission from water tracks might arise due to their receiving location on the landscape, extended thaw season, and strong hydrologic connection to soils, forming a convergence zone for water, heat, dissolved gases, nutrients, and substrates (Rushlow et al., 2020). Gas emissions from water tracks might be expected to increase over longer timescales, because feedbacks among warming soils, snow accumulation, and subsidence of the ground surface cause expansion of water tracks (Osterkamp et al., 2009). In addition to increasing the areal coverage of water tracks, maturing channel networks support increased runoff within the network (Evans & Ge, 2017; Godin et al., 2014; Liljedahl et al., 2016). Thus, we predict that development of water track networks will increase the relative importance of gaseous C fluxes from fluvial networks in the Arctic C budget during periods of ground thaw and network expansion.

Acknowledgments

We gratefully acknowledge A. Bjärhall, K. Blake, C. Cook, R. Giesler, M. Jaeger, E. Longano, S. Ludwig, R. Risser, C. Rushlow, and M. Väisänen for their contributions to data collection. Alex Webster provided comments on the manuscript. We thank staff of CH2MHill Polar Services and Toolik Field Station for logistical support. This research was supported by the U.S. National Science Foundation (OPP-1108200); the International Network for Terrestrial Research and Monitoring in the Arctic (INTERACT, Grant Agreement No. 730938 funded by the EU); the Swedish Research Council (2013-5001); and the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (2014-00970). We also acknowledge six anonymous reviewers and editors for comments that improved the manuscript.

Open Research

Data Availability Statement

Data are available from the Arctic Data Center (https://arcticdata.io/catalog/view/doi:10.18739/A23R0PS5P).