Patterns and isotopic composition of greenhouse gases under ice in lakes of interior Alaska

Study lakes are located in the YFB, a sub-basin of the broader Yukon River Basin (figure 1). Detailed site descriptions are available in Bogard et al (2019). Lakes were selected to cover the range in environmental conditions observed in the region. Briefly, lakes sampled in the YFB region are situated in a semi-arid, flat landscape. The YFB is a broad plain of poorly drained deposits with a high-water table and discontinuous permafrost, and lake chemical and physical conditions range greatly due to heterogeneous alluvial deposits (Heglund and Jones 2003). Most lakes in the region are rich in nutrients and naturally eutrophic or hypereutrophic (Heglund and Jones 2003), and are relatively productive (Bogard et al 2019). Average landscape slopes ranged from 0 to 2.2%, with annual precipitation spanning from 174 to 403 mm yr⁻¹ and soil C content from 33.6 to 64.2 kg m⁻² (to 3 m depth). Vegetation in the region is heterogeneous, with spruce, willow, alder and shrubs and mosses are distributed differently along gradients of topography, hydrology, geology and soil structure, and disturbance by fire and thermokarst (Jorgenson et al 2013).

Zoom In Zoom Out Reset image size

A total of 13 lakes were sampled by float plane to measure under ice conditions in April 2017. Sampling reflects late winter conditions, since ice-cover typically initiates in October to November, and ice melt begins in May (late May, in 2017). Water temperature, pH, dissolved oxygen (DO) concentrations and fraction saturation were measured with a Yellow Springs Instruments EXO II probe. Sensors were calibrated before and after sampling and no drift in outputs were observed. Lake surface elevations were obtained from Google Earth. Lake depth values were taken from Bogard et al (2019), as the maximum depth at sampling sites. Water from pelagic regions of each lake was collected for laboratory analyses and was pumped across a filter on site (0.45 μm capsule filters, Geotech Environmental Equipment). Samples for analysis of deuterium and oxygen isotopic composition of water were stored in the dark in 20 ml plastic bottles, filled to capacity free of air. Samples for dissolved CO₂ and CH₄ stable isotopic composition were collected in 60 ml syringes using the headspace equilibration technique following standard methods (Striegl et al 2012). The partial pressures of CO₂ and CH₄ in the water were measured on site using a Los Gatos Research (LGR) Ultraportable Greenhouse Gas Analyzer. Gases were stripped from water by pumping water continuously through a closed showerhead system. Equilibrated gases were circulated into the LGR in a closed loop system following the general setup of (Webb et al 2016).

Filtered water was analyzed for DOC following (Aiken 1992). Ultraviolet absorbance and SUVA₂₅₄ were measured following (Weishaar et al 2003). Samples for the isotopic composition of water were analyzed at the U.S. Geological Survey Reston Stable Isotope Laboratory in Reston, VA, USA, using standard methods (Reéveész et al 2008). Isotopic composition of dissolved gases was analyzed at the University of Washington using a Picarro G2201-I Isotopic Analyzer. Triplicate samples were collected and are reported here as average values. Equilibrated gas samples were directly injected into the Analyzer using the small sample introduction module 2. In some cases, samples with extremely high gas concentrations were diluted with ultra-pure Nitrogen (N₂) to reach target concentrations. After testing to confirm no effect on isotopic signatures, N₂ was also added to some samples in cases where sample volumes were less than the required 17 ml of gas. Standards of CO₂ (31.48‰) and CH₄ (−38.3 ‰) were injected prior to sample processing. Standards were also injected at the end of some runs, confirming that instrument drift did not occur. A systematic offset in δ¹³C-CH₄ was corrected using a two-point calibration curve (y = 0.9578x—0.1791; r² = 1).

Stable isotopic composition of both O and C are reported in the text using delta notation, in equation (1):

$$\begin{equation}{\delta ^{18}}O{\text{ }}or{\text{ }}{\delta ^{13}}C = \left( {\frac{{{R_{sample}}}}{{{R_{standard}}}}} \right) {-1}\end{equation} \tag{ 1 }$$

where R_sample and R_standard are ratios of heavy to light isotopes (¹⁸O:¹⁶O or ¹³C:¹²C) in samples (R_sample) and in standards (R_standard) of Vienna standard mean ocean water or Vienna Pee Dee Belemnite.

To depict lake positions in the landscape, we developed a map of the YFB region using ArcGIS 10.4 (ESRI). All statistical analyses were performed using R version 3.3.3 1.1.463 (R Development Core Team 2017). To assess the relationships between gas content, isotopic composition, and environmental properties (including lake depth, elevation, and physico-chemical properties), we first tested for normality using the Shapiro Wilks test, and used Spearman correlation analyses to evaluate relationships for variables with non-normal distributions. The Pearson correlation test was used on log₁₀-transformed data when comparing CO₂ and CH₄ concentrations, and on comparisons with normal data distributions. To explore which methanogenic pathway dominated in lakes, we used Keeling plot and Miller Tans plot approaches to evaluate the approximate isotopic composition of freshly produced CH₄ prior to oxidation in the lakes under ice (Pataki et al 2003, Miller and Tans 2003, Zobitz et al 2006). The Keeling plot involves using isotopic values and CH₄ concentrations to find an isotopic source signature via the y-intercept. With the Miller-Tans plotting approach, the slope of the regression model obtained by plotting the product of δ¹³C and the concentration of either gas provides the average isotopic signature of CH₄. Regression relationships were calculated using a major axis regression approach with the 'lmodel2' package (Legendre 2013), because both x and y variables are subject to measurement error. Based on this regression approach, two distinct lake groups were identified using the Keeling and Miller-Tans plots, and the between-group differences in CH₄ sources were summarized with box plots and compared statistically with student t-tests.

The Keeling plot method is typically used in simple systems with initial certain background δ¹³C-CH₄ and a single source added over time, and therefore not entirely transferable to cross-system surveys with variable sources and sinks of CH₄. As such, the Keeling plot was only used to visually assess isotopic patterns. The Miller-Tans approach is more robust in cross-system applications because the test accounts for inconsistencies in background constants, and incorporates uncertainties and errors in the collection and analysis process (Miller and Tans 2003). From this approach, we cautiously interpreted potential differences in slopes (i.e. source CH₄ isotopic composition) among identified lake groupings. To ensure these applications were as accurate as possible, we restricted our explorations to periods of ice cover for two key reasons that simplify our interpretation of sources and sinks, and thereby make this approach more appropriate: First, during lake capping by ice, CH₄ from individual loss pathways (diffusive, ebullitive) are trapped and homogenized in the water column (Elder et al 2019), enabling clear source identification. Second, sampling during ice cover eliminated the second potential removal pathway of CH₄ (atmospheric loss), and thus we assume only microbial oxidation removed CH₄ from lakes. This simplification of source and sink pathways helps to ensure our interpretation of isotopic composition was as accurate as possible. Still, we only qualitatively interpret these outcomes (i.e. using lake groupings).

Across our 13 sample lakes, partial pressures of CO₂ and CH₄ under ice were not normally distributed (figure 2, table 1), with CH₄ having a more skewed distribution than CO₂. CO₂ content ranged from 454 to 19 000 μatm with a mean of 6673 μatm and a median of 5300 μatm, and CH₄ content ranged from 1.60 to 29 000 μatm with a mean of 7040 μatm and a median of 1152 μatm. We observed a weak positive correlation between the log₁₀ transformed CO₂ and CH₄ (figure 2; p= 0.18, r= 0.39). This relationship improved with the exclusion of an extreme outlier (Greenpepper Lake, Lake 6) (r= 0.72, p = 0.009). We present the data, however we suspect that this lake reflects a landscape position unique from others in this dataset, where it is perched within higher elevation conditions along the northern slopes of the White Mountain range. Furthermore, as outlined in Bogard et al (2019), this lake is shown to have the oldest mean DOC age (i.e. most depleted Δ¹⁴C signatures of DOC), and the most enriched values for the isotopes of water (table S1 (stacks.iop.org/ERL/15/105016/mmedia)), suggesting very little connection to the adjacent terrestrial and aquatic landscape.

Zoom In Zoom Out Reset image size

Like many other studies in the northern circumpolar region, all lakes studied here were supersaturated in CO₂ and CH₄ (Elder et al 2018, Serikova et al 2019), but the range in under ice gas concentrations was among the most extreme in the literature. The pCO₂ range across the lakes (454 to 19 000 μatm) spans beyond the range reported in a recent comprehensive survey of under ice pCO₂ in Swedish lakes (median = 2961, 95% C.I. = 919 to 11 162 μatm (Denfeld et al 2016). Yet Striegl et al (2001) found that lakes in north-temperate U.S. had under ice pCO₂ values reaching as low as 85 μatm, indicating that the low-end values observed in YFB lakes is not anomalous, and within the expected range of late winter CO₂ content. All but one lake had CO₂ concentrations exceeding CH₄, which is a general trend in many northern lake regions and across distinct seasons (Anthony et al 2014, Rasilo et al 2015, Elder et al 2018, Denfeld et al 2018). The range in under ice pCH₄ was also extreme, spanning 1.8 to 29 000 μatm. Very wide ranges in under ice CH₄ content have been shown for other Alaskan lakes (Kling et al 1992, Sepulveda-Jauregui et al 2015, Elder et al 2018). Yet, CH₄ content was heterogeneous (more so than for CO₂) with only four lakes having concentrations >5000 μatm (figure 2). Heterogeneity of under-ice CH₄ content is commonly observed, making it difficult to predict patterns and controls on winter CH₄ accumulation (Denfeld et al 2018) as opposed to open water periods, where ice-free ebullition rates scale with incoming solar radiation (Wik et al 2014). Overall, despite the small regional area covered in our study, we still observed GHG gradients similar to those in more extensive geographic surveys in other regions (Townsend-Small et al 2017, 2018, Elder et al 2019). This heterogeneity, even within a localized survey of lakes, underscores the need to better define the mechanistic controls of GHG content in lakes in winter.

Across an elevation range from 102 to 303 m asl, we observed a strong, non-linear relationship between GHG content and lake elevation. The highest concentrations of CO₂ and (to a lesser extent) CH₄ were found in lakes located below 150 m asl, situated closer to the Yukon River (figures 3(a) and (b); CO₂: p = 0.03403, ρ = −0.6; CH₄: p = 0.1516, ρ = −0.42). Only three lakes were located at elevations >150 m asl, though none of these sites had CO₂ or CH₄ content above ∼5000 μatm (figures 3(a) and (b)). In addition to elevation, we also examined hydrologic connectivity using δ¹⁸O-H₂O as a proxy. Overall, many of the lakes exhibited enriched values of δ¹⁸O-H₂O, from −18.67 to −10.36‰, indicating the dominance of evaporative processes in many lakes during ice free periods. Lakes with stronger connections to hydrologic networks (rivers, groundwater) are more depleted in δ¹⁸O-H₂O with values corresponding to approximately −21‰ for river and groundwater and −27‰ for snow and permafrost (Anderson et al 2013, Halm and Griffith 2014). The enriched isotopic values observed here further indicate the lakes are weakly connected to hydrologic flows, we found no clear relationship between δ¹⁸O-H₂O values and either pCO₂ (p= 0.28, ρ= −0.32) or pCH₄ (p= 0.64, ρ = 0.14).

Zoom In Zoom Out Reset image size

We observed a wide range of relationships between under-ice gas content and the limnological properties of each study lake (figure 4). Maximum depth at the sampling site (Z_max) ranged from 0.9 to >12 m (figures 4(a) and (b)), and GHG content scaled inversely with depth (CO₂: p = 0.0005, ρ = −0.83, CH₄: p = 0.006, ρ= −0.72). Concentrations of CH₄ were ∼ 5 orders of magnitude higher in lakes that were <2 m deep, while pCO₂ among lakes scaled more linearly with depth. DO concentrations in the lakes ranged from near anoxia to near saturated (0.9 to 9.56 mg L⁻¹; figures 4(c) and (d)). For both CO₂ and CH₄, concentrations were much greater at lower DO content (CO₂: p = 0.003, ρ = −0.78, CH₄: p = 0.009, ρ= −0.71). Yet this relationship was less linear for pCH₄, which declined dramatically in environments with >2 mg L⁻¹ DO. Consistent with an earlier survey of northern lakes (Kortelainen et al 2006), we found no relationship between organic matter and CO₂ content. There was a wide, cross-lake DOC concentration range (14.6 to 181 mg L⁻¹), and although no relationship to pCO₂ was observed, lakes with the greatest DOC concentrations tended to have the highest pCH₄ values (figures 4(e) and (f)); (CO₂: p= 0.34, ρ= 0.29; CH₄: p =.004, ρ= 0.75). Not only the quantity, but the composition of dissolved organic matter ranged widely among lakes, as indicated by SUVA₂₅₄ values spanning from 0.99 to 4.00 (figures 4(g) and (h)), yet values were uncorrelated with CO₂ or CH₄ content (CO₂: p= 0.29, r =0.31; CH₄: p= 0.91, r =0.03). When compared to pH (not shown), pCH₄ showed no relationship (p = 0.26), and the weak correlation with pCO₂ (p = 0.02, r = −0.64) likely reflected the direct effect of elevated CO₂ concentrations on pH. Although algal biomass as chl a pigment concentration does not capture total ecosystem productivity (i.e. macrophyte and benthic algal growth), it provides a general indication of the wide range in trophic status of our study lakes (figures 4(i) and (j)). While not correlated with lake pCO₂, pCH₄ tended to increase in more productive lakes with higher chl a concentration.

Zoom In Zoom Out Reset image size

Overall, lake surface elevation (dictating the general position in the surveyed landscape) and depth appear as important controls determining in which lakes elevated CO₂ and CH₄ content is observed. It is likely that these controls over-rode the influence of factors of secondary importance, since the three most shallow lakes (lakes 5, 10, and in some cases 2; Table S1) appeared as outliers with extremely high GHG concentrations in relationships between GHG content and other predictors (DOC, SUVA₂₅₄, chl a concentration). This importance of landscape position in structuring GHG content has been demonstrated in global-scale analyses (Lapierre et al 2012) and makes sense at the regional scale in the YFB, first because lake depths typically become shallower moving from steeper, high elevation terrain, to flatter, low elevation terrain (Heathcote et al 2015), and second because shallow lakes tend to have greater content of CO₂ and CH₄ under ice (Ducharme-Riel et al 2015b, Denfeld et al 2016, 2018). The absence of a correlation between GHG content and water isotopic signatures reflects the fact that many lakes in the YFB region have weak hydrologic connectivity to riverine or groundwater flows, and receive comparatively little terrestrial C input (Johnston et al 2019). The weak or absent relationship between overall DOC aromaticity (as SUVA₂₅₄) and CO₂ and CH₄ (respectively; figure 4) is in line with the independence of under-ice GHG content from external C loading in many of the lakes (Bogard et al 2019). While elevated lake GHG content was primarily observed at low elevations, CH₄ content was additionally dependent on lake trophic status (chl a content), and DOC availability, which likely enhanced winter respiration and oxygen depletion (figures 4(d) and (j)). These secondary controls likely made under ice pCH₄ content even more heterogeneous at low elevations than pCO₂. Taken together, the variability in under ice GHG content appears to track general environmental characteristics (lake surface elevation and depth) and (for CH₄) lake trophic status, which are predictable features that could be used in future studies (with more extensive lake sampling) to build broader regional models of under ice GHG content and spring emissions at ice-out, as has been done in better-studied boreal landscapes (Guo et al 2020).

When compared to partial pressures, the δ¹³C values of each gas revealed distinct cross-lake patterns for both CO₂ and CH₄ (table 1, figures 5(a) and (b)). Values of δ¹³C-CO₂ were more constrained, ranging from −10.13 to −23.02‰, and showing no relationship to gas concentration (figure 5(a); p= 0.86, ρ= −0.05). Conversely, values for CH₄ ranged widely, from −19.59 to −71.87‰, and pCH₄ was correlated non-linearly to isotopic composition (figure 5(b); p= 0.051, ρ= −.55), with isotopic values most enriched, and increasing sharply at concentrations below ∼2400 μatm. We found no correlation between the isotopic composition of CO₂ and CH₄ (figure 5(c); p = 0.67, ρ = −0.13).

Zoom In Zoom Out Reset image size

We used Keeling and Miller Tans plots (Keeling 1958, Miller and Tans 2003, Campeau et al 2017) to explore the potential source isotopic signature of fresh CH₄ prior to microbial oxidation. CH₄ is largely produced through two main pathways in freshwater ecosystems, where hydrogenotrophic methanogenesis generates CH₄ with δ¹³C values from approximately −90‰ to −30‰, while the acetoclastic-derived source CH₄ may range from −30‰ to −10‰ (Claus et al 2009). As detailed in the methods section, both plotting approaches use the concentration of methane and the δ¹³C-CH₄ to predict the isotopic signature of source CH₄. Using the Keeling approach (figure 6(a)), we plotted δ¹³C-CH₄ values from each lake against the inverse of respective gas concentrations. Inverse values were log-transformed due to the extreme range and non-normal distribution of concentrations, therefore the intercept value of the regression model associated with the Keeling plot was not interpretable as the exact source value. Thus, we used this method in a more qualitative manner to compare lake categories. Major Axis (type-II) regression models applied in the Keeling plot showed two distinct groups of lakes (figure 6(a) group A: y = 16.48x-27.2, r²= 0.91, p = 0.0008; group B: y = 7.7x-63.5, r²= 0.96, p = 0.0006). Next, we used the Miller-Tans plotting approach to identify source isotopic signatures for CH₄ in the two distinct lake groups identified with the Keeling plot. Here, the value of δ¹³C-CH₄ was multiplied by the concentration of CH₄, then compared to CH₄ concentrations (figure 6(b)). Consistent with the Keeling plot groupings, lake distributions fell into two distinct groups (group A: y = −58.96x + 55.29, r²= 0.99, p <0.00001; group B: y = −71.8x + 2.2 29, r²= 0.99, p <0.00001).

Zoom In Zoom Out Reset image size

These two lake groupings in the Miller-Tans plots (figure 6(b)) indicate that lakes in the YFB appear to have a range of inputs from both acetoclastic and hydrogenotrophic pathways, with lakes in group B predominantly generating CH₄ from the hydrogenotrophic pathway, and lakes in group A reflecting a mix of both CH₄ source input pathways and increased relative importance of acetoclastic methanogenesis. This wide range in isotopic composition of δ¹³C-CH₄ is consistent with other northern lake studies that showed a mix of pathways possibly supporting lake CH₄ accumulation and processing via oxidation. In Eastern Canadian Arctic thermokarst lakes, hypolimnetic δ¹³C-CH₄ signatures also ranged across lakes from −47.1 to −67.0 ‰, likely sourced from multiple pathways (Matveev et al 2018). In an unrelated study of Alaskan lakes, Elder et al (2018) used similar isotopic methods to show that under ice δ¹³C-CH₄ ranged from −38.8‰ to −73.3‰. Acetoclastic methanogenesis uses acetate as a reactant to produce CH₄ and CO₂. Hydrogenotrophic methanogenesis produces CH₄ through CO₂ reduction. Therefore, differences in net CO₂ production and consumption linked to methanogenesis in lakes of groups A and B could hypothetically alter the overall winter GHG budgets of the lake groups. Although there was no clear distinction in GHG content between the two groups of lakes with distinct isotopic signatures of fresh CH₄ (i.e. signature prior to oxidation), generally the highest concentrations of CO₂ and CH₄ were found in those lakes that appear to produce CH₄ primarily through the acetoclastic pathway, most notably in lakes 5 and 10 (figure 6(b)).

We categorized the lakes by their identified groups (A, B; figure 6), and explored whether lake and landscape features differed among groups. Lakes with isotopically depleted source CH₄ dominated by hydrogenotrophic methanogenesis were found across the elevational gradient but on average, were located at higher elevation than lakes with enriched source CH₄ (figure 7(a); t = −9.3, p < 0.0001). They were also often deeper (figure 7(c); t = 5.6, p < 0.0001), more oxygenated (figure 7(d); t = 6.1, p < 0.0001), had somewhat lower chl a concentrations (figure 7(g); t = 2.6, p = 0.02), and higher pH (figure 7(h); t = −31.5, p < 0.0001). While DOC concentrations were not different among groups (figure 7(e); t = −0.9, p = 0.39), and SUVA₂₅₄ was uncorrelated to pCH₄ across lakes (figure 4(h)), we found a significant difference in SUVA₂₅₄ among the two lake groups. Group A lakes (isotopically enriched source CH₄) had less aromatic organic matter, as inferred from somewhat lower SUVA₂₅₄ values (figure 7(f); t = −3.7, p = 0.001). The differences among both groups in terms of hydrologic connectivity were statistically significant, but variable (figure 7(b); t = 25.9, p < 0.0001). Taken together, it appears that the dominance of individual methanogenic pathways may vary in a predictable way during winter months for lakes in the YFB region. These landscape patterns in CH₄ isotopic composition (figure 7) generally makes sense, since hydrogenotrophic methanogens are often more O₂-tolerant than obligate anaerobe acetoclastic methanogens (Oremland and Capone 1988, Zehnder 1988, Whiticar 1999), and declining pH can favor acetoclastic over hydrogenotrophic pathways (Conrad 2020). Differences in SUVA₂₅₄ and chl a concentration also may reflect the fact that more productive environments rich in bio-labile organic matter can support higher relative rates of acetoclastic methanogenesis, while decomposition of complex aromatic materials favours hydrogenotrophic pathways (Conrad 2020).

Zoom In Zoom Out Reset image size

The winter ice-covered period represents an extended period of time for which we currently have a weak understanding of northern lake C biogeochemistry (Denfeld et al 2018). Here, we show that the patterns of winter under ice CO₂ and CH₄ content are heterogeneous across a suite of boreal lakes spanning a landscape gradient in interior Alaska, and that divergent stable isotopic signatures of CH₄ indicate a potential contribution from different biogeochemical pathways. Therefore, these observations fill a critical gap in our understanding of C cycling in remote and understudied arid, flat, and circumpolar lake landscapes (Bogard et al 2019). The content of GHGs under ice in winter is extremely variable, with concentrations ranging from near- atmospheric equilibrium to extreme supersaturation. The degree of inter-lake variability in winter CO₂ content shown for the YFB region is consistent with observations from other boreal and temperate lakes (Ducharme-Riel et al 2015a). This variability may lead to a wide range in importance (in an annual context) of early-spring emissions pulse at ice out, from one lake to the next, though such an exercise is beyond the scope of this research. While this variability complicates our ability to predict region-wide emissions pulses at ice out, the observed correlations between lake properties (Z_mix, DO, DOC, SUVA), and catchment properties (elevation, δ¹⁸O-H₂O) suggests there is potential for future studies to empirically model lake CO₂ emissions at ice-out in the YFB or similar landscapes. We also found a comparably wide range in CH₄ concentrations in YFB lakes. Here, the much weaker relationship between CH₄ and lake or catchment properties suggests that the construction of comparable models to predict the CH₄ pulse at ice out will be more difficult for the YFB region (but see Guo et al 2020). More detailed sampling is likely needed to constrain these relationships. The fact that under ice CO₂ and CH₄ content was positively correlated in most lakes, and gas content was highest at lowest elevation in the region, provides a first step to constraining the potential locations for emissions hotspots during spring thaw. Overall, because of the limitations on sampling in the winter in these remote regions (mostly due to lack of access), our results add critical data to fill a major gap in our understanding of winter limnology and C biogeochemistry.

To date, few studies have employed stable isotopic surveys of lake GHGs that build up under ice, especially for CH₄ (Townsend-Small et al 2017, 2018, Elder et al 2019). Here we demonstrate that even within a small geographic region, the importance of individual biochemical pathways supporting winter CH₄ content and emissions at ice-out can vary greatly from one lake to the next. Such differences in methanogenic pathway dominance may have consequences for the overall radiative effect of lake emissions, because acetoclastic and hydrogenotrophic methanogenesis pathways respectively consume and produce CO₂. While our survey provides a first-order exploration of potential differences in the relative importance of these pathways, more detailed isotopic and microbial studies are needed to constrain the overall contributions from these distinct pathways, relative to oxidation of CH₄. Furthermore, the extreme range in isotopic signatures of CH₄ under ice, even in the limited geographic region surveyed here, indicate that it will be exceedingly difficult to distinguish lake emissions from other sources (terrestrial, riverine) at spring thaw using airborne technologies and top-down isotopic surveys.

There remains much to be learned regarding the patterns and controls of under ice GHG accumulation in boreal and arctic lakes. More research must be done to formally construct accurate C budgets (Denfeld et al 2018), and to better define the mechanisms regulating net GHG accumulation. Here, our study was designed to target the latter issue. Our isotopic evaluation of CH₄ composition, and our correlations between under ice GHG content and environmental variables both fill an important research gap related to winter C cycling in lakes from this understudied arid, low-relief region that is representative of up to a quarter of northern high-latitude lake landscapes worldwide.