A pan-Arctic synthesis of CH₄ and CO₂ production from anoxic soil incubations

Data collection and site classification

We compiled a database of anaerobic incubations that were conducted on soil samples within the geographic permafrost area (Fig. 1, Table 1; Brown et al. 1998, revised 2001). We identified potential published studies for inclusion in the database using Web of Science with the keywords ‘anaerobic or anoxic or methane or incubation’ and ‘arctic or boreal or permafrost’, and we actively solicited contributions of published and unpublished studies through the NSF-funded Vulnerability of Permafrost Carbon Research Coordination Network as a parallel study to Schädel et al. (2014). Twenty studies (16 published, 4 unpublished) fell within the northern high-latitude permafrost area and met the following criteria: (i) created anaerobic headspace using oxygen-free gas (most commonly N₂ or He); (ii) measured CH₄ on a per gram soil dry weight or per gram C basis; (iii) reported sample depth and incubation temperature. We included soils with either intact or homogenized soil structure but excluded soil slurries and incubations of lake sediments due to difficulties of comparing methods. These efforts resulted in a dataset of 2270 CH₄ production measurements across 54 sites in Alaska (USA), Canada, Sweden, and Siberia (Russia) between 1984 and 2013 (Fig. 1). We extracted CH₄ and CO₂ production data from published studies using a data extraction program (Plot Digitizer), calculated the molar CO₂:CH₄ production ratio, and collected ancillary data that were commonly reported across studies.

We extracted information about site type and local conditions from the published studies and unpublished datasets. We classified sites by biome (tundra or boreal) and landscape position (drained lake basin, active floodplain, wetland, lowland, upland) based on site descriptions. The wetland classification included bogs, fens, mires, marshes, swamps, and polygonal tundra, while the lowland classification included low-lying black spruce forests without standing water. We classified dominant vascular vegetation within each landscape position at each site into graminoid, shrub, and tree categories using vegetation descriptions. Moss cover for each site was classified into three categories: no moss, dominance of Sphagnum mosses, or dominance by other mosses (e.g., Polytrichum spp., Drepanocladus spp.), based on site-level vegetation description. We also categorized water table position of each soil sample based on sample depth relative to reported water table depth. The three water table categories were dry (>2.5 cm above the mean water table depth), fluctuating (mean water table fell within 2.5 cm of the sample depth), and inundated (mean water table depth was within the sample depth). We classified soil type according to % soil C; soils with >20% C were classified as organic soil, while soils with <20% C were classified as mineral. Two samples with cryoturbation features were classified as a mix of organic and mineral soil (O/M). We also classified sites using two categories for permafrost: permafrost present/absent and active layer/permafrost layer. Permafrost present/absent included measurements only from active layer soils when permafrost was present, and the active layer/permafrost layer comparison excluded measurements from sites without permafrost. Finally, we extracted mean annual air temperature and mean annual precipitation (1950–2000) from WorldClim database (Hijmans et al., 2005). Vascular vegetation, moss cover, and water table position were not classified for samples from >1 m depth, and these samples were omitted from the vegetation and water table position analyses (N = 18).

Data assimilation and data processing

The anaerobic laboratory incubation studies ranged in incubation length from 1 to 1238 days, incubation temperatures ranged from −0.5 °C to 35 °C, and sample depth ranged from the soil surface to 10 m. There were often multiple C production measurements collected during an incubation for each sample (i.e., a unique site × depth × incubation temperature combination). Instead of using multiple production measurements taken over multiple days in the analysis, maximum CH₄, mean CO₂ production rates, and mean daily CO₂:CH₄ production ratios prior to 150 days were used as a single measurement for each sample (see Supplemental Materials), hereafter referred to as the aggregated dataset. We used maximum CH₄ production rate from each sample instead of mean production (as we did for CO₂) because different environmental conditions and potential differences in microbial communities resulted in varying lag times for CH₄ production that, in part, drove mean CH₄ differences. Similarly, lag time effects precluded us from obtaining relevant cumulative CH₄ production rates. We excluded daily production measurements that showed CH₄ uptake (consumption) from our calculations of the mean daily CO₂:CH₄ production. After finding mean and maximum CH₄ and CO₂ production rates for each sample, we identified 44 samples in the aggregated dataset that showed no CH₄ production over the course of the incubation; all of these samples were removed from our study either because the incubation time (<7 days) was too short (33 samples) to recover from disturbance to the microbial communities due to sampling, transport, and handling (Nilsson & Oquist, 2009) or due to insufficient replication to validate the zero measurement (11 samples). This resulted in 303 measurements of CH₄ production and 219 anaerobic CO₂ production measurements in the aggregated dataset.

The dataset was highly unbalanced with regard to temperature, a known control on CH₄ and CO₂ production rates. We did not standardize measurements for incubation temperature using a Q₁₀ derived from the full dataset because of the high variability associated with the calculation; Q₁₀ derived from individual studies with multiple incubation temperatures (Table 1) ranged from 0.96 to 3.10 for CH₄ production (median: 1.16) and from 0.67 to 4.10 for anaerobic CO₂ production (median: 1.39). Instead, we included temperature as a covariate in our statistical analysis.

To directly compare the potential radiative forcing of CH₄ and CO₂ production, we calculated the CO₂ equivalent (CO₂e) of CH₄ production using a GWP of 28 kg CO₂ equivalents kg⁻¹ CH₄ over a 100-year time horizon (Myhre et al., 2013). While some of the CH₄ produced under field conditions is oxidized before reaching the atmosphere (and thus has a GWP of 0), the calculation of CO₂e allows us to compare the warming potential of these two greenhouse gases.

We estimated % C and bulk density values using an empirical relationship and calculated CO₂ and CH₄ production per gram soil C. Percent C was not reported for ~ 30% of measurements, and bulk density was not reported for ~ 70% of measurements. We estimated missing % C and bulk density values using an empirical relationship between depth, soil type, and % C or bulk density (Table S3, Fig. S5). We calculated missing CH₄ or CO₂ production per gram C from the reported C production per gram dry weight to investigate the relative differences in C quality among samples. We use the term, C quality, to refer to decomposability as determined from laboratory incubations, as has been performed in a recent synthesis of aerobic soil incubations (Schädel et al., 2014). While we mainly present results on a per gram C basis, results per gram dry weight followed similar trends. Carbon production rates on an areal basis were calculated to a 1 m depth using maximum CH₄ and mean CO₂ production per gram dry weight soil from the aggregated dataset and measured or estimated bulk density (see Supplemental Methods).

Lag analysis of CH₄ production

To understand dynamics of CH₄ production over time, we analyzed the lag time from the start of the incubation to maximum rates of CH₄ production, hereafter referred to as lag time. Lag time integrates both redox potential and microbial community dynamics within the soils, as maximum rates of CH₄ production are achieved under conditions when alternate electron acceptors are depleted and methanogen communities are established (Knorr & Blodau, 2009; Nilsson & Oquist, 2009). Microbial communities, and thus lag times, can be affected by disturbance when sampling, handling, and processing during the experimental setup. While the level of disturbance may vary across studies as a result of methodological differences, we do not expect systematic differences across sample or site types, which is the focus of our analyses. Lag time was calculated using a subset of the data with three or more CH₄ production measurements during the incubation experiment (n = 120; Fig. S3).

K-value decay rate calculation

In addition to quantifying lag times, we calculated the exponential decay rate (k) of CH₄ production following maximum rates of CH₄ production using the same dataset. The exponential decay rate (k) represents the rate of relative decrease in CH₄ production rates from maximum production rates over time to better understand potential changes in CH₄ production due to substrate depletion of the labile C pool. We calculated relative decrease in CH₄ production following maximum rates using the percentage of daily CH₄ production relative to the maximum CH₄ production among samples, which had several orders of magnitude of variation in maximum production rates (Fig. S1, Fig. S3). We then fit a log-linear regression to obtain k, the exponential decay rate of CH₄ production over time. We tested for interactions between time and incubation temperature, vegetation, sample depth, and relative water table position that would indicate differences in the rate of labile C consumption among groups within those factors. In field conditions, substrate for methanogenesis is generally highly labile C compounds and recent photosynthates (King et al., 2002; Prater et al., 2007). k represents the depletion of labile C pool due to anaerobic metabolism. This approach assumes that CH₄ is only derived from the labile C rather than multiple C pools of varying quality, as has previously been performed for CO₂ production during incubations (Schädel et al., 2014). However, k provides little information on the absolute size of the labile C pool or about methanogenic substrate shifts from recently fixed plant photosynthates and highly labile C compounds to more recalcitrant compounds.

Temperature sensitivity analysis

We performed a temperature sensitivity analysis to better understand the response of CH₄ and CO₂ production to temperature for each soil sample. Apparent temperature sensitivity (Q₁₀) was calculated using the equation: Q₁₀ = (R₂/R₁) ^10/(T₂ ^− T₁⁾, where T₁ and T₂ represent two temperatures and R₂ and R₁ represent C production rates at temperatures T₁ and T₂. We calculated Q₁₀ using the subset of data with incubations conducted at multiple temperatures as well as the entire dataset using the regression between incubation temperature and log-transformed mean CO₂ and maximum CH₄ production. We compared differences in apparent temperature sensitivity among groups using one-way anovas. There were insufficient data to evaluate differences in Q₁₀ due to incubation temperature range (Hamdi et al., 2013).

Statistical analyses

We investigated differences in maximum CH₄ production, mean CO₂ production, CO₂:CH₄ production, CO₂e, and lag times among environmental controls (incubation temperatures, relative water table position, pH), substrate controls (vegetation type, depth, permafrost, soil type), and landscape patterns (biomes and landscape position). We used mixed effects modeling (lmer, R package: lme4, Bates et al., 2014) with maximum likelihood estimation to account for the random effects of site and fixed effects of incubation temperature, pH, and depth. While we could have included these fixed effects as blocking factors in a linear modeling approach, the dataset was highly unbalanced and the mixed modeling approach was preferable. We tested for differences among groups using a chi-squared test against a null model including only fixed and random effects (incubation temperature, pH, and site). The chi-squared test is known to be conservative and agreed with model comparisons using AIC.

We hypothesized that CH₄ and CO₂ production could be predicted by a combination of environmental controls, soil characteristics, and site characteristics. We used a forward-selection stepwise multiple regression to identify which predictors explained the most variance among rates of maximum CH₄ production, mean CO₂ production, anaerobic CO₂:CH₄ production, and lag times based on AIC scores (stepAIC, R package: MASS, Venables & Ripley, 2002). Variables were selected for the model in the order of largest improvement in AIC scores from the null model with only an intercept and calculated the relative importance of the accepted predictors (calc.relimp, R package: Relaimpo, Grömping, 2006). Variables that resulted in higher AIC scores (i.e., poorer model fit) than the null model were rejected. Prior to running the stepwise regression, we tested for correlation among predictor variables and found that all correlations were <0.4. An additional ordination analysis was used for data visualization (Figs S8 and S9).

Data were log-transformed (+1 μg C gC⁻¹ day⁻¹) to meet assumption of normality for all statistical analyses. All statistical analyses were conducted using R (R Development Core Team, 2008). We were unable to use meta-analysis techniques on this dataset because there was no common incubation temperature or length across all incubations (i.e., no ‘control’ treatment; Table 1).

General trends

The median value of the maximum CH₄ production rates from incubations of northern permafrost zone soils was 0.6 μg CH₄-C g soil⁻¹ day ⁻¹ and normalized per gram soil C was 3.2 μg CH₄-C gC⁻¹ day ⁻¹. Maximum CH₄ production per gram soil was positively related to % C in the soil (F_1,205 = 36.7, P < 0.001, R² = 0.15; Fig. S6). On an areal basis, median of the maximum CH₄ production rates from laboratory incubations was 0.05 g CH₄-C m⁻² day ⁻¹.

The median rate of anaerobic CO₂ production from incubations of northern permafrost zone soils was 18.0 μg CO₂-C g soil⁻¹ day⁻¹ and normalized per gram C was 58.3 μg CO₂-C gC⁻¹ day⁻¹. As with CH₄ production, anaerobic CO₂ production per gram soil was positively related to % C in the soil (F_1,172 = 36.7, P < 0.001, R² = 0.18; Fig. S6). On an areal basis, median anaerobic CO₂ production from laboratory incubations was 1.5 g CO₂-C m⁻² day⁻¹. Hereafter, we report all results for anaerobic CO₂ and CH₄ production as per gram soil C to account for differences in the amount of soil C among samples, unless otherwise noted.

CO₂ and CH₄ production rates varied among hierarchical landscape units. The CO₂:CH₄ production ratio was >50 times higher in boreal sites (median = 387) than tundra sites (median = 7; χ² = 18.5, df = 1, P < 0.001), and while CO₂ production was higher in the boreal biome than the tundra biome (153 ± 17 and 109 ± 16 μg CO₂-C gC⁻¹ day⁻¹), these differences were not statistically significant (χ² = 0.45, df = 1, P = 0.50). Anaerobic CO₂ production and the CO₂:CH₄ production ratio differed among landscape positions (χ² > 14, df = 4, P < 0.01; Table 2). Anaerobic CO₂ production was largest from the wetlands while the CO₂:CH₄ production ratio was largest from the lowland forests and smallest from the wetlands (Table 2). Maximum CH₄ production did not differ significantly between boreal and tundra sites (χ² = 2.36, df = 1, P = 0.12) or among landscape positions (χ² = 6.6, df = 4, P = 0.16; Table 2). However, the decay rates (k) of CH₄ production differed significantly among landscape positions (Table 2).

Effects of substrate

Maximum CH₄ and mean CO₂ production rates per gram soil C were significantly different among mineral and organic soils and across soil depths. Maximum CH₄ production was more than five times greater in organic soils than mineral soils (χ² = 3.6, df = 1, P = 0.06; Table 3), and anaerobic CO₂ production was three times greater in organic soils than mineral soils (χ² = 20.5, df = 1, P < 0.001; Table 3). There was a trend toward lower anaerobic CO₂:CH₄ in mineral soils (median = 11.6) than in organic soils (median = 19.0; χ² = 2.3, df = 1, P = 0.13). The decay rate (k) of CH₄ production was significantly higher in organic soils than mineral soils (14.6 and 2.9 × 10⁻⁴ day⁻¹, respectively; F_3,748 = 51.8, P < 0.001). Lag times were significantly longer in mineral soils than organic soils (403 ± 72 days and 27 ± 5 days, respectively; χ² = 4.5, df = 1, P = 0.03).

Both CH₄ and anaerobic CO₂ production decreased as a function of soil depth (Fig. 2). Maximum CH₄ production was significantly higher in soil from the top 20 cm than deeper soils (>20 cm; χ² = 17.0, df = 2, P < 0.001; Table 2). Anaerobic CO₂ production also decreased at depth (χ² = 6.5, df = 2, P = 0.04; Table 2), while CO₂:CH₄ production ratio increased with depth (Fig. 2c; Table 2). Soil depth also affected lag time, which was less than 45 days for surface soils and over 2 years for soils from >1 m depth (χ² = 13.0, df = 2, P = 0.001; Table 2). The CO₂e of anaerobic decomposition was largest from 20 to 100 cm depth soils, followed by surface soils (<20 cm), and then by deep soils (>100 cm) (1530, 680, and 340 μg CO₂e g C⁻¹ day⁻¹, respectively; χ² = 26.6, df = 2, P < 0.0001). The decay rate (k) of CH₄ production significantly decreased with depth (Table 2; F_5,746 = 35.3, P < 0.0001).

Rates of CH₄ production and anaerobic CO₂ production differed significantly among vegetation types. Maximum CH₄ production was more than five times greater in soils from graminoid- and shrub-dominated sites compared to soils from treed sites (Fig. 3a; χ² = 32.2, df = 2, P < 0.001). Maximum rates of CH₄ production from sites with Sphagnum and other moss species were more than double sites without moss (Fig. 3b; χ² = 21.1, df = 2, P < 0.001). The CO₂:CH₄ production ratio differed significantly among vascular plant types (Fig. 3d; χ² = 188, df = 2, P < 0.001) and moss cover (Fig. 3e; χ² = 78.4, df = 2, P < 0.001), as did CO₂e (Fig. 3g–h; χ² > 325, df = 2, P < 0.001). Lag times between the beginning of the incubation and day of maximum CH₄ production were <30 days for graminoid- and tree-dominated sites, but were ~7 months for shrub-dominated sites (χ² = 64.8, df = 2, P < 0.001). Decay rates (k) of CH₄ production were larger in graminoid and shrub sites than treed sites (0.12 ± 0.02 and 0.04 ± 0.05 day⁻¹, respectively).

Effects of environmental conditions

Environmental conditions – both within the incubation experiments themselves (i.e., temperature) and site-level conditions (pH and water table status) – significantly affected rates of CO₂ and CH₄ production. Incubation temperature significantly and positively related to CH₄ and anaerobic CO₂ production but explained relatively little variability compared to all variables considered (Table 4). The temperature sensitivities of anaerobic CH₄ and CO₂ production differed among studies with multiple incubation temperatures and across the whole synthesized dataset. The median apparent Q₁₀ for CH₄ production across all sites, depths, and temperatures was 1.18, and the median apparent Q₁₀ for anaerobic CO₂ production was 1.41. Lag times and decay rates (k) of CH₄ production did not differ among incubation temperatures (Table 2).

The relative water table position of a soil sample in the field also affected CO₂ and CH₄ production during the incubation experiments. Maximum CH₄ production was more than five times higher from inundated samples than from dry samples (Fig. 3c, Table 2; χ² = 43.8, df = 2, P < 0.001). Mean anaerobic CO₂ production was more than three times higher in fluctuating water table samples than from inundated samples and more than five times higher in fluctuating samples than from dry soils (Table 2; χ² = 77.3, df = 2, P < 0.001). The CO₂:CH₄ production ratio differed significantly among water table positions (Fig. 3f; χ² = 123, df = 2, P < 0.001). Lag time was <20 days for inundated saturated samples or periodically inundated soils but >45 days for dry unsaturated soils (Table 2; χ² = 113, df = 2, P < 0.001). Decay rates (k) of CH₄ production rates were greatest in the zone of fluctuating water table followed by inundated samples (Table 2; Fig. S3).

Soil pH was significantly and positively correlated with CH₄ production. Furthermore, the relationship between CH₄ production and pH was dependent on incubation temperature (χ² = 35, df = 4, P < 0.001; Fig. S7). The CO₂:CH₄ production ratio was negatively correlated with pH (slope = −0.64 ± 0.18, χ² = 84, df = 1, P < 0.001).

Effects of permafrost

Presence of permafrost, both within and across sites, affected anaerobic CO₂ and CH₄ production. Maximum CH₄ production rates from sites without permafrost were nearly double those from sites with permafrost (Fig. 4a; χ² = 4.3, df = 1, P = 0.04). In sites with permafrost, maximum CH₄ production rates from the active layer were nearly four times greater than from the permafrost layer (Fig. 4a; χ² = 1.7, df = 1, P = 0.20). Anaerobic CO₂ production was also higher from sites without permafrost than with permafrost (Fig. 4b; χ² = 8.5, df = 1, P = 0.004) and was more than twice as large from the active layer than from the permafrost layer at sites that contained permafrost (Fig. 4b; χ² = 3.1, df = 1, P = 0.08). In sites with permafrost, anaerobic CO₂:CH₄ production was significantly larger in the permafrost layer than the active layer (median = 1163 and 37, respectively; χ² = 6.6, df = 1, P = 0.01). Lag times did not differ among any permafrost classification (P = 0.8, P = 0.28).

Relative importance of substrate, environmental, and landscape controls

The relative importance of substrate, environmental, and landscape controls differed for anaerobic CH₄ and CO₂ production. All variables considered explained 39% of CH₄ production rates, with the highest amount of variance explained by landscape-level categorization (21%), followed by substrate controls (i.e., moss type, vascular vegetation type, and soil type) (11%) and incubation temperature (7%; Table 4). All variables considered explained 75% of anaerobic CO₂ production. Environmental controls including incubation temperature (13%) and relative water table position (16%) explained 29% of variance, followed by substrate controls (moss type, vascular vegetation, pH, C:N ratio, and soil type) (28%) and landscape factors such as biome (5%), landscape position (9%), and mean annual air temperature (MAAT, 5%; Table 4). Site-level and landscape factors, such as biome, landscape position, and precipitation, explained most of the relative variance in CO₂:CH₄ production ratio (43%), followed by substrate controls (27%; Table 4). Lag time was best predicted by landscape factors (47% of total variance), followed by substrate controls, specifically depth (10% of total variance; Table 4). Permafrost, both within and across sites, was not a significant predictor of anaerobic CO₂ production, CO₂:CH₄ production ratio, or lag time.

Methane production in permafrost soils: insights from the synthesis

This synthesis of anaerobic incubations studies showed the complexities and also the generalizable patterns in CH₄ and anaerobic CO₂ production rates in soils across the permafrost region. Production of methane and anaerobic CO₂, as observed from laboratory incubation studies, is largely controlled by factors that can be grouped into a few general categories: (i) substrate, (ii) environmental conditions, (iii) landscape-level factors, and (iv) microbial community dynamics (not examined in this synthesis; Table 4). The relative importance of environmental, substrate, and landscape controls differed for CH₄ production and anaerobic CO₂ production. We observed numerous legacy effects on CH₄ production and anaerobic CO₂ production from site factors such as landscape position, biome, relative water table position, and vegetation type (Fig. 3). These observed differences in CH₄ and CO₂ production among sites, under common incubation conditions, highlight the importance of soil and substrate composition for greenhouse gas production.

Substrate controls on CH₄ production

Substrate composition controls the overall decomposability of soil organic matter, and therefore, substrate is a primary determinant of potential CH₄ and CO₂ production. Substrate quality often differs between depths due to fresh litter inputs from leaves and stems at the surface, and roots belowground. Because all soils were incubated under a strict anoxic atmosphere, we were able to examine the effect of organic matter quality at different depths on CH₄ and CO₂ production. Differences in organic matter quality on gas production rates were evident among depths and vegetation types. We observed higher CH₄ and anaerobic CO₂ production from surface soils as well as higher decay rates (k) of CH₄ production, indicating higher organic carbon quality in this zone (Fig. 2, Table 2). Generally, higher quality organic C is expected closer to the ground surface due to root and litter inputs (Agren & Bosatta, 1996; Nilsson & Oquist, 2009); however, high-quality organic matter can be preserved under some permafrost formation conditions (e.g., Schuur et al., 2008).

Substrate controls on CH₄ production are especially important in the permafrost zone because labile C can be added to soils not only from recently fixed plant inputs, but also as a result of permafrost thaw. Results from our synthesis show that vegetation type was a better predictor of CH₄ production than sample depth (Table 4), that CH₄ production generally decreased with depth (Fig. 3), and that production was higher from active layer soils than from permafrost. Previous studies have found relatively high rates of C production at depth (Waldrop et al., 2010; Treat et al., 2014) due to relatively high soil C lability, which depends on permafrost formation history (Waldrop et al., 2010; Harden et al., 2012; Lee et al., 2012; Treat et al., 2014) and processes such as cryoturbation that can mix undecomposed, relatively labile organic matter in mineral soils (Schuur et al., 2008) or in patterned tundra (Repo et al., 2009). Finally, landscape factors such as biome and landscape position affect substrate through vegetation composition, C and N inputs, and losses through leaching, as well as through permafrost formation history. For example, soil organic matter in permafrost was much more labile in tundra soils than boreal soils due, in part, to a legacy of more decomposition prior to permafrost formation in boreal soils (Treat et al., 2014), which may explain why landscape factors were strong predictors of anaerobic C production in this synthesis rather than depth and environmental conditions (Table 4).

Vegetation type also influenced CH₄ production rates from incubations, similar to trends from field emissions (Olefeldt et al., 2013). Our synthesis results show that maximum CH₄ production rates occurred in graminoid-dominated sites followed by shrub-dominated sites, while little to no CH₄ production occurred in tree-dominated sites (Fig. 3). Similar patterns have been observed along a thaw gradient from thawed, wet, graminoid-dominated areas to dry palsa with intact permafrost due to changes in organic matter chemistry (Hodgkins et al., 2014). A synthesis of field measurements of CH₄ fluxes showed similar trends with highest CH₄ fluxes observed in graminoid-dominated sites and substantially lower CH₄ fluxes from shrub- and tree-dominated sites (Olefeldt et al., 2013), which may be a function both of differences in organic matter composition and vascular transport of CH₄ in sedge-dominated ecosystems. Tundra decomposition studies have shown that graminoid species decompose most quickly, followed by shrubs and finally mosses, a pattern tied to the high lignin:N ratios of woody tissues (Hobbie, 1996); our synthesis results show that k-values for CH₄ production were largest in graminoid- and shrub-dominated sites and smallest in treed sites. Thus, dominant site vegetation resulted in differing substrate favorability for CH₄ production in incubations through differences in the labile C pool, and this mechanism may be partially responsible for differences in CH₄ fluxes among vegetation types in field measurements (e.g., Olefeldt et al., 2013). However, field measurements also capture effects of vegetation on gas exchange that are not reflected in laboratory incubations. For example, the proliferation of sedges can enhance CH₄ emissions to the atmosphere through transport of CH₄ through aerenchymous tissue (King et al., 1998) and also enhance CH₄ oxidation through transport of oxygen to roots (Strom et al., 2005).

Favorable environmental conditions and microbial processes

Favorable environmental conditions for CH₄ production include temperature, pH, and deficiency of oxygen and alternate electron acceptors (not addressed in this analysis); these factors act as a secondary control on CH₄ production. Environmental conditions explained the most variance among CO₂ production rates, but were less important for CH₄ production rates (Table 4). Higher temperatures result in increased rates of CH₄ and anaerobic CO₂ production (Table 2), as has been shown in previous studies (Dunfield et al., 1993; Bergman et al., 1998; Whalen & Reeburgh, 2000; Yavitt et al., 2006; Lupascu et al., 2012). Previously, CH₄ production had been found to be positively and negatively correlated with pH (c.f. Bergman et al., 1998), perhaps due to interactions between pH, temperature, and substrate type (Valentine et al., 1994; Bergman et al., 1998) or due to a pH optima of methanogenesis (Dunfield et al., 1993). This synthesis found a positive relationship between CH₄ production and pH, and that relationship differed among incubation temperatures; pH can affect multiple processes simultaneously, including methanogenic pathway (Kotsyurbenko et al., 2007), and so although the specific mechanisms may be complex and interactive, pH is seen as a key variable controlling anaerobic metabolism across multiple landscape units.

Redox conditions play an important role for C cycling in anoxic soils. Dry, aerobic conditions or drying and re-wetting can result in a substantial lag time prior to CH₄ production due to regeneration of alternate electron acceptors (Dise & Verry, 2001; Knorr & Blodau, 2009). In this synthesis of incubation studies, the lag time before maximum CH₄ production was shortest in inundated sites, intermediate in fluctuating water table sites, and longest in dry sites (Table 2). Additionally, in chronically dry soils, methanogen populations may be substantially smaller than soils that experience inundation (Tveit et al., 2013) and require longer periods for population size to grow prior to measurable CH₄ production. Observations of lag times suggest that environmental conditions, such as temperature and redox status, as well as methanogen populations, may limit CH₄ production in fluctuating water table zones that can change over the course of a growing season in permafrost areas due to perched water tables. Furthermore, landscape changes associated with permafrost thaw, such as inundation, may not result in CH₄ production until environmental conditions are suitable and methanogen communities are present.

Recommendations, scaling up, and climate change effects

Climate change in northern regions is affecting ecosystems in multiple ways, including deepening active layer depths, shifting vegetation dynamics, growing season length, and hydrology (Serreze et al., 2000; Hinzman et al., 2005; Myers-Smith et al., 2011; Overland et al., 2013). All of these factors have the potential to affect CH₄ and anaerobic CO₂ production. Our synthesis results show both direct effects of temperature on CH₄ and anaerobic CO₂ production (Table 2), as well as indirect effects of ecosystem shifts that may be associated with climate change.

CH₄ production rates were more than twice as high in active layer soils compared to permafrost as well as from sites without permafrost compared to those underlain by permafrost, both of which suggest that CH₄ production will increase as permafrost thaws. However, there are also important indirect effects of permafrost that will have profound impacts on anaerobic CO₂ and CH₄ production and emissions. Thermokarst thaw and permafrost thaw frequently result in wetter soils, especially in lowlands. This synthesis suggests that flooding and periodic inundation of surface soils due to permafrost thaw may result in higher CH₄ fluxes, as has been observed in field CH₄ emission studies (e.g., Olefeldt et al., 2013 and references therein). Vegetation changes, such as a transition from dry, treed sites to wet, graminoid sites, or from graminoid tundra sites to shrub or tree tundra, may also alter anaerobic CO₂ and CH₄ production rates (e.g., Hodgkins et al., 2014), although the effects of vegetation on CH₄ production via substrate quality will operate at timescales of a growing season to decades while the environmental controls on CH₄ production operate on daily to seasonal scales.

There is substantial uncertainty in the spatial and temporal variability of estimates of CH₄ fluxes and CO₂ exchange in the permafrost region (McGuire et al., 2009, 2012). This synthesis has several implications for modeling anaerobic CH₄ and CO₂ fluxes from permafrost region. Our results indicate that models of greenhouse gas emissions from thawing permafrost should consider several sources of variability in methanogenesis that are generally not taken into account: variation (i) between organic and mineral horizons, (ii) with depth within soil horizons, (iii) between active layer and thawed permafrost layer of soils in sites with permafrost, (iv) between permafrost sites and sites without permafrost, and (v) among plant functional types. Models generally consider some of the effects of water table position, but this study indicates that it is important to distinguish between chronically dry surface soil horizons and seasonally or chronically inundated soil horizons (e.g., Zhuang et al., 2004; Wania et al., 2010; Riley et al., 2011; Bohn et al., 2013). Similarly, models generally have different parameterizations for different anaerobic ecosystems, but there is no generally accepted scheme as to how those parameterizations should be developed. This synthesis indicates that parameterizations should be hierarchically organized by biome, landscape position, and vascular/moss vegetation types (Table 4). While models generally consider the effects of pH and temperature on methanogenesis, this study provides some additional insight into the relative importance of these effects on anaerobic metabolism (Table 4). The information on lag times in methanogenesis provided by this study can inform models that simulate the effects of thermokarst disturbance (e.g., from forested permafrost plateau to collapse scar bog) on methanogenesis because of possible lags in microbial colonization and population dynamics, which are ultimately shorter than the persistence of thermokarst wetlands. Finally, most models do not consider anaerobic CO₂ production in modeling methane dynamics, and this study provides substantial insight into the variability in the ratio of anaerobic CO₂ to CH₄ production.

To improve estimates of CH₄ production from soils across the permafrost zone, future incubation study designs should be modified to include longer incubations, a wider variety of sites, additional measurement parameters, and to include methane oxidation. The substantial lag times before maximum rates of CH₄ production (Table 2) suggest that longer duration anaerobic incubation experiments are required to assess maximum CH₄ production from permafrost soils; we suggest 30-day minimum incubation length and longer time periods if low incubation temperatures are used. Second, a sampling bias toward hot spots of CH₄ production across the landscape (McClain et al., 2003), that is, wetlands, could affect our estimates of lag time and maximum CH₄ production rates among some classifications we used, especially for permafrost presence/absence. Increased sampling of additional mineral or upland soils might reduce bias in our estimates of maximum rates of CH₄ production and lag times in the permafrost zone. New methodologies need to be used to identify pathways of anaerobic CO₂ and CH₄ production, including use of stable isotopes and chemical characterization of organic matter (Corbett et al., 2013; Tfaily et al., 2013; Hodgkins et al., 2014). Furthermore, measurements of methanogen communities, redox status and pH in mineral or upland soils and both dry and saturated wetland soils may help clarify whether long lag times prior to maximum CH₄ production rates are due to redox conditions or methanogen community abundances. Furthermore, we did not consider CH₄ oxidation in this synthesis; additional research on rates of oxic and anoxic CH₄ oxidation (Whalen & Reeburgh, 2000; Blazewicz et al., 2012; Haroon et al., 2013) will better constrain CH₄ budgets from the permafrost zone.

Results from this synthesis highlight the importance of understanding the complex interactions among geochemistry, methanogen communities, and organic matter quality for determining ecosystem scale patterns of CH₄ and CO₂ production as the thaw regime and vegetation dynamics continue to change in northern latitude systems. Our results suggest that the largest source of anaerobic CO₂ and CH₄ produced will be from seasonally saturated surface soils rather than from the decomposition of newly thawed substrate found in permafrost soils. Therefore, increased wetland area associated with permafrost thaw will result in larger CH₄ emissions than deeper seasonally thawed soils and underscores a need for capturing hydrologic changes associated with ecosystem transitions.