Effects of soil warming and drying on methane cycling in a northern peatland mesocosm study

2.1. Source Sites

[6] Source sites for the mesocosms consisted of a bog and a fen in northeastern Minnesota, USA (47°N, 92°W). These sites are typical of northern bogs and fens, and have been described elsewhere [Bridgham et al., 1998; Chapin, 1998; Bridgham et al., 2001]. The peat at the bog site is approximately 3.5 m deep with a basal radiocarbon date of 10,040 ± 70 years BP. The upper 60 cm is derived largely from Sphagnum moss. Vegetation includes ericaceous shrubs [Chamaedaphne calyculata (L.) Moench., Andromeda glaucophylla Link., Kalmia polifolia Wang., Vaccinium oxycoccos L., Rhododendron groenlandicum (Oeder) Kron and Judd], bryophytes [Sphagnum fuscum (Schimp.) Klinggr., S. capillifolium (Ehrh.) Hedw., S. magellanicum Brid., Polytricum juniperinum Var. affine (Funck) Brid.], and spruce [Picea mariana (Mill.) BSP]. The fen is composed of approximately 4.4 m of sedge peat over about 2 m of unconsolidated aquatic (limnic) peat with a radiocarbon date of 9,730 ± 70 years BP [Bridgham et al., 1998]. To maximize the contrast with the bog, the source areas selected were low areas (flarks) dominated by graminoids (Rhynchospora alba (L.) Vahl, R. fusca (L.) Ait. f., Carex limosa L., C. lasiocarpa Ehrh., C. livida (Wahl.) Willd.) with minimal bryophyte coverage.

2.2. Mesocosm Experimental Design

[7] The mesocosm facility was constructed during winter of 1993–1994. Twenty seven intact cylindrical peat monoliths (2.1-m² surface area, ∼0.6-m depth) were removed each from the bog and fen, and were placed in insulated plastic tanks that had been buried in an open field near the source sites at the University of Minnesota Fens Research Facility, 70 km north of Duluth, MN, USA. Frozen, intact peat monoliths were carefully placed into the tanks with minimal disturbance of the living vegetation. Treatments were randomly assigned to each experimental mesocosm (2 peatland types × 3 water table treatments × 3 IR loading treatments × 3 replicates = 54 mesocosms). For this study, in order to reduce stable isotope analytical costs, we sampled only the extremes for each treatment (2 peatland types × 2 water table treatments × 2 IR loading treatments × 3 replicates, 24 mesocosms). The heated mesocosms (IR = ‘high’) were treated year-round with overhead IR lamps to augment atmospheric inputs; the net radiation increase was ∼90 W m⁻² [Noormets et al., 2004]. The unheated mesocosms (IR = ‘ambient’) received no supplemental IR radiation. Mean monthly soil temperatures in the heated mesocosms exceeded ambient soil by approximately 1.6 to 4.1°C at 15-cm depth during the growing season [Bridgham et al., 1999], within the range of temperature increases predicted by global climate models. Fen mesocosms were on average 0.8–1.0°C warmer than bog mesocosms during the growing season due to ecosystem-dependent controls on soil energy fluxes [Noormets et al., 2004].

[8] Water table levels were controlled by a drainage system at the base of the tank and connected to a vertical J-tube manostat inside a smaller second sump tank. Water table levels were measured as the distance between the average of the peat surface in the fen mesocosms and the dense mat of Sphagnum mosses in the bog mesocosms and the top of the manostat tube. This average surface was determined in mid-growing season of each year by placing a frame over the mesocosms and measuring the distance of ∼110 points per mesocosm to the appropriate surface. Water was pumped from a ditch draining a nearby bog and added to the mesocosms at approximately weekly intervals if needed to bring the water tables to the set level. The water chemistry of the ditch water was similar to bog pore water [Updegraff et al., 2001]. Water table levels were kept relatively constant relative to the lip of the mesocosm tanks but allowed to change relative to moss surface (bog) or peat surface (fen) as the mesocosms gained or lost carbon, such that water table level relative to the peat surface was a dynamic response variable of the experiment [see Bridgham et al., 2008]. Mean water table levels (WL) in the bog mesocosms were −21 cm in 1998 and −22 in 1999 in the wettest treatment (WL = ‘wet’) and −31 cm in 1998 and −33 cm in 1999 in the driest treatment (WL = ‘dry’). For fen mesocosms, mean water table levels were −5 cm in both 1998 and 1999 in the wettest treatment (WL = ‘wet’) and −23 cm in 1998 and −25 cm in 1999 in the driest treatment (WL = ‘dry’). More details of mesocosm construction and design are provided by Bridgham et al. [1999], Weltzin et al. [2000, 2003], and Noormets et al. [2004].

2.3. Sampling Schemes

[9] During the period of this study, 1998 and 1999, methane fluxes were measured at two- to three-week intervals during the growing season (May through September), using static flux chambers. Flux chambers enclosed the entire mesocosm, and methane emissions were sampled several times over 30 to 40 min of accumulation. Soil temperature was unaffected by incubation times used for flux measurements. Further details of the static chambers and trace gas analyses are provided elsewhere [Updegraff et al., 2001]. The sampling dates for methane fluxes and for pore water sampling were off-set by one to two weeks, so we used linear interpolation to estimate methane fluxes on the pore water sampling dates. Gas for isotopic measurements was pumped from flux chambers into 1-L evacuated cylinders to a pressure of 15 psi.

[10] Pore water samples were collected from mesocosms every 3–4 weeks during the growing season. Pore water samples were drawn from the bottom of mesocosms using gas-tight glass syringes after flushing manostat systems to ensure representative pore water. Pore water syringes were capped immediately, packaged in dry ice, and shipped overnight to the Biogeochemical Laboratories at Indiana University for stable isotope analyses and to Pennsylvania State University for analysis of methane, acetate and sulfate concentrations, and to the University of Minnesota-Duluth for analyses of dissolved organic carbon (DOC), ammonia (NH₄⁺-N), nitrate (NO₃⁻-N), and phosphate (PO₄⁻³-P).

2.4. Pore Water Chemistry and Belowground Net Primary Productivity

[11] Dissolved methane in pore waters was stripped into an equal volume of air in a gas-tight syringe after vigorous shaking for 3 min. Headspace gas samples for methane analysis were then injected directly into a 2–ml sample loop of a Shimadzu GC-14A gas chromatograph with a flame ionization detector, and carried by helium (60 ml min⁻¹) onto a Poropak Q column (80/100 mesh, 1 m by 3 mm) at 70°C. Pore water samples for analysis of acetate and sulfate were preserved by filtering with a 0.45-μm membrane filter, adjusting the sample pH to >10 with NaOH, and storing at 4°C until analyzed. Acetate and sulfate concentrations in pore waters were analyzed and quantified simultaneously with a Dionex 500 ion chromatography system, utilizing a Dionex AS-11 analytical column with self-regenerating autosuppression. Sample analysis was achieved with a gradient run designed to separate low molecular weight organic anions and inorganic polyvalent anions. Eluent (1 ml min⁻¹) ranged from 0.5 to 70 mM NaOH, sample loop size was 20 μl, and detection limits for acetate and sulfate were 5 μM and 0.3 μM, respectively. Pore water for the remaining analyses was passed through Fisherbrand G4 glass-fiber filters. Colorimetric analysis by autoanalyzer was used to measure concentrations of NH₄⁺-N, NO₃⁻-N, and PO₄⁻³-P. We developed an empirical relationship between DOC concentration determined directly with a TOC analyzer (Doorman DC-80) and UV absorbance at 320 nm wavelength measured with a Perkin Elmer Lambda 3B Spectrophotometer, and then used UV absorbance to estimate DOC for the remainder of the samples [Pastor et al., 2003].

[12] Belowground net primary productivity (BNPP) was measured for the growing seasons of 1998 and 1999 using root ingrowth cores. Three cores of root-free homogenized peat were inserted into each mesocosm just after soil thaw in early June and were harvested at the end of the growing season in September. Seasonal root production was calculated volumetrically to a depth of 27 cm and expressed on an areal basis (g m⁻²). Root production below 27 cm was minimal [Weltzin et al., 2000]. Aboveground net primary productivity (ANPP) was measured nondestructively on all dominant vascular plant species within each mesocosm based upon species-specific allometric equations [Weltzin et al., 2000]. Bryophyte ANPP was only measured in the bog mesocosms because bryophytes were unimportant in the fen mesocosms. It was determined based upon basal cover, lineal shoot growth, and density of shoots and species-specific shoot mass/shoot length relationships [Weltzin et al., 2000].

2.5. Stable Isotope Analyses

[13] The stable isotopic signature of carbon and hydrogen in biogenic methane is dependent on the production pathway and on methanotrophic activity. There is a kinetic isotope effect (KIE) associated with each of these microbial transformations, leading to significant isotopic shifts for both the hydrogen and carbon in methane (expressed as ɛ_H and ɛ_C, respectively; units of ‰) [Avery et al., 1999; Whiticar, 1999; Snover and Quay, 2000; Penning et al., 2005]. In autotrophic methanogenesis, typical ɛ_C values are around −64‰, which yields light methane around −75‰ if one assumes the typical values for C in source carbon dioxide of around −11‰. With acetoclastic methanogenesis, the acetate carbon is fractionated to a lesser degree and thus, the isotopic signature of carbon in methane is similar to the source acetate (about −18‰ [Heuer et al., 2006]).

[14] Methanotrophy also selectively removes isotopically light C and H in methane (through the KIE of methane monooxygenase activity), enriching the residual methane pool with the heavier isotopes [Coleman et al., 1981; Bergmaschi and Harris, 1995; Whiticar, 1999]. Note that this KIE is in the opposite direction relative to methanogenesis. The KIE of methanotrophy is larger for hydrogen than for carbon. Thus, the corresponding isotopic shift in carbon is about 20‰, while the shift in hydrogen can be as high as 280‰ [Whiticar, 1999]. The values of ɛ_H are less well constrained than ɛ_C [Waldron et al., 1999]. Note that the δ¹³CO₂ is unlikely to be altered by methanogenic activity. The isotopic signature of carbon dioxide in pore water does not change significantly over seasons in peatland soils because there is little or no KIE for the respiration pathways that yield the majority of the carbon dioxide.

[15] We measured the coexisting isotopic compositions of pore water methane and carbon dioxide and of emitted methane in each mesocosm over the course of the growing season in order to determine pathways of methanogenesis and the extent of methane oxidation. Pore water gases were transferred from gas-tight syringes to gas-tight serum vials within 48 hours of field collection. Pore water was analyzed for δ¹³C_CH4, δ¹³C_CO2, δD_CH4 and δD_H2O. Emitted methane was analyzed for δ¹³C_CH4 and δD_CH4. Methane and carbon dioxide were prepared for mass spectrometry using cryogenic distillation on a methane combustion vacuum line. Pore water gases were transferred from serum vials to the vacuum line using ultra high-purity helium (30 ml min⁻¹). Flux chamber gases, contained in pressurized cylinders, were introduced to the vacuum line through a mass flow controller at 30 ml min⁻¹. The entire amount of gas in vials and cylinders was transferred to the vacuum line so as to avoid any isotope fractionation associated with sample removal. Gases first passed into a liquid nitrogen (LN₂) trap (−192°C) for removal of water and carbon dioxide. Methane moved through this trap and was combusted at 800°C to carbon dioxide and water in a platinum catalyzed CuO₂ furnace. Combustion water was trapped in an acetone slush trap (−112°C) and carbon dioxide was collected in a LN₂ trap. Helium carrier gas was pumped away by vacuum and the methane combustion products, carbon dioxide and water, were transferred into separate glass tubes and flame sealed. In a similar manner, the dissolved inorganic carbon δ¹³C_CO2 was prepared by cryogenically distilling the carbon dioxide from the first LN₂ trap into glass tubes. A zinc catalyst was used to reduce combustion water hydrogen to H₂, in vitro, by heating tubes to 500°C for 40 min [Coleman et al., 1982]. We included VSMOW (Vienna Standard Mean Ocean Water) and SLAP (Standard Light Antarctic Precipitation) standards with each set of samples processed for water reduction. The overall methane oxidation and purification efficiency of the procedure was consistently >95%.

[16] Purified gas samples were analyzed for ¹³C/¹²C and D/H using a Finnigan MAT 252 isotope ratio mass spectrometer using dual-inlet. The detection limits for this instrument are about 1 μmol CO₂ and 4 μmols H₂; analytical precision was generally better than ±0.05‰ and ±0.1‰ for ¹³C/¹²C and D/H, respectively. VSMOW and SLAP standards were used to correct δD_CH4 values according to the method outlined by Coplen [1988, 1996]. Based upon the repeated analysis of methane isotope standards, the accuracy of the entire procedure was ±0.5‰ and ±2‰ for ¹³C/¹²C and D/H, respectively.

2.6. Statistical Analyses

[17] We used three-way ANOVAs to examine the fixed effects of wetland type (bog versus fen), water table setting (dry versus wet), and IR loading (ambient versus high heat) on methane pore water concentrations, methane flux, δ¹³CH₄ in pore water and emitted methane, and the δD of pore water methane and emitted methane. We used mean values of the response variables over 1998 and 1999 for each mesocosm for the ANOVAs because seasonal and interannual effects on methane fluxes in this experiment have been reported elsewhere [Updegraff et al., 2001]. It is important to note that our data do not overlap with the previous study [Updegraff et al., 2001], which only spanned 1995–1997. Repeated measure ANOVAs (not shown) gave very similar results to ANOVAs on mean treatment values. Often there were significant interactions between wetland type and water table setting and IR loading, so we also conducted two-way ANOVAs for bog and fen mesocosms separately. If necessary, response variables were transformed prior to analysis to conform to assumptions of normality and equal variances. A square root transformation was conducted for pore water concentrations of methane, and a natural log transformation was used for methane fluxes and pore water δD values.

[18] We also did regression analyses to examine the response of methane flux and pore water methane concentration to water table depth, average soil temperate at 15-cm depth at the time of methane flux measurements, and pore water concentrations of acetate, sulfate, NO₃⁻-N, NH₄⁺-N, PO₄⁻³-P, and DOC. For regression analyses, we used separate mean values for 1998 and 1999 for each mesocosm. We also examined pore water pH and concentrations of total nitrogen and phosphorus, but these variables were highly autocorrelated with other predictor variables, so we eliminated these three variables from further consideration. We examined the individual relationships of the predictor variables to methane flux with simple regressions. We then used Akaike Information Criterion (AIC) to determine the best set of multiple regression models for methane flux and pore water methane concentrations. AIC is based upon information theory [Anderson et al., 2000; Burnham and Anderson, 2002]. Instead of choosing a single ‘best’ model as in stepwise multiple regression, it ranks all possible regression models based upon minimizing the residual sums of squares but with a penalty for incorporation of additional explanatory variables (i.e., more parsimonious models are favored). Following accepted practice, we calculated the regression model with the minimum AIC value and then accepted any model with an AIC score within 2 units of this minimum. We also used corrected AIC values (AIC_c) to account for small sample sizes. Thus, this method allows one to examine a set of essentially equivalent multiple regression models. To examine the weighting of the variables in the multiple regression models, we calculated the partial correlation coefficients of each variable in the accepted AIC_c models and then averaged the partial correlation coefficients for all models in which that variable entered. The 1023 possible multiple regression models and the partial correlation coefficients were determined in JMP v. 5, and then the AIC_c values were calculated in a spreadsheet. Last, we examined the multivariate relationships among the variables in a principal components analysis (PCA) with quartimax rotation. We entered both methane fluxes and pore water methane concentrations into the same PCA analyses to reduce the number of analyses reported (and to directly compare the two response variables), but when considered separately, the results were qualitatively very similar in relation to the predictor variables. Because of the large differences in the responses of the bog and fen mesocosms, we performed all regression analyses on all mesocosms combined and then separately for the two community types. With the exceptions of the partial correlation coefficients mentioned above, all statistical analyses were performed with Systat v. 11.

3.1. Methane Concentrations in Mesocosm Pore Waters

[19] Pore water methane concentrations were more than three times as high in bog mesocosms than in fen mesocosms (328 μM versus 96 μM, respectively, p < 0.0001, Figure 1a). In the three-way ANOVA, there was a suggestion that the effects of water level and IR loading differed in the bog and fen mesocosms (water level × wetland type interaction p = 0.096; IR loading × WL × wetland type interaction p = 0.120, Table 1). When the bog mesocosms were examined separately, pore water methane concentrations were higher in the dry (420 μM) than in the wet (236 μM) WL treatment (p = 0.002), and greater with ambient (375 μM) than with high IR (281 μM, p = 0.007). Thus, the greatest pore water methane concentrations in the bog mesocosms occurred in dry, cool treatments, and the lowest concentrations occurred in wet, warm treatments (Figure 1a). In fen mesocosms, only the IR treatment affected pore water methane concentrations (p = 0.026), with greater concentrations in the ambient (129 μM) versus the high (62 μM) IR treatment.

[20] Wetland type, WL, and IR individually had significant effects on pore water δ¹³CH₄ values (Table 1). Mean values of pore water δ¹³CH₄ were significantly heavier in fen mesocosms than in bog mesocosms (fens = −35.7‰, bog = −41.0‰), heavier for wet WL than for dry WL (wet = −34.6‰; dry = −42.2‰), and heavier in mesocosms with high versus ambient IR (high = −35.7‰, ambient = −41.0‰) (Figures 1c and 1e). There was no differential effect of IR and WL between the two wetland types on pore water δ¹³CH₄. Wetland type was the only factor that affected δD values of pore water methane (p = 0.050, Table 1 and Figure 1e), with mean δD value of −190‰ in bog mesocosms and −168‰ in fen mesocosms.

[21] Using regression analyses, we found that bog and fen mesocosms often had very different relationships between pore water methane concentrations and the various predictor variables that we examined (Figure 2). Pore water methane concentration was negatively correlated with soil temperature when the bog and fen mesocosms were considered together (r² = 0.17; p < 0.01), but the relationship was insignificant when the mesocosms were examined separately. Pore water methane concentrations exhibited a moderately strong negative correlation with water table depth in the combined bog and fen mesocosms (r² = 0.42; p < 0.001) and in the bog mesocosms alone (r² = 0.36; p < 0.01), but the relationship was insignificant in the fen mesocosms alone.

[22] Pore water methane concentrations were positively correlated with ANPP only when the bog and fen mesocosms were considered together (r² = 0.15; p < 0.01). Irrespective of mesocosm grouping, pore water methane concentrations were not significantly correlated with BNPP and pore water nitrate concentrations. However, pore water methane concentrations were positively correlated with pore water ammonium concentrations in only the fen mesocosms (r² = 0.36; p < 0.01). Pore water methane concentrations were significantly correlated with pore water phosphate, sulfate, and DOC concentrations only when the bog and fen mesocosms were considered together, but this relationship was positive for phosphate (r² = 0.31; p < 0.001) and DOC (r² = 0.37; p < 0.001) and negative for sulfate (r² = 0.13; p < 0.05). Pore water methane concentrations were negatively correlated with pore water acetate concentrations in the combined bog and fen mesocosms (r² = 0.09; p < 0.05) and the fen mesocosms alone (r² = 0.21; p < 0.05) (Figure 2).

[23] We evaluated candidate multiple regression models of the square root of pore water methane concentrations against various predictor variables, and report average partial correlations obtained with the corrected Akaike Information Criterion (AIC_c). Nine multiple regression models with four to six variables were accepted for predicting pore water methane concentrations when the bog and fen mesocosms were considered together in our AIC analysis, with R² values ranging from 0.82 to 0.84 (Figure 3). Pore water sulfate concentration, soil temperature, and water table depth entered into all models, whereas pore water ammonium concentration entered into eight of the nine models. The partial correlations were negative for sulfate (average partial r [APR] = −0.74), soil temperature (APR = −0.42), and water table depth (APR = −0.85), but positive for ammonium (APR = 0.19). BNPP entered into four models (APR = −0.22), pore water acetate concentrations into three models (APR = −0.19), ANPP into three models (APR = 0.22), and DOC concentrations into one model (APR = 0.22).

[24] Only a single multiple regression model with four variables was accepted based upon our AIC criteria for the bog mesocosms, with an R² of 0.73 (Figure 3). Pore water methane concentration had negative relationships with water table depth (APR = −0.56), soil temperature (APR = −0.55), and DOC concentration (APR = −0.58), whereas this relationship was strongly positive with pore water nitrate concentration (APR = 0.69) (Figure 3).

[25] Five multiple regression models with three to six variables were accepted for the fen mesocosms, with high overall R²'s ranging from 0.85 to 0.90 (Figure 3). Pore water sulfate (APR = −0.83) and acetate (APR = −0.70) entered into all five models with negative loadings, whereas pore water ammonium loaded into all models with a high positive loading (APR = 0.90). BNPP (APR = −0.43) entered into three models, and soil temperature (APR = −0.44) and pore water nitrate (APR = −0.34) each entered into two models, all with negative loadings.

3.2. Methane Fluxes From Mesocosms

[26] Similar to pore water methane concentrations, methane fluxes from bog mesocosms were significantly greater (p < 0.0001) than fluxes from fen mesocosms (8.3 versus 3.1 mmols m⁻² d⁻¹, respectively, Figure 1b). The effect of the WL treatment depended on wetland type, with wet fen mesocosms emitting more methane than dry fen mesocosms (5.2 versus 0.9 mmols m⁻² d⁻¹, respectively, p < 0.001), but with no significant effect of the WL treatment in the bog mesocosms. In contrast to the significant effect of IR treatment on pore water methane concentrations, IR treatment did not affect methane fluxes. In the three-way ANOVA, there was a strong first-order interaction between wetland type and WL, with dry mesocosms having much larger differences in mean methane fluxes between bogs and fens (wetland type × WL interaction p = 0.001, Table 1 and Figure 1b).

[27] The mean δ¹³CH₄ values of emitted methane exhibited no main treatment effects of wetland type, WL or IR (Table 1). However, we observed a significant first-order interaction between wetland type and WL (wetland type × WL interaction p = 0.049, Table 1), with dry fen mesocosms yielding 15‰ heavier δ¹³CH₄ than wet mesocosms (dry = −21.5‰, wet = −36.2‰) (Figure 1d). Wetland type was the only main factor that affected δD values of emitted methane (p = 0.048, Table 1 and Figure 1f), with mean δD value of −195‰ in bog mesocosms and −141‰ in fen mesocosms.

[28] Several factors were correlated with methane fluxes, but these relationships varied depending on whether the bog and fen mesocosms were examined together or separately (Figure 4). Soil temperature was weakly correlated with methane flux when the bog and fen mesocosms were combined (r² = 0.07, p = 0.073). In bog mesocosms, the correlation was significant and positive (r² = 0.22, p < 0.05), while in fen mesocosms it was insignificant (p = 0.16). Methane flux was positively correlated with water table depth in the bog and fen mesocosms when examined separately, but this relationship was quite different for the two community types so that when they were combined the correlation was insignificant. The water table depth was particularly effective in accounting for variations in methane flux in fen mesocosms (r² = 0.87), but less so in bog mesocosms (r² = 0.33) (Figure 4).

[29] Methane flux increased with increasing ANPP in the mesocosms when considered together (r² = 0.28; p < 0.001) and in the fen mesocosms alone (r² = 0.32; p < 0.01), but there was no statistically significant relationship in the bog mesocosms. In contrast, correlation of methane flux with BNPP was insignificant, irrespective of mesocosm grouping.

[30] Bog and fen mesocosms represented very different biogeochemical conditions. Note that the ranges in pore water concentrations of nitrate, ammonium, sulfate, and acetate were much narrower in bog mesocosms than in fen mesocosms (Figure 4). Methane fluxes were strongly negatively correlated with pore water concentrations of ammonium in both the bog and fen mesocosms for all cases (r² = 0.34 to 0.73). Methane fluxes were negatively correlated with pore water nitrate concentrations for all mesocosms (r² = 0.33) and in the fen mesocosms (r² = 0.59), but the relationship was insignificant in the bog mesocosms.

[31] The relationship between methane flux and pore water phosphate was strongly nonlinear when the bog and fen mesocosms were considered together. This relationship was marginally significant and negative in bog mesocosms alone (r² = 0.15, p = 0.064), but insignificant in the fen mesocosms. Methane fluxes were strongly negatively correlated with pore water sulfate concentrations for all cases (r² = 0.70; p < 0.001) and in the fen mesocosms (r² = 0.83; p < 0.001), but the relationship was insignificant in the bog mesocosms alone. Methane fluxes were not significantly correlated with pore water acetate concentrations in any case. Methane fluxes were positively correlated with pore water DOC concentrations overall (r² = 0.49; p< 0.001) and in the fen mesocosms (r² = 0.27; p < 0.05), but the relationship was insignificant in the bog mesocosms.

[32] Nine multiple regression models with four to six variables were accepted for predicting methane flux when the bog and fen mesocosms were considered together in our AIC analysis, with high overall R² values ranging from 0.91 to 0.93 (Figure 5). Pore water concentrations of ammonium, sulfate, acetate, and DOC entered into all models. The partial correlations were negative for ammonium (APR = −0.69), sulfate (APR = −0.64), and acetate (APR = −0.36) and positive for DOC (APR = 0.65). Pore water nitrate concentrations entered into five of the nine models (APR = −0.27), BNPP entered into four models (APR = −0.24), water table depth into four models (APR = 0.27), and soil temperature into one model (APR = 0.22).

[33] Only a single multiple regression model with five variables was accepted based upon our AIC criteria for the bog mesocosms, with an R² of 0.85 (Figure 5). Pore water concentrations of ammonium (APR = −0.86) and sulfate (APR = −0.65) had strong negative relationships with methane flux, whereas soil temperature (APR = 0.84), pore water phosphate concentrations (APR = 0.72), and BNPP (APR = 0.49) had strong positive relationships with methane flux.

[34] Four multiple regression models with three to five variables were accepted for the fen mesocosms, with high overall R² values ranging from 0.94 to 0.95 (Figure 5). Pore water acetate concentration (APR = −0.61) and BNPP (APR = −0.56) entered into all four models with negative loadings, whereas water table depth loaded into all four models with a positive loading (APR = 0.87). Pore water nitrate (APR = −0.43) and sulfate (APR = −0.37) each entered into two of the models with negative loadings.

3.3. Principal Components Analysis of Methane Flux and Pore Water Methane

[35] In the PCA for the bog and fen mesocosms together, the first and second axes explained 32% and 27% of the total variance, respectively (Table 2). Methane flux loaded strongly onto the first axis (axis 1 score = 0.85) but only weakly onto the second axis (score = 0.38) (Figure 6a and Table 2). In contrast, pore water methane concentration loaded strongly onto the second axis (axis 1 score = 0.89). Thus, it appears that the first PCA axis primarily explains methane flux and the second axis explains pore water methane concentrations. ANPP, water table depth, and pore water DOC and acetate concentrations loaded in a similar direction as methane flux on axis 1, whereas pore water concentrations of nitrate, ammonium, and sulfate, and secondarily soil temperature, loaded in the opposite direction to methane flux. Similar to pore water methane concentration, pore water DOC and phosphate concentrations had strong positive loadings on axis 2, whereas BNPP and ANPP had moderate positive loadings on this axis. Water table depth had a strong negative loading on axis 2, and acetate concentration and soil temperature had moderate negative loadings on this axis.

[36] In the PCA of the bog mesocosms, the first and second axes explained 28% and 26% of the total variance, respectively (Table 2). Methane dynamics were primarily described on the first axis, with methane flux loading strongly negatively (score = −0.79) and pore water methane concentration loading strongly positively (score = 0.71) on this axis (Figure 6b and Table 2). Water table depth had a strong negative loading on axis 1, similar to methane flux, whereas pore water concentrations of ammonium and phosphate and BNPP had moderate-to-strong positive loadings on axis 1.

[37] In the PCA of the fen mesocosms, the first and second axes explained 44% and 18% of the total variance, respectively. Both methane flux and water table depth had strong positive loadings on the first axis (score = 0.91 and 0.96, respectively) (Figure 6c and Table 2). However, ANPP and pore water concentrations of DOC and acetate all had moderately strong positive loadings on this axis, whereas pore water concentrations of sulfate, nitrate, and ammonium all had strong negative loadings on axis 1. Pore water methane concentration had moderate negative loadings on both the first and second axes (axis 1 score = −0.42, axis 2 score = −0.39), and so it was somewhat more difficult to interpret relative to the predictor variables. Pore water concentrations of phosphate, nitrate, and ammonium tended to have loadings in the same directions as pore water methane, and pore water acetate and DOC concentrations and ANPP tended to have loadings in the opposite direction.

4.1. Direct Effects of Water Level on Pore Water Methane Concentrations and Methane Emissions

[40] Water level is known to be an important factor affecting rates of methane production, oxidation and emission from peatland ecosystems [Shannon and White, 1994; Nykanen et al., 1998]. Low water levels often suppress methane flux rates in wetlands due to an expansion of the methane oxidation zone associated with aerobic unsaturated soil and a contraction of the methanogenesis zone associated with anaerobic saturated soil [Whalen and Reeburgh, 1996, 2000]. However, pore water methane concentrations may remain elevated in anaerobic peat layers below the water table. In our study, water table depth had opposite effects on pore water methane and methane flux, and these effects often varied by peatland type. Drier bog mesocosms had higher pore water methane concentrations, but water table depth had a minor effect on pore water methane in the fen mesocosms (Figures 1a, 2, and 3). As expected, higher water levels caused higher methane fluxes in both the bog and fen mesocosms (Figure 1b), and this effect was particularly strong in the fen mesocosms, with water table depth alone explaining 87% of the variance in methane fluxes (Figure 4). Thus, our results support water table depth as the dominant factor controlling methane fluxes in the fen, probably mainly through its effect on depth of the zone for aerobic methane oxidation. This is consistent with other studies that have measured the highest flux rates from peat soils with the greatest inundation and thus the smallest zone of unsaturated aerobic peat [Shannon and White, 1994; Nykanen et al., 1998; Bellisario et al., 1999]. However, water table level affected a number of other factors, and the indirect affects of these factors on methane emissions were likely of secondary importance in the fen and appeared to be more important than direct water table effects on methane emissions in the bog mesocosms. Similarly, the negative correlation between pore water methane concentrations and water table depth overall and in the bog mesocosms likely involved the indirect effects of water table level on the proximate controls over methane production in the anaerobic zone. These indirect effects are discussed below.

4.2. Direct Effects of Soil Warming on Pore Water Methane Concentrations and Methane Emissions

[41] Methane production in soils and sediments is typically a strong and positive function of temperature [Bridgham and Richardson, 1992; Updegraff et al., 1995a, 1995b; Kotsyurbenko et al., 1996; Nozhevnikova et al., 1997; McKenzie et al., 1998; Lafleur et al., 2005; Rivkina et al., 2007]. Previously we found a strong seasonal temperature effect on methane emissions in both the bog and fen mesocosms [Updegraff et al., 2001]. By focusing on growing season means in the current study, this seasonal temperature effect is eliminated and we examine only the effects of the IR treatments. Hence, our results should be interpreted in the context of the effect of climatic warming on seasonal mean methane dynamics.

[42] When examined with ANOVA, the IR treatments did not affect methane emissions (Table 1). However, methane emissions increased with seasonal mean soil temperature in the bog mesocosms (Figures 4 and 5), reflecting a range of soil temperatures within each IR treatment that result from the complicated biotic-energy flux interactions that have been previously demonstrated in this experiment [Bridgham et al., 1999; Noormets et al., 2004]. In contrast, there was a strong negative temperature effect on pore water methane in both the bog and fen mesocosms (Figures 1a, 2, and 3). As with the water table treatment, we suggest below that many of the temperature effects on methane dynamics were indirect.

4.3. Multivariate Controls Over Pore Water Methane Concentrations and Methane Emissions

[43] Our experimental design allowed us to examine the interactions among manipulated soil temperature and water table level, aboveground and belowground plant productivity, pore water chemistry, pore water methane, and methane flux. Our multivariate analyses suggest that the effects of the warming and water table treatments on methane dynamics was at least partially mediated through changes in these other factors. Despite the fact that this was a manipulative experiment, such complicated sets of intercorrelations make ascribing cause/effect for methane dynamics difficult. The regression models for methane fluxes in the two peatland types consisted of different sets of predictive variables (Figure 5). As discussed above, water table level appeared to be a dominant control over methane flux in the fen mesocosms. However, pore water sulfate, nitrate, and acetate and BNPP had significant negative partial correlations with both methane emissions and pore water methane concentrations in the fen mesocosms (Figures 3 and 5). In contrast, water table depth did not enter into the single accepted multiple regression model for methane flux in the bog mesocosms. BNPP had a positive effect on methane flux in the bog mesocosms and a negative effect on methane flux in the fen mesocosms. Similarly disparate sets of predictive factors entered into the multiple regression models for pore water methane concentrations in the bog and fen mesocosms (Figure 3).

[44] Some of these discrepancies between the two peatland types are probably explained by the low variance in the concentrations of many of the pore water parameters in the bog mesocosms; hence one might not expect them to have high predictive power for methane dynamics. This is particularly true of nitrate, ammonium, sulfate, and acetate. When the bog and fen mesocosms are considered together, a more consistent pattern begins to develop of the inhibitory effects of pore water ammonium, nitrate, sulfate, and acetate and positive effects of DOC on methane fluxes. Higher water table levels were also consistently associated with higher methane fluxes, even if the strength of this effect was uneven across community types.

[45] In the bog mesocosms, the highest concentrations of pore water ammonium and phosphate occurred at the lowest water table levels (Figure 6b). Similarly, in the fen mesocosms the highest concentrations of pore water nitrate, sulfate, and ammonium occurred at the lowest water table levels (individual r = −0.72 to −0.93), whereas pore water concentrations of DOC and acetate increased with water table level (Figure 6c). Higher soil temperatures under high IR were associated with higher concentrations of pore water nitrate and DOC in the bog mesocosms, and with lower BNPP and pore water phosphate concentrations in the fen mesocosms. Lower water tables and higher soil temperatures may have enhanced mineralization rates and oxidation processes, with the oxidized products diffusing into anoxic pore waters from the unsaturated zone above. We suggest that the water table and warming effects on these variables explain the observed negative relationship between soil temperature and water table level and pore water methane concentrations (Figures 2 and 3).

[46] One of the most consistent effects that we found was a negative relationship between pore water sulfate concentrations and pore water methane concentration and methane flux. Normally, elevated sulfate concentrations inhibit methanogenesis as more energetically favored sulfate-reducing bacteria out-compete methanogens for common substrates (mainly acetate and H₂) [Zinder, 1993]. Shannon and White [1996] found that seasonally low water tables in a Michigan bog led to high sulfate concentrations in pore water, and that methane production was delayed until after sulfate concentrations decreased. Several other studies confirmed the inhibitory effect of elevated sulfate on methane emissions from peatland soils [Vile et al., 2003; Gauci et al., 2002, 2004].

[47] Ammonium and nitrate also consistently showed a negative relationship with methane flux (Figure 4). However, the story with pore water methane concentrations was more complicated, with ammonium having a strong positive relationship and nitrate a negative relationship in the fen mesocosms, and nitrate having a positive relationship in the bog mesocosms (Figure 3). The interactions between N cycling, ammonium availability, methanogens, and methanotrophs involve a complex set of feedbacks between plant communities and microbial communities [Bodelier and Laanbroek, 2004]. Mineralization of organic N and an increase in the concentration of ammonium in pore waters may have had a stimulatory effect on both methanogen and methanotroph activity. This seems the most likely explanation for the positive effect of ammonium on pore water methane concentrations in the fen mesocosms, along with the negative effect of inorganic N on methane emissions in the bog and fen mesocosms. Normally, nitrate is inhibitory toward methanogenesis because, as with sulfate, it is a more energetically favorable electron acceptor [Zinder, 1993]. The positive effect of nitrate on pore water methane concentrations in the bog mesocosms (Figure 3) may have been through its effect (or simply correlation) with some other variable, e.g., DOC.

[48] Fertilization effects on methane cycling have been investigated directly in peatlands. Keller et al. [2006] demonstrated that high nitrogen concentrations inhibited methane oxidation in bog peat during 5-week lab incubations. However, long-term 6-year fertilization with lower concentrations of nitrogen stimulated rates of methane oxidation in bog peat. In contrast, no nitrogen effects on methane oxidation were observed in fen peat [Keller et al., 2006]. Overall, this prior study and our current study suggest that the role of nitrogen in methane oxidation can differ over the short and long-term and is strongly mediated by peatland type.

[49] Pore water acetate also had a negative relationship with both methane flux and pore water methane concentration in the fen mesocosms (Figures 3 and 5). Despite the fact that acetate is the substrate for the acetoclastic methanogenic pathway, there is abundant evidence that it can be inhibitory at even moderate concentrations to methanogenesis due to its conversion to acetic acid, particularly at low soil pH [Bridgham and Richardson, 1992; Bräuer et al., 2004]. We were apparently observing this inhibitory effect at the higher pore water acetate concentrations observed in the fen mesocosms (seasonal average up to 108 μM), although this effect was not apparent at the low, narrow range of acetate concentrations observed in the bog mesocosms (22–46 μM).

[50] Pore water DOC showed a positive relationship with both methane flux and pore water methane concentrations over the full range of DOC concentrations observed in the bog and fen mesocosms (Figures 2 and 4). The mean annual DOC concentration was almost twice as high in the bog mesocosms as in the fen mesocosms (94 versus 55 mg C L⁻¹, respectively). The IR and water table treatments had large effects on DOC concentrations and export that have been described previously [Pastor et al., 2003]. DOC is composed of a heterogeneous set of organic compounds, many of which are recalcitrant humic and fulvic acids [Mathur and Farnham, 1985], but it also contains lower concentrations of more labile organic compounds that apparently are important for fueling methanogenesis. DOC had a negative partial correlation with pore water methane in the bog mesocosms (Figure 3), which may reflect correlation with other variables or an actual inhibition by the very high DOC concentrations observed in these mesocosms.

[51] The rhizosphere represents a critically important environment for methane cycling in inundated soils. The oxic/anoxic interface around root surfaces likely contains microhabitats for both methane-oxidizing and methanogenic microorganisms, while the roots also provide a conduit for plant-mediated emission of methane [Brune et al., 2000]. Methanotrophs access oxygen leaking from roots while methanogens benefit from the labile substrates generated by root exudation [King, 1994; Watson et al., 1997; Gilbert and Frenzel, 1998; Bodelier et al., 2000; Chin et al., 2004]. In northern peatlands, increases in belowground productivity of roots has been shown to increase substrate supply to soil microbial communities while also enhancing plant-mediated transport of methane [Joabsson and Christensen, 2001; Ström et al., 2003, 2005; Ström and Christensen, 2007]. Our study underscores these complicated effects of the rhizosphere. BNPP had a positive association with methane flux in the bog mesocosms, but a negative association with methane flux and pore water methane concentrations in the fen mesocosms (Figures 3 and 5). BNPP was also higher in the bog than in the fen mesocosms. Our results suggest that an expansion of the rhizosphere in bogs may ultimately lead to an increase in methane emission rates, even though the species accounting for much of the root growth in our mesocosms (woody shrub species Andromeda glaucophylla and Chamaedaphne calyculata [Weltzin et al., 2000, 2003]), are likely to be relatively low emitters of methane [Shannon et al., 1996]. In contrast, enhancement of methane oxidation by oxygen leakage from roots appears to predominate in the fen mesocosms.

[52] There was a significant positive effect of ANPP on methane flux and pore water methane concentrations when these variables were examined in simple regressions (Figures 2 and 4), but that effect was not carried over to the multiple regression models for methane flux and entered into a small subset of the models for pore water methane concentrations in all mesocosms combined (Figures 3 and 5). Thus, even though aboveground plant litter typically was not in the anaerobic zone, it may have provided a labile carbon source for methanogenesis.

4.4. Methane Dynamics in Bogs Versus Fens

[53] The above discussion attempts to illuminate the different controls over methane dynamics in the bog and fen mesocosms, but it does not explain the 2.7 times higher methane flux and 3.4 times higher pore water methane concentrations observed in the bogs. This was despite the fact that the bog mesocosms were on average much drier than the fen mesocosms because of a rapid gain of soil carbon in the bogs and loss of soil carbon in the fens over time, with the corresponding changes in water table levels [Bridgham et al., 2008]. Most field studies report greater methane emissions in fens, but water table position and other factors are confounded in these studies. Lab incubations of bog and fen peat typically yield much higher methane production potentials in fen peat than in bog peat. In peat slurries from the source sites in this study, Bridgham et al. [1998] reported methane production rates that were 50 times greater in fen peat than in bog peat.

[54] Several factors may help explain these differences between peatland types. Given the highly recalcitrant nature of bog peat [Bridgham et al., 1998], it is likely that recent carbon inputs from plants is driving much of the methane production, as shown in other peatland studies [Chanton et al., 1995; Joabsson and Christensen, 2001; Ström et al., 2003, 2005; Ström and Christensen, 2007]. Thus, the greater methane fluxes that we observed in the bog mesocosms may be due to their higher BNPP and ANPP.

[55] Pore water chemistry was also very different between the bog and fen mesocosms, and variables associated with pore water had some of the strongest partial correlations with methane flux and pore water methane concentrations in our multiple regressions. The bog mesocosms had lower concentrations of nitrate, ammonium, sulfate, and acetate than the fen mesocosms, all of which showed strong inhibitory effects on methane fluxes. The bogs also had higher DOC concentrations, which may be an important carbon source for methanogenesis.

4.5. Stable Isotopic Evidence of Dynamic Methane Cycle

[56] Pore water methane in peatlands is produced through an autotrophic pathway (CO₂ reduction) and an acetoclastic pathway (acetate fermentation), and it is consumed by methanotrophic bacteria [Krumholz et al., 1995; Segers, 1998; Galand et al., 2005]. The stable isotopic signature of methane is affected in different ways by these microbial processes. Sugimoto and Wada [1993] reported δ¹³CH₄ from acetoclastic methanogenesis of −36‰. Lansdown et al. [1992] reported a mean δ¹³CH₄ = −73‰ in peatland pore waters where all of the methane derived from CO₂ reduction. Other published values of δ¹³CH₄ in peatland ecosystems have ranged from about −30‰ to −80‰ [Kelley et al., 1992; Martens et al., 1992; Avery et al., 1999, 2002]. In the absence of coincident methanotrophy, the end-member δ¹³CH₄ value in peatland pore waters is likely to be −32‰ for acetate fermentation and −78‰ for CO₂ reduction [Avery et al., 1999; Whiticar, 1999].

[57] The δ¹³CH₄ in pore water for mesocosms receiving high IR load was somewhat heavier, although the effect of warming appeared to be overridden by low water levels, so that mesocosms with low water levels had lighter pore water δ¹³CH₄ values regardless of IR loading. The effect of high IR loading exerted itself only in the mesocosms with a high water table, where the warm and wet fen and bog mesocosms contained pore water methane with mean δ¹³CH₄ values of −27.1 and −29.3‰, respectively; nearly 13‰ heavier than other treatments (Figure 1c). Avery et al. [1999] found that, in soils from similar peatland systems, acetate fermentation yielded mean δ¹³CH₄ of around −44‰ (heaviest values of around −32‰). Generally, δ¹³CH₄ values heavier than approximately −32‰ are associated with methane pools subject to losses from methanotrophic activity [Whiticar, 1999]. Thus, our results support significant methane oxidation in the warm, wet mesocosms.

[58] Isotope crossplots of δ¹³CH₄ versus δD-CH₄ for emitted methane (Figure 7a) and pore water methane (Figure 7b) provide an effective means of demonstrating treatment effects on methane production and oxidation pathways. With oxidation of methane being controlled by only one pathway, an isotopic “vector” can be used to represent the carbon and hydrogen isotopic shifts in methane resulting from methanotrophic activity. The two components of the oxidation vector are the ɛ_C and the ɛ_H of methanotrophy. Specifically, methane monooxygenase (MMO) shifts the residual methane pool to heavier values; ɛ_C ∼ 20‰ and ɛ_H ∼ 280‰ [Whiticar, 1999]. As methane is oxidized, the remaining methane pool will shift isotopically along this vector [Whiticar, 1999; Snover and Quay, 2000]. Values of ɛ_H are less well constrained than ɛ_C [Waldron et al., 1999]. The presence of methanotrophic activity in soil incubations from bog ecosystems has been confirmed using similar methods [Avery et al., 2002].

[59] The majority of the flux samples from the bog and fen mesocosms had coexisting C and H isotopic signatures consistent with the KIE of MMO activity, suggesting extensive oxidation (Figure 7a). In particular, fen mesocosms with lower water table emitted methane with distinctly heavy carbon and hydrogen isotopic compositions (mean δ¹³CH₄ = −21.5‰; mean δD-CH₄ = −120‰). It appears that lower water tables stimulated oxidation of methane within the peat of mesocosms, before emission of methane to the atmosphere.

[60] The isotope trajectories of pore water methane in bogs suggest a significant effect of the methanogenesis pathway in addition to oxidation effects. Note that the majority of the pore water values for δ¹³CH₄ and δD-CH₄ in bog mesocosms were consistent with acetate fermentation as the primary production pathway (Figure 7b). Previous studies have found that the acetoclastic fermentation pathway predominates in fens and the autotrophic pathway predominates in bogs [Keller and Bridgham, 2007], although there is evidence that the acetoclastic pathway may predominate in bogs during the spring temperature transition [Avery et al., 1999, 2002]. Pore water acetate concentrations in bogs in this study were consistently between 20 and 40 μM (Figure 2), which is sufficient to support acetoclastic methanogenesis [Fukuzaki et al., 1990; Schulz and Conrad, 1996], particularly in low-temperature terrestrial ecosystems [Kotsyurbenko, 2005]. The greater belowground shrub root production in the dry bogs during the preceding 4 years [Weltzin et al., 2000, 2003] may have enhanced the availability of labile substrates associated with root exudates that are critical to methane production [Joabsson and Christensen, 2001]. If root production is stimulated by warmer and drier conditions in bogs, there may be an associated increase in acetoclastic methane production in these systems. Increased availability of labile C seems to favor acetoclastic methanogenesis over autotrophic methanogenesis in peatlands, particularly during the growing season [Avery et al., 1999].

[61] In fens, the majority of the pore water values for δ¹³CH₄ and δD-CH₄ fell outside the production zone (Figure 7b), suggesting that methanotrophic activity was affecting methane pools of deep pore waters, particularly in warm, wet mesocosms (Figure 1c). This result is consistent with the strong effect of water table level on methane emissions in the fens found in our other analyses. While acetoclastic methanogenesis probably predominated in the fens, as found in other studies, the acetate appears to have accumulated to inhibitory levels. Acetate accumulation in the anaerobic zone of peatlands is a poorly understood, if seemingly relatively frequent occurrence [Duddleston et al., 2002; Bräuer et al., 2004; Keller and Bridgham, 2007].