Controls on ecosystem and root respiration across a permafrost and wetland gradient in interior Alaska

Permafrost is common to many northern wetlands given the insulation of thick organic soil layers, although soil saturation in wetlands can lead to warmer soils and increased thaw depth. We analyzed five years of soil CO2 fluxes along a wetland gradient that varied in permafrost and soil moisture conditions. We predicted that communities with permafrost would have reduced ecosystem respiration (ER) but greater temperature sensitivity than communities without permafrost. These predictions were partially supported. The colder communities underlain by shallow permafrost had lower ecosystem respiration (ER) than communities with greater active layer thickness. However, the apparent Q10 of monthly averaged ER was similar in most of the vegetation communities except the rich fen, which had smaller Q10 values. Across the gradient there was a negative relationship between water table position and apparent Q10, showing that ER was more temperature sensitive under drier soil conditions. We explored whether root respiration could account for differences in ER between two adjacent communities (sedge marsh and rich fen), which corresponded to the highest and lowest ER, respectively. Despite differences in root respiration rates, roots contributed equally (∼40%) to ER in both communities. Also, despite similar plant biomass, ER in the rich fen was positively related to root biomass, while ER in the sedge marsh appeared to be related more to vascular green area. Our results suggest that ER across this wetland gradient was temperature-limited, until conditions became so wet that respiration became oxygen-limited and influenced less by temperature. But even in sites with similar hydrology and thaw depth, ER varied significantly likely based on factors such as soil redox status and vegetation composition.


Introduction
Wetlands represent a major land cover class in boreal and subarctic regions, occupying about 25% of the total boreal Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. region land area (Vitt 2006, Wieder et al 2006. Wetlands in the discontinuous permafrost zone range from poorly drained conifer bogs with shallow permafrost to permafrost-free inundated marshes and fens that are too wet to support trees. These wetland types could be impacted differently by climate change, particularly because of strong relationships between soil temperature, soil moisture, and active layer depth. Frozen ground impedes the drainage of surface soils, and thus maintains saturated conditions that reduce rates of decomposition (cf Ford and Bedford 1987, Kane et al 2010, Roulet and Woo 1986 and inhibit soil carbon losses through fire combustion (Trumbore and Harden 1997). In these systems, increasing thaw depth throughout the growing season regulates soil drainage but also exposes more organic carbon to decomposition.
Permafrost in interior Alaska has been showing widespread evidence of thawing (Osterkamp et al 2000, Jorgenson et al 2001. Because deep peat impedes soil drainage, changes in active layer depth in wetlands typically lead to more saturated conditions either in the form of thermokarst ponds, collapse scar bogs, or fens. Permafrost thaw in hydrologically connected uplands also can influence runoff or upwelling in wetlands, contributing to saturation. Permafrost may be particularly vulnerable to thaw where heat flow from groundwater movement maintains warmer soil conditions at depth; for example where heat can be transferred from groundwater fens to adjacent permafrost in forest margins (Racine and Walters 1994). Groundwater influences can also contribute to the maintenance of taliks or unfrozen zones in permafrost (Racine and Walters 1994). Overall, permafrost thaw related to wetlands and soil saturation in the discontinuous permafrost zone have the potential to alter soil climate, organic layer thickness, and vegetation structure, all of which are likely to influence respiration and CO 2 release from soils.
Soil water content and soil temperature interactions with thaw depth are likely to generate positive relationships between water availability and soil temperature in northern wetlands (i.e., warm and wet conditions occur with increasing thaw depth). Warming usually stimulates heterotrophic respiration and increases soil CO 2 fluxes, but can have variable effects on autotrophic respiration (cf Dorrepaal et al 2009, Bronson and Gower 2010, Hicks Pries et al 2013. More saturated soil conditions generally inhibit microbial respiration due to lack of O 2 but also could inhibit plant productivity and respiration due to reduced nutrient availability in the rooting zone. Likely as a result of these interactions, several studies have found no or weak relationships between ecosystem respiration (ER) and variation in water table position (Silvola et al 1996, Lafleur et al 2005, Dimitrov et al 2010. Here we analyzed growing season ER and relationships with seasonal thaw depth, soil temperature, and soil moisture among five wetland communities that varied in permafrost and hydrologic conditions. Our measurements were collected along a physical gradient spanning from a conifer permafrost forest to an open saturated fen. Ecological gradients are useful because they capture variation in community structure associated with long-term processes along with variation in important environmental variables such as soil moisture and soil temperature (Muller 1998). Within and across wetland communities, we hypothesized that ER would be lower but more temperature sensitive in the vegetation communities with permafrost than those that lacked frozen material in the upper 1 m of soil. Given that soil saturation can occur when seasonal ice is shallow (when soil interception is constrained by near surface ice) or when soil thaw is deeper (i.e., taliks that do not freeze due to deep water, thermokarst features that become saturated), we predicted that ER would exhibit a unimodal relationship with thaw depth across communities along the wetland gradient. Within each wetland community, we predicted that ER would show a positive relationship with depth to seasonal ice, since increasing thaw depth corresponds to warmer soil temperatures and greater unfrozen organic soil stocks.

Description of wetland gradient
Data were collected at the Alaskan Peatland Experiment (APEX) sites located near the Bonanza Creek Experimental Forest in Alaska (64.82 • N, 147.87 • W). The mean annual air temperature for the Tanana River Valley ranges from −5 to −7 • C and mean annual precipitation is between 215 and 300 mm (Hinzman et al 2006). Our gradient is composed of five wetland communities that extend from the toe slope of an adjacent upland forest into a large fen complex (Waldrop et al 2012). Adjacent communities are separated by 25-50 m, including (in order from toe slope to center of the large fen complex): (1) black spruce forest underlain by permafrost with a understory dominated by feather moss and ericaceous shrubs, (2) shrub system dominated by willow (Salix spp.), birch (Betula spp.) and Sphagnum species (

Atmospheric and soil environmental variables
Air temperature, relative humidity, and precipitation data were collected at the site continuously using CR10X data loggers (Campbell Scientific Inc., Logan, UT) and are available in the Bonanza Creek LTER data archive (table 2). Soil temperatures were measured manually at 10 cm at each community during each ER flux campaign using thermistors. Soil moisture was measured continuously at various depths (5, 25, 50 cm) using CS615 and CS616 TDR soil moisture probes (Campbell Scientific Inc., Logan, UT). We calibrated the TDR probes using the method for organic soils outlined in Bourgeau-Chavez et al (2010). Surface soil moisture also was measured manually during each ER flux using a ThetaProbe soil moisture sensor (Delta-T Devices, Cambridge, England) inserted 10 cm vertically into the surface organic layer. During each flux measurement, we manually measured water table position in shallow wells (∼1 m depth) permanently placed in the peat surface and seasonal thaw depth using tile probe inserted into the peat. Soil environmental variables were analyzed using repeated measures analysis of variance and Tukey's post hoc comparison of means (SAS Proc Mixed), with ecosystem and year as fixed effects. Table 1. Dominant species and mean growing season soil temperature ( • C) at 10 cm, soil moisture (% VMC) at 5 cm, organic soil layer depth (cm), and seasonal thaw depth (cm) for each community along the wetland gradient. Below, communities along the gradient are listed in order from the toe slope of an adjacent forest to the center of the large fen complex. A visual schematic of this wetland gradient can be found in Waldrop et al (2012

Ecosystem respiration (ER) fluxes
Ecosystem respiration of CO 2 was measured at each community from mid-May to September approximately every two weeks from 2007 to 2011. Details on the flux methods are provided in Chivers et al (2009). Briefly, one gas flux collar constructed of galvanized steel was permanently installed into the surface soils (∼5 cm depth) of each community except at the rich fen, where three gas flux collars were installed as part of a larger experiment. Due to the lack of spatial replication within most of the community types found along the wetland gradient, the results are specific to the particular locations of each flux collar, which were installed in conditions representative of each habitat type. At the start of each flux measurement, a clear chamber (dimensions of 0.37 m 2 × 26 cm deep) constructed of an aluminum frame with Teflon siding (American Durafilm, Holliston, MA) was placed on each gas flux collar. The chamber was covered with a dark shroud to block any incoming PAR. We used a PP-systems EGM-4 infrared gas analyzer (IRGA, Amesbury, Massachusetts) to measure CO 2 concentrations inside the chamber every 1.6 s for 2-3 min. Temperature, relative humidity, and PAR were recorded continuously inside the chamber using a PP-systems TRP-1 sensor attached to the inside of the chamber. A CO 2 flux (µmol CO 2 m −2 s −1 ) was calculated from the slope of the linear relationship between time and the CO 2 headspace concentration.
To analyze differences in mean ER across the five communities and study years (2007)(2008)(2009)(2010)(2011), we used a repeated measures analysis of variance (SAS Proc Mixed) and Tukey's post hoc comparison of means. We modeled the dependence of ER, including both autotrophic and heterotrophic respiration, on seasonal thaw depth as: where ER is the observed CO 2 flux (µmol CO 2 m −2 s −1 ), R max is the estimated maximum ER when seasonal thaw depth is optimal for plant and heterotrophic respiration (µmol CO 2 m −2 s −1 ), F is the observed thaw depth (cm), uR is the estimated optimal seasonal thaw depth for respiration (cm), and tR is an estimated measure of the width of seasonal thaw depth amplitude (cm). This nonlinear model was modified from Tuittila et al (2004) and here is used to examine data across communities along the gradient. The relationship between seasonal thaw depth and ER within each community was examined using linear regression models. The temperature dependence of ER was modeled as: where A is the estimated ER at 0 • C (µmol CO 2 m −2 s −1 ), Q 10 is the estimated temperature dependence of ER (unitless), and T is the observed soil temperature ( • C) at 10 cm below the peat surface. We estimated parameters from equations (1) and (2) first using instantaneous ER and environmental variables measured during the flux measurements. Parameters also were estimated using ER and environmental data averaged by month across years. All estimated parameters throughout the paper were calculated using Proc NLIN in SAS. When analyzing the temperature dependence of ER, we identified a statistical outlier for the grass community that represented a very low ER value for a high soil moisture level (greater than 99%). This single outlier was removed from all subsequent analyses. For all linear and nonlinear analyses, we used an alpha value of 0.05 to determine significance.

Root respiration and vegetation measurements
We conducted two root experiments, the first of which was conducted in 2010. We installed five new flux collars in the rich fen and measured ER fluxes for each collar around peak biomass (mid-July) following the methods above. This was repeated daily for two weeks, after which the per cent cover of mosses and vascular vegetation was visually estimated inside each of the collars. A 10 cm × 10 cm aboveground biomass sample was randomly harvested inside each collar. We also collected vascular stem density inside each collar, and measured vascular green area (VGA) on all of the species found within the collar (m 2 of vascular leaf area m −2 ) following Chivers et al (2009). Root respiration was then measured on the three dominant vascular species (Carex spp., Equisetum fluviatile, Potentilla palustris) within each collar by destructively harvesting one of each species found within the collar, and picking all fine roots. The three dominant species each comprised between 20 and 40% of the total aboveground vascular biomass in all collars for both years. Fine roots were rinsed and immediately placed inside of a 5 cm diameter (0.0082 m 3 ) aluminum root cuvette (following Burton and Pregitzer 2003) and a flux was measured using the EGM-4 infrared gas analyzer, which measured the CO 2 concentration inside of the chamber every 1.6 s until concentrations reached around 1000 parts per million (ppm). This took 5-10 min. In all cases, fluxes were linear and there was no evidence of higher respiration rates at lower headspace CO 2 concentrations (cf Burton and Pregitzer 2002). Temperatures were continuously monitored inside the cuvette, which allowed us to later standardize all fluxes to the same soil temperature. Roots were placed in a cooler and transported back to the lab where field moisture content was determined by weighing and then drying at 65 • C to a constant mass. These data allowed us to calculate root respiration for each species on a per g dry mass basis.
After root fluxes were measured in the field, three soil cores (5 cm diameter, 20 cm length) were collected from within each collar, placed in a cooler, and transported back to the lab. Cores were cut in half length-wise and half of each core was analyzed for bulk density while fine roots were separated from soil with tweezers from the other half of each core, dried to a constant mass, and weighed to obtain a measure of root biomass (g root biomass m −2 ).
We conducted this experiment as a pilot study in the rich fen in 2010. Because our analyses showed that mean ER was highest at the sedge marsh and lowest at the rich fen, we repeated this experimental design in both of these communities in 2011. For the 2011 experiment, we installed five new collars at the rich fen and also installed five collars at the sedge marsh, and repeated the methods outlined for the 2010 experiment. Root respiration fluxes were standardized to a g m −2 basis using a weighted average based on species per cent cover and root biomass estimates at a collar level. Root respiration was also standardized to the average 10 cm soil temperature during flux measurements to account for differences in temperature between the soil and the chamber atmosphere (collected using thermistors). During our 2010 measurements, one collar had root respiration values that exceeded ER, and a higher variance in root respiration relative to the other collars. The studentized residual error for this measurement location was −3.30, thus we removed this data as an outlier. For all other data, we analyzed root respiration as a predictor of ER using a linear regression (SAS Proc GLM).

Patterns in soil environment and ER across a wetland gradient
While surface soil temperature, volumetric soil moisture content, seasonal thaw depth, and water table position all varied by a community × year interaction (table 3), communities tended to differ predictably in environmental conditions. In general, soils were colder and drier in the black spruce and shrub communities and were warmer and wetter in the more open marsh and fen (table 1). Seasonal thaw depth was shallowest at the black spruce community and increased along the gradient into the rich fen. These patterns in soil conditions were consistent among sampling years with a few exceptions. The sedge marsh and rich fen had much deeper seasonal thaw depths in 2008 and 2009, when the general study region experienced a severe natural flood (Wyatt et al 2012), than in the other years.
Averaged across years, ER was highest in the sedge marsh and grass communities, which did not differ from one another and averaged 4.85 ± 0.27 µmol CO 2 m −2 s −1 . Ecosystem respiration in the black spruce, shrub, and rich fen communities did not differ from one another, and averaged 2.70 ± 0.09 µmol CO 2 m −2 s −1 . While instantaneous ER varied by a community × year interaction (table 3), this general trend across communities was consistent in most study years. Across all vegetation community types and years, mean seasonal ER peaked in July-August (least squared mean value of 4.29 µmol CO 2 m −2 s −1 ), compared with May-June (least squared mean value of 3.15 µmol CO 2 m −2 s −1 ) or September (least squared mean value of 2.94 µmol CO 2 m −2 s −1 ).

Effects of soil environmental variables on ER
Across communities, there was a significant unimodal relationship (following equation (1) 2). Across all vegetation communities and years, depth to seasonal ice was positively correlated with surface soil temperature (Pearson correlation coefficient = 0.63, p < 0.0001) but was not significantly correlated with water table position or soil moisture. Surface soil temperature was positively correlated with surface soil moisture content (Pearson correlation coefficient = 0.41, p < 0.0001).
Instantaneous ER was more sensitive to temperature in the sedge marsh and shrub communities than the other communities (table 4). Not surprisingly, the predictive ability of these relationships declined with smaller Q 10 values (table 4). Monthly averaged ER had lower temperature sensitivity (estimated via Q 10 ) in the rich fen than the other communities (figure 3). Estimates of A (respiration at 0 • C) based on instantaneous data were highest at the grass community and lowest at the shrub and rich fen communities (table 4). While estimates of A based on monthly averaged data were similar to those reported for the instantaneous ER models, Q 10 values from the monthly model in each community tended to be higher than those derived from instantaneous data (table 4). Mean monthly Q 10 values increased as the mean depth to water table increased across all communities (figure 4). Mean monthly Q 10 was not related to depth to seasonal ice (F = 1.43, p = 0.28).

Contributions of root respiration to ER
Averaged across collars, ER flux during the experimental campaigns was 0.143 ± 0.015 mg CO 2 m −2 s −1 at the rich   (2)) analyzing the dependence of ecosystem respiration on soil temperature. Instantaneous ER was analyzed using soil temperature at the time of flux measurements, while mean monthly ER was analyzed using mean monthly soil temperatures across the entire growing season as well as peak biomass (not possible to run at the rich fen).
A (µmol CO 2 m −2 s −1 ) Q 10 P; F (num,den) ; R 2  fen and 0.262 ± 0.015 mg CO 2 m −2 s −1 at the marsh. Root fluxes averaged 0.003 ± 0.001 µmol CO 2 s −1 g −1 at the rich fen and 0.004 ± 0.000 02 µmol CO 2 s −1 g −1 at the sedge marsh. We used root biomass (rich fen: 494.13 ± 281.72 g dry mass m −2 ; sedge marsh: 605.15 ± 123.38 g m −2 ) to scale root respiration. On a m 2 basis, mean root fluxes were 0.059 ± 0.02 mg CO 2 m −2 s −1 at the rich fen and 0.095 ± 0.007 mg CO 2 m −2 s −1 at the sedge marsh. We observed no change in respiration rates as chamber CO 2 concentrations changed from ∼400 to 1000 mg l −1 , which suggests little to no CO 2 acclimation effect (Burton and Pregitzer 2002). Root respiration contributed to 40% of ER in the rich fen in both years, and 37% of ER at the marsh. In 2011, when we measured root respiration in both systems, both root respiration and ER fluxes were higher in the marsh than in the rich fen. Across both communities and years (rich fen only), root respiration (mg CO 2 m −2 s −1 ) was a strong predictor of ER, explaining 60% of the variation in ER (figure 5). At the rich fen, soil temperature and seasonal thaw depth during our root respiration experiment (mid-July measurements only) explained 60% and 20% of the variation in ER respectively (figures 6(a) and (b)). We found that ER was not related to surface soil moisture (figure 6(c)) or water table position (not shown). The two communities differed in their relationships between ER and plant biomass. At the rich fen, there was a positive relationship between ER and root biomass, while at the sedge marsh this relationship was not significant (figure 6(d)). Conversely, there was a trend towards a positive relationship between VGA and ER at the marsh, but not at the rich fen (figure 6(e)).

Discussion
This study analyzes a 5-year dataset of ER across a wetland gradient of distinct vegetation communities varying in permafrost and soil moisture. Due to the lack of spatial replication within each community type (except the rich fen), we do not know how generalizable our results are to these community types in general. Because the wetland gradient encompasses environmental variation driven by long-term community assemblage, our results are likely to differ from shorter-term manipulation of environmental conditions in wetlands. For example, while short-term flooding in wetlands can enhance soil nutrient pools and stimulate ecosystem productivity, CO 2 transport, and respiration (cf Wyatt et al 2012), longer-term flooding and inundation often inhibits respiration due to anoxia. Overall, several of our hypotheses regarding environmental controls on ER were supported. As predicted, communities with permafrost on average had reduced ER relative to the communities without permafrost. Also as predicted, across the gradient ER peaked at intermediate soil thaw depths. Finally, because we found that mean ER and apparent Q 10 values varied considerably between the sedge marsh and rich fen, two adjacent communities that lacked permafrost, we examined the contributions of root respiration to ER in these adjacent systems. We found that despite differences in mean root respiration rates, root respiration contributed similarly to ER in both communities. These findings have implications for modeling C fluxes in high latitude wetlands, and are described in more detail below.

Thaw depth, soil temperature, and ecosystem respiration
We predicted that ER would peak at intermediate soil thaw depths across the wetland gradient, given that soil saturation can result from both shallow and deep soil thaw depth scenarios, potentially suppressing both heterotrophic and autotrophic respiration by limiting oxygen and nutrient availability. Our results supported this prediction, as across wetland communities we observed a unimodal relationship between soil thaw depth and ER. Estimates of uP (optimal soil thaw depth supporting maximum ER) were similar for relationships between thaw depth and instantaneous ER (63 ± 34 cm) and mean monthly ER (51 ± 4 cm; figure 1).
The presence of soil ice (seasonal ice or permafrost) also can govern ER fluxes by regulating soil temperature and access to unfrozen substrate. As a result, we predicted that within each community, there would be positive relationships between ER and soil depth. Though frost depth and soil temperature were positive correlated, we found no evidence of linear relationships between frost depth and ER within any of the communities (figure 2). Instead, within each community, the highest ER tended to occur either at intermediate or shallow soil thaw depths. Relationships between ER and soil thaw depth can be confounded by the seasonal development of vegetation, given that plant respiration (aboveground or belowground) is an important contributor to ER. Without specific data on seasonal changes in biomass or leaf area, we were not able to test how vegetation influenced ER across the gradient, and whether small ER values under deep thaw conditions were due simply to vegetation senescence. Examining the relationship between ER and soil thaw depth across three seasonal periods (figure 1) clearly shows evidence of some, but not complete, confoundment between season and thaw depth.
Finally, we predicted that permafrost communities would have lower ER but greater temperature sensitivity than communities lacking in surface permafrost. Across the gradient, the black spruce and shrub communities with shallow permafrost (maximum seasonal thaw depth of less than 1 m) had colder surface soils (between 1 and 4 • C colder during the growing season) than the communities with maximum seasonal thaw depth greater than 1 m (grass, sedge marsh, and rich fen). Likely at least in part due to these differences in soil temperature, ER in the black spruce and shrub communities was lower than the other communities averaged across study years.
While both our mean ER and apparent Q 10 values generally fall within the range of reported values for northern wetlands (Raich and Schlesinger 1992, Lafleur et al 2005, Bronson et al 2008, our results clearly show that the wetland communities diverged in the functional relationship between ER and soil temperature. Monthly ER in the black spruce, shrub, grass, and marsh communities fell along the same relationship with soil temperature, corresponding to a gradient-wide apparent Q 10 of 5.2 (5.6 based on mid-season data only, excluding periods of green-up or plant senescence). This Q 10 is higher than many previous soil respiration studies, but is not atypical for boreal systems (Bronson et al 2008). The Q 10 of ER was smaller in the rich fen, which could simply reflect decreasing temperature sensitivity with increasing temperatures (Lloyd and Taylor 1994). This also could indicate that other factors that co-vary with soil temperature along this gradient could be posing constraints on respiration processes. Increasing substrate quality (i.e. decreasing soil C/N/P ratios) also can result in lower temperature sensitivities due to lower activation energies for metabolic processes (Davidson and Janssens 2006). However, surface soil C/N ratios along this gradient have been shown to be lowest in the sedge marsh and not the rich fen (see Waldrop et al 2012; sedge marsh termed 'emergent fen' in that study), which is not consistent with the observed lower temperature sensitivity within the rich fen. Alternatively, soil saturation and anaerobiosis could reduce temperature sensitivities within the rich fen (Davidson and Janssens 2006). The rich fen had the highest soil moisture content (table 1), the shallowest average water table depth (Waldrop et al 2012), and the highest rates of CH 4 flux (Waldrop et al 2012) relative to the other vegetation communities along this gradient. These saturated and more anaerobic conditions could explain lower temperature sensitivities within the rich fen as oxygen constraints are placed on the microbial community. Moreover, we observed a strong negative relationship between water table position and apparent Q 10 values across all communities (figure 3), showing a gradient-wide pattern of ER becoming increasingly sensitive to temperature with lower water table positions beneath the moss surface (drier). This pattern can be explained by (1) removal of anaerobic constraints on respiration as water table declines and Q 10 increases, (2) seasonally warmer soils when water table depth is lowest and/or (3) deeper active regions of microbial activity accessing lower quality substrates and therefore producing higher apparent temperature sensitivities. These results indicate a complex interactive set of explanatory variables related to anaerobiosis, temperature, and substrate quality, but it is a complex behavior that can be captured within a simple ecosystem metric, water table depth. Similar to trends in soil thaw depth, relationships between mean monthly ER and temperature can be influenced by the seasonal development of vegetation. However, two lines of evidence suggest that the cross-community Q 10 value (figure 3) is robust to this seasonal confoundment. First, when we constrained our mean monthly model to periods of peak biomass (July and August; excluding periods of leaf-out and senescence), the gradient-wide Q 10 value was not reduced (table 4). During this period, observed ER was much lower at the rich fen than at the other communities. Second, within each of the vegetation communities except the rich fen, mean monthly Q 10 values tended to increase over time, with declines at the end of the growing season in some community types (table 5). At the rich fen, mean monthly Q 10 varied little through the growing season, ranging from 1.1 ± 0.2 to 1.7 ± 0.4. Averaged across all vegetation community types and years, mean seasonal Q 10 peaked in July-August (mean ± SE of 4.9 ± 12.2), compared with May-June (of 1.2 ± 0.7) or September (of 1.3 ± 3.8).
These results have implications for modeling soil CO 2 fluxes from high latitude wetlands. We speculate from our collective results that ER was regulated by temperature across most of the wetland gradient, including all of the permafrost habitats. In the rich fen, soil conditions were warmer, likely driven by heat transfer of groundwater inputs, and temperature was less relevant to ER. This suggests that modeling CO 2 fluxes across a wide range of vegetation and permafrost conditions requires an understanding of this 'tipping point' where the system shifts from being mainly temperature-limited to being limited by other conditions such as oxygen.

Contributions of root respiration to ER
Ecosystem respiration is composed of both microbial and plant (above and belowground) respiration and thus to fully examine the environmental controls on ER we partitioned ER between these sources (belowground plant respiration only) within the sedge marsh and rich fen communities. While rates of plant and microbial activity are generally optimal at an intermediate level of soil moisture, these relationships may differ among these two groups such that microbial respiration may be more limited than plants at high soil moisture and plants may be more limited than microbes under dry conditions (Wickland and Neff 2007). For example, although saturated conditions can inhibit microbial respiration and also can cause plant stress due to reduced availability of oxygen and buildup of toxins in the root zone (Pezeshki 2001, Mitsch andGosselink 2007), hydric plant adaptation has led to the development of large air spaces in the cortex of the rhizomes and roots, which allows the transport of O 2 to the roots, thus mitigating the impact of soil saturation on plant processes (Justin and Armstrong 1987, Armstrong et al 1991. The response of ER to changing water availability will depend both on the response of autotrophic versus heterotrophic respiration as well as the characteristics of the vegetation community. Few studies have explicitly considered the sensitivity of these components of ER to changing moisture conditions in northern wetlands (Sulman et al 2012).
Our research site gave us a unique opportunity to examine variation in root respiration among different wetland communities, given that the highest and lowest mean ER was observed at the adjacent marsh and fen communities. We explored whether varying contributions of root respiration could account for these differences in ER. Our estimates of root respiration showed that root respiration was higher in the rich fen compared to the marsh (0.003 ± 0.0001 µmol CO 2 s −1 g −1 dry mass at the rich fen and 0.004 ± 0.0002 µmol CO 2 s −1 g −1 dry mass at the marsh) and are slightly higher than estimates from an Alaskan lowland conifer forest (Ruess et al 2003). Our excised root respiration measurements may be higher than total intact root respiration rates because we selected for relatively young < 2 mm diameter roots, which have been shown to have higher respiration rates than older, larger diameter roots (Billings et al 1977;see Lawrence and Oechel 1983 and references therein). However, this bias would not be a strong factor in wetlands dominated by grasses and sedges, which exhibit more consistent root morphology. When root respiration is scaled to a m 2 basis, our flux estimates in the rich fen (0.143 ± 0.015 mg CO 2 m −2 s −1 ) are significantly lower than the marsh (0.262 ± 0.015 mg CO 2 m −2 s −1 ), and are similar to previous estimates for a boreal black spruce forest (Vogel et al 2005). Our experiment indicated that root respiration across the sedge marsh and rich fen communities contributed 14-62% (mean 40%) to ER. This agrees with other high latitude studies that have found that root respiration contributes approximately 40-60% of ecosystem respiration (Silvola et al 1996, Hanson et al 2000, Ruess et al 2003, Bond-Lamberty et al 2004, Griffis et al 2004, Crow and Wieder 2005. While fluxes of both root respiration and ER were higher at the marsh community than the rich fen, the fraction of ER that was root respiration was similar in both systems. Thus, ER fluxes in the rich fen likely are low relative to the other gradient communities because of both reduced autotrophic and heterotrophic respiration.
Despite similar levels of root biomass and VGA, we found a positive relationship between root biomass and ER at the rich fen but not the marsh. On the other hand, ER in the sedge marsh might be responding more to variation in aboveground vascular biomass than in the rich fen. Yet given the similar levels of plant biomass in the two communities, total ER in the rich fen does not appear to be constrained by low values of plant biomass, either aboveground or belowground. It seems more likely that high soil moisture is limiting both microbial respiration and plant belowground activity in the rich fen, likely through constraints on oxygen and nutrient availability.
Determining how sources of autotrophic respiration (aboveground or belowground) and heterotrophic respiration contribute to CO 2 flux or respond differently to environmental variables is important for modeling studies predicting the fate of carbon in future scenarios with climate change. Our results show that ER in adjacent wetland communities with similar permafrost and soil moisture conditions can vary tremendously, and demonstrate different relationships with vegetation components. Direct measures of soil redox status or oxygen availability, nutrient availability, and CO 2 fluxes from different plant tissues would provide a stronger mechanistic understanding of variation in CO 2 fluxes from northern wetlands. This is particularly important given that increasing thaw depth and loss of permafrost are altering both soil environments and vegetation structure in northern wetlands.