Methane fluxes between terrestrial ecosystems and the atmosphere at northern high latitudes during the past century: A retrospective analysis with a process-based biogeochemistry model

2.1. Model Framework

[6] We have developed an hourly time step methane dynamics module (MDM) for TEM that explicitly considers the process of CH₄ production (methanogenesis) as well as CH₄ oxidation (methanotrophy) and the transport of the gas from the soil to the atmosphere. We have coupled the MDM with several existing TEM modules (Figure 1a): the core carbon and nitrogen dynamics module of TEM 5.0 (CNDM) [Zhuang et al., 2003], the soil thermal module (STM) that includes permafrost dynamics [Zhuang et al., 2001], and an improved and expanded hydrological module (HM) [Zhuang et al., 2002] that simulates water movement across an atmosphere-vegetation-soil continuum. For northern ecosystems, the soil component of the HM considers moisture dynamics explicitly in moss, organic soil, and mineral soil layers [Zhuang et al., 2002], and is designed to consider fluctuations in water table depth.

2.1.1. Methane Dynamics Module

[7] Fluxes of methane between soils and the atmosphere depend on the relative rates of methane production and oxidation within the soil profile and the transport of methane across the surface of soils. We assume that soils can be separated into an upper unsaturated zone and a lower saturated zone according to the water table depth. Methanotrophy (methane oxidation) occurs in the unsaturated zone and methanogenesis (methane production) occurs in the saturated zone. Because methanotrophy reduces soil methane concentrations in the unsaturated zone and methanogenesis increases soil methane concentrations in the saturated zone, the resulting concentration gradient causes methane to diffuse from the saturated zone into the unsaturated zone. If the rate of methanogenesis is larger than the rate of methanotrophy within the soil profile, such as occurs in wetland soils, methane will be emitted to the atmosphere through diffusion. There are two other pathways in addition to diffusion that can be important for CH₄ transport to the atmosphere. Soil CH₄ can be transported from deep layers in sediments and soils through “hollow tubes” running from the roots through the stems to the leaves of some plants (plant-aided transport). If the water table is above the soil surface, methane can also move in bubbles through the overlying water and escape to the atmosphere. This transport process is known as ebullition.

[8] If the rate of methanotrophy is greater than the rate of methanogenesis within the soil profile, then most, if not all, of the methane produced in the saturated zone will be oxidized in the unsaturated zone and little or no CH₄ will be emitted from soils. Indeed, if the rate of methanogenesis is negligible, methanotrophy may cause a concentration gradient to develop that causes methane to diffuse from the atmosphere into the soil, such as occurs in well-drained upland soils. In this situation, soils are said to “consume” atmospheric methane.

[9] To simulate methane dynamics within the soil, we divide the soil column into a layered system with 1-cm increments from an upper boundary (i.e., the soil surface or water surface if the water table is above the soil surface) to a lower boundary (L_B), which represents the depth of microbial activity (Figure 1b). The L_B depends on active layer (i.e., unfrozen soil) depth as simulated by the soil thermal module. If the active layer depth is deeper than the maximum depth of microbial activity prescribed for an ecosystem (L_MAXB; see Table 1), the L_B is equal to L_MAXB; otherwise the L_B is equal to the active layer depth.

Table 1. Parameters of the Methane Dynamics Module for Major Ecosystem Types in the Northern High Latitudes^a

	Alpine Tundra/Polar Desert		Moist/Wet Tundra		Boreal Forest		Unit
	Wetland (Toolik-D)b	Upland (Tundra-NS)	Wetland (Toolik-W)	Upland (Tundra-UI)	Wetland (SSA-FEN)	Upland (B-F)
LMAXB	100	100	100	100	110	100	cm
Methanogenesis
MGO	0.45	0.45	1.0	0.45	1.3	0.8	μmol L−1 h−1
NPPMAX	100	100	150	100	250	250	gC m−2 month−1
PQ10	3.5	3.5	4.0	3.5	4.5	7.5	–
TPR	−3.0	8.0	−5.5	8.0	10.0	7.0	°C
Methanotrophy
OMAX	35	1.0	30	2.0	15	1.0	μmol L−1 h−1
KCH4	5.0	10.0	5.0	5.0	5.0	15	μmol L−1
OQ10	3.5	0.8	2.2	1.1	1.9	1.5	–
TOR	−3.0	5.0	−5.5	5.5	10.0	5.4	°C
MVMAX	1.0	0.9	1.0	0.7	1.0	1.0	% volume
MVMIN	0.0	0.0	0.0	0.0	0.0	0.2	% volume
MVOPT	0.5	0.4	0.5	0.3	0.5	0.6	% volume

• a See text or notation section for the definition of variables.
• b Names in parentheses represent sites used to calibrate the methane dynamics module; see Table 2.

[10] Within each soil layer, changes in CH₄ concentration are governed by the following equation:

where C_M(z, t) is the soil CH₄ concentration in μmol L⁻¹ at depth z (centimeters) and time t (time step = 1 hour), M_P(z, t) is the CH₄ production rate, M_O(z, t) is the CH₄ oxidation rate, R_P(z, t) is the plant-aided CH₄ emissions rate, and R_E(z, t) is the ebullitive CH₄ emissions rate. The term, , the flux divergence, represents the net change in methane concentration resulting from the diffusion of methane into soil layer z from the surrounding soil layers or the atmosphere (if z = 0) and the diffusion of methane out of soil layer z into the other soil layers or the atmosphere (if z = 0). The rates of diffusion and the emissions calculated for each soil layer within the soil profile are then used to determine the CH₄ flux at the soil or water surface. The CH₄ flux between the atmosphere and the soil (F_CH4(t)) is the total of the fluxes at the soil/water-atmosphere boundary via the different transport pathways,

where F_D(z = 0, t) is the diffusive flux of CH₄ between the atmosphere and the soil surface, F_P(t) is the sum of all the plant-aided CH₄ emissions, and F_E(t) is the sum of all the ebullitive CH₄ emissions. By numerically solving equation (1) for all the soil layers simultaneously, we obtain F_D(z = 0, t) which will be positive if methane diffuses from the soil out to the atmosphere and will be negative if methane diffuses from the atmosphere into the soil. We determine F_P(t) by integrating R_P(z, t) for all soil layers between the soil surface and the rooting depth. Similarly, F_E(t) is obtained by integrating R_E(z, t) over all soil layers in the saturated zone if the water table is at or above the soil surface. Otherwise, the F_E(t) term will equal 0.0. Emissions of CH₄ from soils occur when F_CH4(t) is positive and CH₄ consumption by soils occurs when F_CH4(t) is negative.

[11] As both biological activity and soil transport properties influence our estimates of CH₄ fluxes at the soil/water surface, we describe below how we obtain the terms in equation (1) in more detail.

2.1.1.1. Methane Production

[12] Methane production is modeled as an anaerobic process that occurs in the saturated zone of the soil profile. We estimate hourly methanogenesis (M_p(z, t)) within each 1-cm layer of the soil profile as follows:

where M_G0 is the ecosystem-specific maximum potential production rate (Table 1); f(S_OM(z, t)) is a multiplier that enhances methanogenesis with increasing methanogenic substrate availability, which is a function of net primary production of the overlying vegetation; f(M_ST(z, t)) is a multiplier that enhances methanogenesis with increasing soil temperatures using a Q₁₀ function [Walter and Heimann, 2000] with Q₁₀ coefficients (P_Q10) and reference temperatures (T_PR) that vary across ecosystems (Table 1); f(pH(t)) is a multiplier that diminishes methanogenesis if the soil-water pH is not optimal (i.e., pH = 7.5) as described by Cao et al. [1996]; and f(R_X(z, t)) is a multiplier that describes the effects of the availability of electron acceptors which is related to redox potential on methanogenesis. To simulate f(R_X(z, t)), we use the relationships of Zhang et al. [2002] and Fiedler and Sommer [2000] where f(R_X(z, t)) diminishes methanogenesis linearly if redox potential is greater than −200 mV; otherwise, f(R_X(z, t)) is equal to 1.0. With the exception of f(S_OM(z, t)) which is described in section 2.1.4, the components of equation (3) are described in more detail in Appendix A.

2.1.1.2. Methane Oxidation

[13] Methane oxidation is modeled as an aerobic process that occurs in the unsaturated zone of the soil profile. We estimate hourly methanotrophy (M_O(z, t)) within each 1-cm layer of the soil profile as follows:

where O_MAX is the ecosystem-specific maximum oxidation coefficient (Table 1) that typically ranges between 0.3 and 360 μmol L⁻¹ h⁻¹ [Segers, 1998]; f(C_M(z, t)) is a multiplier that enhances methanotrophy with increasing soil methane concentrations using a Michaelis-Menten function with a half-saturation constant (K_CH4) that varies across ecosystems (Table 1); f(T_SOIL(z, t)) is a multiplier that enhances methanotrophy with increasing soil temperatures using a Q₁₀ function [Walter and Heimann, 2000] with Q₁₀ coefficients (O_Q10) and reference temperatures (T_OR) that vary across ecosystems (Table 1); f(E_SM(z, t)) is a multiplier that diminishes methanotrophy if the soil moisture is not at an optimum level (M_vopt); and f(R_OX(z, t)) is a multiplier that enhances methanotrophy as redox potentials increase linearly from −200 mV to 200 mV [Zhang et al., 2002]. As redox potentials become greater than 200 mV, f(R_OX(z, t)) is set equal to 1.0. Methanotrophy is also assumed to cease if soil moistures reach a critical minimum (M_vmin) or maximum (M_vmax) soil moisture. These critical soil moistures along with the optimum soil moisture (M_vopt) are assumed to vary among ecosystems (Table 1). The components of equation (4) are described in more detail in Appendix B.

[14] The inputs to equations (3) and (4) are either prescribed for a site or provided by the other modules of the coupled model. Net primary production (NPP) is estimated by the CNDM. Soil temperatures are provided by the STM. Soil-water pH is prescribed for a site. Redox potential is calculated based on root distribution, the fraction of water-filled pore space, and the position of the water table [Zhang et al., 2002; Segers and Kengen, 1998]. Root distribution is prescribed for a site based on ecosystem type. The fraction of water-filled pore space, the position of the water table, and soil moisture are provided by the HM.

2.1.1.3. Methane Transport

[15] In the model, we consider three pathways by which CH₄ can be transported from the site of production in wetlands to the atmosphere: (1) diffusion through the soil profile (F_D(z, t)); (2) plant-aided transport (R_P(z, t)); and (3) ebullition (R_E(z, t)). In upland soils, we assume that diffusion of atmospheric methane into soils is the sole method of moving methane through the soil. We assume that soil diffusion follows Fick's law with a diffusion coefficient that varies with soil texture [Walter et al., 2001a] and moisture status (i.e., saturated or unsaturated) of the soil layers [Walter and Heimann, 2000]. Plant-aided transport depends on vegetation type, plant density, the distribution of roots in the soil, and soil CH₄ concentrations [Walter and Heimann, 2000]. Ebullition occurs in saturated soil layers where the CH₄ concentration is greater than 500 μmol L⁻¹ to allow bubbles to be formed [Walter and Heimann, 2000].

[16] The amount and timing of CH₄ emissions depend on the pathway used to transport methane to the atmosphere. Diffusion is relatively slow such that CH₄ produced in the lower saturated zone may be oxidized in the unsaturated zone before it can reach the atmosphere. In contrast, methane emissions from plant-aided transport or ebullitions, if the water table is at or above the soil surface, may reach the atmosphere from anywhere in the soil profile in a single hourly time step. However, if the water table is below the soil surface, bubbles formed in the saturated zone will contribute methane to the soil layer just above the water table. This methane will then continue to diffuse upward in the unsaturated zone where it may also be oxidized before reaching the atmosphere. Similar to Walter and Heimann [2000], we also assume that a portion of the methane transported by plants will be oxidized before the gas reaches the atmosphere. However, we assume that only 40 percent of the methane is oxidized as compared to the 50 percent assumed by Walter and Heimann [2000]. We describe how we modeled each of these transport pathways in more detail in Appendix C.

2.1.4. Carbon/Nitrogen Dynamics Module

[21] We assume that the production of root exudates during the growing season enhances methanogenesis by increasing the availability of organic carbon substrate. To capture the effect of the spatial and temporal variations in root exudates on methanogenesis, we use monthly net primary productivity (NPP) estimates from the carbon/nitrogen dynamics module (CNDM) of the Terrestrial Ecosystem Model (TEM) [Zhuang et al., 2003]. The NPP estimates are used as an indicator for the seasonal and interannual variations in methanogenic substrate as follows:

where NPP(mon) is monthly net primary productivity (g C m⁻² month⁻¹); NPP_MAX represents the maximum monthly NPP expected for a particular vegetation type (Table 1); f(C_DIS(z)) is a multiplier that describes the relative availability of organic carbon substrate at depth z (centimeters) in the soil profile; and t represents time (hour). While organic substrates associated with fine root mortality are assumed to be available throughout the year, the ratio of NPP(mon) to NPP_MAX is used to represent the additional availability of root exudates during the growing season (i.e., NPP greater than 0.0). Hence the first term on the right-hand side of equation (5) is assumed to equal 1.0 during the dormant season. We assume the simulated monthly NPP remains constant throughout the month. As a result of root mortality, we assume that f(C_DIS(z)) is equal to 1.0 throughout the rooting zone (i.e., z is above the rooting depth). If z is below the rooting depth, the effect of f(C_DIS(z)) is assumed to decrease exponentially with depth [Walter and Heimann, 2000] as follows:

where R_D is the rooting depth (centimeters) as determined by the soil texture and the vegetation type [Vörösmarty et al., 1989] found at the site.

2.2. Methane Dynamics Module Parameterization

[22] We parameterize the methane dynamics module (MDM) using measurements of CH₄ fluxes and key soil and climate factors made at six field sites in North America between 53°N and 68.5°N (Table 2). For wetland ecosystems, we parameterize the MDM by minimizing the differences between observed fluxes and simulated fluxes at the Toolik-D, Toolik-W, and SSA-FEN field sites. For each site, we start the parameterization procedure with an initial set of parameter values determined by a review of the literature. Each individual parameter has been adjusted to be within a range of values provided from the literature review until the root-mean-square error (RMSE) between the daily simulated and observed CH₄ fluxes is minimized. This procedure is conducted sequentially for all parameters with the result that RMSE for the Toolik-D, Toolik-W, and SSA-FEN parameterizations are 20, 52, and 42 mg CH₄ m⁻² d⁻¹, respectively.

[23] Unlike the wetland sites, we do not have a daily time series of CH₄ flux data for the other three upland sites (B-F, Tundra-NS, and Tundra-UI). Therefore we parameterize the MDM for upland ecosystems such that the difference between the simulated and observed maximum daily CH₄ consumption rate is minimized at these sites. Specifically, we alter the parameters of the methane module until the simulated CH₄ consumption by soils reaches the maximum consumption rate of 0.95, 1.2, and 2.7 mg CH₄ m⁻² d⁻¹ at the B-F, Tundra-NS, and Tundra-UI sites, respectively. Because the meteorological observations of some sites are not available to us, we use climatic data from other sources (see Table 2), and it is possible that this may lead to biases in the parameterization. In addition, our approach of adjusting a single parameter at a time may lead to biases in parameterizations. The ecosystem-specific parameters for the MDM based on these site calibrations are documented in Table 1.

2.3. Model Testing at the Site Level

[24] To test the model and validate our parameterizations, we conduct simulations for a boreal forested wetland site (NSA-FEN) in Canada and a tundra site (Tundra-F) at Fairbanks, Alaska (Table 2), which are not used during our parameterization process. To evaluate model performance, we compare the simulated daily CH₄ fluxes to observed fluxes at these sites. The SSA-FEN parameterization is used for the NSA-FEN site simulations, and the Toolik-W parameterization is used for the Tundra-F site simulations.

2.4. Regional Simulations Using Geographically Explicit Data

[25] To make spatially and temporally explicit estimates of CH₄ emissions and consumption in the northern high latitudes (north of 45°N) with our new version of TEM, we use spatially explicit data of climate, land cover, soils, daily climate, and monthly leaf area index (LAI) from a variety of sources. The model is applied at the spatial resolution of 0.5° latitude × 0.5° longitude and at a daily time step for the period 1900 through 2000.

[26] The static data sets include potential vegetation [Melillo et al., 1993], soil texture [Zhuang et al., 2003], the distribution of wet soils and the fractional inundation of wetlands [Matthews and Fung, 1987], and soil-water pH [Carter and Scholes, 2000]. Similar to earlier versions of TEM, the vegetation and soil texture data sets are used to assign vegetation-specific and texture-specific parameters to a grid cell. The remaining spatially explicit data sets are needed to provide inputs into the new MDM. The wet soils and the fractional inundation of wetlands data sets are used to derive the proportions of wetlands and uplands within each 0.5° × 0.5° grid cell. The soil-water pH data set is used to estimate methanogenesis across the study region.

[27] The daily climate data sets are derived from the historical monthly air temperature, precipitation, vapor pressure, and cloudiness data sets (T. D. Mitchell et al., A comprehensive set of high-resolution grids of monthly climate for Europe and the globe: The observed record (1901–2000) and 16 scenarios (2001–2100), submitted to Journal of Climatology, 2004) (hereinafter referred to as Mitchell et al., submitted manuscript, 2004) of the Climatic Research Unit (CRU) of the University of East Anglia in the United Kingdom. We linearly interpolate the monthly air temperature and vapor pressure to daily data using three consecutive month's data. To determine a current month's daily air temperatures, for example, we assume that (1) the value of day 15 is equal to the current month's mean air temperature, (2) the value of the first day is equal to the average monthly air temperature of the current month and the previous month, and (3) the value of last day is equal to the average monthly air temperature of the current and the next month. The temperatures for the other days are linearly interpolated using values of the first, fifteenth, and last days. To convert monthly precipitation into daily rainfall, we use the statistical algorithm of Li and Frolking [1992] and Liu [1996]. The algorithm converts the monthly precipitation into a number of rainfall events of different duration and intensity based on air temperature and the correlation of monthly precipitation with the frequency of heavy, intermediate, and small rainfall events.

[28] In the HM, monthly LAI is used to estimate transpiration (Appendix D) [Zhuang et al., 2002]. We use monthly LAI data sets derived from satellite imagery for the period 1982 to 1999 [Myneni et al., 1997, 2001] to prescribe LAI for each 0.5° latitude × 0.5° longitude grid cell. From 1900 to 1981, we use the LAI of 1982 to represent LAI during this period. We also use the LAI of 1999 to represent LAI during 2000. During our simulations, LAI is assumed to remain constant within a month.

[29] To develop regional estimates of CH₄ exchange from 1900 to 2000, we simulate the methane dynamics and estimate daily CH₄ fluxes from both wetland and upland ecosystems in each 0.5° latitude × 0.5° longitude grid cell. These ecosystem-specific CH₄ flux estimates are then area-weighted for each grid cell, as defined by the fractional inundation data set of Matthews and Fung [1987], to determine the CH₄ fluxes from each 0.5° latitude × 0.5° longitude grid cell.

3.1. Site-Specific Testing

[30] At the test site Tundra-F, the simulation captures the interannual and seasonal variations of the net CH₄ emissions. The simulated annual emissions are 12.2, 10.4, 7.6, and 12.1 g CH₄ m² yr⁻¹ for 1987, 1988, 1989, and 1990, respectively, compared to observed emissions of 8.05 ± 2.5, 11.38 ± 2.88, 8.11 ± 1.80, and 13.64 ± 1.20 g CH₄ m² yr⁻¹ for the same years [see Whalen and Reeburgh, 1992]. The geometric mean regression statistics [Sokal and Rohlf, 1981] shows a significant (P < 0.01; N = 48 months) relationship between the simulated and observed monthly emissions with R² = 0.77, slope = 0.86 ± 0.06, and intercept = 0.17 ± 0.10 g CH₄ m⁻² month⁻¹ (Figure 2a). Overall, the simulations tend to have higher emissions compared to the observations during the spring of each year (Figure 2b). This discrepancy occurs because the model assumes that the winter snowpack insulates the soil from the frigid air temperatures during the winter such that soil temperatures remain relatively high. The higher soil temperatures lead to an earlier spring thaw, which in turn leads to earlier CH₄ production at the site. In 1990, the model underestimates the emissions in August and September. This is primarily because the simulated water table ranges from 27 to 28 cm, which is deeper than the measured maximum depth of 23 cm. The deeper water table leads to less CH₄ production and emissions.

[31] Similarly, at the NSA-FEN test site, the model is able to capture the interannual and seasonal dynamics of net CH₄ emissions in 1994 and 1996. A geometric mean regression between the monthly simulated and observed net emissions is significant (P < 0.01; N = 10 months) with R² = 0.90, slope = 0.73 ± 0.07, and intercept = 0.35 ± 0.23 g CH₄ m⁻² month⁻¹ (Figure 2a). The model slightly underestimates the emissions from June to September in 1996 (Figure 2c). Our analyses suggest that the lower emissions in our simulation are primarily due to the lower soil temperatures resulting from the low soil thermal conductivity prescribed for the model at this site. The deviation is also partially due to the climate data used to drive the model. Owing to the lack of in situ meteorological data at the site, data from the Thompson station of the Canadian Atmospheric Environment Service (AES) has been used to drive the model for this analysis.

3.2. Contemporary Regional and Subregional Fluxes

[32] Overall, our simulations estimate that the Pan-Arctic region has been a mean source of about 51 T g CH₄ yr⁻¹ during the 1990s. This estimate is in the same range as a number of other recent estimates that have been made using a variety of approaches (Table 3). Differences between our estimates and those of other studies may be a result of using different geographical boundaries or assuming different importance of various ecosystems in contributing methane to the atmosphere. For example, Walter et al. [2001b] considered areas north of 30°N in developing their regional estimates rather than the 45°N boundary used in this study. Several studies considered only tundra, boreal forests, or wetlands when developing their regional estimates. In our study, we estimate that the source strength varies over the Pan-Arctic and that large regions have actually been small net sinks of atmospheric CH₄ (Figure 3).

[33] In our simulations, wetlands act as a net source of CH₄, whereas upland areas act as a net sink. We estimate that wetlands across the Pan-Arctic emitted about 57 T g CH₄ yr⁻¹ during the 1990s. Wetlands within boreal forests have the highest rates of emissions (23 g CH₄ m⁻² yr⁻¹), but the large areas of wetlands within wet tundra cause these ecosystems to be the largest contributor of atmospheric CH₄.

[34] In addition to the estimates of net CH₄ emissions from wetlands, our simulations estimate that soil microbes in upland areas have consumed about 6 Tg CH₄ yr⁻¹ across the Pan-Arctic during the 1990s. This estimate is higher in comparison to most other studies of methane consumption (Table 3), which estimate the consumption rate to be between 0 and 5.5 Tg CH₄ yr⁻¹. An exception is the Born et al. [1990] study, which suggested a consumption rate of up to 15 Tg CH₄ yr⁻¹. In developing our estimates of CH₄ emissions, we do not consider the potential effects of moisture hindrance on methane diffusion through unsaturated soil. As a result, our model may overestimate actual consumption rates. Upland areas within wet tundra have the highest consumption rates (0.27 g CH₄ m⁻² yr⁻¹) because the simulated soil moisture in wet tundra is closer to the optimum soil moisture for methanotrophy than that simulated for upland boreal forests.

[35] The simulated CH₄ emissions and consumption vary across the region as a result of the distribution of wetlands as well as the spatial variability in climate (Figure 3). For the 1990s, our simulations estimate that terrestrial ecosystems within Russia, Canada, and Alaska are the major sources of CH₄ emissions in the Pan-Arctic, which are contributing 64%, 11%, and 7%, respectively, of the total net CH₄ emissions per year (51.0 Tg CH₄ yr⁻¹, Table 4). Within Russia, we estimate that the net CH₄ emissions from the West Siberia wetlands are 21g CH₄ m⁻² yr⁻¹, for a total of 12 Tg CH₄ yr⁻¹, which is close to the high end of the mean estimates of 0.3–14 Tg CH₄ yr⁻¹ by Smith et al. [2004], but lower than the mean estimate of 26 g CH₄ m⁻² yr⁻¹ by Friborg et al. [2003] for this region. Consumption of methane is more evenly distributed across the Pan-Arctic. The soils of Russia, Canada, and Alaska account for 38%, 25%, and 5%, respectively, of the total CH₄ consumed per year (6.3 Tg CH₄ yr⁻¹, Table 4) in this region.

[36] Our simulations indicate that 60% of the net CH₄ emissions come from the latitude band of 45°N–60°N as compared to 40% of total emissions from the region of 60°N–75°N, Table 5). This pattern is probably due to the larger areas of wetlands in the southern Pan-Arctic compared to the middle Pan-Arctic. However, wetlands represent a larger proportion of the land area in the middle Pan-Arctic than the southern Pan-Arctic such that the mean net emissions per square meter are actually higher in the middle Pan-Arctic. The consumption in the southern Pan-Arctic is also 2 times larger than in the middle Pan-Arctic, which is primarily due to the larger forest area in the southern Pan-Arctic.

3.3. Twentieth Century Trends

[37] During the past century, our simulations estimate that net CH₄ emissions have increased at a rate of 0.08 Tg CH₄ yr⁻¹ estimated as the slope of the linear regression between the annual net emissions and year from 1900 to 2000. For the 1980s, however, the model simulates a larger increasing trend in CH₄ emissions (∼1.0 Tg CH₄ yr⁻¹, estimated as the slope of the linear regression between the annual net emissions and year from 1980 to 1989). The increased trend of net emissions during this period is consistent with the increased trend (11.6 ± 0.2 ppbv yr⁻¹) of the observed atmospheric CH₄ concentrations during 1983–1991 in the Northern Hemisphere [Dlugokencky et al., 1994].

[38] While methane consumption rates remain fairly constant throughout the study period, net CH₄ emissions vary from decade to decade (Table 6) with relatively large emissions in the 1920s–1930s, 1950s, and 1980s–1990s. The decadal net CH₄ emission rates are correlated with decadal variations in climate and its derived variables, namely, soil temperature, water table depth, and NPP. Our analyses indicate that net CH₄ emissions are more significantly correlated with air temperature (R² = 0.81; P < 0.01, N = 10 decades) than precipitation (R² = 0.40; P < 0.01; N = 10 decades). The correlations between decadal net CH₄ emissions and water table depth, soil temperature, and NPP are significantly (P < 0.01) high, with R² values of 0.65, 0.82, and 0.65, respectively. These analyses suggest that changes in climate and its influence on ecosystem production and the soil environment could significantly influence the dynamics of CH₄ emissions.

[39] Decadal changes of simulated monthly emissions from the 1900s to 1990s also show an increasing trend in the magnitude of net CH₄ emissions during the growing season (May through September; see Figure 4) when root exudates provide additional carbon for methanogenesis. As shown in Table 6, NPP has increased over the period. The enhanced NPP increased the input of root exudates to enhance methanogenesis and CH₄ emissions over this time period. Our simulations show that the peak emissions occurred in July, which is consistent with the results of recent inverse modeling studies [Houweling et al., 2000; Chen, 2004] and other process-based modeling [Cao et al., 1996]. This peak in monthly CH₄ emissions corresponds to the seasonal peak in NPP.

[40] Our simulations also show that large interannual variability in net CH₄ emissions occurred during the twentieth century (Figures 5a and 5b). For example, our simulations estimate that the net CH₄ emissions decrease from 50 Tg CH₄ yr⁻¹ in 1991 to 40 and 45 Tg CH₄ yr⁻¹ in 1992 and 1993, respectively, after the Mount Pinatubo eruption in 1991 (Figure 5c). This pattern of CH₄ emissions has also been observed in the inverse modeling study of Dlugokencky et al. [1994], and the modeling study of Walter et al. [2001b]. During 1998, there was a large positive anomaly in the global growth rate of atmospheric methane concentrations; Dlugokencky et al. [2001] attributed this anomaly in part to increased emissions from wetlands in northern high latitudes resulting from warm conditions in 1998 due to the strong El Niño phenomena. Our simulation results support this interpretation and indicate that the region released 55 Tg CH₄ in 1998, an amount that is 8–11 Tg higher than the net CH₄ emissions in 1999 and 1997. However, we acknowledge that the atmospheric CH₄ concentrations are influenced by a variety of factors including atmospheric transport and atmospheric CH₄ oxidation as well as CH₄ emissions from other sources such as fires, landfills, industrial processing, fossil-fuel burning, and rice paddies. Thus net CH₄ emissions from natural wetlands can only partially explain the changes in atmospheric CH₄ concentrations and may at times show opposite trends. For example, our simulated net CH₄ emissions for the year 2000 are the second highest emissions estimated during the last decade or so in response to interannual climate variability, but the atmospheric CH₄ concentrations did not show a corresponding peak due to a number of different sink and source dynamics. A decrease in fire disturbances north of 38°N [van der Werf et al., 2004] and their associated emissions during 2000 may have compensated for the increases in net CH₄ emissions from wetlands to influence atmospheric CH₄ concentrations.

3.4. Sensitivity of Net CH₄ Emissions to Active Layer Depth

[41] Previous studies [Zhuang et al., 2001; Romanovsky and Osterkamp, 1997] have indicated that the active layer depth may be significantly overestimated if soil thermal models do not consider the possibility of soil freezing upward from the permafrost boundary. Because the estimated active layer depth is used to determine the lower boundary of microbial activity including methanogenesis in soils of permafrost regions, the overprediction of the active layer depth will result in higher estimates of methane production and emissions. To test this hypothesis, we increase the depth of the lower boundary by 10 cm in a sensitivity analysis. The resulting regional estimate of net CH₄ emissions for the Pan-Arctic region is 38% larger than that of our control simulations. This suggests that the consideration of two freezing fronts, i.e., freezing of soil upward from the permafrost boundary as well as downward from the soil surface, is important when modeling methane emissions from regions underlain with permafrost.

3.5. Conclusions

[42] In this study, we couple key aspects of soil thermal, hydrological, and carbon dynamics of terrestrial ecosystems with methane cycling to estimate CH₄ fluxes between the atmosphere and the soils of the Pan-Arctic region. By considering the ability of soils to produce methane in wetland soils and to oxidize methane in both wetland and upland soils, we have developed more comprehensive regional estimates of CH₄ fluxes than provided by earlier studies using either process-based models or field estimates. Our analyses suggest that CH₄ emissions are more sensitive to changes in climate, particularly air temperature, than consumption, such that natural ecosystems may become a larger source of atmospheric CH₄ with future global warming. In addition, our analyses suggest that changes in root exudates associated with climate-induced enhancements in plant productivity may also increase CH₄ emissions. However, reductions in the area of wetlands in the Pan-Arctic region [e.g., McGuire et al., 2004] as a result of alterations of the hydrological cycle may decrease methane production and allow methane consumption by soils to become more important. Because the areal extent of wetlands has been kept constant in this study, we have not been able to evaluate the importance of this negative feedback on regional estimates of net CH₄ emissions.

[43] Our regional estimates of net CH₄ emissions from natural ecosystems are 10–20% higher than those estimated from an inverse modeling study based on spatial and temporal changes in atmospheric CH₄ concentrations [Chen, 2004]. To help resolve this discrepancy and to better understand the role of natural ecosystems in the global methane budget, it is desirable to couple our spatially explicit estimates of CH₄ fluxes to an atmospheric transport model to simulate seasonal and interannual changes in atmospheric CH₄ concentrations. This approach has already been taken with CO₂ fluxes and has proved helpful in evaluating and improving the simulation of the various aspects of the carbon cycle including terrestrial carbon sequestration [McGuire et al., 2000; Dargaville et al., 2002; Zhuang et al., 2003].