Interannual variations in tundra methane emission: A 4-year time series at fixed sites

Abstract. This paper summarizes 4 years (1987-1990) of weekly net CH4 flux measurements at permanent sites representing important plant components of Arctic tundra. The data coincide with variations in precipitation and temperature of interest in regional and global modeling efforts and are useful in placing bounds on the role of tundra in the global CH4 budget. Precipitation in the study area during the summer emission period ranged from twice to half the long-term mean, and air temperature anomalies were about +2 °C. This data set also permits consideration of temporal (seasonal to interannual) and spatial variability in CH4 flux. We studied the relationship between the net CH4 flux and subsurface properties (water table depth, thaw depth, soil temperature, /pCH4 distributions) at these permanent sites during the 1988 and 1989 emission periods. Net CH4 emission and subsurface properties are largely unrelated. Relationships between soil temperature (or any single variable) and emission are site specific and are of little value as flux predictors. Parameters that integrate conditions influencing flux appear to be the best flux predictors over the emission period. We estimate that Arctic wet meadow and tussock:shrub tundra presently emit about 42 ± 26 Tg CH4 yr−1 to the atmosphere. This estimate has a North American bias, but it is supported by measurements in a range of locations, transect studies, and model calculations.


INTRODUCTION
Tundra environments occupy some 7% of the Earth's land area and contain about 13% of its stored soil carbon [Post et al., 1982]. The fate of this stored soil carbon under altered climate is a major question [Billings, 1987]. The role of high-latitude wetlands in the global CH4 budget was poorly known until recently. Prior to 1986, only three publications [Svensson, 1976;Svensson and Rosswall, 1984;Sebacher et al., 1986] reported measurements of CH4 fluxes from tundra environments. Reports of increases in atmospheric CI-I4 concentrations except C-1 and C-11, where thermistors extended to 40 and 70 cm, respectively. Mean soil temperatures in the thawed layer at T, M, and BH primary stations were calculated by integrating temperature data from the soil surface to permafrost. Mean soil temperatures to the deepest thermistor were calculated at C primary stations. Thaw depths were measured by inserting a 3-mmdiameter stainless steel rod to the frozen soil horizon.
The thaw season reported at each station (Tables 1-4)   is the period from initial rod penetration in the spring to surface soil freeze-up in the fall. The "warm rim effect" [Likens and Johnson, 1969] limited permafrost formation in the lake margin, so no thaw depth determinations were made at C stations. Soil CI-In profiles were sampled with equilibration samplers [Hesslein, 1976] within 0.5 rn of primary stations during the 1988 and 1989 thaw seasons.
The 1989 soil CI-h measurements were made at station C-7. The samplers were made of lucite and had equilibration wells (3 cm3) spaced at 2.5-cm intervals along the sample probe. The equilibrafion   IV  I  II  III   1988   IV  I  II  III  IV  I  II  III  1989  Methane fluxes (mg m-2 d-I) for each station (Figures 1 -4) were averaged, and these were integrated (trapezoidal role) over time to give annual emission estimates for each site (Table 5). Methane emission from the T site showed little interannual variability, ranging over a factor of only 1.7. In contrast, annual emissions at the BH and M sites ranged over factors of 7 and 10, respectively, and the data for the C site varied by a factor of 74.
Annual flux estimates for each tundra component in Table 5 were weighted as in the work by Whalen and Reeburgh [1988] to estimate global tundra CH4 emission for 1987 through 1990 (Table 6). Overlap in percent ground cover occupied by each site type causes total areal coverage to exceed 100%. Nonetheless, three points are clear in Table 6  Second, the C site (Carex) accounts for most of the CI-I4 emission from wet meadow tundra; it is the dominant plant form and annual CHn emission was roughly equal to or exceeded that of the M site in all study years except 1988 (Table 5)

. Finally, the annual source strength of tundra in the global CI-h budget is highly variable. The mean global emission calculated from the 1987-1989 estimates is 42 + 26
Tg CI-h yr-1. Including the 1990 global emission estimate, which contains the unusually high Carex emission rate (Tables 5 and 6), gives a global emission estimate of of 57 + 39 Tg yr-1.

Environmental Variables
Air temperature and precipitation data from the study area are presented in Figures 5 and 6. Air temperatures during the study were generally higher than the long-term mean. Only 9 of 48 months showed negative temperature anomalies. Air temperature anomalies were about +2 øC during the June through September emission period. Ctmaulative annual precipitation varied from 16 cm (1987) to 47 cm (1990), or from 60 to 170% of the long-term mean of 28 cm. Summer rainfall is critical in regulating water table position in the muskeg. Soils are water-saturated at freeze-up, and the snow pack is largely exported by runoff before surface soils thaw, so that little standing water remains. Summer rainfall during 1987 through 1989 was low; precipitation during this period was only 56 to 67% of the the long-term mean value of 16 cm. In contrast, 1990 summer precipitation totaled 30 cm, or about 190% of the long-term mean.
Tables 1-4 contain summary data for environmental variables at permanent sampling stations from the 1987-1990 thaw seasons, respectively. Annual ranks of the duration of thaw, maximum thaw depth, and mean soil temperature for sites followed a predictable pattern based on local topography. Tussocks are taller than the surrounding vegetation, so they thaw earlier.  I  II  III  IV  I  II  III  IV  I  II  III  IV  I  II  III  IV   1987  1988    Water table levels at C stations are reported for 1990 only. Stations were initially established on the Smith Lake margin midway between the Fall 1986, low water mark and the historic high water mark. The water table level was below the 70-cm-deep gauging wells late in the low summer precipitation years of 1988 and 1989. The new C stations were established at roughly the same elevation for 1990 sampling; the water table was above the soil surface throughout this summer of high rainfall (Table 4).

. Precipitation (in centimeters, water-equivalent) in the University of Alaska Arboretum. Long-term mean and conditions during study period. (Top) Cumulative precipitation. (Bottom) Precipitation anomaly (difference in long-term mean and monthly mean values) during study period. Shading denotes the major emission period.
Data from NOAA [ 1986,1987,1988,1989,1990] records for Fairbanks International Airport, 6 km south of study sites. bMean to depth of permafrost.
CAbsolute value of product of thaw depth and mean soil temperature to permafrost.

. (Top) Net CI-I4 fluxes, (middle) water table and thaw depths, and (bottom) subsurface pCH4 distributions for (A) T-l, (B) BH-1, and (C) M-1 stations, June -September, 1988. Carex (C-l) station was not sampled during 1988. Each of the pCH4 versus depth panels integrates conditions up to the sampling date.
Water table and thaw depth were measured when the equilibration samplers were retrieved.
increase throughout the season at T-1, while emission remained generally constant (Figure 8). Methane emission and soil CH4 distributions appear to be uncoupled at BH-1 ( Figure 8) and C-7 ( Figure   8).

Global Tundra Emission Estimates
One of our goals was to determine net CH4 emission and its temporal and spatial variability in representative tundra floral associations.We view the annual emission from each site (Table 5) as a "reagent grade" budget constituent of known uncertainty that can be appropriately combined and weighted to produce regional and global CI-h emission estimates. We have extended site emissions from this study (Table 5) using our previous weighting scheme [Whalen and Reeburgh, 1988] to annual global CH4 emission estimates for 1987-1990 in Table 6. These emission estimates incorporate uncertainties in areal coverage as well as the variations in annual site fluxes (Table 5).
The representativeness of sites in a heterogeneous landscape is crucial in regional extrapolation of fluxes to global budgets [Matson et al., 1989]. We are confident that our weighting scheme, which considers the floral or site composition in two major tundra subdivisions (tussock:shrub and wet meadow tundra), provides a masonable framework for several reasons. First, CI-I4 fluxes from these major tundra subdivisions are dominated by two site types: Eriophorum ressocks (T) and Carex (C). These sites account for 24-45% and 80-90% of the areal coverage and contribute 50-94% and 89-100% of the annual CH4 flux, respectively, from the two tundra subdivisions (Table 6) Table 6) The data suggest a range of variations in annual CI-I4 emission of about twofold for years with summer precipitation similar to the long-term mean and about fourfold for years with unusually high summer precipitation. It is unlikely that differences between the 1990 annual flux estimates and those from previous years resulted from station relocafion. Parallel sampling of new and old stations on a single occasion during 1990 showed no statistically significant differences in fluxes for new and old stations of any site type (Mann-Whitney Test; p>0.05; data not shown). New C stations were established at roughly the same elevation as previous stations. We conclude that the increased emission at C stations in 1990 (Tables 5   and 6) resulted from a -40 cm rise in the summer water table due to record rainfall. Increased CI-h emission from this site type was largely responsible for the high 1990 emission estimate, as noted above.
We also present the first data for interannual variations in CI-h emission from representative tundra CH4 budget constituent sites. Although summer air temperature and precipitation were roughly comparable in 1987-1989, there were obvious differences in the site fluxes (Table 5) and the relative contribution of C, M, and BH site types to the annual CH4 emission estimate (Table 6). We have no ready explanation for these differences but suspect they are due to interannual variations in the submeter scale interactions of biological, physical, and chemical parameters and processes. Improved understanding of these interactions is necessary to further resolve the twofold interannual differences in overall annual CH 4 emission.

Correlations With Environmental Variables
Other investigators have reported strong relationships between CI-h flux and temperature. We had limited success in correlating mean soil temperature with CI-h emission at primary sites (Table 7). Correlation coefficients were not improved by selecting temperature at a single depth or by use of a running mean temperature as the independent variable (data not shown  Bartlett et al., 1992], and exponential ] relationships between CI-h flux and air or soil temperature have been reported. The variety of functions used suggests that the relationship between 155 CI-h flux and temperature is not unique or straightforward. The lack of statistical correlation between water table depth and seasonal CI-h emission at our muskeg sites (Table 7) appears to result from analysis of data for the entire thaw season. A low late summer water table corresponds with reduced CI-h flux (Figures 7 and 8). However, early season (May) emissions from waterlogged soil were also low as a result of a thin thawed soil zone. Combining all data for a season results in a low rs. Highest CH4 emissions are generally observed during midsummer, when the thawed zone is rapidly increasing, but the water table is falling. Seasonal studies in temperate swamp [Wilson et at., 1989] and wetland [Yavitt et al., 1990] soils also report poor correlation between water table level and CI-h flux. In contrast, transect studies showed a closer relationship between CI-h flux and water table position (r = 0.42 to 0.54) [Sebacher et al., 1986;Hatriss et al., 1988;Whalen and Reeburgh, 1990a] or soil moisture content (r = 0.77 to 0.93) [Svensson, 1976[Svensson, , 1980. It is likely that the CI-h emission response to changing environmental conditions occurs rapidly and is masked in whole-season correlations.

The parameters thaw depth and centimeter-degrees
show the best correlation with CI-h emission (Table   7) because these variables integrate conditions important in determining net CI-h emission. Specifically, centimeter-degrees is a reasonable index of net microbial activity because it accounts for the mean soil temperature and the depth of the active zone.

Controls on CH4 Emission
The measurements of net CH4 emission reported here are the difference between CH4 production and microbially mediated CH4 oxidation. This oxidation occurs before the CH4 is emitted to the atmosphere, and is an important control on tundra CH4 emission. Reeburgh

Soil pCH 4 Distributions and CH 4 Flux
Soil pCH4 distributions suggest a surface zone of rapid CI-I4 oxidation and a subsurface zone of CH4 production (Figures 7 and 8 Sequential soil CH4 distributions are an integrating indicator of microbial activity. We expected CH4 emission to track changes in the soil CH4 pool, which can be seen in temporal sequences from single stations (Figures 7 and 8). Instead, we found that CI4_n flux ,•nd pool size were generally unrelated over time. Similar results have been reported for a boreal fen  and Subarctic tundra [Bartlett et al., 1992]. This decoupling could be the result of comparing measurements that respond to different time scales. The static chamber flux determinations are essentially instantaneous rate measurements (snapshots), while the soil CH4 determinations result from integration o•er the equilibration period (time exposure). However, the boreal fen CI-h distributions cited above were instantaneous (pore waters syringe-sampled) and CI4_n pool sizes determined in a salt marsh with equilibrafion samplers were well correlated with CH4 flux [Bartlett et al., 1987]. It is likely that this decoupling is an additional indication of the complex biotic and abiotic factors determining net CH4 emission.

Whalen and Reeburgh: Tundra Methane Emission Predictors and Feedbacks
Several lines of evidence suggest that the tundra system is too complex for a a single environmental variable to serve as a universal and successful indicator of CH4 flux. First, a variety of functions have been used to relate CH4 flux and temperature (see above). Second, significant relationships often involve special conditions. For example, Sebacher et al. [ 1986] reported good correlation between water table depth to 10 cm and CH4 emission in a transect study, and Wilson et al.
[1989] required a step function to relate soil temperature to emission. Third, relationships reported in Table 7 were best when data were simply ranked using a nonparametric statistic. Finally, lx)oling all data for one variable by station or year or by station and year resulted in few significant correlations, despite the fact that the increase in sample size allows a smaller value of rs for statistical significance. Moore et al. [ 1990] noted that correlation between CH4 flux and a single environmental variable in high-latitude fens may, in part, be specious because variables influencing flux are not entirely independent. This view is supported by our observation that CH4 emission was best correlated with variables that serve as environmental integrators (Table 7). Attempts to identify predictors of CI-h emission should focus on parameters that integrate factors important in CH4 production and consumption.
Recent papers that have addressed feedbacks on trace gas emissions [Khalil and Rasmussen, 1989;Lashof, 1989] assume that increased temperatures will result in higher CH4 fluxes, i.e., a positive feedback. These studies focus on the influence of a single variable (temperature) on a single biological process (methanogenesis) and do not consider other effects of climate change, namely regional changes in water table level [Mitchell, 1989] and rates of CH4 oxidation. Functions that successfully relate CH4 flux to temperature are largely limited to inundated wetlands (see above) where CI-I4 production is the dominant biological process influencing emission. We expect that CH4 flux-temperature relationships will deteriorate as water tables fall below the soil surface and oxidation becomes important. The ratio of oxidized:reduced soil zones will increase, favoring both aerobic decomposition and CH4 oxidation. Laboratory studies with packed peat columns [Moore et al., 1989] showed that CH4 emission decreased logarithmically, while CO2 emission increased linearly as water table decreased. Moreover, changes in soil temperature will not be uniform with depth and the effect of increased soil temperature will be greater for CI-I4 consumption than production. Finally, the increased depth of thaw in permafrost areas will not have as great a stimulatory effect on methanogenesis as predicted [Khalil and Rasmussen, 1989;Lashof, 1989], because deeper peat is likely to be refractory ['Fan•sh and Grigal, 1988; Yavitt et al., 1988]. A temperature feedback will be important in coupled biogeochemical-atmospheric models, but water table position will be a better integrator, waterlogged soils are zones of CI-I4 production, while moist soils are areas of CH4 consumption. Carefully executed field experiments involving manipulations of water table and temperature are needed to resolve the treatment of feedbacks in coupled biological models. One simple hydrologic model for northern fens indicates that a falling water table has a greater impact than increased soil temperature on CI-I4 emission [Roulet et al., 1992].

SUMMARY
On the basis of a 4 year study at permanent sites we estimate that tundra contributes 42 + 26 Tg yr-1 to the net atmospheric CH4 budget. This large, multiyear data set, obtained under a range of moisture and temperature conditions, restricts the range of tundra CH4 source strength estimates and also shows the effects of local extrema. Results from multiple permanent sites also suggest that single-parameter relationships used to predict CI-I4 flux are sitespecific and point to the need for an integrating predictor.