Micrometeorological measurements of CH 4 and CO 2 exchange between the atmosphere and subarctic tundra

,


EXPERIMENT
Fluxes of CI-I4 and CO2 were measured directly using the eddy correlation method.A three-axis sonic anemometer, capable of measuring wind velocities and ambient temperature at 20 Hz [Fitzjarrald and Moore, this issue], was mounted at the top of the micromet tower.The sensor was rotated to keep the tower downwind to minimize interference.Air was sampled at -7 dm3min -x through an inlet located 0.5 m behind of the sonic cell cooled thermoelectrically to about 0øC, and filled with glass beads.A mercury marioseat (Gilmont Absolute Pressure Control) was used to dampen pressure fluctuations at high sample flow rate as required to achieve fast response.The gains for CO2 and CI-I4 instruments were measured by adding to the inlets small flows of concentrated standard mixtures.The 90% response time was slightly better than 1 s for THC, CI-I4 and CO2 measurements.
Measurements were collected at 8 Hz.
Methane and CO2 concentrations were measured consecutively at 8 altitudes on the tower, using duplicate insmunentation for CH 4 and a Binos nondispersive infrared analyzer for CO2.Air samples were drawn through Teflon tubes (4.5 mm ID) at 10.8, 8.5, 6.1, 4.3, and 3.1 m above ground, on booms extending 1 m from the tower, and from tubes fixed to the guy wire, 7-8 m NW of the tower, at 1.5, 0.5, and 0.02 m above the ground (see Figure 2).Each level was sampled for 4 min, requiring 32 min to obtain a profile.Temperature and dew point of the air were precondirioned as for the flux measurement.The instruments were periodically calibrated with reference gases using the same pressure, temperature, and dew point as for air samples, and at the same flow rate through the FID (100 em3min -a).

The Flame Ionization Detector
The flame ionization detector (FID for Gas Chromatograph GC-6A, Shimadzu) measures current between its electrodes car-

Computation of Eddy Correlation Fluxes
Eddy correlation fluxes were computed from covariances sorption measurement can therefore be obtained by alternately between vertical wind (w) and concentrations (c) of trace gases, tuning the laser to each of these frequencies.with the coordinate system rotated to make the fluctuating corn-The instrument, shown schematically in Figure 3, may be ponent of w perpendicular to the streamlines [McMillen, 1986].described in terms of four subsystems: the laser, the control and Eddies important for vertical transport typically had time scales signal processing electronics, the optical system (including the from a few seconds to a few minutes.A net upward or downward multipass sample cells), and the gas handling system.The refer-flux of a trace gas is represented by a statistically significant posience cell, nominally identical to the two sample cells, contained a five or negative covariance.In this study we calculated fluxes by standard gas mixture of approximately 1.6 ppm CHn in air, at the first subtracting 4-rain running means from 8-Hz measurements of Figure 11 shows means and standard deviations for net exchange of THC in each group.We excluded 14 classes with fewer than 4 hours of measurement.Influences from adjacent sectors were tested by examining data falling within the average standard deviation of the wind direction during a 1-hour interval, -15 ø.Shifting the boundaries between sectors by this amount had little effect on the analysis (<10% change in the NEE associated with a particular vegetation type).In sectors 0-120 ø and 170ø-230 ø, where lakes were located a short distance from the tower, the emission rate for CH 4 depends significanfiy on wind speed.The relation is weaker, and observed only at low wind speed, in the sector 2300-360 ø , where lakes were small and farther away.In the sector 120ø-170 ø , with mostly dry tundra and no lakes, dependence on wind speed was not apparent.Mean emissions from the predominantly dry sector (120ø-170 ø ) were smaller than from the wettest sector (170ø-230 ø ) or from the lake-dominated sector (0-120 ø) (see Figure 12b).The conditional averages obtained here may be combined with information on tower footprint, derived in the appendix (see Figure 13), to quantify the factors controlling CH4 emissions on spatial scales of -103m.This analysis is carded out in section 5. 1983] and an English bog [Clymo and Reddaway, 1971] indicated relase of 11% and 1-6% of NPP as CH4, respectively.

Methane Flux Derived From Nighttime Variations of CI'h and CO2
Concentrations of CI-I4 and CO2 usually increased after sunset when the surface layer became stratified, rising at times to more than 2500 ppb and 400 ppm, respectively.These correlated concentration changes, shown in Figure 14 2b gives the optimal values for {a•, b• }, fit to the data for 75% confidence level) than obtained in models I-1II, suggesting 30 combinations of wind direction and stability class with statisti-that emissions from wet meadow tundra may increase at higher cally significant mean fluxes (Figure 11).In order to determine wind speeds.This dependence could be an artifact reflecting the which parameters account for the sample variance, four models prevalence of wet tundra at lake margins: some pixels labelled ("hypotheses") were considered where some of the ai and b• were wet meadow likely include lake surfaces (and vice versa).set to zero.In model I, all three surface types (lake, dry upland The accuracy of the probability density functions adopted here tundra, and wet meadow tundra) are allowed finite surface emis-for the tower footprint cannot be verified, and we would therefore sion at zero wind and nonzero response to increasing wind veloci-like to determine the sensitivity of model results to details of the ty.The regression indicates insignificant emissions from lakes at functions shown in Figure 13.A simple test is to use exclusively zero wind speed (a 1), and negligible influence of wind speed on distribution curves for each stability class, ignoring information on fluxes from dry tundra (b2-0).Model 11 therefore set b2 equal to the observed stability, and to examine the coefficients for the same zero; b2 and al were set to zero in hypothesis HI.Coefficients for four models in each case as shown in Table 2b

Comparison of Enclosure, Aircraft, and Micrometeorological
Methane Flux Measurements      Differences of order 2 x arise in scaling-up chamber data to the in Table 5.The estimates span a wide range due to adoption of tower footprint using the 20 x 20 m pixels of the SPOT image to different global areas for the various types of bogs and fens and classify vegetation type.Tundra ecosystems are very heterogene-various rates and periods for CI-14 emission from each type of ous, with dramatic variations of soil moisture and plant assera-vegetation.Arctic tundra lakes are important sources of methane, blages on length scales from centimeters to kilometers.Pixels about half the flux in the Yukon-Kuskokwim delta, but these are classified as "dry" unavoidably include some wet soils, and vice not considered in most global estimates.versa.Lake margins and drainage channels are likewise often not Present results suggest that global sources of CH4 from tundra resolved.The SPOT image provides resolution higher than typi-should be at the lower end of the range shown in Table 5.The cally used in regional remote sensing studies, and the magnitude ABLE 3A site is considerably wetter than average, and should of the differences clearly indicates the need for caution in scaling-therefore produce more CH4/me/yr than global mean tundra.up point data to regional scales using remote sensing data.
Nevertheless    than reported in many earlier data sets, even though the site was wetter than the global mean tundra and therefore expected to be origin at the tower and with the x axis pointing in the upwind direction.

Interpretation of F in terms of surface properties requires an evaluation of g (x,y). Recent theoretical studies [
tracks is expected to b• and vertical concentration profiles for CH4 and CO2 on a 12-m significantly larger than in surrounding vegetation [Chapin et al., micrometeorological tower at an Arctic tundra site, obtained as 1988].Water levels declined as the season advanced in all areas, part of NASA's Arctic Boundary Layer Experiment (ABLE 3A).and the depth to l•rmafrost increased, from about 12 cm initially The dependence of methane emission rates on time of day, wind in the dry tundra to about 20 cm at the end of the experiment.dixection, and advancing season is examined, allowing Us to deter-Locations of areas covered by lake, upland and wet meadow mine the influence of incident solar radiation, wind speed, and tundra, shown in Figure la, were derived [Bartlett et al.; this isvegetation type on ecosystem uptake of CO 2 and emission of CH4.sue] using a System Probatoire d'Observation de la Terre (SPOT) The fluxes derived from eddy correlation measurements on the satellite image (resolution 20 m x 20 m).Areas of each type are tower are compared with airborne eddy correlation flux data summarized in Figure lb for sectors selected to discriminate [Ritter et al., this issue] and with enclosure measurements eagled among measurements obtained with the wind traversing different out simultaneously at the site [Bartlett et al., this issue; Whiting et types of vegetation.The sector betwe,•n 120 ø and 170 ø was domal., this issue].inated by dry tundra, 170ø-230 ø was mostly wet tundra with a small lake (Lake Biospherics Re. search: Emissions from Weftands 2. STUDY SITE (BREW)), and 2300-360 ø was a mixture of dry and wet tundra without significant lakes.The sector 0-120 ø was influenced by The experimental site was located in the Yukon-Kuskokwim two large lakes and includes the generator.Lake ABLE, the River Delta in southwestern Alaska, 50 km NNW of Bethel.Fig-closest, served as runway for float planes that provided transportaure la shows the spatial distribution of vegetation types at the site, tion to the nearest town (Bethel).The lakes were 0.5-1 m deep, classified as discussed below.A 12-m micrometeorological tower with relatively lush growth of vegetation (Carex spp., riophorurn (the origin in Figure la), located at the west end of the camp, was spp.) at the margins.the main platform for measurements.A diesel generator was lo-We examined flux measurements for wind directions between 0 cated 300 m to the east of the tower, connected to the tower by a and 120 ø, for chemical (CI-I4, CO2, and O3 [Jacob et al., this iswooden boardwalk (for details of the geometry, see Figure 1 in sue]) and physical tracers (momentum, sensible heat, and latent Fitzjarrald and Moore, [this issue]).There were five camp tents heat; [Fitzjarrald and Moore, this issue]), and concluded that the set up along the boardwalk, with most instruments housed in the data were not systematically biased by the presence of the tent tent closest (20m)to the tower.structures at the site.We include data from this sector in our Bartlett et al. [this issue] classified the vegetation into two analysis, excluding brief periods of pollution from the generator broad categories: wet meadow tundra and dry upland tundra.Wet using observations of NO r as an indicator (see below).

Fa_
Fig. la.Three surface types (lake, dry upland, and wet meadow tundra) are distinguished for the site, based on a SPOT satellite image (resolution 20 x 20 m).The tower is located at the origin.Four sectors are divided to represent regions of the Arctic tundra with different distributions of surface vegetation.The circle is 1000 rn radius.
Fig. 2. Schematic drawing of the sampling tower.Points labeled are 1, the chemical sensors.Atmospheric CH4 was measured using a inlet for eddy correlation instruments at 12 m, lashed to the rotating fast response flame ionization detector (FID) and by a prototype boom that carried the tri-axial sonic anemometer, 2-9, inlets for profile fast response HeNe laser methane monitor.Ambient CO 2 was instruments at 10.8, 8.5, 6.1, 4.3, 3.1, 1.5, 0.5, and 0.02 m altitude, measured using a Beckman (Model 865) nondispersive infrared respectively; 10 inlet selection box for profiling.Inlets were protected CO2 analyzer.Air samples were preconditioned to a constant from rain by small funnels.For locations of additional instrumentation, temperature and dew point (2øC) using a Teflon-coated aluminum see Harriss et al. [this issue].

Fig. 11 .
Figure 12a shows the average NEE for CO2 for day and night, chamber measurements of CI-I4 emission rates into three classes for each sector.Nocmmal respiration is essentially the same in all based on surface type: dry upland tundra, wet meadow tundra, directions, but daytime uptake is markedly smaller in the sector and lakes.Mean CH 4 emissions from wet tundra were 80 dominated by dry tundra (120ø-170ø).The smaller rate for pho-mgCH4/m2/d, while dry tundra emitted only 3 mgCH4/m2/d.tosynthesis in dry tundra, as compared to wet tundra, is consistent Lake emissions, estimated by Bartlett et al. [this issue] from surwith observations [Whiting et al., this issue] based on chamber face concentrations and an assumed piston velocity, were intermeasurements.Significant photosynthesis occurred in the lake-mediate, 44 mgCH4/m 2/d for Lake ABLE.These data were metidominated sector (0-120ø), likely associated with emergent vege-culously extrapolated to regional scale using a Landsat scene tation at the lake margins.where each 500m x 500m pixel was assigned to one of these sur-The average methane emission rate for the whole experiment at face types and assumed to have the associated mean emission rate.the tower, 25 mgCH4/m2/d, represents 6% of the net daily uptake We may help elucidate the uncertainties associated with extraof CO 2 (Figure 8a), similar to values obtained for analogous sys-polating point measurements to large scales, as needed to apply retenas.Sebacher et al. [1986] estimated that methane losses in a mote sensing techniques, by performing a similar scaling-up exercoastal tundra corresponded to 8.6% of net prim• productivity cise for the tower site and comparing to direct flux measurements.(NPP).Similar estimates for a minerotrophic wetland [Svensson, Tower observations respond to ecosystem activity over scales Fig. 14.Nighttime concentrations of THC and CO2 are well correlated when the gases are emitted from the surface and accumulated in the surface layer.Levels 1-7 were located between 0.5-10.8m and level 8 at 0.02 m above ground.
wet mea•0w tundra, respectively, J• is the effective area-Table 3 compares fluxes from the three surface types derived weighted fraction, and u is the wind speed.Models differ in the from eddy correlation measurements and from the enclosure stu-number of significant or independent variables as shown (see text).dies of Bartlett et al. [this issue].The tower data indicate substan-Here, rse, residual standard error of the regression; r 2, fraction of variance explained by the model.tially higher emissions from "dry tundra" pixels than reported by " of fitted parameters for the tower data and of measurements for the enclosure data.Enclosure data for lake flux is calculated from surface CI-I4 concentrations and assumed exchange velocity [Bartlett et al., this issue].the enclosure studies (11 versus 3 mgCI-h / m 2 / d) and significantly lower emissions from "wet tundra" pixels than observed over wet tundra (29 versus 79 mgCH4/m2/d).Contributions from lakes were similar, a somewhat surprising result, since Bartlett et al. [this issue] had to infer fluxes indirectly for lakes.
, methane fluxes measured on the tower over the 30-The errors incurred in scaling-up flux data could perhaps be mi-day period averaged 25_+1 (s.e.) mgCH4/m e/d, lower than, or tigated if information on fractional coverage of surface types equivalent to, data used for global mean tundra in most estimates.could be incorporated into satellite imagery.A vast expansion of A low emission rate is supported by aircraft, tower, and chamber the available data would be required, however, to significantly ira-data.prove large-scale extrapolations.Moreover, even fine-scale vege-If the mean tower emission rate applied to the global tundra tation classification cannot account variance observed within a area of about 7.3 x 10 lem e [Matthews, 1983], for a typical active surface type; for example, fluxes at wet meadow sites ranged over period of 120 days [Bartlett et al., this issue; Whalen and Reetwo orders of magnitude, apparently reflecting variations in burgh, 1988; Sebacher et al., 1986; Matthews and Fung, 1987], a methane-producing substrates, soil temperatures, nutrient inputs, total of 22 Tg/a would be released globally by arctic umdra.This primary productivity, and other factors [Harriss and Sebacher, is only about 5% of the global CH4 source.The careful extrapola-1981; Sebacher et al., 1986; Bartlett et al., this issue].Seasonal tion by Bartlett et al. [this issue], which accounts for the prevariations may also be substantial.valence of dry tundra over the globe, suggests about half as much Table 4 compares fluxes from aircraft overflights with tower CH4 derived from global tundra.These results point toward a reflux data for 03 and CHq.The tower data were averaged over the latively minor role for tundra in the global CH4 budget.This conafternoon of each flight, to account for averaging associated with clusion should apply unless umdra lands elsewhere are for some turnover of the planetary boundary layer.We focused on aircraft reason much more productive than the relatively wet terrain in the measurements after solar noon (1500 local time), corresponding to Yukon-Kuskokwim River delta. the smallest flux divergence in the boundary layer [Ritter et al., this issue].The tower and aircraft data are remarkably dose for 6.SUMMARY both the ozone and methane fluxes.Methane fluxes were, however, about twice as high in the coastal tundra as measured from the Eddy correlation flux measurements and concentration profiles tower site, reflecting largely differences in soil moisture and sur-for THC and COe were combined to provide a comprehensive face vegetation [see Figure 18 in Ritter et al., this issue].record of atmosphere-biosphere exchange for these gases over a It is important to note that the aircraft flux flights took place on 30-day period in July-August 1988, in the Yukon-Kuskokwim warm, rain-free days in the afternoon, times favoring peak emis-River Delta of Alaska.A prototype methane monitor successfully sions (see Figure 8b).The average tower flux for these after-measured CH4 concentrations and fluxes, showing that net ecosysnoons, 55 mgCH4/me/d, was more than twice as large as the tern exchanges of THC were >90% due to methane.grand mean of the tower flux measurements, implying that aircraft Lakes and wet meadow tundra provided the major sources of observations (averaging 50 mgCHn/me/d) were likely biased, by methane.Emissions from lakes were strongly dependent on the roughly a factor of 2, as compared to the mean for the region in surface wind speed.The average fluxes from lake, dry tundra, and the growing season.This bias should be taken into account when wet tundra, identified from 20 x 20 m pixels in a SPOT satellite scaling-up aircraft data to obtain regional fluxes.image [Bartlett et al., this issue] were 11ñ3, 29ñ3, and 57ñ6 Estimates for methane emissions from the global tundra, report-mgCH4/m e / d, respectively.The mean emission rate for the site ed from previous studies based on enclosure measurements was 25 mgCH4/me/d during the 30-day period.Mean fluxes [Svensson and Rosswall, 1984; Sebacher et al., 1986; Whalen and depended on wind direction reflecting the angular distribution of Reeburgh, 1988, 1990; Bartlett et al., this issue], are summarized surface types at the site.The average emission rate was lower tLake fluxes are excluded to be consistent with previous flux estimates; lakes provided -50% of the flux at the tower site, significantly more than expected for global tundra.
tower results are in harmony with measure-Here u is the wind speed, and o• and o, are Pasquill-Gifford ments of CI-I4 flux at the site obtained using traditional chamber dispersion parmeters which are functions of x and of atmospheric methods [Bartlett et al., this issue] and with aircraft eddy correla-stability.We parameterize % and 0, to the footprint of the Table A1 lists values of the coefficients L J, K corresponding to tower measurements (10am).The scaling procedure relied on as-the Pasquill-Gifford stability classes C, D, E (slightly unstable, signment of a surface type and associated CI44 flux to each 20 x neutral, and slightly stable conditions, respectively).Using the 20 m pixel in the SPOT image.At this resolution, heterogeneous flux-gradient assumption, we derive g (x,y) directly fxom methane-producing habitats were not fully resolved; tower fluxes •(d2cxo)/•z: for pixels assigned to wet or dry tundra differed from chamber data for that surface type by approximately a factor of 2, apparently as a result of the admixture of habitats at satellite resolution.g (xx,y) is the surface emission flux at location (x,y), and g(x,y) is a probability density representing the contribution of point (x,y) to the footprint.The coordinate system is defined with

Fn
Figur• A1 shows th• two-dimensional field of g (x,•) computed on th• integral probability density •, r•pr•s•nting th• fraction of from equation (A6) for a ngutral atmosphere.Thg cross-wind gx-the towgr footprint contributed by surface type i: tent of the footprint is small.W• find that >99% of th• footprint is contained within 15 ø of the x-axis for stability classes C and 3 higher.Highly unstable conditions (classes A and B) would cause F = •J•i (AS) a wider cross-wind spread, but such conditions are infrequent in i=l the surface layer.Figure 13 shows the cross-wind integral f (x) computed from We compute values of j• for three stability classes (C,D,E) and equations (A3) and (A6).Under neutral conditions we find that four wind sectors (see section 5): 50% of the flux is contributed by surfaces less than 300 m from the tower, with a maximum probability density for Xm• 150 m.In comparison, Schuepp et al. [1990] obtained Xm•, = 240 m for a j• = If (x)Sq(x)dx (A9) roughness height zo = 0.5 cm (as measured at the ABLE 3A tower [Fitzjarrald and Moore, this issue].Leclerc and Thurtell [1990] presented results from a Lagrangian numerical simulation with Here So(x) is the fractional area occupied by surface type i at disconditions zo = 0.6 cm and h = 9 m, similar to those at the ABLE rance x from the tower in sector j, and f (x) is taken from Figure 3A tower.They found that 50% of the flux was contributed by 13.Use of the cross-wind integral f (x) in equation (A9) is surfaces less than 400 m from the tower.Our calculated footprints justified by the short cross-wind extent of the footprint.Data for are slightly shorter than those calculated by Schuepp et al. [1990] S•i(x ) are taken from a map with 20 x 20 m 2 resolution (Figure 1).and Leclerc and Thurtell [1990]; a possible explanation is that the The resulting values of j• are listed in Table2a.

TABLE 1 . Results of Least Squares Fit to Nighttime (2300-0700 LT) Concentrations of THC Versus CO2
Here, std, standard deviation of the slope; rse, residual standard error.