Three-dimensional model synthesis of the global methane cycle

The geographic and seasonal emission distributions of the major sources and sinks of atmospheric methane were compiled using methane flux measurements and energy and agricultural statistics in conjunction with global digital data bases of land surface characteristics and anthropogenic activities. Chemical destruction of methane in the atmosphere was calculated using three-dimensional OH fields every 5 days taken from Spivakovsky et al. (1990a, b). The signatures of each of the souwes and sinks in the atmosphere were simulated using a global three-dimensional tracer transport model. Candidate methane budget scenarios were constructed according to mass balance of methane and its carbon isotopes. The verisimilitude of the scenarios was tested by their ability to reproduce the meridional gradient and seasonal variations of methane observed in the atmosphere. Constraints imposed by all the atmospheric observations are satisfied simultaneously by several budget scenarios. A preferred budget comprises annual destruction rates of 450 Tg by OH oxidation and 10 Tg by soil absorption and annual emissions of 80 Tg from fossil sources, 80 Tg from domestic animals, and 35 Tg from wetlands and tundra poleward of 50øN. Emissions from landfills, tropical swamps, rice fields, biomass burning, and termites total 295 Tg; however, the individual contributions of these terms cannot be determined uniquely because of the lack of measurements of direct fluxes and of atmospheric methane variations in regions where these sources are concentrated.

variations in the composition of these isotopes provide additional quantitative constraints on the budget of methane.Thus an indirect approach to study the atmospheric methane cycle would employ models of varying complexity to test the consistency between atmospheric methane variations and hypotheses about the sources and sinks.These models include globally averaged, height-dependent photochemical models for investigating the changing balance between sources and sinks [e.g., Chameides et ai., 1977;Thompson and Cicerone, 1986; Quay et ai., 1988], and photochemical-transport models in one, two, and three dimensions [e.g., Mayer et al., 1982;Crutzen and Gidel, 1983;lOtalii and Rasmussen, 1983;Blake, 1984;Jones and Pile, 1984;Fraser et ai., 1986a;lsaksen and Hov, 1987;Wahlen et al., 1989;Quay et al., 1991;Taylor et al., 1991].The range of source strengths in these models corresponds to the range of estimates of chemical lifetimes of methane in the atmosphere.Furthermore, the magnitude and locations of the sources are tested in toto, and partitioning among the individual source terms using the indirect approach is not unique.
In this study, we present a synthesis of the global methane cycle in an attempt to derive a justified global budget of methane for the 1980s.Variations of methane and its isotopic composition in the atmosphere (section 2) are used to constrain hypotheses about the sources and sinks.We have embarked on an effort to map the geographic and seasonal distribution of all the major sources of methane.For a few of the sources (for example, wetlands) there is some information about their strengths as well as geographic distributions; while for other sources (for example, natural gas venting), geographic distribution can be deduced but direct estimates of the global source strength are difficult.A global three-dimensional tracer transport model (section 3) is used to simulate the atmospheric methane response to each of the sources and sinks (section 4).Candidate budgets that satisfy all the constraints (section 5) are defined methane calibration scale provides confidence that the concentration gradients indicated by the data, and which we are attempting to simulate, are real.
Full details of the methodologies used for obtaining methane concentrations from the NOAA/CMDL network, and the first 2 years of data, have been described by Steele et al. [1987].Further details and additional data can be found in the works of Lang et al. [1990a, b].The NOAA/CMDL methane data are also available as a numeric data package [Steele and Lang, 1991].
Methane data covering the 4-year period 1984-1987 are used in this work.For 16 of the 19 sites a complete 4-year record is available.Because of some gaps in the flask record, there are only three complete years (1984,1985,1987) of data for PSA (65øS) (cf.Table 1 for the explanation of the station codes and locations of the stations).Also, because of later starts in flask sampling, only three years (1985)(1986)(1987) of data from CHR (2øN) and 2 years (1986)(1987) of data from ALT (82øN) were used.For comparison with the model results the methane flask data have been processed to extract two major features.
The first feature of the methane data is the annual mean concentration at each site, averaged over the period of observations.To derive this quantity we first calculated, for each site, the annual mean concentration for each calendar year as the arithmetic mean of the monthly average values.Each monthly average concentration is itself the arithmetic mean of the individual flask values for that month [see Steele et al., 1987].Then to obtain the average annual mean concentrations for the period, the long-term growth rate of atmospheric methane must be removed from the data.To a very good approximation, the globally averaged growth rate of methane during this 4-year period was identical to the growth rate observed at SPO (90øS).Thus we subtracted the year-to-year growth increments seen at SPO from the annual means at each site, effectively normalizing all of the data to the annual average concentration measured at SPO in 1984.At all of the sites except for SPO, this procedure results in a set of detrended annual means which may then be averaged to obtain the average and standard deviation of the annual mean concentration.
The other feature derived from the data set is the average seasonal variation of atmospheric methane over a full year at each site.The procedure adopted here is the same for each site.
The time series of monthly average methane concentrations were used to calculate 12-month running mean values.A least squares linear fit to the 12-month running means was determined and subtracted from the time series of monthly data.The detrended monthly values were then averaged by month, i.e., if there were 4 years of data for the site, all four of the detrended values for January were averaged, and so on for the other months.Average values and standard deviations were calculated.
Figure 1 shows the latitudinal and seasonal variations of atmospheric methane in the remote marine boundary layer derived from the NOAA/CMDL network.Annually averaged methane concentration is highest at BRW (71øN), which is 150 ppbv (parts per 109 by volume in dry air) above that of SPO (90øS).At the southern hemisphere middle-to high-latitude sites, methane peaks in September/October and minimizes in 15oo February, with a peak-trough amplitude of -30 ppbv.At the northern hemisphere middle-to high-latitude sites, methane appears to show a complex double maximum in the year.It is clear that the average seasonal cycle is both simpler and better defined in the southern hemisphere than at most of the northern hemisphere sites.The annual mean concentrations and seasonal cycles averaged over the 4 years are given in Table 2.In addition to observations from the NOAA/CMDL network we also used methane measurements made at two locations by Central to the use of

TRANSPORT MODEL
The study of tracer distributions using a three-dimensional global general circulation model (GCM) was pioneered by Mahbnan cmd Moxim [1978].The tracer transport model employed in this study [Russell and Lo?•er, 1981] uses winds generated from the GCM at the Goddard Institute for Space Studies (GISS) [Hansen et al., 1983]   There is a problem in the location of the Intertropical Convergence Zone in the Atlantic Ocean in the GCM [Druyan andRind, 1988, 1989] that affects the simulation at ASC (8øS).

Most of the measurement sites listed in
While ASC is observed to be influenced by southern hemisphere air, it is in the northern hemisphere circulation regime in the model.Since there is little latitudinal gradient in methane concentrations in the southern hemisphere [Steele et al., 1987], the model results for the gridbox southward of ASC are used.In this way, interpretation of the methane simulations is not obfuscated by deficiencies in the meteorology in the GCM.

ATMOSPHERIC SIGNATURES
A first step in using the geographic variation of atmospheric methane to infer sources and sinks is some knowledge of the patterns of the sources and sinks themselves.We have compiled, at 1øxl ø resolution for the globe, a series of digital data bases of land surface characteristics and parameters important for the study of biogeochemical cycles.The primary data bases include vegetation and land use [Matthews, 1983], soils [Zobler, 1986], and countries and states [cf.Lerner et al., 1988].Built upon these are data bases of natural wetlands [Matthews and Fung, 1987], animal populations [Lerner et al., 1988], rice cultivation [Matthews et al., 1991], human population densities and associated activities such as municipal solid waste disposal.In addition, locations of coal mines, oil wells, and natural gas wells have been digitized from available atlases.
While we have relatively detailed information about source locations, making estimates of source strengths is difficult.For A 4-year model experiment was run for each source scenario, by which time a stable north-south gradient in atmospheric methane is obtained in the model.Because emission rates are independent of the atmospheric concentration, each model experiment was started from a globally uniform background concentration of zero.The resultant increasing atmospheric concentrations were linearly detrended using the global, annually averaged growth rate in the experiment, thus preserving the geographic and seasonal variations of the simulated atmospheric distributions.The calculation of chemical destruction rates, which is dependent on the total methane concentration, is described separately in section 4.11.
To facilitate the analysis, the simulated three-dimensional distribution of methane concentration X(,k,O,ty,t), is separated   4.
For the stations listed in Table 1 we

Natural Wetlands
In our global digital data base of natural wetlands [Matthews and Fung, 1987], there are five major wetland ecosystems grouped according to characteristics important for methane emission: forested bogs, nonforested bogs, forested swamps, nonforested swamps, and alluvial formations.The global wetland area is 5.3 x 1012 m 2, with -50% of this area occurring as forested and nonforested bogs between 50øN to 70øN.A data set of natural wetlands has also been compiled by Aselmann and Crutzen [1989] who obtained a global area of 5.7x 1012m2, within 10% of ours.Relative regional distributions of areas are also very similar between the two works: their areas are -6% higher between 50øN to 70øN and --6% higher in the tropics.Given the scattered distribution of wetlands over most of the globe, agreement between the two data sets is considered good.The data set of Matthews and Fung [1987] was used in this study.
We investigated two hypotheses about the seasonal emission characteristics from wetlands and their impact on atmospheric methane variations.In experiment W1 the limited series of field measurements of emission rates published prior to 1987 were broadly extrapolated to represent emission rates for the entire wetland group.Peat-rich bogs were assumed to have "typical" emission rates of 0.2 g CH4/m2/d, while forested swamps and nonforested swamps were assumed to have lower   See Table 3  For this study we modeled both emission seasons and emission rates based on the climatology of monthly surface air temperatures (Ts) and precipitation [Shed, 1986].The modeling was carried out separately for bogs (experiment WB), swamps/alluvial formations (experiment WS), and dry tundra (experiment WT).
Most wetlands experience seasonal wet/dry cycles and/or freeze/thaw cycles.In our model, methane is emitted only during wet and/or thaw seasons.For a location that experiences above freezing temperatures throughout the year, wet season was assumed to occur when monthly precipitation exceeds monthly potential evaporation.Potential evaporation is the maximum demand for moisture and is calculated from Ts using Thornthwaite's formula (1948).Thaw season was assumed to begin when monthly mean Ts rises above 5øC and assumed to end when monthly mean Ts falls below 0øC.The Ts criteria for freeze/thaw were chosen to match the observations of Whalen and Reeburgh [1988] at the University of Alaska site.With this definition, the duration of methane emission ranges from 12 months per year at the headwaters of the Amazon River to 5 months at its mouth, and from 6 months in the maritime south coast of Alaska to 4 months on the north coast.
It remains to determine the variation of emission rates during the emission season.Seasonal time series of methane fluxes from tussock and carex sites in Alaska [Whalen and Reeburgh, 1988] show that methane emission rates appear to peak with Ts in August.We therefore assumed, as a first approximation, that methane emission rates for forested and nonforested bogs in experiment WB are temperature-regulated.The temperature dependence is assumed as:

(14øS) and ASC (8øS) of -4 ppbv above that at SPO (90øS).
There is very little concentration difference between the hemispheres.
The peak-trough amplitudes of the seasonal cycles resulting from wetland emissions are illustrated in Figure 4a      In Europe and South Asia where there are high densities of cattle, dairy cows, and water buffalo, the simulated surface methane concentration is >20 ppbv above the global mean (Figure 2f).The central plains of North America and the Sudeste (southeastern) region of Brazil, both with high cattle populations, also exhibit elevated surface methane concentrations.The north-south profile of annual mean methane concentration simulated at the NOAA/CMDL sites is shown in Figure 3b for AN(80).The results for AN(50) are included for comparison with other aseasonal mid-latitude sources.For AN(80), fairly uniform methane concentrations (-20 ppbv higher than the South Pole) are simulated in midlatitudes in the northern hemisphere, and the concentration decreases monotonically toward the South Pole.As may be expected from an aseasonal source, intra-annual fluctuations in methane concentration due to seasonally varying circulation patterns are small (Figure 4b).The peak-trough range of methane oscillation due to this source is -5 ppbv for the NOAA/CMDL sites in middle to high latitudes in the northern hemisphere, -4 ppbv in the tropics and subtropics (KUM (20ø1'4), SMO (14øS), ASC (8øS)), and -2 ppbv at CGO (41øS).

Landfills
Anaerobic environments may be found in landfills that are several years old.Decomposition of biodegradable organic material in these landfills produces carbon dioxide and methane which may escape to the atmosphere.Bingeruer and Crutzen [1987] have made a survey of the production of municipal and industrial waste and the biodegradable carbon content of the waste.Assuming that the methane proportion of landfill gas is constant and that all the methane in landfill gas escapes to the atmosphere, they estimated the global production of methane in landfills to be 30-70 Tg CH4/Yr.The amount of methane that escapes to the atmosphere may vary from site to site, and may depend on, inter alia, the carbon content of the landfill and the porosity of the cover.Explosive levels of methane have been found at some sites (C.Moore, personal communication, 1988), while undecomposed organic material is found at others [Rathje, 1989].Methane consumption in landfill cover soil also reduces the emissions [Whalen et al., 1990].Bingeruer and Crutzen [1987] expect the emission estimates to be lower than the production estimates of 30-70 Tg.Most of the methane from landfills is from North America, where both the per capita production of municipal solid waste and the biodegradable fraction of the waste are high.
We combined our global data base on human population densities with per capita statistics on biodegradable carbon given by Bingeruer and Crutzen [1987]   For GV(50) the simulated concentration is -15 ppbv higher at the northern hemisphere NOAA/CMDL sites than at the southern hemisphere sites (Figure 3b).This north-south difference is smaller than that caused by sources more proximate to the monitoring sites, for example, LF(50).Similarly, the seasonal cycles simulated at the sites for GV(50) are small compared to the observations (Figure 4b).

Natural Gas Consumption
With global annual natural gas consumption equivalent to -1200 Tg CH4/Yr , transmission loss or pipeline leakage of natural gas may be a nonnegligible source of atmospheric Transmission loss of methane is assumed to be a globally uniform fraction of gas consumption.In experiment GL a basis source strength of 50 Tg was used.
The atmospheric methane response to GL(50) is shown in Figure 2a'.The near-surface methane concentration is elevated by -30 ppbv over Eastern U.S.A. and by -40 ppbv over Central Europe where natural gas is in common use.The concentration is -25 ppbv higher at middle latitudes in the northern hemisphere monitoring sites than at SPO (90øS) (Figure 3b).Because this source is widely distributed in North America, it gives the highest north-south gradient at the monitoring sites of all the aseasonal mid-latitude sources investigated.

Coal
Methane is the major component of coal gas, and is released to the atmosphere in the mining and processing of coal.We digitized, at løxl ø resolution for the globe, the locations and approximate sizes of coal mines [Espenshade, 1978;Central Intelligence Agency, 1978;1986 (Figure 3c).The peak-trough amplitude also peaks at -6 ppbv at CHR (2øN) (Figure 4c).
Termite habitats are generally located in tropical grasslands and forests [Zimmerman et al., 1982;Fraser et al., 1986b].We mapped the potential habitats of termites using the global vegetation data base of Matthews [1983].Assuming an aseasonal source of 50 Tg/yr, we distributed the emission among ecosystems according to the tabulations of Zimmerman et al. [1982Zimmerman et al. [ , 1983] ]  shows maximum elevations of 6-8 ppbv above the global mean in the tropical regions.The effect on mid-latitude concentrations is small.The peak concentration in the tropics is 6 ppbv above that of SPO (Figure 3c).Correspondingly, the simulated peak-trough amplitudes at the monitoring sites are small and due entirely to variations in atmospheric transport (Figure 4c).

Hydrates/Clathmtes
The potential release of methane from hydrates/clathrates is highly uncertain.While methane hydrates are known or inferred to be found on the continental shelf at all latitudes [Kvenvolden, 1988], it is generally hypothesized that those hydrates located in the Arctic, in particular the offshore permafrost, are subject to destabilization due to climate warming [Bell, 1983;Revelle, 1983;Kvenvolden, 1988].
To test the atmospheric methane signature from such a source, we simulated the atmospheric methane distribution in response to two hypothetical methane sources in the Arctic.For the case of multiple sources, the concept of separate simulations for each source followed by a linear recomposition of the signal at an observing site is mathematically rigorous.In the case of chemical destruction in the atmosphere, methane loss is linearly proportional to its concentration and hence cannot be strictly separated from its local abundance.The errors in the approach adopted here are limited by the variations in methane about the assumed value.Within each hemisphere, average latitudinal variations in methane are obsemed to be less than the interhemispheric gradient of 6% (see Figure 1).Similarly, obsemed seasonal variations over the year at remote sites in mid-latitudes are at most $% (50 ppbv).N, ear and above source regions, however, the abundance may be elevated by 20% (330 ppbv) or more, but the fraction of the atmosphere (and also the chemical loss) in such regions is small at the global scale.Any such errors will be further damped by the inherent feedbacks in the OH-CH4-CO system [Sze, 1977] where high values of CH 4 and CO depress OH concentrations.Thus we estimate that errors in the total budget or in the latitudinal and seasonal patterns of CH 4 presented here under the assumption of fxxed losses are less than 5% of the calculated signal (for example, the seasonal amplitude at high N and S latitudes is 40+_2 ppbv).The distribution of chemical loss of CH 4 is summarized in Table 5 for   The atmospheric methane response is shown in Figures 2p, 3b  and 4b, where for convenience the response was plotted with a sign change for direct comparison with the mid-latitude aseasonal sources.Concentrations are depressed by -18 and -28 ppbv below the global mean over boreal forests of North America and Siberia.At the monitoring sites the largest northsouth gradient, -26 ppbv, is found between STM (66øN) and SPO (90øS).The largest seasonal amplitude is again found at high latitudes due to the seasonality of the circulation.

Other Sources and Sinks
We have not carried out three-dimensional simulations for scenarios of several methane sources and sinks that are extremely poorly known or whose contribution to geographic and intra-annual variations may be very small.'l"hese sources/sinks include (1) oceans and lakes [Ehhalt, 1974] 3. Rj for each source was selected from within the measured range to satisfy Equation ( 6).
Seven candidate budget scenarios are listed in Table 6.

ANALYSIS OF BUDGET SCENARIOS
Seven budget scenarios are constructed according to the constraints C1-C7 outlined above (Table 6).Budget scenario 1 is that given by Cicerone and Oreroland [1988].Scenarios 2-4 used a chemical destruction rate of 500 Tg/yr, while scenarios 5-7 used a destruction rate of 450 Tg/yr.For each chemical destruction rate, fossil fractions comprising -27, 20 and 16% of global sources were investigated.The total atmospheric methane distribution corresponding to each of the budget scenarios was calculated as the linear combination of the responses to the individual source/sink functions.In this section, we apply constraints 8 and 9 and test whether the modeled atmospheric distribution matches that observed.
In scenario 1 the wetland source is W1, with a global strength of 110 Tg, and seasonality of emission given by Matthews and Fung [1987].A tundra emission of 5 Tg (WT) was included.As can be seen in Figure 5, the simulated pole-to-pole difference in concentration is -200 ppbv, an overestimate of -60 ppbv.Figure 6  The atmospheric concentrations simulated with scenarios 2-7 compare well with the observations, both in terms of the annual mean north-south gradients (Figure 5) and the seasonal cycles at the stations.Figure 7  To examine the geographic constraints on methane sources and sinks, it is easiest to start with regions with a single source/sink, or regions with distinctive atmospheric signatures.Such a region is the mid-latitudes in the southern hemisphere Hence, even if their source strengths were equal, they would produce north-south concentration profries with similar shapes but different magnitudes at the monitoring sites (of. Figure 3b).The north-south profile of the annual mean atmospheric methane concentration alone thus presents some, but not a unique, constraint on these mid-latitude sources.
Together with the ratio of lnC/•2C in atmospheric methane, We consider a fossil fraction of 27% highly improbable unless inadvertent and unaccounted-for natural gas escaping at wells and refineries comprises a significant term in the global methane budget, or unless there is a large clathrate source remote from the monitoring sites, such as in the Soviet Arctic.We hence reject scenarios 2, 3, and 5, as the vented component would comprise over 100% of the natural gas flared and vented at the wells.Similarly, scenario 6, with a vented fraction of 33%, is considered unl•ely, although there are no data to eliminate it det'mitively.The magnitude of the fossil sources inferred here in scenarios 4 and 7 (16% of the global source strength) do not disagree with the direct estimates of Barns and Edmonds [1990].
It remains to seek constraints on tropical sources (swamps, termites, biomass burning, and rice cultivation).Tropical sources have little impact on the north-south gradient record• at the current network (cf. Figure 3c).Also, while their emissions may be seasonal, the majority of monitoring sites are too distant to record their impact.Hence one constraint on their total strength is the residual in the global budget.Table n, It is difficult to anticipate the vertical distribution of methane in the atmosphere as the major source and sink terms all maximize near the surface: a surface source would contribute to a negative vertical gradient with methane decreasing with height, while a surface sink would lead to a positive gradient.We show in Figure 9  In most of the upper troposphere (Figure 9b), the sign of the methane vertical gradient is similar to that of the lower troposphere: concentrations decrease with height by 10-20 ppbv in -200 mbar in the northern hemisphere and increase with height by --10 ppbv in --200 mbar in the southern hemisphere.
An interesting exception is over the continental source regions in the southern hemisphere where the vertical concentration gradient changes sign from that in the lower troposphere.Here and elsewhere in the southern hemisphere, northern hemisphere methane is advected aloft across the equator and enhances the concentration increase caused by less effective OH destruction with height.Thus the vertical gradient is steeper in the upper than in the lower troposphere in the southern hemisphere.Figure 9b shows also a large decrease of methane over Greenland and Antarctica.This is an artifact of using concentrations at layer 7, which represent in these regions stratospheric air with very low methane concentrations.
There are very few methane profries to support the modeled vertical distributions.The only long time series aircraft data available for the period of investigation here are the CSIRO aircraft data over Southeast Australia [Fraser et al., 1986a].The data show that, averaged over 1979-1984, the annual mean concentration of methane increases by -6 ppbv from the surface to 3.5-5.5 kin, and by another -9 ppbv from 3.5-5 to -7-10 kin.This is in reasonable agreement with the modeled increases of -5 and --10 ppbv shown in Figures 9a and 9b Detailed mapping of the locations of the sources and sinks was carried out using available information on vegetation, mils, land use, and economic statistics.The most important natural source is wetlands, and our previous work presented a detailed compilation of wetland distribution using vegetation, mils, and inundation information.Because of their economic importance, the distribution of rice cultivation areas, animal populations, coal mines, and natural gas wells can be derived from national and global land use and energy atlases and statistics.Similarly, we assumed that landf 'all sizes and natural gas consumption within a count• or political unit are proportional to the human population density within the unit and thus can be mapped using the distribution of population density.We consider these distributions to be fairly well known, although we have no measure to express our confidence quantitatively.While it is generally acknowledged that biomass burning occurs mainly in the tropics, the areal extent, intensity and other characteristics of the fires are not known on a global, annual basis.We assumed that this source is colocated with the sources of CO 2 from deforestation and land-use modification.The information on termite populations and soft absorption is highly speculative, and these sources/sinks are crudely mapped based on their association with vegetation.
Over haft of the surface sources/sinks were assumed to be constant through the year.For the seasonal sources the timing of the emissions was either derived directly from available statistics (rice cultivation) or modeled using climate information (wetlands, biomass burning).
There The methane budget cannot be uniquely deemed using the data currently available.We presented six global methane budget scenarios that appear to satisfy all the constraints imposed by the atmospheric observations.While a single scenario cannot be selected objectively from the suite, we prefer scenarios with a low fraction of fossil sources as we demand that surface fluxes implied by a particular source strength are within reasonable bounds established directly from the statistics on emissions.It turns out that this choice hinges critically on the amount of methane vented at natural gas wells, a term for which there are no direct measurements.
The low fossil fraction is consistent with the direct estimates of Barns and Edmonds [1990] who also found very little information on the uncombusted fraction of methane vented and flared at wells.
Thus while we have a preferred budget scenario, the supporting evidence for some of the terms is weak and the scenario may be revised when we learn more about the treatment and fate of natural gas at the wells.
In summary, we prefer scenario 7 as the most likely budget.This scenario comprises an annual destruction of 450 Tg by OH oxidation and 10 Tg by soil absorption, and an annual emission of 80 Tg from fossil sources, 80 Tg from domestic animals, and 35 Tg from wetlands and tundra poleward of 50øN.Emissions from landf'dls, tropical swamps, rice fields, biomass burning, and tenTLites total 295 Tg; however, the individual contributions of these terms cannot be determined uniquely.This budget is preferred for the following reasons.First, the methane lifetime corresponds to a CH3CCI 3 lifetime of 6.3 yr [Prinn et al., 1987] and is consistent with the lower rate coefficient derived recently •t•cVaghjiani and Ravishankara [1991].Second, the total -free emission, the lowest of all the scenarios investigated, is consistent with that from direct estimates, even though the direct estimates are based on very sparse information.Third, we note that this scenario also matches successfully the northsouth and seasonal variation of 613CH4 in the atmosphere reported by Qua), et al. [1991].
Large uncertainties exist, in particular, about tropical and subtropical sources (rice, biomass burning, swamps, and termites), which make up > 50% of the total sources.Data available as yet are too scanty to be organized into any systematic framework to quantify emissions on a global scale: the measured fluxes span orders of magnitude, and there is no clear way to rank the different factors found to impact increased methane consumption due to the drying of softs [Whalen and Reeburgh, 1990b] or decreased consumption due to expansion of agricultural land [Keller et al., 1990] must not be ignored.

Fig. 1 .
Fig. 1.Three-dimensional representation of the global distribution of the background concentration of methane measured in the marine boundary layer at the NOAA/CMDL cooperative flask network for 1984-1987.The arrow denotes the equator.The raw data have been smoothed to provide a uniform grid spacing of 10' in latitude and 10 days in time.See text and Table I for a discussion of the data used.
sites were treated in the same way as the NOAA/CMDL data described above to obtain the annual mean and seasonal cycle at each site.The data were used to expand the observational base of the seasonal variation of atmospheric methane.However, as these data have not been precisely intercalibrated against the NOAA/CMDL data, they were not included in the analysis of the north-south gradient of atmospheric methane.Data on isotopic content of atmospheric methane and the isotopic signatures of methane resulting from the production and destruction of methane by different processes provided additional information about the global methane cycle.These are summarized below.A 14C content of atmospheric methane of -122 percent modern (pM) was reported for the northern hemisphere for 1987-1989 by Wahlen et al. [1989] and Quay et al. [1991].In the southern hemisphere for the same period, both these studies reported values of 120-121 pM, close to the values reported by Manning et al. [1990].The determination of the fraction of fossil sources in the global methane budget depends on the model and parameters used and is further complicated by uncertainties in 14CH4 emission from pressurized water reactors.These 14C values are interpreted as 16, 21, and 24.3% •4C-free methane in the total emissions by Quay et al. [1991], Wahlen et al. [1989] and Manning et al. [1990], respectively.Uncertainties in these estimates are large, of the order of 9%.This fraction of fossil carbon is derived primarily from coal and natural gas sources and includes any methane from clathrate destabilization.Methane depleted in •4C may also be derived from rice paddies, peat-rich wetlands and tundra.Direct measurements of 14CH4 fluxes show close-to-modern methane emanating from these biogenic sources in North America [Wahlen et al., 1989; Quay et al,.1991].Recent measurements show that the active layer of peat in Alaska has a reported values of -47.7+_0.2ø/oo for both hemispheres in 1978-1983.More recently, globally averaged values of -46.7 ø/oo for 1986-1987 were presented by Wahlen et al. [1989] and -47.2 ø/oo for 1986-1989 by Quay et al. [1991].Stevens [1988] and Stevens and Engelkemeir [1989] have found 13CH4 trends at several stations in each hemisphere for the period 1978-1989.Also, an increasing 613C trend in the past 100 years has been determined from analysis of air bubbles trapped in Greenland ice cores [Craig et al., 1988].Because the sampling sites are few and the spatial variability of •3CH4 in the atmosphere is unknown, we cannot evaluate if the reported values are hemispherically or globally representative.
statistics of the atmospheric circulation.Full seasonal and diurnal cycles are resolved in the GCM.Fourhourly winds, convection, and rainfall statistics, as well as 5-day or monthly patterns of convective frequencies from the GCM were saved for use in the tracer model.In this study, circulation statistics from 1 year are used and are repeated for multiyear experiments.The tracer model has been applied to the study of atmospheric distributions of CO 2 [Fung et al., 1983; Fung, 1986; He#harm et al., 1986; Fung et al., 1987; Tans et al.simulation is needed for the north-south gradient to stabilize (assuming an interhemispheric exchange time of 1.0 year).In order to facilitate a large number of model experiments we have made several simplifications to save computer time.First, a coarseresolution model (8øx10 ø) was constructed using the winds and mixing parameters from the 4øx5 ø GCM.The coarseresolution tracer model thus retains the circulation statistics of the 4øx5 ø GCM but uses -22 min CPU per tracer year.Second, we note that the largest methane variations observed are small (150 ppbv pole to pole) compared with the global annual mean (1625 ppbv in 1984).Hence the methane simulations can be reduced to a quasi-linear problem, where the total response can be approximated by the linear sum of the responses to each of the source/sink functions.For sources, the linear approximation is exact; for chemical losses, or other sources/sinks which depend on the absolute atmospheric concentration, the error in such an approximation is at most -10% of the signal.With such an approximation we can simulate separately the atmospheric response to a large number of hypotheses about each source/sink and examine their effects on the total atmospheric distributions in the diagnostic analyses after the model experiments.This will be discussed further in section 4.
animals and natural wetlands, methane flux rates and seasons from a few measurement programs are extrapolated via these data bases to obtain the global distributions of methane emission.These are used directly in the simulations.Large uncertainties are associated with methane emissions from rice cultivation, landfills, natural gas production and consumption, coal mining, biomass burning, and termites. of the linear decomposition discussed in section 3, atmospheric response to different source strengths or to a combination of sources can then be scaled proportionately.We shall use the shorthand WS(xx) to denote the case where results from experiment WS were scaled to a source strength of xx Tg/yr.In the construction of a global methane budget scenario (section 6), we held fixed the strengths of the better known sources and sinks, and varied the strengths of the lesser known ones until the combined atmospheric response satisfied all the constraints imposed by the atmospheric observations.In other words, the strengths of the sources and sinks are interdependent and are determined, in this study, from their combined signatures in the atmosphere.
shall use the shorthand &(STN) and •(STN,t), where STN is the station code, to denote the local annual mean and seasonal concentrations.For both model and observations, north-south profiles are defined relative to the South Pole, using O•(STN) -O•(SPO).For comparison, the simulated north-south profiles and seasonal cycles of atmospheric methane at the monitoring sites in response to the different sources are displayed in groups: seasonal sources (bogs, tundra, rice), predominantly northern hemisphere mid-latitude aseasonal sources (animals, landfills, natural gas production, natural gas consumption, coal mining) tropical sources (swamps and alluvial formations, biomass burning, termites), and arctic sources (clathrates).
Figure 3a for WB(30) and WT(5) and in Figure 3c for WS(50).The annual mean concentration at each of the observing sites is plotted relative to that at SPO (90øS).As is expected from the wetland distribution, concentrations are highest at BRW (71øN) for WB, and at MBC (76øN) for WT.In WB(30), northern hemisphere high-latitude sites are elevated by as much as 8 ppbv above the mid-latitude sites, which in turn are higher than the southern hemisphere by ---15 ppbv.Similar patterns are found for WT(5) where •)(ALT) -•(SPO) is -6 ppbv.By comparison, emissions from W$(50) are mainly from the southern hemisphere tropics, and so the NOAA network would register a peak concentration at SMO

FigureRH
Figure 2e shows the annually averaged near-surface distribution qb(•,0) of methane simulated for RH(50).Concentrations are elevated above background values by -16 ppbv at TKB (36øN) and -25 ppbv over India and China.By contrast, except at GMI (13øN), the concentration in the northern hemisphere NOAA/CMDL sites is elevated by only -3 ppbv.Because of the remoteness of the monitoring sites from the source regions, the north-south gradient simulated at the NOAA/CMDL sites for RH(50) is about half that for WB(30) (Figure 3a).The concentration is relatively high at middle latitudes in the northern hemisphere, -13 ppbv above SPO (90øS), and decreases sharply by -7 ppbv from GMI (13øN) to SMO (14øS).In RH(50) the simulated peak-trough amplitudes are -8 ppbv at the northern hemisphere sites and -3 ppbv in the southern hemisphere (Figure 4a).We also carried out a sensitivity experiment where the daily emission is constant at 0.25 g CH4/m2/d throughout the growing season to yield an annual total emission of 50 Tg/yr.In this experiment the total emission is larger for a long growing season (mainly for single-crop systems) than for a short growing season.The simulated distributions at the monitoring sites are practically indistinguishable from those in experiment RH(50).4.3.AnimalsMethane is produced by enteric fermentation in animals, mainly ruminants.We have applied emission rates cited byCrutzen et al. [1986] to our global data bases of animal populations to obtain the global distribution of methane emissions from animals [Lerner et al., 1988].The global emission rate was 78 Tg in 1984.Because of the economic importance of animal husbandry, this is one of the better known sources in the methane budget; the uncertainties in the magnitude and distribution of this source have been emphasized by Crutzen et al. [1986] and Lerner et al. 1988].We carried out
for different regions to produce a global map of methane emission from landfills.For comparison with other sources, a global source strength of 50 Tg/yr is used in experiment LF.The atmospheric methane distribution resulting from this source is shown in Figure 2g.Because of the high production of biodegradable carbon, surface methane concentrations are >30 ppbv above the surface mean over the large population centers in Northeastern U.SA. and Eastern Canada and > 25 ppbv over Western Europe.By contrast, the surface concentration is elevated by only -12 ppbv over China, with approximately equivalent population densities.The north-south gradient simulated at the NOAA/CMDL sites is typical of northern hemisphere mid-latitude sources (Figure 3b).For a source of 50 Tg the annually averaged concentration at middle to high latitudes in the northern hemisphere is elevated by -25 ppbv relative to SPO (90øS).The sharpest gradient, -15 ppbv, is found between KUM (20ø1'/) and SMO (14øS).With this aseasonal source the seasonal fluctuations at the NOAA/CMDL sites caused by seasonal circulation are -4 ppbv in the northern hemisphere and -1 ppbv in the southern hemisphere (Figure4b).
gross production of natural gas was 71,963 billion cubic feet and marketable production was 62,382 billion cubic feet [U.S.Department ofEnergy, 1986a].Methane content of natural gas is on average 90% and so the marketable production figures are equivalent to -1200 Tg CH4/Yr.Thus any escape of natural gas during production and consumption processes may be a significant source of atmospheric methane.Natural gas can escape during several production stages, the most significant of which is associated with venting at oil and natural gas wells.Other venues such as escape during field exploration, waste at field use, and leakage from abandoned wells and coal mines are largely undocumented and assumed here to be of little economic value and therefore negligible.We mapped the global distribution of natural gas production using production statistics [U.N.Department of International Economic and Social Affairs, 1986] and locations of major oil and gas wells [e.g.,Espenshade, 1978; Central Intelligence Agency, 1978; 1985; Seydlitz Weltatlas, 1984].As is expected, there is a high density of production sites in the Middle East, and in the Ob region of West Siberia.Some of the flaring sites were verified using a poster of nocturnal light intensities composited from satellite images from the Defense Meteorological Satellite Program [Sullivan, 1985].Barns and Edmonds [1990] have documented the history of natural gas production since 1950.The vented and flared fraction shows large fluctuations: it was as high as 12% in the early 1970s and was at its minimum (-5%) in the early 1980s.Data on amounts of natural gas flared and vented [U.S.Department of Energy, 1986a, b] show a distinct geographic pattern.In 1984, venting and flaring was only 1% of marketed production in North America, 2% in Western Europe, 2% in Eastern Europe and the USSR, 11% in Asia and Australia, 18% in Central and South America, 40% in Africa, and was as high as 61% in the Middle East.Globally, venting and flaring was 3,533 billion cubic feet and corresponded to 5.7% of the marketable production in 1984.To estimate the atmospheric methane from this source, it is necessary to determine the uncombusted fraction of methane that escapes to the atmosphere during flaring and venting at wells.Darmstadter et al. [1987] and Barns and Edmonds [1990] assumed that 20% of the flared/vented natural gas escapes directly to the atmosphere, thus yielding a methane source of -15 Tg/yr.There are no direct measurements of this fraction and hence the contribution of this term to the global budget is highly uncertain.The modeled atmospheric methane response to an assumed source strength of 50 Tg/yr (experiment GV) is shown in Figure 2h.Of note are the elevated concentrations of -20 ppbv in the large fields in the River Ob Region of West Siberia and -30 ppbv from the venting in the Middle East.
under ideal conditions.)Coal gas is -95% methane.While the amount of coal gas increases with the depth of coal deposits[Darmstadter et al., 1987], no attempt was made to distinguish among anthracite, bituminous, brown and other types of coal, as such information was not consistently available for most of the coal-producing countries.With -15 L of coal gas per kilogram of coal, there were -30 Tg CH 4 in coal produced in 1984.Barns and Edmonds [1990] estimated 25 Tg/yr.To obtain the methane emission pattern, we assumed that coal was processed in the vicinity of the mines, i.e., the methane emission pattern was the same as the mining pattern.As with the other sources, an arbitrary source strength of 50 Tg was used in the three-dimensional simulations.Figure2jshows the near-surface methane concentrations from the CL(50) source are elevated by -30 ppbv over the coal mining regions of the Eastern U.S.A. (West Virginia, Kentucky, and Pennsylvania) and Europe (Poland and West Germany).Elevated concentrations of > 20 ppbv are also found over the River Ob region in the state of Russia and over China.The north-south gradient (Figure 3b) and peak-trough amplitudes (Figure 4b) show patterns similar to the other midet al. [1988] have used the 13C/12C ratio of atmospheric methane to infer that the methane flux from biomass burning is 52 Tg/yr.Cicerone and Ore•nlw•d [1988] assumed this source to be 55 Tg/yr.The geographic information is scanty.We mapped the distribution of CO 2 release due to land use modification in the tropics [Houghton et al., 1987] and assumed that methane release due to biomass burning is proportional to CO 2 release [Crutzen et al., 1979; Crutzen andAndreae, 1990].The CO 2 release pattern is similar to that compiled by Detweiler and Hall [1988].The use of a single CH4/CO 2 proportionality constant for the whole globe is not strictly valid, as the deforestation CO 2 source is not only direct inputs from fires, but also includes the net release of CO 2 due to oxidation of soil carbon over regrowth of vegetation.Also, in South and Southeast Asia, where commercial logging rather than slash and burn agriculture may be an important mode of deforestation, this assumption may overestimate the methane source due to biomass burning.Furthermore, CH4/CO 2 ratios from biomass burning are found to depend on vegetation and fire characteristics [e.g., Corer et al., 1990].Nevertheless, the methane source mapped here is similar to that compiled independently by Hao et al. [1990] who made similar assumptions.Experiment BB assumes that the biomass burning release is during the dry season, determined as the months when precipitation is less than potential evaporation (cf.section 4.1).The annual mean methane distribution near the surface resulting from the BB(50) source is shown in Figure 2k.Concentration elevations of > 10 ppbv are simulated over Brazil, tropical Africa, and Southeast Asia.The north-south gradient at the NOAA/CMDL sites shows a maximum of 9 ppbv at CHR (2øN) in the tropics relative to SPO (90øS) and Fraser et al. [1986b], with 41% emanating 13,048 FtJNG ET AL.: GLOBAL METHANE CYCLE from savannas and 35% from tropical forests.Because the assumed habitat is widely distributed in the tropics and subtropics over an area of ---100 x 1012 m 2, the local methane flux is weak, <0.002 g CHn/m2/d for TM(50).The resulting atmospheric distribution for experiment TM(50) (Figure 2/) HV distributed 10 Tg in the coastal areas of the Soviet Arctic, from 90øE to 140øE and from 72øN to 80øN.Experiment HZ distributed 10 Tg in a zonally uniform belt extending fro m 80øN to 88øN.The annual mean concentration variations near the surface are shown in Figures 2m and 2n.ALT (82øN) is elevated -38 ppbv above SPO (90øS) in HZ(10) and by -13 ppbv in HV(10) (Figure 3d).The largest concentration is found in the Arctic: •(ALT) -&(BRW) is 16 ppbv in HZ(10) and is --2 ppbv in HV(10).Farther south, near CBA (55'N) on the south coast of Alaska, the concentrations are similar in the two cases: the elevation relative to SPO (90øS) is -8 ppbv at CBA (55øN) and --5 ppbv at KUM (20øN).The annual cycle at the high-latitude sites reflects the spring breakdown of the polar vortex: the concentration peaks in April at ALT (82øN), with a peak-trough amplitude of --24 ppbv in HZ(10) and --10 ppbv in HV(10).4.11.Chemical Destruction Methane is destroyed predominantly in the troposphere by reaction with the hydroxyl radical OH.The rate coefficient k levels also vary diurnally.We lack the direct observations of OH that would allow us to integrate the global methane loss.One approach in calculating chemical budgets has been to apply a theoretical model for the tropospheric photochemistry and to test the resulting OH fields using a well observed species with known sources, such as CHaCC1 a [e.g., Logan et al., 1981; Rasmussen and lOtalii, 1984b; Pdnn et al., •87].The global distributions of OH used here were taken from Spivakovsky et al. [1990a, b].They calculated the three-dimensional global distribution of OH every 5 days over a year as a function of sunlight and trace gas concentrations.The local independent variables needed to compute OH were taken from the parent GCM (5-day averages of pressure, temperature, water vapor, and cloud cover), and from observed climatologies for CO, O3, CH4, NOx, water vapor above 500 mbar, and the overhead ozone column.The data base is insufficient to define four-dimensional fields for these species, so the model relied on monthly mean concentrations, and used zonal average distributions with smooth variations over latitude, altitude, and season.The exceptions were CO and NOx, for which higher values were adopted in the continental boundary layer than in the marine boundary layer, and ozone, for which higher values were used over the tropical continents than over the oceansthe photochemical signal in methane is based on losses evaluated for a fixed distribution (1700 ppbv in the northern hemisphere and 1600 ppbv in the southern hemisphere).The model calculation was initialized with zero methane concentration in the atmosphere and integrated for 4 years until a steady pattern of chemically driven variations developed.The tropospheric losses are countered by a tropospherically uniform fill of methane that is equal to the annual globally integrated sink.In other words, at the end of each year's cycle the globally averaged methane abundance is zero.
a global sink strength of 450 Tg CH4/Yr, corresponding to a scaling factor of 0.85 for the OH field.Unlike many of the sources which are highly localized at the surface, methane loss occurs throughout the entire atmosphere.The destruction rate is highest in the lower tropical troposphere: OH concentrations peak where high values of water vapor and temperature coincide with the brightest intensity of sunlight.Stratospheric losses of methane are of secondary importance and contribute about 7% to the

Figure
Figure 20 shows the annually averaged near-surface methane distribution resulting from chemical destruction alone (experiment CH(-450)).The concentration in the tropics is depressed on average -30 ppbv below the global mean, with the depression slightly larger over land areas.The maximum latitudinal gradient (equator to pole), for example, CHR (2øN) to SPO (90øS), is -35 ppbv (Figure 3e).The peak-trough amplitude of the seasonal cycle at mid-latitudes is comparable to the magnitude of the latitudinal gradient (-40 ppbv).The that the difference between the mass M of methane in the atmosphere between year n + 1 and year n is the algebraic sum of the annual sources S, and sinks.The sinks included are OH destruction and soil al•sorption and are given by the destruction rate kou and absorption rate ksoit, respectively, multiplied by the mass of methane in the atmosphere.Isotopic composition of the sources and sinks and of atmospheric methane provide additional constraints on the global budget [Stevens and Engelkemeir, 1988we constructed global methane budget scenarios and tested their ability to reproduce the observed geographic observations.To reduce the degrees of freedom in the construction of the scenarios, we maintained several conditions imposed by the concentration and isotope data, some of which have been discussed by Cicerone and Oreroland [1988]: Condition 1 (C1): The current rate of increase, M n+•assumed the fossil fraction comprised emissions from coal mining, natural gas venting and leakage, and clathrate alestabilization.Condition 5 (C5): We used a global annually averaged 613C of atmospheric methane of -47.2.%0.1 ø/oo [Quay et al., 1991].This value differs by -0.5 ø/oo or larger among different investigators.Because this uncertainty encompasses the trend in •3CH4 reported by Stevens [1988] and Stevens and Englekemeir [ 1989] for this period, we assumed that R n + • • , as this is perhaps one of the better known sources in the budget[Crutzen et al., 1986].The acceptance of candidate budget scenarios was based on the ability of the scenario to match, in addition, observed geographic and temporal features in the atmosphere: condition 8 (C8): the annual mean northsouth gradient of atmospheric methane measured at the NOAA/CMDL network; and condition 9 (C9): the methane seasonal cycle at each of the observing stations.

FUNG
shows the comparison of the modeled and observed seasonal cycles at a few sites.The left panels show the modeled seasonal cycle •C(STN,t) for each source/sink with the strength specified in the budget scenario.These individual cycles are summed to form the total seasonal cycle modeled for this scenario, which is compared with the observed cycle in the right panel.The modeled seasonal cycle agrees well with the observed cycles at SPO and other middle-to high-latitude stations in the southern hemisphere: the seasonal cycle is dominated by that of chemical destruction which maximizes in summer (cf.Table 5).At MLO (20øN) the interaction between the cycles of chemical destruction and seasonal emissions from wetlands and rice paddies results in a complex annual cycle which resembles that observed: a broad peak of -8 ppbv above the annual mean from January to March, a trough of -15 ppbv in July/August as a result of the seasonality in chemical destruction, and a second peak of -14 ppbv in November dominated by biological emissions in the summer.
shows the simulated and observed annual cycle at each of the observing sites for scenario 7. The annual cycles simulated in the other budget scenarios are very similar.The composition of the modeled annual cycles (left panels) show clearly the dominance of chemical destruction in the seasonal variations of atmospheric methane in the southern hemisphere, where seasonal signals from both wetlands and rice fields are diluted.In the northern hemisphere, reasonable agreement between modeled and observed cycles was found at all the stations.Compared to scenario 1, the improved agreement with the observations at CBA (55øN) and BRW (71øN) comes from the improved description of seasonal emissions from high-latitude bogs.It is clear that while six budget scenarios presented above satisfy all the available constraints, they do not represent a comprehensive set of budget scenarios.For example, because the 813C's of methane from tropical swamps and from rice fields are similar, all the constraints could have been met equally with a shift of, say, 20 Tg from the rice source to the swamp source.In Table 6, individual source terms range by a factor of 2 among the scenarios presented.This range is comparable to that presented by Ehhah [1974], Cicerone and Orernland [1988] and others, who used mainly constraints C1-C7.It is wrong, however, to conclude that geographic and temporal variations of methane in the atmosphere (constraints C8 and C9) present no additional constraints on the global methane budget.By comparing the fluxes of methane directly measured or estimated with the average fluxes implied by each of the sources in scenarios 2-7, we were further able to eliminate scenarios 2-6 as likely global methane budgets.In the following we describe how we arrived at scenario 7 as the preferred budget scenario.

Fig. 5 .FromFig. 6 .
Fig. 5. Comparison of the observed annual mean methane concentrations at the observing sites with those simulated for global budget scenarios 1-7 (Table 6).All concentrations are defined relative to the concentration at the South Pole, •6(STN) -•6(SPO).The thick solid lines and shaded region denote +la range of the NOAA/CMDL data in the marine boundary layer.The thin solid lines connect values modeled at the NOAA/CMDL sites, leaving modeled values at CPT, JB1, and TKB off the line.
) would exaggerate the north-south gradient at the monitoring sites located mainly in the Pacific Ocean.The partitioning among the three terms is of course not unique, and is affected by the uncertainties in the landifil and animal sources as well as in the soil sink.In scenarios 2 and 5 the fossil fraction is 27%.In order to match the north-south methane gradient at the monitoring sites, 80 Tg of the fossil sources are required to come from locations remote from the monitoring sites, i.e., from the natural gas venting source.Statistics presented in section 4.5 on natural gas production give a methane source of 15 Tg/yr if 20% of the natural gas vented and flared at production sites escapes to the atmosphere as methane.Thus 75 Tg/yr is the absolute but unlikely upper bound on methane emissions from this source.

Fig. 7 .
Fig. 7. Comparison of the observed seasonal cycle of atmospheric methane at all the observing sites with those modeled using scenario 7. Refer to Figure 6 for an explanation of the scales.

Fig. 8 .
Fig. 8. Latitude-longitude distribution of the annual mean methane concentration •(A,a) --960 mbar for scenario 7. Concentrations are defined relative to the global annual mean, and dashed lines denote negative values.Stars identify measurement locations listed inTable 1. Unit is ppbv.
are far from enough field measurements to attempt a specification, on a global basis, of the magnitude and timing of the methane fluxes at each location for each source/sink.The only poss•le exception is emissions from ruminants and other domestic animals.Because methane production in these animals is considered energy wastage in commercial animal husbandry, methane fluxes from animals, at least in developed economies, are among the best known in the methane cycle.For the remainder of the surface sources and sinks we assumed values for the global annual source/sink strengths and tested how these values combine together to satisfy the constraints of mass balances, and to produce, in a three-dimensional atmospheric tracer transport model, concentration variations in the atmosphere that resemble those observed.The strengths of the sources and sinks are thus interdependent, and are determined from the combined signatures in the atmosphere.The primary sink for methane is chemical destruction, and was modeled using four-dimensional OH fields derived by Spivakovsky et al. [1990a, b], who validated the fields by a simulation of atmospheric CH3CC13.Uncertainties in the rate coefficient for the methane destruction affect conclusions about the lifetime and budget of methane.We used the 1990 J'PL Panel evaluation of rate coefficients in the calculation, but examined the global budget taking into account uncertainties in the calculated OH fields as well as recent laboratory measurements of the reaction rate coefficent for OH with CH 4 Vaghjiani and Ravishankara [1991].
a priority of measurements for defining and refining a global methane budget: 1. OH oxidation of methane in the troposphere is the largest single term in the global methane budget; it defines the total source strength.Narrowing the uncertainties in the magnitude and distn'bution of this term must be of prime importance.This would involve further refinements in the OH field through theoretical investigation, better definition of the climatologies of the atmospheric parameters and chemical species that affect OH, field experiments measuring OH concentrations under a range of atmospheric conditions, and further validation of the derived OH field using tracers such as the new synthetic HCFCs and HFCs [Prather and Spivakovsky, 1990].We prefer a destruction rate of 450 Tg/yr, yielding a chemical lifetime of 10.! yr for methane, which is within the range inferred using methyl chloroform and consistent with the new evaluation of the rate coefficient for the reaction with OH [Vaghjiani and Ravishankara, 1991].However, uncertainties in threedimensional distributions of H20 , 03, and NO x in the atmosphere might affect some of the conclusions presented here which relied on the geographic information.Similarly, our preferred budget used a single value [from Canwell et al., 1990] for the associated kinetic isotopic fractionation ratio (kls/k•2)OH.Confirmation of this value and its lack of temperature dependence (especially at temperatures <273K) would give confidence to the analysis which followed from these values.2. The 14C content of atmospheric methane defines the methane input associated with our energy demands and hence a large portion of the north-south gradient of atmospheric methane.While the range of estimates of inc-free methane has narrowed in recent years, uncertainties in the fossil fraction of the sources remain large.Improvements in the accuracy of 14C measurements and a characterization of the spatial variability and global representativeness of the 14CH4 data are necessary for an improved budget.3.There must be an expanded measurement program to understand the controls and causes of variability of the methane fluxes found in the tropics.This is particularly important for emissions from rice fields and from biomass burning, two terms likely to play an important role in the increasing abundance of methane in the atmosphere.this way, regional fluxes can be constrained in a manner similar to that presented here.Regions where measurements of boundary layer methane concentrations are sorely needed include: the emergent and established agricultural regions of South and Southeast Asia, the mid-latitude population centers, and the hypothesized region of clathrate destabilization in the Soviet Arctic.Also, we should begin to investigate what information about the competition between increasing sources and decreasing sinks can be extracted from the temporal changes of the vertical profile.An update of theEhhalt and Heidt [1973] profiles and continuation of the South and Southeast Australian profiles are paramount.The increasing trend in atmospheric methane was not the focus of study here.Increased emissions associated with food and energy demands of a growing population, and decreased cleansing capacity of OH in the troposphere are the prime candidates in causing the trend.Feedbacks between climate change and natural systems are also likely to impact the methane trend [e.g.,Khalil and Rasmussen, 1989], as is evident in the glacial-interglacial fluctuations of methane deduced from ice core records[Raynaud et al., 1988;Stauffer et al., 1988;  Chappellaz et al., 1990].In the coming decades, high-latitude warming is hypothesized to enhance clathrate destab'dization as well as wetland and tundra emissions.At the same time,