A STUDY OF THE SOURCES AND SINKS OF METHANE AND METHYL CHLOROFORM USING A GLOBAL 3-DIMENSIONAL LAGRANGIAN TROPOSPHERIC TRACER TRANSPORT MODEL

Sources and sinks of methane and methyl chloroform are investigated using a global three- dimensional Lagrangian tropospheric tracer transport model with paranteterized hydroxyl and temperature fields. By comparison with methyl cldoroform observations a global average tropo- spheric hydroxyl radical concentration of 6.4 x 10 s cm -3 was found to be consistent with published methyl chloroform emission data for the year 1980. Published methyl chloroform emissions data for 1981-1984 were found to be inconsistent with the observed methyl chloroform concentration increases. A large decrease in hydroxyl radical concentrations could explain the disagreement be- tween the emission data and atmospheric methyl chloroform concentrations, but this is unlikely. Using the hydroxyl radical field calibrated to the methyl chloroform observations, the globally av- eraged release of methane and its spatial and teinporal distribution were investigated. Two source function models of the spatial and temporal distribution of the flux of methane to the atmosphere were developed. The first model was based on the assumption that (cid:127)nethane is enfitted as a proportion of net primary productivity (NPP). With the average hydroxyl radical concentration fixed, the methane source term was computed as -(cid:127) 623 Tg 0}f4, giving an atmospheric lifetime for methane ,(cid:127)8.3 years. The second model identified source regions for methane from rice paddies, wetlands, enteric fermentation, termites, and biomass burning based on high-resolution land use data. This methane source distribution resulted in an estimate of the global total methane source of ,(cid:127)611 Tg OH4, giving an atmospheric lifetime for methane -(cid:127)8.5 years. The most significant difference between the two models were predictions of methane fluxes over China and South East Asia, the location of most of the world's rice paddies, indicating that either the assumpt.ion that a mziform fraction of NPP is converted to methane is not valid for rice paddies, or titat NPP is underestimated for rice paddies, or that present methane emission estimates from rice paddies are too high. Using a recent measurement of t]te reaction rate of hydrox. yl radical and methane by Vaghjiani and Ftavishankara (G. L. Vaghjiani and A. Ft. 1Xavishankara, l]ate coefficient for the reaction of Ott and CIt4: hnplications to the atmospheric lifetime and budget of methane, submitted to Nature, 1990) (hereinafter referred to as Vaghjiani and Ftavishankara, 1990) leads to estimates of the global total methane source for SF1 of -(cid:127)524 Tg OH4 giving an atmospheric life- time of ,(cid:127)10.0 years and for SF2 ~514 Tg OH4 yielding a lifetime of ,(cid:127)10.2 years. These results are provisional pending any revision of the reaction rate for hydroxyl radical and methyl chloroform. The availability of good estimates of observed global wind trace species on the sphere, particularly when applying a fields at a fine resolution (2.5øx 2.5 ø) at a number of finite difference approach to a model transport. As we pressure levels was preferred to using General Circulation approach the poles, the constant longitudinal grid size in Model (GCM) wind fields. However, GCM wind fields degrees produces a dramatically shrinking cartesian dis-could readily be substituted if desired. The mean vet-rance which requires that the model time step be reduced tical motion is included within the European Centre for in order to ensure numerical stability (the Courant con-Medium Range Weather Forecasting (ECMWF) data set. dition), and in Eulerian and semi-Lagrangian finite dif-A component of the rapid vertical transport associated ference tracer transport, to ensure conservation of tracer with convective cloud transport is embodied included in mass. Alternative approaches, such as the spectral trans-this data set. Mixing processes are also modeled through port approach, have also been found to be seriously de-inclusion of a random component in the computation of ticlent in advecting concentration fields of such quantities each air parcel velocity. only layer values impleInentation of diffusive mixing provides consistent are shown. results where the number of air parcels ranges frmn n = 10,000 to n = 1,000,000 on the basis of model simulations with an F-11 like tracer. Calibration of such a.n approach to diffusive mixing, other than to the spatial distribution of tracer concentration [Taylor, 1989], may be possible by comparing observed and predicted autocovariances of first as a result of interference proble(cid:127)ns and changes where CN is the average northern hemisphere coneentra- in calibration, and they may have been made in polluted, tion, Cs is the average southern hemisphere concentra- dominate the releases of methane in the northern lati-model surface layer (75 hPa) methane concentration. The tudes. Concentrations rise in high northern latitudes in methane concentration rises in the northern hemisphere summer due to the large release of methane even though summer and falls in the northern henrisphere winter. This loss to reaction with the hydroxyl radical is at a peak. model predictions observations disagreement with observations than in atmospheric concentration than of concentrations at northern hemisphere differ-emissions. If these data also require revision upward than ent latitudinal gradients. It is this difference in the a concomitant increase in hydroxyl radical concentrations latitudinal distribution of methane sources in combina-and the total methane emissions ;viii be required. After tion with latitudinal variations in hydroxyl radical con-completion of this study we beca(cid:127)ne aware of the upward centration which produces the difference in the estimates revision of estimated emissions of methyl chloroform of of the total methane source required to explain the ob-Midgley (1990).


INTRODUCTION
Methane is a key chemical species in the chemistry of both the troposphere and the stratosphere. Recent analysis of air in dated ice cores provides very clear evidence that methane concentrations have been rising rapidly since the industrial revolution [Craig and Chou, The •nodel takes advantage of modern computer ar-truncation, and dispersion errors associated with spectral chitectures, particularly array processing. For example, transport [Williamson and Rasch, 1989]. Williamson and model simulations with methyl chloroform and methane Rasch [1989] also note that computational fixes are reof 1 year in duration with 100,000 air parcels on the quired to remove physically unrealistic negative water va-CRAY X-MP/48 at the National Center for Atmospheric pot concentrations, implying a failure to conserve tracer Research require only ~ 140-180 s of central processing mass, from spectral transport models. In the Lagrangian time for completion. The location of air parcels at each formulation, as described above, the computation of tracer time step may also be stored if so desired to avoid the re-concentration proceeds independently of the calculation of calculation of the transport component during subsequent the transport of the air parcels thereby ensuring the conmodel runs. This approach will be used in future stud-servation of tracer mass. As the poles are approached ies involving more complete atmospheric chemistry •nod-lines of longitude become closer, longitude loses meanels which will require short •nodel time steps in order ing, and the Lagrangian trajectories will tend toward reto properly model the spatial and temporal concentration producing the zonal mean displacements rather than the distribution of short-lived chemical species.
individual longitudinal grid displacements.

The wind velocities in (2) were obtained by nearest
The location L, of particle p, in a grid cell located at neighbor interpolation. The nearest neighbor interpolation latitude i, longitude j, level l, and at time t is evaluated appears initially to be an unsophisticated form of inter- in the next Lagrangian air parcel to arrive at the grid Equation (3) illustrates a common problem associated square. The residual flux was found to remain, at each with the mathematical representation of the transport of time step, less than 1% of the total flux. Mixing ratios were derived by translgting the at each time step. These data were plotted as a function Lagrangian parcel coordinates to Eulerian grid coordi-of model time step. The results of a least squares fit nates. The Eulerian grid was based on the wind field yielded a. correlation coefficient of I and a. slope equal grid. The divisions in the vertical were centered about the to the specified emission of tracer. Figure  results where the number of air parcels ranges frmn n = 10,000 to n = 1,000,000 on the basis of model simulations with an F-11 like tracer. Calibration of such a.n approach to diffusive mixing, other than to the spatial distribution of tracer concentration [Taylor, 1989] and the analysis procedures used to generate the data are • described in detail by Loteric [1981]. Rather than use the • 6 :> ECMWF data directly within the model we reduced the •' m 60 data to a set of coefficients. In this way the cmnputer to •nodel did not spend the majority of its execution time • 58 reading the wind field data. On theoretical grounds • 56 the components of the wind field should be normally distributed [Justus, 1978]. To take into account the non-54 stationarity in the wind fields due to seasonality, the year was divided into six-bimonthly intervals for which the parameters of the normal distributional model were estilnated. It was considered that at least problems of severe nonstationarity could thus be avoided. An approximately sixtyfold reduction in wind field parameters required by the model could be achieved through this data reduction scheme. The primary disadvantage of this approach is that only the mean spatial and temporal to the southern hemispheres. We obtained a mean value Summer midday averages of northern mid-latitude meaof rex of 1.2 years based on the 48-monthly estimates surements, obtained at the Earth's surface, of the hyof the last 4 years of the model run. We discarded the droxyl radical concentration produced estimates in the first 12 months of the model run to remove the effects of range 7-87 x 105 molecules cm -3 [Platt et al., 1988; the model initialization on the estimates of rex. We also Perner et al., 1987]. These values are in agreement •vith observed a seasonal variation in rex with the minionurn the average hydroxyl radical concentrations reported in values of rex occurring twice yearly during the northern Table 1. The measured midday hydroxyl radical conand southern hemisphere summers and maximum values centrations represent maximum possible hydr9xyl radical occuring during spring and autumn. This observation is concentrations which the model predictions of the averconsistent with the movement of the intertropical convet-age hydroxyl radical concentration do not exceed. Model gence zone from one henrisphere to the other. The value predictions of the two-dimensional hydroxyl radical field of rex -  [1987]. Note that hydroxyl radical field. Rather than compute the full threethis value is obtained without introducing an additional dimensional hydroxyl field, which would require a subdiffusion term associated with convective activity in the stantial chemical mechanism •vith concomitant increase in tropics, as was done for example in the model of Prather computational requirements, a zona.lly averaged hydroxyl et al. [1987]. The intensity of interhemispheric exchanges radical concentration field was computed using the twoseexns therefore to be adequately represented despite the dimensional model of Brasscur et al. [1990]. In order to fact that convection is not explicitly simulated. This may reproduce the seasonal variatior in hydroxyl radical conbe due, in part, to the fact that a substantial fraction of centration, monthly averaged hydroxyl radical fields were the rapid vertical transport is included in the ECMWF calculated. Figure 4 illustrates the resulting hydroxyl taddata set.
ical concentrations (molecules cm -3) for the months Jan-At the end of the 5-year model run we also computed uary and Jul3'. These hydroxyl fields are in good qualirathe number density of air parcels as a function of the 2.50 tire agreement with those calculated by Crutzen and Gidel latitudinal bands. The maximum variation in the number [1983], Vole et al. [1981], and Allam et al. [1981]. Idedensity of air parcels occurs with respect to latitude. ally, a full three-dimensional hydroxyl radical field would Figure 3 shows the theoretical mean number of air parcels, be preferred; however, xneridional rather than zonal variawhich is just the volume of the latitudinal band divided tions are likely to dominate in the hydroxyl radical field. by the volume of an air parcel, plotted against the Hydroxyl radical concentrations in Figure 4 vary from number of air parcels actually present within the model near zero in the polar night to values greater than 106 2.50 latitudinal bands at the final model time step of molecules cm -3 at midtroposphere over the equator and the 5-year model run. Some statistical variation about near the surface at midnorthern latitudes. Until meathe theoretical mean number density values is anticipated surements of hydroxyl radical concentrations are obtained due to the Monte Carlo component of air parcel velocity routinely and they can be considered to be representative computations. Relative to the mean this variation will of a large volume of the troposphere, quantitative cornincrease as we approach the poles. Figure 3 shows, even parison of model predictions and observations will not be after a 5-year model integration, that the number density useful. Unfortunately, such measurements are not likely of the air parcels is very close that expected.
to become available in the near future.
The global mean hydroxyl radical concentration is TROPOSPHERIC HYDROXYL RADICAL FIELD important in determining the lifetime of many trace In view of its central role in atmospheric chemistry, species, but the actual spatial distribution of hydroxyl estimating the global tropospheric average hydroxyl tad-radical in relation to the sources of the trace gases is ical concentration has been the subject of much study. also critical. For those trace gases that react with OH, A number of published esti•nates of the hydroxyl radical lifetimes will be shortest if releases coincide with the concentration are listed in Table 1. While some estimates highest hydroxyl radical concentrations, such as over the of the average hydroxyl radical concentration indicated tropics, whereas the opposite is true if trace gas sources that a value in the range of 10-100 x 105 molecules are located toward the polar regions where hydroxyl cIn -3 was likely [Weinstock and Chan•l, 1974; Neel•t and radical concentrations are near zero in winter. Plonka, 1978], more recent results based on the analysis were an order of magnitude higher than more described below. Table 2 shows that hydroxyl radical con-

METHYL CHLOROFORM
It has been recognized for some time that methyl chloroform (CH3CC13) could be used to estimate the average hydroxyl radical concentration [Lovelock, 1977;Singh, 1977]. Our objective behind modeling methyl chloroform was to deduce the average hydroxyl radical concentration and thus estimate the loss of methane due to reaction with hydroxyl. Methyl chloroform is a solvent which has been increasingly used by industry as a substitute for the more toxic trichloroethylene. Since 1970 annual production and release of methyl chloroform to the atmosphere have more than tripled [Neely and Plonka, -1978;Prinn et al., 1987]. As a consequence, concentrations have been observed to be rising rapidly (,,0 6% per annum) within the troposphere [Prinn et al., 1987]. This increase in concentration has been the focus of concern -in two areas. Once reaching the stratosphere methyl chloroform provides a source of atomic chlorine through photodissociation. Atomic chlorine eventually leads to the the production of "reactive" chlorine (C10•) which is thought to be the trace species prima. rily responsible for the reduction in the ozone layer [e.g., Anderson et al., 1989]. Methyl chloroform is also a greenhouse gas [Ramanathan et al., 1987].

METHYL CHLOROFORM SOURCE DISTRIBUTION
Methyl chloroform is considered to be entirely anthropogenic in origin. Detailed information concerning the spatial distribution of releases of methyl chloroform to the atmosphere is lacking. In this study we employ the emissions grid originally developed by Prather et al. [1987] for the chlorofluorocarbons (CFC), F-11 and F-12. Prather et al. [1987] based their exnissions grid on electric power consumption because electricity consumption and CFC production are both products generally associated with technologically advanced countries. It was also necessary to divide the world into three economic groupings which related the pattern of CFC use to electricity production. Weighting factors were computed covering the period 1970-1975 and 1976-1982     This proble•n has been investigated in two ways. First, the average hydroxyl radical concentration required to the change in tropospheric hydroxyl radical concentration oping countries. Calibration of the average hydroxyl conobtained by Thompson and Cicerone [1986] and Isaksen centration to the 1985 methyl chloroform emissions data and Hov [1987]. Thompson and Cicerone [1986]  mosphere, for the month of January 1984. Source regions A second approach to resolving this inconsistency was over Europe and North America, South Africa, Japa.n, to compute the source term required to explain the oh-and Australia are clearly indicated. Southern Hemisphere served growth in methyl chloroform concentration based oceans are predicted to have a relatively uniform methyl on a fixed hydroxyl concentration. Table 4

Figure 7 sho•vs the model predictions of methyl chloroform concentration along with the methyl chloroform observations obtained at four Atmospheric Lifetime Experiment/Global Atmospheric Gases Experiment network sites namely Adrigole (Ireland), Cape Meares (Oregon), Ragged Point (Barbados), and Cape Grim (Australia) for the period 1980-1984 as reported by Prinn et at. [1987]. It
should be noted that the methyl chloroform concentration measurements reported by Prinn et al. [1983aPrinn et al. [ , 1987 have been subject to data selection procedures aimed at determining long-term trends. Accordingly, Prinn et al. [1983aPrinn et al. [ , 1987 have sought to remove high-frequency variation in the methyl chloroform record obtained at these monitoring sites. This high-frequency variation is attributed to "local" pollution. This makes the direct cozn. parison of any model predictions with observations problematic as model predictions include all model events whereas reported measurements do not. Measurements are also collected at coastal sites with the objective of sa•npling rine air. Measurements are •nade at particular times of the day and under particular meteorological conditions. Measurements may also be missing as a result of instru-•nental or other problems. These factors bias reported concentration measurements . However, it is clear that this data selection procedure is warranted when the determination of long-term trends is the goal of the measurement program. Also, data selection is required to remove the effects of local air pollution emanating from sources within a few kilometers of the measurement station. measurements exceed 300 ppt before data selection. Re- Figure 9 shows the zonally and yearly averaged latgardless of these problems, the model correctly predicts itudinal gradient predicted by the model averaged over the long-term trend and the key features of the seasonal lowest 75 hPa of the troposphere for 1984. Peaks in concycle at Adrigole, through considerably amplified in corn-centration occur at midnorthern latitudes and to a much parison to the observations. lesser extent at 30 o south corresponding to the releases of At Cape Meares, Oregon, observed concentrations are methyl chloroform. A strong latitudinal gradient is preoffset about 10 ppt above the model predictions and share dicted in the northern hemisphere with far less variation a similar mean concentration as that at Adrigole, Ireland. in concentration in the southern hemisphere as would be Model predictions also show less variability than the expected from the distribution of sources of methyl chlo-     . While we shall be investigating the total ganic material that is converted to methane may be some amount of methane released to the atmosphere consistent years old. with the known sinks, which will require adjustment of NPP h• been modeled by relating measured NPP to the total amount released, the individual sources will be readily available environmental data such as temperature adjusted in proportion to the change in the total methane and precipitation, evapotranspiration, and the length of emission. In view of the large uncertainty associated with the growing season [Lieth, 1973[Lieth, , 1975. Lieth [1975] h• all the individual source estimates, such an approach is reviewed the above approaches to modeling NPP and reasonable. Table 11   The Mia. mi model is based upon two simple e•npirical relationships derived from NPP, precipitation, and temperature data collected at 52 sites representing the world's key ecosystems. The elnpiricM formula derived by Lieth [1973,1975] relating NPP (g dry matter/m2/yr) to mean annual tctnperature (øC), T, is stated as

The methyl chloroform measurements at
;vhile the formula relating annual precipitation (millimeters), P, is NPPij = 3000(1 -½-0.000664P) (12) As two estimates of NPP are computed for each grid point, one value is selected on the basis of the assumption that one factor, either precipitation or te•nperature, limits NPP. Accordingly, the minimmn value of NPP, derived fom (xx) (X2) bov½, is doptd the best estimate of NPP.
Equations (11) •nd (X2) •bove, as originally developed by Lictb [1973,1975] used annual average va,lues for temperature and precipitation. However, (11) and (12) are very close to linear over most of the range of expected values for temperature and precipitation. By using monthly mean values for temperature and precipitation the monthto-month variation in NPP ca.n be computed. NPP estimates derived using (11) must also be modified to account for the change from annual to monthly averaged data by dividing estimates by 12. This model of NPP implies that during months of lowest rainfall and temperatures, low NPP values will result. During the warm months with high rainfall the largest values of NPP are predicted.
Both precipitation and surface telnperature data are available as monthly means on a 2.5øx 2.50 grid. Oort [1983] describes the temperature fields. Shed [1986] has prepared global maps of precipitation based on a compilation of data from a number of sources covering the period 1950-1979. Shed [1986] notes that precipitation data for the southern oceans are particularly unreliable.
Fortunately, we only wish to compute NPP using (11) and (12) over the land areas for which the data are most reliable. Figure 12 shows the distribution of NPP computed for the •nonths January and July.
For January the highest values of NPP occur in the southern hemisphere with NPP falling to zero for large areas of the northern hemisphere. In July the situation is reversed with the high values for NPP occurring in the northern hemisphere with small values for NPP accompanying southern hemisphere winter conditions.  In each case the data sets were coinpiled using different methods. This variation in methodology is reflected in the wide range and number of vegetation types used to classify land cover and the differences in resolution. In no case have these vegetation data sets been prepared with the aim of estimating the flux of methane to the atmosphere.
One sontee of systematic error associated with these land cover data sets arises from the a.pproad, of reporting only the major ecosystem present with a grid cell (•  129.7 temperature and precipitation this study 50% land cover). This leads to a systematic underesti-Biomass burning has been recognized for some time to mation of land cover for an ecosystem that is intermixed be a source of atmospheric methane [Crutzen et al., 1979].    should be reduced by a factor of 4-10.

Henderson-Sellers [1985] data base. This implies that the flux of methane may be underesti•nated from temperate regions by a few percent. Table 11 lists the land cover codes of IVilson and Henderson-Sellers [1985] and their
where Ai is the land area (meters squared) associated with land cover type, l, within a grid cell at longitude, i, and latitude, j, and ek represents the incthane e•nission rate corresponding to the land cover code, l, as listed [1982].
in Table 11. Figure 15 shows the spatial distribution of the annual total flux of methane to the atmosphere derived from termites. Figure 14 Figure 14 shows the latitudinal distribution of the A nu•nber of data bases are available which include the •nethane flux from wetland areas derived from the THYextent of natural wetlands. Both ;Viison and Henderson-DRO wetland area data set of Cogley [1985]. These data

Sellers [1985] and Olson et al. [1983] include a code for of Coghy [1985] lead to a latitudinal distribution of the natural wetlands in their respective data bases. Both flux of methane frown wetlands with a slightly greater emof these data bases only provide estimates of the extent phasis on tropical regions than that obtained by Matthews of natural ;vetlands when natural wetlands are either and Fu,•g [1987]
. Figure 17 shows the geographical distrithe primary or secondary land cover within each grid bution of the annual total methane emissions from wetcell. Unfortunately, as wetlands are widely scattered this lands. The distribution of methane emissions from wetapproach to estimating the extent of naturM wetlands lands presented in Figure 17 is qualitatively similar to leads to a systematic bias toward underestimation. that obtained by Matthews and Fung [1987].  ION ................................................... ' ............................................. '•}}•. ............ .•..'.: ....................... ' .......... •':a¾•}• ........... Methane is produced as a byproduct of digestion in Fig. 18. The spatial distribution of the flux of methane animals as a consequence of the microbial activity of (108g CH4 per 2.5øx 2.50 grid square)to the atmobacteria present within the animal gut [Hungate, 1966]. sphere from enteric fermentation. In the study reported here the distribution of the flux of associated with oil and gas exploration and recovery, parinethane from animals is based on a simple model. Using ticularly offshore activities [Cicerone and Oreroland, 1988]. the land cover codes of Wilson and Henderson-Sellers tlowever, the total methane released from these latter [1985], land cover categories likely to support agricultural tivities is currently thought to be small at around 14 x activity were identified. The land cover categories, a 0•2 description of ;vhich appears in the ;york by Wilson and 1 g CH4 yr-or less, and methane emission estimates from individual exploration and production facilities are Henderson- Sellers, were codes 13, 16, 21, 23, 24, 26,   27, 31, 32, 33, 34, 35, 36, 37, 39, 40, and 41. The lacking, making it difficult to develop a source function for the spatial distribution of methane from coal raintotal annual methane exnission from enteric fe•'mentation was then distributed uniformly with respect to the total ing and gas drilling and venting. As better information ' becomes available on this source of methane it shall be land area. Figure 18 shows the global distribution of the annual total flux of methane arising from enteric incorporated into the model. fermentation. Figure 14 includes the annual average It should be noted that in regard to the fraction of the latitudinal distribution of methane emissions from enteric total release of methane attributed to fossil fuels, analysis of •4C isotope measurements of methane can provide an fermentation. Comparing Figure 14 Figure 14. However, given the of atmospheric methane. They found that methane from uncertainty associated in evaluating methane emissions fossil fuels could be as much as 50% higher than is inover different geographical •'egions as noted by Crutzen et dicated in Table 5 pleted methane from old organic matter may explain the unexpectedly low atmospheric 14CH4 measurements.
The seasonal variation of methane emissions from en-Methane may be released from landfills in amounts teric fermentation was modeled assuming tha. t the emis-that are globally significant but existing estimates are uncertain [Bingemet and Crutzen, 1987]. Cicerone and become available. Fortunately, such a large uncertainty is Oreroland [1988] noted that further work with regard to associated with only a small source of methane.
landfill emissions of methane is required in the areas of The exchange of methane between the ocean and atmothe amounts of waste material and trends, composition sphere is modeled as a uniform release of methane from of waste material, landfill management practices, and the the oceans. This simple model may not be appropriate for ilnpact of methane oxidation and burning. Little data the continental shelf regions where the flux of methane to are available regarding the spatial distribution of the the at•nosphere may be considerably higher per unit area flux of methane from landfills. Landfills are associated than over the open ocean [Ehhalt, 1974]. However, given with populated areas of industrial countries. Bingemet the small magnitude of the source and the lack of reliable and C7'utzen [1987] have estimated that the majority of measurement data distinguishing between the release of methane released from landfills (78%) is from industriM methane from the open oceans and the continental shelf countries.
regions, a more complex model for the release of methane Accordingly, to model the spatial distribution of fos-from the oceans does not seem warranted. sil fuel and land fill releases of methane to the atmo-No seasonality in the release of methane to the atmosphere, their combined emissions have been assigned a sphere from the oceans has been included in the model. single spatial distribution. Fossil fuel emissions of CO2 Instead, a constant release rate over time has been ashave been comprehensively studied [Marland et al., 1985; stoned. Again, lack of sufficient data and the small mag- of concern as a potentially large source of atmospheric No seasonal cycle for these releases was incorporated meth'ane [Kvenvolden, 1988 Ehhalt [1974]. They found that the origi-All simulations with methane used the spatial and ternhal data employed by Ehhalt [1974] was limited and was poral distributions of the hydroxyl radical conce•trations obtained when 15-20% less metha. he was present in the derived from the model of Brasscur ½t al. [1990]. The atmosphere compared with present-day methane concen-mean hydroxyl concentration, 6.37 x 105 molecules cm -3, trations. Cicerone and Orerelated [1988] suggest that the was that obtained from the methyl chloroform modeling flux of methane from the ocean to the atmosphere has studies, as reported earlier. The hydroxyl radical concenreduced greatly, or even reversed in sign, and recommend tration specifies the most important loss term for atmothat extensive sampling of marine surface waters be un-spheric methane. Using this specified loss term for atdertaken.
•nospheric methane along with a fixed loss of methane to We adopt the value reported by Cicerone and Oreroland the stratosphere, the total source of methane required to [1988] for the flux of methane from the oceans to the reproduce the observed atmospheric amounts and growth atmosphere. However, we caution that this estimate may rate in atmospheric concentration can be derived. The torequire substantial revision if new measurements of the ta.1 source of methane can be computed for any specified flux of methane from the oceans to the atmosphere should methane spatial and temporal source distributions. Brasseur et al. [1990] for in the available observational data sets. the month of January. The three-dimensional Lagrangian Figure 20 shows the seasonal cycles averaged over tracer transport model with methane sources and sinks the model surface layer (75 hPa), for the northern and was then integrated for 1 year. A second year of model southern hemispheres and over oceanic regions of the integration generated the model results reported here. respective hemispheres. The hemispheric averages include

Model initialization was achieved by specifying a latitu-the source regions which is included in the model estidinal and vertical distribution of methane based upon the mates of the seasonal cycle but which is not represented two-dimensional model results of
The two methane source functions used in this study both the source regions and the tropics, whereas the are summarized in Table 11. For source function 1 oceanic averages are computed using the model grid (SF1) the NPP-based approach, a total annual release of squares over the Pacific ocean at mid-latitudes. These methane to the atmosphere of ~623 x 1012 g CH4 was plots of the predicted methane concentration over the required to explain the observed growth in atmospheric oceans show the importance of the sink of atmospheric methane concentration. Whereas, for source function 2 methane and the seasonality of that sink.
(SF2) a total annual release of methane to the atmosphere Figure 21 shows a contour plot of the predicted methane of ~ 611 x 1012 g CH4 xvas needed to reproduce the concentrations over the model surface layer (75 hPa) for observed growth in atmospheric methane concentration. both January and July. As SF1 is largely based on a These two estimates of the total source of methane model of NPP a strong seasonality is evident in model are in excellent agreement. Hmvever, the predicted predictions. This seasonality is particularly evident in the spatial and seasonal variation in the atmospheric methane northern hemisphere. In January, only methane from fosconcentrations differ. We shall consider each source sil fuel is being released in high northern latitudes. In function in turn.
July the biospheric sources of methane are predicted to    I  I  I  I  I  _1  I  I   I  I  I  I  I  I  I  I dicted concentrations over the model surface layer (75 hPa) for both January and July for SF2. The major differences between SF1 and SF2 are apparent in these plots. A lower seasonal variation is predicted for SF2 in the high northern latitudes. Hoxvever, the most striking difference lies in the methane concentrations eredieted over Southeast Asia. Here SF2, which includes a separate emission function for rice paddies, predicts much higher methane concentrations than SF1. Clearly, the assumption that a uniform fraction of NPP is converted to methane is not valid •vhere rice paddies occur or NPP is grossly underestimated in these areas using the NPP model based on the Miazni model of Lieth [1973,1975] or the assumed methane releases from rice paddies (Table 5) are too large. Figure 22 also illustrates the two-dimensional, zonally averaged, contour plot of the model predicted methane concentrations for SF2. Figure 22 shows that the model predicted vertical gradient is reversed over the southern hemisphere, where methane concentrations decrease with altitude, when compared with the northern hemisphere gradient during the month of January. During the month of July the vertical profile of methane concentration over the southern henrisphere is predicted to be more nearly constant. A similar result has been obtained for methyl chloroform as illustrated in Figure 10 and  four northern hemisphere and four southern hemisphere NOAA/GMCC sites. In general, similar agreement be-of methyl chloroform coincide with the highest hydroxyl tween model predictions and observations was obtained radical concentrations then a much lower average hydroxyl using SF2 and SF1. At the northern latitude sites SF2 concentration will be required to produce the same growth precludes a larger disagreement with observations than in atmospheric concentration than if methyl chloroform SF1. Again, overprediction of methane concentrations at emissions occurred where hydroxyl radical concentrations northern hemisphere sites may have occurred. are at their lowest. Accordingly, as the hydroxyl radi-

DISCUSSION
The magnitude of the methane sources deterxnined in this study are sensitive to the assumed sinks of atxnospheric methane. The total mass of methane released to the atmosphere has been determined by balancing the cal concentration varies over several orders of magnitude with respect to latitude, altitude, and season, accurate calculation of the average hydroxyl radical concentration requires at least two-dimensional atmospheric tracer transport models which take into account these variations in methyl chloroform emissions and hydroxyl radical concentration.
loss of methane to reaction with the hydroxyl radical and The assumption that a constant fraction of NPP is the relatively small loss to the stratosphere so that the released as methane to the atmosphere appears, from a observed growth in tropospheric methane concentration is comparison of model results with available atmospheric reproduced. Should other sinks of tropospheric methane metha. he measurmnents, to provide a reasonable source function for describing the spatial and temporal distribe demonstrated to be significant (for example consumption by soils), then the total of all sources of atmospheric bution of •nethane. Only at high northern latitudes are methane would need to be revised upward. Titis poten-the model predictions of the seasonal cycle of atmospheric tial systematic error confounds assessment of the error methane concentration not in agreement with the observed associated with the total sources of atmospheric methane. methane concentrations. However, this result is also true Itowever, it can be assumed that the total source esti-for SF2. mates reported here are more likely to represent a lower This disagreement between model predictions and obbound than an upper bound, if the average hydroxyl tad-servations indicates that either the previously predicted ical concentration remains valid.
sources of •nethane are too large at high northern lati- The magnitude of the sources of atmospheric methane tudes during summer for SF1 and to a greater extent for is also closely tied to the average hydroxyl radical concen-SF2 or that the observations of methane are not representration. We have estimated this quantity by using the re-tatlye of the actual seasonal cycle of methane because of ported methyl chloroform emissions data and atmospheric the data selection required to estimate long-term trends. concentration data. However, our analysis indicated that Measure•nents directed at determining the true seasonal a single constant value for the hydroxyl radical concentra-cycle of atmospheric methane at high northern latitudes tion could not be found that ;vould lead to a predicted will be required to resolve this problem. If the methane tropospheric concentration consistent with the estimated source functions must be modified to reduce the high sources and concentration measure•nents. This result ira-northern latitude sources with a concomitant increase in plies either of two possibilities (1) that a larger drop in the release of methane in low latitudes, then the total hydroxyl radical concentrations has occurred than model methane source will need to be increased. This increase studies would predict [Thompson and Cicerone, 1986; Isak-in the total methane release to the atmosphere is required sen and Hov, 1987]; or more likely (2) that the recent to account for the increased loss of methane due to the methyl chloroform emissions data •nay need revision up-higher hydroxyl radical concentrations at low latitudes so ward. The need to revise methyl chloroform e•nissions that the observed increase in atmospheric methane conupward is based upon a constant hydroxyl radical concert-centration is correctly predicted. tration calibrated to earlier estimates of methyl chloroform The two source functions also produce slightly differemissions. If these data also require revision upward than ent latitudinal gradients. It is this difference in the a concomitant increase in hydroxyl radical concentrations latitudinal distribution of methane sources in combinaand the total methane emissions ;viii be required. After tion with latitudinal variations in hydroxyl radical concompletion of this study we beca•ne aware of the upward centration which produces the difference in the estimates revision of estimated emissions of methyl chloroform of of the total methane source required to explain the ob-Midgley (1990). served growth in methane concentration. While SF1 more The importance of methyl chloroform emissions data closely matches the latitudinal variation in annual averand tropospheric concentration measurements for an in-age methane concentrations observed at NOAA/GMCC dependent determination of the average hydroxyl radical sites, true zonally averaged concentration estimates would concentration warrants a greater effort to evaluate the greatly assist in determining the correct latitudinal distritotal emissions of methyl chloroform and its spatial and bution of methane sources and sinks. temporal distribution more precisely. Clearly, the charac-A comparison of the methane concentration estimates terization of the emissions of other trace gases which are obtained from the two atmospheric methane source funcentirely anthropogenic in origin, and for which the major tions shows that the models produce very similar resink is reaction with hydroxyl radicals in the troposphere, suits except where rice paddies are concentrated. Here would provide independent estimates of the average by-SF2 predicts much higher atmospheric methane concendroxyl radical concentration.
trations. Measurements of atmospheric methane concen-The average hydroxyl radical concentration that is de-trations near or above the rice paddies of Asia will be duced using methyl chloroform data will also be sensitive required to determine which source function provides a to the spatial and temporal distribution of the sources realistic representation of the release of methane from rice of methyl chloroform. If the majority of the emissions paddies.
Unfortunately, only a limited number of tracer concentration measurements are presently available to validate tracer transport model predictions. Available measurements of atmospheric trace gas concentrations currently provide important information regarding the long term trends. However, the practice of reporting selected data makes the direct comparison of such observations with the output of atmospheric tracer transport models difficult. Monitoring sites are also located away from source regions making validation of model predicted concentrations over such regions very difficult. In order to more fully validate atmospheric tracer transport models a measurement program aimed at obtaining latitudinal and zonal profiles, both over the continents and oceans, is required. Measurements could consist of either point measurements at different altitudes or, tropospheric column averages, or vertical profiles. Measurement techniques which produce area or column averages rather than point values would be preferred, as transport model predictions represent large volume averages.
Finally, a recent measurement of the reaction rate of hydroxyl radical and methane by Vaghjiani and Ravishankara [1990] indicates that the rate is lower than previously measured. Using the rate coefficient of Ravishankara in the three-dimensional model, we calculated that the sources listed in Table 11 should be multiplied by a factor of 0.83 if the observed growth in atmospheric methane concentrations is to be reproduced. This would lead to an estimated atmospheric residence time of 10.0 and 10.2 years for methane for the respective source funct. ions. The major difference in predicted methane concentrations will be a reduction in the amplitude of seasonal cycle of methane concentrations. This leads to improved model predictions of observations collected at most NOAA/GMCC monitoring sites. These results are provisional, pending any revision of the reaction rate for OH + C2 H3C13.

CONCLUSIONS
A global three-dimensional Lagrangian tracer transport model has been eInployed to study the sources and sinks of atmospheric methane. A global average tropospheric hydroxyl radical concentration of 6.4 x 105 cm -3 was determined using published methyl chloroform emissions and atmospheric concentration data. Southern hemisphere averaged OH was 7.2 x 105 cm -3 and northern hemisphere OH was 5.5 x 105 cm -3. For the years 1981-1984, methyl chloroform emissions data were found to be inconsistent with the observed growth in atmospheric concentration if a constant or near-constant average hydroxyl radical concentration is assumed. The hydroxyl radical concentration must fall ~ 25% to produce the observed growth in methyl chloroform concentration over the period 1981-1984. The more likely situation is that the methyl chloroform emissions for 1981-1984 need upward revision.
Two source functions describing the spatial and temporal distribution of the flux of methane to the atmosphere were developed. The first model, SF1, assumed that the releases of methane from rice paddies, wetlands, enteric fermentation, termites, and biomass burning were proportional to net primary productivity. Model-predicted methane concentrations were in good agreement with available observations except at high northern latitude NOAA/ GMCC sites. Including releases of methane from fossil fuels, landfills and the oceans an annual source of ~ 623 Tg CH4 was required to explain the observed growth in atmospheric methane concentration. A globally averaged atmospheric lifetime of 8.3 years is deduced from SF1. With a new, slower rate constant for OH plus methane, this lifetime is raised to 10.0 years.
The second source function, SF2, separately identified the releases of methane fi'om rice paddies, wetlands, enteric fermentation, termites and biomass burning. The major difference between SF1 and SF2 is the prediction of an intense release of methane corresponding to the rice paddies of Asia and Southeast Asia by SF2. SF2 leads to a different latitudinal distribution of the release of methane with a greater release of methane occurring at higher latitudes than SF1. Accordingly, SF2 leads to a lower estimate of the total methane source of ~ 611 Tg. A globally averaged atmospheric lifetime ooe 8.5 years is deduced from SF2. With a new, slower rate constant for OH plus methane, this lifetime becomes 10.2 years. SF2 also produces estimates of atmospheric methane concentration which are in poorer agreement with NOAA/GMCC observations of methane at high northern latitudes than SF1.
On the basis of the model, results obtained using both methane source functions developed in this study it would appear that further investigation of the sources and sinks of methane at high northern latitudes is warranted. Measurements of atmospheric methane with the objective of determining methane.sources and sinks rather than just long-term trends are also required.