We compare modeled and observed atmospheric methane (CH4) between 1996 and 2001, focusing on the role of interannually varying (IAV) transport. The comparison uses observations taken at 13 high-frequency (∼hourly) in situ and 6 low-frequency (∼weekly) flask measurement sites. To simulate atmospheric methane, we use the global 3-D chemical transport model (MATCH) driven by NCEP reanalyzed winds at T62 resolution (∼1.8° × 1.8°). For the simulation, both methane surface emissions and atmospheric sink (OH destruction) are prescribed as annually repeating fields; thus, atmospheric transport is the only IAV component in the simulation. MATCH generally reproduces the amplitude and phase of the observed methane seasonal cycles. At the high-frequency sites, the model also captures much of the observed CH4 variability due to transient synoptic events, which are sometimes related to global transport events. For example, the North Atlantic Oscillation (NAO) and El Niño are shown to influence year-to-year methane observations at Mace Head (Ireland) and Cape Matatula (Samoa), respectively. Simulations of individual flask measurements are generally more difficult to interpret at certain sites, partially due to observational undersampling in areas of high methane variability. A model-observational comparison of methane monthly means at seven coincident in situ and flask locations shows a better comparison at the in situ sites. Additional simulations conducted at coarser MATCH resolution (T42, ∼2.8° × 2.8°) showed differences from the T62 simulation at sites near strong emissions. This study highlights the importance of using consistent observed meteorology to simulate atmospheric methane, especially when comparing to high-frequency observations.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Methane (CH4) is a radiatively and chemically important trace gas whose global budget is only partially understood. It contributes about 27% of the total 2.6 W m−2 increase in radiative forcing caused by the anthropogenic release of long-lived greenhouse gases in the industrial age [Hansen and Sato, 2001]. Methane is also important in terms of atmospheric chemistry, playing a key role in the CO-NOx-hydrocarbon-OH oxidation cycle. High-precision long-term atmospheric CH4 measurements that are useful for global modeling studies are conducted at approximately 80 worldwide locations. Measured concentrations of methane are jointly controlled by the CH4 sources and sinks, and atmospheric transport. For the past 25 years, direct observations have recorded the long-term global deceleration in the rate of increase of the atmospheric methane burden, broadly consistent with an approach of methane to steady-date [Dlugokencky et al., 1998]. Significant year-to-year variations in the methane global growth rate have also been observed, such as the large CH4 increase in 1998 [Dlugokencky et al., 2001]. These global fluctuations are likely due to changes in sources and/or sinks, which can in principle be estimated from the observations [Cunnold et al., 2002]. As additional measurements have become available, there has been an increased effort to estimate methane emissions for specific spatial regions. Sources of methane include both natural (e.g. wetlands, termites) and anthropogenic (e.g. rice, domesticated animals, coal mining, natural gas) processes, most of which have significant quantitative uncertainties. Methane observations implicitly contain information about the surface emissions of methane, which can be extracted and quantified through the use of atmospheric models and statistical inverse methods. This “top-down” approach complements “bottom-up” estimates of CH4 emissions which typically rely on the extrapolation to global scales of point flux measurements and/or on process-resolving models.
 Most methane observing sites use flask sampling, in which air is collected in flasks at approximately weekly intervals [e.g., GlobalView-CH4, 2001]. The actual methane measurement is then made at a central laboratory at a later date. Flask measurements allow efficient global coverage since a sample need only be collected, rather than actually measured, at any given location. Most flask measurements are designed to capture baseline air, as distinct from polluted air due to continental emissions. The weekly frequency of the observations necessarily does not capture short term transport and methane emission events, or inadvertently captures weak pollution events that may be difficult to interpret. Global atmospheric modeling studies of methane have typically relied on smoothed, monthly mean data based on multiple years of data. In an early study, Fung et al.  used average seasonal values from global flask measurements to test different methane emission scenarios. Most atmospheric CH4 inverse studies have also relied on these seasonal monthly means [e.g., Houweling et al., 1999; Hein et al., 1997]. Cunnold et al.  estimated changes in global emissions between 1985–1997 based on monthly means observations using on 2- and 3-D modeling. The simulations in these earlier studies have also used climatological model winds, or observed meteorology of a single year, which may or may not be applicable to seasonal means based on flask sample averages. Two more recent studies have simulated methane flask data using observed interannually-varying meteorology: Dentener et al.  examined anthropogenic emission trends between 1979–1993 using interannual monthly CH4 averages based on smoothed flask data, and Warwick et al.  performed a multi-year analysis of changes in CH4 growth rates based on monthly means of smoothed flask data at 9 sites, specifically looking at the role of transport IAV.
 Over the past two decades, several high-frequency (e.g. every 40 minutes) in situ monitoring stations which measure atmospheric methane have become active. The Global Atmospheric Gases Experiment (GAGE) and Advanced Global Atmospheric Gases Experiment (AGAGE) network, for example, have measured CH4 mole fractions at 5 latitudinally distributed locations since 1986 [Prinn et al., 2000]. These high-frequency measurements capture the variability of CH4 concentrations, including the effects of subweekly synoptic transport events, much more effectively than flask observations. They offer, in principle, much greater information about methane sources, sinks, and transport compared to the weekly flask measurements. The simulation of high-frequency measurements in general requires the use of high-resolution 3-D chemical transport models. Cunnold et al.  estimated changes in global emissions between 1985–1997, using monthly means based on both high and low-frequency observations. Janssen et al.  and Wang and Bentley  have used high-frequency observations and meso-scale models to estimate regional methane emissions. Houweling et al.  simulated high-frequency measurements at Mauna Loa and Barrow for a single year using consistent meteorology in a global model. No published study has yet used a global model to simulate worldwide high-frequency CH4 measurements using interannual transport. High-frequency measurements potentially allow a much more precise study of the impacts of interannually varying transport than studies based solely on modeling or flask observations. For example, do large scale atmospheric transport phenomena, such as El Niño, affect CH4 observations in a predictable fashion?
 In this study, we compare modeled and observed methane at 13 high-frequency and 6 low-frequency measuring sites using temporally consistent analyzed observed meteorology between 1996–2001. We test the ability of the chemical transport model (MATCH) to simulate worldwide CH4 observations, which have very different spatial and temporal signatures. This includes comparison to methane observations in polluted air, thus testing the common perception that a global model can only replicate smoothed, baseline observations. These model-observational comparisons also provide insight into the processes controlling the observations, especially where the agreement is good. Available methane time series data were used from relevant data archives for high (in situ) and low frequency (flask) observations. For the model runs discussed here, a time-varying but annually repeating methane surface map based on an optimally estimated set of methane sources was used [Chen, 2003]. For the methane sink, we used a monthly-varying OH distribution produced by a version of MATCH which contains comprehensive photo-chemistry. The magnitude of this OH field was scaled to reproduce methyl chloroform (MCF, CH3CCl3) observations over the late 1990s. Both the emissions and the OH fields vary by month but not year-to-year. MATCH was run forward with these emissions and OH fields using NCEP meteorology between 1993–2001. We then compared the forward run to observations after three years of spin-up, examining both general features and specific events for 1996–2001. Our overall goal was to elucidate the effects of transport variability on short (hours to days) and long (year to year) time scales on the observed methane concentrations. Interannual changes in methane behavior at any particular site are governed by changes in transport, sources, and sinks. By choosing smoothly varying OH (diurnally and seasonally) and monthly varying sources, neither of which change year-to-year, we thus ensure that the model variability on the above short and long time scales is associated primarily with transport and not the assumed methane sources and sinks. We further compare model simulations at two model resolutions: T42 (2.8° × 2.8°) and T62 (1.8° × 1.8°).
 Comparison of the model to high-frequency observations is generally straightforward because observations and model output frequency are fortuitously similar. The model-observation comparison using flask measurements is more difficult because of the challenge of accurately simulating isolated individual CH4 observations. We also examine monthly mean model-observation comparisons based on both in situ and flask sampling. In general, these studies of the role of transport IAV and the effect of high versus low-frequency observations for CH4 may also extend to other long-lived constituents such as CO2, CF2Cl2, and N2O that also have large surface fluxes in the northern hemisphere.
Sections 2 and 3 contain brief descriptions of the MATCH model and the CH4 observations, respectively. Sections 4 and 5 describe the methane sources and sinks, respectively, used in the MATCH simulation. Section 6 provides the main results of the observational-model comparison at all sites. Further analysis of the importance of modeled transport IAV is contained in section 7. Section 8 summarizes model-observational comparisons for monthly means for in situ and flask CH4 data. Section 9 provides a summary and the conclusions.
2. Model Description
 The MATCH model was developed to realistically simulate atmospheric constituents using observed analyzed meteorology [Rasch et al., 1997; Mahowald et al., 1997b; Lawrence et al., 1999]. For the methane simulations presented here, MATCH is driven by National Centers for Environmental Predication (NCEP) reanalysis meteorology at T62 spectral horizontal resolution, which corresponds to approximately a 1.8° × 1.8° latitude-longitude grid. In the vertical, NCEP has 28 levels between ∼1000 and 2.9 mb. The surface (bottom) layer varies between 50 to 100 meters in height. The time-varying planetary boundary layer determined by MATCH contains several of the bottommost layers. Eight meteorological inputs drive MATCH in our configuration: horizontal winds (U, V), temperature (T), surface pressure (PS), surface heat and water vapor fluxes (SHFLX, WVFLX), and surface stresses (TAUX, TAUY). These inputs are available at a time resolution of 6 hours from the NCEP archived meteorology. MATCH linearly interpolates this data to the 30 minute model time-step used for T62 resolution. The water vapor field (Q) is initialized using NCEP data at the first time step, after which MATCH predicts its own hydrological cycle. Rasch and Lawrence  introduced the mass-conserving SPITFIRE advection scheme, replacing the original semi-Lagrangian (SLT) advection scheme. Mahowald  examined the sensitivity of transport to different moist convection schemes within MATCH. Our version uses convection schemes based on Zhang and McFarlane  and Hack . The planetary boundary layer (PBL) parameterization is based on Holtslag and Boville  and is similar to the NCAR CCM3 scheme [Kiehl et al., 1996]. MATCH has been used in the study of numerous gases, including Rn, CFCl3, and CH3Br, [Mahowald, 1996; Mahowald et al., 1997b; Jensen, 1999]. It has also been used in studies of the sulfur cycle and aerosols [Lucas, 2003; Lucas and Prinn, 2003; Rasch et al., 2000]. Lawrence et al.  added a comprehensive photochemical module into MATCH. Kuhlmann et al.  have recently modeled ozone using an updated version of this model. The OH fields used in our methane and methyl chloroform (CH3CCl3) simulations were produced by this latest photochemical version.
 A useful test of transport involves the comparison between modeled and observed values of long-lived, anthropogenic gases that have relatively well-known emission magnitudes and distributions. These comparisons are useful to carry out before the simulation of more complex biogenic gases such as CH4 and CO2, which also have long-lifetimes compared to global mixing, but have much more uncertain emissions. Mahowald et al. [1997a] compared simulated and observed interhemispheric gradients of CCl3F, using MATCH with NCEP winds at T42 (∼2.8° × 2.8° horizontal) resolution between 1990–1991. CCl3F (CFC-11) is produced anthropogenically for refrigerant and foam-blowing applications, has a lifetime of ∼50 years, and like CH4 and CO2, is dominated by Northern Hemispheric emissions. Mahowald et al. [1997a] found that the simulated interhemispheric gradient (IHG) of 14 ppt was statistically in agreement with the observed IHG of 13 ppt. Another long-lived tracer of anthropogenic origin, sulfur hexafloride (SF6), has been simulated using a similar version of MATCH at T21 (∼5.6° × 5.6° horizontal) resolution [Jockel, 2000]. This compound is emitted by leakage from electrical switching equipment; its global emissions are based on electrical power usage and population distributions. Jockel  compared MATCH simulations to SF6 observations for 1993 following the Transcom 2 protocol described in Denning et al. . The T21 simulation reproduces the SF6 IHG (defined at flask 20 sites) to within 10%. The ability to reproduce the observed IHGs of these anthropogenic compounds adds confidence to this aspect of MATCH global transport when simulating more complex tracers such as CH4 and CO2.
3. Methane Observations
 The locations of all methane time series data used in this study are shown in Figure 1 and listed in Table 1. We chose 13 high-frequency in situ and 6 low-frequency flask sites active between 1996 and 2001. The 13 high-frequency stations (red circles) measure methane mole fractions between 24 and 36 times per day using in situ automated gas chromatographs with flame ionization detectors (CG-FID). Although more than 13 high-frequency stations currently measure CH4, we chose stations that: (1) had readily available data, (2) had data free from possible local contamination, and (3) covered most of our simulation period. Flask sampling occurs at approximately 80 sites, overlapping in situ sites at approximately 7 locations at any one time, thus allowing flask in situ intercomparisons. We also chose 6 flask sites which sample regions not covered by the in situ measurements and representative of different meteorological and geographical regimes. In situ and flask sampling represent the two complementary approaches to global methane sampling: one providing high sampling frequency and the other providing high spatial coverage. Most of the approximately 80 CH4 observing sites are located in the Northern Hemisphere, while tropical land regions in particular are undersampled.
Table 1. List of Methane Observing Sites Used in This Study, Including Laboratory and Adjustment Factor for Conversion to the AGAGE Standarda
High-frequency sites where flask samples are also taken are indicated with an asterisk. AES, Atmospheric Environment Service [Worthy et al., 2000, 1998]; AGAGE, Advanced Global Atmospheric Gases Experiment (earlier phases were Atmospheric Lifetime Experiment (ALE) and Global Atmospheric Gases Experiment (GAGE) [Prinn et al., 2000]); CMDL, Climate Monitoring and Diagnostics Laboratory (NOAA) [Dlugokencky et al., 1994a; GlobalView-CH4, 2001]; NIES, National Institute for Environmental Studies [Tohjima et al., 2002]; JMA, Japan Meteorological Agency.
High-Frequency Stations (13 Locations)
Mace Head, Ireland
Cape Ochi-Ishi, Japan
Trinidad Head, CA
Tenerife, Canary Islands
Mauna Loa, Hawaii
Ragged Point, Barbados
Cape Matatula, Samoa
Cape Grim, Australia
Flask Only Sites (6 Locations)
Cold Bay, Alaska
Niwot Ridge, Colorado
Mahe Island, Seychelles
Halley Bay, Antarctica
 The standards used for the absolute calibration of methane mole fractions differ between certain laboratories. Intercomparisons of these different standards have been performed [e.g., Prinn et al., 2000; Cunnold et al., 2002; GlobalView-CH4, 2001], allowing conversion of methane data to a common standard. Table 1 lists the calibration factor used to convert other CH4 measurements to the AGAGE CH4 standard for this study. The absolute calibration used by AGAGE is based on the Tohoku University gravimetric technique for methane, as described in Cunnold et al. . This scale is also used for many of the Japanese NIES sites [e.g., Tohjima et al., 2002]. Dlugokencky et al. [1994b] describe the Climate Monitoring and Diagnostics Laboratory (CMDL) calibration which is used by the majority of flask sampling networks, as described in GlobalView-CH4 . Comparisons between the AGAGE/NIES and CMDL absolute CH4 calibrations indicate that the CMDL calibration is about 1.1% lower (∼20 ppb). Different calibration multiplicative factors between AGAGE and CMDL are reported in the literature: factors of 1.0119, 1.01069, and 1.0152 are reported by Cunnold et al. , Prinn et al. , and GlobalView-CH4 , respectively. We choose the 1.0119 value since it is the most recently reported inter-calibration factor that has been used to compare field data. In this work, all measurements based on the CMDL calibration are multiplied by this value. Cunnold et al.  noted that, after correction, some CMDL measurements are still about 1 ppb lower than the corresponding AGAGE measurements at the same locations. This small offset can be either attributed to a small error in the inter-calibration factor, or to small differences in the exact sampling time and location. Our major conclusions in this paper are not sensitive to these small differences in absolute calibration. The reported precision due to random instrumental error of most methane measurements is between 0.07–0.2% [Cunnold et al., 2002; CMDL, 2001] which corresponds to an uncertainty range of about 1–3 ppb. This precision is typically determined by repeated measurements of the same sample. The spread in measured values due to instrumental imperfection determines the measurement precision. When comparing to simulations, this error is generally small compared to the mismatch (or “representation”) errors expected between a point observation and a model value which is an average over a large grid volume.
 The five GAGE/AGAGE stations provide the largest number, and longest running set of high-frequency in situ methane measurements on the global scale. The five locations are Mace Head (Ireland), Trinidad Head (California), Ragged Point (Barbados), Cape Matatula (Samoa), and Cape Grim (Tasmania). As described in Prinn et al. , methane mole fractions are measured in situ by GC-FID approximately every 40 minutes (36 times per day). The now inactive GAGE instruments (predecessor to AGAGE instruments) measured methane approximately every 2 hours. However, GAGE methane measurements are generally considered to be less reliable than the AGAGE measurements [Prinn et al., 2000]. For the purposes of this study, we used CH4 data from AGAGE only, with methane measurements starting between August 1993 (Cape Grim) and July 1996 (Cape Matatula). The AGAGE sites, with about 1000 measurements during one month, effectively sample large volumes of the lower troposphere across the major latitudinal zones. The black curves in Figure 2 show methane data for the five different sites (with 100 ppb offsets for each station progressively from Cape Grim for clarity). Each measuring station has a distinctive seasonal cycle that depends on the seasonality of emissions, destruction by OH, and transport. A slow methane growth rate of approximately 0.5% annually is observed at all sites, although it is difficult to discern from the scale of this figure. For Cape Grim, which is relatively distant from strong seasonal emissions, the cycle is dominated mostly by the seasonal cycle of OH. The seasonal cycle at Cape Matatula is similar in phase to Cape Grim, but smaller in amplitude. The local late summer trough at Cape Matatula is partially masked by the intrusion of Northern Hemispheric winter air which has higher methane concentrations at this time of year. This intrusion is less during El Niño which is best seen by examining CH3CCl3 data [Prinn et al., 1992, 2000]. The month-to-month variations at the three Northern Hemispheric sites are more complicated due to the influences of seasonal CH4 emissions. The late summer troughs in their cycles, however, are consistent with maximum OH concentrations during the summertime. Very large methane peaks of short duration (i.e. less than a few days) usually represent air that has just passed over strongly emitting anthropogenic source regions (i.e., “polluted” or “non-baseline”). The CH4 peaks evident at Mace Head, Trinidad Head, and Cape Grim are caused by strong emissions from continental U.K./Europe, California, and Southeastern Australia, respectively. In some cases, large methane variations do not represent emissions from a specific nearby region but rather a very different air mass origin. For example, the large synoptic-scale CH4 fluctuations between January and April at Cape Matatula represent shifts between Northern and Southern Hemispheric air, which contain high and low methane values, respectively, relative to its seasonal cycle. This is a consequence of the ITCZ's proximity to Samoa during those months, as discussed in section 6.1. The CMDL, NIES, AES, and JMA high-frequency sites usually exhibit characteristics similar to at least one of the AGAGE sites and are also discussed in section 6.1.
 Given fixed resources, flask sampling allows greater spatial coverage than high-frequency sampling because samples need only be collected, rather than actually measured, at a particular site. The locations of these six additional flask sites we will use to augment the real-time sites are shown in Figure 1 (as crosses). The flask measurements at these sites have been obtained from the flask measurements contained in GlobalView-CH4  which include data from several different laboratories. The general method of flask sampling is as follows: at each site, two identical flasks are filled with air; these duplicates are then sent to a central measuring laboratory for analysis a few weeks later. For example, CMDL uses GC-FID as the CH4 measuring technique, with a comparable precision (0.1% [Dlugokencky et al., 1998]) to the similar AGAGE on-site measurement technique (0.2% [Cunnold et al., 2002]). The CH4 mole fraction difference between duplicate (or triplicate) flasks add an additional uncertainty to the CH4 average value at each time. At most sites, collection occurs at times that maximize the chances of capturing air free of pollution (i.e. baseline air), which can otherwise lead to highly variable methane mole fractions. This is often attempted by sampling when the winds are coming from a specific direction (e.g., from the presumed clean marine boundary layer) and at a specific time of day (e.g., during the day when the planetary boundary layer is thicker and better mixed). The definition of baseline air varies from site to site, because many locations have methane peaks that do not represent pollution from strongly emitting regions. Non-baseline air is sometimes inadvertently sampled, as it is difficult to forecast the true origin of air masses. In this work, we have discarded those samples which are flagged as having obvious problems in sample collection or analysis, but retain those samples labeled as non-baseline (as well as baseline). Duplicate or triplicate measurements have also been averaged to produce single CH4 mole fractions at each time.
 In addition to the flask data, the GlobalView CH4 data set [GlobalView-CH4, 2001] contains a data product based on flask observations collected between 1983 and the present. Flask measurements above a certain threshold value (site-specific) have been removed according to a specific algorithm in order to generate a smoother methane data set [CMDL, 2001; Masarie and Tans, 1995]. In addition to these filtered flask measurements, GlobalView CH4 includes interpolated and extrapolated data to fill missing data at each site between 1984 and the present as described in Masarie and Tans . This study uses only actual observations for the following study, and not these smoothed/interpolated data, since our interest is in comparing simulated and observed values at specific times.
4. Methane Emissions
 The individual methane emissions that were input into MATCH are described in Table 2. These emissions are based on an optimally estimated set of emissions using a prior estimates from the literature, CH4 observations, and a Kalman filter with the MATCH model as described in Chen . The spatial distributions of these emissions are derived from in situ and process-modeling studies of several different methane emitting processes. Figure 3 shows the annually averaged spatial distribution of these combined surface fluxes at T62 resolution. This input flux pattern contains significant month-to-month variability due to the strong seasonal variations in the optimally estimated (and in the original a priori) methane emissions. Most of the emissions are concentrated in the Northern Hemisphere, resulting in the large observed methane interhemispheric gradient (IHG). Nearly all emissions are continental, although the logarithmic emission scale in Figure 3 captures the very small amounts of emissions due to international shipping. The spatial distribution of individual emissions in Table 2 are derived from data sets from EDGAR , Fung et al. , and Hao and Liu . The EDGAR  and Fung et al.  data sets (1° × 1° spatial resolution) contain many of the same emission processes, although their distributions and magnitudes differ. For most of the overlapping sources, we use the more current EDGAR 3.0 database. The EDGAR fields do not include seasonal variations in methane emissions from rice and biomass burning emissions, and so the other data sources were used for the a priori source fluxes. The biomass burning spatial distribution is based on the Hao and Liu  spatio-temporal distribution of biomass burning (5° × 5° spatial resolution) in tropical Africa, Asia, and America. Due to their relatively small contributions and globally distributed emissions, termites and “Other Anthropogenic” emissions were not optimized by the inversion and instead assumed as constant fluxes. The “Other Anthropogenic” source includes emissions from industrial activity and biofuel burning.
 The magnitudes of the emissions used here are obtained from an inverse method as described in Chen . Briefly, the optimization used a time-dependent inversion technique to estimate flux magnitudes of most known sources of methane using MATCH simulations at T42 resolution (2.8° × 2.8°). The total model methane surface flux is largely dependent on the global OH magnitude, which we have independently adjusted to fit methyl chloroform observations as described in the next section. As described in Chen , the inversion optimally estimated individual monthly fluxes between 1996–2001 for 3 wetland (North American and Eurasian bogs, and swamps), 1 rice, and 3 biomass burning (Africa, Asia, Americas) areas assuming distributions in each area from the above prior studies. Three time-invariant (aseasonal) components, consisting of assumed global distributions for animals/waste, natural gas, and coal emissions were optimally estimated as constant emissions over the 1996–2001 period. The global surface flux is ∼600 Tg yr−1, which is at the higher end of previous modeling studies, although this global surface flux depends on the global OH concentration. As noted above, our global OH values are adjusted to fit global CH3CCl3 observations. The literature range for the individual methane sources are also shown in Table 2. Most individual process emissions are within the Intergovernmental Panel on Climate Change (IPCC) range, which include both bottom-up and top-down studies. One source that is significantly higher in our optimized emissions than a priori bottom-up studies are the total rice emissions. This may arise from a real difference or the inability of the inversion to separate collocated rice and wetland emissions from regions of China, India, and South-East Asia. In general, bottom-up estimates of rice emissions do not include emissions from nearby inundated areas not directly associated with rice production [e.g., Yan and Cai, 2003]. These spatially collocated regions cannot be resolved at the size of a MATCH grid cell, which is discussed in more detail in Chen .
 To generate the T62 emissions fields used here, we first used the optimally estimated flux magnitudes determined at T42 resolution to adjust the flux magnitudes of original emissions fields. The original fields, at 1° × 1° spatial resolution except for biomass burning (5° × 5°), were subsequently regrided to T62 resolution. The T42 MATCH inversion results show general consistency with preliminary inversion results at T62 resolution, suggesting that the T42-derived flux magnitudes are applicable in the following T62 model-observational comparisons. The optimized emissions include the monthly interannual variability from seasonally varying sources. For this study which focuses on defining the effects of interannually varying synoptic-scale transport, we have averaged this optimally estimated interannual variability in emissions into an annually repeating set of emissions, with the same seasonality from year to year. This removes all interannual variability in the methane budget components from the simulation except for transport IAV. In terms of total emission magnitude, approximately half the total has a seasonally varying component, while the other half varies much more slowly and as noted above is assumed to emit constantly throughout the year. The emission values for certain processes, such as from biomass burning and swamps, have significant uncertainties even after their optimal estimation. Much of this uncertainty is associated with the sparseness of the measuring network, which prevents adequate discrimination of certain emissions within similar latitude bands. However, this study relies more on the global spatial and temporal emission distribution than the partitioning among individual processes. Slight inaccuracies in the magnitude or distribution of individual processes should not adversely affect the model-observational comparison of synoptic scale and interannual events as analyzed here.
5. Methane Sink
 We use a seasonally varying OH field based on a version of MATCH described in Lawrence et al. , Jockel , and Kuhlmann et al. . This MATCH version included a detailed chemical scheme, including non-methane hydrocarbons, for studies of tropospheric photochemistry. We use monthly mean results from the meteorological year 1997 at T62 resolution to define this OH field. The original field has an average annual OH concentration of approximately 0.8 × 106 and 0.9 × 106 molecule cm−3 within the troposphere averaged by mass and volume, respectively. This is at the lower end of the range of the optimized OH fields of Prinn et al.  (updated by J. Huang, personal communication, 2003) using methyl chloroform (MCF, CH3CCl3) observations. A diagnostic of the magnitude of an OH field is its ability to reproduce observed MCF observations, because the major sources of MCF are relatively well known, and its sink, like CH4, is dominated by reaction with OH. Jockel  simulated the observed time-history of atmospheric MCF at the five ALE/GAGE stations using MATCH at T21 resolution and an OH distribution similar to ours. His results show overestimates of MCF at most of the sites, suggesting that the model OH concentration is too low.
 To estimate an OH global scaling factor to best fit MCF observations, we simulated high-frequency MCF observations between 1978–2001 at the 5 ALE/GAGE/AGAGE sites [Prinn et al., 2000] using constant scaling factors of 1.0, 1.1, 1.2, 1.3, 1.4, and 1.5 to multiply the original MATCH OH field. This simulation was conducted at T42 resolution for computational efficiency, but should yield similar results to T62 resolution due to the global nature of the model-observation comparison. Historical MCF emission strengths were taken from EDGAR  with regional and global totals based on the work of McCulloch and Midgley , who estimated annual MCF emissions using industrial production data. The spatial distribution of MCF emissions within each broad geographic region (e.g. Europe, N. America, etc.) was assumed to be constant over the above period as described by Midgley and McCulloch . Briefly, total emissions were first subdivided regionally, followed by a country-by-country subdivision based on GDP. Population density was then used as a spatial proxy for emissions within countries. We combine this spatial distribution with historical estimates of MCF emission magnitudes for input into MATCH. To initialize the runs, we have used an initial atmospheric MCF distribution for 1974 (J. Huang, personal communication, 2003). Given the approximately 4.9 year lifetime for MCF, we expect any small inaccuracies in this initial condition to have been attenuated by the mid-1980s. The Jet Propulsion Laboratory (JPL)  OH + CH3CCl3 temperature dependent rate constant has been used. A diurnal cycle scaled to the solar zenith angle is further applied to the daily average OH concentrations interpolated in the MATCH runs from the monthly mean OH concentrations from Kuhlmann et al. . This ensures low nighttime values while maintaining the daily average OH concentration.
Figure 4 shows the modeled and observed MCF time history at the 5 ALE/GAGE/AGAGE stations between 1995–2001. Modeled MCF mole fractions at each station reproduce the observed increase in MCF that occurred until about 1992, followed by the strong decrease resulting from the ban on MCF emissions as part of the Montreal protocol (not shown). For the late 1990s, a global scaling factor between 1.2 and 1.3 for MATCH OH best reproduces the observations (Figure 4). That all stations show a similar scaling factor suggests that the spatio-temporal characteristics of our adopted MATCH OH field are broadly correct. Assuming accurate emissions, interannual changes in OH are reflected in the shifts in observed MCF towards the 1.2 or 1.3 times reference OH simulated curves. Consistent with the goals of this paper, an inter-annually invariant OH factor is used for our methane simulations. In the above experiment, we have not included the oceanic sink of MCF, estimated to be about 5% of the total sink [Butler et al., 1991]. Addition of this sink would result in an approximately 5% lower OH scaling factor. We therefore chose an OH factor of 1.2 for the T62 OH fields used in our methane runs. This corresponds to a tropospheric annual average OH concentration of ∼1.1 × 106 molecules cm−3 weighted by mass.
 For the methane simulation, we used the 2-parameter JPL  CH4 + OH rate constant. The OH sink is estimated to account for approximately 90% of the total atmospheric methane loss [IPCC, 2001]. Additional sinks include consumption of methane by soil bacteria, and small amounts of destruction by O(1D) and Cl in the stratosphere. Using a process based model, Ridgwell et al.  estimate an average uptake of 20–51 Tg yr−1 by soils. The soil uptake spatial distribution shares some similarities with the wetland and biomass burning emission distributions, although there are significant uncertainties in the soil uptake spatial distribution and magnitude. The methane emission pattern used in the simulation does not explicitly include this sink. Our implicit assumption is that areas of soil uptake share a similar temporal and spatial distribution to the other emissions, such as wetland and biomass burning, with the net flux (derived optimally from observations) driving MATCH. Only large differences in the spatial-temporal distribution of soil uptake from the overall emission pattern would significantly impact our modeled atmospheric mole fractions.
6. Modeled Versus Observed Methane
 We simulated CH4 observations by integrating the methane emissions in MATCH between 1993 and 2001, with comparison to observations beginning in mid-1995. The model runs were initialized in January 1993 using a CH4 atmospheric distribution based on a multi-year run of the previously described emissions source; this distribution was then adjusted by a global scaling factor to match the observations at Alert, Mauna Loa, and the South Pole in January 1993. We first examined the model simulation of the north-south gradient. For this purpose, Figure 5 shows the modeled and observed latitudinal variation in mole fraction at the 13 high-frequency and 6 flask locations listed in Table 1 and shown in Figure 1. Here, we have annually averaged all observed and model values between 1996–2001, with the model values interpolated to match the exact observational times. Figure 5 shows that the interhemispheric (IHG) gradient of methane can be reproduced by MATCH using a realistic set of methane emissions. Certain Northern Hemispheric sites show model overestimates of the observed methane values. Some of these overestimates are due to a mismatch (or “representation”) error between the MATCH averaged grid value and the observation point measurement (i.e. the station lies in the same MATCH grid volume as nearby large sources). In addition, MATCH has relatively low boundary ventilation rates compared to other atmospheric models [Gurney et al., 2003], perhaps associated with its higher vertical resolution, vertical diffusion scheme, or meteorological inputs. These factors could result in spuriously higher modeled mole fractions than observed, especially where an observational site is located in a strongly emitting region. In the following we examine the ability of MATCH to reproduce actual high-frequency and flask observations, focusing on its ability to capture short term synoptic events. We also examine the importance of IAV by examining whether the model successfully reproduces observed interannual changes in the fluctuations in CH4 mole fractions. Finally, simulations using a coarser horizontal model resolution (T42, 2.8° × 2.8°) are compared to the T62 resolution to examine the influence of resolution on our conclusions.
6.1. Comparison to High-Frequency (in Situ) Observations
 The simulated mole fractions at the 5 AGAGE sites are compared to observations in Figure 2. The 1-hourly MATCH output has been linearly interpolated to exactly match the observational times. As can be seen, MATCH captures very well the distinct seasonal cycles at each site. The overall methane variance at each site is also nearly replicated, including non-background air (pollution events) evident at Mace Head and Cape Grim. The timing and usually the magnitude of most methane peaks and troughs are reproduced. This indicates that much of this short-term behavior must then be due to the rapidly varying NCEP-driven model transport, since model CH4 emissions and OH concentrations are varying smoothly and are annually repeating. Figure 2 also shows that the modeled values share a similar long-term growth rate as the observations. The model increase is due to the slow approach of methane mole fractions to a steady-state level that is determined by the emissions and OH sink.
Figure 6 compares the MATCH simulations to each of the 13 high-frequency stations listed in Table 1 in terms of daily averages, ordered by latitude. The model values are inverted compared to the observations to facilitate comparison; a good fit is obtained when model values are the mirror image of the observations. These inverted plots allow the high-frequency variability and seasonal cycle to be more clearly compared. The following discussion divides the 13 high-frequency sites into northern, tropical, and southern regions.
 The northern stations sample air from unpolluted terrestrial and marine regions, as well as air influenced by strong terrestrial emissions. The northernmost site, Alert, has a well defined seasonal cycle that mostly follows the high latitude OH cycle of destruction in the NH (i.e. high CH4 loss in the summer, low in the winter). Although Alert is relatively distant from strongly emitting sources, there are enhanced CH4 mole fractions for certain days during the winter. These peaks (e.g. 1997 and 1999 winter months) can be linked to overall southerlies, while non-peak years (e.g. 1998 and 2000 winter months) experience stronger northerlies at Alert, as verified by the MATCH NCEP-wind fields. The southerlies bring air with higher methane mole fractions arising from anthropogenic emissions from North America, while the northerlies bring lower methane air from the Arctic Ocean and Siberia during the winter. MATCH successfully captures the presence or absence of these transport induced peak events over these years. Barrow (brw), located on the northern coast of Alaska, samples air from clean marine areas and emitting wetland areas during the summer [Dlugokencky et al., 1995]. The seasonal cycle at Barrow is similar to Alert, but with a greater number of peaks between June and September due largely to wetland emissions. The overall variability is captured reasonably well by MATCH, although the simulation overestimates the general amplitude of the seasonal cycle. Mace Head (mhd), located on the west coast of Ireland, samples both clean air from the Atlantic and air that has passed over Europe. Peak events, which are simulated very well by MATCH, depend largely on the synoptic-scale transport which brings either polluted European or clean Atlantic air to Mace Head. The interannual variability of this transport is discussed in the next section. The Hudson Bay Lowlands in Canada, which surrounds the Fraserdale (frd) site, are a strong source of methane during the summer [Worthy et al., 1998]. Although the overall cycle of methane is captured at Fraserdale, MATCH overestimates the magnitude of the peaks, particularly during the winter months. Transport errors or misallocated emissions in MATCH could lead to these overestimates in CH4 mole fractions during the winter. The next station, Cape Ochi-Ishi (coi), mostly samples continental Asian and Pacific Ocean air in the winter and summer, respectively, as described in Tohjima et al. . MATCH captures the seasonality at coi during most of the year, but overestimates the depth of the trough during July and August. Assuming the transport is being accurately reproduced, the OH sink is either being overestimated in the summer months or the upwind CH4 sources are being underestimated during these months. The seasonality at Trinidad Head (thd), located on the western coast of California, samples mostly Pacific air and occasionally continental U.S. air with higher CH4 mole fractions. MATCH reproduces the seasonality at thd, although there are more simulated peak events in the winter than observed. This may arise from the greater surface trapping of tracers within the MATCH boundary layer, or the inadequate resolution of MATCH for resolving strong local sources that influence observing sites as discussed earlier.
 The Northern Hemispheric sites located near or within the tropics include (from north to south) Izana (iza), Hateruma (hat), Minamitorshima (mnm), Mauna Loa (mlo), and Ragged Point (rpb). These sites generally have well-defined OH-driven seasonal cycles, since most are distant from emitting regions. MATCH captures the overall seasonality and variance at these sites, in addition to specific peak events. The seasonal cycle at iza, located on a mountain plateau, is dominated by the OH cycle. Individual peak events arise from air masses originating from North American or European emitting regions [Bergamaschi et al., 2000]. Air masses originating over Northern Africa or the North Atlantic result in significantly lower measured CH4 values. MATCH can reproduce the overall variability of these peak events; for example, the large observed peak of ∼1870 ppb in late 1999 is simulated quantitatively by MATCH. Hateruma (hat) station experiences a large seasonal cycle (almost 100 ppb) due to the monsoon-derived seasonality of its transport during the year [Tohjima et al., 2002]. During the summer, southerlies bring low CH4 air masses from the relatively clean Pacific Ocean that have been depleted by OH; in the winter, northerlies bring air originating over strongly emitting continental Asian sources when OH is low. MATCH reproduces the magnitude of this cycle, in addition to specific peak events caused by synoptic transport. Peak methane events during the beginning and end of the flat trough periods are replicated in timing, if not in magnitude. For example, MATCH simulates the double peak in June 1997, the single peak in May 1998, and the double peaks in June/July 1999.
 Minamitorshima (mnm) is located several thousand kilometers to the west of Hateruma (hat) in the open Pacific. Its seasonal cycle is similar to hat, but is less intense due to its remoteness from strongly emitting regions. The magnitude of its peak events are also less intense. As at hat, MATCH is able to simulate specific peak events, which represent the effects of shifts in north-south transport in conjunction with the latitudinal CH4 gradient. The Mauna Loa station is situated on a steep 4 km high volcanic peak which experiences significant diurnal variability in transport, alternately sampling sea level and high altitude air masses [Dlugokencky et al., 1995]. Since MATCH can only simulate the average mole fraction within a relatively large grid volume, this transport cannot always be realistically reproduced. Although MATCH can simulate the timing of some CH4 peaks at Mauna Loa, especially during the winter months, the correspondence is worse than at iza, hat, and mnm. The unusual seasonality observed at Ragged Point (rpb) due to its proximity to the ITCZ and sensitivity to the large hurricane season wind shifts is also simulated quite well by MATCH. Both high peak and low trough CH4 mole fractions at Ragged Point are observed and simulated. The three peaks in early 1998, for example, are simulated very well by MATCH. Very low mole fractions are sometimes observed between July–December; these low values arise from intense short-period south-to-north transport events associated with hurricane activity as reported for CH3CCl3 and CCl2F2 by Prinn et al. . The observed sharp troughs in October 1999 are at least partially simulated by MATCH. The anomalously high observed values during this period are likely due to highly local sources near rpb which are not resolvable by MATCH.
 The two Southern Hemispheric stations, Cape Matatula (smo) and Cape Grim (cgo), have seasonal cycles with maxima in the Southern Hemispheric winter, except during La Niña years at Cape Matatula. The amplitude of their cycles are smaller than those found in the Northern Hemisphere. MATCH can reproduce the overall seasonality at both sites, including the timing, if not the magnitude, of most peak events. The summer trough period at smo has very strong peaks occasionally, due to the intrusion of northern hemispheric air. The year-to-year frequency of these peaks is related to the El Niño, and is discussed in the next section. The peak events at cgo are due to large CH4 emissions from Southeastern Australia. These events are captured well by MATCH, indicating that the NCEP winds can correctly replicate wind trajectories at this location. These peak events occur throughout the year, with no obvious seasonality.
6.2. Large-Scale Interannual Transport Effects
 We now focus on the ability of MATCH to capture interannually varying short lived methane events at Cape Matatula and Mace Head. These sites provide clear examples of the impact of transport IAV on year-to-year changes in observations. The Samoan station, located near the ITCZ in the Pacific, is well situated to capture observational changes due to large scale El Niño–Southern Oscillation (ENSO) events. This phenomenon has a profound effect on the transport of air masses in the equatorial Pacific, and can be expected to impact observed methane mole fractions in this region. Prinn et al.  have reported systematic changes in methyl chloroform observations at Cape Matatula between El Niño and La Niña years. Basically, cross-equatorial transport near Samoa in the NH winter is suppressed in El Niño years; consequently, there is less influence from cross-equatorial transport of air containing higher Northern Hemispheric methane mole fractions. Figure 7 compares CH4 mole fractions at Cape Matatula between successive strong El Niño (1997/1998) and La Niña (1998/1999) years for the months between September and May. The bottom of Figure 7 includes the Southern Oscillation Index (SOI), which corresponds to the normalized pressure difference between Tahiti and Darwin. There is a strong contrast between the 1998 and 1999 SOI as compared to the entire observational period addressed in this paper. Compared to the La Niña summer period, the El Niño summer observations contain a better resolved trough, representative of Southern Hemispheric methane observations. Excluding the very large peaks, the El Niño observations show a lower variance compared to the La Niña year, suggesting that SH air predominates, with only infrequent but intense intrusions of NH air. The greater variance of the La Niña observations suggest that both NH and SH air masses contribute over the observed period. As shown by the model values in Figure 7 (inverted scale), MATCH reproduces the general variability as well as many of the individual peaks in both periods. Again, a good fit is indicated by a mirror image between observations and model across the dashed line. The average surface wind vectors during the trough period (January to April) for 1998 and 1999 show striking year-to-year differences in transport at Cape Matatula (see Figure 7). In 1998, southeasterly winds dominate, with only a few high mole fraction (northerly wind) methane events occurring on top of the consistently lower CH4 mole fractions. In 1999, the average wind vector at Cape Matatula is slightly northeasterly, suggesting that the air at Cape Matatula is a mix of NH and SH air. This is consistent with methane mole fractions which oscillate frequently between high (northerly) and low (southerly) values, resulting in a noisier CH4 mole fractions. For other years, observed and simulated CH4 mole fractions at Cape Matatula display a behavior intermediate to 1998 and 1999, although closer to the latter. MATCH also predicts a large sensitivity to ENSO at other sites in the equatorial Pacific such as at Christmas Island and in Pacific Cruise data. Unfortunately, most of these flask sites were inactive during the early spring of 1998, precluding a similar observation-model comparison to that at smo. In addition, a weekly flask sampling frequency is often insufficient to resolve the variability of mole fraction changes as shown in Figure 7, making these flask observations more difficult to interpret. The Seychelles in the Indian Ocean is another station that is also predicted by MATCH to have anomalous tracer behavior during the El Niño year. This site is examined in section 6.3 which compares observed and modeled flask data.
 Significant interannual variations in CH4 mole fractions are also observed at Mace Head, as shown in Figure 8. The low observed mole fractions are attributed to relatively clean air coming off the North Atlantic. These baseline levels are interspersed by non-baseline pollution events of varying magnitude and duration. Most of these large peaks (i.e., 50–100 ppb above baseline) can be attributed to air recently exported from European regions with strong CH4 emissions. O'Doherty et al.  examined Mace Head chloroform (CHCl3) observations and removed pollution peaks correlated with halocarbons, which are of purely anthropogenic origin. Like methane, chloroform has natural as well as anthropogenic sources. Even after this procedure, they observed a correlation between CHCl3 peaks and low wind velocities during the nighttime, suggesting that nighttime decreases in the planetary boundary layer thickness may enhance observed mole fractions of those trace gases with nearby local sources relative to the daytime. Although this phenomenon may also contribute to some of the observed methane peaks, the majority of the peaks result from large scale transport from the U.K. and continental Europe. Figure 8 compares the mole fractions between 1996 and 2000 for the months December to April. There is a striking difference between the relative number and intensity of pollution events between the two years. The 2000 observations show very few peak events compared to the 1996 observations, which is also captured by the MATCH simulation (inverted scale) in Figure 8. As can be seen, MATCH captures the timing of most pollution events at Mace Head, as shown by the near mirror image between the model and observations across the dashed line. There is no obvious seasonality to the high peak events, which occur throughout the year. Depleted CH4 levels below 1800 ppb are also successfully modeled. The local origin of these depleted values is transport of clean oceanic air to mhd from the subtropical Atlantic.
 Since the MATCH simulation reproduces the 1996 and 2000 observed peaks but does not contain interannually varying emissions or OH, transport IAV must be the dominant factor causing the substantial differences between the two years. The December–April averaged surface wind vectors at Mace Head are shown for the two different years below the mole fraction plots. The 2000 average indicates the presence of strong south-westerlies which contain relatively clean air from the North Atlantic. This source is consistent with the observed low CH4 mole fraction baseline interspersed with few peak events emanating from Europe. The average winds during 1996 show a weak southeasterly component, in contrast, indicating that continental sources are a significant component of the observations. The result is that there are many more and stronger peaks during 1996. The large difference between the two periods can be linked to the North Atlantic Oscillation (NAO), which dominates large scale transport variability in this region. The NAO index is shown below the surface wind plots in Figure 8. It is computed as the normalized pressure difference anomaly between the Icelandic (polar) low and Spanish (subtropical) high pressure system between December and March. A positive index corresponds to a strong Icelandic low and subtropical high, which results in strong westerlies on a more northward track across the Atlantic. A negative index corresponds to a weaker Icelandic low and subtropical high, resulting in a weakened westerly flow on a more southward track. There is a significant difference between the NAO index for December-March 1995–1996 and 1999–2000. This difference is reflected in the NCEP averaged wind fields in Figure 8, which show strong westerlies and weak easterlies affecting Mace Head during these two periods, respectively.
6.3. Comparison to Low-Frequency (Flask) Observations
 This section describes observation-model comparisons at six methane flask locations (Table 1). Figure 9 contains observed and modeled CH4 mole fractions measured at approximately weekly frequency between 1996–2001. Again, the model values are shown on an inverted scale relative to the observations. These sites lie across a range of latitudes, and sample regions different than the high-frequency stations previously discussed. Duplicate and triplicate flask samples have been averaged to produce single CH4 mole fractions at each time. For comparison, model values at the exact time of CH4 sampling have been interpolated from the nearest two bounding MATCH CH4 mole fractions. The northernmost site, Cold Bay station (cba) located at the tip of the Alaskan peninsula, samples air mostly from the North Pacific. It has a late summertime trough which follows the OH concentration maxima in the early summertime; this trough also exceeds methane increases due to summertime wetland emissions. MATCH captures the relatively well defined seasonal cycle, although it overestimates the depth of the brief summertime trough. The Baltic Sea (bal) and Niwot Ridge, Colorado (nwr) sites are two continental sites surrounded by strongly emitting regions. The Baltic Sea site in particular shows significant scatter presumably due to nearby European sources, although the seasonal cycle due to OH destruction is still discernable. MATCH captures the phase of the bal and nwr seasonal cycles, but the timing of individual observations are much more difficult to reproduce. MATCH reproduces some of the observed peak events in late 1997 at bal, although the sharpness of the peak is not as well simulated. MATCH often overestimates the overall variance in CH4 mole fractions at both sites, suggesting that the observations sample cleaner air than what is being modeled due to inadequate model resolution or boundary layer ventilation.
 The next subplot shows CH4 time series at the Seychelles (sey), which is located in the Indian Ocean. It has an unusual seasonal cycle due to its proximity to the ITCZ. During the early part of the year (the Southern Hemispheric (SH) summer) the ITCZ is shifted relatively southward; consequently, sey experiences strong northerlies from the Indian subcontinent. These northerlies carry air with enhanced methane mole fractions from multiple South Asian emission sources to sey, resulting in a strong seasonal methane peak. During the SH fall the ITCZ shifts northward, resulting in net southerlies at sey. These southerlies contain relatively clean SH air from the Southern Hemisphere, resulting in a strong methane decline. For the rest of the SH fall and winter the Seychelles mimic the low-amplitude seasonality of the SH stations, with a local peak in August and September during the period of lowest OH in the SH. Interannual variability in the sey observations is most obvious for the seasonal peaks in the SH summer. MATCH reproduces, at least qualitatively, much of this behavior suggesting that these changes are due mostly to large scale transport changes from year to year, as opposed to source/sink changes. For example, the lower 1998 peak relative to 1997 and 1999 is captured by MATCH, although the peaks are defined by a only a few measurements. MATCH also reproduces the reduced SH low amplitude seasonal cycle observed in 1997 (immediately following the large seasonal peak defined by 3 measurements) compared to subsequent years. The effect of El Niño on cross-ITCZ transport over the Indian ocean has not been as well studied as for the tropical Pacific. The reduced low-amplitude cycle in 1997 followed by a reduced peak in early 1998 suggests that cross-ITCZ transport is actually enhanced at the Seychelles, with neither northerlies or southerlies dominant. The NCEP wind vectors show some evidence of weaker north-south wind vectors during 1998 compared to adjacent years. The MATCH high-frequency output (not shown) shows a significantly reduced methane peak at sey during the early 1998 period. However, a definitive corroboration of this result would require high-frequency CH4 observations at the Seychelles.
 The mole fractions at Ascension Island (asc), located off the coast of West Africa, are similar to other SH sites, with low and high values during the SH summer and winter, respectively. Biomass burning in central Africa during the months of August–September may also contribute to the winter peak amplitude of the observed cycle. MATCH captures the seasonal cycle at this station relatively well. Methane mole fractions are elevated during the early 2000 trough in the model, and to a lesser extent in the observations. The final site, Halley Bay (hba), has a very clean seasonal cycle that is well reproduced by MATCH. Most sites south of 30°S are remote from strong source regions and have a similarly smooth cycle that is governed by destruction by OH and replenishment by transport from the north.
 To summarize, Figure 9 shows that MATCH can capture the general seasonal behavior at flask sites, which have very different characteristics based on transport, emissions, and OH destruction. The model-observation agreement at sites remote from highly emitting region is generally good. The comparison deteriorates for flask measurements near or within strongly emitting regions. MATCH usually over predicts mole fractions at these sites (e.g., bal and nwr), but can also underpredict them. Mountain sites which experience well-mixed surface air in the daytime and free tropospheric air at nighttime are also difficult to model by MATCH due to the subgrid scale nature of this transport. In general, it is difficult to simulate single weekly measurements, especially in regions of high CH4 variability. This creates challenges in attempting to definitively attribute specific CH4 flask observations to transport events. For example, correlations to El Niño transport are more difficult to investigate at the Seychelles (low-frequency observations) than at Cape Matatula (high-frequency observations). Despite their low sampling frequency, flask sites are nevertheless currently useful because they provide most of our tropical observing locations. These stations provide at least some information about those tropical emitting regions that are not sampled by the high-frequency measuring network.
6.4. Effects of Using Lower Model Resolution
 In this section, we compare methane simulations conducted at a reduced horizontal resolution of T42 (∼2.8° × 2.8°), and compare the model-observational results to the T62 (∼1.8° × 1.8°) simulations. The T42 simulations take approximately a factor of 3–4 less computer time, due to the fewer number of grid cells and the longer time-step. This computational saving can be significant, particularly for inverse studies that require multiple trace gas-surface flux sensitivity simulations. It is thus worthwhile to compare the ability of T42 and T62 simulations in reproducing methane observations. The T42 simulations were performed using the same set of optimized emissions as the T62 runs, as described in Table 2. Consistent rules were used for both T42 and T62 model output for sites where adjacent grid cells were sampled rather than central grid cells. In terms of horizontal area, there are approximately four T42 grid cells for every nine T62 grid cells, while the vertical spacing remains the same for both resolutions. This difference in grid volumes and locations often leads to small offsets between T62 and T42 mole fractions at the same measurement site locations. In the following comparisons, we focus on differences in the T62 and T42 simulations of the observed seasonal cycles, rather than on these small offsets. Modeled and observed monthly means are used to facilitate the comparison.
 For remote oceanic sites, such as at Ragged Point (Barbados), Cape Matatula (Samoa), and Halley Bay (Antarctica), the T42 simulations are nearly identical to T62. A greater difference is observed for island and coastal sites downwind of emitting regions. Monthly mean model-observational residuals are shown in Figure 10 at four sites for T62 and T42. At Trinidad Head (thd), T42 consistently overestimates the summertime trough more than T62, which is indicative of greater amounts of CH4 mole fractions from the marine boundary layer at this coastal location. The T62 simulation more successfully captures the amplitude of the trough, which is influenced by summertime wetland emissions. At Hateruma (hat), downwind of East Asia, T42 values follow the T62 values relatively closely, although both peak and trough overestimates are sometimes even larger, as seen in late 1996 and 1998, respectively.
 The greatest difference between T42 and T62 CH4 simulations is seen for continental sites within strongly emitting regions. At these sites, MATCH generally has greater difficulty reproducing methane observations due to the mismatch errors between point observations and the relatively large grid volume average. The observations often represent air volumes that are much smaller than that resolved by MATCH. This may be due to the effect of local emissions not well resolved by the MATCH resolution of input emissions fields. This error is, as expected, larger for the coarser T42 resolution. The increase in mismatch errors from T42 to T62 is more apparent at locations such as Barrow (brw) and Baltic Sea (bal) (based on flask observations). In general, the T42 simulation exacerbates the T62 over- and underestimates of the observed CH4 mole fraction. A strong monthly overestimate is notable at Barrow in mid-2000, when T42 overestimates CH4 mole fractions by nearly 50 ppb, compared to the T62 overestimate of 30 ppb. At Baltic Sea in late 2000, T42 mole fractions are overestimated by an additional 30 ppb compared to the T62 overestimate.
 At most mid-continental sites with a significant model-observational difference, T42 simulations usually overestimate the CH4 mole fractions compared to T62. The simulated CH4 mole fractions for these regions are strongly influenced by the input surface emissions within MATCH. Differences between T42 and T62 CH4 mole fractions are linked to the differences in the surface emission magnitudes of the grid cell containing an observational site. Since the primary methane emission data sets are generally at higher resolution (e.g. 1° × 1°) than T62 and T42 resolution, there is a “smearing” of the emissions upon regridding. This “smearing” is more pronounced for the T42 case due to its coarser resolution. Assuming accuracy of the primary emissions, the T62 emission will be a more accurate representation of the methane emissions field, particularly for more localized emissions. At most observing sites in polluted regions, T42 overestimates the observations more than the T62 simulations. Here, the greater “smearing” of emissions leads to greater emissions within the relevant grid cell compared to T62. At nearly all mid-continental sites, T62 more closely reproduces the CH4 observations than T42, especially in terms of seasonal and shorter-term structure. This implies that more highly resolved emissions, and hence smaller model grid cells, are preferred for simulating observations within strongly emitting regions.
7. Modeled Impact of Transport IAV on CH4 Surface Mole Fractions
 In this section we use the MATCH simulations to quantify the effect of transport IAV at all surface locations. This is accomplished globally by using the simulated monthly mean values. As shown at Mace Head and Cape Matatula, monthly mean CH4 mole fractions can be significantly affected by interannual transport. The modeled monthly means (ynm) are first computed at each surface grid cell for all months in the simulation. This time series is then detrended by removal of linear growth rates. The monthly mean value for each month (i.e. Jan, Feb, …) is then computed from this time series to produce the multi-year average monthly mean, . Next, the root-mean squared deviations are taken between the monthly means of the full time series and the multi-year monthly averages. The root-mean squared value is then divided by the amplitude (half of the peak-to-trough difference) of the seasonal cycle (Amp) at each gridpoint to provide normalized deviations (denoted RMS) to the seasonal cycle.
where ynm is the detrended MATCH monthly means, is the multi-year monthly mean of ym, n is the year index (5 years total), m is the month index, and Amp is the amplitude of seasonal cycle. In equation (1), the difference (ynm − ) represents a deviation from the mean seasonal cycle at month nm caused by transport IAV. If only a single year of transport had been used in the MATCH simulation, then (ynm = ), since sources and sinks are annually repeating and we have removed any long-term linear trend associated with global source-sink imbalances. For this case, RMS would be zero everywhere.
 RMS values for all model surface grid cells are shown in Figure 11. Large RMS values (red and yellow) denote regions where transport IAV has a significant impact on the seasonal cycle. Low RMS values (blue) denote regions where transport IAV has minimal impact. Large RMS values require the presence of both strong transport IAV and significant CH4 gradients, the latter generated by either strong nearby sources or by transport barriers. The impact of transport IAV is greatest across a tropical band that approximately follows the ITCZ. The gradient between the northern and southern hemispheric tropical methane mole fractions is greatest across this transport barrier (Figure 11). Consequently, year-to-year shifts in the ITCZ will impact CH4 mole fractions at sites that straddle it. This is especially pronounced between El Niño and La Niña years, as discussed previously for Cape Matatula. This high RMS band is concentrated in tropical ocean regions, with generally less extension over land regions (with the exception of coastal South America). The CH4 monthly mean mole fraction variations over these tropical land regions appear to be less sensitive to year-to-year changes in cross-equatorial transport compared to the oceanic ITCZ regions. This is partially due to the larger emissions over land, which lead to larger seasonal variations (i.e. larger Amp), and hence, smaller RMS values according to equation (1). In addition, the methane gradients across the ITCZ are less sharp over land compared to ocean regions, due to the presence of large underlying land CH4 emissions on both sides of the ITCZ. Certain regions of the Northern Hemisphere also show large RMS values. Many NH sites are near strongly emitting sources such as wetlands, coal mines, and natural gas facilities. In many of these locations large scale wind changes play a role in controlling observed mole fractions, as shown earlier for the effect of NAO at Mace Head (Figure 8). Enhanced RMS values are seen along the Northern coasts of North America and Western Eurasia. In these regions, strong methane gradients are caused by the juxtaposition of strongly emitting wetland or industrial regions and non-emitting oceanic regions. The interannual variability of transport then causes significant year-to-year changes in monthly mean CH4 values. Elevated RMS values within Asia are linked to strongly emitting MATCH grid cells with CH4 values also susceptible to transport IAV. Certain oceanic regions downwind of strongly emitting sources, such as east of the Northeastern United States and Japan/China, have very low RMS values. This is due to the low impact of transport IAV relative to nearby source effects on year-to-year monthly mean CH4 mole fractions. Some highly emitting regions such as India and Southeast Asia have very small RMS values, which are related to the extremely large source-dominated seasonal amplitudes (Amp) in this region. The areas least sensitive to transport IAV are in the SH far from strong sources, due to the relative homogeneity of the CH4 mole fractions in this region. Although SH transport IAV itself may be large, weak SH methane gradients prevent any strong changes in the monthly mean variations. Cape Grim also shows low sensitivity to IAV despite the frequent occurrence of pollution events from Southeastern Australia (Figure 6). This is a result of the apparent relative constancy of the number and amplitude of pollution events from year-to-year.
 In addition to large scale IAV transport, local meteorological conditions such as changes in convective transport and planetary boundary layer (PBL) thickness will affect the year-to-year variations in mole fractions and consequently, the RMS values in Figure 11. A particularly low local winter surface temperature, for example, could lead to increased stability and thus near-surface trapping of methane emissions. Under such conditions, the modeled methane mole fraction at the surface would be enhanced relative to other years. These local transport changes would have a greater impact for grid cells containing strong emissions, and in regions which have strong summer/winter changes in PBL thickness. In general, changes in PBL thickness are less pronounced in tropical regions, which experience less temperature change and substantial mixing throughout the year.
Figure 11 also shows the relative variability at the 19 sites emphasized in this study. For the five AGAGE sites, transport IAV affects Mace Head and Samoa strongly. The relative impact is smaller at Trinidad Head, Ragged Point, and Cape Grim, which are less sensitive to large scale transport effects from NAO and ENSO. The low sensitivity of Halley Bay (CMDL) and Cape Grim to IAV suggests that their monthly means could be adequately modeled using a single year of winds. For stations such as Cape Matatula and Mace Head, however, transport IAV should be used in order to correctly simulate interannual variations in monthly means. Those locations more sensitive to transport IAV changes should also be sampled at a high enough frequency to capture the effects of transport IAV.
8. Model-Observational Comparison at Coincident in Situ and Flask Sites
 In this section, we compare model and observed monthly means based on high-frequency (in situ) and low-frequency (flask) measurements. Recent inversions of methane and other trace gases have utilized monthly mean observations and simulations to estimate interannual fluxes [Houweling et al., 2000; Gurney et al., 2002]. It is therefore useful to compare model results with observed monthly means based on in situ and flask sampling. Flask sampling occurs at 7 of the 13 high-frequency sites used in this paper: alt, brw, mhd, mlo, rpb, smo, and cgo. The goal here is not a direct comparison of in situ and flask observations at these sites, since we do not expect correspondence between monthly means based on ∼1000 measurements (in situ) and 3–5 measurements (flask). Instead, we compare model-observational residuals based on in situ and flask sampling.
 The primary way to generate a model monthly mean is to interpolate MATCH output (30 minute time-step) to the exact observational time (as shown in Figures 2 and 9), followed by monthly averaging. The observed monthly mean based on high-frequency observations is well-defined, and can be compared directly to the corresponding modeled high-frequency average. We denote this model – observational difference in monthly means as the HF residual. The monthly mean based on flask measurements is less well-defined due to the once per week sampling frequency. This 3–5 value average can nevertheless be taken to represent a monthly mean. The difference between flask and model monthly means are denoted as Flask1. Based on model-observational comparisons of atmospheric CO2 monthly means, Haas-Laursen et al.  suggested that using all model time-steps within a month was preferable to sampling the model at the exact flask time. Most other studies, including inverse studies, have used this approach [e.g., Hein et al., 1997; Houweling et al., 1999]. To test this approach, as opposed to the primary way discussed above, we have further compared observed flask monthly means (based on 3–5 observations) to simulated monthly means derived from all model time-steps contained in that month (based on ∼1000 model time-steps). These model-flask residuals are denoted as Flask2. Note that the only difference between Flask1 and Flask2 are in the modeled CH4 monthly means. For the HF, Flask1, and Flask2 residuals, an additional comparison can also be made using filtered, or ‘non-polluted’ observations. Filtered observations are typically generated using a statistical algorithm which removes polluted observations which exceed a certain baseline. For the AGAGE stations, an iterative process is used to remove polluted values using a baseline polynomial fit [Cunnold et al., 2002]. A similar process is used for the CMDL flask sites, as described for CO2 by Conway et al. . Observations at alt and smo do not contain pollution events, as both sites are remote from emitting regions. For HF and Flask1, the simulated model output is sampled at the same times as the filtered observations, to maintain consistency. The full model output is still used in the Flask2 residual. As removal of polluted data almost always decreases the observed monthly mean, the Flask2 residuals become more positive when moving from unfiltered to filtered cases. In general, filtered observations are assumed to be more readily modeled by global models (with similar filtering), since polluted CH4 mole fractions are generally more difficult to simulate due to inadequate model resolution, imperfect subgrid-scale transport, and other issues. Tests of this assumption are included in the following analyses.
 The residuals between model and observed monthly CH4 mole fractions at six observing sites are contained in Figure 12. We have not included the Mauna Loa site in Figure 12 because it has similar characteristics to the other tropical stations. For each site, there are three main curves, corresponding to the HF, Flask1, and Flask2 residuals. In general, there is qualitative similarity in the shapes of residuals among the three cases, although the magnitudes and fine structures differ. The northern sites do not exhibit a strong seasonal residual, but have shorter term deviations. Some of the longer term residual structures, such as the extended trough for smo during 1998–2000, can be linked to the large global increases in CH4 mole fractions during this time, which cannot be reproduced by our annually repeating sources and sinks. The seasonal structure in the residuals for rpb and cgo are indicative of small differences between the modeled and observed seasonality. For most sites, the HF residual is less variable than for the Flask1 and Flask2 residuals. In particular, the Flask1 and Flask2 residuals can have very strong deviations, as seen for Flask1 at brw in late 1999, cgo in mid 1998, and mhd in mid 2001. For each case, the dotted curves represent the corresponding filtered residual. At alt and smo, all measurements are considered to be pollution-free, so the filtered curves overlap with non-filtered curves and hence do not appear. There are clear difference between non-filtered and filtered residuals at brw and mhd. The filtered residuals at rpb do not show the very low residual in late 1999 seen for all the non-filtered cases. During this time, very high CH4 observations were measured at rpb, by both in situ and flask sampling. At cgo, there is a difference between polluted and filtered residuals for the HF case only, since the flask observations are taken under predicable, pollution free conditions.
 A quantitative metric to complement the residual plots in Figure 12 are χ2 values, which are contained in Table 3. At each site, the χ2 value has been computed using the relationship:
In this equation, and y represent the modeled and observed monthly CH4 mean mole fraction at month m and year n (5 total years). The average difference for the residual at each site has also been removed, using
which removes any constant model-observational differences. This offset is removed to focus on the ability of the model to reproduce shorter term variations in CH4 observations, rather than the average mole fraction for the entire period. The standard deviation in the monthly mean, Rk, for each month and site is computed from the filtered high-frequency observations, which have less month-to-month variation than the unfiltered standard deviations. These monthly standard deviations are averaged into a seasonal average (12 monthly values) for each site, as there are a few month-long gaps in the high-frequency record. A lower χ2 indicates a better model to observational correspondence. For alt and smo, at which there are no pollution events, the filtered and unfiltered χ2 will be the same.
Table 3. χ2 Values Corresponding to the Monthly Mean Residual Comparisons Shown in Figure 12
8.1. Alt (Alert) and Brw (Barrow)
 At alt, Figure 12 indicates that the HF residual has less variability than both Flask1 and Flask2 residuals. This is also seen by its lower χ2 value in Table 3. The model-observational comparison is better for Flask2 than Flask1. This suggests that taking the full model monthly mean (∼1000 time-steps), better replicates the observed flask monthly mean, which is based on 3–5 measurements. Note that alt is relatively distant from strongly emitting regions, especially during the winter when Northern Hemispheric wetlands are inactive. There are thus no pollution events recorded by either HF or Flask observations. The χ2 comparison at brw shows some similarities to alt, where HF outperforms Flask1 and Flask2 (except for the Flask2 filtered case). Again, Flask2 has a much better correspondence than Flask1. There is a large residual peak for Flask1 in June 1999. At the flask observing times during this month, the model substantially overestimates the observed CH4 mole fractions. This overestimate is much smaller for HF and Flask2. For Flask2, using the full model monthly mean avoids the strong overestimate in Flask1 at this time. Because peak CH4 events can be of extremely short duration, (i.e. less than 1 day), a small temporal mismatch between the model simulation and actual transport event can lead to a large model-observational difference. At brw (and alt), the high-frequency modeled monthly mean avoids these peak mismatch errors. Filtering of HF improves the residual comparison at brw, suggesting that the model can better reproduce monthly means after removal of large, polluted air mole fractions. The χ2 (Table 3) for Flask1 and Flask2 filtered residuals at brw also decrease; however, the large peak in June 1999 for Flask1 remains. Here, the flask observations are not considered to be polluted, but the model still overestimates the June observations at these exact times, possibly due to errors in transport. In this case, the use of all model time-steps for flask comparison is again preferable.
8.2. Mhd (Mace Head)
 At mhd, HF again has the smallest residual. Both Flask1 and Flask 2 show greater positive and negative residuals. Flask1 has large positive residuals in March 1996 and 2001, while Flask2 has a large negative residual in March 1996. Comparison between the Flask1 and Flask2 χ2 values indicates that the Flask1 comparison is superior to Flask2 (for both filtered and unfiltered values) in contrast to alt and brw. For mhd, sampling the modeled data at the exact time of flask sampling is preferred to using all modeled data over a month. The flask sampling at Mace-Head occurs during clean air conditions (i.e. when the wind is westerly). MATCH apparently simulates the CH4 mole fractions relatively well under these conditions. The model averages using all model time-steps, which contain air from all wind sectors including polluted air, lead to overestimates of the monthly mean. As before, using the filtered observations lowers the χ2 values for all cases. The residual is lowered most for the HF case, indicating that MATCH can very accurately reproduce the non-polluted observations. The improvement is smaller for the Flask cases. For Flask1, using the filtered values actually increases the residual peak in March 2001 in Figure 12, indicating that the MATCH is overestimating certain non-polluted values. For Flask2, the signs of the residual peaks are shifted significantly with filtering from negative to positive, with March 1996 as a clear example. Here, the removal of pollution events (arising from continental European emissions) decreases the observed monthly mean; since the modeled means in Flask2 are still the full model average, the residuals will become more positive. The change is smaller in Flask1 since both observed and monthly mean mole fractions are altered by filtering.
8.3. Mlo (Mauna Loa), rpb (Ragged Point), smo (Samoa), and cgo (Cape Grim)
 These stations sample the relatively clean air found in the tropics and the southern hemisphere (Figure 1). At mlo, (Table 3, but not shown in Figure 12) HF again has the smallest χ2. Note that filtering at mlo results in an increase in the χ2. There is no filtering effect for the Flask1 and Flask2, since no flask observations are considered to be polluted. At rpb, there is a significant negative residual during November 1999, which is caused by anomalously high CH4 observations in both in situ and flask records that are not modeled by MATCH. These high residual values are lessened to some degree for the HF case, due to the hundreds of other baseline CH4 observations in the July monthly mean. Use of filtered observations substantially reduces the χ2 at rpb by removing the very high July observations, especially for the Flask cases. At smo, where there are no polluted values, the correspondence is best for HF. Note that the Flask1 case does slightly better than the Flask2, suggesting that interpolating the model output to the exact flask time is preferable. As was shown for the high-frequency model-observational comparison at Matatula in Figure 7, MATCH can capture many of the short term events due to transient changes in interhemispheric transport. It is therefore more realistic to use model data at the exact time of flask sampling to produce a monthly mean, rather than all model time-steps, at smo. The comparison at cgo is similar to smo, with decreasing residuals (in order) for HF, Flask1, and Flask2. Again, MATCH can also reproduce the transient CH4 peaks (Figure 6) at cgo; thus it is better to use the exact time of flask sampling for model-observational comparison. At this station, there is an improvement when using filtered values for the HF case. Here, the pollution events from southeastern Australia have been removed. Figure 12 indicates that the residuals become smoother, due to the removal of pollution peaks, whose magnitudes cannot always be accurately modeled by MATCH, despite correct timing (Figure 6). The flask sampling procedures at cgo are designed to avoid pollution events based on wind direction, and thus flask filtering is not necessary at this station.
8.4. General Discussion
 Using atmospheric CO2 simulations, Haas-Laursen et al.  cautioned against the use of modeled data sampled at the exact time of flask observations. They concluded that such sampling could exacerbate bias errors in the model representation of observed flask averages. This analysis was conducted, however, using uncorrected general circulation model winds which in general cannot realistically capture the transport effects on individual measurements as realistically as a model using analyzed observed winds. Figure 9 shows that MATCH can sometimes replicate individual flask measurements using analyzed-observed NCEP meteorology. Figure 12 and Table 3 indicate an ambiguity in using model values at the time of flask sampling (Flask1) versus using all model time-steps (Flask2). In terms of χ2, Flask1 does better than Flask2 at three out of the seven sites in Table 3, for both filtered and non-filtered cases. The Flask1 comparison is better at mhd, where interpolating to the exact flask time does better. The reverse is true at brw, where using all model time-steps produces a monthly mean closer to the flask observations. The lack of a clear advantage between Flask1 and Flask2 highlights the uncertainties in modeling flask monthly means. The ability of MATCH to model mole fractions at specific times is the most important criteria influencing Flask1 and Flask2 differences. Flask samples are usually taken depending on specific wind direction, in addition to time-of-day criteria. An alternative strategy to using model output at the flask sampling time would have been to sample observed and simulated high-frequency data at times when wind direction is the same as the flask sampling. Unfortunately, wind direction information at the flask sites is generally not available, preventing consistent model selection based on this criteria. By using the exact flask sampling time, however, we have implicitly accounted for this sampling criteria, assuming consistency between the modeled and observed wind direction.
 At the relevant sites, MATCH almost always does better using filtered values (Table 3). This is not surprising, given that monthly means using filtered CH4 observations omit the pollution observations which are more challenging to reproduce using a global model. This is not, however, an argument to only use filtered observations, which removes much significant information on nearby CH4 sources/sinks. As shown in Figure 8 for Mace Head, MATCH can reproduce many of the high-frequency pollution events. An additional complication is the process of defining baseline and polluted observations. As more observations are taken in continental regions, it is possible that the entire observational signal may be significantly influenced by nearby emissions. Thus, in situ measurements will become increasing important in mid-continental sampling in order to capture the full variability of observations.
9. Summary and Conclusions
 This study compared simulated and observed CH4 mole fractions, focusing on high-frequency observations and the impacts of interannually varying transport on these observations. We also examined the effects of model resolution and the usefulness of low frequency flask measurements. For the model methane source, an annually repeating methane emissions field derived from an optimally estimated set of emissions was used. For the global methane sink, an annually repeating OH field from a previous run of MATCH was adjusted globally to match methyl chloroform observations. These source and sink fields were input into MATCH which was run between 1996 and 2001 using interannually varying NCEP meteorology. Predicted CH4 mole fractions were then compared to high-frequency and flask observations over this time period. The total global methane source of approximately 600 Tg yr−1 allowed MATCH outputs to closely follow the observed average CH4 growth rate between 1996–2001. By using annually repeating and smoothly varying sources and sinks, we isolated the role of modeled synoptic-scale and interannual variability in transport in explaining the methane simulations.
 Model and observed CH4 mole fractions were compared in detail at 13 high-frequency and 6 flask sites. Using analyzed observed NCEP winds, MATCH captures the general shape of the methane seasonal cycle at nearly all sites in addition to short-term synoptic-scale variability. Observations at stations located in the remote marine boundary layer and on the coasts are more accurately reproduced than those in strongly emitting regions. At the 13 high-frequency sites, MATCH can reproduce specific observed synoptic events, including pollution events, that vary from day-to-day and year-to-year. This shows that a global model can indeed simulate the non-baseline CH4 observations, which contain significant information on nearby methane sources and sinks. MATCH can also reproduce the transport effects of global climatic phenomena, such as ENSO and the NAO on methane mole fractions (specifically at Cape Matatula and Mace Head, respectively). Thus, realistic transport IAV, in addition to accurate source/sink surface fluxes, is an essential component for models to accurately simulate methane observations. At a given location, the effect of transport IAV on year-to-year deviations from the average methane seasonal cycle depends on the proximity to large methane gradients and strength of transport IAV. These transport IAV effects apply even to locations far from strong CH4 emitting regions (e.g. Samoa). A similar behavior should also apply to many long-lived species which have atmospheric distributions similar to methane such as chlorofluororocarbons, CO2, and N2O. At the 6 flask sites considered, MATCH reproduces the general seasonal cycle for both oceanic and continental locations. Specific pollution events are also simulated, although direct attribution is made difficult by the sparsity of the flask data. Individual flask observations are extremely difficult to model within strongly emitting continental regions due to the high-variability of the methane mole fractions. Flask sites in the far Southern Hemisphere such as in Antarctica, which are less susceptible to rapid changes in methane mole fractions can be modeled adequately.
 We have focused on the impact of synoptic scale transport variability on CH4 observations in this study. In addition to the resolved scales of transport, the treatments of subgrid-scale changes in vertical mixing and planetary boundary layer (PBL) thickness can also significantly affect methane observations. This is especially true at sites that are within or surrounded by strongly emitting areas [e.g., Worthy et al., 1998]. Due to the high variability of CH4 mole fractions at these locations, high-frequency measurements would be essential for comparison with models. High-frequency vertical profiles would also be extremely useful for testing simulated planetary boundary layer mixing. At these sites, possible diurnal variability in CH4 emissions may also need to be modeled (e.g. wetland emissions depend on near-surface water temperature, which can vary throughout the day). Accurately, simulating the effects of PBL transport changes from the diurnal to interannual time-scales challenges the capabilities of current global 3-D models. Using the coarser T42 resolution, simulations of CH4 mole fractions in clean locations were similar to the T62 simulations, but became worse in regions with strong methane emissions. With greater availability of mid-continental CH4 observations in the future, higher model spatial resolution will likely become increasingly important for their interpretation.
 We also compared model and observed monthly means at seven coincident in situ (∼1000 observations/month) and flask (3–5 observations/month) sites. For nearly all cases, the in situ comparison is better than the flask comparison. The flask comparison is worse for those sites that experience significant CH4 mole fraction variability due to the difficulty of quantitatively simulating single time and single point observations. High-frequency measurements are highly preferable in these locations. For the flask comparison, we showed that use of high-frequency model monthly means instead of model means based on the flask sampling times improved the model-observational agreement at 4 out of the 7 sites. There is thus an ambiguity in terms of how best to interpret the observed flask monthly mean, in addition to the uncertainty associated with defining monthly means using only 4–5 measurements. This highlights another advantage of in situ measurements, which samples at sufficiently high-frequency to largely avoid this problem. A more precise quantification of sampling errors associated with monthly means is addressed in a forthcoming paper which focuses on the inverse modeling of methane emissions. In the future, in situ measurements will likely become more important than flask measurements in inverse studies which estimate inter- and intra-annual CH4 emissions. In terms of new CH4 measuring locations, the most sensitive and hence useful sites for emissions estimates will likely have high CH4 mole fraction variabilities that can be modeled successfully. These locations will generally be downwind and some distance from the emission region of interest. High-frequency observations at these sites, in conjunction with realistic models using analyzed observed meteorology for interpretation, should help us to better understand the sources, sinks, and transport of atmospheric methane.
 We thank Phil Rasch and Brian Eaton for help in using the MATCH model, and Mark Lawrence for the OH field. We further thank the AGAGE and CMDL observational groups, and the GAW data archivers. Additional observations were provided by Doug Worthy and Yasunori Tohjima. We also thank Don Lucas for helpful comments. This research was supported by NSF grant ATM-0120468, DOE grant DE-FG02-94ER61937, and NASA grants NAG5-12099 and NNG04GJ80G. Y.-H. Chen was also partly supported by a National Defense Science and Engineering Graduate Fellowship, the industry and foundation sponsors of the MIT Joint Program on the Science and Policy of Global Change, and the MIT PAOC Houghton Fund.