TransCom model simulations of CH4 and related species: linking transport, surface flux and chemical loss with CH4 variability in the troposphere and lower stratosphere.

. A chemistry-transport model (CTM) intercomparison experiment (TransCom-CH 4 ) has been designed to investigate the roles of surface emissions, transport and chemical loss in simulating the global methane distribution. Model simulations were conducted using twelve models and four model variants and results were archived for the period of 1990–2007. All but one model transports were driven by reanalysis products from 3 different meteorological agen-cies. The transport and removal of CH 4 in six different emission scenarios were simulated, with net global emissions of 513 ± 9 and 514 ± 14 Tg CH 4 yr − 1 for the 1990s and 2000s, respectively. Additionally, sulfur hexaﬂuoride (SF 6 ) was simulated to check the interhemispheric transport, radon ( 222 Rn) to check the subgrid scale transport, and methyl chloroform (CH 3 CCl 3 ) to check the chemical removal by the tropospheric hydroxyl radical (OH). The results are compared to monthly or annual mean time series of CH 4 , SF 6 are consistent within the models, revealing vertical transport differences between models. An average IH exchange time of 1.39 ± 0.18 yr is derived from SF 6 time series. Sensitivity simulations suggest that the estimated trends in exchange time, over the period of 1996–2007, are caused by a change of SF 6 emissions towards the tropics. Using six sets of emission scenarios, we show that the decadal average CH 4 growth rate likely reached equilibrium in the early 2000s due to the ﬂattening of anthropogenic emission growth since the late 1990s. Up to 60 % of the IAVs in the observed CH 4 concentrations can be explained by accounting for the IAVs in emissions, from biomass burning and wetlands, as well as meteorology in the forward models. The modeled CH 4 budget is shown to depend strongly on the troposphere-stratosphere exchange rate and thus on the model’s vertical grid structure and circulation in the lower stratosphere. The 15-model median CH 4 and CH 3 CCl 3 atmospheric lifetimes are estimated to be 9.99 ± 0.08 and 4.61 ± 0.13 yr, respectively, with little IAV due to transport and temperature.


Introduction
The variability of atmospheric CH 4 depends on the spatiotemporal variations of the surface fluxes, atmospheric transport, and destruction due to OH, Cl and O 1 D chemistry. In recent years, measurements of CH 4 and related species are being conducted at an increasingly large number of sites at hourly or daily time intervals and with high instrumental precision (Rasmussen and Khalil, 1984;Aoki et al., 1992;Dlugokencky et al., 1998;Cunnold et al., 2002;WDCGG, 2010, for a complete list of observational programs). Satellite CH 4 observations from SCanning Imaging Absorption SpectroMeter for Atmospheric CartograpHY (SCIAMACHY), Atmospheric Infrared Sounder (AIRS), and Greenhouse Gases Observing SATellite (GOSAT) are also becoming available, albeit at a lower precision (Frankenberg et al., 2008;Xiong et al., 2008;Yoshida et al., 2011). Significant developments in understanding the distributions, trends and interannual variations of CH 4 emissions and sinks have been achieved in the past two decades through forward modeling (e.g. Fung et al., 1991;Gupta et al., 1996;Houweling et al., 2000;Dentener et al., 2003;Wang et al., 2004;Patra et al., 2009b). Inverse model results show the ability of the models to reproduce the observed atmospheric CH 4 trends and variabilities within the uncertainty of the processes involved (Hein et al., 1997;Houweling et al., 1999;Mikaloff-Fletcher et al., 2004;Chen and Prinn, 2006;Bousquet et al., 2006;Bergamaschi et al., 2009). However, further improvements (reduction in the posterior emission uncertainty) of inverse modeling results depend on a better quantification of (the errors in) the prior emissions and sinks, and on error reductions in forward model transport. Presently, inverse estimates of global CH 4 emissions range between 500 to 600 Tg yr −1 , depending on the transport properties and the chemical loss parameteriza-tion in the forward models. Bottom-up estimations of Individual flux components vary by even greater percentages (e.g. Matthews and Fung, 1987;Yan et al., 2009).
The performance of atmospheric transport models has been investigated within the TransCom project since the early 1990s for the non-reactive tropospheric species, such as sulfur hexafluoride (SF 6 ) and carbon dioxide (CO 2 ) (Law et al., 1996Denning et al., 1999). Convective parameterizations in CTMs have been tested through simulation of Radon ( 222 Rn), which has a radioactive decay halflife of 3.8 days (e.g. Jacob et al., 1997). The full chemistry model simulations of reactive species with a focus on ozone (O 3 ) chemistry have also been tested using multiple CTMs, where CH 4 is treated as a tracer with a prescribed concentration evolution (Stevenson et al., 2006, and references therein). Note that most, if not all, full chemistry models do not treat CH 4 in an interactive manner, because of its long lifetime. The CH 4 lifetime ranged from 6.3 to 12.5 yr due to large range of simulated OH concentrations in the participating models . A more conservative estimate of CH 4 lifetime is required for calculating the global warming potential (GWP) for CH 4 and its impact on climate change, or developing effective emission mitigation policies. According to Shindell et al. (2009) the 100-yr integrated GWP of CH 4 is sensitive to changes in oxidantaerosol precursor emissions and to OH-feedbacks of CH 4 emissions itself. Proper understanding of the CH 4 budget is crucial for these assessments.
The aim of the TransCom-CH 4 experiment is to quantify the role of transport, emission distribution and chemical loss in simulating the interhemispheric (IH) gradient, seasonal cycle, synoptic variation and diurnal cycle of CH 4 . Only the first two, IH gradient and seasonal cycles, are discussed in this paper. The dependence of the CH 4 budget on model vertical transport in the stratosphere is also analysed. We setup a long simulation period (1988-2007, including two years of spin-up) for the following reasons: (1) in the 1990s and 2000s methane growth rates have fluctuated between 15 ppb yr −1 to −5 ppb yr −1 (Dlugokencky et al., 1998;Simpson et al., 2006;, and (2) we would like to obtain a better understanding of the role of emissions (using a set of six CH 4 emissions scenarios), chemical loss, and transport model characteristics, such as the stratospheretroposphere exchange (STE) and the IH exchange time on CH 4 concentration variations in the troposphere. The proposed 18-yr simulation period allows a proper quantification of the removal fluxes in the troposphere and stratosphere and of the influence of transport processes on these removal rates. Since the previous TransCom intercomparison experiments spanned only a few years, the 18 yr of SF 6 simulation allows us to track the interannual variability (IAV) in the IH exchange time for the first time. We also discuss the dependence of CH 4 and CH 3 CCl 3 lifetimes on the model grid structure and transport, as well as the transport and temperature as drivers of IAVs in lifetimes. In Sect. 2, we describe the experimental protocol, followed by the key information on the participating models and analysis methodology. We focus this analysis (Sect. 3) on the comparison of model results with atmospheric observations of SF 6 , CH 3 CCl 3 and CH 4 at 8 surface sites and the salient differences in model properties. An attempt is also made to understand possible implications of (1) inert tracer (SF 6 ) transport or short-lived radioactive tracer ( 222 Rn) transport on the lifetimes and distributions of chemically active species (CH 3 CCl 3 , CH 4 ), and (2) the effect of the OH abundance, as constrained by CH 3 CCl 3 , on CH 4 . Simulations of SF 6 , 222 Rn, CH 3 CCl 3 and six CH 4 emission scenarios are commonly referred to as model tracers. Scope for further analysis using the TransCom-CH 4 database and conclusions are given in Sects. 4 and 5, respectively.

Models, measurements and methods
Previous TransCom experiments focused on chemically nonreactive species (SF 6 , CO 2 , 222 Rn). A CH 4 intercomparison requires the introduction of atmospheric chemistry. Additionally, the sources and atmospheric lifetime of CH 4 are distinctly different from CO 2 , which may provide a different view on transport model differences. Detailed doc-umentation of the requested simulation is available in the TransCom-CH 4 protocol (Patra et al., 2010). Chemistrytransport model simulations were requested for the period of 1 January 1990 to 31 December 2007, after a spin-up of 2-yr (1988)(1989) using analyzed or atmospheric general circulation model (AGCM) meteorology or a combination of both (referred here as AGCM-nudged). A schematic diagram of TransCom-CH 4 model intercomparison set-up is shown in Fig. 1.

Photochemical and surface loss processes
The following chemical removal reactions for CH 4 (R1-R3) and CH 3 CCl 3 (R4-R6) are prescribed in the forward simulations. The temperature-dependent reaction rates (k; units: cm 3 molecule −1 s −1 ) are taken from the JPL synthesis of chemical kinetics (Sander et al., 2006). The monthly-mean OH fields are provided here for online calculation in the model by combining the semi-empirically calculated tropospheric (Spivakovsky et al., 2000) and 2 (Velders, 1995). The tropospheric OH field is reduced by 8 %, an amount that was required to optimize the agreement between the TM5 simulated and observed CH 3 CCl 3 decline since 2000 (Huijnen et al., 2010). The model simulations performed here allow us to verify whether observations of CH 3 CCl 3 can also be reproduced for a longer simulation period, i.e. 1990-2007, by TM5 and a variety of other models. The supplied OH field has about equal OH abundance in the Northern Hemisphere (NH) and the Southern Hemisphere (SH) (Spivakovsky et al., 2000). Since the NH/SH OH-ratio in full chemistry model simulations varies between 1.1 and 1.5 (Krol et al., unpublished data, 2008, based on the model intercomparison described in Shindell et al., 2006) we encouraged modelers to submit another set of simulations using their preferred OH field, e.g. obtained by a full chemistry version of their model. For CH 3 CCl 3 (MCF), the photolysis rates J due to solar UV radiation are provided from an Atmospheric general circulation model-based CTM (ACTM; Patra et al., 2009b) and interpolated on each model's grid. Because the resolution in the stratosphere varies widely between models, it is necessary to scale the stratospheric loss of MCF to a common value. This value is calculated by mass-weighted averaging: Here, M(i,j,k) denotes the air mass in gridbox (i,j,k) from ACTM. Modelers were required to scale their interpolated J MCF field to match the J av,MCF field of the mass-weighted annual and global mean J av,MCF value of 7.959 × 10 −8 s −1 . Similarly, the annual and global mean rate constant for CH 4 oxidation due to stratospheric Cl and O 1 D combined is rescaled to 2.069 × 10 −10 s −1 .
The monthly deposition velocities (deposition CH 3 CCl 3 ; units: m s −1 ) of CH 3 CCl 3 to ocean surfaces are provided by Krol et al. (1998;see also Kanakidou et al., 1999). This sink should be applied in the model as: where, dz = atmospheric lowest layer depth (m), dt = timestep (s), and subscript 0 indicates initial concentration. Radon decays in the atmosphere with a half-life of 3.8 days, and this decay is calculated in the model at each timestep, following 222 Rn= 222 Rn where 222 Rn is the radon mixing ratio at all gridpoints. This setup follows the recommendation of World Climate Research Programme (Jacob et al., 1997). Due to the long timescales of CH 4 and CH 3 CCl 3 oxidation and vertical transport in the stratosphere (age-of-air ∼5 yr), several years of spin-up are required to establish realistic CH 4 and CH 3 CCl 3 vertical profiles throughout the model atmosphere. A set of 3-D initial conditions, prepared following a 10-yr spin-up simulation by ACTM, were made available for 1 January 1988 for CH 4 , SF 6 and CH 3 CCl 3 . CH 4 , SF 6 and CH 3 CCl 3 concentrations at South Pole (SPO) are 1655 ppb, 1.95 ppt and 130 ppt, respectively, for January 1988. Radon will be spun-up quickly due to its half-life of several days. Hence, its initial concentration is set to zero.

Fluxes
The typical seasonal variations of the six CH 4 emission scenarios are shown in Fig. 2a. Annual total emissions time series are depicted in Fig. 2b for CH 4 , and in Fig. 2c for SF 6 and CH 3 CCl 3 . The following source and sink components of CH 4 were considered in the six different scenarios listed in Table 1: 1. Interannually varying anthropogenic emissions (IAV ANT), based on annual mean 1 • × 1 • maps from the Emission Database for Global Atmospheric Research (EDGAR; version 3.2/FT) (Olivier and Berdowski, 2001). The combination of different emission categories and the inter-/extra-polation of EDGAR emission maps for the years 1990, 1995, 2000 are described elsewhere (Patra et al., 2009b (Ito and Inatomi, 2011) fluxes are still under evaluation, but included here since no other bottom-up wetland emission scenario was available with IAV for the full simulation period at the time the intercomparison protocol was released. VISIT is driven by climate variables from the Climate Research Unit time series version 3.0 (CRU TS3.0) dataset (Mitchell and Jones, 2005; updated values) and NCEP/NCAR reanalysis (Kalnay et al., 1996) for the periods of 1988 cycling in the inundated areas is modeled using the scheme of Cao et al. (1998, and references therein). 3. Cyclostationary natural emissions (CYC NAT), such as those from all types of natural wetlands, domestic and large-scale biomass burning, and termites, based on the GISS inventory (Matthews and Fung, 1987;Fung et al., 1991), and emissions due to rice paddies taken from Yan et al. (2009). All these emissions are scaled as in Patra et al. (2009b). Though predominantly anthropogenic, emissions from rice cultivation are included in this category because its seasonal cycle is controlled by seasonal rainfall and temperature. The emissions due to oceanic exchange (∼10 Tg CH 4 yr −1 ) are distributed over the coastal region (Lambert and Schmidt, 1993;Houweling et al., 1999) and mud volcano emissions are based upon Etiope and Milkov (2004).
4. Wetland emissions with interannual variation (IAV WL) have been derived from the wetland emission module of the ORganizing Carbon and Hydrology in Dynamic EcosystEms (ORCHIDEE) terrestrial ecosystem model (Ringeval et al., 2010). This model uses satellitederived area of inundation for the period of 1994-2000 (Prigent et al., 2007). The emission is scaled by a multiplication factor of 0.76 to match the wetland emission component in CYC NAT. An average seasonal cycle is used for the rest of the simulation periods (1988-1993 and 2001-2008).
5. Second set of wetland and rice emissions (IAV WLe) is obtained from the Vegetation Integrative Simulator for Trace gases (VISIT) terrestrial ecosystem model (Ito and Inatomi, 2011), which calculates inundated area based on analyzed rainfall, temperature (Mitchell and Jones, 2005). The rice and wetland emissions are scaled by 0.895 and 0.69, respectively, to match with CYC NAT.
6. Biomass burning emissions (IAV BB) are taken from the Global Fire Emission Database (GFED version 2), representing mainly forest and savannah burning . Since this dataset is available only after 1997, an average seasonal cycle is used for the 1988-1996 period. Unlike the wetland emissions, global total IAV BB emissions are lower than those incorporated in CYC NAT. Thus biomass burning emissions in CYC NAT is only partially replaced by IAV BB (see Table 1). This methodology is likely to double count some of the open burning and to underestimate the emissions from closed burning.    Levin et al. (2010). The 2005 distribution is used from 2006 onwards. SF 6 emissions increased from 4.77 Tg yr −1 in 1990 to 6.79 Tg yr −1 in 2007.  , 1997). Radon emission fields were not rescaled to match a global total source, but are expected to produce a global radon source of approximately 2.2 × 10 −6 mol s −1 .
3. The annual mean CH 3 CCl 3 emission distribution is taken from EDGAR3.2 and linearly corrected for the global totals following McCulloch and Midgley (2001) for the period 1988-1998. Emissions for 1999 to 2002 are taken as 27.5, 26.0, 17.7, and 16.1 Gg yr −1 , respectively. After 2002, the regional emission trends follow an exponential decay with a timescale of 5 yr updated).

Participating models and output
Twelve chemistry-transport models and four of their variants (2 at higher horizontal resolution and 2 using different OH, Cl and O( 1 D) fields) have submitted simulation results for the period 1990-2007 (Table 2). Half of these models (ACTM, CCAM, IMPACT, LMDZ, PCTM, TM5) also participated in the previous TransCom continuous experiment, where they were tested for interhemispheric transport using SF 6 , and synoptic and diurnal scale variability using continuous CO 2 measurements at surface sites Patra et al., 2008). Six other models (ACCESS, CAM, GEOS-Chem, MOZART, NIES08i, TOMCAT) participated for the first time in a TransCom experiment. The model horizontal resolution varied from 1 • × 1 • longitude × latitude to 6 • × 4 • . In the vertical, 19 to 67 levels were employed. Salient features of each model configurations (resolutions, meteorological fields) are given in Table 2 for guidance purpose only, and do not automatically link with model performance as evaluated for various features in this study.
Concerning the wind field and other meteorology, all models, except ACCESS, used meteorological fields from weather forecast models either by interpolation (offline models) or by nudging towards horizontal winds (U , V ) and temperature (online models). Most models generated output as  Onogi et al., 2007) and parameters used in nudged AGCMs are given. f The averaging period for IMPACT 1×1.25 (2002-2007) and TM5 1×1 (2003-2007) differ. $ The tropospheric OH field is taken from CHASER full chemistry model (Sudo et al., 2002) and scaled by × 0.88, and stratospheric OH is taken from AGCM (Takigawa et al., 1999) as discussed in Patra et al. (2009a). SF 6 emissions are used from EDGAR4.0 without scaling the global totals to match with Levin et al. (2010).
The TransCom-CH 4 experiment archived model simulations for 18 yrs and 9 tracers. We have sampled model output at 280 surface sites and 115 vertical profile sites (at all model levels within the troposphere) at hourly time intervals. 3-D output at 17 standard pressure levels for monthly-means for 1990-2007, and noon-time daily values for 2001-2007 are also archived.

Observational data sources and processing
Selected sites from the Advanced Global Atmospheric Gases Experiment (AGAGE; http://agage.eas.gatech.edu) and the NOAA Earth System Research Laboratory, Global Monitoring Division (http://www.esrl.noaa.gov/gmd) networks are used in this study. These sites all have simultaneous measurements of CH 4 , SF 6 , and CH 3 CCl 3 covering the 1990s and 2000s (Table 4). Unfortunately, radon measurements are not available for most of these sites. Monthly or annual mean observations have been calculated from continuous (hourly averages) or flask sampling (events) measurements data available from the World Data Center for Greenhouse Gases website (WDCGG, 2010). NOAA flasks are usually sampled under clean air (or baseline) conditions; this is usually onshore flow at coastal sites. The AGAGE continuous records have been flagged to remove local and regional pollution events.
The model outputs are extracted for the corresponding sites and sampling time from the hourly surface data files. All the models sampled concentrations for BRW and CGO stations at the nearest full ocean grid, i.e. BRWOCN and CGOOCN, respectively, to better represent baseline conditions. However, note that some models, e.g. TM5, interpolate model output to the site locations, and submit identical values at the land and ocean grids for the coastal sites. The results of HBA (75.6 • S, 26.5 • W, 10 m) site are used as a replacement for the SPO site for PCTM. These selections are made as per the modeler's advice.
For CH 4 , the NOAA-2004 calibration scale (Dlugokencky et al., 2005) agrees within 5 ppb with the AGAGE/Tohoku University calibration scale (Aoki et al., 1992). The calibration scales for CH 3 CCl 3 are as per the Scripps Institution of Oceanography (SIO) 2005 (Prinn et al., 2005)  The AGAGE and NOAA scales for SF 6 are in excellent agreement (NOAA-AGAGE = 0.02 ± 0.01 ppt, compared to repeatability of the analytical system of 0.04 ppt) (Rigby et al., 2010). The systematic NOAA-AGAGE difference of ∼ 4 ppt in 1992 reduces linearly to around 0 ppt in 2001 for the CH 3 CCl 3 concentrations as determined from co-located measurements at CGO, SMO and MHD by both the networks (P. Krummel and T. Arnold, personal communication, 2010).
As a check for the stratospheric CH 4 distribution simulated by the models, a climatology of CH 4 vertical profiles measured by the ACE-FTS (Atmospheric Chemistry Experiment -Fourier Transform Spectrometer) instrument onboard the SCISAT-1 satellite in the upper troposphere and stratosphere altitudes has been used (De Mazière et al., 2008). Please note that the data coverage of the ACE-FTS instrument in the tropical region is sparse. The HALOE/UARS (Halogen Occultation Experiment onboard the Upper Atmosphere Research Satellite) (Park et al., 1996) had a denser coverage and measurements of stratospheric CH 4 from this are also used for validating simulated vertical gradients in the tropical stratosphere (100-10 mb).
For this analysis, the models were sampled within 1 to 3 h of the sampling times of the measurements. Time series were constructed of monthly or annual mean samples for verification of model simulated seasonal cycles and interannual variability, respectively. The Pearson's moment correlation (Press et al., 1986) analysis is performed to evaluate the agreements between the simulated and observed time series for seasonal cycles and interannual variations at the 8 selected sites.

Calculation of IH gradients, IH exchange time, and atmospheric lifetimes
The IH exchange time (τ ex ) is estimated from the SF 6 annual mean concentration time series and the ratio of emission in the NH (E n ) and SH (E s ) using (Patra et al., 2009a, and references therein): c s and c n are the average concentrations of SH and NH sites, and c n−s is IH concentration gradient. Derivation of this equation assumes that c n−s is in steady state (Jacob et al., 1987). We used two sites in the NH (BRW, MLO) and SH (CGO, SPO) to estimate the hemispheric average concentrations c n and c s , respectively, at yearly time intervals (dt). Because SF 6 measurements at ALT are not available after 2005, BRW is chosen for this analysis. The values c n − c s are shown as IH gradients of each species. We have tested that the calculated τ ex does not depend strongly on the E n E s ratio.
We also estimated CH 4 and CH 3 CCl 3 lifetimes (τ ) using the mass balance equation B, E and L are the annual total atmospheric burden, net surface emission and photochemical loss, respectively. As approximate estimates of B, average concentrations of CH 4 and CH 3 CCl 3 for 8 sites are multiplied by the concentrationto-mass conversion factors of 2.845 Tg CH 4 ppb −1 and 23.689 Gg CH 3 CCl 3 ppt −1 , respectively. Ideally, vertical distribution properties of each species should be accounted for with appropriate air mass factors for calculating B. However, we note that the average lifetimes over the period of 2000-2007 calculated using ACTM simulated gridded loss rates agree within 0.01 yr for CH 3 CCl 3 with the lifetimes calculated using Eq. (5) suggesting that the simplification is acceptable. Figure 3 compares the latitude-pressure variations of the zonal mean CH 4 CTL scenario in the troposphere and lower stratosphere to representative ACE-FTS measurements (please refer to Figs. S1-S17 for individual model comparison plots of 222 Rn, SF 6 and CH 4 along the 70 • E and 180 • E longitudes, and zonal mean CH 3 CCl 3 distributions corresponding to the years 1994 and 2005). Generally, all models exhibit similar large-scale features, equator-pole latitudinal gradients, and vertical gradients in the lower stratosphere. Most significantly different, however, is the transition between the troposphere and stratosphere and the heights at which the vertical gradients maximize in the models (the green-to-blue shaded regions). The ACCESS/GEOS-Chem/NIES-08i and CAM/MOZART/CCAM models show the maximum and minimum decrease rates of CH 4 with increasing height, respectively, in the height range of 100-50 mb. CAM, CCAM and MOZART models have only few vertical layers above 100 mb and use the same reanalysis wind fields from NCEP, and the NIES model employs isentropic coordinate system in the stratosphere. The formulation of models is known to affect the simulation of tracer    (Dlugokencky et al., 1998;Butler et al., 2004) (Cunnold et al., 2002;Prinn et al., 2000) gradients in the upper troposphere/lower stratosphere region. A version of the TOMCAT/SLIMCAT model, which uses isentropic coordinates in the stratosphere, produces stronger tracer gradients and a more realistic Brewer Dobson circu-lation than the p-coordinate version used here (e.g. Hossaini et al., 2010, and references therein). The "tropical pipe" (Plumb, 1996) along the upward transport branch of the Brewer-Dobson circulation appears more "leaky" in the Fig. 4. Time series of differences between observed and simulated annual mean CH 4 in CH 4 CTL scenario (top row), SF 6 (middle row) and CH 3 CCl 3 (bottom row) at two selected sites: MLO (left column) and CGO (right column). Time series of annual mean values and tropospheric model values, averaged over 1000-200 mb and all latitudes/longitudes, are shown in Figs. S18 and S19, respectively. models than in the limb-viewing remote sensing observations by HALOE and ACE-FTS. As a result, the simulated concentration isopleths appear flatter compared to the observations with increasing latitudes in both hemispheres. The CH 4 meridional gradients between tropics (Eq-10 • N) and northern extratropics (20 • N-30 • N) are calculated to be 140 ± 72 ppb for the models and 225 ppb for HALOE in the height range of 70-30 mb.

Zonal mean concentrations
In the troposphere, vertical transport of the NH emission varies between the models, most prominently in the tropical region, where deep cumulus convection is prevalent. Table 3 shows mean vertical gradients and between model variabilities in three broad latitude ranges. Model to model differ-ences are much less distinct at the mid and high latitudes than in the tropics for SF 6 , CH 4 and CH 3 CCl 3 . However, two main categories of models can be identified based on the density of CH 4 isopleths in the height range between 350 and 200 mb. The position and concentration gradient across the tropopause differs considerably among models. These model features are also present in the SF 6 simulations suggesting the predominant role of transport in the simulation of the CH 4 vertical distributions. The penetrative mass flux due to deep cumulus convection in tropical latitudes is strongest in ACTM and GEOS-Chem (low 222 Rn concentration difference of ∼ 5.63 × 10 −21 between 850 and 200 mb), and relatively weaker in NIES08i, PCTM and TOMCAT (high 222 Rn concentration difference > 12 × 10 −21 ). This is clearer from the simulated 222 Rn distributions over the South Asian monsoon region at 70 • E during boreal summer (Figs. S2 and S4). It has been shown in Patra et al. (2009a), based on ACTM simulations with and without cumulus parameterization, that deep cumulus convective transport is the main cause for rapid vertical transport of tracers to the upper troposphere (seen as higher 222 Rn concentrations compared to the lower troposphere). Feng et al. (2011) showed that the online convection scheme used in the TOMCAT runs for this study underestimated the ECMWF archived convective mass fluxes especially in terms of the altitude extent of deep convection in the tropics. The higher horizontal resolution versions of both IMPACT and TM5 resulted in higher 222 Rn concentrations in the middle-upper troposphere (i.e. smaller difference between 850 mb and 400/200 mb) compared to their respective lower resolution simulations. This is suggesting that some convective processes are being resolved in higher resolution models that are not present in the lower resolution models. Figure 4 shows the time series of annual mean concentration differences between simulated and observed CH 4 , SF 6 and CH 3 CCl 3 at two selected sites (MLO and CGO). First SF 6 is considered, which has no chemical loss (middle row). Typical model behaviour is similar at all 8 sites (not shown). For most models, the simulated concentrations divert from the measurements by 0.2 ppt in 1995, after which differences remain at that level. The offsets between models can be explained by initial values assumed by each models. Only AC-CESS shows increasing differences in time until 2006. The ACTM OH case, which uses EDGAR4.0 emissions without scaling between 1988-2005, and 2005 emission for 2006 and 2007 and produces a slower increase in the model concentration compared to observations after 2000. This suggests that the global total emissions estimated by Levin et al. (2010) and later confirmed by Rigby et al. (2010) are adequate also for independent state-of-the-art transport models. The CH 3 CCl 3 simulations are of an intermediate level of uncertainty, as this species is emitted to the atmosphere by a relatively well quantified industrial use as a solvent. However, uncertainties remain for its loss by reaction with OH in the troposphere, photolysis in the stratosphere and the rate of STE. Until 1990, CH 3 CCl 3 emissions were increasing, followed by near-exponential decrease due to stringent restriction of its production/use by the Montreal Protocol (WMO/SAOD, 2003). The lifetimes of CH 3 CCl 3 due to photochemical removal is much longer in the stratosphere (∼28.6 yr) than in the troposphere (∼5.8 yr) (estimates from ACTM simulated loss rates at model grids; similar estimates of lifetimes using TM5 1×1 are 37.6 and 5.8 yr, respectively). Thus the troposphere to stratosphere transport plays a minor role in the global total budget of CH 3 CCl 3 after the late 1990s, because the concentration gradients across the tropopause reduced to less than 10 ppt (ref. Table 3). Although these lifetimes in the stratosphere and troposphere are within the range of independent estimates, 38 +15 −11 and 6 +0.5 −0.4 yr, respectively (Prinn et al., 2005), both the mean values are lower for ACTM. Despite the fact that the models were initialized using the same concentration at 1 January 1988, significant differences in the simulated concentrations are found already for 1990, after two years of simulation. The established model-measurement differences in 1990 persist until the end of the simulations (LMDZ CH 3 CCl 3 being the only outlier) even though the CH 3 CCl 3 concentrations become very small towards 2007. Given the CH 3 CCl 3 lifetime of less than 5 yr, these differences are caused by differences in transport and removal rather than the initialization. CH 4 is the most complicated species considered in the TransCom-CH 4 experiment, because the CH 4 surface emissions and effect of STE are less certain than for CH 3 CCl 3 . Generally, the models that simulate lower CH 3 CCl 3 concentrations compared to the multi-model mean, also yield lower CH 4 concentrations (such as MOZART, CCAM). For example, the high and low-resolution TM5 simulations show the highest concentrations of both CH 4 and CH 3 CCl 3 at MLO. However, it is interesting to note that GEOS-Chem (with TransCom OH) calculates the lowest CH 4 concentrations among all models, while the simulated CH 3 CCl 3 levels are not distinctly different. LMDZ is among the models that most strongly underestimate the observed level of CH 3 CCl 3 , whereas the opposite is true for CH 4 (Fig. 4). These contrasting behaviors among models for various CH 4 and CH 3 CCl 3 simulations clearly suggest that CH 4 loss due to the reaction with OH in the troposphere is not the only control on the CH 4 budget differences between models and that other factors such as transport differences also play a role (details in Sects. 3.3 and 3.4). Figure 5 shows model to measurement comparisons of the seasonal cycles of CH 4 , SF 6 and CH 3 CCl 3 at three selected sites (MLO, SMO and CGO) for the period 2002-2003. These sites have been selected because they are at large distances from the continental emissions for each of these species. To highlight differences in seasonality, approximate linear trends and offsets corresponding to the period 2002-2003 have been subtracted from the monthly-mean values. All models capture the salient features in the seasonal cycles at very high statistical significance (ref. Table 4 for correlation coefficients at 8 sites), except for SF 6 at MLO, where the measurements show unusual fluctuations and a large data gap during 2002. Even for the years with dense data coverage (2005)(2006), low average correlation coefficients (r) of 0.3 for the SF 6 seasonal cycles at MLO are obtained due to a very small seasonal cycle of less than 0.04 ppt. In contrast, the clear seasonalities (amplitude >0.04 ppt) at ALT, MHD, SMO, CGO and SPO are well reproduced by the models (r > 0.4; significant at P = 0.05 in two tailed Student's ttest for 24 data points). The fact that the models are able to reproduce the observed seasonal cycles indicates that even though the signals are weak, they nevertheless provide useful information for model validation. All models reproduce the CH 3 CCl 3 seasonal cycle fairly well at all 8 sites, both in phase and amplitude (r > 0.8).

Seasonal cycles
For CH 4 , the influence of the surface flux on the simulated seasonal cycles can be studied using the 6 different CH 4 scenarios. The corresponding correlation coefficients, listed in Table 4, suggest that the CH 4 seasonal cycle depends strongly on the implemented wetland and biomass burning fluxes. The CH 4 INV simulations consistently pro-duce higher correlation coefficients at all the NH sites (ALT, BRW, MHD, MLO and RPB), which was expected because the atmospheric-CH 4 inversion used data from these sites for flux optimization. CH 4 EXTRA results are next best in comparison with measured seasonal cycles for the NH high latitude sites. For any given tracer, the correlation coefficients are highest at remote SH sites, CGO and SPO, compared to all other sites. The use of ACTM OH and GEOS-Chem DOH or the higher horizontal resolution versions of IMPACT 1 × 1.25 and TM5 1 × 1 do not always improve the agreement between model and observations compared to the default implementation. These results suggest a need for improving our understanding of the CH 4 flux seasonality in the Northern Hemisphere land regions, noting that the OH loss is realistically represented as seen in the simulated CH 3 CCl 3 seasonal cycles. Because ACCESS (blue line) was not run with analyzed winds and temperature, the simulated seasonal cycles are not as good as other models for CH 4 and SF 6 highlighting the role of meteorology in simulating tracer concentrations. The role of meteorology is less pronounced for CH 3 CCl 3 because the emissions are weak in the 2000s. A more detailed analysis of seasonality at a larger number of sites will be conducted in a future study.

Interannual variability (IAV)
We calculated growth rates for all tracers as the difference between annual mean concentrations for two adjacent years. The growth rate at January 2000 is shown as the difference between 1999 and 2000 mean concentrations. The simulated and observed SF 6 growth rates (not shown) decreased from ∼0.25 ppt yr −1 in 1997 to ∼0.2 ppt yr −1 in 2000. Afterwards, the growth rate steadily increased to ∼0.25 ppt yr −1 in 2006 (please refer to Table 4 for correlation coefficients; r ∼ 0.7 for 5 sites). The length of time series considered for the correlation calculation of the IAVs is 1990-2007 for CH 4 (except for CH 4 WL BB, which has 7 yr of IAV, 1994IAV, -2000 and CH 3 CCl 3 , and 1996-2007 for SF 6 (the period when observations are available). The correlation coefficients greater than 0.44 and 0.53 are statistically significant at P = 0.05 for 18 and 12 data points, respectively. Average CH 3 CCl 3 growth rates (not shown) hovered around 0 to 5, −12 to −17 and ∼ −2.5 ppt yr −1 during 1991, 1997 and 2007, respectively, with gradual changes in between. These temporal variations are well simulated by all models (r > 0.9). Because both SF 6 and CH 3 CCl 3 emissions are of purely anthropogenic origin, their production, consumption and release to the atmosphere vary relatively smoothly in comparison to the natural components of CH 4 emissions. Figure 6 shows the model-observation comparison of the IAV in CH 4 growth rates for three broad latitude regions: the NH, tropics and SH corresponding to the CTL and EXTRA emissions (refer to Figs. S20 and S21 for others). Averaged observed CH 4 growth rates for the 1990s are 5.25, 5.06 and 7.01 ppb yr −1 in the NH, tropics and SH, respectively. Almost no increase in concentration is observed for the 2000s (except for 2007). Additionally, it can be seen that the IAVs in the growth rate are higher at the NH sites compared to the tropical and SH sites. Figure 6 shows that although the CH 4 CTL simulations capture the observed reduction in the decadal average growth rates, the IAV is not well reproduced. Most prominent is the 1997/1998 El Niño event . During this event the observations show an increase, while the simulations show a decrease in the growth rate. Interestingly, the CH 4 CTL emission and OH concentration (both without IAV) cannot explain this model behavior, which is therefore attributed to the increased CH 4 + OH reaction rate as modified by CH 4 transport and temperature in the model. This, in turn, is caused by the El Niño induced higher air temperatures (Reaction R1), resulting in faster removal of CH 4 from the troposphere and thus a decrease in the growth rate. Indeed, the 1998 CH 4 CTL lifetime (9.82 yr) is estimated to be the shortest among all simulation years.
As seen from Table 4, inclusion of biomass burning emission IAV (CH 4 BB) improves the IAV model-observation agreement at all sites compared to CH 4 CTL. However, when wetland emissions are included (CH 4 WL BB), the correlations tend to deteriorate. Compared to CH 4 BB, only CH 4 EXTRA produces better model-observation agreement for growth rates (Fig. 6, right panels). The wetland CH 4 emission simulated by the VISIT ecosystem model included in CH 4 EXTRA displays a large positive anomaly on top of the emissions from biomass burning during 1997/1998. Combined, these emissions compensate for the extra CH 4 loss due to higher air temperatures. As a consequence, these emissions result in an excellent agreement with the growth rates observed at the SH and NH sites (except that the simulated tropical signal is a bit too strong in 1998). The IAV of the multi-model average CH 4 growth rate at BRW did not correlate significantly with the observed IAV because this site is located close to the Alaskan wetland region, and the site representation error is large for the coastal sites in coarse resolution global models .
The decreasing growth rate in the 1990s, near zero growth rates in early 2000s and the reappearance of positive CH 4 growth in the late 2000s have drawn considerable interests for developing emission inventories. For example, Lamarque et al. (2010) suggested a decrease of CH 4 emissions by about 40 Tg CH 4 from 1990 to 2000 for simulating the zero CH 4 growth rate in the early 2000s using the CAM-Chem model. Their estimate is largely inconsistent with our results, which is suggesting that a steady state is achieved between CH 4 chemical destruction and emissions during the early 2000s . The EDGAR4.0 anthropogenic CH 4 emission increase of ∼4 Tg CH 4 yr −1 during 2001-2007, synchronized with the Chinese economic growth, produces inconsistencies between observed and simulated growth rates during 2003-2007 (thus the lowest correlation coefficients for the growth rate IAVs in Table 4; see also Fig. S20). This indicates that forward simulations using multiple forward transport models are useful for the verification of emission inventories. Figure 7 shows the concentration gradients between two NH sites and two SH sites obtained using annual mean observed and modeled time series. All models except for the TM5s, simulate the observed SF 6 IH gradient within the measurement accuracy of ± 0.057 ppt ( √ 2 × measurement precision). These gradients translate to an average IH exchange time (τ ex ) of 1.39 ± 0.18 yr (for all models and years), which is an indication of close model-model agreement (Fig. 8). This model-model spread is much smaller compared to the model pool of the 1990s (Denning et al., 1999), which gave τ ex range of 0.8-2.0 yr. In this intercomparison, the τ ex range between models is 0.62 ± 0.06 yr compared to 1.2 yr in the TransCom experiment during the 1990s. The average τ ex of 1.39 yr is in excellent agreement with the estimates of 1.3 yr (Geller et al., 1997) and 1.5 yr (Levin and Hesshaimer, 1996), derived using measured SF 6 time series. The underestimation by ACTM OH version is due to smaller SF 6 emissions (note: this version used EDGAR4.0 without scaling), highlighting the role of the emission strength in the forward model simulations.

Interhemispheric gradients and exchange times
A tendency towards faster IH exchange rate as calculated from Eq. (4) is seen for both the observations and the simulations (a decrease in exchange time by about 0.2 and 0.15 yr, respectively) between 1996-1999 and 2004-2007. A decrease in τ ex could have large implications for estimating fluxes of long-lived species by inverse modelling, especially for the trends in hemispheric emissions and sinks ratio. To investigate the influence of possible shifts in the global SF 6 emission distribution (I. Levin, personal communication, October 2011) sensitivity tests were performed in ACTM using three emission scenarios: (1) same as in TransCom-CH 4 Fig. 7. Interhemispheric gradients (IHGs) for CH 4 CTL , SF 6 and CH 3 CCl 3 concentrations between the NH (BRW, MLO) and SH (CGO, SPO) sites. The values at CGO (AGAGE) are adjusted to NOAA scales by adding an offset of 0.02 ppt for SF 6 , and multiplied by 1.0003 and 1.0333 for CH 4 and MCF, respectively (see text in Sect. 2.4 for further details). Please note that adjustment of the AGAGE data to NOAA scale is made just for convenience.  Figure 9a shows the trends in SF 6 emissions within three broad latitude bands. The emissions in the NH mid-high latitudes (circles) remained fairly constant  (within 14 %) in EDGAR4.0/Levin scenario, while the emissions from the NH tropical latitudes (triangles) increased by more than 100 % during 1996 to 2007. The estimated τ ex using three emission scenarios are shown in Fig. 9b. These results clearly suggest that the decrease in τ ex is caused by the trends in the SF 6 emission distribution rather than interhemispheric transport. The decrease in τ ex becomes an order of magnitude smaller as well as statistically insignificant (r = −0.34, N = 12) when SF 6 emissions were kept constant (EDGAR Y2000). The decrease in τ ex remained significant for scenario EDGAR Y2000/Levin (red square/line). This is probably because the SF 6 IH gradients do not reach steady state, since the emissions decreased during 1995-1998 and increase thereafter till 2008 (Fig. 2c), even though the E n E s ratio was fixed at 36.9, corresponding to the year 2000. Our results indicate that the time evolution of the SF 6 emissions should be introduced at a higher spatial resolution (than the presently used 2-box model), when calculating τ ex .
All three long-lived species show a similar relationship for the IH gradient: the model that produces a larger (smaller) SF 6 IH gradient generally also produces a larger (smaller) IH gradients for CH 3 CCl 3 and CH 4 in comparison with the observations (Fig. 10). The intriguing exception is MOZART, which exhibits an excellent match for SF 6 IH gradient, but produces one of the largest CH 4 IH gradients (127 ppb compared to an observed value of 101 ppb) and one of the smallest CH 3 CCl 3 gradients (0.13 ppt) during 2003-2007. Similar contrasting behaviour is also seen for several other models at lesser distinction, e.g. GEOS-Chem DOH, NIES-08i lie below the fitted line for CH 4 , but lie above the fitted line for CH 3 CCl 3 . The CH 4 IH gradients are best reproduced using the CH 4 INV emissions: deviations are within 5 ppb for 7 models (Fig. 10). Taking into account the IH gradient of all three species, TM5/CCAM, ACTM/IMPACT 1×1/PCTM and LMDZ/NIES-08 showed systematically higher, similar and lower IH gradients, respectively, compared to the observations.
Two models submitted simulations using different OH, which can be used to further investigate the sensitivity of the simulated CH 4 and CH 3 CCl 3 IH gradients to the IH difference in OH. The NH/SH ratio of hemispheric total tropospheric OH are (1) 1.32 for ACTM OH and (2) 1.11 for GEOS-Chem DOH, while that (Spivakovsky et al., 2000) used in ACTM is 0.99. The observed CH 4 IH gradient is 100.97 ppb, while the simulated gradients from ACTM and ACTM OH are 99.99 and 87.45 ppb, respectively. In order to simulate the observed CH 4 IH gradient precisely, the NH/SH OH ratio for ACTM would need to be 0.97 [= 0.99 + (1.32 − 0.99) × (99.99−100.97) (99.99−87.45) ]. This is close to the NH/SH OH ratio derived by Spivakovsky et al. (2000). Note also that the ACTM simulations of both CH 3 CCl 3 and SF 6 agree very well with the observations. Similarly, to simulate the observed CH 4 IH gradient, the NH/SH OH ratio for GEOS-Chem model should be ulated SF 6 IH gradient agrees very well with the observed value. Given the small number of alternative OH distributions in models, and the remaining uncertainties in CH 4 , SF 6 and CH 3 CCl 3 emissions, our best judgement at the moment is that we cannot falsify the NH/SH gradient (0.99) derived by Spivakovsky et al. (2000).

Photochemical removal of CH 4 and the role of transport
The calculated photochemical loss of CH 4 varies between 490 and 509 Tg CH 4 yr −1 during the first eight years (1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999), and between 497 and 513 Tg CH 4 yr −1 during the last eight years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007) of the simulation.   . 12. Time series of median (symbols) and ranges (broken lines) of atmospheric lifetimes of CH 4 (black) and CH 3 CCl 3 (red). Note that the mean CH 4 lifetimes for EXTRA (9.96 ± 0.08 yr) and CTL (9.99 ± 0.08 yr) fluxes agree within their interannual variability. A comparison of CH 3 CCl 3 lifetimes calculated using Eq. (5) and gridded photochemical destruction rates is shown in Fig. S22 (ACTM only). the troposphere is an order of magnitude shorter than that in the stratosphere. Although these relationships are again quite loose during the 1990s, as the models attain their steady state, statistically significant correlations (r = −0.69) are found for the period of 2000-2007 between the CH 4 growth rates and vertical gradients in the lower stratosphere. While the growth rates decrease from the 1992-1999 period to the 2000-2007 period in all models, the modeled gradients remain remarkably constant. Based on this analysis we suggest that the simulated concentration growth rates at the surface sites are linked to the troposphere to stratosphere transport rate of CH 4 . Figure 12 shows the temporal variability in the estimated lifetimes (Eq. 5) for CH 4 CTL and CH 3 CCl 3 (ref. Table 2 for time-averaged model specific lifetimes). The median CH 4 lifetime due to atmospheric loss processes (Reactions R1-R3) is 9.99 ± 0.08 (1σ for interannual variability) years and ranges from 9.50 ± 0.10 to 10.27 ± 0.14 yr for the different models. Here it should be remembered that all, but ACTM OH and GEOS-Chem DOH, models used the same OH distribution. The median CH 4 lifetime for 16 models or model variants agrees well with the lifetime (10.0 ± 0.17) estimated from the measured mean concentrations at 8-sites (Eq. 5). This is because a close balance is achieved between the modeled atmospheric loss and net surface emissions (513 ± 9 and 514 ± 14 Tg CH 4 yr −1 for the 1990s and 2000s, respectively). The median CH 3 CCl 3 lifetime due to all loss processes (Reactions R4-R6) is estimated to be 4.61 ± 0.13, 4.59 ± 0.18 and 4.62 ± 0.02 yr during the  simulates a lifetime of 4.87 ± 0.03 yr, which is close to the estimates using the measured CH 3 CCl 3 concentrations of 4.9 ± 0.3 yr (Prinn et al., 2005), 5.0 yr (WMO/SAOD, 2003) or 4.94 ± 0.15 yr (this study based on 8 sites). All other models calculated a shorter lifetime, by an average value of 0.3 yr. CAM, CCAM, LMDZ and MOZART calculate lifetimes of 4 yr or shorter. The interannual variation in the estimated CH 3 CCl 3 lifetimes during the 1990s is an order of magnitude higher than that in the 2000s. We find up to 5 % variability in the modeled CH 3 CCl 3 lifetimes during 1992-1999, a period with substantial emissions. During 2000-2007, when the emissions of CH 3 CCl 3 dropped significantly, less than 0.4 % variability is simulated. This implies that inversions to estimate OH from CH 3 CCl 3 observations are less uncertain since 2000, a finding in good agreement with Montzka et al. (2011).

Further work and data accessibility
For this analysis we used results of chemical tracer simulations at only 8 selected sites with measurements of atmospheric CH 4 , SF 6 and CH 3 CCl 3 , which is a very small subset of the 280 surface sites for which output is available. In addition, vertical profiles of chemical tracers and several meteorological parameters have been archived at 115 sites. More analyses on the basis of the TransCom-CH 4 simulations are planned focusing on (1) CH 4 vertical profiles measured using aircraft, (2) analysis of vertical column averaged CH 4 concentrations using TCCON observations, (3) using increase, decrease and exponential decay of CH 3 CCl 3 for optimizing tropospheric OH abundance. We also welcome use of this data set by the measurement community. In an effort towards ease of access, time series at a subset of surface sites are archived at JAMSTEC (http://ebcrpa.jamstec.go.jp/ dav/prabir/transcom-ch4/sites data/). Information on how to access the full dataset is available in the experimental protocol, archived on this website (http://transcom.project.asu. edu/T4 methane.php).
In addition to the site-specific data, gridded output at monthly intervals at the model horizontal resolution at standard pressure levels are archived for the period 1990-2007. Afternoon averages (12:00-15:00 LT -Local Time) at daily intervals are also archived for the period 2001-2007. We believe these sets of model output, their extension to recent years, will be useful for comparing the model simulations with satellite observations (SCIAMACHY, AIRS, GOSAT) and aircraft observations (e.g. HIPPO -HIAPER Pole-to-Pole Observations of carbon cycle and greenhouse gases study, Wofsy et al., 2011).

Summary and conclusions
We analyzed concentration time series of CH 4 , CH 3 CCl 3 , SF 6 and 222 Rn simulated by 16 chemistry-transport models as part of the TransCom-CH 4 intercomparison experiment. We focused the analysis on the model-to-model differences in: 1. The vertical redistribution of tracers, based on 222 Rn simulations and comparisons to CH 4 satellite observations in the upper troposphere and lower stratosphere.
3. Simulated seasonal cycles, by comparing to observed seasonal cycles at remote background stations.
4. Inter-annual variations in the simulated CH 4 growth rate, by focusing on the results of six different CH 4 emission time-lines.
5. The role of removal by OH on the simulated CH 4 and CH 3 CCl 3 concentrations.
The main conclusions can be summarized as follows: i. Although the simulated zonal mean 222 Rn concentrations agree between models, significant differences are observed in regions of deep cumulus convection, e.g. the south Asian summer monsoon domain. Unfortunately, observational evidence to check the model behavior is lacking. Models also differ in the simulated height of large troposphere-stratosphere concentration gradients of CH 4 . Compared to CH 4 satellite observations in the upper troposphere, most models appear to be too diffusive around the tropical tropopause.
ii. The IH exchange time, calculated from the simulated SF 6 distributions, ranges from 1.79 to 1.17 yr (average over 1996-2007) for the different models. The modelaverage value of 1.39 yr is in close agreement with earlier studies and observational evidence. Models that show faster IH exchange for SF 6 , also exhibit faster exchange (smaller IH gradients) for CH 4 and CH 3 CCl 3 . This multi-tracer evidence provides clear directions for the improvement of specific models. The IH exchange time calculated from the simulated and observed SF 6 time series suggest an acceleration of the IH exchange in the 2000s compared to the 1990s. However, a sensitivity study using 3 different SF 6 emission scenarios show that the acceleration is caused by an increased share of SF 6 emissions in the tropical land regions, compared to the midlatitudes, of the NH as as well as nonsteady state IH gradients in SF 6 concentrations. The estimated IH exchange time remained constant over the period of our analysis (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007) for a time invariant SF 6 emission scenario.
iii. All models reproduce the observed seasonal cycles of CH 4 and CH 3 CCl 3 at background sites very well. The simulated CH 4 seasonal cycles depend on the destruction by reaction with OH, concealing the seasonality of the underlying emissions. The simulated seasonal cycles of CH 4 are influenced by the photochemical destruction by OH: without destruction, the simulated seasonal cycles would more strongly reflect the seasonal cycles present in the emission scenarios. Two of the six simulated CH 4 scenarios (CH 4 INV and CH 4 EXTRA) show a higher correlation with the observed seasonal cycle.
iv. A set of six CH 4 flux representations was used to investigate the role of specific processes in reproducing the observed interannual variations in the CH 4 growth rate.
The control emission case (CH 4 CTL) without a significant increase in anthropogenic emissions and no interannual variability in natural emissions for the period 1990-2007 reproduces the declining growth rate in the 1990s, followed by the stabilization in the 2000s. Inclusion of interannual variation in emissions from the wetland and forest fires (CH 4 EXTRA) most successfully simulates the observed interannual variations in CH 4 . It was also suggested that the higher tropospheric temperatures during the 1997/1998 El Niño resulted in larger CH 4 destruction, whereas the observations clearly show a rise in CH 4 concentration. To match the observations, either enhanced emissions are required (as in CH 4 EXTRA), or less than average OH should have been present in this period.
v. The simulation of CH 3 CCl 3 is used to check the consistency of the employed OH abundance and distribution. Two models used an alternative OH field next to the prescribed field. Generally, models that simulate a low abundance of CH 3 CCl 3 also simulate a low abundance of CH 4 . However, there are exceptions, which indicates that CH 4 loss due to OH in the troposphere is not the only cause of the modeled CH 4 differences. Thus, horizontal and vertical transport differences may also be important.
vi. Further analysis reveals that the simulated CH 4 growth rate shows (weak) correlations with the modeled vertical gradient in the equatorial lower stratosphere. This suggests that differences in vertical mixing of the emissions and in stratosphere-troposphere exchange are the main causes of the model-to-model differences. Next to the interhemispheric transport in models, this issue requires further analysis, e.g. based on the archived model output.
vii. Finally, the multi-model lifetime estimates for CH 4 and CH 3 CCl 3 were found to be fairly constant over the simulation period with median values of 9.99 ± 0.08 and 4.61 ± 0.13 for the period 1992 to 2007. This underscores the fact that OH (assumed constant in the simulations) is the driving factor in the budgets of these gases, and that transport and temperature (affecting the reaction rate) differences play a smaller role. We find net CH 4 emissions to the atmosphere of 513 ± 9 and 514 ± 14 Tg CH 4 yr −1 (soil sink subtracted) for the 1990s and 2000s, respectively, are consistent with the atmospheric losses accounted for due to OH, O 1 D and Cl in order to simulate CH 4 concentrations and growth rates at the surface sites.