Transport mechanisms for synoptic , seasonal and interannual SF 6 variations and “ age ” of air in troposphere

We use an atmospheric general circulation model (AGCM) driven chemistry-transport model (ACTM) to simulate the evolution of sulfur hexafluoride (SF 6) in the troposphere. The model results are compared with continuous measurements at 6 sites over 71 ◦ N–90 S. These comparisons demonstrate that the ACTM simulations lie within the measurement uncertainty over the analysis period (1999– 2006) and capture salient features of synoptic, seasonal and interannual SF 6 variability. To understand transport timescales of SF 6 within the troposphere, transport times of air parcels from the surface to different regions of the troposphere (“age”) are estimated from a simulation of an idealized tracer. The age estimation error and its sensitivity to the selection of reanalysis meteorology for ACTM nudging or the tracer transport by deep cumulus convection as represented in the model are discussed. Monthly-mean, 2-box model exchange times ( τex) are calculated from both the observed and simulated SF 6 time series at the 6 observing sites and show favorable agreement, suggesting that the ACTM adequately represents large-scale interhemispheric transport. The simulated SF 6 variability is further investigated through decomposition of the mixing ratio time-tendency into advective, convective, and vertical diffusive components. The transport component analysis illustrates the role of each process in SF6 synoptic variability at the site level and provides insight into the seasonality of τex . Correspondence to: P. K. Patra (prabir@jamstec.go.jp)


Introduction
Sulfur hexafluoride (SF 6 ) represents a powerful tracer of atmospheric transport (Maiss et al., 1996).Emitted at the earth's surface as a byproduct of industrial activity (primarily as a dielectric material in electrical switching equipment), SF 6 has no known production or loss in the troposphere or stratosphere (Ravishankara et al., 1993).Moreover, SF 6 emissions have no significant seasonal cycle.These characteristics render SF 6 ideal for illustrating the effects of different transport processes (e.g.mean meridional transport, convection, vertical diffusion) on the tropospheric distribution of trace species.Like other minor tropospheric constituents, including 85 Kr and chlorofluorocarbons (CFCs) (Jacob et al., 1987;Prather et al., 1987), the potential for SF 6 to elucidate tropospheric transport mechanisms has led to its incorporation into model studies, especially as a metric for model intercomparison (e.g.Denning et al., 1999;Law et al., 2008).
While early applications of SF 6 to transport model diagnosis tended to focus on large-scale features such as the interhemispheric gradient, more recent observation-model comparisons of SF 6 (e.g.Peters et al., 2004;Gloor et al., 2007) have demonstrated that current generation transport models are able to resolve finer-scale latitudinal gradients as well as location-specific vertical distributions.The increased capacity of forward transport models to replicate observed SF 6 can be attributed in large part to improvements in the models, including the treatment of diurnally varying planetary boundary layers (PBLs); the refinement of cumulus convective tracer mass transport; and the use of meteorological fields at higher spatial and temporal resolutions to drive advection (see, e.g.Rind et al., 2007 and references therein).1210 P. K. Patra et al.: SF 6 , age of air, transport processes in troposphere In the present study, we analyze the characteristics of SF 6 transport as simulated by the Center for Climate System Research/National Institute for Environmental Studies/Frontier Research Center for Global Change (CCSR/NIES/FRCGC) ACTM (hereafter, ACTM for brevity) on daily-to-interannual time scales and local-tohemispheric space scales.The investigation of ACTM transport variability across multiple temporal and spatial scales is crucial for assessing the model's overall validity for applications such as source/sink inversions of atmospheric CO 2 .For example, the inclusion of near-surface land region CO 2 measurement sites in CO 2 inversions may introduce uncertainties into estimated carbon fluxes because of deficiencies or errors in the representation or sampling of such sites in coarse resolution global transport models (Patra et al., 2006(Patra et al., , 2008)).The present analysis is motivated in part by a recent high-frequency model-observation intercomparison study, TransCom-4 (see Law et al., 2008;Patra et al., 2008) focusing on hourly and daily average CO 2 variations.An important conclusion of TransCom-4 is that the 25 participating models show some skill in simulating the observations, considering uncertainties or errors in CO 2 flux distribution and intensity.However, in that study, detailed analysis of SF 6 as a potential constraint on high-frequency, transport-induced variability was not performed; moreover analysis was further restricted to the period of 2002-2003.Because there are relatively few sites for which high frequency SF 6 data exist, we focus on the continuous observations of SF 6 from 6 sites (WDCGG, 2008; see also Table 1 for a summary of site information), namely Point Barrow (BRW), Schauinsland (SCH), Niwot Ridge (NWR), Mauna Loa (MLO), Samoa (SMO) and South Pole (SPO).
Apart from validating the ACTM's capacity to replicate the SF 6 records from the 6 observing sites, we further examine the roles of different transport processes in explaining variations of SF 6 as simulated by the ACTM.In particular, model simulations are used to analyze the vertical and horizontal features of tropospheric SF 6 transport.To quantify the vertical transport, we make use of an approach com-monly applied to the stratosphere (see, e.g.Bischof et al., 1985;Hall and Plumb, 1994) to estimate the age of air in the troposphere.While the age of stratospheric air is usually defined as the time since leaving the tropopause layer, the age of tropospheric air is defined here as time since leaving the earth's surface.The large-scale horizontal transport, readily characterized in terms of an interhemispheric transport time (e.g.Jacob et al., 1987), is computed for both observed and simulated SF 6 data.A more detailed component analysis of ACTM-simulated SF 6 transport elucidates the transport pathways involved in both regional and large-scale atmospheric transport and how these pathways vary on different timescales.

AGCM-based Chemistry Transport Model (ACTM)
The CCSR/NIES/FRCGC AGCM is nudged with reanalysis meteorology using a simple Newtonian relaxation method (Hoke and Anthes, 1976) for driving the chemical tracer transport.The nudging technique forces the AGCM-derived meteorology towards the reanalyzed horizontal winds (U and V components) and temperature (T) with relaxation times of 1 and 2 days, respectively, for every 6-h interval of the reanalysis (except the top and bottom model layers).For the simulations discussed here, we use reanalysis products (U, V, T) from National Center for Environmental Prediction (NCEP)/DOE AMIP-II Reanalysis (Kanamitsu et al., 2002; hereinafter referred to as NCEP2) and European Centre for Medium-Range Weather Forecasts (ERA40, Uppala et al., 2005).To compute the heat and moisture exchange fluxes at the earth's surface, the AGCM is also supplied with interannually varying monthly-mean sea ice and sea-surface temperature (SST) fields from the Met Office Hadley Centre observational datasets (Rayner et al., 2003).This forward transport model has been adapted for simulations of greenhouse gases (CO 2 , CH 4 , N 2 O, CFCs, SF 6 etc.) that have Fig. 1.Distribution of SF 6 emissions (left column; in pg-SF 6 m −2 s −1 ; 1 pg=10 −12 g) at T42 resolution (∼2.8 • ×2.8 • latitude-longitude), with an annual total emission of 8.8×10 9 g-SF 6 yr −1 for the year 2002.The measurement site locations are marked as numbers (1-6), with abbreviated names given at bottom-left (see Table 1 for site details).The MLO marker is displaced to the northeast of the site so that the island and emission grid are visible.In the right column are magnified views of the SF 6 emissions around the two continental sites (NWR and SCH).
negligible photochemical production or loss in the troposphere and stratosphere.The ACTM is run in "online" transport mode with both the AGCM meteorology and chemical tracers simulated at the same integration timestep (∼20 min).Although computationally expensive, running the ACTM in online mode offers some advantages over less demanding "offline" chemistry-transport models.Specifically, over the period of the analysis, the AGCM generates a detailed and consistent meteorology as represented by the grid-and subgrid-scale processes, surface processes (e.g.PBL height and mixing), above-PBL dynamics (e.g.convection) and interhemispheric gradients.
The basic physical and dynamical features of the ACTM have been described in (Hasumi et al., 2004).Advective transport of moisture and tracers is obtained from a 4th order flux-form advection scheme using a monotonic Piecewise Parabolic Method (PPM) (Colella and Woodward, 1984) and a flux-form semi-Lagrangian scheme (Lin and Rood, 1996).Mass fluxes around the polar caps are calculated with a semi-Lagrangian scheme in polar stereo projection.Subgridscale vertical fluxes of heat, moisture, and tracers are approximated using a non-local closure scheme based on Holtslag and Boville (1993) used in conjugation with the level 2 scheme of Mellor and Yamada (1974).The cumulus parameterization scheme is based on Arakawa and Schubert (1974) with some simplifications described in Numaguti et al. (1997).The updraft and downdraft of tracers by cumulus convection are calculated by using the cloud mass flux estimated in the cumulus parameterization scheme.

Fluxes and data for SF 6 , and curve fitting
We have simulated SF 6 for overall evaluation of regional and interhemispheric scale atmospheric tracer transport by the ACTM.The SF 6 emissions are taken from the Emission Database for Global Atmospheric Research (EDGAR) (Olivier and Berdowski, 2001), with the yearly emission change scaled to the global SF 6 growth rate estimated from the Earth System Research Laboratory/National Oceanic and Atmospheric Administration, USA (ESRL/NOAA) observations (Geller et al., 1997).The observed concentrations of SF 6 at daily time intervals are taken from the NOAA/ESRL halocarbon in situ network (Butler et al., 2004) and the Air Monitoring Network of Umweltbundesamt, Federal Environmental Agency, Germany (UBA/FEA).The geographic distribution of SF 6 emissions at the ACTM's horizontal www.atmos-chem-phys.net/9/1209/2009/P. K. Patra et al.: SF 6 , age of air, transport processes in troposphere resolution is displayed in Fig. 1 along with the locations of the 6 continuous measurement sites.There is some information loss with the conversion of the raw emission data (at 1 • ×1 • resolution) to T42, e.g.emissions from the Korean peninsula and Japan are not easily distinguishable, and similarly high emissions from Europe and the United States are smeared out.These errors are denoted hereafter as site representation errors.
Fitted curves and long-term trends for each daily average time series are derived using a Butterworth filter of order 16 (Nakazawa et al., 1997) with a cut-off length of 24 days.The time series are then decomposed into seasonal cycles (data or fitted curves -long-term trends), growth rates (time derivative of the long-term trends) and synoptic variations (datafitted curves).Statistical assessments of modeled and observed data are determined using original values (without fitting) but excluding missing data periods.

"Age of air" tracer
The mean "age of air", defined as the time required for an air parcel to transit from the earth's surface to the layers above (Kida, 1983), is calculated as the difference between surface and upper air concentrations normalized by the concentration increase rate at the surface using a Green's function method (Hall and Plumb, 1994).The Green's function is estimated from the simulation of an idealized transport tracer with uniform surface fluxes, linearly increasing trend, and no loss in the atmosphere.Simulation of all tracers was initialized on 1 January 1960 (with the ACTM nudged to ERA40 meteorology; data available up to 2002), and the analysis presented here covers the period of continuous SF 6 observations (ACTM nudged to NCEP2 meteorology; data available from 1979).For the overlapping period of simulations (the 1990s), no significant differences in SF 6 simulations have been found for the two reanalysis products.In a more general sense, systematic differences in simulated synoptic variations of longlived tracers (e.g., CO 2 ) using various analysis and reanalysis products in transport models are not evident (see also Patra et al., 2008).

Two-box model of interhemispheric (IH) exchange time (τ ex )
The interhemispheric exchange time (τ ex ) has been widely used to diagnose large-scale model transport properties and has been previously estimated from both measured and modeled trace constituents (Jacob et al., 1987;Maiss et al., 1996;Levin and Hesshaimer, 1996;Denning et al., 1999;Geller et al., 1997).τ ex is computed from a simple mass balance equation using the mean mixing ratios and growth rates in the Northern Hemisphere (NH) and Southern Hemisphere (SH) (see, e.g.Prather et al., 1987;Jacob et al., 1987): Here, c n and c s are the average mixing ratios for the NH and SH, respectively; E n and E s are hemispheric total tracer emission; α is an emission-to-mixing ratio conversion factor; c n−s is the north-south IH difference in tracer mixing ratio; and τ a is the tracer's atmospheric lifetime.Since SF 6 has no known loss in the atmosphere up to about 50 km and its lifetime as estimated to be about 3200 years (Ravishankara et al., 1993), the last term in both equations can be neglected.Elimation of α from Eqs. ( 1) and ( 2) gives an expression for τ ex : All terms on the right hand side in Eq. ( 3) are estimated directly from either measured or modeled SF 6 time series.We used the fitted time series and monthly average values for calculation of monthly-mean τ ex .

Separation of mass transport due to advection, convection and vertical diffusion
Tracer transport in the ACTM comprises numerical solution of the continuity equation that describes the mass conservation for a chemical tracer in the atmosphere: where ∇ is the 3-dimensional divergence operator and F is the tracer mass flux (including both the direct large-scale advective effect and parameterized convection and diffusion terms).P and L are production and loss in the atmosphere, respectively; for SF 6 , L is neglected in all model layers, and P =0 at all layers but the surface layer, where the emission occurs.Thus, in the interior of the troposphere, the flux divergence term (∇•F ) associated with various transport processes drives the net time-tendency in c.We decompose the flux divergence term into three components in this study: (1) advection by grid-scale air flow calculated from the flux-form transport scheme (Lin and Rood, 1996); (2) lifting through cumulus convection as parameterized by the Arakawa and Schubert (1974) scheme; and (3) vertical diffusion calculated using the turbulent closure method of Mellor and Yamada (1974).The component tendencies, which are positive (negative) if a model grid gains (loses) mass, are utilized to examine the impact of different transport mechanisms on the temporal evolution of SF 6 .

Results and discussions
3.1 Model-observation comparison of SF 6 time series   1, indicate little disagreement between the observed and modeled latitudinal gradients, especially over the Pacific Ocean where the remote background sites BRW, MLO, and SMO are located, and between the continental sites in North America (NWR) and Europe (SCH).In fact, differences between the simulated and observed time series fall within the measurement uncertainty of 0.04 ppt.The amplitude and phasing of the high-frequency variability at all sites are also generally well captured.Some model-observation mismatches do occur, likely arising from spatial representation errors in SF 6 emissions and ACTM transport as well as measurement quality issues.
Over the 1999-2006 period, some changes in observationmodel agreement at high-frequencies are evident as instrumentation was altered.For example, marked improvements at SMO and MLO are evident around mid 2000 and mid 2002, respectively, following a change in electron capture detector (ECD).The observed synoptic or finer time scale variability decreased significantly after the detector change, resulting in fluctuations more comparable to the simulations.Also, an ECD change in early 2004 clearly brings the model and observations into closer agreement at SPO, although the agreement deteriorates later in the record.At BRW the modeled concentrations were lower by about 0.05 ppt in 1999 and are higher by about 0.07 ppt in the recent years compared to the observations.Such systematic differences, reflected in the BRW growth rate, may stem from errors in regional emission trends: while the rate of emissions increase in our simulation is globally uniform, the actual trends are likely regionally heterogenous, e.g., larger in rapidly developing countries than in the developed countries.The overall agreement in the latitudinal gradients (Fig. 2, left column) between the simulations and measurements highlights realistic representation of the interhemispheric scale model transport in the ACTM.
The seasonal cycle and its interannual variability (IAV), as derived using the digital filtering method, are fairly well captured by the ACTM simulation at all remote background sites (Fig. 2g, j-l).By contrast, at the continental sites (NWR and SCH), there is no clear seasonality in SF 6 as the IAV is quite large.While the timings of SF 6 increases/decreases generally match between the observed and modeled data, their amplitudes are generally underestimated in the latter, likely as a result of the ACTM's coarse horizontal resolution.Since the SF 6 emissions input to the ACTM lack seasonality, the seasonal variations of simulated SF 6 are driven entirely by atmospheric transport.Poorer agreement at the 2 continental sites compared to the 4 remote sites presumably arises from site representation error within the ACTM's coarse horizontal resolution, i.e. the gridpoint at which the model is sampled may not adequately represent the conditions seen locally at the observing site.The degree to which this type of error contributes may further depend on how the features of local meteorology interact with nearby source emissions (with the latter generally negligible for remote observing sites).
We further illustrate the model-observation agreement of SF 6 synoptic variations at 5 sites (Fig. 3).Here, SPO is excluded because of low signal in the simulation (amplitude variability of ∼0.005 ppt or less).As a measure of the phase similarity of modeled and observed SF 6 , Pearson's correlation moment (r) values of daily mean SF 6 variability are significant at the 95% confidence interval only at SCH and SMO (r>0.28, for N >600 data points in Student's 2-tailed test).Normalized standard deviations (NSDs; =SD observation /SD model ; =1 when modeled and observed variability amplitudes are equal regardless of the phase) are systematically >1 at all sites, indicating smaller variability in the model than in the observations.In the daily-averaged observed time series (thin lines), there are many positive and negative spikes lasting only for a day.Since the ACTM is likely incapable of simulating such sharp spikes because of the smoothed emissions, we also show 5-day (pentadal) running means (thick lines).Visually, the match between simulated and observed synoptic variability improves considerably for the pentadal time series, with both correlations and NSDs improved when 5-day moving window averages (MWAs; 1 average value for each 5-day block in the time series) are considered.Use of 5-day MWAs does not introduce spurious serial autocorrelations into the time series (as is the case for running means) although the length of the time series is reduced by factor of 0.2.Overall, the 5-day MWA correlations at all sites except BRW are found to be statistically significant at the 95% confidence interval (r>0.41, for N∼125 data points), while the NSDs are closer to 1 than those estimated from 1-day averages.
From the model-observation comparison, we conclude that the nudged ACTM at T42 horizontal resolution adequately simulates SF 6 interhemispheric mixing ratio gradients, seasonal cycles at remote sites, and synoptic variations at 5-day time scales.Of course, we have also identified some limitations of the coarse resolution global models for simulating the highest-frequency fluctuations of SF 6 .In the remainder of this study, we emphasize the simulated transport characteristics that are likely to be significant for SF 6 distributions on regional to global scales and compare some transport diagnostics estimated from observations and model simulations.
3.2 Mean "age of air" in the troposphere and its relation to tracer transport One of the ways to analyze model transport properties is the age distribution in troposphere, a concept that has been widely applied to tracers in the stratosphere (Bischof et al., 1985;Hall and Plumb, 1994).Here, the mean age has been estimated using the idealized tracer simulation described in Sect.are much less than typical values estimated for the stratosphere (i.e., 1-5 years; Bischof et al., 1985), where transport is much slower.The transport model reproduces the upper tropospheric "mixing barrier" around 30 • in both hemispheres in most longitude bands and seasons (Erukhimova and Bowman, 2006).The upper tropospheric locations of the steepest meridional age gradients are coincident with the rapid change in zonal winds (black contours) corresponding to the equatorward edges of the subtropical jet steams at approximately 30 • latitude in both Hemispheres and between ∼500-150 mb in the vertical.The steep gradient in the ageof-air in the subtropical upper troposphere may be associated with weak breaking of Rossby waves around the level of the θ=350 K potential temperature isotherm (Postel and Hitchman, 1999), as the strong westerlies generally suppress Rossby wave breaking and thus slow mixing.The effect of this mixing barrier has also been discussed in the context of atmospheric CO 2 (Miyazaki et al., 2008).
For the longitude interval spanning the core of the Indian monsoon zone (Fig. 4b), the mixing barrier (defined by large age gradient) lies further north in NH summer because of the extensive cumulus convection near the Himalaya/Tibetan plateau region between 30-40 • N.This region bears the signature of younger air at relatively higher altitudes (∼150 mb or higher) in the NH compared to the SH at similar latitudes.Over the Pacific region (Fig. 4c, d), the zone of maximum convection and the upper tropospheric mixing barrier oscillate along with seasonal changes in solar insolation in the absence of heterogeneous orography.Note also that latitudinal gradients in isochrones (contours of constant age) are much steeper in the middle-upper troposphere (∼400 mb and above) of the Indian monsoon region in July (Fig. 4b) compared to January conditions or the mid-Pacific in either January or July (Fig. 4a, c and d).  5.This result is produced using the ACTM nudged to ERA40 meteorology.
Our estimation of the mean age in the tropical upper troposphere, 20-30 days (at the 150 mb model layer; longitude: 87-80 • W; latitude: 7-11 • N; time: January-February), is within the range of the mean age of air (26±3 days) entering the tropical tropopause as estimated from CO 2 measurements over Central America (latitude: <11 • N; height: 14-18 km; time: January-February; Park et al., 2007), although a more extensive data set is necessary to validate the age ranges implied by the modeled age distribution.The smallest vertical age gradients within the troposphere occur mainly where vertical θ gradients (Fig. 4e, f; black contours) are small, i.e.where dynamically or thermodynamically unstable conditions prevail.In the ascending zones of the Hadley circulation, the energy required for air parcels to cross isentropes is mainly supplied through diabatic convective heating/cooling.(The role of individual mechanisms in SF 6 transport is elabo-rated in Sect.3.4).The sharp rise in age above the tropopause (θ >400 K and 380 K in the tropics and midlatitude, respectively; note the unequal contour interval with height) is associated with the slower cross-isentropic transport under the strong thermal stratification of the lower stratosphere (i.e., − ∂θ ∂P large).Thus, within the tropics, our age distribution calculation supports the conjecture of a vertical mixing barrier in the altitude range of 14 km and tropopause or potential temperature range ∼360-390 K (Folkins et al., 1999), with age increases from a few days below to >100 days above.
We checked the sensitivity of ACTM-simulated age of air by: i) nudging to both NCEP2 and ERA40 meteorology; and ii) disabling tracer transport via parameterized cumulus convection.In the latter (denoted hereafter as NoCumTransp), tracer transport associated with parameterized deep cumulus convection is switched off while the underlying AGCM Fig. 5. Latitude-pressure cross-sections of mean age of air as simulated by the ACTM without transport by parameterized cumulus convection for the age tracer (shaded, NoCumTransp case), differences in age between two ACTM simulations by nudging to ERA40 and NCEP2 reanalyzed meteorology (in %; black contour; ERA40-NCEP2), and potential temperature (in K; light blue contour).The seasons (top row: January; bottom row: July) and the locations of longitudinal cross-sections (left column: 70-90 • E; right column: 170-190 • E) are same as those in Fig. 4a-d.
meteorology is unchanged and nudged to ERA40 meteorology.Figure 5 shows an overall small impact (mostly within 5%; white area) of switching reanalysis meteorology, although there are localized maxima of ∼20% near the latitudes of peak convection.(These localized differences are reduced in the annual mean).On the other hand, rather large differences in age are obtained in the absence of tracer transport by deep cumulus convection.Notably, the age values in the height range of 250-100 mb increase by >60% in the equatorial upper troposphere and up to 10% in the high latitudes in the NoCumTransp case (Fig. 5) relative to the control case (Fig. 4).For these simulations, the effect of cumulus convective tracer transport on monthly mean age is effectively restricted to tropospheric layers and in the tropical lower stratosphere (T-LS; 100-70 mb, 30 • S-30 • N).The age difference in the T-LS region can be up to 20%, i.e. an absolute difference as large as 30 days.This difference in absolute age is propagated throughout the stratosphere, but it is not significant in a relative sense because of the large age values.
To illustrate the interplay between circulation and age of air, we show the latitude-longitude distributions of age values, vertical/pressure velocity (ω), and horizontal winds at 200 mb in Fig. 6.In the control case, (panels a, b), the regions of youngest air (darkest blue shading) at this height are always colocated with the strongest grid-scale upward motion (red contours), particularly when the age signatures indicate tropospheric air.In the NoCumTransp case, there is a weak association between small age values and upper level horizontal wind divergence, with non-neglible gradients only near the Asian monsoon/western Pacific warm pool region (where convection is strongest) (Fig. 6c, d).These results underscore the importance of parameterized convective tracer transport for species with atmospheric residence times comparable to the age of air in the troposphere: chemical constituent with atmospheric residence times less than 30 days (i.e., the approximate age of air at 200 mb for NoCum-Transp case) would be unable to reach the upper troposphere in the absence of transport via parameterized cumulus convection.Even in the control, the transport efficiency of chemical species with short residence times (days to weeks) from the earth's surface to various parts of the upper troposphere/lower stratosphere (UT/LS) will strongly depend on the location of emissions and season.
In ACTM framework, cumulus convective transport in the tropics is far more effective in vertical mass delivery from earth's surface to the upper troposphere at short timescales than is the large-scale advective transport shown by the winds fields.This is presumably because of the lack of a connection between the mixing in lowermost troposphere by vertical diffusion and relatively strong winds divergence in the upper troposphere.Note also that the large-scale monsoon anticyclones, centered near 130 • E, 15 • S in January and 90 • E, 20 • N in July, produce sharp age gradients by restricting air mass transport across the southern and northern edges, respectively.At the same time the aniticyclones transport low age air across the equator.
For the long-lived trace constituents like SF 6 , the link between increasing age of air in the (lower) stratosphere and decreasing mixing ratio is clearly seen (Fig. 4), as the isochrones and SF 6 isopleths are parallel to each other for the altitude range where age is greater than 100 days.In the NH troposphere for latitudes poleward of 40 • N, age is increased and SF 6 mixing ratios decrease with altitude, particularly during the NH winter when convective transport is weaker.The SF 6 isopleths tend to become vertical on encountering the NH mixing barrier.In the tropics SF 6 is always well mixed vertically, and a strong latitudinal gradient is established around the subtropics.The meridional mixing ratio gradient weakens in the SH mid-and high-latitudes.These characteristics of SF 6 make it a powerful tracer for studying interhemispheric exchange in the troposphere as well as large-scale dynamical processes in the lower stratosphere.

Interhemispheric exchange time
Monthly-mean τ ex estimated from both ACTM-simulated and observed SF 6 data appear in Fig. 7.For comparative purposes, hemispheric average mixing ratios were calculated from a few different combinations of the 4 NH and 2 SH sites.For case 1 (Fig. 7a, light blue), all sites in each hemisphere were used in the averaging, while in case 2 (black) only the remote sites (BRW, MLO, SMO, and SPO) were used.The τ ex values are lower in case 2, which reflects exclusion of the non-background, high-concentration sites at NWR and SCH.Somewhat smaller values of τ ex are evident for case 2 (black), in which only the SF 6 mixing ratios at the remote sites were used in averaging.For case 3 (dark blue), which used MLO and SMO concentrations as proxies for the NH and SH hemispheric averages, the lowest mean exchange times were obtained, resulting from the relatively small difference in mixing ratio between these close proximity sites, although the variability is clearly the largest.An additional case (4, black line; Fig. 7b), calculated from the ACTM simulation only using the average surface mixing ratio for all NH and SH gridpoints, produces similar τ ex estimates to case 2, suggesting that SF 6 hemispheric averages consisting of BRW and MLO in the NH and SMO and SPO in SH are representative of the whole hemispheric averages.
Values of annual mean τ ex , computed for the various cases, are summarized in Table 2 along with several estimates from prior studies.Both the observed and simulated results obtained in cases 1 and 2 compare well with the estimates of Bottom panel (b); black line: Case 4, Red line: Case 5 shows exchange times estimated using model based the hemispheric mean mixing ratio at the surface and total mass in the troposphere (surface to 100 mb).Note the difference between two y-axis scales.Levin and Hesshaimer (1996) and Geller et al. (1997), despite the interannual variability in τ ex .We point out that some differences between different studies may be tied to the selection of station sets for estimating the hemispheric average mixing ratios.The ACTM-derived τ ex are also in the range of earlier SF 6 model-based estimates of 0.76-1.97years obtained from multiple transport models (Denning et al., 1999).However, the more recent TransCom-4 intercomparison (Law et al., 2008) demonstrates much tighter agreement in SF 6 IH gradients.Specifically, 17 out of 20 global models examined produced IH gradients in the range of 0.21-0.29 ppt (0.24 ppt for the ACTM at T42) compared to an observed value of 0.23 ppt (for 2002), implying less spread among the TransCom-4 τ ex estimates.The closer agreement among models in the more recent intercomparison likely reflects improvements in forward transport modeling; moreover, all TransCom-4 models used analyzed meteorology to drive model transport, while in the Denning et al. intercomparison, modeling groups chose the meteorology divers independently (e.g.reanalysis or GCM winds).For case 4 as well as an analogous calculation using the total 3D mass distribution within the troposphere (case 5), annual-mean τ ex values of 1.37 (case 4) and 0.7 (case 5) years are obtained.The case 5 estimate is within the range of 0.55-1.26years noted by Denning et al. (1999) and more closely agrees with recent model estimates of ∼0.7-1.20 years (ref. Table 2).
The monthly-mean τ ex manifest pronounced seasonal cycles in all cases considered here.In particular, the seasonality is dominated by a semi-annual periodicity for cases 4 and 5 (Fig. 7b).The primary and secondary maxima, corresponding to relatively slow IH exchange rates, are found during April and October, respectively, and minima occur during January and July (case 4; note the reversal of the primary and secondary maxima in case 5.) This seasonality is similar to that described in Lintner et al. (2004) but is distinct from the seasonality shown in Levin and Hesshaimer (1996).Broadly, such seasonality can be understood in terms of the seasonal variations in the Hadley circulation and the meridional displacement of Intertropical Convergence Zone (ITCZ), with the solisticial seasons of strong and equatorially asymmetric Hadley cells producing strong cross-equatorial flow and more rapid exchange of tracer mass (Lintner et al., 2004).In fact, idealized model studies suggest that seasonal oscillation of the Hadley circulation may be responsible for a significant portion of IH mixing (Bowman and Cohen, 1997).A more complete picture of τ ex seasonality in the ACTM is developed below.
The model-observation mismatches in April are mainly caused by the variability in the Samoa (SMO) site data.This site is situated between the western Pacific ITCZ and South-Pacific Convergence Zone (SPCZ) and is significantly influenced on interannual timescales by El-Nino Southern Oscillation (ENSO)-induced transport variability.For instance, we find the smallest dC s /dt values at SMO during 2001 and 2002 April, which is coincident with a period of sharp increase in Multivariate ENSO Index (MEI).Low mixing ratios at SMO during El Niño events have been attributed to locally-enhanced southeasterly flow at the surface associated with a northward shift of the SPCZ (Hartley and Black, 1995).Omitting anomalous SMO outliers in the averaging yields observational (simulated) τ ex values of 2.7 (2.3), 2.1 (1.6) and 1.8 (1.5) years for Case 1, 2 and 3, respectively.

Analysis of SF 6 transport pathways and seasonality in τ ex
Tropospheric tracer transport in the ACTM consists of advection, convection, and vertical diffusion as described in Sect.2.5. Figure 8 shows latitude-pressure distributions of Eulerian mean SF 6 mass transfer rates of each of these components.Generally, the advection term dominates in most parts of the troposphere, with the intensity maximized (≥1 pptm/mon) over 30 • S-60 • N, where the meridional gradient in SF 6 is the largest (black contours in Fig. 8a-d).tropical and sub-tropical latitudes where it efficiently transfers mass from surface levels to the upper troposphere (see also Donner et al., 2007).Vertical diffusion plays a major role only near source emission regions (land areas between 20-60 • N latitude; see Fig. 1) where low-level loadings of SF 6 are maximized.Figure 8e-h and f-i indicate that the tracer transport associated with parameterized convection is limited to the top of the Hadley cells, which has been suggested as a cause for the vertical mixing barrier in the tropical upper troposphere (see also Folkins et al., 1999).The average upper height limit for convective detrainment is located around 14 km, coinciding with the largest age gradient (>25 day/km) below the tropical tropopause (altitude range of 14-16 km).
To summarize the typical transport pathway connecting the source regions to remote portions of the troposphere, SF 6 emissions are initially mixed through vertical diffusion near the source regions (shown in red in Fig. 8i-l).Except for the emission layer (model level #1), the model levels up to ∼500 mb between roughly 30-60 • N gain SF 6 through vertical diffusion.From here, SF 6 is lofted deeper into the upper troposphere by convective transport from the lower troposphere (Fig. 8e-h).In regions of strong convection near the equator, the spacing between isentropes is reduced, facilitating efficient and strong vertical mass transport in this part of the troposphere.Generally, increases to upper tropsopheric SF 6 mixing ratios via convective transport are restricted to the north side of the upward branch of the Hadley circulation; to the south, upper tropospheric SF 6 values are reduced by convective uplift of low tropospheric air masses since the near-surface SH air is relatively deficient in SF 6 .Negative values of the Eulerian mean pressure velocity (ω), contoured in the right column of Fig. 8, correspond to largescale ascent, with the strongest upward motion of the Hadley circulation located around 15 • S (15 • N) during boreal win-ter (summer).Note that the Eulerian mean circulation does not accurately represent the Lagrangian (tracer) circulation in the extratropics (see Bowman and Carrie, 2002 for details).Upon its delivery to the middle and upper troposphere, SF 6 is advected longitudinally by strong zonal winds.Miyazaki et al. (2008) discuss detailed mechanisms of tropospheric CO 2 transport, with the results presented here supporting their conclusions.Relative to CO 2 , the SF 6 -based analysis is easier to visualize because of simpler emission statistics (i.e.well-defined source emission regions with no seasonality), and the simulated data can be readily validated against observations through simple transport diagnostics like τ ex .
The Hadley circulation and its associated meridional winds advect SF 6 into the Tropics.The seasonality of τ ex mirrors seasonalities seen in the convective and advective transport components of SF 6 near the equator.Figure 8e and g indicates that the maximum SF 6 transport to the upper troposphere associated with parameterized convection occurs near the equator, while Fig. 8a, c shows upper tropospheric meridional spreading of SF 6 by advection across mixing ratio isopleths.This situation accounts for the faster tracer mass transport across the equator during January and July.On the other hand, during April and October, the regions of strong advective transport in the tropical upper troposphere do not cross the SF 6 mixing ratio isopleths.The apparent isolation of SF 6 transport on either side of the equator during the equinoctal seasons is consistent with the larger 2-box exchange times estimated from measured and simulated SF 6 .Results of a recent study by Aghedo et al. (2008) suggest that the seasonal asymmetry in τ ex should be opposite for tracer emissions localized to the SH, with the primary (secondary) maximum in October (April), although the reason for such sensitivity is unclear.
The location of mixing barriers, particularly over the NH, determine the intensity of meridional tracer advection.When  (e-h) (middle column): convective transport, i-l (right column): vertical transport by diffusion (shaded).The black contours shown denote SF 6 mixing ratio (a-d; unit ppt); potential temperature (e-h; unit K); and vertical velocity ((i-l) unit: mb hr −1 ).These contours are provided for interpreting relationships between the meteorological conditions and transport components as discussed in Sect.3.4.
the NH mixing barrier is located at about 15 • N during January (at the downwelling branch of the Hadley circulation), the cross-equatorial mass transfer in the upper troposphere is minimized and intensified in the lower troposphere (below 700 mb).In fact, the strong subsidence leads to reduction (enhancement) of high SF 6 in the upper (lower) troposphere.The situation is opposite during July, when the NH mixing barrier is located over 40 • N (the region of strongest SF 6 emission).The upwelling branch of the Hadley circulation (between 0-25 • N latitude) transfers SF 6 rich air to the upper troposphere from lower troposphere, which is then transported to the SH by the meridional branch of the Hadley cell in the height range of 150-250 mb.

Contributions of transport pathways to SF 6 synoptic variations at surface sites
The transport components for the 5 continuous monitoring sites excluding SPO are presented in Fig. 9.It is clear that, depending on site location with respect to major emission regions and atmospheric transport regimes, the dominant mechanism(s) for SF 6 variability differs.The tendency components also exhibit distinct seasonality at some sites.At BRW, transport by advection (red) dominates in January when the winds are predominantly northerly.By contrast, during February-March, under the influence of stronger southerly flow, both advection and vertical diffusion (blue) are comparable and generally of opposite phase, thereby attenuating the total synoptic-scale variability.At SCH, the antiphasing of synoptic SF 6 transport via advection and vertical diffusion is clearer.Advective and vertical diffusive tendencies are largest during the boreal winter, when PBL ventila-tion is weak, and smallest during the boreal summer, when PBL ventilation is strong: at SCH (located at 1205 m altitude), the maximum PBL height simulated by the ACTM is typically 400 m during winter, but it is as high as 1500 m during the summer.It is important to note that the simulated SF 6 at SCH was sampled at model level 3, even though the site is actually located on a mountain top, i.e. locally the earth's surface, which strictly corresponds to model level 1 in the sigma-pressure coordinate.At the other continental site, NWR, the synoptic variability of advective transport is much larger than the vertical diffusive component; however, the magnitude of variability at NWR is several times smaller than that at SCH because of differences in SF 6 emission in their proximity (cf.Fig. 1; right column).At the remote MLO site, the synoptic variation in SF 6 is almost entirely driven by advection.Only at the remote marine site SMO is the convective transport (in cyan) of comparable magnitude to the advective transport at synoptic timescales.Interestingly, these two components are out of phase, with the tendencies changing sign seasonally, e.g.advection (convection) shows positive (negative) tendencies during January-February and negative (positive) tendencies during March-September.These features can be understood from changes in SF 6 vertical structure and the Hadley circulation as seen in Fig. 8.However, it is worth noting that an accurately simulated SF 6 distribution alone does not imply that the relative influences of advection and convection are accurately modeled in ACTM.Checking some fundamental model properties against observations provides some confidence in the fields used to drive tracer transport.For example, comparison of the simulated outgoing longwave radiation (OLR) aginst the NOAA non-interpolated OLR (source: www.cdc.noaa.gov) on daily and monthly timescales suggests reasonable simulation of the intensity and location of tropical convection zones.Additionally, a multi-tracer (CO 2 , CO, O 3 ) analysis using nudged and free (without U, V, T nudging) meteorology showed closer agreements between the simulated and observed daily-averaged tracer and OLR variations for the nudged ACTM results compared to the free running ACTM (Patra et al., 2007).Further relevant observables to identify and validate the relative partitioning of advection and convection in models, e.g., vertical profiles of trace species with varied atmospheric residence time and well constrained surface fluxes, are clearly desirable.

Conclusions
The CCSR/NIES/FRCGC AGCM-driven online transport model (ACTM) has been applied to simulate atmospheric SF 6 .These simulations prove comparable to the observed behavior at six continuous measurement sites, with mixing ratios increasing by about 1.6 and 1.8 ppt and at SPO (90 • N) and BRW (71 • S), respectively, between 1999 and 2006.Both the modeled and measured time series are decomposed into synoptic variations, seasonal cycles, and growth rates, with the synoptic and seasonal variations of simulated and observed SF 6 data correlating significantly at most sites during the analysis period.The ACTM is thus a useful tool to study daily-to yearly-transport mechanisms in the troposphere.
An illustrative transport metric for the ACTM discussed here is the tropospheric "age of air", with age values agreeing with available observation-based estimates of order 20-30 days in the tropical upper troposphere.However, more observations are required for spatially extensive validation of tropospheric age estimates.We have shown small error could arise in age estimation of tropospheric air due to nudging the ACTM with NCEP2 or ERA40 meteorology.On the contrary, removing the effect of cumulus convection on tracer transport in the ACTM results in significant lengthen-ing of age estimates; thus, the accuracy of cumulus convection parameteriztions and their impact on tracers have large implications for the distributions of chemical sepcies in the upper troposphere/lower stratosphere.Additionally, the 2box interhemispheric exchange times (τ ex ), estimated using observed (∼1.3 year) and modeled (∼1.2 year) surface SF 6 mixing ratios, agree well with prior estimates based on earlier simulations/observations.While the exchange time is found to be dependent on the details of the hemispheric mean SF 6 mixing ratio estimates, the combination of BRW and MLO in the NH and SMO and SPO in the SH reasonably represents their respective hemispheric mean surface mixing ratio for SF 6 .The seasonality in τ ex is shown to arise primarily from the seasonal migration of the zonal mean meridional advective transport across the equator: stronger isolation of the NH (high SF 6 air) and the SH (low SF 6 air) occurs during boreal spring (April) and autumn (October), resulting in the slowest net cross-equatorial exchange of tracer mass.
From the tracer transport component tendency perspective within the ACTM, vertical diffusion accounts for much of the delivery of SF 6 from its surface emission regions into the lower troposphere, after which tracer mass transport to the sub-tropical and tropical upper troposphere is accomplished largely by the effect of parameterized cumulus convection.The maximum mass exchange between the hemispheres occurs in the upper troposphere region via advection, which gives produces smaller 2-box exchange times if the total SF 6 hemispheric mass is considered instead surface-only values.The transport component analysis is further illustrates the relative roles of the three different transport processes at individual sites, with sizable differences in the relative transport partitioning evident for different locations.
Overall, our study demonstrates that the ACTM adequately simulates the transport characteristics of potential relevance to inverse modeling or data assimilation of atmospheric trace species such as CO 2 , CH 4 , and N 2 O.In a follow-up study, we use the ACTM to perform a CO 2 inversion analysis of weekly measurements to estimate highfrequency regional CO 2 fluxes.Given the ACTM's capacity to replicate observed SF 6 without notable bias or error at regional and hemispheric scales, we have some confidence that inverse-estimated regional CO 2 fluxes will reflect proper treatment of atmospheric transport.

Figure 2
Figure 2 compares observed and simulated SF 6 molar mixing ratios at the 6 continuous observing sites as well as the

Fig. 3 .
Fig.3.Synoptic variations in SF 6 at 5 sites (excluding SPO, see Sect.3.1 for details), derived by subtracting the fitted curve from daily mean values for both models and observations.Daily-and 5-day running means are shown as thin and thick lines, respectively, and the correlation coefficients (r) and normalized standard deviations (NSDs) are given in the panel title for each site.Note the variable y-axis scales in each panel.

Fig. 4 .
Fig. 4. Latitude-pressure cross-sections of mean age of air (in days) distribution in the troposphere during the NH winter and summer of year 2000 (shaded).Also shown are zonal wind speed (black contour, levels: −20 to 20 at an interval of 5 m s −1 , and 30 and 40 m s −1 ) and SF 6 mixing ratio (light blue contour, levels: 4.1 to 4.7 at an interval of 0.5 ppt) as simulated by the nudged ACTM run.Longitudinal averages are taken along the center of the Indian monsoon zone (left column; panels a, b) and the central Pacific (middle column; panels c, d).Approximate altitudes (range: ±1 km) corresponding to the pressure levels are labeled on the right y-axis of panel (d).Distributions of potential temperature are shown in Fig.5.This result is produced using the ACTM nudged to ERA40 meteorology.

Fig. 6 .
Fig. 6.Longitude-latitude distributions of ACTM simulated vertical velocity contours (at −1.0, −1.0, −0.5 (red line; upward motion), 0 (black line), 0.5, 1.0, 1.5 (blue line; downward motion) mb hr −1 ) overlaid on the age (shaded) at 200 mb are shown in top two panels (referred as Control simulation, CTL).Bottom two panels (c, d) are showing the age distribution for NoCumTransp case and horizontal wind streamlines being overlaid.Note that the horizontal and vertical velocities are identical for CTL and NoCum-Transp simulations.Age values >100 days, denoting stratospheric air, are whitened.

Fig. 7 .
Fig. 7. Climatological average (period: 2001-2005) monthly-mean inter-hemispheric exchange time as estimated from SF 6 model simulation (line) and observations (symbol) at 6 sites and their three combinations to calculate hemispheric mean concentrations (top; panel (a); light blue: Case 1, black: Case 2, dark blue: Case 3).Bottom panel (b); black line: Case 4, Red line: Case 5 shows exchange times estimated using model based the hemispheric mean mixing ratio at the surface and total mass in the troposphere (surface to 100 mb).Note the difference between two y-axis scales.

Fig. 8 .
Fig. 8. Monthly and zonal average distributions of component SF 6 mass mixing ratio (MR) tendencies (in pptm month −1 ) as modeled in the ACTM for the year 2000; panels (a-d) (left column): transport by grid-scale advection,(e-h) (middle column): convective transport, i-l (right column): vertical transport by diffusion (shaded).The black contours shown denote SF 6 mixing ratio (a-d; unit ppt); potential temperature (e-h; unit K); and vertical velocity ((i-l) unit: mb hr −1 ).These contours are provided for interpreting relationships between the meteorological conditions and transport components as discussed in Sect.3.4.

Fig. 9 .
Fig.9.Total tendencies of modeled SF 6 mixing ratio and the three transport components at 5 sites are shown (in ppt/day).Averages are taken over 5 model grids (nearest sampling grid plus 4 neighboring grids at the same sigma-pressure level) in order to reduce high frequency variations.

Table 1 .
Statistics of model-observation comparison of SF 6 ; average difference in daily concentrations (Diff.; observation -model), 1-σ standard deviation (SD), and number of observations (N) after taking into account the missing data for two different analysis time periods are given.Note the accuracy in continuous SF 6 measurements is about 0.04 ppt.