Levels of ozone depleting substances (ODSs) in our atmosphere are determined by production, emission, and loss processes. However, atmospheric models are forced by the specified mixing ratios of these ODSs rather than the more fundamental emissions-based forcing. To more accurately represent the physics and chemistry of climate change on atmospheric circulation and ODSs, and therefore future ozone recovery, it is desirable to switch from the current highly constrained mixing-ratio-based forcing to emissions-based forcing in general circulation models (GCMs). As a first step of this model transition, we have conducted a 45-year (1960–2005) emissions-based simulation of the three primary chlorofluorocarbons (CFC-11, -12, -113) using the GEOS coupled chemistry-climate model (CCM). The simulated CFC concentrations and their seasonal cycles are compared with AGAGE and NOAA-GMD observations to evaluate emissions and atmospheric transport. The simulated CFC-12 agrees well with the observations, indicating a good estimate of emission and atmospheric loss. The simulated CFC-11 and CFC-113 shows high biases due to overestimate of emissions. Using tagged CFC tracers to track recent surface emissions and aged air masses transported downward from the stratosphere separately, we quantify the relative contribution of stratosphere-troposphere exchange (STE) and tropospheric transport to the seasonal cycles of CFCs in the lower troposphere. The seasonal cycles of CFCs in the lower troposphere are dominated by tropospheric transport of recent emissions during 1985–1994. The 1995–1999 period marks the transition period when variations due to fresh emissions and STE become equally important. Seasonal cycles of CFCs at most surface sites in the 2000–2004 period are dominated by STE. Seasonal cycles of CFCs due to STE show a late winter/early spring maximum and a summer/fall minimum. Seasonality of the tropospheric transport component at individual stations is governed by seasonal transport variations of fresh emissions from the polluted regions.
 While current coupled chemistry climate models (CCMs) are forced by specifying surface mixing ratios recommended by the World Meteorological Organization (WMO)/United Nations Environmental Program (UNEP), recent modeling studies show that increasing greenhouse gases will speed up the mean atmospheric circulation and result in shorter mean age of air [Austin and Li, 2006; Butchart et al., 2006; Garcia et al., 2007; A. R. Douglass et al., The relationship of loss, mean age and the distribution of CFCs to model circulation and implications for atmospheric lifetimes, submitted to J. Geophys. Res., 2008]. This suggests that ozone-depleting substances (ODSs) will be removed more quickly in a future climate [Butchart and Scaife, 2001; Douglass et al., submitted manuscript, 2008]. Therefore to accurately predict the future ozone (O3) recovery and its interaction with climate change, it is desirable to advance from the current commonly adopted mixing-ratio-based forcing to emissions-based forcing in CCMs. One major difficulty in implementing this model transition is demonstrating model's adequacy in simulating the mean atmospheric circulation and the atmospheric lifetimes of ODSs. A model with a poor circulation will not simulate well the annual atmospheric loss, leading to an inaccurate projection of the future ODS concentrations and stratospheric ozone. An additional difficulty lays in the uncertainties of the emissions of ODSs, e.g., chlorofluorocarbons (CFCs) and hydrochlorofluorocarbons (HCFCs). Even with well-regulated man-made ODSs like CCl3F (CFC-11, trichlorofluoromethane) and CCl2F2 (CFC-12, dichlorodifluoromethane), there are significant differences in emission estimates from the existing unvented air conditioners, refrigerators, and plastic foams (referred to as “banks”) [WMO, 2003, Chapter 1; Daniel et al., 2007]. Emission estimates made using a “bottom-up” approach, e.g., estimates from Intergovernmental Panel on Climate Change/Technology and Economic Assessment Panel (IPCC/TEAP) , are based on production and release rates, therefore subject to uncertainties in the bank sizes, annual release rates and service lifetimes of banks [McCulloch et al., 2001]. Another type of estimate derives a CFC emission with box models that use the reported production, observed atmospheric CFC concentration and that CFCs' lifetime as constraints (top-down approach), e.g., WMO . With such an approach, uncertainties arise due to systematic reporting errors as well as uncertainty in the atmospheric lifetimes of the various CFCs [Daniel et al., 2007].
 As a first step of transition from a mixing-ratio-based CCM to an emissions-based CCM, we have conducted an emissions-based simulation of the three principal CFCs (CFC-11, CFC-12, and CCl2FCClF2 (CFC-113)). CFC simulations provide a good initial effort for tackling the emissions problem for two reasons: (1) simple chemistry: CFCs are destroyed in the middle and upper atmosphere through photolysis with well-known rates and (2) single source: CFCs are man-made pollutants released at the surface with reasonably good-reported production and emission release rates, in contrast to many other ODSs that have natural sources and emissions that are highly uncertain.
 Surface CFC observations at many Advanced Global Atmospheric Gases Experiment (AGAGE) sites show prominent seasonal cycles [Prather et al., 1987; Prinn et al., 2000]. Early studies have attributed the seasonal variability to tropospheric transport and interhemispheric exchange [Prather et al., 1987; Prinn et al., 2000]. Nevison et al.  suggested that due to the phase-out of CFCs in the 1990s, observed CFC seasonality in the tropospheric surface stations after mid-1990s should primarily reflect stratosphere troposphere exchange (STE) via the stratospheric Brewer-Dobson circulation. Using the Whole Atmosphere Community Climate Model (WACCM) with mixing-ratio-based forcing, they investigated the seasonality of CFCs in the troposphere and concluded that the stratosphere exerts a coherent influence on the tropospheric CFC mixing ratios. A model's ability to reproduce the amplitude and phase of the seasonal cycle of CFCs in the troposphere is a useful indicator of its capability to simulate atmospheric transport. In addition, many other long-lived atmospheric trace gases, i.e., N2O and CO2 that are important in the projection of ozone recovery and climate change, show similar seasonality as CFCs due to the same governing atmospheric transport processes [Shia et al., 2006; Jiang et al., 2007; Nevison et al., 2007]. A good understanding of the seasonality of CFCs is useful in source characterization of these trace gases.
 Here we present results from a 45-year (1960–2005) emissions-based simulation of CFC-11, CFC-12, and CFC-113 using the three-dimensional GEOS Chemistry Climate Model (GEOS CCM) [Stolarski et al., 2006; S. Pawson et al., Goddard Earth observing system chemistry-climate model simulations of stratospheric ozone-temperature coupling between 1950 and 2005, submitted to J. Geophys. Res., 2008]. We use the simulated CFCs to evaluate the model's atmospheric circulation plus the magnitude and distribution of CFC emissions. The model description and a summary of observations used are included in section 2. The simulated CFCs are compared with observations in section 3. Discussion of the seasonal cycles of CFCs in the troposphere and its implications on the simulated atmospheric transport and CFC emission distributions is presented in section 4, followed by summary in section 5.
2. Model and Observations
2.1. Model Description
 We have conducted a 45-year emissions-based simulation of CFC-11, CFC-12, and CFC-113 from January 1960 to December 2004 using the Goddard Space Flight Center (GSFC) GEOS CCM, described by Stolarski et al.  and Pawson et al. (submitted manuscript, 2008). The 1 January 1960 initial mixing ratios of CFCs and other atmospheric constituents are taken from a prior mixing-ratio-based simulation to account for accumulation of emissions prior to 1960. The horizontal resolution of the simulation is 2° latitude by 2.5° longitude, with 55 layers extending from the surface to 0.01 hPa. The GEOS CCM couples the GEOS-4 GCM [Bloom et al., 2005] with a stratospheric chemistry module [Douglass et al., 1996]. A comprehensive evaluation of several CCM simulations of the period 1960–2005 shows that the GEOS CCM simulation agrees well observations in many of the meteorological, transport-related, and chemical diagnostics [Eyring et al., 2006]. Pawson et al. (submitted manuscript, 2008) further validate the GEOS CCM ozone and temperature with the observations and show the strengths of the model in simulating ozone, temperature, and the temperature-ozone coupling. Of particular relevance to this study, Douglass et al. (submitted manuscript, 2008) and Waugh et al.  presented detailed analyses of the simulated CFC-11 and CFC-12 from the GEOS CCM and a Chemical Transport Model (CTM) which uses the same chemistry and meteorological fields generated by the same GEOS-4 GCM. These two studies show that GEOS CCM reproduces well the observed concentrations and vertical profile of CFC-11 and -12 in the atmosphere, which indicates a good representation of the mean atmospheric circulation with realistic age-of-air, and therefore a realistic loss and atmospheric lifetime for CFCs.
 The GEOS CCM simulation couples the mixing-ratio-based ODSs, including CFCs, into the radiation code. The Cly in the chemistry simulation comes from the mixing-ratio-based CFCs. The simulation also includes three independent tracers: CFC-11, 12, and 113 calculated with specified emissions. These are transported by the CCM circulation and removed by photolysis and reaction with O(1D) in the middle and upper atmosphere with the same rates as the mixing-ratio-based CFCs, but they do not feedback into the radiative heating code. In addition, while the emissions-based CFCs may differ from the mixing-ratio-based CFCs, the difference does not feedback on Cly or ozone.
 Emissions of CFCs are released at the lowest layer in the model using the bottom-up emission estimates of CFCs from IPCC/TEAP , which are based on compilations of global production [AFEAS, 2004], sales into each end-use and time-dependent release from each end-use [McCulloch et al., 2001, 2003]. Historically, CFC-11 was used primarily in aerosol propellant mixtures with prompt release to the atmosphere within a year after production but later mainly used in rigid plastic insulating foams of which leakage is small and slow [McCulloch et al., 2001]. CFC-12 was used primarily within refrigeration and air conditioning systems with additional usage as aerosol propellants [McCulloch et al., 2003]. CFC-113 was used as a solvent and cleaning agent for electronics and metal [Environmental Protection Agency (EPA), 1980]. The annual global emissions of CFC-11, 12, and 113 for each year between 1960 and 2005 are shown in Figure 1. In our simulation, we release CFCs with the same magnitude throughout the year, assuming no seasonal dependence. For simplicity, we deploy a regional distribution of CFC-11 and 12 that is compiled based on the fractions of world total production for individual countries as described by McCulloch et al. [2001, 2003], with updated production information for the year 2000 from the Global Emissions Inventory Activity (http://www.geiacenter.org). The same distribution fraction is adopted throughout the entire simulation period. 90.0% of the CFC-12 and 98.9% of the CFC-11 are emitted in the Northern Hemisphere (NH), with the remaining released in the Southern Hemisphere (SH). An example distribution of CFC-11 for 2000 is shown in Figure 2. Note that an alternative choice is to use the UNEP consumption data as a proxy for CFC emissions, which is possibly a better representation and is available for multiple years. For CFCs that have long atmospheric lifetimes (>45 years), compared to an interhemispheric exchange time of a few months in the tropics [e.g., Bowman and Cohen, 1997; Gupta et al., 2001] to ∼1 year for mixing to the extratropics [Bowman and Cohen, 1997], we do not expect the differences in the distributions to affect the long-term simulation of atmospheric CFCs.
 To quantitatively understand the relative contribution of the stratosphere-troposphere exchange (STE) and tropospheric transport to the seasonal cycle of CFCs in the troposphere, we have also added two sets of tagged tracers in the simulation, one set of tropospheric tracers to track CFCs that have been recently emitted (CFC-11T, CFC-12T, CFC-113T) and another for CFCs that were at some time in the stratosphere (CFC-11S, CFC-12S, and CFC-113S). The two sets are transported as independent tracers in the CCM. Fresh CFCs emitted at the surface are labeled as “tropospheric CFC tagged tracers” and being transported in the CCM circulation. When the tropospheric tagged tracers are transported across the tropopause into the stratosphere, they become “stratospheric CFC tagged tracers”. The majority of these stratospheric tracers circulates through the middle and upper atmosphere and re-enters the troposphere via STE. However, these tracers keep their identities as stratospheric tracers to separate the old CFCs from new emissions. Since the regional emission distribution is for 2000, we limit our tagged simulation between January 1995 and December 2004 for a meaningful comparison between the simulations and the observations. Initial concentrations of tropospheric and stratospheric tracers for the tagged simulation are obtained with a 5-year spin-up before 1995 to obtain a reasonable partition between the two.
 Long-term observations of CFC-11, CFC-12, and CFC-113 are made by the Advanced Global Atmospheric Gases Experiment (AGAGE) [Prinn et al., 2000] and the National Oceanic and Atmospheric Administration – Global Monitoring Division (NOAA-GMD) [Montzka et al., 1999; Thompson et al., 2004]. We use surface measurements of CFCs at five stations from the AGAGE network: (1) Ireland, Adrigole (ADR, 10.0°W, 53.0°N), and Mace Head (MCH, 9.5°W, 53.4°N); (2) Cape Meares, Oregon (CME, 123.6°W, 45.3°N), and Trinidad Head, California (TRH, 124.9°W, 41.3°N); (3) Ragged Point, Barbados (RGP, 59.3°W, 13.1°N); (4) Cape Matatula, American Samoa (MAT, 170.3°W, 14.2°S); (5) Cape Grim, Tasmania (CGR, 144.4°E, 40.4°S) (Figure 2). The NOAA-GMD observations are obtained at 10 sites: (1) Alert (ALT, 62.3°W, 82.3°N), (2) Barrow (BRW, 156.4°W, 71.2°N), (3) Park Falls (LEF, 90.2°W, 45.6°N), (4) Harvard Forest (HFM, 72.2°W, 42.5°N), (5) Niwot Ridge (NWR, 105.4°W, 40.3°N), (6) Grifton (ITN, 77.2°W, 35.2°N), (7) Cape Kumukahi (KUM, 154.5°W, 19.3°N), (8) Mauna Loa (MLO, 155.4°W, 19.3°N), (9) Tutuila (SMO, 170.3°W, 14.2°S), and (10) South Pole (SPO, 24.5°W, 89.6°S). The locations of the stations are shown in Figure 2.
3. Simulated Results and Evaluation of the Emission Estimates
Figure 3 shows the observed and the simulated surface monthly mean CFCs in the NH and the SH between January 1960 and December 2004. The hemispheric mean surface mixing ratios are area-weighted (cosine of the latitude) average values. The simulated CFC-12 (top) agrees well with the observation in both hemispheres and the average differences (simulation-observation) are −0.1% in the NH and −0.6% in the SH, with annual biases (simulation-observation) range from −2.4% to 1.6%, comparable to the 2% differences between the independent laboratory measurements [WMO, 2007]. This indicates that the simulation presents a good estimate of annual emissions and photolytic loss in the stratosphere.
 The simulated CFC-11 (middle) in the NH shows a high bias and the overestimate increases from an average of 4.9 pptv (+2.7%) in 1978–1984 to 10.4 pptv (+4.1%) in 1985–1994 and 18.1 pptv (+6.9%) in 1995–2004. The simulated CFC-11 in the SH agrees well with the observation before 1989, with annual differences <3.5 pptv (+1.5%), but the overestimate increases significantly afterward and reaches an average of 15.6 pptv (+6.0%) in 1995–2004.
 The simulated CFC-113 (bottom) in both hemispheres is consistently higher than the observation with the overestimate steadily increasing from 4.5 pptv (+11.3%) in the NH and 3.8 pptv (+12.9%) in the SH in 1984 to peak values of 18.0 pptv (+22.2%) in the NH in 1992 and 15.8 pptv (+19.2%) in the SH in 1995. As emission decreases rapidly after the early 1990s (Figure 1), the simulation overestimate also decreases (+17.0% and +16.7% in the NH and the SH, respectively, for 1995–2004).
 The comparison between the simulated and observed CFCs at individual surface sites is consistent with the comparison between the hemispheric means. Figure 4 shows the midlatitude sites, Adrigole/Mace Head in the NH and Cape Grim in the SH. The model reproduces well the observed CFC-12 at both sites (top panels), with low biases, −3.4 pptv (−1.0%) at Adrigole/Mace Head and −5.5 pptv (−1.2%) at Cape Grim, respectively. Simulated CFC-11 bias (middle panels) increases slowly from an average of +4.2 pptv (+2.3%) at Adrigole/Mace Head and +1.0 pptv (+0.6%) at Cape Grim before 1985 to +21.3 pptv (+8.2%) and +7.6 pptv (6.8%) in 1995–2004, respectively. The simulated CFC-113 (bottom panels) is consistently higher than the observation, with average biases of +15.4 pptv (+19.3%) at Mace Head and +11.4 pptv (+16.7%) at Cape Grim between 1983 and 2004.
 We diagnose the simulation biases by examining the simulated atmospheric lifetimes of CFCs. We calculate photolysis rates offline using archived overhead nitrogen (N2), oxygen (O2), and O3 column, and the reaction rates of CFC + O(1D) with saved O(1D) concentration and temperature fields. The average lifetimes of CFC-11 and CFC-12 calculated in the GEOS CCM are 57 years and 100 years, respectively, with photolysis accounting for >99% of the loss. CFC-113 has a lifetime of 97 years, with 90% of the loss due to photolysis and 10% via reaction with O(1D).
 The accurate simulation of the observed CFC-12 mixing ratios and an atmospheric lifetime of 100 years, similar to that in Chapter 1, Table 1–4., in the WMO  assessment, suggests reasonable actinic fluxes, and hence photolytic loss of CFCs, in the model [Waugh et al., 2007; Douglass et al., submitted manuscript, 2008]. This accuracy is essential to correctly simulating the atmospheric transport, fractional release, and photochemical age of CFCs (Douglass et al., submitted manuscript, 2008). The aforementioned implies that the overestimate in CFC-11 mixing ratios is probably due to an overestimate of emissions from banks. Compared with CFC-12 (most of its bank resides as refrigerants with an average characteristic release time of 3–5 years), more than 90% of the CFC-11 banks are in close-cell plastic foams with complex and highly variable immediate release fractions from 4% to 95%, leaking rates from 0.5% to 5% a−1, and service time from 12 to 50 years [McCulloch et al., 2001]. All these uncertainties propagate into emission estimates. The result is large uncertainty in the CFC-11 emission rate, especially when the primary usage switches from aerosol propellants to plastic foam blowing agents. This is consistent with our simulation results that show increasing overestimate in CFC-11 after 1990, when emission of CFC-11 is dominated by release from plastic foam banks.
 Unlike CFC-11 with significant quantities stored in the banks, 99% of CFC-113 is released to the atmosphere within 6 months of sale and has negligible banks [AFEAS, 1995; Fisher and Midgley, 1993]. CFC-113 emission is estimated using the fraction of CFC-113 in total CFCs in the AFEAS database and scaled up accordingly based on the total CFCs in the UNEP database to include additional emissions from “nonreported” countries, e.g., China, Czech Republic, India, Korea, Taiwan, Romania, and Russia [McCulloch et al., 2001]. This derivation method could introduce error in emission estimates if the fractional composition of individual CFCs is different between the reported AFEAS database and the nonreported countries. The two possible contributors for the high concentrations of CFC-113 in the simulation are overestimates of emissions and/or underestimates of losses. We compare the model-simulated CFC-113 burden in the atmosphere with observations in Figure 5. As >90% of the burden resides in the troposphere and CFC-113 concentration is relatively uniform in the troposphere, the surface mixing ratio of CFC-113 is well-correlated (r = 0.9999 in the model) with the total atmospheric burden. We infer the observed burden from the observed surface observations based on a model-derived relationship. Before 1983 when CFC-113 observations are unavailable, the surface CFC-113 concentrations recommended by A1B scenario of the WMO  are used instead. The simulated CFC-113 burden exceeds the WMO A1B burden in 1964 and the overestimate continues to increase until reaching a maximum of 23% in 1990, coincides with the emission maximum in 1989. If we decrease the CFC-113 lifetime from the simulated 97 years to 50 years, the calculated atmospheric CFC-113 burden matches the observed burden in 2004 while the accumulation rate before 1994 and the decrease rate after 1994 are too fast compared with observations. When emission is decreased to 85% of what is used in the simulation, the calculated atmospheric burden agrees well with observations between 1960 and 2004, indicating that the overestimate of CFC-113 in the simulation is most likely due to an overestimate in emission.
 The interhemispheric gradients (IHG) of CFCs are governed by the NH-SH emission partitioning, interhemispheric transport and STE in both hemispheres. The observed interhemispheric differences decrease from 1980s (21.2 pptv, 12.9 pptv, and 6.8 pptv for CFC-11, -12, and -113, respectively) to 1990s (11.7 pptv, 6.4 pptv, and 3 pptv for CFC-11, -12, and -113, respectively) (Figure 3). The IHG of CFC-12 in the simulation agrees relatively well with the observations (an average of 23.9 pptv in 1980s and 14.8 pptv in 1990s). The simulated CFC-11 hemispheric difference (18.5 pptv in 1980s and 9.4 pptv in 1990s) is ∼40% too high compared to the observations. The simulated CFC-113 IHG (9.9 pptv in 1980s and 4.6 pptv in 1990s) is 50% higher than the observed before 2000 and becomes comparable (0.3 pptv) to that of the observations in 2000–2004. The model simulates well the time lag for the concentration in the SH to reach that in the NH for all three CFCs (the observed = 1.2–1.5 years, the simulated = 1.2–1.6 years) when emissions were steadily increasing during the 1980s and the early 1990s, indicating a good representation of the NH-to-SH transport of emissions. The model also reproduces well the variations in surface CFCs due to STE and the contrast in the amplitude of the variation in two hemispheres (the amplitude in the NH doubles that in the SH) (section 4.1). This implies a good representation of the strength of stratosphere-to-troposphere flux in individual hemispheres and hence its impact on IHG. We examine how the NH–SH emission partition ratio and the magnitude of emission would impact the IHG. The NH–SH emission partition ratio of CFC-12 has decreased from 99.0%–1.0% in 1980–1990 (http://www.geiacenter.org) to 90.0%–10.0% in 2000. Though the same 90.0%–10.0% ratio was used for the entire 1960–2004 period, the simulation captures well the observed IHG. In contrast, the CFC-11 NH–SH emission partition ratio has remained relatively constant throughout the time (99.2%–0.8% in 1980–1990, 98.9%–1.1% in 2000). The simulated CFC-11 IHG is high-biased compared to the observed even though a reasonable NH–SH emission partition ratio is used. In addition, the magnitude of IHG of CFCs decreases proportionally as emission decreases from 1980s to 1990s. These suggest that for CFCs, of which the majority of emissions (>90%) resides in the NH, the IHG is dominated by the magnitude of emission.
4. Seasonal Cycle of CFCs
 To examine the seasonality of CFCs at ground-based stations, we remove long-term variations from the observed and simulated CFCs, as well as the CFC tagged tracers, by applying a 13-month high-pass filter which uses a 13-month boxcar average algorithm. The filtered monthly mean anomalies are averaged for each month of the year to obtain the long-term climatological seasonal cycles. Figure 6 shows the CFC-12 time series (top) and filtered anomalies (bottom) between 1995 and 2004 for two example sites, Mace Head (Figure 6a) in the NH and Cape Grim in the SH (Figure 6b). The simulation reproduces relatively well the observed seasonal variations of CFC-12 at the NH Mace Head site (Figure 6a) between 2000 and 2004, but shows a phase-delay of a few months during 1995–1999. The bias in simulating the phase of seasonality during 1995–1999 is also present at a few other NH midlatitude and tropical sites, Trinidad Head, Mauna Loa, and Cape Kumukahi (not shown), which are influenced by continental pollution outflow [e.g., Prospero and Savoie, 1989; Prospero et al., 2003; Heald et al., 2003; Liang et al., 2004]. The simulated seasonal cycle of CFC-12 at the SH Cape Grim, as well as three other SH sites (not shown), agrees with the observations both in magnitude and phase. Because of rapid decreases in emissions in North America and Europe in the 1990s, the regional emission distribution has changed significantly from the 1980s to the 2000s. Error in regional emission distribution is likely to introduce bias in simulating the seasonal variation of tropospheric transport of fresh CFCs to the above NH sites. Model transport errors may also contribute to the biases. Therefore we first limit our comparison of simulated versus observed seasonal cycles for 2000–2004 and then examine how seasonal cycles of CFCs change as emissions change from 1985 to 2004 to understand the governing processes of CFC seasonality and possible causes of simulation biases.
4.1. Seasonality of Surface CFCs and Governing Processes
 The 5-year averaged seasonal cycles of CFCs and the tagged tracers for 2000–2004 are shown at four selected stations in the NH in Figure 7a and three stations in the SH in Figure 7b. Significant seasonal variability of CFCs is observed at many surface stations. Observed CFCs show a late winter/early spring maximum (January–February in NH and September–October in SH) and a summer/early fall minimum (July–September in NH and March–May in SH) (Figures 7a and 7b). The amplitudes (maximum–minimum) of the seasonal cycles of the CFCs are ∼0.4% (∼2 pptv for CFC-12, ∼1 pptv for CFC-11, and ∼0.4 pptv for CFC-113) of the absolute mixing ratios at the NH sites and reduce to ∼0.2% at the SH sites. The simulation reproduces well the observed phase and amplitude of the seasonal cycles of CFC-11 and CFC-12 at most of the sites. The seasonal maxima/minima of simulated CFC-113 at most stations are shifted 1–2 months from that observed, and amplitudes of seasonal variations are underestimated.
 We further decompose the seasonal cycle of CFCs to quantify the contributions of stratosphere-to-troposphere transport of old air and tropospheric transport of recent emissions between 2000–2004 by examining the two sets of tagged tracers (Figures 7a and 7b). Seasonal variations of CFC-11 and CFC-12 at the surface sites are dominated by variations of the stratospheric component (CFC-11S and CFC-12S) throughout the globe, except at the southern tropical sites (Tutuila, American Samoa), with contributions from tropospheric transport (CFC-11T and CFC-12T) varying from a comparable magnitude to almost zero at individual sites. Seasonality of CFC-11 and CFC-12 at Tutuila is dominated by the tropospheric component due to active NH–to–SH transport. The simulated seasonal variation of CFC-113 closely follows that of CFC-113S as emissions are negligible, resulting in very small values (<0.1 pptv) of CFC-113T during 2000–2004.
 The stratospheric tracers show early spring maxima and summer/early fall minima at the surface in both hemispheres. The amplitudes of the seasonal cycles of the stratospheric CFC tracers are largest in the NH middle and high latitudes: ∼2.0 pptv for CFC-12S, ∼1.6 pptv for CFC-11S, and ∼0.4 pptv for CFC-113S. The amplitudes reduce by half in the NH tropics and the SH middle and high latitudes, and are small in the SH tropics.
 The seasonal variations in the stratospheric component of CFCs are associated with the STE [Nevison et al., 2004]. The winter downwelling of old, CFC-depleted air from the middle and upper stratospheric results in a CFC seasonal minimum in the polar lower stratosphere (the blue anomalies in Figures 8 and 9a). The amplitude of the associated seasonal variations is stronger in the NH compared with the SH because of the stronger wave driving mixing rates and downward Brewer-Dobson stratospheric circulation [Haynes et al., 1991; Appenzeller et al., 1996; Holton et al., 1995]. In the NH, the seasonal cycle in the lower stratosphere slowly migrates to the troposphere. The descent in the Arctic has a seasonal minimum anomaly of −6 pptv in April–May below the tropopause region (Figure 9b). However, it is difficult for stratospheric air masses to penetrate into the Arctic lower troposphere [Stohl et al., 2006; Law and Stohl, 2007]. The lower tropospheric minimum occurs in July–August (Figure 9a), approximately 3 months behind the near tropopause region's minimum and 4–5 months behind the lower stratosphere one. The amplitude of the seasonal cycle in the lower troposphere is significantly reduced (1 pptv for CFC-12) compared to the lower stratosphere (70 pptv) and near tropopause region (5–10 pptv). This seasonal minimum in the lower troposphere is reached in the NH extratropics in summer (July–August period in Figure 9c). Compared to the NH, the SH has weaker orography-induced wave activities, less cyclogenesis and tropopause folding events, and therefore weaker STE [Haynes et al., 1991; Appenzeller et al., 1996; Beekmann et al., 1997]. Hence the downward migration from the lower stratosphere to the upper troposphere is weaker and slower in the SH than the NH, resulting in weaker troposphere minimum (Figures 8 and 9). The surface minimum in both hemispheres is consistent with net mass flux out of the stratosphere during the spring and early summer [Olsen et al., 2004]. Furthermore, the weaker annual cycle in the SH is also consistent with the weaker SH mass flux out of the stratosphere [Olsen et al., 2004].
 Since emissions have decreased rapidly after the early 1990s and reduced to very low levels by 2000, percentage contribution of recent emissions (<4%) to the tropospheric burden of CFC-12 is minor compared to aged emission that has circulated through the stratosphere (Figure 6). However, tropospheric transport of recent emissions still contributes significantly to the seasonal cycles in the lower troposphere (Figure 7). Figure 10 shows the annual-mean global distribution of the simulated tropospheric tracer, CFC-12T, averaged between 0–2 km for 2000. In the NH, distribution of CFC-12T in general follows emissions with hot spots of high mixing ratios concentrated in high emission regions, i.e., Europe, China, and the eastern United States (Figure 2). CFC-12T in the NH middle and high latitudes shows a seasonal variation of ∼1 pptv, 5–10% of absolute mixing ratios, with a maximum in winter as surface emissions are capped in the lower troposphere by strong subsidence during winter, in particularly over Europe and Asia [Stohl et al., 2002; Liang et al., 2004; Stohl, 2006]. In addition, mixing ratios of the tropospheric component of CFCs at individual stations vary as winds vary in strength and direction in individual seasons, bringing air from different continents that differ in CFC levels as the emission magnitudes are different. In the tropics, 10°N–20°S, CFC-12T mixing ratios exhibit significant seasonality (∼1 pptv) with maximum during NH winter due to the NH–to–SH transport which occurs primarily in the near-surface layer in winter [Gupta et al., 2001]. In the SH, distribution of CFC-12T is zonally uniform. Mixing ratios decrease poleward, and exhibit little seasonal variation.
 The GEOS CCM simulates well the seasonality associated with STE [Olsen et al., 2004] and the NH–to–SH transport, therefore reproduces well the amplitude and phase of seasonal cycles of CFC-11 and CFC-12 at sites that are dominated by these two processes, i.e., Park Falls, Tutuila, Cape Grim, South Pole (Figure 7). At several NH observation sites where long-range hemispheric transport contributes significantly to the seasonality of CFCs, the simulated seasonal cycles of CFC-11 and CFC-12 show a slight phase-shift or overestimate in amplitude, suggesting biases associated with emission distribution and tropospheric meteorological transport. The biases in the simulated CFC-113 are more apparent (Figure 7). The phase and percentage amplitude of CFC-113S agrees well with those of CFC-11/CFC-12, indicating well simulated seasonality due to STE. A significant increase in the amplitude of CFC-113T is needed to reconcile the difference between the observed and simulated CFC-113 seasonal cycles. This suggests that while total CFC-113 emission between 1960 and 2004 is overestimated, the emission for 2000–2004 is underestimated. This is consistent with the fact that the simulated CFC-113 is decreasing at a faster rate (−1.21%/a) than the observed CFC-113 (−1.01%/a) after 1995.
4.2. Evolution of CFC Seasonality and Implications for Emissions
 In this subsection, we examine how annual cycles of CFCs have changed between 1985 and 2004 as emissions decreased. We use the AGAGE data set which excludes pollution events to examine variations driven by the seasonal change in the large-scale transport of polluted air from the continents versus clean air from the remote regions by the mean surface winds. When the pollution events are included, seasonal cycles of CFCs remain relatively unchanged except at Adrigole and Cape Grim during the 1980s when emissions were high. These two stations are close to the populated regions and susceptive to frequent local pollution events. The inclusion of the pollution events at the two sites significantly increases the amplitude of the seasonal cycle but only has a minor impact on the phase of the seasonal cycle.
Figure 11 shows the 5-year averaged seasonal cycles of CFC-11 for 1985–1989, 1990–1994, 1995–1999, and 2000–2004 (CFC-12 is similar). Note that the magnitude of seasonality of the stratospheric tracers of CFCs is proportional to their atmospheric loading as governed by the Brewer-Dobson circulation. Since the concentration of CFCs in the middle and upper atmosphere during 2000–2004 is comparable or larger than the earlier years, the amplitude of CFC-11 seasonality due to STE should be ≤1.6 pptv in the NH (≤1 pptv in SH) before 2000. During 1985–1994, seasonal amplitude of CFC-11 reaches ∼4–5 pptv at the NH stations, implying that the seasonal variation of CFC-11 is dominated by fresh emissions. As emission decreases, seasonal variation of CFC-11 is significantly reduced in amplitude. 1995–1999 marks the transition period when both tropospheric transport and STE contribute significantly to the seasonal cycles of CFC-11 at the surface. Seasonal variation of CFC-11 is dominated by STE during 2000–2004 (section 4.1).
 The hemispheric transport of anthropogenic emissions from populated continents to the remote AGAGE/GMD sites contributes to the observed seasonality of carbon monoxide (CO) [e.g., Cape et al., 2000; Li et al., 2002; Heald et al., 2003; Liang et al., 2004]. Since anthropogenic emissions of CO and CFCs share common source regions, the contribution of emissions from different regions to the seasonality of CO at the remote stations lends valuable information in understanding the magnitude of regional emissions of CFCs over Europe, North America, and Asia. Mace Head (Figure 11, top row, middle) is mostly influenced by anthropogenic emissions from Europe [Simmonds et al., 1997; Cape et al., 2000] with peak influence in fall and winter [Li et al., 2002]. As European emissions were reduced, the seasonal cycle changed from a peak in October (reflecting dominant influence from fresh European emissions) to a peak in February due to STE. European emission outflow dominates at Barrow (Figure 11, top row, left) from winter to early spring, but Asian emissions become important in late spring [Liang et al., 2004]. The seasonal cycle of CFC-11 gradually decreases in amplitude and shifts from a peak in December during 1985–1989 to a peak in March during 1995–2000, indicating a transition from European-emission-dominated to Asian-emission-dominated. During 2000–2004, the observed seasonal cycle of CFC-11 at Barrow is close to that due to STE, with a maximum in February and a minimum in September. The two sites in the Pacific sector, Trinidad Head and Mauna Loa are, in general, influenced by North American emissions in fall and trans-Pacific transport of Eurasian emissions in spring [Heald et al., 2003; Liang et al., 2004], and hence show a peak in October and a secondary peak in May during the peak emission time. During 2000–2004, amplitudes of observed seasonal cycle of CFC-11 at these sites are greatly reduced with phases approaching those due to STE. The simulated seasonal cycles show an overestimate in spring at these sites as well as Barrow, indicating that the Asian emission used in the simulation is possibly too high. The comparison of CFC-11 emission in 1986 with 2000 shows that European emission decreased five times from 1.76 × 105 tons/a (53% of global total) to 0.32 × 105 tons/a (37% of global total) and North American emission decreased 70% (0.96 × 105 tons/a and 29% of global total in 1986, 0.27 × 105 tons/a and 30% of global total in 1986). Though Asian emission also decreased from 0.52 × 105 tons/a in 1986 to 0.26 × 105 tons/a in 2000, its relative contribution to global emission doubled (15% in 1986, 29% in 2000). These changes in regional emission are consistent with the changes in seasonal cycles at the above stations.
 Interestingly, the seasonality of CFC-113 at most of the available sites display a very different signature compared to CFC-11 and CFC-12 during its peak emission time period, 1980–1989. The seasonal maximum occurs during May–July at Mace Head, Trinidad Head, and Cape Grim (Figure 12). This implies that either the usage of CFC-113 during the 1983–1989 period has a distinctive seasonal dependence or its regional distribution is very different from that of CFC-11 and -12. However, the fact that continuous CFC-113 observations are only available at a few sites before emissions are significantly decreased makes it difficult to address the uniqueness of the seasonality of CFC-113. After 1990, the seasonal cycle of CFC-113 is similar to that of CFC-11 and CFC-12, and gradually reduces to a seasonal cycle that is dominated by STE. The magnitude of the simulated CFC-113 seasonal cycle is smaller than the observed at most sites with additional phase shifts at a few sites, i.e., Barrow, Tutuila, and South Pole, for 2000–2004, implying a missing component from tropospheric transport, as we have discussed in section 4.1.
 An emissions-based approach to the ozone depleting substance (ODSs) forcing of a CCM is a more fundamentally sound technique in examine the effects of the change in circulation on ODS concentrations than constraining a CCM with a mixing ratio boundary condition. The relaxation of the mixing ratio constraint generally assumed in a CCM provides an important step in the evaluation of the governing dynamical and chemical processes that determine the time evolution of ODSs. During this evaluation process, ground observations can be used as an important piece of information to validate and improve emission, tropospheric transport, and STE. In the long-run, emissions-based CCM is necessary to accurately predict future ozone recovery. As the first step toward switching from mixing-ratio-based forcing to emissions-based forcing, we have conducted a 45-year emissions-based simulation of the three primary CFCs (CFC-11, -12, and -113) between January 1960 and December 2004 using the GEOS-4 CCM. The simulation results are compared with surface observations to evaluate emission sources, photolytic losses, and atmospheric transport processes.
 The GEOS CCM simulation reproduces well the observed CFC-12 mixing ratios at the surface with average biases of −0.1% in the NH and −0.6% in the SH, indicating a good estimate of emission and atmospheric loss. The simulated CFC-11 in general shows a high bias (5% in the NH and 4% in the SH) and this overestimate increases with time, possibly due to overestimate of emission from banks. If we assume no significant biases with the reported production quantities, the overestimate of CFC-11 emission from banks implies that the current estimate of release rates from plastic foams is too high and significant emissions will extend further into future years than originally expected. The simulated CFC-113 shows a consistent high bias for the 1983–2004 period due to overestimate in emissions. An atmospheric CFC-113 burden calculated with 15% decrease in emission matches well with the observations. The magnitude of emission plays an important role in accurately simulating the interhemispheric gradient of CFCs. As a result, the overestimate in CFC-11 and 113 emissions also results in an overestimate of the simulated interhemispheric gradient.
 Surface observations of CFCs at many AGAGE and NOAA-GMD sites show significant seasonal variations. Seasonal cycles of CFCs in the lower troposphere are dominated by tropospheric transport of recent emissions during the 1985–1994 period with peak-to-trough amplitude of as much as 6 pptv for CFC-11, 10 pptv for CFC-12, and 3 pptv for CFC-113. The 1995–1999 period marks a transition during which variations due to fresh emissions and STE are equally important with comparable magnitude but distinctive phases. Seasonal cycles of CFCs in the 2000–2004 period are dominated by the STE, but tropospheric transport still contributes significantly at many observation sites and affects the amplitude and phase of CFC seasonality. These results are consistent with the conclusions from earlier studies by Prather et al. , Prinn et al. , and Nevison et al. . Seasonal cycles of CFCs due to the STE show a late winter/early spring maximum and a summer/fall minimum, with amplitudes of ∼0.4–0.5% of the absolute mixing ratios are the surface. Individual observation sites show distinctive seasonal cycles of CFCs as governed by seasonal variations of surface emissions transported from the polluted continents. Amplitude and phase of seasonal cycles of CFCs at these sites change accordingly as emissions over Europe, North America, and Asia change from the 1980s to 2000s, suggesting that the seasonality of CFCs is a sensitive indicator of the magnitude of CFC emission as well as its regional distribution. Regional tagging technique and inverse modeling have been used extensively in the past decade to improve the emission inventory of short-lived anthropogenic pollutants such as CO [e.g., Bey et al., 2001; Palmer et al., 2003]. Applying such techniques in CCMs with emissions-based forcing will provide a useful tool in obtaining a top-down estimate of global emissions of CFCs and other long-lived ODSs.
 This research was supported by an appointment to the NASA Postdoctoral Program at the Goddard Space Flight Center, administered by Oak Ridge Associated Universities through a contract with NASA. We thank Archie McCulloch and Guus Velders for providing CFC emissions and distributions. We thank the reviewers for their helpful comments.