We present methane mixing ratio and δ13C time series measured at Baring Head, New Zealand, and Scott Base, Antarctica, over the years 1991–2003. These data demonstrate that the apparent kinetic isotope effect (KIE) of the methane atmospheric sink (derived from the amplitudes of the mixing ratio and δ13C seasonal cycles) is generally much larger than would be expected if the sink were the hydroxyl radical alone and has changed significantly during the observation period on a timescale of ∼3 years. We show using a global transport model that this technique for deriving the KIE should be quite accurate for a single atmospheric sink and that the change with time is unlikely to arise from El Niño–Southern Oscillation transport effects. We infer that a sink in addition to hydroxyl is required. A strong candidate for this extra sink is atomic chlorine in the marine boundary layer (MBL). We derive the amplitude of the chlorine concentration seasonal cycle that would fully account for the apparent KIE. This amplitude ranges from ∼104 atom cm−3 in 1994–1996 to about 3 × 103 atom cm−3 in 1998–2000. If the KIE is enhanced throughout the free troposphere, the seasonal mean concentrations of atomic chlorine required in the MBL would be about 3 × 104 atom cm−3 in 1994–1996 and ∼104 atom cm−3 in 1998–2000.
 Methane (CH4) is a significant species in the atmosphere, both from its effect on the Earth's radiation balance and from its role in atmospheric chemistry. Model estimates suggest that CH4 accounts for ∼20% of the incremental trace gas radiative forcing since preindustrial times [Myhre et al., 1998]. CH4 plays an important role in the regulation of stratospheric ozone levels and consumes ∼25% of hydroxyl (OH•) radicals in the troposphere, thus becoming an in situ source of carbon monoxide (CO) and ozone (O3) [Thompson, 1992].
 The mean mixing ratio of CH4 in the atmosphere (abbreviated from now on as MR) has more than doubled over the last 150 years [Etheridge et al., 1998]. Over the last 40 years the average atmospheric growth rate of CH4 has been nearly 1% per year [e.g., Etheridge et al., 1998]. However, the average growth rate has slowed over the last 20 years to <0.5% per year and was approximately zero from 1999 through 2002 [Dlugokencky et al., 1998, 2003]. These changes are at present not fully understood, a situation that is directly related to uncertainties in the size, spatial distribution, and trends of identified methane sources. Measuring the stable isotopic (13C) composition of atmospheric CH4 can reduce these uncertainties because several CH4 source types can be distinguished by their characteristic isotope signatures [e.g., Bergamaschi et al., 2001].
Allan et al. [2001a] extended the utility of isotope measurements to assess indirectly the properties of CH4 sinks in a limited region of the atmosphere, namely, the extratropical Southern Hemisphere (ETSH) (south of 23.5°S). For CH4 the 13C/12C ratio is reported through the δ13C ratio defined by
where R0 = (13C/12C)PDB has an accepted value of 0.0112372 for the isotope standard, Peedee belemnite (PDB) [Craig, 1957], and δ13C is scaled by a factor of 1000, being reported as per mil, “‰.” Allan et al. [2001a] showed that when the change in δ13C(CH4) is plotted versus the relative change in MR measured in the ETSH over a seasonal cycle, an ellipse-like figure (a “phase ellipse”) is obtained. Their modeling study demonstrated that the shape of the phase ellipse is modified by source effects but that the slope of the ellipse major axis appeared to be a robust measure of the kinetic isotope effect (KIE) ɛ of the oxidation process removing CH4 from the ETSH lower troposphere. Here ɛ = (k13/k12) − 1, where k13 and k12 are the rate constants for 13C and 12C removal, respectively, by the oxidation process. Note that from now on, 13C and δ13C should be taken as referring to the carbon in CH4.
Allan et al. [2001a] modeled this using OH• as the sole oxidant and found that ɛ derived from the simulated phase ellipse major axis slope was consistent with the value of ɛOH supplied as input to the model. However, comparison with measurements in New Zealand and Antarctica showed that ɛ derived from observed ellipses was significantly larger in magnitude than expected for OH• oxidation alone (−3.9‰ [Saueressig et al., 2001]), with a value of about −13‰. We refer to this observed ɛ as ɛA, the “apparent KIE.” Allan et al. [2001b] used a simple chemical box model to demonstrate that a plausible seasonal cycle of chlorine radicals (Cl•) in the marine boundary layer (MBL) acting together with the expected OH• seasonal cycle could potentially account for the magnitude of ɛA.
 A mean of the data from the years 1993–1996 was used for the above comparison, as these years yielded phase ellipses with similar major axis slopes. However, a time series of measurements from 1991 to 2003 is now available, and it is apparent that there is considerable interannual variation in the structure of the phase ellipses over this period. In particular, the period 1997–2003 is significantly different from the period 1993–1996 considered in our earlier work. In the present paper, we describe the complete data set and give example phase ellipses at two observing sites in the ETSH. We employ two new methods, reduced major axis (RMA) regression and nonlinear least squares fitting of sinusoids, to derive ɛA for the years 1991–2003, including uncertainties in ɛA that could not be obtained in our earlier work. These uncertainties show that ɛA varies significantly on a timescale of ∼3 years. We employ a new version of a general circulation model to show that the large and variable values of ɛA inferred from the data are not a result of El Niño–Southern Oscillation (ENSO) transport effects. This adds strong support to our earlier hypothesis that a CH4 sink mechanism additional to OH• is required in the MBL and that this sink is probably Cl•. We then show that such a Cl• sink for CH4 would also have to vary significantly on a timescale of ∼3 years to account for the observed ɛA variation. We derive the Cl• concentrations in the MBL that would be required to produce this effect.
2. Experimental Techniques and Results
 Analyses of 13C in atmospheric CH4 are made as described by Lowe et al. . Samples are collected approximately every 2 weeks at Scott Base (SB), Antarctica (78°S, 167°E), and at Baring Head (BH), New Zealand (41°S, 175°E). Methane in the air samples was converted to CO2 by combustion in a platinum furnace. Quantitative conversion during this process ensures that the carbon isotopic composition of the CO2 is the same as that of the parent methane. Using a highly modified 6 inch Nuclide isotope ratio mass spectrometer (IRMS) [Lowe et al., 1994], δ13C measurements of methane in air samples collected from 1989 to 1994 were made. Subsequent measurements used a Finnigan MAT 252 IRMS. The 13C/12C ratios are measured against a laboratory standard, itself calibrated against Peedee belemnite using a quality-controlled protocol as described by Lowe et al. . The overall precision (1σ) of the δ13C determinations in atmospheric CH4 is estimated to be 0.1‰ (1989–1991), 0.05‰ (1991–1995), and 0.02‰ (1996–2003).
 MR measurements were made using HP5890 series II and HP6890 gas chromatographs with a flame ionization detector as described by Lowe et al. . MRs are referenced to CH4 standard reference materials from the U.S. National Institute of Standards and Technology (NIST). These reference materials have been intercompared with reference gases defined on a scale from the U.S. NOAA/Climate Monitoring and Diagnostics Laboratory (CMDL) [e.g., Lang et al., 1992], and our estimate of the ratio of the NOAA/CMDL scale to the NIST values is 0.986 ± 0.001. The 1σ precision of these measurements is ∼3 ppb, or ∼0.2% of ambient MR.
 For the analyses in the present work we have detrended the MR and δ13C data using the seasonal and trend decomposition (STL) procedure described in section 3 and accumulated them in monthly bins to minimize the effects of unavoidable gaps in the observational data. The results are shown in Figure 1 for BH (Figures 1a and 1c) and SB (Figures 1b and 1d). Seasonal cycles are clear at both sites, but there is more variation in the seasonal amplitude of δ13C than of MR. At both BH and SB the MR mean value increases from 1675 to 1740 ppb during the period shown, and the δ13C mean value at both sites is −47.14‰ over that period. Note that there are some missing δ13C data at SB (Figure 1d) resulting from Antarctic logistical problems.
3. Phase Ellipses
 The observational phase ellipses presented by Allan et al. [2001a] were obtained by binning and averaging 4 years of data into 26 points per year and by using the STL procedure of Cleveland et al.  to obtain detrended smoothed seasonal cycles for MR and δ13C. This process is not suitable for the present work as we are interested in changes from year to year. We therefore work directly with the monthly binned data in Figure 1.
Figure 2 shows phase ellipses for the years 1996 and 1999 at BH and SB. We have chosen these years to illustrate ellipses with very different characteristics at BH and also to show how different the ellipses can be between BH and SB in the same year. Ellipses for individual years are much more irregular than the 4 year averaged and smoothed ellipses of Allan et al. [2001a]. Also shown in Figure 2 are straight lines fitted by RMA regression [Sokal and Rohlf, 1981] to the ellipse points, representing approximations to the ellipse major axes from the slopes of which ɛA can be estimated.
 The most recent laboratory measurement [Saueressig et al., 2001] of ɛOH is −3.9 ± 0.4‰. If we take this as the most likely value for the atmosphere and assume that OH• is the only atmospheric sink for CH4, then we would expect the ellipses to provide an apparent KIE of the same magnitude. Comparisons with the KIE lines for ɛOH = −3.9‰ (also in Figure 2) show that the RMA regression lines generally have significantly larger slopes than the ɛOH lines. For 1996 and 1999 at BH, ɛA values derived from the RMA lines are −19.0 ± 3.1‰ and −9.1 ± 2.0‰, while for 1996 and 1999 at SB, ɛA values are −13.4 ± 3.6‰ and −14.2 ± 2.3‰. Other years have similar values, with comparable interannual variability.
4. KIE Time Variation
 The phase ellipse approach is excellent to obtain a visual “feel” for the tilt and shape of ellipses and how these vary with time. However, the irregular “ellipses” in Figure 2 contain variability related to short-term local transport effects as well as the underlying seasonal sink variation. RMA regression includes all variability, and the assumption that the resulting fitted line represents the major axis of the underlying sink-driven ellipse is likely to overestimate the major axis slope and hence the magnitude of the derived KIE. An alternative method that minimizes the effect of short-term transport variability is the fitting of sinusoidal seasonal cycles to the separate detrended time series for MR and δ13C. The underlying model is that the atmospheric sink has a sinusoidal seasonal cycle. This is reasonable, as the OH• fields of Spivakovsky et al.  have an ETSH spatially averaged value that is well represented by a sinusoidal function [see also Allan et al., 2001b]. If the ETSH were an isolated box containing a steady CH4 source of fixed isotope ratio, such a sink would produce a sinusoidal seasonal cycle in MR, and an approximately sinusoidal δ13C seasonal cycle in antiphase with it, giving KIE lines like those shown in Figure 2. We fit the sinusoids to the data points using the Marquardt nonlinear least squares gradient expansion algorithm [Bevington and Robinson, 1992]. The period is fixed at 1 year, and the fitting process allows both the amplitude and phase of the sinusoid to be determined. Uncertainties in the amplitude and phase correspond to the square roots of the diagonal terms in the error matrix.
 The modeling results of Allan et al. [2001a] showed that for any individual CH4 source the “phase ellipse” in the ETSH is always a straight line with slope determined by the KIE of the single atmospheric sink assumed in the model. When sources with different seasonal cycle amplitudes and distinct 13C compositions are combined in the model, the size of the minor axis of the ellipse changes, broadening or narrowing the ellipse. This is equivalent to shifting the phase of the δ13C sinusoid relative to the MR sinusoid. In no case was there any significant change in the slope of the major axis from which the atmospheric sink KIE is derived. Thus we suggest that changes in the amplitudes of the δ13C and MR seasonal sinusoids relate dominantly to sink effects and changes in the relative phase between these sinusoids relate dominantly to source effects.
 Therefore the apparent KIE ɛA can be obtained from the amplitudes of the seasonal cycles fitted to the MR and δ13C time series [Allan et al., 2001a]:
where δ0 is the mean δ13C for the year, ΔMR is the amplitude of the MR seasonal cycle sinusoid, MR0 is the mean MR for the year, and Δδ is the amplitude of the δ13C seasonal cycle sinusoid. An uncertainty estimate for ɛA can then be calculated from the uncertainties in the amplitudes of the two sinusoidal fits by adding the relative uncertainties of the individual amplitudes to give a relative uncertainty for the quotient in (2). Figure 3 shows such fits for the 1996 and 1999 years at BH using the detrended data. The amplitudes of the fitted MR cycles are similar for the 2 years, but the amplitudes of the δ13C cycles are very different. Using (2), the values of ɛA at BH for 1996 and 1999 are −17.4 ± 2.9‰ and −6.1 ± 2.8‰, respectively. Note that as expected, the magnitudes of the ɛA values obtained from nonlinear least squares fitting are somewhat smaller than those obtained from RMA regression, with similar and overlapping uncertainties.
Figure 4a shows ɛA derived at BH for the years 1991–2003, and Figure 4b shows ɛA derived at SB for late 1991–2002. Values have been plotted at the midpoints of the years for which 1 January was the starting point (as in Figure 3). Some years at SB are affected by missing data. For 1991 and 2002 we have shifted the starting point of the fit to make use of the available data and have plotted the value at the appropriate midpoint of the shifted year. Insufficient data were available at SB in 2003 to derive a useful value. There is interannual variation at both BH and SB, but the variation is different at each site. The uncertainty bars for successive years often overlap, but if the values are considered over an interval of ∼3 years, there is a significant change with time. For example, the period 1994–1996 has an apparent KIE of about −15‰, and the period 1998–2000 has an apparent KIE of about −7‰, with well-separated uncertainty bars.
5. Model Simulations
5.1. Global Modeling
 A limitation of the Allan et al. [2001a] analysis is that the TM2 model used only 1987 wind fields, so that the effect of interannual transport variation on the phase ellipse major axis slope could not be obtained. To investigate this and other source and transport effects in detail, we are currently developing a modification of the unified model (UM) [Cullen and Davies, 1991], a general circulation model used for weather forecasting and climate prediction in the United Kingdom. This modification includes the surface emission, atmospheric transport, and in situ destruction of CH4. The dynamical model used is the version 4.5, atmosphere-only, configuration of the UM with a horizontal resolution of 2.5° (latitude), 3.75° (longitude), and 19 vertical levels [Pole et al., 2000]. Two methane isotopic species (12CH4 and 13CH4) are transported by the dynamical model and are chemically destroyed via reaction with an OH• climatology based on Houweling et al. . We intend to include a seasonally varying chlorine sink in the MBL to investigate the detailed effect of this; however, model development and testing will not reach this stage for some time.
 For initial model tests we included surface CH4 sources and the OH• sink from the International Geosphere-Biosphere Program Global Atmospheric Methane Synthesis scenario (GAMeS) (S. Houweling, personal communication, 2001), taking ɛOH = −3.9‰. The CH4 sources (biomass burning, rice cultivation, termites, bogs, swamps, coal, oil, gas, animals, and landfills) are specified at each horizontal grid point of the model, as is a soil sink. The biomass burning, rice, bog, and swamp sources and the soil sink vary on a monthly basis, with the same annual cycle of sources being repeated each year of the model integration. The other sources are constant. The total annual CH4 source is 580 Tg per year. The constant δ13C values for the sources are assigned as specified in the GAMeS scenario, with a weighted mean source δ13C(CH4) of −51.5‰.
 The model was initialized using a zonally uniform CH4 field whose latitudinal variation was empirically fitted to data from the NOAA/Climate Monitoring and Diagnostics Laboratory network (available at ftp://ftp.cmdl.noaa.gov/ccg/ch4/). The details of the empirical fit are not important as the initialization is intended only to speed up the model's convergence to its final steady state. The δ13C(CH4) value was initially set to −47‰ throughout the model domain, this value being close to the current atmospheric average.
 We drove the UM with the composite 4 year El Niño and La Niña sea surface temperature (SST) cycle developed by Spencer et al. . Spencer and Slingo  compared the UM's atmospheric response to a composite of five El Niño and five La Niña events with National Centers for Environmental Prediction/National Center for Atmospheric Research reanalysis data. The UM was forced by observed SSTs and sea ice data taken from the Global Sea Ice and SST data set. Anomalies in sea level pressure, precipitation, 200 hPa velocity potential, and 200 hPa asymmetric stream function averaged over the five composites were compared for December, January, and February and March, April, and May. Good agreement was found between the model and reanalysis fields in the tropics. Some discrepancies in the model's extratropical response were associated with the limited vertical resolution of the 19 level version of the model. Future work will include testing the transport characteristics of a 30 level version of the model.
 Using the 19 level version, we ran the model for 25 composite ENSO cycles (100 years). At this point the model had been in steady state for a considerable time as measured by constant global mean CH4 MR and δ13C. From the results we inferred the tropospheric CH4 lifetime against OH loss to be 9.6 years, with a whole atmospheric lifetime of 8.5 years. Both are consistent with Intergovernmental Panel on Climate Change estimates [Prather et al., 2001].
 We applied the sinusoidal least squares fitting approach described in section 4 to the MR and δ13C model output for each of the 4 years of the 25th ENSO cycle and obtained the following estimates of the apparent KIE ɛA: year 1 (El Niño), −4.2 ± 0.4‰; year 2 (El Niño), −4.4 ± 0.4‰; year 3 (La Niña), −3.9 ± 0.4‰; and year 4 (La Niña), −3.8 ± 0.3‰. There is no significant difference between these estimates, and the overall weighted mean is −4.03 ± 0.18‰, a good estimate of the input KIE ɛOH = −3.9‰. These results show that (1) the KIE for a single atmospheric sink can be derived using the MR/δ13C technique from a completely different transport model to TM2 and (2) changing transport effects during the composite ENSO cycle do not significantly affect the derived KIE. We are therefore confident that the measured variations of the apparent KIE shown in Figure 4 do not arise from spurious variations in the apparent KIE for the OH• sink and so most likely relate to a CH4 sink mechanism additional to OH•.
5.2. Local Box Modeling
Allan et al. [2001b] applied a simple local box model to examine oxidation of 12CH4 and 13CH4 in the MBL. Seasonally varying OH [Spivakovsky et al., 2000] and Cl• sinks were specified as drivers of the system, with the form of the Cl• seasonal cycle in the MBL based on the dimethylsulfide (DMS) seasonal cycle but with a variable seasonal trough-to-peak amplitude ΔCl• in the concentration of Cl•. They presented a plot showing how ɛA increased in magnitude from the chosen OH value of −5.4‰ to larger values as ΔCl• increased. They showed that the addition of a fixed (aseasonal) Cl• sink in the MBL with a Cl• concentration of the order of 103–104 atom cm−3 had only a small effect on the amplitude of the MR seasonal cycle and no effect on the δ13C seasonal cycle. However, they also showed that a Cl• sink with ΔCl• ∼ 7 × 103 atom cm−3 in tandem with a Spivakovsky et al.  OH• cycle (with ɛOH = −5.4‰) could account for an observed ɛA ∼ −13‰. Note that the Cl• concentrations quoted are averages over day-night cycles. Because the production of Cl• is a photochemical process, peak daytime values of Cl• would be more than twice as large as the quoted values.
 We modified the box model of Allan et al. [2001b] to use the more recently published value ɛOH = −3.9‰ rather than the value −5.4‰ used by them. The resulting relationship between ΔCl• and ɛA is given by the following cubic function (in units of 103 atom cm−3) fitted by least squares to the model output:
This is valid in the ɛA range −3.9 to −21.3‰ and assumes that the Cl• sink has the seasonal shape described by Allan et al. [2001b], with an extended minimum between June and October and a sharp maximum peaking in January. The change in ɛOH from −5.4 to −3.9‰ results in an increase of ∼103 atom cm−3 in the ΔCl• required to reach a given value of ɛA. Therefore the results are relatively insensitive to the value of ɛOH used. Figure 5 shows the inferred ΔCl• for BH and SB derived using (3) and the values of ɛA in Figure 4. The error bars were obtained by passing the extreme values of the ɛA error bars in Figure 4 through function (3).
Figure 4a shows that over the period 1991–2003 at Baring Head, apparent KIE values for CH4 derived from the MR/δ13C technique vary considerably with time and are always larger in magnitude than expected from OH oxidation. Although uncertainty ranges tend to overlap from year to year, when considered in periods of ∼3 years, there appears to be a significant variation of the KIE magnitude with time. For example, the period 1994–1996 has an apparent KIE of about −15‰, while the period 1998–2000 has an apparent KIE of about −7‰. The results for Scott Base in Figure 4b are broadly similar but differ in detail (noting that the small apparent KIE in 2002 at SB has large uncertainty).
 If we interpret the apparent KIE values in terms of a possible Cl• sink in the MBL, Figure 5 shows the ΔCl• that sink would require in terms of our simple box model of the MBL. In Figure 5a the inferred ΔCl• at BH has a maximum of ∼104 atom cm−3 in the period 1996–1998 and minima of (<2 ± 2) × 103 atom cm−3 in 1992 and 1999. These latter values are indistinguishable from zero given the size of the uncertainty. Again, although adjacent yearly values have uncertainties that overlap, the periods 1994–1996 and 1998–2000 appear to have distinctly different values of ΔCl•.
 We have no way of estimating the seasonal mean value of Cl•, as our technique is sensitive only to ΔCl•. To the extent that the seasonal maximum is roughly twice the seasonal mean, Allan et al. [2001b] suggested that half the maximum value of ΔCl• could give a rough indication of the seasonal mean of Cl•. If so, the BH 1996–1998 maximum would give a Cl• seasonal mean concentration of about 5 × 103 atom cm−3. The largest inferred ΔCl• at SB in Figure 5b is similar to that at BH, but the smallest reliable value (excluding 2002, which has a zero value with large uncertainty) is about 3 × 103 atom cm−3, giving a seasonal mean of about 1.5 × 103 atom cm−3. Because free tropospheric measurements suggest that the observed enrichment of 13C in CH4 may spread throughout the troposphere, it is possible that the Cl• mean concentration in the MBL could be significantly larger than the values suggested above. This was discussed by Platt et al. . They estimated that for the enrichment of 13C in CH4 to be spread uniformly through the free troposphere, an increase by a factor of ∼6 in the Cl• mean concentration in the MBL would be required. This implies that the BH seasonal mean for 1996–1998 might actually be equivalent to about 3 × 104 Cl• atom cm−3 in the MBL. Although this value is relatively large, it is within the range of values quoted in the literature [e.g., Graedel and Keene, 1995]. Using tetrachloroethylene as an indicator, the global approach of Singh et al.  inferred Cl• concentrations in the MBL to be <0.5–1.5 × 104 atom cm−3. Note also that we infer seasonal mean values of ∼104 Cl• atom cm−3 in the period 1998–2000 using the troposphere factor of 6 above. The reason for the apparent variation of MBL Cl• concentrations on a timescale of ∼3 years is at present unknown. We intend to investigate these ideas in detail using the unified model described in section 5.1.
 The values of ΔCl• discussed above were derived under the assumption [Allan et al., 2001b] that the seasonal cycle of Cl• was closely related to the form and phasing of the DMS seasonal cycle measured at Cape Grim, Tasmania [e.g., Ayers et al., 1995]. This follows from the possibility that oxidation products of DMS in the MBL could acidify sea-salt droplets and allow the acid-catalyzed release of Cl• [e.g., Vogt et al., 1996]. This process is quite likely to occur over regions of the Southern Ocean [e.g., Ayers et al., 1999]. However, acids other than H2SO4 may be important in other regions, for example, HNO3 and carboxylic acids in the vicinity of Amsterdam Island in the southern Indian Ocean [Moody et al., 1991]. The exact form of the seasonal cycle of Cl• at middle to high latitudes is therefore uncertain. However, the evolution of Cl• in the MBL is a photochemical process, so a seasonal cycle of significant amplitude at middle to high latitudes seems likely, whatever the acidifying mechanism. Allan et al. [2001b] showed that the ΔCl• inferred from δ13C measurements was relatively insensitive to the exact form and phasing of the Cl• seasonal cycle, although for given ΔCl• a cycle related to that of DMS somewhat reduced the Cl• seasonal mean concentration compared with a sinusoidal cycle.
 We have presented CH4 mixing ratio and δ13C data measured at Baring Head and Scott Base over the years 1991–2003. From these, seasonal phase and amplitude relationships can be derived, and hence the “apparent kinetic isotope effect (KIE)” of the CH4 atmospheric sink can be obtained. Derived time series of this apparent KIE suggest considerable variability and greater magnitude than can be accounted for by OH• as a lone sink. The uncertainties in our results are such that the changes appear to be significant over periods of ∼3 years. In 1994–1996 the apparent KIE was about −15‰, and in 1998–2000 it was about −7‰. We have shown using a global transport model that the technique we use to derive the apparent KIE should be quite accurate and that the KIE variation with time is unlikely to arise from ENSO transport effects. We infer that an atmospheric sink in addition to OH• is required.
 Assuming that this extra sink is Cl• in the MBL, we have derived the amplitude of the Cl• concentration seasonal cycle that would be required to give the observed effect in an isolated MBL. This amplitude ranges from 104 atom cm−3 in 1994–1996 to about 3 × 103 atom cm−3 in 1998–2000. If the measured enrichment of 13C in CH4 extends throughout the free troposphere, the seasonal mean concentrations of Cl• required in the MBL would be about 3 × 104 Cl• atom cm−3 in 1994–1996 and ∼104 Cl• atom cm−3 in 1998–2000.
 We thank the reviewer for detailed comments that have improved this paper. We thank Hilary Spencer (University of Reading) for providing the composite El Niño and La Niña SST data and Jeff Cole (University of Reading) for help with inclusion of surface tracer emissions in the UM. We thank Sander Houweling (National Institute for Space Research, Netherlands) for providing the GAMeS scenario fields. We thank Keith Lassey (National Institute of Water and Atmospheric Research) for many helpful comments and a critical reading of the manuscript. This work was supported by the New Zealand Foundation for Research, Science and Technology under contract C01X0204.