A gas chromatography–mass spectrometry (GC–MS) system, designed to make high precision measurements of pmol mol-1 level compounds in small air samples, was used for this work. A pre-concentration system with an Agilent 6890 GC was coupled to a VG/Waters EBE tri-sector mass spectrometer. Details of previous work conducted using this system can be found in Laube et al. (2010b) and Sturges et al. (2012).

Magnesium perchlorate (Mg(ClO4)2) was used to dry the air introduced to the GC-MS system. Trace gases from 0.2 L (293 K, 1 bar) of air were pre-concentrated on Hayesep D (80/100 mesh), which was cooled to -78 ∘C by immersion in a dry ice/ethanol mixture. The sample was then trapped on a 49 m GS-GasPro column (internal diameter 0.32 mm) at -10 ∘C for 2 min. Heating to 200 ∘C at 10 ∘C min-1 released the trapped compounds to the MS according to their retention strength.

The MS was operated in EI-SIR (electron impact-selected ion recording) mode at a mass resolution of 1000. The fragment ions measured were C(35Cl)2F+ (m/z 101), C35Cl37ClF+ (m/z 103) and C(37Cl)2F+ (m/z 105).

In this study, isotope deltas (δ), expressed in per mill (‰), are used to denote the relative 37Cl / 35Cl ratio difference of CFCs in sample air with respect to a standard, δ(37Cl)=RsampleRstandard-1, where R represents the 37Cl / 35Cl abundance ratio of a standard or sample. There is no internationally recognised isotope standard for these measurements. For this reason, all δ values in this study are relative to a laboratory standard (AAL-071170). This is a background air sample collected in 2006 at Niwot Ridge, Colorado, by the National Oceanic and Atmospheric Administration (NOAA).

This study considers isotopologue current ratios of the CCl2F+ fragment, rather than directly measuring Cl+ ions. Equation (2) describes how the C(35Cl)2F+ (m/z 101), C35Cl37ClF+ (m/z 103) and C(37Cl)2F+ (m/z 105) fragments can be used to calculate the 37Cl / 35Cl isotope ratio from the CCl2F measurements, R=A(37Cl)A(35Cl)=A(C35Cl37ClF+)A(C35Cl2F+)+2A(C37Cl2F+)A(C35Cl2F+)2+A(C35Cl37ClF+)A(C35Cl2F+), where A represents the abundance of a given fragment. The C(37Cl)2F+ fragment was included in the analysis of 172 out of 448 samples. In all cases, δ values calculated using ratios derived from Eq. (2) agreed within 1σ with δ values calculated directly from C35Cl37ClF+ / C(35Cl)2F+ ratios (without the inclusion of the C(37Cl)2F+ fragment). We use the linear regression slope of the C35Cl37ClF+ against the C(35Cl)2F+ ion current over the duration of the peak to derive R and, based on this, δ (Eq. 1). Typically, two sample measurements were made between standard runs, allowing instrumental drift to be quantified and corrected for.

No chromatographic interferences were found for these ions at the retention times corresponding to CFC-11 and CFC-12. However, CFC-113a has grown to a current atmospheric abundance of ∼ 0.5 pmol mol-1 since its emergence in the 1960s (Laube et al., 2014) and partially co-elutes with CFC-113. Interference from CFC-113a can be detected above the surrounding noise in one sample. This peak was manually excluded from the linear regression slope and a δ value calculated using Eq. (1). Within 1σ, this value is the same as the original δ value calculated, without removing the CFC-113a interference.

Poor-quality measurements were defined as a measurement with an Rsample standard deviation (from repeat measurements) of more than 3 times the average standard deviation of the Rstandard values during an analysis period. Less than 1 % of the measurements were rejected as poor quality. Measurements were also rejected if there was uncertainty regarding the integrity of the sample; small system leaks can be identified through the analysis of other trace compounds (e.g. sulfur hexafluoride, SF6). Two percent were excluded due to a suspected lack of sample integrity.

A static dilution series was prepared by diluting background air with research-grade nitrogen. These samples were used to quantify the response behaviour of the analytical system. Nonlinear isotope ratio responses were found at small mole fractions and were corrected (details in the Supplement). Ninety-nine percent of the measurements were unaffected by this nonlinearity and did not have to be corrected. The smallest peaks measured for each species define the effective detection limit for this study. One CFC-113 sample was below this limit and rejected.

The stratospheric samples were used to calculate εapp for each species. This was done by assuming a Rayleigh-type mechanism and applying this linear relationship to the data (Kaiser et al., 2006): ln⁡1+δ37Cl≈εappln⁡(yyT), where y and yT are the stratospheric and tropospheric mole fractions, respectively.

A statistical “bootstrap” analysis technique was used to determine εapp values and their uncertainties (Volk et al., 1997). In this technique, a data pool is created by describing each sample with three values (the measured δ value and this value ± 1σ). The bootstrap tool draws 500 random samplings from the complete stratospheric data pool (including the possibility of drawing the same value repeatedly) to produce an overall εapp value and a robust associated uncertainty range. Anomalous stratospheric samples were omitted using an iterative least squares method, reducing the εapp standard error in all cases. Any sample more than 2 times the standard error of the ln⁡(1+δ) estimate away from the linear regression of the Rayleigh fractionation plot was removed. This process was repeated until all data points fell within 2 times the standard error of the ln⁡(1+δ) estimate. Of the measurements, 21 % (CFC-11), 10 % (CFC-12) and 7 % (CFC-113) were rejected using this method.

Following Kaiser (2002) and Röckmann et al. (2003), a two-box model of the atmosphere is used to project the changing tropospheric isotope signature. The following mass balance equations hold for the model atmosphere, partitioned into a stratosphere (index S) and a troposphere/lowermost stratosphere (index T), where chemical loss only occurs in the stratosphere nTdyTdt=E-FTS+FST=E-F(yT-yS),nSdySdt=FTS-FST-L=FyT-yS-JnSyS. The symbols have the following meaning: n – amount of air (in mol), y – mole fraction (in mol mol-1), F – bulk air flux between troposphere and stratosphere and vice versa (in mol a-1), FTS – CFC flux from troposphere to stratosphere (in mol a-1), FST – CFC flux from stratosphere to troposphere (in mol a-1), E – trace gas emissions (in mol a-1), L – trace gas loss (in mol a-1) and J – loss rate coefficient (in a-1).

The global average lifetime is calculated as the ratio of total burden and total loss: τ=nSySnTyTJnSyS=nSyS+nTyTFyT-yS-nSdySdt. The parameters nT, nS and F are taken from Holton (1990), Trenberth and Guillemot (1994) and Appenzeller et al. (1996). Using the tropospheric mole fraction history yT(t) (Velders and Daniel, 2014) and an estimate of the global average lifetime τ, Eqs. (4), (5) and (6) can be solved for yS, J and E. In practice, we adjust J so that the average τ for the years 1999 to 2008 equals the lifetime estimates as per SPARC (2013), and we use a fixed value of J for all model years.

To derive the corresponding isotope budget, we write Eqs. (4) and (5) for another isotopologue (primed quantities): nTdyT′dt=E′-FyT′-yS′, which can be written in δ notation (relative to an arbitrary reference material). We have yT′ = yT(1+δT)Rstandard, yS′=yS(1+δS)Rstandard and yP′=yP(1+δP)Rstandard. Substituting this into Eq. (7) gives Eq. (8): nTdyT(1+δT)dt=E1+δE-FyT1+δT-yS1+δS. Together with Eq. (4) this gives the tropospheric isotope budget: nTyTdδTdt=EδE-δT+FySδS-δT. The equivalent equation for the stratosphere is nSySdδSdt=FyTδT-δS-nSySJεJ(1+δS), where the photochemical isotope fractionation is εJ=J′J-1, with J′ being the loss rate coefficient for the isotope in the numerator of the isotope ratio (here, 37Cl).

εJ (or J′) are not known, so again following the analogue of N2O (Röckmann et al., 2003), we use the observed apparent Rayleigh isotope fractionation in the stratosphere (εapp), to solve the isotope-specific part of the budget. Based on this apparent Rayleigh isotope fractionation, the relative isotope ratio difference between the stratosphere and the troposphere δST can be calculated according to δST=1+δS1+δT-1=δS-δT1+δT=-δTS1+δTS=ySyTεapp-1. εapp is the apparent stratospheric isotope fractionation, which is empirically determined for the years 1999 to 2008 (reflecting the time of stratospheric sample collection and the age of stratospheric air). It depends on both chemistry and transport (Kaiser et al., 2006), which may have changed over the period of CFC emissions. εJ is a model parameter that is tuned to give εapp for the corresponding model years. It has no immediate physical interpretation and is only meaningful in the context of the chosen two-box model. We adjust εJ so that the average εapp value for the years 1999 to 2008 equals our stratospheric observations and derive the projected atmospheric history of δT and δS from Eqs. (10) and (12), assuming a constant δE value that results in δT(2006)=0, corresponding to the year our tropospheric air standard was filled.

The migration of gases from the atmosphere through firn is largely controlled by diffusional and gravitational effects, such that less diffusive gases move through the firn more slowly than more diffusive gases and heavier gases and isotopologues are enriched at depth relative to lighter gases and isotopologues. Consequently, there is no unique “age-of-air” at a given depth in the firn, rather an age distribution is calculated for a specific gas or isotopologue. Also, a gas can undergo significant isotope fractionation during its movement through the firn. A model of gas transport in firn must therefore include both diffusive transport and gravitational separation to reconstruct changes in atmospheric abundances and isotope ratios over time. These reconstructions have been accomplished here using the model developed at the Laboratoire de Glaciologie et Géophysique de l'Environnement (LGGE) and Grenoble Image Parole Signal Automatique (GISPA-lab). The physical basis of the model is described in Witrant et al. (2012). The model has been further refined to provide a new and more robust definition of the optimal solution, and a more rigorous treatment of isotopes based on a forward firn model written in terms of δ values (Witrant and Martinerie, 2013). The diffusive behaviour of gases is initialised for each firn column by using a set of gases with well-known atmospheric histories (Witrant et al., 2012).

More details on the firn modelling are given in the Supplement. The dates shown here for the firn data are based on median ages for each gas. The age spectrum associated with these median ages is also discussed in the Supplement. When plotting mole fractions and isotope δ against median age, it is necessary to correct for gravitational and diffusive fractionation. In the case of mole fractions, gravitational settling is calculated as the difference between normal model runs and runs in which gravity is set to zero. For within-firn isotope fractionation, a correction (of 1–2 ‰) is calculated using the reconstructed scenario of the gas mole fraction and assuming a constant atmospheric isotope δ. After these corrections, any discrepancies between projected and measured firn isotope δ reflect changes in atmospheric isotope δ over time.

Thermal diffusion was not included in the model, despite an observed gradient of 2 to 3 ∘C in the firn column at NEEM, due to recent local warming in northwest Greenland (Carr et al., 2013). This is because although the thermal diffusion effect on the CFC mole fractions is on the order of 0.2 % at NEEM, the effect on δ(37Cl) is estimated to be less than 0.02 ‰ (Leuenberger and Lang, 2002).

After outlier correction (Sect. 2.2), a total of 31 and 28 (CFC-11), 33 and 54 (CFC-12) and 36 and 54 (CFC-113) measurements remain, for samples collected in the mid- and high-latitude stratosphere, respectively.

Negative apparent isotope fractionations (εapp) have been calculated for all three species (Fig. 1). This indicates that sink reactions discriminate against heavier isotopes, in other words, the heavier isotopologue is broken down more slowly, causing the observed enrichment in 37Cl. Trend lines in Fig. 1 correspond to apparent fractionations (εapp) of -2.4 (0.5) and -2.3 (0.4) ‰ (CFC-11), -12.2 (1.6) and -6.8 (0.8) ‰ (CFC-12) and -3.5 (1.5) and -3.3 (1.2) ‰ (CFC-113) for mid- and high latitudes, respectively. As the sampling point moves from mid- to high latitudes, the absolute CFC-12 εapp value decreases. This effect is qualitatively the same for N2O (Kaiser et al., 2006). CFC-11 and CFC-113 both show mid- and high-latitude εapp values that are indistinguishable within 1 standard deviation. It could be that the latitude dependence seen in CFC-12 and N2O also exists in these compounds, but is obscured by the size of analytical uncertainties, compared to the relatively small isotope changes being tracked.

Laube et al. (2010a) calculated the tropical CFC-12 εapp to be -12.1 (1.7) ‰, which agrees with the mid-latitude value of -12.2 (1.5) ‰ presented here. Given the lower εapp value calculated from high-latitude samples, a difference between εapp values at mid-latitudes and the tropics might be expected. The relative difference between mid-latitude/tropical and polar εapp is much larger for CFC-12 than for N2O. Due to the similar atmospheric lifetimes of CFC-12 and N2O, this is unlikely to be due to dynamic effects as suggested for N2O (Kaiser et al., 2006) and is at present unexplained. In the absence of globally representative εapp values, we use the mid-latitude values to project the tropospheric isotope δ changes over time.

Based on historical and predicted atmospheric mole fractions (Velders and Daniel, 2014), our two-box approach estimates global emissions for CFC-11, CFC-12 and CFC-113 from 1950 to 2050. Figure 2 shows good agreement between these values and previously published estimates (Velders and Daniel, 2014). The proliferation of CFC production and consumption caused an initial sharp increase in emissions, followed by sustained high emissions during the 1970s and 1980s. The Montreal Protocol and its subsequent amendments ensured a significant reduction in CFC production, causing the concurrent decrease in emissions of all three species at the end of the 1980s.

Figure 3 shows how the tropospheric δ(37Cl) of these three compounds would have changed with time, assuming no source variations. From their first release until the present day, tropospheric CFC-11, CFC-12 and CFC-113 δ(37Cl) values are predicted to have increased by 1, 3 and 1 ‰, respectively. Maximum rates of increase are predicted to have occurred since 1990, which coincide with the sharp reduction in CFC emissions due to the introduction of legislation phasing out their production and consumption (Montzka et al., 2011). A drop in the anthropogenic input means that the existing atmospheric pool will become increasingly enriched, as demonstrated by the predicted acceleration of tropospheric δ(37Cl) increase after 1990.

The shading in Fig. 3 represents the uncertainty envelope. Approximately equal contributions to the uncertainty arise from (1) the differing reference interfaces for stratosphere–troposphere exchange used by Holton (1990) and Appenzeller et al. (1996), namely the 100 mbar and the 380 K isentrope, respectively; (2) the 1σ uncertainty in εapp values; (3) the range of “most likely” lifetime values from SPARC (2013).

Using the GC-MS instrument discussed in this work, Laube et al. (2010a) verified that their isotope ratio measurements were not biased by the response behaviour of the analytical system. They measured CFC-12 to a minimum of 77 pmol mol-1. The present study covers a wider range of CFC-12 mole fractions and also includes CFC-11 and CFC-113, meaning that a more extensive effort was necessary to remove the possibility of a bias in this work (details in the Supplement).

From three independent data sets, a total of 44 (CFC-11), 74 (CFC-12) and 48 (CFC-113) δ(37Cl) measurements are presented, covering the last 60, 50 and 45 years, respectively (Fig. 3). In general, there is good agreement between the Cape Grim and firn air measurements. This agreement is in line with expectations from these well-mixed gases. The predicted trends for all three species are small (Fig. 3), with the largest isotope changes expected to have occurred in the last 2 decades. From 1970 to the present day, the projected changes can be accounted for within analytical uncertainties (Fig. 3). This means that for these three CFCs, our observations are consistent with an isotopically invariant source signature, but the high measurement uncertainty does not allow us to preclude the possibility that it has changed over time. Despite the noted parallels in atmospheric chemical behaviour between CFC-12 and N2O, their atmospheric isotope δ histories differ because N2O has been present in the atmosphere for at least 800 000 years longer than CFC-12 (Spahni et al., 2005; Schilt et al., 2010). When anthropogenic emissions began, atmospheric N2O isotope δ were perturbed from a near steady-state source-sink system, whereas the industrial release of CFC-12 was not preceded by any atmospheric burden. This has caused δ(15N, N2O) and δ(18O, N2O) to decrease over the last century (Röckmann et al., 2003), while δ(37Cl, CFC-12) is predicted to have increased slightly (Fig. 3).

Before 1970, there were two (CFC-11), one (CFC-12) and two (CFC-113) measurements between 5 and 10 ‰ higher than expected, which could represent source δ(37Cl) changes. Emissions of these gases earlier in the record represent a larger proportion of the total atmospheric load, meaning that our measurements are more sensitive to source δ(37Cl) changes at this time. As the oldest samples, these five measurements contain the lowest CFC mole fractions, producing higher-than-average analytical uncertainties. Also, a small correction was applied to four of them, introducing an additional error (details in the Supplement). The limited number and precision of these measurements make it premature at this stage to report a source δ(37Cl) change in these gases.

Zuiderweg et al. (2013) describe how the manufacturing processes used to synthesise chlorofluorocarbons have altered through industrial advances in the 20th century. The most significant change occurred in the production of the main feedstock (CCl4). Traditionally it was synthesised through the chlorination of CS2, but in the last 50 years CH4 has been used in this chlorination. Zuiderweg et al. (2013) cite these methodological changes as the most likely cause of the significant 13C enrichment of CFC-12, but there is no direct evidence to confirm that changes in production processes have changed the isotope signature of the resultant CFCs.