A record of atmospheric halocarbons during the twentieth century from polar firn air

Measurements of trace gases in air trapped in polar ﬁrn (unconsolidated snow) demonstrate that natural sources of chloroﬂuorocarbons, halons, persistent chlorocarbon solvents and sulphur hexaﬂuoride to the atmosphere are

Chloro¯uorocarbons (CFCs) and other halocarbons have been used as refrigerants, propellants and solvents since the early to middle twentieth century. Both leakage and direct emission, in combination with long atmospheric lifetimes, have resulted in a steady accumulation of these gases in the atmosphere, allowing them to increase the burden of atmospheric chlorine (Cl) from ,0.5 nmol Cl mol -1 (that is, 0.5 parts per billion, p.p.b.) to a peak of over 3.5 nmol Cl mol -1 by 1992±94 1,2 . Suggestions that a build-up of CFCs and long-lived halocarbons in the atmosphere could result in the depletion of stratospheric ozone prompted the need for systematic measurements 3 , but only in the mid-1970s were routine atmospheric monitoring programmes begun 4,5 . Industrial production and emission data, together with atmospheric lifetimes, have been used to estimate the atmospheric burden of CFCs before their detection and measurement 6,7 . Until now, however, no data have been available for testing these early estimates.
Natural sources have been suggested for many of the halocarbons involved in the depletion of stratospheric ozone 8 . Atmospheric methyl chloride (CH 3 Cl), for example, is believed to be predominantly of natural origin, although it does have a few anthropogenic sources and there appear to be unidenti®ed sources of this gas 9 . The division of methyl bromide (CH 3 Br) emissions into natural and anthropogenic components has generated considerable scienti®c and regulatory controversy during recent years 10,11 . Some studies have suggested that natural sources may be signi®cant for atmospheric CCl 4 (ref. 12). Detection of CFCs in air sampled near volcanoes has lead to the suggestion that volcanism might be a signi®cant source of these gases in air 13 . The general absence of CFCs in the ocean's deep waters virtually precludes this possibility, but no reliable measurements in air taken before the onset of anthropogenic emissions have been made to date. Air samples from ice cores have been too small to obtain the precision necessary to address these questions for halocarbons. Previous collections of ®rn air either were contaminated during sampling 14 or extended back no more than a few years 15 .
Here we use samples of air collected from ®rn (unconsolidated snow; Table 1) at the South Pole and Siple Dome in Antarctica and at Tunu in Greenland to obtain high-precision measurements spanning the full history of anthropogenic emission of most halocarbons (data reported here are available; see Supplementary Information). Samples of ®rn air from all three sites were analysed for CCl 3 F (CFC-11), CCl 2 F 2 (CFC-12), CH 3 CCl 3 , CCl 4 , N 2 O, SF 6 , CBrF 3 (halon H-1301), CBrClF 2 (halon H-1211), CH 3 Br and CH 3 Cl. Newly introduced hydro¯uorocarbons (for example, CF 3 CH 2 F or HFC-134a), hydrochloro¯uorocarbons (such as CH 3 CCl 2 F or HCFC-141b, CH 3   effects, and in some instances chemical effects, the composition of ®rn air at any given depth does not correspond exactly to the atmospheric composition at some time in the past. Nevertheless, because the behaviour of each gas in the ®rn is governed primarily by its own diffusivity and molecular mass, we are able to estimate histories of both anthropogenic and naturally occurring halocarbons during a time of rapid population, agricultural and industrial growth.

Concentration±depth pro®les of trace gases
Firn is a porous matrix in which constituent gases diffuse at rates governed by both the structure of the ®rn and the nature of the species in question. Pro®les of d 15 N values of N 2 versus depth in the ®rn (Fig. 1) re¯ect the processes governing transport and storage of air 19,20 . Below the in¯uence of seasonal temperature gradients and other surface-related effects (roughly 30 m), d 15 N of N 2 increases linearly with depth, owing to gravitational settling. The linearity indicates a porous column in which transport is governed predominantly by molecular diffusion 21,22 . The interval of constant d 15 N of N 2 at the base of the ®rn shows the lock-in zone 23 , where dense winter layers substantially impede vertical diffusion, while summer layers retain enough open porosity to allow sample collection. Gases within the porous column atop the lock-in zone communicate with the atmosphere through diffusion, creating a smoothed record of atmospheric changes. In contrast, vertical diffusion is considerably restricted within the lock-in zone, which tends to isolate this air from the diffusive column. Thus, air in the lock-in zone ages at approximately the same rate as the surrounding ice. At Siple Dome and Tunu, where the diffusive column is relatively short and smoothing is minimal, the composition of the overlying atmosphere is recorded in the lock-in zone without extensive diffusive mixing. At the South Pole, the deeper diffusive column signi®cantly smooths the atmospheric history, but also allows for collection of very old samples 16 . By comparing results from these sites with different diffusive structures, we are able to reconstruct robust atmospheric histories. Variations in concentration with depth re¯ect both long-term trends and recent perturbations of trace gases in the atmosphere (Figs 2±4). Concentrations of compounds known to have sig-ni®cant anthropogenic sources over the past few decades (for example, CFCs, solvents and halons) decrease considerably with depth, where the ®rn air is older. Concentrations of CFC-11, CFC-12, CFC-113, H-1301, H-1211 and SF 6 all decreased to levels that did not differ signi®cantly from zero by the bottom of the depth pro®les (Fig. 2). This indicates that these gases were not present in the atmosphere during the early twentieth century and con®rms their exclusively anthropogenic origin. Previously, the lowest values reported for CFC-11 and -12, also from ®rn air measurements, were 16 and 18 pmol mol -1 (that is, parts per trillion or p.p.t.) 14 . The earliest direct measurements of atmospheric CFC-11 and -12 date from the early to mid-1970s, with values of roughly 50 and 200 pmol mol -1 (refs 12, 24±26). CFC-113 has been measured in archived air from Cape Grim, Australia, as far back as 1978, for which a value of 13 pmol mol -1 was reported 27 . Systematic SF 6 measurements began in 1978 at levels of 0.6 pmol mol -1 , or about 15% of the current atmospheric burden 28 , although the lowest reported value was for 1976, at 0.24 pmol mol -1 (ref. 29). Early measurements of the halons are also available since 1976, with amounts of the order of 0.5 pmol mol -1 (ref. 30).
The concentrations of two abundant chlorocarbons, CH 3 CCl 3 and CCl 4 , also decreased rapidly with depth ( Fig. 3). Both of these gases reached concentrations near the bottom of the pro®les that were at or near detection limits, suggesting that natural sources are insigni®cant and that both gases were essentially absent in the pretwentieth-century atmosphere. The detection limit for CH 3 CCl 3 was ,1.5 pmol mol -1 , although for CCl 4 it was ,5 pmol mol -1 because of a co-eluting peak on the gas chromatography/electroncapture detection (GC/ECD) system (see Supplementary Information). Thus natural sources, if they are signi®cant at all, could not be responsible for more than 5 pmol mol -1 of CCl 4 in the atmosphere. Of course, this requires that CCl 4 is stable in ®rn air, a question that is raised in our attempts to reconcile an atmospheric history of CCl 4  from both Siple Dome and South Pole pro®les. Although absolute calibration was a concern for some of the early real-time measurements of these two gases 12,24 , values from the late 1970s were reported at around 60 pmol mol -1 for CH 3 CCl 3 and 90 pmol mol -1 for CCl 4 , after normalizing to recent calibration scales 2,31 . Also apparent in the data from Siple Dome (Fig. 3) and Tunu is the 1991±93 turnaround of CH 3 CCl 3 in the atmosphere that resulted from decreased anthropogenic emissions 1,31 . Those HCFCs and HFCs that have been used only recently appear exclusively in the upper layers of the ®rn, as they have not been present in the atmosphere long enough to have been isolated in the lock-in zone 1 at concentrations above our detection limits (Fig. 2).
Concentrations of CH 3 Cl are signi®cant in the deepest samples, and increase gradually toward the surface of the ®rn (Fig. 3). This is evidence that this gas has both natural and anthropogenic sources or at least sources that existed before the twentieth century. The concentrations of CH 3 Cl in samples collected between 20 m and 45 m depth are similar to the mean annual concentrations at Cape Grim, Tasmania, in recent years 18 . The decrease of 15± 20 pmol mol -1 observed in the upper 10±12 m of the ®rn is consistent with the seasonal variability observed for CH 3 Cl in the Southern Hemisphere, as lower mixing ratios are observed in summer when atmospheric loss processes outweigh sources. Concentrations between 20 and 45 m depth are about 10% higher than at the bottom of the pro®le. If we assume that CH 3 Cl is not being degraded slowly over time in the ®rn, these results indicate that the atmospheric mole fraction of CH 3 Cl has increased by about 40 pmol mol -1 over the past century.
The ®rn data for CH 3 Br are more dif®cult to interpret. At the two Antarctic sites, CH 3 Br concentrations at the top of the ®rn air pro®les are 20±25% higher than at the bottom, implying a sig-ni®cant pre-twentieth-century source and an increase in the atmospheric burden of this gas over the past 50±100 years (Fig. 4). Unlike CH 3 Cl, however, the observed difference of 5±10% between the surface samples and those below 10 m is not explained by seasonally varying mixing ratios; no signi®cant seasonality is observed in the atmosphere at sites in the Southern Hemisphere 32 . This anomaly remains a mystery, as it was present at both Antarctic sampling sites and appears not to be a sampling or analytical artefact. More disturbing, however, are the samples from Tunu. At this seasonally warm and coastally in¯uenced Greenland site, the concentration of CH 3 Br was very high near the bottom of the pro®le, reaching mixing ratios of nearly 50 pmol mol -1 at the ®rn±ice transition (Fig. 4). We are con®dent that this increase in concentration at depth is not an artefact of sample collection, storage or analysis. CH 3 Br±depth curves were essentially identical in our two Tunu ®rn-air pro®les, although the second pro®le did not extend as deeply as the ®rst. Tests of the sampling equipment showed no evidence whatsoever of contamination by CH 3 Br (CFCs were present in our equipment, but our sampling design and procedure still allowed for concentrations of zero or nearly zero in deep samples.) Samples in steel and glass¯asks yielded similar pro®les for CH 3 Br and CH 3 Cl, and chromatographic results, obtained by both electron-capture or mass spectrometry, were independent of analytical technique. This leads us to believe that the observed high values (up to 48 pmol mol -1 ) for CH 3 Br in the ®rn at Tunu are real, although not necessarily of atmospheric origin. Concentrations of other marine, biogenic gases (such as CH 2 Br 2 and CH 3 I) also increased at depth in the Tunu ®rn, although the anomalies were smaller than for CH 3 Br.   density±porosity±diffusivity relationships 19 . These relationships, however, are poorly constrained, so we prefer to calibrate diffusivity with the measured atmospheric history of CO 2 deduced from real-time¯ask samples and the Law Dome ice core 33 . We refrain from modelling the Tunu pro®le because of the extensive convective zone and the possibility that the ®rn-air CO 2 pro®le at that site is erroneously offset by covariance between seasonal convective mixing and seasonal variations in atmospheric CO 2 (`recti®er' effects). To account for the effects of diffusion in the ®rn on each species at the South Pole and Siple Dome, we use the simple onedimensional ®rn diffusion model described in detail by Battle et al. 16 . As in that investigation, we infer a site-speci®c diffusivity± depth relationship by adjusting diffusivity until the model reproduces the observed CO 2 ®rn-air pro®le when driven by the atmospheric CO 2 history. We then calculate halocarbon±depth pro®les using this adjusted diffusivity±depth pro®le in the diffusion model. However, we examine these as halocarbon±CO 2 rather than halo-carbon±depth pro®les; the former are much less sensitive to errors in the inferred pro®les of diffusivity versus depth. Our results remain sensitive to errors in the diffusivities of halocarbons relative to the diffusivity of CO 2 .
We apply the ®rn diffusion model with one of two goals, depending on the halocarbon: (1) to test an independently derived atmospheric history, or (2) to infer an atmospheric record for species with uncertain or unknown histories. The ®rst approach is suitable for gases such as the CFCs, halons and SF 6 , for which atmospheric histories have been derived from emission estimates and real-time measurements. We use these histories and the ®rn-air diffusion model to predict concentration±CO 2 pro®les for these gases and to compare the predictions with observations in the Antarctic ®rn. The second approach is used for those chlorocarbons and methyl halides for which there is insuf®cient information to estimate or calculate complete atmospheric histories. In this approach we choose one or more suitably general mathematical descriptions of concentration as a function of time, with consideration given to recent direct atmospheric measurements as well as ®rn-air data. The free parameters of the mathematical expressions are then adjusted using inverse methods to optimize agreement between measured and modelled concentration±CO 2 pro®les at one or both sites.
We cannot eliminate the possibility that these gases are very slowly produced or degraded in the ®rn. However, we have strong  evidence that such production or degradation is not signi®cant for most of them. Agreement of emission records with the ®rn model for those gases with available data provides one level of assurance. Consistent histories inferred from measurements at Siple Dome and the South Pole provide additional assurance; Siple Dome and the South Pole are sites suf®ciently different in ®rn properties that any physical, biological or chemical effects on the gas concentrations should be expressed differently at these two sites.

Tests of independently derived atmospheric histories
Concentration±CO 2 pro®les predicted for CFCs, halons and SF 6 based on emission histories 6,34 , real-time measurements 1,35±37 and ®rn-air diffusion modelling are in good agreement with ®rn-air data (Fig. 5). This agreement, together with zero or near-zero concentrations in the oldest samples, is consistent with solely anthropogenic sources of these gases. It also suggests that the emission models used 7,36 yield reasonably accurate estimates of the early atmospheric histories. The one signi®cantly inconsistent species is halon H-1211. Both modern measurements and ®rn results suggest either a shorter lifetime than that given solely by atmospheric chemistry models (11 compared to 20 years) or lower emissions 37,38 . The atmospheric curve and model output in Fig. 5 incorporate the 11-year lifetime. For the CFCs, for which signi®cant emissions began in the 1940s, analyses of ®rn air provide data for long periods where few or no real-time measurements are available 36 . For the halons, with sig-ni®cant emissions beginning in the 1960s, real-time measurements span a larger fraction of their atmospheric history, although calibrations of earlier measurements are less certain 30,37 . For SF 6 , with signi®cant emissions beginning in the 1970s, the concentration history is derived almost entirely from direct atmospheric measurements 28,35 , and thus is insensitive to errors in estimates of emissions or lifetimes.

Inferred atmospheric records of chlorocarbons
CCl 4 appears to be derived solely from anthropogenic sources, but this conclusion must be tempered by the analytical uncertainty of 65 pmol mol -1 for the lowest-concentration samples, owing to an interfering peak in the GC/ECD system. This inference is based primarily upon the longer South Pole ®rn-air record (Fig. 6), although the concentration of CCl 4 also does not differ from zero at the bottom of the Tunu pro®le. The ®rn data do show that CCl 4 has been present in the atmosphere for a longer time than the CFCs, owing to its earlier industrial use. This earlier use also appears in the deep ocean record where CCl 4 is present in the relative absence of CFCs 39 . Although our data from Siple Dome suggest that, before 1960, CCl 4 was substantially lower in concentration than predicted from estimated production and emissions 40 , the data from the South Pole agree quite well with the emission prediction. Emission estimates for CCl 4 , however, are much less certain than for the CFCs or halons. The lowest value at Siple Dome, corresponding to the late 1940s or early 1950s, was 4±7 pmol mol -1 , but the South Pole data suggest a value of about 35 pmol mol -1 for 1950. In an early analysis of CCl 4 emissions, Galbally 41  Errors (shown as purple bars) correspond to a 68% con®dence envelope.
®rn; thus, we cannot make an unambiguous reconstruction of its concentration history. Although the derived atmospheric record of CH 3 CCl 3 is consistent at the two Antarctic sites, which suggests that this gas is stable in ®rn air, two data points in the lower part of the South Pole record do show measurable amounts of CH 3 CCl 3 . These, however, are most probably caused by contamination. The lowest value at each of the three sites, representing years from the 1890s to the 1940s, was ,2 pmol mol -1 (,2% of maximum atmospheric levels). Such concentrations are very near or below our limit of detection (3± 5 j; see Supplementary Information) and probably result from lowlevel contamination in the sampling apparatus. Note also that, although the turnaround in atmospheric CH 3 CCl 3 was recorded in the ®rn at Tunu (Fig. 3), this was not the case at South Pole (Fig. 6). This is because the South Pole was sampled only shortly after the turnaround. There is a time lag of about one year in transporting signals from the Northern Hemisphere to the Southern Hemisphere and wind pumping causes some mixing in the uppermost part of the ®rn, so the signal would be suppressed or absent in the South Pole ®rn in early 1995.
Because the inferred temporal change in CH 3 Cl is small relative to its variability in the ®rn air, a detailed history cannot be estimated reliably from these data (Fig. 6). Nevertheless, the difference of 5± 10% between bottom and mid-hole samples remains signi®cant and suggests that the atmospheric burden of CH 3 Cl has increased by about this amount during the past century. CH 3 Cl is the most abundant chlorine-containing gas in the atmosphere and is emitted from the ocean 42 , from biomass burning 43 , and by fungi 44 . Its current atmospheric budget remains uncertain. Real-time measurements of atmospheric CH 3 Cl extend back only to 1980 9 and show wide seasonal oscillations. The ®rn data represent a much longer time span, and are compatible with the real-time measurements.

Methyl bromide
Methyl bromide measurements from the South Pole and Siple Dome suggest a simple atmospheric history, with a rate of growth increasing from 0.01 pmol mol -1 yr -1 in the early 1900s to 0.05±0.06 (60.01) pmol mol -1 yr -1 (90% con®dence limits) during the 1970s and 1980s. These two sites differ in snow accumulation rate, ®rn depth and temperature (Table 1). Thus the similarity of the two CH 3 Br±CO 2 pro®les suggests that the record in the ®rn is atmospheric in origin. Taken at face value, such a record implies that the concentration of CH 3 Br was increasing slowly in the atmosphere through the ®rst half of the century, perhaps as a function of increased biomass burning or some other global change. The upturn in the mid-1960s coincides with the onset of CH 3 Br use as an agricultural fumigant, which suggests that the increase since the 1960s could indicate the expected reduction, about 15±25%, from the international ban on the use of this gas as a fumigant 45 . However, the response of CH 3 Br in the atmosphere to a ban on agricultural use is also dependent upon how other sources have changed since 1960. The budget of CH 3 Br for the modern atmosphere, calculated from what is known of sources and sinks, is grossly out of balance. Based upon current understanding 46,47 , there is a large source of this gas that has not been identi®ed. From the atmospheric concentrations implied by the deepest samples at the South Pole, it appears that this`missing' source was also present at the turn of the century. These results suggest that the unidenti®ed source is separate from modern fumigation practices, or that the overall sink, calculated for CH 3 Br from atmospheric reactions and loss to the oceans and soils, is currently overestimated.
The main problem in interpreting the CH 3 Br data comes from the Tunu site, where concentrations of CH 3 Br increased with depth (Fig. 4). The Tunu record does not appear to re¯ect past atmospheric changes, and it casts some doubt upon the utility of the Antarctic data. The simultaneous occurrence of CH 3 Br at 50 pmol mol -1 in the Northern Hemisphere and 5 pmol mol -1 in the Southern Hemisphere is highly improbable, as it would require an atmospheric lifetime of 0.1 year or less for CH 3 Br. All evidence today points to a lifetime of 0.5±1.2 years (ref. 47). Locally increased atmospheric concentrations are possible, but unlikely at these levels in remote areas. The most logical explanation is that CH 3 Br has been injected into the ®rn air at Tunu, perhaps by desorption or in situ production. Tests with our ®rn diffusion model indicate that one can account for the anomalous enrichments only by invoking net injection of CH 3 Br at or near the ®rn±ice transition (see Supplementary Information). We know of no physical, chemical or biological mechanism that could easily explain this injection. Simple adsorption followed by subsequent release during densi®cation would have injected CH 3 Br into the ®rn air at all three sites. Abiotic chemical transformation at such cold temperatures (-45 8C) cannot be ruled out, although attempts to identify possible source materials are speculative. Tunu has signi®cant marine in¯uence and is located in an area affected by arctic haze, which is known to contain signi®cant amounts of bromine 48 . There is no evidence to date suggesting that organisms grow at the low temperature of the ®rn, but it is possible that loosely bound material could be slowly released from enzymes or cell surfaces. Whatever the source of the anomalous CH 3 Br, we cannot prove it to be absent in Antarctica. However, if the process is driven at the ®rn±ice transition zone, as it most probably was at Tunu, then it could not have been signi®cant at the South Pole or at Siple Dome, which yield observed pro®les of concentration decreasing monotonically with depth to the bottom of the holes. Even a very low net production at the lock-in zone would have changed the shape of the pro®les noticeably, and also would have elevated the bottom concentrations substantially. Competing, unknown production and degradation processes throughout the ®rn might have altered the ®rn-air concentrations of CH 3 Br, but the probability is low that these processes would produce similar pro®les at the two physically dissimilar Antarctic sites.

Summary
Antarctic and Greenland ®rn can provide reliable archives of many atmospheric halocarbons dating back to the late 1800s, extending records of gases that have been measured in real time only during the past two decades. Robust records of CFCs, halons and SF 6 extracted from the measurements of ®rn air agree with real-time estimates in the later parts of the pro®les, corroborate emission models for some species and are useful in evaluating lifetimes and emission estimates for others. Although there are still important uncertainties, our data re®ne halocarbon histories and give experimental evidence about their pre-anthropogenic concentrations. It appears that for many species the cold, dry air in the ®rn preserves the gases. Except for CH 3 Cl and CH 3 Br, which have signi®cant natural sources, all of the halocarbons studied appear to have been derived entirely from emissions during the twentieth century. CH 3 Br and perhaps CCl 4 are, for unknown reasons, not always conserved in ®rn air, and other reactive halocarbons could conceivably behave similarly.

Methods
Two holes were drilled at each site. At the South Pole, both boreholes reached the ®rn±ice transition; at Tunu and Siple Dome the second holes ended in the diffusive zone. We only consider data from the ®rst hole at Siple Dome for our model because of inconsistencies in the sampling protocol. Low-pressure samples (,120 kPa) were collected into duplicate, 2.5-litre glass¯asks at all three sites. Higher-pressure samples (,380 kPa) were also collected into single or duplicate 2.4-litre stainless-steel¯asks at Tunu and Siple Dome. Details on drilling and¯ask ®lling can be found in Battle et al. 16 . Glass¯asks were ®lled at all three sites with a pump (MB-158, Metal Bellows, Senior Flexonics, Sharon, Massachusetts) attached to nylon or Dekabon tubing, and analysed by GC/ECD for 11 trace gases. CO 2 and d 15 N of N 2 also were measured in air from the glass¯asks. Surface air was excluded from the borehole by in¯ating a natural rubber bladder with stainless-steel end-caps. After the bladder was in¯ated, ,500±1,500 litres of ®rn air was extracted and pumped to waste, followed by¯ushing of the sample¯asks with 10 volumes of ®rn air. Stainless-steel¯asks ®lled with a pump (KNF Neuberger, Princeton, New Jersey) at Tunu and Siple Dome were analysed for over 20 halocarbons by GC/ECD and by gas chromatography with mass spectrometric detection (GC/MS). Measurements were calibrated with gravimetrically prepared standard gases, typically spanning a range of 20±150% of current atmospheric values. GC/MS calibration curves were linear in all cases and no zero-corrections were required; those for GC/ECD were predominantly linear. Zero air, a 20/80 mixture of puri®ed, synthetic O 2 and N 2 , was analysed to con®rm values obtained at the bottom of the pro®les and to establish detection limits for both instruments. Detection limits for the CFCs typically were within 1 or 2 pmol mol -1 of zero and, for SF 6 and the halons, within 0.1 and 0.2 pmol mol -1 of zero.