Accelerating growth of HFC-227ea (1,1,1,2,3,3,3-heptafluoropropane) in the atmosphere

We report the first measurements of 1,1,1,2,3,3,3-heptafluoropropane (HFC-227ea), a substitute for ozone depleting compounds, in air samples originating from remote regions of the atmosphere and present evidence for its accelerating growth. Observed mixing ratios ranged from below 0.01 ppt in deep firn air to 0.59 ppt in the current northern mid-latitudinal upper troposphere. Firn air samples collected in Greenland were used to reconstruct a history of atmospheric abundance. Year-on-year increases were deduced, with acceleration in the growth rate from 0.029 ppt per year in 2000 to 0.056 ppt per year in 2007. Upper tropospheric air samples provide evidence for a continuing growth until late 2009. Furthermore we calculated a stratospheric lifetime of 370 years from measurements of air samples collected on board high altitude aircraft and balloons. Emission estimates were determined from the reconstructed atmospheric trend and suggest that current "bottom-up" estimates of global emissions for 2005 are too high by a factor of three.

Published by Copernicus Publications on behalf of the European Geosciences Union.
Additional desirable properties of CFC replacements are relatively low atmospheric lifetimes and low Global Warming Potentials (GWPs). Here, HFC-227ea shows some disadvantages as both its estimated lifetime as well as its reported GWP are relatively high (34.2 years and 3220 on a 100-year time horizon, Forster et al., 2007) and comparable to those of CFC-11. Despite these perhaps not optimal properties HFC-227ea is being used, and emissions (EDGAR, 2009) reported. Missing so far has been the actual detection of HFC-227ea in the remote atmosphere, a determination of its rate of growth and "top-down" emission estimates.

Experimental methods
Air samples were collected from a range of platforms and at various locations: Please refer to Brenninkmeijer et al. (2007), Kaiser et al. (2006), Laube et al. (2008) as well as to http://neem.nbi. ku.dk/ for further details on sample collection.
All samples were analysed using gas chromatography with mass spectrometric detection (GC-MS). After drying using an on-line drying tube with Mg(ClO 4 ) 2 , condensable trace gases were pre-concentrated from about 300 ml of air at −78 • C in a 1/16 ′′ sample loop filled with an adsorbent (Hayesep D, 80/100 mesh) which was heated to 100 • C immediately after injection. Separation was carried out with an Agilent 6890 gas chromatograph using an Agilent GS-GasPro column (length 30 m, ID 0.32 mm) coupled to a high sensitivity tri-sector (EBE) mass spectrometer (Micromass/Waters AutoSpec). This instrument has a proven detection limit < 1 attomole and was operated in EI-SIR (Electron Impact-Selected Ion Recording) mode using a mass resolution of 1000. The GC column temperature was ramped from −10 • C to 200 • C at 10 • C per minute. CF 3 CHFCF 3 eluted after about 13.6 min and was measured by means of the fragment ions C 3 HF + 6 (m/z 151.00) and C 2 HF + 4 (m/z 101.00). No chromatographic interferences were found for these ions at the given retention time window. In addition, due to the relatively high mass resolution (as compared to common single quadrupole mass spectrometers) a possible interference from an unknown co-eluent is very unlikely.
In order to confirm the identity of the compound and to assign mixing ratios to the air samples we prepared static dilutions of CF 3 CHFCF 3 in Oxygen-free Nitrogen (OFN) obtained from BOC Gases, UK. For this purpose we constructed a dedicated drum dilution system based on an existing system described in Fraser et al. (1999). To evaluate this system CF 2 Cl 2 was added as an internal reference and diluted similarly. The resulting mixing ratios (which are dry air mole fractions) agreed with the internationally recognized calibration scale of NOAA (2001 scale) within 1.3%. Summing up all uncertainties that could affect the calibration values gives about 14%. However, taking into account the good agreement with the NOAA scale for CF 2 Cl 2 as well as the small variability between calibrations for HFC-227ea (< 4%) we estimate our scale uncertainty to be not larger than 5%. The air standard used to assign mixing ratios to the samples was found to contain 0.354 ppt HFC-227ea and the respective average 1 σ measurement precision was 1.4%. More details on configuration and evaluation of the calibration system can be found in the supplemental information.
To ascertain the potential presence of contaminants, one of the dilutions was measured while running the MS in scan mode across the range m/z 47 to 200. The chromatogram confirmed the purity of both compounds as no significant amounts of other halocarbons were observed. The obtained spectrum can be found in Table 1 and was compared to the one reported by Reizian-Fouley et al. (1997). All major peaks were present except one at m/z 117 which we did not observe. However, no ion is assigned to this peak in Reizian-Fouley et al. (1997) and there is no possible primary fragment of 1,1,1,2,3,3,3-heptafluoropropane with such a mass to charge ratio. Therefore we suggest that its occurrence is very likely to have been an instrumental artefact. The observed relative abundances are of limited comparability due to the different experimental setups. We generally observed much higher relative abundances for ions with lower mass to charge ratios which could be caused by different mass discrimination effects of the mass spectrometers or the use of a nonfluorinated internal reference compound (i.e. n-hexadecane) in our study.
The spectrum was background-corrected, but some minor residual peaks resulting from the internal mass axis calibration gas (i.e. nhexadecane) remain. The major peaks are the same as in the mass spectrum reported by Reizian-Fouley et al. (1997) except for an additional peak reported at m/z 117 which is a very untypical mass fragment for HFC-227ea and was not observed in this study. should be able to separate the compounds (boiling point difference of about 2 K between isomers). However, as an additional check we calculated the ratio of the two ions measured in SIR mode (m/z 151.00 and 101.00). It was found to be comparable for all samples and dilutions.

Firn air measurements and results
The Greenland firn air samples showed good measurement precisions for HFC-227ea and revealed that mixing ratios were monotonically decreasing with increasing depth (Fig. 1). As the age of the air also increases with depth this indicates a recent growth in the atmosphere. Moreover, the very low mixing ratios are observed around 74 m indicate a recent onset of emissions and thus an entirely anthropogenic origin of the compound. The relationship between past atmospheric time trends and concentration profiles in firn can be established using models of trace gas transport in firn (e.g. Schwander et al., 1993;Butler et al., 1999;Sturrock et al., 2002 and references therein). Please note, that all of the trends derived from measurements of air entrapped in firn are dependent on the assumption that the respective compound is chemically (no destruction) and physically (no adsorption or dis- solution) unaffected in the firn over the time periods represented by the measured trends. Here we use direct and inverse models that are developments of those described by Rommelaere et al. (1997), Fabre et al. (2000) and Martinerie et al. (2009). The following is a brief description whereas a fuller description will be given in a forthcoming publication from the NEEM project community. The firn diffusivity is first evaluated using three reference gases: CO 2 , CH 4 and CH 3 CCl 3 , for which measurements were made by NOAA-ESRL, CSIRO Marine and Atmospheric Research, IUP Heidelberg and UEA and atmospheric time trends estimates have been determined. For the latter we used publicly available atmospheric data from NOAA-ESRL (http: //www.esrl.noaa.gov/gmd), AGAGE (http://cdiac.ornl.gov/ ndps/alegage.html, Prinn et al., 2000), the GAW database (WMO Global Atmospheric Watch, World Data Centre for Greenhouse Gases, http://gaw.kishou.go.jp/wdcgg/) and firn and ice records from McFarling-Meure et al. (2006) for CO 2 and CH 4 as well as emission-based atmospheric modelling (Martinerie et al., 2009) for CH 3 CCl 3 . For HFC-227ea we further needed to know its diffusion coefficient in air. We used the Chen and Othmer (1962) calculation to derive an estimate for this, not the more commonly used method of Fuller et al. (1966), due to the reportedly greater robustness of the former (Massman, 1998). For HFC-227ea a diffusion coefficient relative to that of CO 2 of 0.465 was obtained. A very similar ratio of 0.471 was calculated using the Fuller et al. (1966) parameterisation.
Both a Monte-Carlo approach (Bräunlich et al., 2001) and a linear inverse approach (Rommelaere et al., 1997) were used to infer the atmospheric time trend for the HFC-227ea mixing ratios. Consistent results (within the respective model uncertainties) were obtained with the two methods. The time trend obtained with the inverse model (Rommelaere et al., 1997) is shown in Fig. 1 alongside with the actual measurements as a function of depth. It is most likely, that the emissions started in the early 1990s or late 1980s. Please note, that below 0.1 ppt the scenario essentially relies on the last two data points. The data gap in this region of the firn creates high uncertainties and does not allow further constraints on the emission start date or the trend prior to 1999. However, good agreement of the early trend (1993/1994) was observed with a northern hemispheric air sample (Colorado mountains, USA) remotely collected in 1994 (see Fig. 1) although this sample can not be directly compared with the firn air data (different locations). The more recent part of the scenario is better constrained and shows a sustained growth of HFC-227ea since 1999 until mid 2008. The average growth rate accelerated continuously from 0.029 ppt yr −1 in 2000 to 0.053 ppt yr −1 in 2005 and 0.056 ppt yr −1 in 2007. The firn air itself was collected in July 2008, therefore an averaged growth rate cannot be provided for 2008, but mixing ratios continued to increase. Please also note that these growth rates represent only best estimates and contain considerable uncertainties within the envelopes shown in Fig. 1. Additional data from samples collected on board aircraft flying in the Northern Hemispheric upper troposphere in late 2009 clearly indicate a further growth since during this period (see Fig. 1).

Emission estimates
We derived the emissions required to produce the temporal trend in the atmospheric concentrations inferred from the firn air data using a 2-D atmospheric chemistry-transport model. The model contains 24 equal area latitudinal bands and 12 vertical levels, each of 2 km. Thus the latitude band applicable to the NEEM firn site is the most northerly (66.4 • -90.0 • ) and since the altitude of the site is about 2.5 km, the concentrations for the second level (2-4 km) were used. The emissions were assigned predominantly to the northern midlatitudes with only 2% released in the Southern Hemisphere (distribution as in EDGAR, 2009). Using CFC-11, CFC-12 and CH 3 CCl 3 Reeves et al. (2005) showed that the model's  1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year

Annual emissions [Gg]
HFC-227ea modeled emissions EDGAR 2009 Fig. 2. Emission estimates as derived from trends of HFC-227ea in firn air in comparison with those reported by the EDGAR emission database (EDGAR, 2009). The primary sink for HFC-227ea is reaction with the OH radical in the troposphere. Therefore error bars were calculated as the root of mean of the squares of the summed uncertainty in the OH concentration and in the respective reaction rate coefficient as well as firn air related uncertainties (corresponding to the minimum and maximum scenarios in Fig. 1). transport scheme (Hough, 1989) typically reproduces the observed concentrations in the Arctic to within 10%, when constrained to observed southern hemispheric concentrations.
The major known atmospheric sink of HFC-227ea is the reaction with the OH radical (Zellner et al., 1994). The model's OH field was adjusted to give a partial lifetime for CH 3 CCl 3 , with respect to reaction with OH (τ OH ), in agreement with Clerbaux and Cunnold (2007) (6.1 years) when using a reaction rate coefficient of (1.2 × 10 −12 ) exp[−1440/T] cm 3 molecule −1 s −1 (Atkinson et al., 2008). The model then gives a value of 46.5 years for τ OH for HFC-227ea when using a reaction rate coefficient of (5.3 × 10 −13 ) exp[−1770/T] cm 3 molecule −1 s −1 (Atkinson et al., 2008). The diffusive transport out of the top of the model is then adjusted to account for stratospheric loss equivalent to a partial lifetime (τ strat ) of 450 years which is within the range calculated in Sect. 3.3. Overall the model gives a total lifetime of 42.1 years for HFC-227ea. This is higher than the 34.2 years reported in Clerbaux and Cunnold (2007) but the difference is mostly due to recent changes in the reaction rate of HFC-227ea with OH as reported in Atkinson et al. (2008).
Using the model the temporal trend in the global emissions was adjusted to match the trend of HFC-227ea from the firn model (Fig. 2). This has been done for the best, maximum and minimum mixing ratios derived from the firn air. The error bars shown in Fig. 2 are the root mean squares of the uncertainties from these minimum and maximum scenarios plus uncertainties in the OH loss of +59% and −40%. The latter were derived from an uncertainty of 14% in the OH concentrations (Prinn et al, 2001) and an uncertainty range of −37% to +57% in the reaction rate coefficient at 298 K (Atkinson et al., 2008). Given the relatively long lifetime Atmos. Chem. Phys., 10, 5903-5910, 2010 www.atmos-chem-phys.net/10/5903/2010/ of HFC-227ea compared to the period of rapidly increasing emissions, the uncertainty in this loss rate has a negligible impact on the derived emissions. Also shown in Fig. 2 are the emissions from "bottom-up" estimates based on industrial production and use (EDGAR, 2009). The annual estimates for the years 1999 to 2001 from EDGAR are around twice those derived from the firn data and the respective discrepancy increases with time from 2001 on reaching a factor of 3.0 in 2005. The firn and emission model uncertainties as well as the above mentioned uncertainties connected to OH cannot bridge this gap. We suspect that the discrepancy is likely to be caused by an overestimation of the bottom-up emissions. However, the currently available data are of insufficient temporal and spatial coverage to verify this suspicion.

Upper tropospheric data and stratospheric lifetime
Mixing ratios from upper tropospheric air samples collected in October 2009 (CARIBIC) are given in Fig. 3. The transect from 48 • N (Frankfurt) to 30 • S (Cape Town) was nearly true north-south (between 6 and 19 • E) and reveals a compact linear trend of HFC-227ea with latitude. This systematic behaviour with little scatter indicates that the observed air masses had not been in contact with sources recently and were representative for the upper troposphere at the given time and location. Backward trajectories (publicly available at http://www.knmi.nl/samenw/campaign support/ CARIBIC) indicate that, with the exception of the ITCZ region, air masses had been predominantly advected over the Atlantic Ocean from the west. The observed north-south gradient implies that the source regions are in the Northern Hemisphere. To maintain such a gradient also implies that HFC-227ea mixing ratios were still increasing in 2009. Assuming, that the observed mixing ratios are representative of the global upper troposphere, we calculate that the interhemispheric ratio (NH/SH) must be at least 1.1.
The stratospheric data are shown in Fig. 4 as a function of the concomitant CF 2 Cl 2 (CFC-12) mixing ratios. Long-lived organic compounds are known to form such compact correlations in this atmospheric region (Plumb and Ko, 1992). As an example the correlation of CFC-12 with CFCl 3 (CFC-11) is also displayed in Fig. 4. The slope of these correlations at the tropopause can be used to calculate a stratospheric lifetime (Volk et al., 1997). However, vertical stratospheric transport times are very slow (on the order of years). Thus, a correction must be made to account for the change in the entry mixing ratios over time. Here we use a method based on the mean age of air (i.e. the mean stratospheric transit time of a given air parcel) as derived from mixing ratios of SF 6 and its global tropospheric trends from NOAA-ESRL which were obtained from http://www.esrl.noaa.gov/gmd/ (updated from Geller et al., 1997). Please refer to Engel et al. (2002) for further details on the mean age calculation. As global tropospheric trends are only available until early 2009 yet, we complemented it via an extension to the values observed at the tropopause in late 2009. Subsequently the above determined tropospheric trend of HFC-227ea was shifted by 6 months to account for transport into the tropics and then propagated into the stratosphere assuming no chemical degradation en route. The difference between this and the actual measured mixing ratios in the stratosphere is the inorganic fraction released at a given altitude. This inorganic fraction relative to the amount that initially entered the stratosphere is then the Fractional Release Factor (FRF). Thus, FRFs represent the detrended relative inorganic fraction released from a given long-lived compound at a given location and time in the stratosphere. The exact method for their calculation is described in Laube et al. (2010). If such a set of FRFs for two compounds are multiplied by their globally representative mixing ratios it yields their detrended correlation for a given date. After this correction it is possible to calculate the stratospheric lifetime τ strat via the known lifetime of a reference compound τ ref using Eq. (1) (adapted from Volk et al., 1997).
The reference compounds used here were CFC-11 and CFC-12 and their global average atmospheric mixing ratios σ (which take into account the loss in the stratosphere) were inferred by adjusting the values derived by Volk et al. (1997) to the mixing ratios observed at the tropopause in late 2009 (using monthly averaged NOAA-ESRL global tropospheric trends from Montzka et al., 1999 for detrending, as updated and publicly available from http://www.esrl.noaa.gov/ gmd/).The slope of the detrended correlation already indicates that HFC-227ea is significantly longer-lived than CFC-12. Therefore we performed calculations with values ofσ between 95 and 100% of the mixing ratio observed at the tropopause (95% is the value for CFC-12). The final unknown quantity ( dχ dχ ref χ trop ref ) is the slope of the mixing ratio correlation at the tropopause. This was calculated from a least-square quadratic polynomial fit through the data.
Assuming a stratospheric lifetime of 45 years for CFC-11 (taken from Clerbaux and Cunnold, 2007) results in a stratospheric lifetime of 380 years for HFC-227ea forσ = 100% (negligible stratospheric loss) and 360 years forσ = 95% (same loss as CFC-12). As the exact value is likely to be between these values we infer 370 years as the best estimate. Calculations using the correlation with CFC-12 (100 year lifetime), different lifetimes of the CFCs (41 and 79 years, from Volk et al., 1997) as well as the above stated different values forσ give a range of 330 to 490 years. To assess the uncertainties of the correlation slope we performed similar calculations after subtraction and addition of the 1 σ measurement uncertainty from the ten data points next to the tropopause. This test revealed the limitations of this method for very long-lived compounds with correlation slopes close to zero. The corresponding lifetime range was 270 to 840 years. Due to the rather long tropospheric lifetime of HFC-227ea (37.8 years, Naik et al., 2000) uncertainties related to its tropospheric decomposition should have only a minor impact on this range. Thus, our experimentally derived stratospheric lifetime (best estimate of 370 years) agrees with the model-derived 633 years of Naik et al. (2000) within the respective uncertainties. However, it should be noted, that there are further uncertainties related to the approximated global tropospheric trend which can not be accounted for so far.

Conclusions
We report the first observations of HFC-227ea, a replacement for ozone depleting CFCs and halons, in remote regions of the atmosphere (i.e. in firn air and high-altitude samples). A dilution system has been built and evaluated for quantification of this and other compounds. A mass spectrum has been obtained and all high abundant ions of a spectrum reported in the literature were found to be present except one. A reconstruction of the atmospheric history of HFC-227ea from firn air reveals that emissions were very low in the early 1990s. In combination with further measurements in the upper troposphere we find strong evidence for its growth until the end of our available observations in 2009. Emission estimates inferred from this trend do not agree with estimates from the EDGAR database (EDGAR, 2009) and even show a discrepancy that is increasing with time. We can not resolve these disagreements with the observations and the related uncertainties presented here. Further studies, which should include observations from global ground-based networks, are needed to improve the understanding of the emission processes and patterns and the global distribution of HFC-227ea. Finally, compact correlations with other long-lived gases in the stratosphere have been used to derive a stratospheric lifetime of 370 years (range: 270 to 840 years) which agrees with the 633 years reported by Naik et al. (2000) within the uncertainties. It is the first time that this measurement-based approach has been used to evaluate a modelled stratospheric HFC lifetime. In addition we are grateful for contributions from Thijs F. Duindam, all CARIBIC partners, the Geophysica team and the CNES balloon team (sample collection and campaign organisation) as well as from the NOAA-ESRL GMD, AGAGE and WMO GAW teams and from CSIRO Marine and Atmospheric Research, Australia, the Institute of Arctic and Alpine Research (INSTAAR) at the University of Colorado at Boulder, USA and the IUP Heidelberg, Germany (provision of publicly available and also unpublished data).