Spectrometric monitoring of atmospheric carbon tetrafluoride (CF 4 ) above the Jungfraujoch station since 1989: evidence of continued increase but at a slowing rate

Abstract. The long-term evolution of the vertical column abundance of carbon tetrafluoride (CF4) above the high-altitude Jungfraujoch station (Swiss Alps, 46.5° N, 8.0° E, 3580 m a.s.l.) has been derived from the spectrometric analysis of Fourier transform infrared solar spectra recorded at that site between 1989 and 2012. The investigation is based on a multi-microwindow approach, two encompassing pairs of absorption lines belonging to the R-branch of the strong ν3 band of CF4 centered at 1283 cm−1, and two additional ones to optimally account for weak but overlapping HNO3 interferences. The analysis reveals a steady accumulation of the very long-lived CF4 above the Jungfraujoch at mean rates of (1.38 ± 0.11) × 1013 molec cm−2 yr−1 from 1989 to 1997, and (0.98 ± 0.02) × 1013 molec cm−2 yr−1 from 1998 to 2012, which correspond to linear growth rates of 1.71 ± 0.14 and 1.04 ± 0.02% yr−1 respectively referenced to 1989 and 1998. Related global CF4 anthropogenic emissions required to sustain these mean increases correspond to 15.8 ± 1.3 and 11.1 ± 0.2 Gg yr−1 over the above specified time intervals. Findings reported here are compared and discussed with respect to relevant northern mid-latitude results obtained remotely from space and balloons as well as in situ at the ground, including new gas chromatography mass spectrometry measurements performed at the Jungfraujoch since 2010.


Introduction
Carbon tetrafluoride (CF 4 ) or tetrafluoromethane is a perfluorocarbon (PFC-14) whose unambiguous presence in the earth's atmosphere, from the boundary layer (Rasmussen et al., 1979) to the stratosphere (Goldman et al., 1979), as well as its "near inertness" in the atmosphere (Cicerone, 1979) have raised increasing attention and concern among the scientific community since the 1980s.
Its main anthropogenic source is primary aluminum production, during which CF 4 is released through "anode effect" episodes (e.g., Penkett et al., 1981;Khalil et al., 2003).Since the 1980s, non-negligible CF 4 emissions have also been released increasingly by manufacturing of semiconductors and other electronic devices (e.g., "plasma etching"; Tsai  et al., 2002).The only known but poorly quantified natural sources of CF 4 are of lithospheric origin (e.g., Gassmann, 1974;Cicerone, 1979;Harnisch et al., 1996a;Harnisch and Eisenhauer, 1998;Harnisch, 2000).Recently, Deeds et al. (2008) presented the first in situ evidence for this lithospheric flux, which leads to a background contribution to atmospheric carbon tetrafluoride that lies between 35 and 45 ppt (parts per trillion dry air mole fraction) with the lower value favored on the basis of reported accuracies and precisions (for an overview, see Table 2 of Mühle et al., 2010).
PFCs are long-lived species with lifetimes of many thousand years (e.g., Ravishankara et al., 1993).With an atmospheric lifetime estimated to exceed 50 000 yr, CF 4 is by far the longest lived PFC (WMO-2010(WMO- , 2011)).Combined with a high global warming potential of at least 7390 on a 100 yr time horizon (WMO-2010(WMO- , 2011)), this compound is a strong greenhouse gas whose anthropogenic emissions are deservedly targeted for regulation under the Kyoto Protocol (IPCC, 2001).Because of the absence of atmospheric sinks, CF 4 shows a nearly constant mixing ratio profile throughout the atmosphere (e.g., Zander et al., 1992Zander et al., , 1996;;Nassar et al., 2006) and its vertical gradient -as quantified, for example, by Fabian et al. (1996), and Harnisch et al. (1996b) using stratospheric balloon-borne cryogenic air sampling between 1987 and 1995 -is only caused by a delayed propagation of the ground-based emissions to higher altitudes.The presence of CF 4 in the stratosphere was first reported by Goldman et al. (1979) who identified the strong ν 3 band of CF 4 at 1283 cm −1 in a solar limb spectrum recorded in 1978 at 25 km altitude, from aboard a balloon platform.Its vertical profile between 15 and 50 km was derived by Zander et al. (1987) from ATMOS (Atmospheric Trace MOlecule Spectroscopy; Farmer, 1987) solar limb observations during the Spacelab 3 shuttle mission in 1985, from subsequent MkIV FTIR balloon flights (e.g., Toon, 1991;Sen et al., 1996), and from the satellite ACE-FTS instrument (Atmospheric Chemistry Experiment Fourier Transform Spectrometer, e.g., Bernath et al., 2005;Brown et al., 2011).
Recent ground-level air sampling and in situ measurements of CF 4 in both hemispheres (e.g., Khalil et al., 2003;Mühle et al., 2010) or remotely from space (e.g., Rinsland et al., 2006;Brown et al., 2011) have indicated a significant slowdown in the rate of increase of atmospheric CF 4 , attributed to efforts undertaken by the aluminum industry to limit its emissions during "anode effect" episodes (International (Primary) Aluminium Institute, 1996Institute, , 2009)).However, significant uncertainties remain, amongst others due to the increase of Chinese aluminum production and insufficiently defined emission factors for Chinese smelters (International Aluminium Institute, 2013).Additionally, the magnitude and temporal evolution of CF 4 emissions from the semi-conductor industry remain very unclear, despite efforts by the World Semiconductor Council to reduce their emissions (WSC, 2013).Based on the inversion with a 2-D box model of a selected subset of AGAGE (Advanced Global Atmospheric Gases Experiment) ground-level measurements in both hemispheres from the early 1970s to 2008, Mühle et al. (2010, Fig. 4) derived global CF 4 emissions which increased during the 1970s to reach their maximum during the early 1980s (17.5 ± 1 Gg yr −1 ) and subsequently declined progressively to stabilize at about 11 Gg yr −1 by 2000 until 2008.We refer the reader to Mühle et al. (2010) and references therein, for a detailed and exhaustive discussion regarding the evolution of CF 4 in the global troposphere from 1973 to 2008.
This paper reports on the mean evolution of the vertical carbon tetrafluoride loading integrated over the free troposphere and stratosphere above the high-altitude Jungfraujoch station, derived from the spectrometric analysis of Fourier transform infrared (FTIR) solar observations made at that site between 1989 and 2012.Related findings are compared with relevant ones also obtained remotely from space-and balloon-borne solar observations, with new in situ gas chromatography mass spectrometry (GCMS) measurements performed by Empa (Laboratory for Air Pollution/Environmental Technology) at the Jungfraujoch since 2010, as well as with recently reconstructed in situ ground level baseline growth rates of CF 4 in both hemispheres, reported by Mühle et al. (2010).Our concluding remarks include recommendations for improving the relative accuracies of spectroscopic CF 4 line parameters which currently remain estimated at ± 6 %, as compared to the achieved 1-2 % for the in situ data.
Since 1990, the University of Liège research activities are performed within the frame of the Network for the Detection of Atmospheric Composition Change (NDACC; see http://www.ndacc.org).

FTIR remote-sensing measurements and retrieval strategy
The long-term CF 4 time series presented and analyzed in this study has been derived from the analysis of solar spectra recorded between January 1989 and December 2012 under clear-sky conditions at the high-altitude International Scientific Station of the Jungfraujoch (hereafter ISSJ; Swiss Alps, 46.5 • N, 8.0 • E; 3580 m a.s.l.).The recordings were made with two very high spectral resolution FTIR spectrometers, a "home-made" instrument primarily used until 1995, and progressively replaced by a faster, more sensitive commercial Bruker-120 HR instrument (Zander et al., 2008).The initial database investigated here consists of over 5500 spectra recorded with an optical filter covering the 750 to 1400 cm −1 spectral region, thus encompassing the strongest infrared band of CF 4 -the ν 3 centered at 1283 cm −1 .Spectral resolutions (defined as the reciprocal of twice the maximum optical path difference) alternate between 0.004 and 0.006 cm −1 , depending on the rate of solar zenith angle variation during the day, and scanning time of successive recordings.Signal-to-noise (S/N) ratios vary between 150 to more than 2500 (average spectra resulting from several successive individual Bruker scans, predominantly around midday, when solar zenith angles vary slowly).
The spectral analyses were performed with the SFIT-2 v3.91 fitting algorithm, a code based on the optimal estimation formalism of Rodgers (1976) and specifically developed to retrieve vertical column abundances and mixing ratio profiles of atmospheric gases from FTIR observations (Connor et al., 1995;Rinsland et al., 1998).This code has been successfully intercompared with the PROFFIT retrieval algorithm (e.g., Hase et al., 2004;Duchatelet et al., 2010), the other tool in use by the NDACC FTIR community for monitoring numerous tropospheric and stratospheric target gases, worldwide.
Line parameters adopted in the spectral fitting process were taken from the HITRAN 2004 spectroscopic compilation (Rothman et al., 2005), including the August 2006 updates (e.g., Esposito et al., 2007).For CF 4 , we selected a set of pseudo-lines whose intensities and temperature-dependent parameters were derived by one of us (G.C.T.) from a series of high-resolution and high-S/N laboratory spectra recorded by Nemtchinov and Varanasi (2003) under pressure and temperature conditions typical of those encountered in the atmosphere.These pseudo-lines were also used for all CF 4 remote sensing measurements intercompared in Sect. 3 with our ISSJ findings.
The model atmosphere adopted above the 3.58 km Jungfraujoch altitude consists of a 39 layer scheme with progressively increasing thicknesses to reach 100 km altitude.The pressure-temperature profiles are those specifically computed for the ISSJ location on a daily noontime basis by the National Centers for Environmental Prediction (NCEP, Washington, DC; see http://www.ncep.noaa.gov).
While most of the Q-and R-branch features of the ν 3 band of CF 4 can be used for remote sensing retrievals from space between about 15 and 50 km altitude (see Fig. 18 in Zander et al., 1987), strong absorptions by H 2 O, HDO, N 2 O, CO 2 , CH 4 , and weaker ones (i.e., by HNO 3 , ClONO 2 and further H 2 O isotopologues) combine to heavily interfere with CF 4 features in low altitude spectra.Consideration of the relative importance of these interferences in typical solar recordings at the high-altitude, rather dry ISSJ site led us to adopt as "RUN 1" the spectral interval from 1284.73 to 1285.15 cm −1 displayed in Fig. 1, which encompasses 6 of the strongest R-branch features of the CF 4 ν 3 band.Because of the poor fitting over the two middle CF 4 features, severely affected by H 2 O and HDO interferences, we excluded from our initial database all spectra whose water vapor content exceeded 8×10 21 molec cm −2 , as well as those observed at solar zenith angles larger than 75 degrees to also minimize line-wing depression by the strong N 2 O line centered at 1284.7 cm −1 .In addition and as illustrated in Fig. 1, we restricted the subsequent "RUN2" to a composite window in which only the CF 4 profile was further adjusted.Frame B shows that the RMS (root mean square) residuals reduced by about a factor 2 in RUN 2 as compared to RUN 1.This RMS decrease was confirmed over the entire database, while the related CF 4 columns reduced by less than 0.7 % on average.
The a priori vertical concentration profile for the CF 4 target gas was set constant at 72 ppt throughout the atmosphere above ISSJ, consistent with values and associated uncertainties derived for the stratosphere (in the 15 to 45 km altitude range) from nearly 1400 solar occultations performed between February 2004 and mid-2012 in the 36.5 to 56.5 • N latitude zone by the ACE-FTS instrument.As there is poor vertical information content in the individual spectra, the constant a priori profile was simply scaled during our fitting procedure.
For all interfering molecules, averaged mixing ratio profiles based on WACCM (the Whole Atmosphere Community Climate Model; e.g., Chang et al., 2008) model predictions for the 1980-2020 period and the ISSJ station were used as a priori.To reliably account for the weak interfering absorption features by HNO 3 (see Frame A of Fig. 1), its concentration profile was pre-retrieved for each spectrum from a simultaneous fit to two NDACC-dedicated spectral intervals (i.e., 867.05-870.00 and 872.25-874.00cm −1 ; see Wolff et al., 2008).It was then assumed in the consecutive fitting steps RUN 1 and RUN 2. During RUN 1, the vertical dis- and  H 2 18 O were independently scaled while in RUN 2, only the CF 4 profile was further adjusted.A S/N ratio of 500, commensurate with the noise level of most retained spectra, was adopted throughout the spectral fitting calculations.Inspection of an exo-atmospheric solar atlas by Farmer and Norton (1989) revealed only a very weak solar absorption feature at 1284.7666 cm −1 .It was neglected in the SFIT-2 retrievals, after verification that it had negligible influence on the CF 4 results.
After further exclusion of observations with S/N ratios lower than 300 and of high and low individual-carbon tetrafluoride-column outliers falling out of the ± 2.5σ confidence interval, the retained database upon which the results and discussion in Sect.2.2 are based includes 3034 individual CF 4 column measurements above ISSJ, encompassing 1272 observational days.
Table 1 provides an error budget resulting from major instrumental and analytical uncertainties that may affect typical individual CF 4 column amounts above the site.They were evaluated according to referenced comments or perturbations to a representative subset of spectra.As is often the case when dealing with relatively "new anthropogenic" molecules (for which detailed fundamental spectroscopic laboratory studies are in progress and await validation), the largest systematic error in this work results from the CF 4 pseudo-line parameters (intensities and temperature dependences), which we set at ± 6 % on the basis of evaluations discussed by Frame A displays the characteristic absorptions of CF 4 and the five most significant interfering gases in the "Run 1" micro-window selected for our CF 4 retrievals, computed for a typical ISSJ observation on 1 April 2000 (at 63.46 • apparent solar zenith angle and 0.006 cm −1 resolution).Their combination results in the black simulation trace (labeled "Sim.")which, compared to the actual observation (green curve, labeled "Obs.")leads to the red residuals (observed minus calculated signals) in Frame B. Because of strong perturbations by the H 2 O and HDO interferences, this initial "Run 1" was followed by a second composite "Run 2", with corresponding residuals displayed by the thick green trace in Frame B. Irion et al. (2002).A "quality-test" performed with a synthetic line-list covering the R-branch of the CF 4 ν 3 band, recently released by Boudon et al. (2011), will be evoked in Sect.3.

GCMS in situ measurements
The second original data set reported for the first time in this paper results from ground-based gas chromatography -mass spectrometry measurements conducted at ISSJ by Empa.The analyses are performed with a Medusa instrument described in detail by Miller et al. (2008), using 2 L of cryogenically trapped air alternated with measurements of a standard to calibrate the instrument.This leads to one air measurement every two hours.The CF 4 measurements at ISSJ began in 2008 but due to some instrumental difficulties, data are only reported here from 2010 onwards.Their precision is ∼ 0.4 % while the accuracy is estimated at 1-2 %.The measurements are based on the Scripps Institution of Oceanography (SIO) SIO-2005 calibration scale and are tightly linked into the AGAGE network.

Results and discussion
Figure 2 reproduces the daily mean vertical column abundances (expressed in numbers of CF 4 molecules per square cm; left vertical scale) derived above ISSJ between 1989 and 2012.They have been normalized to the mean local pressure monitored at the site during the past decades (i.e., 654 hPa) versus the daily surface pressure measured at noontime.The database reveals the relative sparseness and dispersion of the daily mean columns prior to about 1995, resulting from less frequent observations with the home-made instrument and their lower S/N ratios.However, as no statistically significant difference was observed between day-coincident CF 4 columns by both instruments, they have been merged in the post-1995 daily mean averages.
The right side scale of Fig. 2 reproduces the mean constant mixing ratio above ISSJ as returned by the SFIT-2 code on the basis of the physical P-T model atmosphere adopted for each day.The uncertainty on the conversion from leftto-right scale is less than ± 3 %.These mixing ratios correspond to moist air values.However, correction factors to and (0.98 ± 0.02) × 10 13 molec cm −2 , respectively, for the periods 1989-1997 and 1998-2012.Extrapolation of these increases down to sea level, as done in the previous paragraph, translates into yearly total column changes above sea level equal to (2.14 ± 0.17) × 10 13 and (1.50 ± 0.03) × 10 13 molec cm −2 .Globally, these changes require CF 4 emission rates equal to (15.8 ± 1.25) Gg yr −1 and (11.1 ± 0.2) Gg yr −1 for the above mentioned periods.
When taking into account a reasonable lag time of three years for ground-level emissions to uniformly mix in the free troposphere and in the stratosphere (e.g., Fabian et al., 1996;Waugh and Hall, 2002;Anderson et al., 2000;Stiller et al., 2008;Diallo et al., 2012), these derived CF 4 emission rates are commensurate with the 1986 to 2009 time averaged global emission estimates reported by Mühle et al. (2010, Table 6, i.e., 14.6 and 10.8 Gg yr −1 , for the 1986-1994 and 1995-2009 periods, respectively).A noticeable CF 4 seasonal cycle (close to 2 %, peak-to-peak amplitude) also deduced with the bootstrap re-sampling tool (Gardiner et al., 2008) is essentially ascribable to the seasonal variation of temperature versus pressure ratios in our adopted layered model atmosphere and to resulting impacts upon temperature-and pressure-dependent line intensities and half widths.
In a second evaluation approach, the CF 4 database was "modeled" with both a second order function and a nonparametric least-squares fit, respectively displayed in Fig. 2 by the black-and red curves.For the red curve, a local smoothing technique is applied, assuming a Gaussian weighting function and sampling 20 % of the data points at once.The actual bandwidth progressively considers all the measurements along the x axis (see Rinsland et al., 2003;and reference therein).Discrete six-year time averaged trends determined from tangential derivatives to the black line at 1992.0,Fig. 2. FTIR time series of CF 4 daily mean vertical column abundances above the Jungfraujoch (expressed in numbers of CF 4 molecules per square cm), normalized to a mean local pressure of 654 hPa.The solid black and red curves correspond to a second order fit and a 20 % smoothing function to the data points, respectively.The coefficients of the second order black curve are (from the highest order to the independent term): −0.0194, 78.72 and −79835, in units of 10 13 molec cm −2 .In the same units and rank, the coefficients of the linear regression are 1.385, −2675 and 0.979, −1862, for the 1989-1997 (blue) and 1998-2012 (green) time periods, respectively.The right side scale corresponds to the conversion of our measured columns into mean constant mixing ratios above the site, expressed in ppt (parts per trillion).
1998.0, 2004.0 and 2010.0, and related global emissions are provided in Table 2.The latter are in line with the significant slowing in the CF 4 emission rates at the ground that began during the late 1980s, amplified during the early 1990s and leveled off subsequently, as synthesized by Mühle et al. (2010, Fig. 4).The mean global emission of 17.6 Gg yr −1 derived here for the 1989 to 1994 time interval is in good agreement with the mean increase rate of 18 Gg yr −1 derived by Zander et al. (1996) from upper stratospheric CF 4 measurements made in the northern hemisphere by the ATMOS FTIR instrument during the shuttle missions that occurred between 1985 and 1994 (Gunson et al., 1996).
The small but noticeable differences between the red-and black curves in Fig. 2 are indicative of slow temporal growth regime changes by up to +/−1.5 Gg yr −1 around the second order fit.
A "quality-test" covering our entire database, performed by replacing the adopted CF 4 pseudo-lines by the synthetic line parameters of Boudon et al. (2011), as included in the "supplemental" section of the Hitran 2008 compilation (Rothman et al., 2009), led to the following appreciations: (i) the line positions are good; (ii) on average, the fitting residuals are ∼ 30 % higher; (iii) the retrieved CF 4 columns are consistently larger by ∼ 25 %.We believe that this last observed difference may result from the fact that the Boudon line list, currently restricted to the ν 3 R-branch, is only based on one high-resolution FTIR laboratory spectrum at 296 K and a low pressure of 0.17 torr, while the pseudo-lines were derived from a large set of spectra recorded at numerous typical pressure and temperature sets encountered throughout the atmosphere (Nemtchinov and Varanasi, 2003).However, another possible explanation is that this significant bias might also result from contributions, for example, by underlying hot bands of CF 4 not present in the Boudon line list (Rothman et al., 2009), while they are captured by crosssection measurements.
The conversion of our measured CF 4 column abundances into mean constant mixing ratios above ISSJ (see vertical scales in Fig. 2), assuming that this molecule is uniformly mixed throughout the entire atmosphere, allows some comparison with ground-level in situ-and FTIR solar occultation measurements from balloon-and space-based platforms.A few relevant examples, referring to data obtained at northern mid-latitudes, are displayed in Fig. 3, where the thick black line corresponds to the mixing ratios associated to the second order black curve fitted in Fig. 2 to our daily mean CF 4 columns above the site.The green curve reproduces an excerpt of the 1973 to 2010 assimilated monthly CF 4 mixing ratios for the northern extra-tropics (i.e., 30-90 • N), based on AGAGE archived air samples and in situ measurements  at 1992at .0, 1998at .0, 2004at .0, and 2010at .0. Data source 1989at -1994at 1995at -2000at 2001at -2006at 2007at -2012 Columns above ISSJ in 10 13 molec cm −2 yr −1 1.53 ± 0.41 1.29 ± 0.11 1.05 ± 0.07 0.82 ± 0.08 Corresponding global emissions in Gg yr −1 17.6 ± 4.7 14.8 ± 1.25 12.0 ± 0.8 9.4 ± 0.9 Emissions from in situ measurements in Gg yr   (Fabian et al., 1996).The pink star corresponds to a CF 4 mixing ratio of 77.8 ± 0.6 ppt derived from one air sample collected in Tokyo (35.6 • N) in August 2003, using a specific calibration approach based on the atmospheric 80 Kr abundance as reference; no calibration accuracy is reported (Aoki and Makide, 2005).The second order curve fitted to the pink triangles and the star is a typical representation of numerous ground-based monitoring efforts conducted in situ at northern mid-latitudes during the late 1970s onwards, as illustrated in Fig. 1 of Mühle et al. (2010).The latter showed that the MPAE 86 and UEA (University of East Anglia) calibration scales adopted during these earlier activities have reported uncertainties ranging by up to ± 15 % and that related measurements can thus be reconciled with the recent, much more accurate ones based on the SIO-2005 calibration scale (1-2 %).
The filled orange circles represent stratospheric monthly mean CF 4 mixing ratios between 20 and 45 km altitude, derived from over 1400 solar occultation measurements (Version 3 products; Boone et al., 2013) in the 36.5 to 56.5 northern latitude zone between 2004 and 2012 with the satelliteembarked ACE-FTS instrument (Bernath et al., 2005).The four orange square symbols reproduce updated mean stratospheric CF 4 mixing ratios between ∼ 20 to 40 km altitude, derived from the 1985, 1992, 1993and 1994 ATMOS missions (Version 3;Irion et al., 2002) as reported by Rinsland et al. (2006).Finally, the filled green triangles correspond to the average mixing ratios between 10 and 35 km altitude derived from individual MkIV balloon flights performed over 1990 to 2007, between 33 and 68 • N latitude (e.g., Sen et al., 1996).The error bars associated to the ATMOS, MkIV and ACE data points represent the standard deviations of the means.As all the space-and balloon-borne CF 4 retrievals were performed using the same cross-section parameters, not only for R-branch features but also for the strong Q-branch of the CF 4 ν 3 band, we assumed that they could reasonably be interlinked with a 2nd order fitting, represented in Fig. 3 by the blue curve.It clearly shows the significant but slowing increase of CF 4 throughout the stratosphere over the past decades.The limited number of spectra recorded by ATMOS during the short pioneering US shuttle flights, as compared to the ongoing ACE-FTS mission, clearly shows the advantage of regular, long-term monitoring approaches, which is also true for ground-based investigations.Table 3 provides mean annual mixing ratio increases determined by taking derivatives at 1990.0, 2000.0, and 2010.0 to the continuous curves displayed in Fig. 3.The dashed horizontal line in Fig. 3 corresponds to the natural background level of ∼ 35 ppt recently reported by Worton et al. (2007) and Mühle et al. (2010); it has been drawn here to better illustrate the relative anthropogenic contribution to the total CF 4 atmospheric burden which, since the turn into the 21st century, has overtaken the natural loading.
The differences between the various data sets displayed in Fig. 3 are obviously linked, at least partially, to atmospheric transport and can be ascribed to the fact that the instruments involved sound different layers of the atmosphere, namely the boundary layer for AGAGE and Empa as well as Cape Meares and Tokyo, the free troposphere and stratosphere for our FTIR observations, and the stratosphere only for the satellite and balloon data.Consequently, changes in the emissions of a long-lived gas at the ground will mix in the global troposphere within 1 to 2 yr, and propagate in the stratosphere by upwelling via the tropical pipe, reaching 20 km mid-latitudes within 3 to 5 yr (e.g., Elkins et al., 1996;Stiller et al., 2008).Examples of such time delays have also been reported by Anderson et al. (2000) who found that the inorganic chlorine and fluorine loadings in the upper stratosphere lag the related organic loadings at the ground by 4.5 to 5.5 yr.Inputs from the two-dimensional 12-box model of Mühle et al. (2010) indicate for 30-90 • N a mean difference between the tropospheric and stratospheric mixing ratios of 2 ppt over the last decade and of more than 3 ppt in the early 1980s, when the CF 4 emissions were stronger.In order to assess the adequacy (and possible bias) of having adopted a constant distribution in our column retrieval calculations and mixing ratio conversion, we built an a priori profile accounting for the time needed for CF 4 emitted at the ground to propagate and mix in the stratosphere.Since the 12-box model only included a single stratospheric level, we based the construction of our a priori CF 4 distribution on a profile of mean age of stratospheric air parcels produced by Diallo et al. (2012), using ERA-Interim reanalyses and a Lagrangian model.Their findings have been confronted to satellite, aircraft and balloon observations, showing good agreement, including at northern mid-latitudes.For this region of the atmosphere of interest here, Figs. 2 and 6 of Diallo et al. (2012) provide profiles of mean age of air, with values close to 2 yr at 16 km, 3 yr at 18 km, 4 yr at 20 km and more than 6 yr above 25 km (i.e., showing a gradient of approximately 0.5 yr per km in the lower mid-latitude stratosphere).Assuming the CF 4 annual rate of change which prevailed over the last decade as derived by AGAGE (i.e., 0.7 ppt per year), our a priori profile was built such as to present a mixing ratio difference with the surface concentration of −1.4 ppt at 16 km, −2.1 ppt at 18 km, −2.8 ppt at 20 km, and so on, up to a constant difference of −4.5 ppt for altitudes above 25 km.All the observations of 2009 were refitted with this non-constant vertical distribution, returning total columns on average 2.5 % higher.Conversion of these columns into corresponding surface mixing ratios resulted in values systematically higher by 3.7 %, or +2.9 ppt with respect to our standard run assuming a constant vertical distribution.Applying a vertical shift of +2.9 ppt to the FTIR function for 2009 brings it in very good agreement with the AGAGE and Empa time series, as indicated by the dashed red segment matching the green curve and open circles in Fig. 3.This result gives good credibility on the absolute scale (1-2 %) provided by the SIO-2005 calibration standard as well as by the CF 4 spectroscopic parameters used in our analyses.
Nonetheless, it remains that the uncertainties associated with the various CF 4 remote data sets presented in Fig. 3, namely 7 % for ISSJ (see Table 1), 11 % for ATMOS (Rinsland et al., 2006), 10 % for MkIV (Sen et al., 1996) and 7 % for the recent ACE-FTS data (Brown et al., 2011;Rinsland et al., 2006) have to be maintained as such, until further spectroscopic laboratory and related theoretical investigations narrow these uncertainties.

Summary and conclusions
Since the 1980s, the presence of carbon tetrafluoride (CF 4 ) in the earth's atmosphere has attracted increasing attention for three reasons, namely (i) its continued accumulation in our atmosphere, (ii) its extremely long lifetime, and (iii) its high global warming potential, 7390 times larger than that of CO 2 , justifying priority recommendations by the Kyoto Protocol for CF 4 monitoring and regulation.In response to these recommendations, strongly endorsed by the Network for the Detection of Atmospheric Composition Change, we have reported the first spectrometric measurement from the ground of the atmospheric CF 4 , and their comparison with recently reported ground-level in situ mixing ratios which have a quoted accuracy of 1 to 2 % (Mühle et al., 2010).
Owing to the location of the Jungfraujoch (3.58 km a.s.l.) above the polluted and wet boundary layer of the low troposphere, we have established a special retrieval procedure that minimizes the perturbations by numerous interfering atmospheric gases, in particular the residual H 2 O and HDO above the site (see Fig. 1), and successfully determined with an accuracy of ± 7 % the loading of CF 4 throughout the free troposphere and the stratosphere (i.e., over two thirds of the total mass of the atmosphere) between 1989 and 2012.The spectrometric analysis of a subset of solar spectra recorded at ISSJ during this period (i.e., over 3000 spectra, encompassing 1272 days) has allowed, for the first time, the measurement of the long-term evolution of the CF 4 column abundance from ground-based remote FTIR observations.
As illustrated in Fig. 2, the yearly mean CF 4 column increase above ISSJ was found equal to (1.38 ± 0.11) × 10 13 molec cm −2 between 1989 and 1997, and (0.98 ± 0.02) × 10 13 molec cm −2 from 1998 to 2012.Globally, these increases require 15.8 ± 1.3 and 11.1 ± 0.2 Gg yr −1 , respectively, resulting from anthropogenic CF 4 emissions at the ground in the earlier part of the record, primarily from the aluminum industry, and in the latter part also from the manufacturing of electronic devices.The significant slowing in the rate of increase is probably the result of efforts undertaken by the aluminum industry to comply with recommendations from the Kyoto Protocol.Considering our uncertainty, which is almost entirely due to the quality of the CF 4 spectroscopic parameters adopted here (± 6 %), our findings are in good agreement with results derived by Mühle et al. (2010) based on selected ground level in situ measurements in the Northern Hemisphere from 1973 to 2008, and by new in situ GCMS measurements performed since 2010 by Empa at the Jungfraujoch.Moreover, we showed that the adoption of a vertical distribution for CF 4 , accounting for the time needed for this very long-lived species to propagate and mix in the stratosphere, provided FTIR converted mixing ratios in very good agreement with the AGAGE data sets, giving good confidence in the absolute mixing ratios derived from the analyses involved here for both techniques.
We also note that the CF 4 FTIR time series is in excellent agreement (within 2 %) with solar occultation measurements made from satellites (ATMOS, ACE) and balloons (MkIV).This demonstrates the reliability of the spectral fitting procedures for retrieving CF 4 from the ground, in particular, the handling of the interfering H 2 O lines (which are negligible in solar occultation retrievals above 10 km altitude).
During this study, we noticed that a synthetic CF 4 linelist produced by Boudon et al. (2011) was available in the formal HITRAN 2008 compilation (Rothman et al., 2009).Running our entire data set with these line-by-line parameters showed a reasonably good fit of the CF 4 line positions and contours.However, the retrieved column abundances were consistently larger by ∼ 25 % than those derived with the pseudo-lines adopted here.This corroborates a conclusion by Boudon et al. (2011), stating that "the new linelist is still approximate concerning line intensities".We strongly encourage the continuation of such fundamental efforts aimed at improving the accuracy of line parameters for important atmospheric species.
Finally, we expect that the new in situ local measurements of CF 4 performed by Empa at the Jungfraujoch within AGAGE open interesting possibilities for in-depth statistical intercomparison with our FTIR time series.This sideby-side, high-mountain operation is currently unique, worldwide, and is complementary in terms of techniques involved, vertical atmospheric coverage and, hopefully, long-term regular operation.

334E.
Mahieu et al.: Spectrometric monitoring of atmospheric CF 4 above the Jungfraujoch station since 1989

Fig
Fig. 1.Frame A displays the characteristic absorptions of CF 4 and the five most significant interfering gases in the "Run 1" micro-window selected for our CF 4 retrievals, computed for a typical ISSJ observation on 1 April 2000 (at 63.46 • apparent solar zenith angle and 0.006 cm −1 resolution).Their combination results in the black simulation trace (labeled "Sim.")which, compared to the actual observation (green curve, labeled "Obs.")leads to the red residuals (observed minus calculated signals) in Frame B. Because of strong perturbations by the H 2 O and HDO interferences, this initial "Run 1" was followed by a second composite "Run 2", with corresponding residuals displayed by the thick green trace in Frame B.
et al. (2010), assuming a lag time of 3 yr for ground-level emissions to propagate in the free troposphere and stratosphere.

Fig. 3 .
Fig. 3.Comparison between selected CF 4 mixing ratio time series and trends deduced from infrared remote-sensing and in situ surface measurements (see legend and text for their identification).The various data sets can all be reconciled in term of absolute concentration when accounting for the associated uncertainties affecting the retrieved quantities (calibration scales, line parameters) and time needed for a thorough mixing of CF 4 throughout the atmosphere.See text for details.

Table 1 .
Mühle et al. (2010))dom and systematic errors on typical individual CF 4 total column retrievals above the Jungfraujoch.molefractionswouldbeverysmall,sinceonly the driest observations were retained here (water vapor column of maximum 8 × 10 21 molec cm −2 , for an air column of ∼ 1.4 × 10 25 molec cm −2 above ISSJ).At first glance, two major features emerge from Fig.2, namely i. the large increase of the CF 4 column loading above ISSJ by 2.8 × 10 14 molec cm −2 between 1989 and 2012, corresponding to +35 % when referenced to 1989.This increase is entirely of anthropogenic origin; it jumps to over 80 % when the "natural" background level of ∼ 35 ppt, recently reported byWorton et al. (2007)andMühle et al. (2010), is taken into account.As about one third of the atmospheric mass is located below the Jungfraujoch altitude (mean pressure = 654 hPa) and assuming that the very stable CF 4 gas is uniformly distributed through the atmosphere, its increase above ISSJ translates into a total column change above sea level equal to +4.35 × 10 14 molec cm −2 , from 1989 to 2012.Globally, this has required a cumulated anthropogenic CF 4 emission at the ground totaling nearly 320 Gg over that time frame.
(2008;a tool based on a Fourier series that allows calculation, at the 2σ confidence level, of the long-term linear component as well as the seasonal modulation of a given data set), returned mean linear yearly increases above ISSJ of (1.38 ± 0.11) × 10 13

Table 2 .
Discrete six-year time averaged CF 4 trends based on tangential derivatives to the black line of Fig. 2, and corresponding emissions

Table 3 .
Annual rate of increase expressed in ppt yr −1 for atmospheric CF 4 computed for three reference times.