Satellite observations of stratospheric carbonyl fluoride

The vast majority of emissions of fluorinecontaining molecules are anthropogenic in nature, e.g. chlorofluorocarbons (CFCs), hydrochlorofluorocarbons (HCFCs), and hydrofluorocarbons (HFCs). These molecules slowly degrade in the atmosphere, leading to the formation of HF, COF2, and COClF, which are the main fluorinecontaining species in the stratosphere. Ultimately both COF2 and COClF further degrade to form HF, an almost permanent reservoir of stratospheric fluorine due to its extreme stability. Carbonyl fluoride (COF 2) is the secondmost abundant stratospheric “inorganic” fluorine reservoir, with main sources being the atmospheric degradation of CFC-12 (CCl 2F2), HCFC-22 (CHF2Cl), and CFC-113 (CF2ClCFCl2). This work reports the first global distributions of carbonyl fluoride in the Earth’s atmosphere using infrared satellite remote-sensing measurements by the Atmospheric Chemistry Experiment Fourier transform spectrometer (ACEFTS), which has been recording atmospheric spectra since 2004, and the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) instrument, which recorded thermal emission atmospheric spectra between 2002 and 2012. The observations reveal a high degree of seasonal and latitudinal variability over the course of a year. These have been compared with the output of SLIMCAT, a state-of-the-art three-dimensional chemical transport model. In general the observations agree well with each other, although MIPAS is biased high by as much as ∼ 30 %, and compare well with SLIMCAT. Between January 2004 and September 2010 COF 2 grew most rapidly at altitudes above ∼ 25 km in the southern latitudes and at altitudes below ∼ 25 km in the northern latitudes, whereas it declined most rapidly in the tropics. These variations are attributed to changes in stratospheric dynamics over the observation period. The overall COF 2 global trend over this period is calculated as 0 .85± 0.34 (MIPAS), 0.30± 0.44 (ACE), and 0.88 % year −1 (SLIMCAT).


Introduction
Although small quantities of fluorine-containing molecules are emitted into the atmosphere from natural sources, e.g. volcanic and hydrothermal emissions (Gribble, 2002), the vast majority of emissions are anthropogenic in nature, e.g. chlorofluorocarbons (CFCs), hydrochlorofluorocarbons (HCFCs), and hydrofluorocarbons (HFCs). Most flu-5 orine in the troposphere is present in its emitted "organic" form due to these molecules having typical lifetimes of a decade or longer, however photolysis in the stratosphere, which liberates fluorine atoms that react with methane, water or molecular hydrogen, result in the formation of the "inorganic" product hydrogen fluoride, HF. At the top of the stratosphere (∼ 50 km altitude), ∼ 75 % of the total available fluorine is present as HF Introduction For CFC-113 and more minor sources such as HFCs (e.g. HFC-134a, HFC-152a), the reaction scheme is similar. COF 2 volume mixing ratios (VMRs) slowly increase with altitude up to the middle of 10 the stratosphere, above which they decrease as photolysis of COF 2 becomes more efficient, leading to the formation of fluorine atoms, As mentioned earlier, these F atoms react with CH 4 , H 2 O or H 2 to form HF. Vertical profiles of COF 2 in the atmosphere have previously been determined from measurements taken by the Atmospheric Trace MOlecule Spectrometry Experiment (ATMOS) instrument which flew four times on NASA space shuttles between 1985 20 and 1994 (Rinsland et al., 1986;Zander et al., 1994). Additionally, there have been several studies into the seasonal variability of COF 2 columns above Jungfraujoch using ground-based Fourier transform infrared (FTIR) solar observations (Mélen et al., 1998;Duchatelet et al., 2009 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | trends, and seasonal and latitudinal variability. This work presents the first global distributions of COF 2 using data from two satellite limb instruments: the Atmospheric Chemistry Experiment Fourier transform spectrometer (ACE-FTS), onboard the SCISAT satellite, which has been recording atmospheric spectra since 2004, and the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) instrument (Fischer et al.,

ACE-FTS spectra
The ACE-FTS instrument, which covers the spectral region 750 to 4400 cm −1 with a maximum optical path difference (MOPD) of 25 cm and a resolution of 0.02 cm −1 (using the definition of 0.5/MOPD throughout), uses the sun as a source of infrared 15 radiation to record limb transmission through the Earth's atmosphere during sunrise and sunset ("solar occultation"). Transmittance spectra are obtained by ratioing against exo-atmospheric "high sun" spectra measured each orbit. These spectra, with high signal-to-noise ratios, are recorded through long atmospheric limb paths (∼ 300 km effective length), thus providing a low detection threshold for trace species. ACE has an 20 excellent vertical resolution of about 3-4 km and can measure up to 30 occultations per day, with each occultation sampling the atmosphere from 150 km down to the cloud tops (or 5 km in the absence of clouds). The locations of ACE occultations are dictated by the low Earth circular orbit of the SCISAT satellite and the relative position of the sun. Over the course of a year, the ACE-FTS records atmospheric spectra over a large Introduction The atmospheric pressure and temperature profiles, the tangent heights of the measurements, and the carbonyl fluoride VMRs were taken from the version 3.0 processing of the ACE-FTS data (Boone et al., 2005(Boone et al., , 2013. Vertical profiles of trace gases (along with temperature and pressure) are derived from the recorded transmittance spectra via an iterative Levenberg-Marquardt nonlinear least-squares global fit to the selected 5 spectral region(s) for all measurements within the altitude range of interest, according to the equation In Eq. (1), x is the state vector, i.e. the atmospheric quantities to be retrieved, y the 10 vector of measurements (over a range of tangent heights), S y the measurement error covariance matrix (assumed to be diagonal), λ the Levenberg-Marquardt weighting factor, F the radiative transfer (forward) model, b the forward model parameter vector, i the iteration number, and K is the Jacobian matrix (≡ ∂F/∂x). The microwindow set and associated altitude ranges are listed in Table 1. The VMRs   15 for molecules with absorption features in the microwindow set (see Table 2) were adjusted simultaneously with the COF 2 amount. All spectroscopic line parameters were taken from the HITRAN 2004 database (Rothman et al., 2005). The v3.0 COF 2 retrieval extends from a lower altitude of 12 km up to 34 km at the poles and 45 km at the equator, with the upper limit varying with latitude (see Table 1). For ACE spectra recorded 20 at tangent heights that fall within the selected retrieval altitude range, the initial VMRs (which do not vary with season or latitude) for the least-squares fit are taken from the set of VMR profiles established by the ATMOS mission (Irion et al., 2002). The COF 2 spectral signal in ACE spectra recorded above the upper altitude retrieval limit (see Table 1)  MOS, or a priori, profile in that altitude region; this scaling factor is determined during the least-squares fitting. An ACE-FTS transmittance spectrum in the region of one of the microwindows is plotted in the top panel of Fig. 1. This measurement comes from occultation ss11613 (recorded on 9 October 2005 south of Mexico, over the Pacific Ocean) at a tangent 5 height of 28.9 km. The second panel reveals the calculated contribution to the measurement of COF 2 based on its retrieved VMR (∼ 3 %); three spectral features are clearly due to absorption of COF 2 . The third panel gives the observed -calculated residuals for the retrieval without the inclusion of COF 2 in the forward model; the shape of these residuals matches well with the calculated COF 2 contribution. The bottom 10 panel contains the observed -calculated residuals, indicating the goodness of the fit.

MIPAS spectra
The MIPAS instrument, a Fourier transform spectrometer, measures the thermal limb emission of the Earth's atmosphere in the mid-infrared spectral region, 685-2410 cm −1 .
Launched in March 2002, the first two years of spectra were recorded at an unapodised 15 resolution of 0.025 cm −1 (MOPD = 20 cm). The nominal scan pattern consisted of 17 tangent points per scan (FR17, FR = full resolution) from 6-68 km altitude with a minimum vertical spacing of 3 km. A mechanical degradation of the interferometer's mirror drive led to a cessation in measurements, with a resumption in operations in January 2005 at a reduced resolution of 0.0625 cm −1 (MOPD = 8 cm). The new nominal 20 scan pattern consisted of 27 tangent points per scan (OR27, OR = optimised resolution) over altitude ranges that varied with latitude, from 5-70 km at the poles to 12-77 km at the equator; this variation, which approximately follows the tropopause shape, minimises the number of spectra lost to cloud contamination. The vertical spacing of OR27 scans ranges from 1.5 km at lower altitudes and 4.5 at higher altitudes. Note 25 that the reduction in scan time associated with the lower spectral resolution resulted in an increase in the number of tangent points (an additional ten) within the limb scan, 18133 mal estimation approach is (e.g. Rodgers, 2000): where the new terms x a and S a represent the a priori estimate of x and its error covariance, respectively. However, rather than applying the above equation to the full set of measurements y, MORSE uses a sequential estimation approach (Rodgers, 2000) and applies Eq.
(2) successively to spectral subsets defined by each microwindow at each tangent altitude. For this work, the a priori estimate is taken from IG2 COF 2 profiles (Remedios et al., 2007). The spectral microwindows and associated altitude ranges are listed in Table 3; the retrieval extends from a lower altitude of 7.5 up to 54.0 km.

15
For COF 2 retrievals, the MORSE state vector consists of the profile of COF 2 plus, for each microwindow (see Table 4), a profile of atmospheric continuum and a radiometric offset (intended to remove any spectrally smooth background variations within each microwindow, e.g., due to aerosols or thin clouds as well as any residual altitudedependent radiometric offsets). The forward model uses pressure, temperature and the 20 abundances of major contaminating species (H 2 O, O 3 , HNO 3 , CH 4 , N 2 O and NO 2 ) retrieved earlier from the same spectra (using MORSE), and IG2 profiles for other minor gases. Spectroscopic data were taken from the MIPAS PF3.2 database (Flaud et al., 2006), with the COF 2 data in this compilation coming from the HITRAN 2004 database (Rothman et al., 2005). As with all MORSE VMR retrievals, the diagonal elements of 25 S a were set to (100 %) 2 ; since MORSE retrieves ln(VMR) rather than VMR, the S a diagonal elements are profile-independent. The off-diagonal elements of S a are set assuming a (strong) vertical correlation length of 50 km, which provides regularisation at Introduction the expense of vertical resolution. Finally, cloud-contaminated spectra were removed using the cloud index method (Spang et al., 2004) with a threshold value of 1.8. Note that unlike the ACE-FTS retrievals, MORSE retrieves COF 2 at altitudes well above the VMR maximum, even though the information at high altitude is almost entirely from the a priori profiles. Thus, any special treatment to scale the a priori is not 5 required, although, through the vertical correlation, the effect is similar to that explicitly applied for ACE. Additionally, unlike ACE, MORSE uses MIPAS spectra with the Norton Beer strong apodisation applied, hence S y is banded rather than diagonal. Figure 2 provides a plot that illustrates the COF 2 spectral feature in one of the MIPAS microwindows. The top panel shows an averaged MIPAS radiance spectrum (in black) 10 interpolated to 20 km altitude from equatorial measurements taken in March 2010 for the 772-775 cm −1 microwindow; in red is the averaged calculated spectrum based on the averaged retrieved VMRs, but without the inclusion of COF 2 in the forward model. The second panel reveals the averaged calculated COF 2 contribution to the spectrum. The third panel gives the observed -calculated residuals for the retrieval (in black), 15 again without the calculated COF 2 contribution; the shape of these residuals matches well with the calculated COF 2 contribution in the second panel. Overlaid in red are the overall observed -calculated residuals, indicating the goodness of the retrieval.
Note that for comparison with ACE-FTS v3.0 COF 2 VMR profiles, the COF 2 MIPAS profiles are interpolated from the standard pressure grid onto the same 1 km grid used 20 by ACE.

Infrared spectroscopy of carbonyl fluoride
Retrieving COF 2 VMR profiles from ACE-FTS and MIPAS spectra crucially requires accurate laboratory COF 2 spectroscopic measurements. Uncertainty in the laboratory 25 data can directly contribute to systematic errors in the COF 2 retrievals. reported here make use of three band systems of COF 2 ; these bands largely correspond to the ν 1 (1943 cm −1 ; CO stretch), ν 4 (1243 cm −1 ; CF 2 antisymmetrical stretch), and ν 6 (774 cm −1 ; out-of-plane deformation) fundamental modes.
Both ACE-FTS and MIPAS retrievals make use of the COF 2 linelist first released as part of the HITRAN 2004 database (and remaining unchanged for the HITRAN 2008 5 release), with partition data taken from the TIPS subroutine included in the HITRAN compilation. In particular, the ACE-FTS retrieval makes use of spectroscopic lines in the ν 1 and ν 4 bands, whereas MIPAS uses ν 4 and ν 6 .
HITRAN employs error codes in the form of wavenumber errors for the parameters ν (line wavenumber) and δ air (air pressure-induced line shift) and percentage errors for 10 S (line intensity), γ air (air-broadened half-width), γ self (self-broadened half-width), and n air (temperature-dependence exponent for γ air ). Each error code corresponds to an uncertainty range, but with no information as to how the parameters are correlated.
In HITRAN the parameter δ air (air pressure-induced line shift) is assumed to have a value of 0 cm −1 atm −1 . The same values of γ air (0.0845 cm −1 atm −1 at 296 K), γ self 15 (0.175 cm −1 atm −1 at 296 K), and n air (0.94) are used for all COF 2 spectral lines in HITRAN; according to the error codes these values are averages/estimates. They are taken from the work of May (1992), who determined these average parameters for selected lines in the ν 4 and ν 6 bands from measurements made by a tunable diodelaser spectrometer. For the ν 1 band most of the spectral lines used in the retrievals 20 have stated intensity uncertainties ≥ 20 %, for the ν 4 band between 10 % and 20 %, and for the ν 6 band the errors are listed as unreported/unavailable. After performing the MIPAS retrievals, the latest HITRAN2012 update was released, which revises the ν 6 band and includes several weak hot bands. The listed intensity uncertainties for this band have been revised to between 10 % and 20 %; spectral simulations indicate only 25 minor intensity differences in the ν 6 band Q branch between the two linelists.
As part of the present study, a comparison was made between an N 2 -broadened (760 Torr) composite spectrum of COF 2 (determined from multiple pathlengthconcentration burdens) at 278 K and 0.112 cm −1 resolution, taken from the Pacific Introduction Northwest National Laboratory (PNNL) IR database (Sharpe et al., 2004) with a synthetic spectrum calculated using HITRAN2004 COF 2 line parameters for the same experimental conditions; the maximum systematic error of the PNNL intensities is 2.5 % (1σ). The comparison reveals that the integrated ν 1 and ν 4 band intensities in the PNNL spectrum are ∼ 15 % higher than HITRAN, whereas the integrated intensity of the very 5 strong Q branch in the ν 6 band of the PNNL spectrum is ∼ 20-25 % higher than HI-TRAN. Furthermore, the air-broadened half-width in HITRAN for this Q branch appears to be too large at 760 Torr. May (1992) states that the average pressure-broadening coefficients, which are included in HITRAN, could not reproduce the experimental pressure-broadened spectra satisfactorily over the full Q-branch region. The author suggests this may be a result of the J (rotational quantum number)-dependence of the pressure-broadening coefficients or other effects such as line mixing (Hartmann et al., 2008). When selecting appropriate ACE microwindows from the ν 1 and ν 4 bands, it was noticed that a number of COF 2 lines suffered from systematic bad residuals. Since the 15 COF 2 lines occur in clusters, i.e. are not isolated, there is a strong suggestion that line mixing is playing a role; unfortunately there are no available spectroscopic line parameters that describe line mixing for COF 2 . Although the ACE v3.0 retrieval only employs lines with the best residuals, there could still remain a small contribution to the error from the neglect of line mixing. On the other hand, lines in the ν 6 Q branch (employed 20 in the MIPAS retrievals) are very tightly packed, so if line mixing effects are important, errors arising from their neglect will likely be larger for MIPAS retrievals compared with ACE. Unfortunately it is an almost impossible task to quantify these errors without accurate quantitative measurements at low temperatures and pressures. For the purposes of this work it is estimated that retrieval errors arising from COF 2 spectroscopy are at 25 most ∼ 15 %, however since different bands are used in the respective retrievals, it is likely there will be a relative spectroscopic-induced bias between the two schemes. Introduction

ACE-FTS spectra
The ACE v2.2 COF 2 data product has previously been validated against measurements taken by the JPL MkIV interferometer, a balloon-borne solar occultation FTS (Velazco et al., 2011). Unlike the v3.0 product, the upper altitude limit for the v2.2 retrieval is fixed at 32 km, with the scaled ACE a priori profile used above 32 km. MkIV and ACE 5 v2.2 profiles from 2004 and 2005 agree well within measurements error. However, it must be recognised that both retrievals make use of the same COF 2 spectroscopic data, which has an estimated systematic error of at most ∼ 15 % (see Sect. 3.1). For a single ACE profile, the 1σ statistical fitting errors are typically ∼ 10-30 % over most of the altitude range. These errors are random in nature and are largely deter-10 mined by the measured signal-to-noise ratios of the ACE-FTS spectra, i.e. measurement noise. For averaged profiles, the random errors are small (reduced by a factor of 1/ √ N, where N is the number of profiles averaged) and the systematic errors dominate. Spectroscopic sources of systematic error predominantly arise from the COF 2 HI-15 TRAN linelist (∼ 15 %; see Sect. 3.1), with minor contributions from interfering species that absorb in the microwindow regions. Since the baselines of the ACE-FTS transmittance spectra and the VMRs of the interferers (H 2 O, CO 2 , O 3 , N 2 O, CH 4 , NO 2 , NH 3 , HNO 3 , HOCl, HCN, H 2 O 2 , CCl 4 , ClONO 2 , N 2 O 5 ) are fitted simultaneously with the COF 2 VMR, it is not a trivial exercise to determine how much they contribute to the 20 overall systematic error of the COF 2 retrieval. In this work, the view is taken that the lack of systematic features in the spectral residuals indicates that these contributions are small, at most 1 %. In addition to spectroscopic errors, uncertainties in temperature, pressure, tangent altitude (i.e. pointing) and instrumental line shape (ILS) all contribute to systematic er-Introduction The fractional retrieval error, µ j , is defined as Note that for the ACE-FTS retrievals, pressure, temperature and tangent height are not strictly independent quantities; tangent heights are determined from hydrostatic 5 equilibrium, and so these quantities are strongly correlated. For the purposes of this work, only two of these quantities are altered: temperature is adjusted by 2 K and tangent height by 150 m (Harrison and Bernath, 2013). Additionally, ILS uncertainty is induced by adjusting the field of view by 5 % (Harrison and Bernath, 2013). A small subset of occultations was selected for this analysis. The fractional value estimates of the systematic uncertainties, and their symbols, are given in Table 6. Assuming these quantities are uncorrelated, the overall systematic error in the COF 2 retrieval can be calculated as The total systematic error contribution to the ACE-FTS COF 2 retrieval is estimated to be ∼ 16 %. As discussed in Sect. 3.1, the COF 2 absorption signal in ACE-FTS spectra decreases relative to the noise as the retrieval extends to higher altitude despite the a priori profile indicating that the COF 2 VMRs do not effectively drop to zero until ∼ 55 km. 20 For this reason an upper altitude limit (see Table 1) is set; the retrieval is pushed as high in altitude as possible. The portion of the retrieved VMR profile above the highest analysed ACE measurement (i.e. the spectrum at the highest tangent height, just below the upper altitude limit) is calculated by scaling the a priori profile.
In an ACE retrieval, the calculated spectrum is generated from the sum of contribu-25 tions from the tangent layer up to 150 km. For the highest analysed measurement, the retrieved VMR in the tangent layer is generated from the piecewise quadratic interpolation scheme (Boone et al., 2005(Boone et al., , 2013, while the VMR in every layer above that will 18139 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | come from scaling the a priori profile; the scaling factor largely comes from forcing the calculated spectrum to match as best as possible the measured spectrum for this one measurement. If the shape of the a priori profile above the highest analysed measurement is incorrect, the contribution to the calculated spectrum from that altitude region will be incorrect for the second highest measurement analysed; the VMRs between the 5 tangent layers of the two highest analysed measurements are adjusted in the retrieval to compensate. Therefore, errors in the a priori VMR profile will introduce systematic errors into the highest altitudes of the retrieved profile.

MIPAS spectra
The precision, or random error, of the retrieved COF 2 VMRs is calculated via the prop-10 agation of the instrument noise and the a priori error through the standard optimal estimation retrieval (using the MORSE code). The total retrieval covariance matrix (neglecting systematic errors) is given by (Rodgers, 2000),

15
Note that this expression effectively represents a combination of the noise-induced random error and the assumed a priori error covariance (sometimes called "smoothing error"), and that some caution is required if interpolating error profiles to different grids. Profile levels with random errors larger than 70 %, mostly at the top and bottom of the retrieval range, are discarded from the dataset and not used in the analysis. Since the 20 a priori profiles have an assumed error of 100 %, this ensures that the retrieved profile levels contain, at worst, ∼ 50 % contribution from the a priori. For a single profile, the noise error is typically 5-15 % between 20-40 km, covering the peak of the COF 2 VMR profile; over this range the contribution to the retrieved profiles principally comes from the measurements. Outside this range, the errors increase rapidly as the COF 2 VMR Introduction The total error is computed by propagating a number of independent error sources expressed as spectra through the retrieval gain matrix, including both spectral correlations and correlations through the pressure-temperature retrieval. For a single profile, the primary error sources are the measurement noise followed by assumed uncertainties in the O 3 (stratosphere) and N 2 O (troposphere) concentrations, which typically con-5 tribute 15 % uncertainty in retrieved COF 2 values. Spectroscopic errors in interfering molecules are treated as a single error, assuming a position uncertainty of 0.001 cm −1 , 5 % in line strength and 0.1 cm −1 in the half-width of COF 2 lines. Figure 3 shows the single-profile error budget for COF 2 , with total errors typically 20-30 % between 20-40 km. 10 MIPAS COF 2 profiles are retrieved at the tangent points of each measurement; these vary from scan to scan. The conversion to absolute altitude for comparison with ACE-FTS COF 2 profiles relies on the MIPAS pointing information, which may lead to a vertical offset of a few hundred metres relative to ACE.
The quality of the MIPAS COF 2 retrieval can be measured using the averaging kernel 15 matrix (Rodgers, 2000), A, where I is the identity matrix. In general, for a given profile, rows of A are peaked functions, peaking at the appropriate altitude range for the observation; the width of spacing at profile level i , and A i i is the corresponding diagonal element of the averaging kernel matrix; the vertical resolution of the MIPAS retrievals is ∼ 4-6 km near the COF 2 profile peak, dropping off outside this range.

Global distribution and vertical profiles
For a detailed comparison between ACE-FTS and MIPAS observations, it was decided 5 to focus on one year of measurements between September 2009 and August 2010. Figure 7 provides a comparison between individual profiles for four near-coincident sets of measurements; these are the four closest sets available over this time period. The locations and times of the eight observations can be found in Table 6. The plots also include the a priori profiles, and calculated SLIMCAT profiles for the location and 10 time of each ACE-FTS observation; these calculations will be discussed in Sect. 5. In Fig. 7, the upper altitudes of the ACE-FTS profiles without error bars correspond to the regions where the a priori profiles are scaled in the retrieval (see Sect. 3.2). Although the pairs of measurements were taken at slightly different locations and times of day, near-coincident profiles are expected to be similar but not necessarily identical. COF 2 15 profiles initially show an increase in VMR with altitude, peaking in the stratosphere and then decreasing with higher altitude; the peak location depends on the latitude and time of year. On the whole, the MIPAS and ACE profiles in Fig. 7  were removed. The MIPAS dataset contains substantially more observations over the globe, and as discussed earlier, profile levels with random errors larger than 70 % of the retrieved VMRs were discarded. For each month, a global spike test was applied to all the remaining data. At each altitude the mean and standard deviation of the ensemble were calculated. Any MIPAS profiles with one or more VMRs outside 5σ of the mean 5 VMRs were discarded. This spike test was repeated until all remaining MIPAS profiles were within this 5σ range.
MIPAS observations indicate a minor diurnal variation in COF 2 VMRs, below the measurement error. In this work ACE and MIPAS zonal means were produced without any consideration of the local solar time of the individual measurements. Figure 8 10 provides a direct side-by-side comparison of MIPAS and ACE zonal means for each of the twelve months, revealing the seasonal variation in the COF 2 distribution. The plotted VMRs are the averages for each month of all filtered data at each altitude within 5 • latitude bins. The highest COF 2 VMRs appear at ∼ 35 km altitude over the tropics, which receive the highest insolation due to the small solar zenith angle; these 15 peaks are located ∼ 10 • S for December to April, and ∼ 10 • N for June to October. COF 2 has a lifetime of ∼ 3.8 years (calculated from SLIMCAT; refer to Sect. 5) and is transported polewards by the Brewer-Dobson circulation. As can be seen in  More details on these atmospheric processes will be discussed in Sect. 5, along with a discussion of the SLIMCAT CTM.
Since there are only a maximum of 30 ACE-FTS profiles measured per day, compared to ∼ 1300 for MIPAS (OR27), the global coverage of the ACE observations between September 2009 and August 2010 is poorer and noisier in appearance, partic-5 ularly in the tropics. Despite this, the ACE observations agree well with MIPAS, apart from the apparent high bias in the MIPAS VMRs, which will be discussed later in this section. As examples, note the good agreement at mid-to high-latitudes in the SH be- Since zonal mean plots do not provide an indication of measurement errors, a representative set of individual latitude bins are plotted in Fig. 9 with error bars; all errors are 15 defined as the standard errors of the bin means. Such plots are useful to inspect biases between datasets. Note that SLIMCAT calculations are also included in this figure; these will be further discussed in Sect. 5. ACE random errors are largest close to the tropics at the highest altitudes of the retrieval (where the black error bars are longest). At these altitudes COF 2 features in ACE-FTS spectra are weaker, so the relative noise 20 contribution to the retrieved VMRs is larger. The retrieved ACE VMR profiles at these altitudes have a rather flat appearance, whereas the corresponding MIPAS profiles are peaked. The MIPAS VMRs themselves are biased as much as 30 % higher than ACE, although there is overlap between the error bars. At the very highest altitudes (above ∼ 50 km), the ACE VMRs drop to zero, and the MIPAS VMRs approach ∼ 50 ppt; these 25 differences result from the different a priori profiles used for the two retrieval schemes. A more detailed discussion on this point will be made in Sect. 5. Figure 9 also reveals a bias at high latitudes in the summer, where the ACE and MIPAS profiles peak just above 30 km. (The summer SH high-latitude peak corresponds to a secondary maxi- Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | mum in the VMR distribution; the origin of this will be discussed in Sect. 5.) As in the tropics, MIPAS VMRs at the peak are ∼ 30 % higher than ACE. Note that for these particular months, the ACE-FTS was taking many measurements at high latitudes, hence the smaller error bars.
5 Comparison with SLIMCAT 3-D chemical transport model 5 ACE and MIPAS observations have been compared with output from the SLIMCAT off-line 3-D CTM. SLIMCAT calculates the abundances of a number of stratospheric gases from prescribed source-gas surface boundary conditions and a detailed treatment of stratospheric chemistry, including the major species in the O x , NO y , HO x , Cl y and Br y chemical families (e.g. Chipperfield, 1999;Feng et al., 2007). The model uses 10 winds from meteorological analyses to specify horizontal transport while vertical motion in the stratosphere is calculated from diagnosed heating rates. This approach gives a realistic stratospheric circulation (Chipperfield, 2006;Monge-Sanz et al., 2007). The troposphere is assumed to be well-mixed. For this study SLIMCAT was integrated from 2000 to 2012 at a horizontal resolution 15 of 5.6 • × 5.6 • and 32 levels from the surface to 60 km. The model uses a σ − θ vertical coordinate (Chipperfield, 2006) and was forced by European Centre for Medium Range Weather Forecasts (ECMWF) reanalyses (ERA-Interim from 1989 onwards). The volume mixing ratios of source gases at the surface level were specified using data files compiled for the 2010 WMO ozone assessment (WMO/UNEP, 2011). These A number of these molecules, e.g. HFC-23, are included even though they make no appreciable contribution to the formation of COF 2 compared with the major source gases. Some other HFCs, e.g. HFC-125, which similarly make minimal contribution, are not included in the model. In addition to providing a direct comparison with satellite observa-5 tions, the new SLIMCAT calculations have been used to show where COF 2 is produced and which source gases have produced it. Most COF 2 is produced in the tropics where solar insolation is highest. Figure 10 provides plots of the loss rates (annual mean zonal mean; pptv day −1 ) for the three main source gases which produce COF 2 . As can be seen, the largest contributing COF 2 source at ∼ 30-35 km is CFC-12, followed by CFC-10 113 (approximately a factor of 10 smaller). HCFC-22 is the second largest contributing source gas overall, however its contribution peaks low in the troposphere (not relevant for stratospheric COF 2 ) and higher up in the stratosphere (∼ 40-45 km). CFC-12 and CFC-113 are removed mainly by photolysis ∼ 20-40 km; above this altitude range the abundances of CFC-12 and CFC-113 tend to zero so that they make only a small con-15 tribution to the formation of COF 2 . On the other hand, HCFC-22 is mainly removed from the atmosphere by reaction with OH. Since this reaction is slower, HCFC-22 persists higher into the stratosphere than CFC-12 and CFC-113 and can therefore lead to COF 2 production in the upper stratosphere and lower mesosphere. Individual contributions from molecules other than these three are typically a small fraction of 1 %. 20 In the altitude region below the maximum COF 2 VMRs at all locations there is net production of COF 2 , while at higher altitudes there is net loss. The primary loss of COF 2 in the atmosphere occurs via photolysis, with an additional secondary loss mechanism through reaction with O( 1 D). Figure 10 also contains a plot of the COF 2 annual mean zonal total loss rate.

25
SLIMCAT has also been used to estimate the atmospheric lifetime of COF 2 by simply dividing the total modelled atmospheric burden by the total calculated atmospheric loss rate. The total calculated mean atmospheric lifetime is ∼ 3.8 years. This lifetime varies slightly between the hemispheres, 3.76 years in the south and 3.82 years in the north. In the lower stratosphere COF 2 can be regarded as a long-lived tracer (local lifetime of many years). Therefore, its tracer isopleths follow the typical tropopause-following contours of any long-lived tracer. In this sense, COF 2 is analogous to NO y which is produced from N 2 O. It has been checked as part of this work that a correlation plot of COF 2 with its major source, CFC-12, is compact in the lower stratosphere, at altitudes 5 below the region of COF 2 maxima (Plumb and Ko, 1992). As discussed in Sect. 4, Fig. 7 contains a comparison between individual ACE-FTS and MIPAS profiles for the measurements specified in Table 6. This figure also contains SLIMCAT profiles calculated for the location and time of each ACE-FTS observation. In comparison with the retrieved portion of the ACE profiles (marked by black error bars), the calculated SLIMCAT VMRs are generally slightly lower; the agreement with MIPAS is worse, however it must be acknowledged that the two sets of measurements are not strictly coincident. Additionally, SLIMCAT captures the VMR "dip" observed for ACE occultation sr34898 (at 67.27 • N on the vortex edge, 4 February 2010) near 30 km altitude, confirming that this profile samples air from the polar vortex. This explanation 15 is supported by the corresponding ACE HF profile, which shows an enhancement near 30 km due to the sampling of descended HF-rich upper-stratospheric air from the polar vortex. Figure 11 provides a comparison between SLIMCAT and ACE zonal means. In order to reduce the noise and increase the latitude coverage for the comparison, the plotted  Figure 9 shows a representative set of SLIMCAT profiles in 5 • latitude bins from the September 2009 to August 2010 time period, along with averaged ACE and MIPAS profiles. These demonstrate a very good agreement between the SLIMCAT calculations and ACE observations, although above ∼ 35 km this agreement is somewhat worse, particularly the upper parts of the ACE profiles (without error bars) which are derived 5 from the scaled a priori profile and susceptible to systematic errors (see Sect. 3.2). Whereas the ACE VMRs drop to zero at ∼ 55 km, the SLIMCAT VMRs do not reach zero even near the model top level around 60 km due to the calculated ongoing production of COF 2 from HCFC-22 (see Fig. 10). MIPAS VMRs similarly do not drop to zero, principally because the a priori profiles make a larger contribution to the retrieved 10 VMRs at these altitudes. Unfortunately, neither ACE nor MIPAS measurements are able to validate the SLIMCAT model HCFC-22/COF 2 VMRs near 55-60 km.

ACPD
In autumn when solar heating of the relevant polar region comes to an end, a stratospheric polar vortex begins to form. This is a large-scale region of air contained within a strong westerly jet stream that encircles the polar region. Reaching maximum 15 strength in the middle of winter, the polar vortex decays as sunlight returns to the polar region in the spring. Polar vortices, which extend from the tropopause up into the mesosphere, are quasi-containment vessels for air at cold temperatures and low-ozone content. They play a critical role in polar ozone depletion, more so in the Antarctic, where the vortex is larger, stronger, and longer-lived than in the Arctic. The SLIMCAT 20 September 2009 (09/2009) plot in Fig. 11 demonstrates the presence of a strong SH polar vortex by the low COF 2 VMRs at high southern latitudes; as mentioned earlier this is a consequence of the descent of upper-stratospheric air where COF 2 VMRs are very low. The breakup of the SH polar vortex as simulated by SLIMCAT occurs around November 2009(11/2009  Although some of the COF 2 present at mid-and high-latitudes can be attributed to transport of COF 2 -rich tropical air via the Brewer-Dobson circulation (a slow upwelling of stratospheric air in the tropics, followed by poleward drift through the mid-latitudes, and descent in the mid-and high-latitudes), this cannot account for the secondary maximum in VMR (∼ 31 km) present in the SH polar region for which an atmospheric 5 chemistry explanation is needed. Diagnosis of the model rates shows that in summer, photochemical production of COF 2 extends to the pole in the middle stratosphere (i.e. in polar day). Further diagnosis of the first-order loss rates of the main COF 2 precursors shows that photolysis and reactions with O( 1 D) are symmetrical between the hemispheres. The only precursor loss reaction which shows significant hemispheric 10 asymmetry is the temperature-dependent reaction of CHF 2 Cl (HCFC-22) + OH. As the SH polar summer mid-stratosphere is around 10 K warmer than the corresponding location in the NH, this reaction provides a stronger source of COF 2 in SH summer compared to the Arctic and contributes to this secondary maximum. Indeed, in a model sensitivity run where the production of COF 2 from HCFC-22 was switched off, this sec-15 ondary SH summer peak disappeared. While the first-order loss rates of the COF 2 source gas precursors are generally symmetrical between the hemispheres, this is not true for the source gases themselves. Differences in the meridional Brewer-Dobson circulation, with stronger mixing to the pole in the north and stronger descent in the south, lead to differences in the distribution of COF 2 precursors. This leads to differences in 20 COF 2 production resulting in the observed and modelled hemispheric asymmetry in COF 2 at middle latitudes.

Trends
As mentioned in the introduction, there is evidence that the atmospheric abundance of COF 2 is increasing with time (Duchatelet et al., 2009;Brown et al., 2011 a period when CFC-12 was still increasing the atmosphere, an average COF 2 linear trend of 4.0 ± 0.5 % year −1 was derived (Mélen et al., 1998). COF 2 trends from 10 more recent studies are considerably lower, largely due to the phase out of its principal source gas, CFC-12. A trend of 0.8 ± 0.4 % year −1 has recently been derived from ACE data for 2004 to 2010 (Brown et al., 2011). Since the majority of halogenated source gases reach the stratosphere by upwelling through the tropical tropopause region, the ACE COF 2 trend was determined by averaging measurements in the latitude In this section, ACE and MIPAS time series are derived as a function of altitude and latitude. As discussed previously, e.g. in Harrison and Bernath (2013), ACE latitude coverage is uneven. For data between January 2004 and September 2010 (the last month for which ACE v3.0 data is usable), the 18 10 • latitude bins used for the ACE time series contain, from southernmost to northernmost, 1000, 1323, 5265, 1776, 796, 5 608, 482, 420, 390, 394, 339, 413, 650, 1062, 2012, 4828, 1875, 1315 occultations, respectively, i.e. over three-quarters of the measurements lie in latitude bins poleward of 50 • S/N. On the other hand, MIPAS data coverage over the globe is more even and extensive, apart from some periods during 2004-2006 when nominal mode measurements were not made. such a plot has not been provided for the ACE dataset.) An annual cycle is readily observed, and as expected its phase is opposite in each hemisphere. The amplitude of 15 this cycle is largest near the poles; note that the maxima in the plot at 20.5 km altitude correspond to the descent of COF 2 in winter polar vortices. Close inspection of Fig. 12, particularly the plots above 30 km, also reveals the presence of the quasi-biennial oscillation (QBO) signal, which is strongest in the tropics. Overall, there is good agreement between the MIPAS and SLIMCAT plots in terms of the overall latitude-altitude pattern, 20 however, as noted before, the MIPAS VMRs are biased high; for example, maxima over the tropics as much as ∼ 25 % and maxima near the poles as much as ∼ 50 %. Figure 13 provides the time series for five altitude-latitude bin combinations of ACE, MIPAS and SLIMCAT data; for ease of viewing, this plot does not include errors. In all plots, the main features in the time series agree well. Note the observed QBO signal for 25 all three datasets, stronger in the two tropical plots and weaker in the high-northernlatitude plot. In the top two plots of Fig. 13 MIPAS is biased high, although less so at 20.5 km. As established previously (refer to Fig. 9), this is a feature of the MIPAS dataset at the high southern latitudes. The agreement between ACE and SLIMCAT is Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | somewhat better, agreeing within the errors of the ACE data, although less so at high southern latitudes. COF 2 trends at each altitude for all 18 latitude bins have been calculated from monthly percentage anomalies in COF 2 zonal means, C z,θ (n), defined as 5 where n is a running index from month zero to 80 (January 2004 to September 2010), VMR z,θ (n) is the corresponding mixing ratio at altitude z and latitude θ, VMR z,θ (m) is the average of all zonal means for each of the twelve months, m, and δ nm , although not used in its strict mathematical sense, is one when index n corresponds to one of the 10 months m and is zero otherwise. In order to compare the three datasets, the same time period was used for each analysis. Such an approach essentially removes the annual cycle and the effect of biases in VMRs; the trend is simply equated to the "slope" of the linear regression between C z,θ (n) and the dependent variable n/12. The inclusion of additional terms such as the annual cycle and its harmonics resulted in no additional 15 improvement in the regression. Figure 14 presents the annual percentage trends (January 2004 to September 2010) for ACE, MIPAS, and SLIMCAT as a function of latitude and altitude. The plotting range has been chosen to cover the maximum VMR features in the COF 2 global distributions; this broadly follows the upper altitude range of the actual ACE retrievals and removes 20 portions of the MIPAS profiles which have the largest contributions from the a priori profile. Note that whereas the MIPAS time series used to derive trends contains data for 67 distinct months in all latitudes band, the number of months of ACE data available varies from as low as 15 to as high as 63 in each latitude band. Errors were not explicitly treated in the linear regression of the SLIMCAT outputs, but were for the MIPAS and ACE VMRs. Note that as the MIPAS and ACE trends approach zero, the ratio to their 1σ uncertainties drops well below one. Broadly speaking, the trends for any ACE/MIPAS latitude-altitude region in Fig. 14 which appear predominantly blue or green become more statistically significant when the individual contributions are averaged. The MIPAS plot in Fig. 14 indicates that between 2004 and 2010, COF 2 has in-5 creased most rapidly (approaching ∼ 4 % per year) at altitudes above ∼ 25 km in the southern latitudes and at altitudes below ∼ 25 km in the northern latitudes. The ACE plot broadly agrees with respect to these two regions of largest positive trend, although their magnitudes are slightly lower. Additionally, the ACE trends in the tropical region are predominantly negative, which somewhat agrees with SLIMCAT below 25 km. 10 The SLIMCAT plot contains a number of features which agree with both the MIPAS and ACE plots. In particular, the SLIMCAT plot indicates a decrease in COF 2 in the tropical region (between 20 • S and 10 • N), although the largest decrease occurs at ∼ 27 km and 0 • latitude; ACE agrees better than MIPAS in this region, except for a narrow altitude range ∼ 30 km where the ACE trends are slightly positive. Outside the tropics, the 15 SLIMCAT plot agrees better with MIPAS, in particular for the regions of largest positive trends which occur at high southern latitudes above 30 km and northern latitudes below ∼ 25 km. An additional SLIMCAT run has been performed with dynamics arbitrarily fixed to those for the year 2000; results from this run give a "clean" COF 2 signal without the 20 complication of changes in stratospheric dynamics. Trends have been calculated in the same manner as above, and plotted in the lowest panel of Fig. 14. Compared with trends for the "control" SLIMCAT run, those for the fixed-dynamics run lie predominantly between 0 and 1 %, with a relatively uniform distribution throughout the stratosphere. This indicates that the variations in SLIMCAT trends, and by extension the regions  -12 since ∼ 2003(WMO/UNEP, 2011; note that HCFC-22 produces COF 2 at higher altitudes. However, the absence of any negative tropical trends in the fixed-dynamics SLIMCAT plot indicates that this feature must result from dynamical considerations. The analyses used to force the SLIMCAT calculations provide information on the stratospheric circulation, but do not allow for any rigorous explanation of the changing stratospheric dynamics that are responsible for the observed trends. Interestingly, the two regions of large positive trends in the ACE, MIPAS, and SLIMCAT plots correspond quite well to the regions of positive age of air trends, as reported by Stiller et al. (2012); see Fig. 10. Additionally, the region of positive trends in the tropics ∼ 28-10 35 km, contained in the ACE plot, more-or-less agrees with the corresponding feature in the age-of-air-trend plot. As discussed by Stiller et al. (2012), it is likely that variations in atmospheric mixing have occurred over the observation period. The regions of maximum COF 2 trends must result from increased in-mixing of COF 2 -rich air, possibly due to major sudden stratospheric mid-winter warmings. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | into a small time-dependent negative bias in COF 2 VMR, meaning that the trend derived from ACE v3.0 data is biased low by on average ∼ 0.1 % year −1 , although it is not obvious how the bias varies with latitude and altitude. Plans are currently underway to create a new ACE processing version 4.0, in which it is assumed that the CO 2 VMR increases by 0.5 % year −1 and in which age of air 5 considerations are used to generate the vertical CO 2 VMR profile as a function of latitude and time of year (Toon, personal communication, 2012). It is anticipated that the new v4.0 will enable more accurate trends to be derived. The ACE-FTS continues to take atmospheric measurements from orbit, with only minor loss in performance; it will be possible to extend the COF 2 time series to the present day and beyond.

15
This work reports the first global distributions of carbonyl fluoride in the Earth's atmosphere using infrared satellite remote-sensing measurements by the ACE-FTS, which has been recording atmospheric spectra since 2004, and the MIPAS instrument, which has recorded thermal emission atmospheric spectra between 2002 and 2012. The observations reveal a high degree of seasonal and latitudinal variability over the course of 20 a year, and agree well with the output of the SLIMCAT model, although MIPAS VMRs are biased high by as much as ∼ 30 %.
The maximum in the COF 2 VMR distribution occurs at ∼ 30-35 km altitude in the tropics where solar insolation is highest; this region is dominated by COF 2 formed from the photolysis of CFC-12 and CFC-113. The first-order loss rates of the main 25 COF 2 precursors are symmetrical between the hemispheres, except for the HCFC-22 + OH reaction, which is temperature dependent; a secondary maximum at ∼ 25-30 km Introduction altitude is present at high latitudes in SH summer due to the mid-stratosphere being around 10 K warmer than the corresponding location in the NH summer. There is also asymmetry in the distribution of COF 2 precursors due to differences in the meridional Brewer-Dobson circulation, with stronger mixing to the pole in the north and stronger descent in the south; this results in larger VMRs at mid-and high-latitudes in the SH.
Between January 2004 and September 2010 COF 2 grew most rapidly at altitudes above ∼ 25 km in the southern latitudes and at altitudes below ∼ 25 km in the northern latitudes, whereas it declined most rapidly in the tropics. These variations are attributed to changes in stratospheric dynamics over the observation period. The overall COF 2 global trend over this period is calculated as 0.85 ± 0.34 % year −1 (MIPAS), Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Lowe, R. P., Mahieu, E., McConnell, J. C., McHugh, M., McLeod, S. D., Michaud, R., Midwinter, C., Nassar, R., Nichitiu, F., Nowlan, C., Rinsland, C. P., Rochon, Y. J., Rowlands, N., Semeniuk, K., Simon, P., Skelton, R., Sloan, J. J., Soucy, M.-A., Strong, K., Tremblay, P., Turnbull, D., Walker, K. A., Walkty, I., Wardle, D. A., Wehrle, V., Zander, R., and   Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Figure 3. The single-profile total error budget for the MIPAS COF 2 retrieval. The total error is computed by propagating a number of independent error sources expressed as spectra through the retrieval gain matrix, including both spectral correlations and correlations through the pressure-temperature (PT) retrieval. Note that NESR is the noise equivalent spectral radiance, SHIFT refers to the uncertainty in the spectral calibration (±0.001 cm −1 ), SPECDB (spectroscopic database errors) refers to uncertainties in the strength, position and width of COF 2 infrared lines, and GRA refers to the uncertainty due to an assumed ±1 K/100 km horizontal temperature gradient. More details are contained in the text.