Envisat MIPAS measurements of CFC-11: retrieval, validation, and climatology

From July 2002 to March 2004 the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) aboard the European Space Agency's Environmental Satel- lite (Envisat) measured nearly continuously mid infrared limb radiance spectra. These measurements are utilised to re- trieve the global distribution of the chlorofluorocarbon CFC- 11 by applying a new fast forward model for Envisat MIPAS and an accompanying optimal estimation retrieval processor. A detailed analysis shows that the total retrieval errors of the individual CFC-11 volume mixing ratios are typically below 10% in the altitude range 10 to 25 km and that the system- atic components dominate. Contribution of a priori informa- tion to the retrieval results are less than 5 to 10% and the vertical resolution of the observations is about 3 to 4 km in the same vertical range. The data are successfully validated by comparison with several other space experiments, an air- borne in-situ instrument, measurements from ground-based networks, and independent Envisat MIPAS analyses. The re- trieval results from 425 000 Envisat MIPAS limb scans are compiled to provide a new climatological data set of CFC- 11. The climatology shows significantly lower CFC-11 abun- dances in the lower stratosphere compared with the Refer- ence Atmospheres for MIPAS (RAMstan V3.1) climatology. Depending on the atmospheric conditions the differences be- tween the climatologies are up to 30 to 110 ppt (45 to 150%) at 19 to 27 km altitude. Additionally, time series of CFC- 11 mean abundance and variability for five latitudinal bands are presented. The observed CFC-11 distributions can be explained by the residual mean circulation and large-scale eddy-transports in the upper troposphere and lower strato- sphere. The new CFC-11 data set is well suited for further scientific studies.

during 14.4 orbits per day. Ten scientific instruments aboard Envisat monitor the global state of the Earth system and environment.
The Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) (Fischer and Oelhaf, 1996;Endemann, 1999;Fischer et al., 2007) measures the mid infrared thermal emission of atmospheric constituents arising from the limb. Spectral measure-10 ments cover the range 685 to 2410 cm −1 in five radiance channels. The unapodised spectral resolution of the interferometer is about 0.03 cm −1 . The nominal measurement mode of Envisat MIPAS covers 17 tangent heights in the vertical range 6 to 68 km. The nominal vertical sampling step is 3 km from the the troposphere to the mid stratosphere and increases to 5 to 8 km above this. The instan- 15 taneous field of view covers approximately 3×30 arcmin 2 , corresponding to 3×30 km 2 in vertical and horizontal extent at the tangent point, respectively. The measurement time for a single spectrum is about 4 s. A full vertical limb scan takes 75 s. The horizontal distance between two profiles is about 500 km. During one day Envisat MIPAS measures about 1100 vertical profiles. 20 Measurements from Envisat MIPAS are processed in different steps. The aim of the ESA operational Level-0 to Level-1B data processing is the conversion of instrument raw data to calibrated radiance spectra with corresponding geolocation data. The main steps involved are quality checks of raw data, phase correction and Fouriertransformation of raw interferograms, radiometric calibration, spectral calibration, and 25 evaluation of satellite and instrument attitude data (Lachance, 1999;Nett, 1999;Kleinert et al., 2007). Within the ESA Level-2 processing vertical profiles of several atmospheric parameters are derived from the radiance measurements (Ridolfi et al., 2000;Carlotti et al., 2001;Carli et al., 2004;Raspollini et al., 2006). Target parameters of the 4564 2002 to March 2004. Due to an unexpected technical problem with the interferometer slides in March 2004, the instrument has to be operated at a reduced spectral resolution in a campaign-orientated mode since that time. In this paper we analyse measurements obtained during the continuous observation period and restrict ourselves to measurements where consolidated Level-1B and Level-2 data products are available 10 from ESA (processing software version 4.61 and 4.62). Comprehensive pre-flight and permanent in-flight calibrations were carried out to guarantee a high quality Envisat MIPAS Level-1B data product (e.g. Nett et al., 2002;Kleinert et al., 2007;Kiefer et al., 2007). Envisat MIPAS atmospheric data products have been validated by detailed tests of self-consistency as well as by comparison with measurements obtained by indepen-15 dent in-situ and other remote-sensing experiments (e.g. Piccolo and Dudhia, 2007;Ridolfi et al., 2007).

Spectral signatures of CFC-11
In the mid infrared CFC-11 radiates in broad bands around 800 to 885 cm −1 , 910 to 20 960 cm −1 , and 1045 to 1120 cm −1 (e.g. McDaniel et al., 1991;Varansi, 1991). Several schemes to identify optimal spectral analysis windows have been proposed, which all aim at minimisation of the retrieval error, or equivalently, the maximisation of the Shannon information content (e.g. Rodgers, 1998;von Clarmann and Echle, 1998;Dudhia et al., 2002). In our case, the Shannon information content under consideration Introduction

EGU
interfering species) was chosen as figure of merit to be optimised. The spectral window 844.275 to 850.575 cm −1 was identified as being best suited for the retrieval of CFC-11 from the Envisat MIPAS measurements (Fig. 1). The optimisation is carried out for mid-latitude atmospheric conditions, however tests for other climatological conditions (tropics or polar regions) lead to only minor changes in the obtained optimal spectral 5 window.
Based on forward calculations of spectra for multiple atmospheric conditions, further studies were carried out to determine the radiance contributions of minor interfering trace species. For tangent heights in the range 10 to 40 km, the maximum contributions to total mean radiance in the spectral window 844.275 to 850.575 cm −1 due to 10 CFC-11 are up to 93%. Other contributing species are HNO 3 (up to 55%), O 3 (43%), background aerosols (36%), CO 2 (22%), H 2 O (8.6%), NO 2 (5.7%), CFC-114 (3.1%), OCS (1.7%), and ClO (1.1%). Any other trace gas contributes less than 1% to total mean radiance in all reference calculations.
The CFC-11 retrieval must be accompanied by a retrieval of radiometric background 15 caused by aerosols, optically thin clouds, or other continuum processes. We identify the spectral window 830.350 to 839.475 cm −1 as being best suited for the retrieval of total particle extinction. Radiance contributions of the background aerosols are up to 45%, but increase to over 90% if clouds are present. Other important radiance emitters in this spectral window are O 3 (69%), CFC-11 (56%), CO 2 (26%), ClONO 2 Introduction the band transmittance approximation, the continuum approximation, and the emissivity growth approximation (Weinreb and Neuendorffer, 1973;Gordley and Russel, 1981;Marshall et al., 1994;Francis et al., 2006). A major reduction of CPU-time is achieved in JURASSIC, because the radiative transfer is not computed based on the general monochromatic approach, but by op-10 erating on spectral mean values of emissivity, Planck function, and radiance instead. Spectral mean emissivities are obtained by interpolation from pre-computed look-uptables. The emissivity look-up-tables are derived from exact line-by-line calculations, utilising the MIPAS Reference Forward Model (RFM) (Dudhia, 2004). As an example, Fig. 2 shows the results of a typical JURASSIC forward calculation for mid-latitude 15 atmospheric conditions.
For the spectral windows used in the CFC-11 retrieval, the difference in CPU time required on a conventional workstation by JURASSIC and the RFM is about three orders of magnitude. However, the approximations lead to a small loss of accuracy, which is quantified by comparisons with exact line-by-line calculations. For the spectral 20 windows utilised here, the deviations are well below 0.5% (Fig. 3) and are rather small compared to other uncertainties, e.g. spectroscopic data. Model errors above one percent were found only in the tropical mid troposphere, where the CFC-11 band gets optically thick.
3.3 Data pre-processing 25 Important steps in data pre-processing are apodization and spectral averaging of the Envisat MIPAS radiance spectra. For apodization the "strong function" of Norton andBeer (1976, 1977)  EGU spectral grid points are required to derive exact radiometric filter functions for the forward model. Apodization is necessary to remove side-lobe effects at the boundaries of the radiometric filter functions. Another important step in the pre-processing is cloud filtering. Spectra containing cloud emission (e.g. Fig. 1, lowermost tangent height) cannot be simulated with 5 JURASSIC because scattering is neglected in the current version of the model. The cloud filtering algorithm of Spang et al. (2004) is based on the analysis of the so-called cloud index (CI), which is defined as the ratio of mean radiances in the spectral windows 788.2 to 796.25 cm −1 and 832.3 to 834.4 cm −1 . The long wavelength region is dominated by CO 2 emission. The short wavelength region is part of an atmospheric 10 window and is strongly influenced by aerosols or cloud particle emissions. In the presence of optically thin or thick clouds the CI quickly drops below certain thresholds. In the ESA operational retrievals a threshold of CI<1.8 is used for cloud-filtering. We apply a more strict cloud filtering criterion to remove even optically thin clouds by using a threshold of CI<4.0 (e.g. Glatthor et al., 2006). 15

Optimal estimation retrieval
Retrieval is based on the standard optimal estimation approach (Rodgers, 1976(Rodgers, , 1990(Rodgers, , 2000. The "optimal estimate" of the atmospheric state (i.e. the maximum a posteriori solution of the inverse problem) is found by minimising the deviations between forward model simulations based on the current estimate of the state and the actual radiance 20 measurements, as well as minimising the deviations between the estimate and the a priori state. Deviations are normalised by the measurement error covariance and the a priori covariance, respectively. Since the retrieval problem is moderately nonlinear, a line-search approach based on the Gauss-Newton method is used to find the minimum of the cost function iteratively. A multi-target approach is applied, i.e. all 25 retrieval targets (CFC-11 volume mixing ratios and aerosol extinction coefficients) are derived simultaneously and correlations between the different quantities are fully taken into account in the retrieval covariance. EGU are used to define the retrieval grid. The inverse problem is regularised by means of explicit a priori data. A first-order autoregressive model (e.g. Rodgers, 2000) is used to initialise the a priori covariances of the different quantities and the measurement error covariance, i.e. to describe the spatial and spectral correlations of the a priori data and error data. In this model corre-5 lations decay exponentially depending on the distance of the spectral windows selected for the retrieval and the vertical distance of the tangent heights or atmospheric layers of the individual limb scan, respectively. Except for noise, most correlation lengths are not well-known from experimental side and had to be estimated. The values actually used are reported below. The vertical correlation lengths of the a priori data are very important, because they contribute to the smoothing characteristics of the constraint. We discuss this effect in more detail in Sect. 3.7. When determining the measurement error covariance we do not only consider noise but also include the non-random components (e.g. von Clarmann et al., 2001). Hence, the retrieval results and error analyses are significantly influenced by the spatial and spectral correlations of e.g. un-15 certainties in radiometric calibration or uncertainties in model parameters like pressure and temperature.
The a priori atmospheric state for the retrieval is prepared by mixing results from the ESA operational retrievals as far as available (see Sect. 2) and data from a climatology prepared for the Envisat MIPAS analysis (Remedios et al., 2007). The RAMstan 20 climatology (Reference Atmospheres for MIPAS, standard atmospheres, version 3.1) was compiled from observations of several experiments aboard the Upper Atmosphere Research Satellite as well as tropospheric and stratospheric chemical transport model output. Climatological data are provided for five different atmospheric conditions (tropics, mid-latitude day or night, polar summer, and polar winter). Mean volume mixing 25 ratios and one sigma variability are reported for more than 30 trace gases. Tropospheric profiles for the source gases are updated to year 2000 estimates where possible. CFC-11 data in the climatology are derived from measurements of the satellite experiment CLAES (Cryogenic Limb Array Etalon Spectrometer) (Roche et al., 1998

EGU
The CFC-11 tropospheric volume mixing ratio is set to 265 ppt globally. Climatological data for aerosol extinction coefficients were obtained from 5.26 micron measurements of the Halogen Occultation Experiment (HALOE) (e.g. Hervig et al., 1996). HALOE measurements during the year 2003 were globally averaged to derive an a priori profile. Spectral correction factors for the aerosol extinction are obtained from a Mie-model 5 study.

Error analysis
The error analysis for the retrieval results is based on the concept of linearisation of the transfer function (Rodgers, 1990(Rodgers, , 2000. The total retrieval error is composed of (1) direct radiance measurement errors, (2) forward model parameters errors, (3) forward model errors, and (4) smoothing effects due to a priori data. Applying the transfer function concept, detailed error budgets for the CFC-11 retrievals can be estimated.
The total retrieval error of the individual CFC-11 volume mixing ratios, including all statistical and systematic components, is typically below 10% and about 6 to 7% at best (Table 1, Fig. 4). We will now discuss the individual components of the error 15 budget in more detail.
(1) Direct radiance measurement errors: Noise as well as uncertainties in offset calibration and gain calibration are considered in the error analysis and are all found to make a significant contribution to the budget. The noise equivalent spectral radiance for unapodised spectra is about 30 nW/(cm 2 sr cm −1 ) around 830 to 850 cm −1 (Klein-20 ert et al. , 2007). To determine the noise equivalent radiance for the spectral windows selected for the retrieval, a scaling factor 1/ √ n is applied, where n is the number of grid points within the spectral window. The noise values estimated for the spectral mean radiances are assumed to be uncorrelated between the different spectral windows and different tangent heights. Offset and gain calibration uncertainty were set 25 to 2 nW/(cm 2 sr cm −1 ) and 2%, respectively (Nett et al., 2002;Kleinert et al., 2007). For the uncertainties in radiometric calibration the spectral correlation length is set to 10 cm −1 and the vertical correlation length is set to 10 km.

EGU
Other instrumental errors: Spectral instability and shifts as well as uncertainties in the instrument line shape can be neglected considering the wide spectral windows used for the retrieval. These instrumental errors do not significantly alter the spectral mean radiances. Line-of-sight pointing errors are also not considered, as Level-2 tangent heights are a derived data product, i.e. determined from the retrieved pressure 5 and temperature profiles based on the assumption of hydrostatic equilibrium. However, the CFC-11 retrieval error due to uncertainties in pressure and temperature data will be estimated separately. Since the Level-2 tangent heights also depend on the lowermost tangent altitude provided by the Envisat MIPAS attitude data system (i.e. the Level-1B tangent heights) which show rather large daily drifts (of the order of 1 to 2 km/day), we generally do not use the Level-2 tangent altitudes for further analysis and work with log-pressure altitudes instead. See von Clarmann et al. (2003a) or Kiefer et al. (2007) for a detailed discussion.
(2) Forward model parameter errors and (3) forward model errors: The leading retrieval errors are due to uncertainties in pressure and temperature (set to 3% and 15 2 K, respectively, 5 km vertical correlation length, see Raspollini et al., 2006;Piccolo and Dudhia, 2007), CFC-11 spectroscopic data (set to 3%, 10 km vertical correlation length, see Rothman et al., 2003), and spectral shape of the aerosol extinction data (set to 20%, 10 km vertical correlation length, based on a Mie-model study). Retrieval errors caused by uncertainties in minor interfering trace species as well as CFC-11 and 20 aerosol extinction top column data are small or negligible. Forward model errors are mainly due to the approximations used to accelerate the radiative transfer calculations. The retrieval errors due to the forward model errors (set to 0.5%, 3 km vertical correlation length, 10  EGU of the smoothing error will be incorrect (Rodgers, 2000). Hence, we do not include the smoothing error in the budget. The influence of a priori information on the retrieval results will be discussed in more detail in the following sections.

Contribution of a priori information
The most important diagnostic quantities for characterising the contributions of mea-5 surement information versus a priori information to the retrieval results are the averaging kernel matrices derived within the retrieval process. Integrating over the averaging kernels, i.e. the rows of the averaging kernel matrices, gives an approximate estimate of the amount of measurement information in the retrieval result. Subtracting the area from one gives the amount of a priori information. It is important to realize that this approximation is only valid for small retrieval errors. For the CFC-11 retrieval, the contribution of a priori information is found to be less than 5 to 10% when CFC-11 is present in significant concentrations (Fig. 5a). Even if the retrieval result at an atmospheric height level is mainly determined by measurement information, this information must not necessarily originate from the observations at the corresponding tangent height 15 alone, but may also come from adjacent tangent altitudes. This would cause a smoothing effect of the a priori on the retrieved profile. Hence, the vertical resolution of the retrieval needs to be analysed in more detail.

Vertical resolution
An ideal retrieval would exactly reproduce the unknown true atmospheric state. In 20 reality, the retrieval produces a smoothed version of the true trace gas profiles. The averaging kernels take the form of broad maxima rather than delta distributions. Hence, the averaging kernels also describe the resolution of the observations. We applied the concept of information density of Purser and Huang (1993)  EGU retrieval grid level. Hence, the reciprocal values of the diagonal elements measure the number of retrieval grid levels per degree of freedom, or the "resolution" of the observations. The reciprocal values have to be multiplied by the retrieval grid spacing (in [km]), to get the correct unit. The vertical resolution of the retrieved CFC-11 profiles is about 3 to 4 km and corresponds to the sampling step of the limb scans and the field 5 of view of the Envisat MIPAS instrument (Fig. 5b).

Internal quality measures
For the purpose of internal data validation the consistency of the retrieval results with the radiance measurements and the a priori data is checked by applying a standard χ 2 -test to the final value of the cost function for each limb scan (e.g. Rodgers, 2000). Once the retrieval converged to a solution, the cost function measures the differences between the actual radiance measurements and the forward model simulations based on the optimal estimate of the state, as well as the differences between the optimal estimate and the a priori state. The differences are normalised by the measurement 15 error covariance and the a priori covariance, respectively. In addition, for the χ 2 -test the cost function values are also normalised by the number m of radiance measurements of the limb scan, since m is the number of degrees of freedom. A normalised χ 2 /m around one indicates that the retrieval results are consistent, at least in a statistical sense or for an ensemble of profiles. The χ 2 -test allows us 20 to identify individual limb scans with abnormally poor fits which are probably not part of the ensemble. In this case the retrieval results and corresponding forward model simulations are either inconsistent with the radiance measurements (with respect to the measurement errors) or inconsistent with the a priori data (with respect to the a priori uncertainty). Since the latter may happen in case of an unusual atmospheric event, the corresponding retrievals are not necessarily "bad" and need to be analysed carefully. However, to prepare the CFC-11 data for subsequent scientific studies, we Introduction It is also interesting to study the χ 2 /m-distribution for a large ensemble of limb scans.
A plot of the final χ 2 /m-distribution covering all 425 000 limb scans of this study is 5 shown in Fig. 6. For comparison the normalised χ 2 /m-distribution for the initial guess, i.e. the a priori state in our setup, is shown, too. The final χ 2 /m-distribution clearly indicates that the majority of retrieval results is consistent. On average only 2.6 iterations are required until the individual retrievals converge. Subtracting one iteration required to detect convergence, this indicates that the particular retrieval problem indeed is a 10 moderately non-linear one.

Comparison with other MIPAS retrievals
The CFC-11 data retrieved with the scheme described in this paper were compared with independent Envisat MIPAS retrievals carried out at the University of Leicester, United Kingdom and the Institut für Meteorologie und Klimaforschung (IMK) at the re- proach. Similar to the approach presented in this paper, spectral mean radiances rather than detailed spectra are analysed in the Leicester OPERA scheme (Optimal Estimation Retrieval Algorithm) (Moore et al., 2006;Moore and Remedios, 2007 EGU fast forward model versus full line-by-line radiative transfer, as well as providing a first verification of algorithm correctness. There are only minor differences in the retrieval setups, concerning the choice of spectral windows (842.650 to 845.475 cm −1 for CFC-11 in the OPERA scheme) or the initialisation of measurement error covariance and a priori covariance (diagonal matrices in the OPERA scheme, first-order autoregressive 5 model in this study). The OPERA scheme produces retrieval results for CI<1.8, however since the comparison here is for individual matched profiles, in effect the comparison is performed for the CI<4.0 threshold. A direct comparison of individual retrieved scans reveals good agreement (i.e. better than 2 to 3%) between our result and the OPERA results as expected from the common retrieval approaches.  (Stiller, 2000;Stiller et al., 2002). Since this approach is rather time consuming, KOPRA has mostly been used for the retrieval of short rather than extended time periods (currently, CFC-11 data are retrieved from about 15% of all Envisat MIPAS scans). The KOPRA forward model is well validated by cross-comparisons with the RFM as well as several other radiative transfer models (von Clarmann et al., 2003b). A brief description of the IMK CFC-11 retrieval is given by von Clarmann et al.
. 20 Figure 7a shows a global comparison between our CFC-11 retrieval results and the IMK data (version V3O F-11 8) for several days. The comparison reveals small systematic differences of about −6 to 6 ppt (−1.8 to 3.5% below 25 km). Statistical deviations are in the range of 8 to 24 ppt (7 to 10%). One reason for differences in CFC-11 data are the different temperature data sets used. A comparison of ESA operationally re-Introduction  Fig. 4 such temperature differences map correspondingly onto the retrieval results. A direct comparison indicates good to perfect agreement for many CFC-11 profiles. However, some profiles show large differences. Next to temperature data, possible reasons to explain these differences are uncertainties in pressure data or tangent altitudes, different a priori datasets of interfering species (ESA operational data versus 5 IMK data), as well as different approaches for the retrieval of radiometric background as caused by aerosols and for the regularisation of the retrieval problem. These aspects have to be addressed in future work, but are out of the scope of this paper. However, the mean differences with the IMK data of 3.5% or less (systematic component) and 10% or less (statistical component) are already encouraging.  (Offermann et al., 1999;Riese et al., 1999b;Grossmann et al., 2002). Trends are corrected for in this comparison by mapping all profiles to the middle of the year 2003, i.e. the mid point of the MIPAS measurement period. Profiles are mapped considering global tropospheric 20 trend data derived from ground-based network measurements (Blake, 2005) and age of air data (Waugh and Hall, 2002). Generally, good agreement is found between the results of the different experiments. The shape of the mean profiles is nearly identical in all cases. Small offsets may be explained by systematic measurement errors, data sampling issues, or remaining uncertainties in trend correction. 25 The individual CFC-11 measurements of ATMOS have a high accuracy and precision, i.e. uncertainties are below 5% (Gunson et al., 1996). However, we see more variability in the mean profiles. This is due to the limited number of measurements and  , 2002). The CRISTA-2 profile (measurements in July 1997) corresponds better with the remaining measurements and shows a nearly identical stratospheric gradient to that seen in the MIPAS data. The climatological data from Remedios et al. (2007) shown in Fig. 8 deviate significantly from all other measurements (i.e. about 2 to 3 climatological standard deviations). The climatological data are based on measurements from the satellite experiment CLAES (Cryogenic Limb Array Etalon Spectrometer) (Roche et al., 1998). These measurements, obtained during the years 1991 to 1993 are likely to be even more influenced by the Pinatubo aerosols than the CRISTA-1 measurements. A similar Pinatubo effect for CFC-12 has been shown by Remedios et al. (2007). However, the direct comparison shows that the Envisat MIPAS retrieval 15 results are not or only slightly influenced by the deficits of the climatology, which is used as a priori in our retrieval setup.

Comparison with air-borne in-situ measurements
The retrieved MIPAS CFC-11 data were compared with measurements from the High Altitude Gas Analyzer (HAGAR) (Riediger et al., 2000) aboard the high-altitude re-20 search aircraft M-55 Geophysica. HAGAR measures trace gas concentrations with a gas chromatograph. The HAGAR measurements shown in Fig. 9  EGU distance are minimised. The direct comparison of the HAGAR and MIPAS measurements shows good agreement of the CFC-11 data. Differences are below 5 to 10% for the two flights.

Comparison with ground-based measurements
The tropospheric concentrations of CFC-11 are monitored by several networks of 5 ground-based measurement stations. During the year 2003 the global mean tropospheric volume mixing ratio of CFC-11 was in between 254.9 to 257.3 ppt and the growth rate was in between −1.0 to −0.7 %/year (WMO, 2007). The mid-latitude tropospheric volume mixing ratio derived from the Envisat MIPAS measurements is about 15 to 17 ppt (6%) lower than the global mean from the ground based measurements.
However, this difference remains within the estimated systematic errors of the Envisat MIPAS measurements (compare Fig. 4). The tropospheric mean value for tropical latitudes is in very good agreement with the ground-based measurements, i. e. the difference is only about 6 to 8 ppt (3%). For the polar regions comparisons are not possible since the polar troposphere is not regularly covered by our analysis.

Latitudinal distribution
We compiled the CFC-11 data retrieved from the Envisat MIPAS measurements during the period July 2002 to March 2004 to provide an update for the climatology of Remedios et al. (2007). Hence, the individual measurements were sorted into four groups. 20 The "tropical latitudes" include all measurements in the latitude range 20 • S to 20 • N. The "mid-latitudes" cover all measurements in the latitude ranges 20 • N to 65 • N and 65 • S to 20 • S. The "polar summer" and "polar winter" groups cover all measurements north of 65 • N or south of 65 • S during the months January to March and October to December or April to September, respectively. For each case we report the CFC-11 25 mean volume mixing ratio and the one-sigma variability on a 1 km log-pressure altitude 4578 Introduction  Fig. 10). We only report the data if statistics are based on at least 25 000 individual measurements. Individual CFC-11 volume mixing ratios are not taken into account in this analysis if they contain more than 10% a priori information. The contribution of a priori information to the retrieval results are derived from the averaging kernel data (see Sect. 3.6). The 5 filtering removes all observations where the volume mixing ratios drop below ≈5 ppt. In this case the radiance signal is low compared to the measurement errors and does not provide sufficient information for the retrieval. Filtering also removes all altitudes where the MIPAS line-of-sight was obscurred by clouds and the retrieval result corresponds to the a priori data. Selecting a strict threshold of 10% for the maximum amount of a 10 priori information ensures that the bias in the averaged profiles due to neglected trends in the a priori data is less than 0.5 to 1% (Hoffmann et al., 2005).

Seasonal variability
The time series of CFC-11 mean volume mixing ratio and corresponding variability shown in Fig. 11 illustrate in more detail the atmospheric variability captured in the 21 15 months of nearly continuous Envisat MIPAS measurements. CFC-11 is upwelling from the troposphere into the stratosphere mainly at tropical latitudes. Accordingly, the time series for tropical latitudes shows the highest average concentrations. In the stratosphere tropical air masses are driven to higher latitudes by the residual mean circulation and large-scale eddy transport. However, CFC-11 is destroyed by photolytical de-20 composition and concentrations decrease with increasing altitudes. At polar latitudes the CFC-11 abundances decrease especially in the winter months due to subsidence of air masses in the polar vortex. This effect is more pronounced in the southern polar winter hemisphere due to the more stable polar vortex. The rather rapid increase in mean CFC-11 abundance at south polar latitudes in late spring (November/December Introduction EGU maximum monthly variability of 40 to 50 ppt was seen around 25 km altitude. The variability increases towards mid and polar latitudes due to increased mixing of CFC-11 rich tropical air masses and CFC-11 poor polar air masses by planetary wave activity. The CFC-11 variability reaches up to 60 to 70 ppt around 15 to 20 km altitude. The highest variabilities are found in the winter months, i.e. during November to April in the 5 Northern Hemisphere and August to December in the Southern Hemisphere. A hemispheric difference in variability can be observed at polar latitudes. A maximum CFC-11 variability up to 90 to 100 ppt is found at south polar latitudes, which is about twice the variability of the north polar latitudes. This enhanced variability is most likely due to the more stable vortex in the Southern Hemisphere and much stronger gradients of CFC-10 11 across the vortex boundary. The enhanced variability in the Southern Hemisphere is mainly due to the activity of the planetary wave-1, i.e. the vortex is shifted away from the pole and crosses the 65 • S latitude boundary used to define the south polar region. Reports on CFC-11 variance are rather rare and comparisons are often difficult. However, in a comparison of Envisat MIPAS CFC-11 variability for August 2003 and 15 measurements of the CRISTA-2 satellite experiment in August 1997 we found excellent agreement (Kuell et al., 2004;Hoffmann, 2006). This gives confidence that both satellite experiments are capable of capturing the atmospheric variability on synoptic and planetary scales. However, the Envisat MIPAS time series presented here provide additional information about the seasonal dependence over a period of nearly two 20 years.

Conclusions
Envisat MIPAS radiance measurements are well-suited to retrieving the global distribution of CFC-11 in the upper troposphere and lower stratosphere. Utilizing a newly developed rapid radiative transfer model and an accompanying optimal estimation re- EGU volume mixing ratios is typically below 10% in our retrieval scheme. The retrieval error is dominated by noise and offset calibration errors at upper altitudes and uncertainties of temperature data and gain calibration data at lower altitudes. Retrieval errors due to the use of the rapid approximative forward model are rather small. Retrieval results typically contain less than 5 to 10% a priori information. The vertical resolution of the 5 observations is about 3 to 4 km. Several validation activities, i.e. internal quality measures, comparison with other satellite measurements, an air-borne in-situ instrument, and ground-based measurements, suggest that the retrieved CFC-11 data are reliable. We compiled all retrieved data to provide a new climatology of CFC-11 for four climatological cases (tropical latitudes, mid-latitudes, polar summer and po-10 lar winter). The new climatology shows significantly lower CFC-11 abundances in the lower stratosphere compared with data provided in the RAMstan V3.1 climatology.
The observed differences are up to 45 to 150% at 19 to 27 km altitude.
The new climatological data set will be provided as an electronic supplement to this paper (http://www.atmos-chem-phys-discuss.net/8/4561/2008/ acpd-8-4561-2008-supplement.zip). Time series of CFC-11 mean abundance and variability for five latitude bands were derived from the measurements and can be explained by the residual mean circulation and large-scale eddy-transports in the upper troposphere and lower stratosphere. CFC-11 variability is strongly influenced by the planetary wave activity near the boundary of the polar vortex. The new CFC-11 data 20 set has a great potential for model validation and case studies of transport and mixing processes in the upper troposphere and lower stratosphere on synoptic or planetary scales and up to the inter-seasonal time scale. It is well suited for further scientific studies.