development and spatial distribution at 15–45 km altitude

Abstract. We present a climatology of monthly and 10° zonal mean profiles of sulfur dioxide (SO2) volume mixing ratios (vmr) derived from MIPAS/Envisat measurements in the altitude range 15–45 km from July 2002 until April 2012. The vertical resolution varies from 3.5–4 km in the lower stratosphere up to 6–10 km at the upper end of the profiles, with estimated total errors of 5–20 pptv for single profiles of SO2. Comparisons with the few available observations of SO2 up to high altitudes from ATMOS for a volcanically perturbed situation from ACE-FTS and, at the lowest altitudes, with stratospheric in situ observations reveal general consistency of the datasets. The observations are the first empirical confirmation of features of the stratospheric SO2 distribution, which have only been shown by models up to now: (1) the local maximum of SO2 at around 25–30 km altitude, which is explained by the conversion of carbonyl sulfide (COS) as the precursor of the Junge layer; and (2) the downwelling of SO2-rich air to altitudes of 25–30 km at high latitudes during winter and its subsequent depletion on availability of sunlight. This has been proposed as the reason for the sudden appearance of enhanced concentrations of condensation nuclei during Arctic and Antarctic spring. Further, the strong increase of SO2 to values of 80–100 pptv in the upper stratosphere through photolysis of H2SO4 has been confirmed. Lower stratospheric variability of SO2 could mainly be explained by volcanic activity, and no hints of a strong anthropogenic influence have been found. Regression analysis revealed a QBO (quasi-biennial oscillation) signal of the SO2 time series in the tropics at about 30–35 km, an SAO (semi-annual oscillation) signal at tropical and subtropical latitudes above 32 km and annual periodics predominantly at high latitudes. Further, the analysis indicates a correlation with the solar cycle in the tropics and southern subtropics above 30 km. Significant negative linear trends are found in the tropical lower stratosphere, probably due to reduced tropical volcanic activity and at southern mid-latitudes above 35 km. A positive trend is visible in the lower and middle stratosphere at polar to subtropical southern latitudes.

This discussion paper is/has been under review for the journal Atmospheric Chemistry and Physics (ACP). Please refer to the corresponding final paper in ACP if available.
with estimated total errors of 5-20 pptv for single profiles of SO 2 . Comparisons with few available observations of SO 2 up to high altitudes from ATMOS, for a volcanically perturbed situations from ACE-FTS and, at the lowest altitudes, with stratospheric insitu observations reveal general consistency of the datasets. The observations are the first empirical confirmation of features of the stratospheric SO 2 distribution which have only been shown by models up to now: (1) the local maximum of SO 2 at around 25-30 km altitude which is explained by the conversion of carbonyl sulfide (COS) as the precursor of the Junge layer, and (2) the downwelling of SO 2 rich air to altitudes of 25-30 km at high latitudes during winter and its subsequent depletion on availability of sunlight. This has been proposed as the reason for the sudden appearance of enhanced 15 concentrations of condensation nuclei during Arctic and Antarctic spring. Further, the strong increase of SO 2 to values of 80-100 pptv in the upper stratosphere through photolysis of H 2 SO 4 has been confirmed. Lower stratospheric variability of SO 2 could mainly be explained by volcanic activity and no hint for a strong anthropogenic influence has been found. Regression analysis revealed a QBO (quasi-biennial oscillation)

Introduction
Sulfur dioxide (SO 2 ) is one of the key species determining the aerosol content of the stratosphere (SPARC, 2006). Main sources of stratospheric SO 2 are the conversion of carbonyl sulfide (COS) (Crutzen, 1976;Brühl et al., 2012) and the direct transport of SO 2 across the tropopause. This transport can occur abruptly through major volcanic 5 eruptions or through upwelling predominantly at the tropical tropopause or through the monsoon circulation (Bourassa et al., 2012). However, there is still a lot of uncertainty regarding the major sources of stratospheric SO 2 during volcanically quiescent periods (Deshler et al., 2006;Hofmann et al., 2009;Vernier et al., 2011;Solomon et al., 2011;Brühl et al., 2012). 10 Loss of SO 2 occurs mainly through oxidation by OH radicals into H 2 SO 4 vapour which condenses into liquid sulfate aerosols. Subsequently, sedimentation of particles results in a loss of sulfur from the stratosphere. Large-scale circulation leads to further upwelling of sulfur species in the tropical stratosphere where probably the photolysis of H 2 SO 4 constitutes a further source of SO 2 (Rinsland et al., 1995;Vaida et al., 2003;15 Mills et al., 2005). During winter, downwelling of air in the polar vortices transports elevated amounts of SO 2 from the upper stratosphere down to altitudes of 30 km and below. Here, on availability of sunlight in spring, SO 2 is transformed through reaction with OH into H 2 SO 4 , thus leading to the formation of new particles. These processes have been proposed by Zhao et al. (1995) and Mills et al. (1999Mills et al. ( , 2005 as explanation 20 for the sudden formation of enhanced layers of condensation nuclei (CN) in polar spring (Rosen and Hofmann, 1983;Hofmann and Rosen, 1985;Hofmann, 1990;Wilson et al., 1990). However, there has been no direct experimental evidence for this mechanism until now.
There exists a large dataset of SO 2 observations from space-borne nadir sounding 25 instruments (e.g. Theys et al., 2012, and references therein). While these measurements have a good horizontal resolution, height-resolved profiles cannot be obtained (albeit some information on plume height has recently been derived, Van Gent et al.,

12391
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | including a regression analysis to single out correlations with external parameters, periodic cycles and trends.

Instrument
MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) is one of the instruments on ESA's polar orbiting Envisat satellite (Fischer et al., 2008). Envisat was was operated alternately in different limb scan patterns. For this work we have used the most frequent "nominal" viewing modes. During P1 these consisted of 17 tangent altitudes (6-68 km with 3 km distances up to 42 km, followed by steps of 2 × 5 km and 2 × 8 km) which do not depend on the geographical position. During P2 the number of 15 tangent views per limb-scan was increased to 27. Here tangent altitudes depend on latitude ranging from 5-70 km at the poles to 12-77 km over the equator with steps increasing with height from 1.5 km to 4.5 km. The horizontal distance between two subsequent limb scans in nominal measurement mode was around 550 km during P1 and 420 km during P2 resulting in about 1000 and 1400 limb-scans providing full latitudinal 20 coverage per day.

Retrieval
Retrieval of SO 2 altitude profiles up to the middle stratosphere from MIPAS observations is difficult due to its small spectral signal relative to the spectral noise of MIPAS, especially under volcanically non-perturbed conditions. Therefore, we have chosen to Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | analyse monthly mean limb-spectra binned within 10 • wide latitude bands. This method has already been employed for the detection of stratospheric BrONO 2 in Höpfner et al. (2009a). In detail, first cloud-clearing has been performed on basis of the cloud-index (CI) method (Spang et al., 2004). This is necessary to avoid retrieval errors due to 5 cloud/aerosol signal in the spectra. The CI is the spectral radiance contrast between two wavenumbers. A smaller CI indicates larger probability that the spectrum is contaminated by aerosols or clouds. We use a CI threshold of 4.5, i.e. spectra with CI < 4.5 are discarded. Further, all spectra in this limb sequence are discarded which were measured at lower tangent altitudes than that with CI < 4.5. Beside thick tro-10 pospheric clouds, this approach excludes e.g. also cirrus clouds, polar stratospheric clouds (Höpfner et al., 2009b) and optically thicker aerosol plumes from the subsequent analysis. Depending on atmospheric situation, particle composition and altitude, the applied CI limit equals a particle volume density of about 1-2 µm 3 cm −3 .
In the second step the limb-scans which are averaged within each latitude-and time- 15 bin are selected. Here we have chosen those locations with the lowest available tangent altitude after cloud clearing under the condition that at least 300 limb-scans are averaged, which reduced the noise by a factor of at least 17. The noise reduction could be further improved by increasing the requested number of samples; this, however, would lead to loss of measurement points particularly at low altitudes and during months with 20 sparser observational coverage. A decrease of the requested number of samples, on the other hand, would result in larger noise error contributions. For this analysis, MIPAS version 5 level-1b data have been used. Together with the averaged MIPAS spectra, for the retrieval mean pressure/temperature profiles are calculated from ECMWF analysis data at the position of each single limb scan. Further, mean tangent altitudes are ob-25 tained from the engineering MIPAS tangent altitude values. Errors resulting from these averaging processes are included in the error estimation as described further below.

12394
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | The retrieval scheme is a constrained global fit approach using all tangent altitudes of one limb-scan with averaged spectra simultaneously (e.g., von Clarmann et al., 2003a): which is a variant of the formulation suggested by Rodgers (2000). Here, x is the vector containing the atmospheric and instrumental parameters to be determined. The atmo-5 spheric profiles are gridded at 1 km altitude levels in-between which the volume mixing ratios (vmr) vary linearly with height. y meas are the averaged measured radiances of all tangent altitudes and S y is the measurement noise covariance matrix. y i contains the spectral radiances which are simulated with the radiative transfer model KOPRA (Stiller, 2000) using the results obtained in iteration number i . K i is the Jacobian ma-10 trix, i.e. the partial derivatives ∂y(x i )/∂x i determined in parallel to y i at each iteration step. x a contains the a-priori profiles. For the target species SO 2 , altitude-constant initial guess and a-priori profiles, x 0 and x a equal 10 pptv, have been used whereas for all other trace gases climatological values have been chosen. For the regularisation matrix R a first-order smoothing constraint R = γL T L, where L 15 is a first order finite differences operator (Tikhonov, 1963;Steck, 2002) has been used in order not to bias the retrieval results. Thus, the choice of the altitude-constant a-priori value for SO 2 has no influence on the resulting profiles. The regularization parameter γ depends only on the species but not on the altitude. It was chosen such to avoid instabilities showing up as oscillations in the retrieved atmospheric profiles.
Line-by-line calculations have been based on the HITRAN 2008 compilation including updates until 2010 (Rothman et al., 2009). The spectral intervals (microwindows) used for the retrieval are located in the same wavenumber region (∼ 1366-1377 cm −1 ) within the SO 2 ν 3 -band as those selected for the analysis of ACE-FTS (Doeringer et al., 2012) and of ATMOS data (Rinsland et al., 1995). mental parameters which have been included in the parameter vector x are a spectral shift correction per microwindow and an additive calibration offset per microwindow and tangent altitude. No regularisation has been applied in case of these instrumental quantities. Figure 1 presents an example of the spectral fit achieved in all four microwindows for 10 MIPAS measurements with higher (top 3 rows) and lower spectral resolution (bottom 3 rows). Spectral residuals between measurements and simulation are shown for the case where no SO 2 has been retrieved (2nd and 5th row) and under consideration of SO 2 (3rd and 6th row). It is obvious that in both cases, the fit residual is strongly improved by including SO 2 into the fit parameter vector, and that the residuals without 15 SO 2 fit (black lines in 2nd and 5th row) are mostly produced by SO 2 spectral signatures (green lines in 2nd and 5th row). For characterisation of the vertical resolution of the retrieval the related averaging kernels have been determined as (2) 20 The rows of A represent the contributions of different altitudes to the retrieved vmr profile of SO 2 whereas the columns are the responses of the system to a delta function at the associated altitude. Typical averaging kernels are shown in Fig. 2 for period P1 (top) and P2 (bottom). The overall diagonal structures of A demonstrate that the retrieval of SO 2 behaves well with better vertical resolution at lower compared to higher 25 altitudes due to the better sensitivity (higher pressure) and a denser tangent altitude sampling there. Typical values for the vertical resolution as derived from the width of 12396 the columns of A and from the inverse of the diagonal of A in both measurement modes vary around 3.5-4 km at 20 km, 4-5 km at 30 km and 6-10 km at 40 km altitude. An estimate of the altitude dependent errors is presented in Fig. 3. Typical mean error profiles calculated for one month (January 2003) of period P1 and one month (January 2011) of P2 are shown together with the single error contributions which are 5 combined to the total error by calculating the root sum squared of the single terms.
The following sources of uncertainty have been taken into account: spectroscopic errors of SO 2 line-intensity (spe_int_SO2) and air-broadened half-width (spe_hw_SO2). For the line-intensity an uncertainty of 5 % and for the half-width 15 % has been assumed. Both values are on the conservative side of the errors stated in Rothman et al. 10 (2009). The spectroscopy of interfering trace gas signatures has been handled as an independent error source by assigning also here errors of line-intensity (spe_int_itf) and half-width (spe_hw_itf) of 5 % and 15 %, respectively. Major instrumental uncertainties are estimated as 1 % for gain calibration (gain), 3 % for the instrument line shape (ils) (in terms of linear loss of modulation efficiency toward the maximum optical 15 path difference of the interferometer) and 300 m for tangent altitude pointing (htang). Spectral noise (noise) is based on the actual level 1b dataset as delivered together with the calibrated spectra. For the uncertainty of ECMWF temperatures (Tecm) values of 2 K below and 5 K above 35 km altitude are assumed. A further error term which has to be taken into account for retrievals from averaged spectra is the effect of retrieval 20 non-linearity (nlin). This error contribution is estimated on basis of dedicated retrieval simulations as described in detail in Höpfner et al. (2009a) except that the values used for the line-of-sight uncertainty of single observations have been 400 m for the P1-and 300 m for the datasets during P2. These values have been determined from comparison of engineering tangent altitude values with those obtained by the pointing retrievals

Dataset overview
Figures 4-6 present an overview of all MIPAS results of SO 2 as color-coded crosssections versus time. As mentioned above, the lower altitude limit of the dataset is defined by the condition on the minimum number of limb-scans used to calculate mean spectra. This is determined by (a) the cloud coverage, (b) the scan-pattern of MIPAS 5 and (c) the lower limit of 15 km set to confine the retrievals mainly on the stratospheric situation. The influence of clouds can best be seen at the data gaps during the Antarctic winter season where the MIPAS observations are obscured by Polar Stratospheric Clouds (e.g. top row in Fig. 4). High clouds also restrict some of the observations at low altitudes in tropical regions during P1 until March 2004. Further data-gaps covering the As basis for subsequent discussions the dataset has been reduced by averaging to the mean profiles for four seasons and 8 latitude regions (see Fig. 7, top two rows). These profiles represent background situations since volcanically strongly perturbed 15 periods, as detailed in the caption of Fig. 7 have been neglected during averaging. The related variability is described in the bottom two rows of Fig. 7 by the altitude-dependent standard deviation.

Data quality
Here we assess the quality of the MIPAS SO 2 dataset by comparison with independent 20 measurements. Further the estimation of random errors will be validated by comparison to the variability derived from the dataset itself.

Comparison with ATMOS
The only measurements of SO 2 reaching up to the upper stratosphere so far were published by Rinsland et al. (1995). These three profiles are compared with MIPAS data Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | obtained for similar latitudes and months in Fig. 8. The ATMOS profile of 1992 and, to a lesser extent, also that of 1993, are influenced by the eruption of Mt. Pinatubo in June 1991 (Rinsland et al., 1995). Thus, the one which can best be compared to the MIPAS dataset is the ATMOS mean profile of 1985. Mind that this has also been observed under enhanced stratospheric aerosol levels due to the eruption of El Chichón 5 in March 1982, albeit to a less extend than ATMOS 1991and 1993(SPARC, 2006.
Throughout the altitude range between 33 and 45 km, the MIPAS mean profile is within the uncertainty range of ATMOS 1985. Only at the lowest 2 km MIPAS mean volume mixing ratios of SO 2 are larger than ATMOS by about 20 pptv. However, accounting for the variability, the MIPAS profiles are comparable to those of ATMOS at these altitudes. MIPAS agrees with both post-Pinatubo ATMOS measurements up to altitudes of 32 and 35 km, respectively. Above these heights ATMOS shows strongly increasing concentrations up to values of 400 pptv compared to maximum MIPAS values of 100 pptv. The fact that at lower altitudes around 30 km the values of SO 2 do not 15 deviate from MIPAS as strongly as above 35 km even for volcanically enhanced conditions can most probably be explained with the major part of sulfur being in the form of H 2 SO 4 at those altitudes.

Comparison with ACE-FTS
A direct comparison with observations by ACE-FTS is possible for a volcanically en-

Comparison with in-situ observations
In-situ measurements of SO 2 at comparable altitudes to those presented here for the mean MIPAS profiles are extremely sparse since they have mainly been obtained from aircraft up to the lowermost stratosphere. Table 1 shows a selection of published in-situ datasets in which explicitly stratospheric values have been indicated which can best be 10 compared to the mean profiles as shown in Fig. 7. The only in-situ dataset reaching into the altitude range presented here is the one obtained by Inn and Vedder (1981) up to an altitude of 20.4 km in June/July 1979. The reported values of 36-50 pptv at 15 km are higher (or at the upper end when taking into account the reported error of 50 % of the in-situ data) compared to the MIPAS average background vmr-distribution (around 13±10 pptv) at mid-latitudes (cf. Fig. 7). At around 20 km reported SO 2 mixing ratios of 45 and 51 pptv (±50 %) at 64 • N and 67 • N are clearly much higher than MIPAS mean values which are around 6 pptv. Even under perturbed situations like the eruption of Sarychev, MIPAS monthly mean data at around 20 km altitude and similar latitudes are less than 20 pptv. Thus, either the aircraft data has been obtained within a locally 20 perturbed situation not representative for the mean background or the difference points to problems of either the in-situ data or the MIPAS analysis. Other observations reaching nearly the lower limit of the altitude range of our MIPAS dataset are the first measurements of SO 2 by Jaeschke et al. (1976) at 14 km and various stratospheric data by Meixner (1984), both obtained in northern mid-latitudes.

25
Comparing the data in Fig. 7 at the lower end of the profiles with those observations shows that MIPAS mean values are generally at the lower part of the variability of the in-situ data. However, when considering the generally increasing MIPAS data at 12400 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | northern mid-latitudes with decreasing altitude is seems reasonable that the typically higher in-situ data at lower heights might be compatible. Since the other stratospheric data listed in Table 1 are similar to those already discussed, we conclude that in general the MIPAS values at the bottom end of their altitude range are at the lower limit of in-situ observations. 5

Internal variability
As described in Sect. 3 the instrument noise dominates the error characteristics over a large part of the profile. We have tried to validate this error term by comparing it to the temporal month-to-month variability of the retrieved SO 2 profiles at distinct altitudes for the different latitude bins. This method is similar to the one used for validation of the precision of MIPAS single-scan data products (Piccolo and Dudhia, 2007). Figure 10 shows for each latitude l and altitude h the standard deviation ∆δ l ,h of the noise-error ( l ,h,m ) weighted difference δ l ,h,m of the SO 2 volume mixing ratios x l ,h,m between subsequent months m where M is the number of months: where is the estimated error due to instrument spectral noise. In the ideal case ∆δ l ,h would be equal 1. Values greater than 1 indicate either that the natural variability of SO 2 from month to month is not negligible (thus increasing the numerator of Eq. 4) 20 or the noise errors are underestimated (denominator of Eq. 4). In Fig. 10  in high-latitude regions are very likely due to the natural variability caused by volcanic activity and the downwelling of SO 2 -rich air into the polar vortices during winter. Anthropogenic influence might be a further source for the enhancements at northern latitudes. Excluding those regions, i.e. for 50 • S-50 • N and altitudes ≥ 30 km the mean value of ∆δ l ,h = 1.5 ± 0.1. Thus, it can be concluded that the estimated noise error may 5 underestimate the instrumental random error by no more than 50 %. The difference between estimated noise and month-to-month variability may be explained by either natural variability or by random parts of the total error estimation like variations in the gain-calibration or the ECMWF temperature/pressure dataset which are not included in the pure spectral noise error.
6 Results and discussion

Mid-stratospheric maximum
The main chemical production of SO 2 in the stratosphere appears via conversion from COS leading to a local maximum of SO 2 in the mid-stratosphere (Crutzen, 1976;Brühl et al., 2012 Fig. 6 of Brühl et al. (2012). Here the altitude of the maximum varies around 28-31 km, depending on time between 1999 and 2002. In this time range the vmr values vary around 30-60 pptv which is similar to the temporal variability of our observations as shown in Fig. 4 and 5.

5
The local maximum in MIPAS mean profiles of SO 2 (Fig. 7) can further be compared with the results of five different models as shown in SPARC (2006, Fig. 6.11) (http: //www.sparc-climate.org/publications/sparc-reports/sparc-report-no4/) and Fig. 6.2 of its supplement to Chapter 6. In these simulations the altitude of the tropical maximum is located between 29 and 32 km. The maximum vmr values of 40-50 pptv for four models

Enhanced upper stratospheric values and polar downward transport
In the tropics, the production of SO 2 from COS fades out shortly above 30 km due to the unavailability of COS leading to a decrease of the SO 2 mixing ratio (Brühl et al., 2012). Above about 33 km our measurements show again an increase of SO 2 volume mixing ratios reaching average values of about 80-100 pptv (Fig. 7) at 45 km. At these 20 altitudes, the SO 2 concentrations do not show any large variability with latitude. This increase of SO 2 from the middle to the upper stratosphere is caused by photolysis of H 2 SO 4 which is released due to the evaporation of aerosols (Vaida et al., 2003;Mills et al., 2005). At altitudes of 45 km Brühl et al. (2012, Fig. 6) show values of around 50 pptv of SO x (mostly SO 2 ) being considerably smaller than our data. This can be 25 explained by neglecting in the model SO 2 photolysis bands in the visible and near-IR (Brühl et al., 2013). The models in SPARC (2006, Fig. 6.11) show large differences at these altitudes: in the tropics three of the models underestimate the observed SO 2 12403 Introduction values above the mid-stratospheric maximum, while the LASP model compares rather well. However, at 45 • N this model overestimates the high-altitude MIPAS observations by a factor of 2 while other models are still lower than the observations. In wintertime at high latitudes the MIPAS data show the descent of enhanced upper stratospheric SO 2 concentrations down to 20-25 km altitude (blue curves in Fig. 7). 5 On average at 30 km vmr values of 50-60 pptv and at 25 km 20-40 pptv are reached. This periodic effect is clearly visible in the top rows of Figs. 4 and 5. The mean evolution over the Antarctic from July to December is shown in detail in Fig. 11 which can be compared with the simulations shown in Mills et al. (2005, Fig . 5). In those model runs highest SO 2 values of about 100 pptv at 30 km and 60 pptv at 25 km altitude are 10 reached in the Antarctic polar vortex by end of August. In the observations, highest SO 2 concentrations at these altitudes are observed in July/August with mean values of up to 60 pptv at 30 km and 40 pptv at 25 km altitude (see Fig. 11). In September/October the model shows a sudden decrease of SO 2 concentrations through photolysis leading to the production of small aerosol particles. This behaviour is well reflected in the MIPAS 15 dataset (cf. Fig. 11) where a first decrease is visible below 30 km altitude between August and September followed by a stronger decrease between September and October below about 35 km.

Lower stratospheric variability
In Fig. 6 we have indicated the time and geographical latitude of major volcanic erup-20 tions by orange triangles together with the initials of each volcano as specified in Table 2. As basis for this list we have taken the one compiled by Neely et al. (2013) and added further eruptions being visible in the MIPAS dataset. For that purpose, the retrieval of SO 2 from single MIPAS observations (which is the topic of a separate study to be published) has been used to unambiguously assign elevated amounts of SO 2 to 25 volcanic events. The clear correlation of volcanic eruptions with elevated values of SO 2 in the top four rows of Fig. 6 shows that the major variability of SO 2 in the altitude range below about 22 km is determined by volcanic activity. 12404 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | In the first half of the MIPAS measurement period, until end of 2006, equatorial volcanic eruptions dominate the SO 2 mean distribution in the lower stratosphere near the equator (cf. Figs. 4 and 5) while during the second part mid-and high northern latitudes are mostly affected. Figure 12 compares the temporal evolution of the global SO 2 content derived from MIPAS for the altitude regions 15-23 km and 20-23 km with 5 the total injected mass as derived from nadir sounding instruments (cf. Table 2). In case of the larger eruptions of Kasatochi, Sarychev and Nabro, the mean SO 2 content at 15-23 km in the month directly after the eruption is about 1-2 % of the total injected mass. As shown in Sect. 5.2, due to the similar distribution of SO 2 volume mixing ratios between ACE-FTS and MIPAS after the eruption of Sarychev (Doeringer et al., 2012), 10 also the total mass derived from both instruments would be comparable.
For tropical eruptions, like those of Soufrière Hills, El Reventador or Raboul, the SO 2 mass in the lower stratosphere is in the order of 5 % or more of the total mass, likely due to the general upward transport in the tropical UTLS region. As already indicated in the plots of volume mixing ratio, the lower panel of Fig. 12 shows that altitudes  Fig. 12 and (2) that due to the cloud threshold described in Sect. 3, strongly aerosol contaminated limb-views are excluded from the retrieval leading to a general underestimation of SO 2 . The latter effect, however, only affects few limb-scans directly after the eruption in the vicinity of the major plume such that the monthly mean values are unlikely to be disturbed. 2006) an upwelling of SO 2 -enhanced air is tentatively visible. However, due to the processing of SO 2 during the months directly after the eruptions much smaller values of SO 2 remain for an upward transport. It is merely probable that sulfur is transported in the form of aerosols as has been shown e.g. in Vernier et al. (2011, Fig . 2). The time series of SO 2 show that enhanced values at its mid-stratospheric maximum near 5 28 km and above are present some months after the eruptions. However, as described in the following chapter there is a strong correlation with the quasi-biennial oscillation (QBO) especially at those altitudes. The resulting maxima of SO 2 are superposed to possible enhancements due to the volcanic eruptions and, thus, both effects are difficult to disentangle.

Regression analysis
In this section, the temporal development of the SO 2 dataset is analysed under consideration of a constant term a, a linear term b, several periodics and external parameters with coefficients c, d , and e using a multivariate fit approach as proposed by von Clar- vmr(t) = a + bt + c 1 qbo 1 (t) + d 1 qbo 2 (t) Here qbo 1 (t) and qbo 2 (t) are the Singapore time series of winds at 10 hPa and 50 hPa as provided by the Free University of Berlin (http://www.geo.fu-berlin.de/en/ met/ag/strat/produkte/qbo/index.html). They are used to describe the QBO signal in 20 the time series (Kyrölä et al., 2010). The sine and cosine terms describe annual 12406 (T 1 = 1 yr) and semi-annual (T 0.5 = 0.5 yr) variability including a phase-shift. The solar radio flux at 10.7 cm (F 10.7 (t)) provided by the NOAA Space Weather Prediction Center (http://www.swpc.noaa.gov), serves as proxy for solar activity (Kyrölä et al., 2010). A possible offset between P1 and P2 data is accounted for by adding a fully correlated block matrix to the P1-part of the data error covariance matrix.

5
In Fig. 13 results of the regression analysis are shown for exemplary latitude/height bins illustrating the influence of various partial signals. In the following this Fig. 13 is discussed together with Fig. 14 where the global distribution of the amplitudes of the single regression functions is presented (see caption of Fig. 14 for a more detailed description). 10 The quality of the fit is illustrated by the χ 2 of the regression (see plot "CHI2" in Fig. 14). In general the global fit quality is relatively homogeneous. Main exceptions are the regions of the polar upper stratosphere and the northern lowermost stratosphere with higher values of χ 2 . The latter is probably explained by volcanic activity the influence of which was mainly concentrated at altitudes up to 20 km in the north 15 (see Sect. 6.3). Further, influences by anthropogenic sources may also contribute to the higher values of χ 2 there. Enhanced χ 2 at high latitudes above 30-35 km altitude stem from strongly increased values of SO 2 up to 120-160 pptv during single winter months (see e.g. top row of Fig. 5) which cannot be represented adequately by the regression model. 20 The QBO signal can clearly be detected in the tropics at altitudes of 28-36 km introducing a variability in SO 2 of around ±20 pptv (cf. top right panel in Fig. 14 and example "QBO" in Fig. 13). Here the major part of the QBO signal stems from qbo 1 (t), the Singapore winds at 10 hPa whereas the winds at 50 hPa (qbo 2 (t)) are of minor importance. High values of SO 2 are correlated with the easterly phase of the QBO. Such 25 a QBO signal has also been observed in case of time series of stratospheric aerosol extinction (Trepte and Hitchman, 1992;Vernier et al., 2011).
Large amplitudes (up to about 30 pptv) of the annual modes are present especially at altitudes of 25-35 km at polar latitudes (cf. Fig. 14). These are explained by the Introduction downwelling of SO 2 -rich air from higher altitudes during winter. The N-S asymmetry with larger amplitudes at lower altitudes in the south are presumably due to the more persistent Antarctic polar vortex. The mode of the semi-annual oscillation (SAO) with amplitudes of up to 10 pptv of SO 2 is visible above 30 km altitude at tropical and subtropical latitudes (Fig. 14). This mode can be identified in the example dataset for 20-30 • S at 41-43 km altitude (see panel "Semi-Annual" in Fig. 13) by a second, smaller peak in the second half of each year. Such a seasonal asymmetry with stronger variation during the first cycle has been described as typical feature of the equatorial SAO (Delisi and Dunkerton, 1988;Garcia et al., 1997). Further, a latitudinal asymmetry of the SAO with stronger intensity south-10 wards of the equator as indicated in Fig. 14 has been observed previously in observations of temperature and wind (Belmont and Dartt, 1973;Ray et al., 1998). The global distribution of the reconstructed linear slope is presented in Fig. 14  eral decrease until about 2009 and a levelling off afterwards. As discussed above, this correlates also with the F 10.7 flux time series and it might be that part of the solar variability is mapped into the linear slope (and vice versa). In fact, in Fig. 14 the upper atmospheric part with negative linear slope in the south is correlated with the area of strongest F 10.7 amplitude. To determine more specifically the significance of the trend 25 values we have used the method of model-error correction as described in detail in Stiller et al. (2012, Sect. 7). In short, the error covariance matrix of the data is inflated and mean error correlations are inferred from the fit residuals such as to achieve a unity χ 2 of the fitted trend residuals. This increased error accounts for deficiencies in the linear regression to describe the real atmospheric situation and any underestimation of the measurement errors. The results of this analysis are presented in Fig. 15.
The corrected values of the slope are very similar to the ones determined originally under the assumption of measurement error only. The corresponding significance of the slope in terms of its error σ is shown in the middle left of Fig. 15 and in the bottom 5 left part the remaining data points for which the trend is larger than 2σ are presented. The latter clearly shows that the only larger contiguous area of significant positive trend is located in the Southern Hemisphere reaching from 25-30 km in the subtropics down to the lower stratosphere in polar regions. This area separates two regions of negative slope: one above 35 km at austral mid-latitudes, the other in the tropical stratosphere below 25 km. Single time series of the latter (cf. Fig. 13 right, bottom) strongly suggest that the negative slope is caused by larger tropical volcanic influence at the beginning of the first and of the second MIPAS measurement period compared to the end of both periods.
A connection with solar activity is indicated by a maximum signal of about 20 pptv 15 for the fit with the F 10.7 cm flux at mostly tropical and sub-tropical latitudes above 30-35 km altitude (cf. Fig. 14). This correlation stems from a positive curvature of the 10 yr dataset as visible in panel "F 10.7 " of Fig. 13. Mind, however, that there is a correlation between the long-wave F 10.7 time series, a linear trend and the independent bias values of the first and the second MIPAS measurement period. E.g., if SO 2 vmr values of 20 the first period had a systematic positive offset compared to the second period, the F 10.7 index would partly compensate for it. However, inspection of single time series (like in Fig. 13) reveals that increases after 2010 from a lower constant level during the years 2007-2010 are present in the data sets of the second period, which is in correspondence with the solar cycle. As for the linear trend just described, we have 25 estimated the significance of the solar cycle signal as shown in the right part of Fig. 15.
Here the major contiguous area of significant F 10.7

Conclusions
We have retrieved zonal 10 • mean stratospheric distributions of SO 2 from MI-PAS/Envisat monthly mean spectra for the period 2002-2012. These are the first observations of height-resolved SO 2 covering (1) nearly the entire region of the stratosphere, (2) the whole globe, and, (3) a time period of about 10 yr. Retrieval diagnostics indicate 5 total error bars of 5-20 pptv most of which is dominated by instrumental noise followed by non-linearity uncertainties due to the retrieval approach at lower altitudes and assumptions on temperature at higher levels. Since the a-priori profiles of the retrieval has been chosen constantly zero over the whole altitude range, the shape (apart from some smoothing due to the regularisation type) and the absolute values of the retrieved 10 profiles are entirely based on information from the measurements. The lack of stratospheric SO 2 observations poses a problem for validation of our dataset. To get an estimate on the data quality, we have performed comparisons with the few available measurements. Above about 30 km there is good consistency with observations from ATMOS, the only available information there. In the altitude range In-situ observations are also sparse above 14 km and not globally available. In general, the MIPAS mid-latitude global mean values at the lower heights (about 10-20 pptv with a standard deviation of about 10 pptv) are within, but at the lower limit of the stratospheric in-situ data with values of roughly 10-50 pptv. Apart from any possible bias of the MIPAS or the insitu datasets this might stem from differences in the observing altitudes (in-situ mostly below 15 km) or geographic location for higher altitudes (in-situ mostly over northern continents). Introduction A further hint for the validity of our results is the general consistency with the understanding of the behaviour of SO 2 from model results. In this respect we could observe for the first time the local maximum of SO 2 at around 25-30 km resulting from conversion of COS (Brühl et al., 2012). Further, the MIPAS data corroborate the strong increase of SO 2 above 30-35 km explained by photolysis of H 2 SO 4 (Rinsland et al., 1995;Vaida et al., 2003;Mills et al., 2005). The observed increase is larger than the one modelled in Brühl et al. (2012) which can be explained by neglect of the visible and near-IR photolysis bands of H 2 SO 4 (Vaida et al., 2003;Brühl et al., 2013). The explanation for the fast increase of small aerosol particles at high latitudes in spring as the conversion from SO 2 on availability of sunlight (Mills et al., 2005) can be followed 10 in terms of an annually reoccurring depletion of SO 2 which has been observed for the first time.

Conclusions References
The global height-resolved fields of SO 2 in the lower stratosphere over 10 yr confirm the importance of volcanic eruptions for the distribution of sulfate in that region. No other source of SO 2 comparable to that of volcanoes could be detected. This supports 15 the conclusions of e.g. Neely et al. (2013) and Vernier et al. (2011) on the minor importance of an anthropogenic influence. Further, the current dataset will help to test model simulations, based on total SO 2 injection masses from nadir sounders (Neely et al., 2013;Brühl et al., 2013), on the distribution and total mass entry of SO 2 into the stratosphere. 20 A multivariate regression analysis of the SO 2 time series at different altitude-latitude bins with global coverage revealed a QBO signal at around 30 km in the tropics and a SAO signal above at tropical and sub-tropical latitudes. Further, there appears a possible connection with the solar cycle above 30 km at equatorial and southern midlatitudes. Significant signals of linear trends are detected at altitudes up to 30 km in 25 the Southern Hemisphere (positive), at high altitude southern mid-latitudes (negative), and at lower tropical altitudes (negative). The negative linear trend at low latitudes is caused by the signal of tropical volcanic eruptions leading to peaks of enhanced SO 2 volume mixing ratios until end of 2006.
In this study we have concentrated on the retrieval from average spectra which enabled us to get SO 2 distributions throughout the stratosphere with a moderate temporal and latitudinal resolution. As shown, these results are e.g. useful for validation of the sulfur budget in chemistry-transport and chemistry-climate models above about 15 km. A direct comparison of such a model with our results is investigated in a parallel 5 study (Brühl et al., 2013). The retrieval from mean spectra is, however, not the optimal way to study the behaviour of SO 2 in the upper troposphere and lowest stratosphere where a resolution in longitudinal direction and a retrieval down to lower altitudes is desirable. Thus, as a complementary approach single-profiles retrievals from MIPAS which will cover the altitude range up to about 20 km are currently in progress. With 10 such a dataset it will e.g. be possible to follow the height-resolved evolution of volcanic plumes.
The retrieval of COS distributions from MIPAS, as investigated initially in Burgess et al. (2005), and the reconstruction of sulfate aerosol properties, as shown in case of MIPAS-Balloon by Echle et al. (1998) will be further significant steps forward to reach   Thornton et al. (1999) 1991-1996 W-Pacific 12 60 Jurkat et al. (2010) Oct 2008 Europe 11.  12420  Table S1) and references therein. References for additional eruptions: Soufrière Hills on 12 July 2003: Carn and Prata (2010), Soufrière Hills on 11 February 2010: Cole et al. (2010), Nabro mass: Clarisse et al. (2012), Nabro injection height: Bourassa et al. (2012) and connected discussion (Fromm et al., 2013;Vernier et al., 2013;Bourassa et al., 2013), Merapi: Surono et al. (2012).   Fig. 3. Error estimation of the retrieval for one month during P1 and one month during P2. Considered error sources are the uncertainty of the foreign-broadened half-width and lineintensity of interfering species (spe_hw_itf, spe_int_itf), the knowledge of these parameters for SO 2 (spe_hw_SO2, spe_int_SO2), the uncertainties of the instrumental line-shape and gaincalibration (ils, gain), the errors in the assumed tangent altitudes and temperatures (htang, Tecm), the error due to the applied technique of retrievals from averaged spectra (nlin) and the spectral noise of the instrument (noise). The total error has been determined by quadratic combination of the single error components. Introduction   Table 2