Stratospheric Ozone Trends and Variability as Seen by SCIAMACHY from 2002 to 2012

Abstract. Vertical profiles of the rate of linear change (trend) in the altitude range 15–50 km are determined from decadal O3 time series obtained from SCIAMACHY1/ENVISAT2 measurements in limb-viewing geometry. The trends are calculated by using a multivariate linear regression. Seasonal variations, the quasi-biennial oscillation, signatures of the solar cycle and the El Nino–Southern Oscillation are accounted for in the regression. The time range of trend calculation is August 2002–April 2012. A focus for analysis are the zonal bands of 20° N–20° S (tropics), 60–50° N, and 50–60° S (midlatitudes). In the tropics, positive trends of up to 5% per decade between 20 and 30 km and negative trends of up to 10% per decade between 30 and 38 km are identified. Positive O3 trends of around 5% per decade are found in the upper stratosphere in the tropics and at midlatitudes. Comparisons between SCIAMACHY and EOS MLS3 show reasonable agreement both in the tropics and at midlatitudes for most altitudes. In the tropics, measurements from OSIRIS4/Odin and SHADOZ5 are also analysed. These yield rates of linear change of O3 similar to those from SCIAMACHY. However, the trends from SCIAMACHY near 34 km in the tropics are larger than MLS and OSIRIS by a factor of around two. 1 SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY 2 European environmental research satellite 3 Earth Observing System (EOS) Microwave Limb Sounder (MLS) 4 Optical Spectrograph and InfraRed Imager System 5 Southern Hemisphere ADditional OZonesondes


Introduction
The stratospheric O 3 layer shields the Earth from UVB and UVC radiation in the 240-320 nm range. The absorbed energy is the main heat source of the stratosphere, and also drives the atmospheric circulation in the upper atmosphere (e.g. Weber et al., 2011). The abundance of stratospheric O 3 is sensitive to both chemical and dynamical factors and has been altered by anthropogenic activity. The most prominent example of human impact on the O 3 layer is the Antarctic O 3 hole, which is caused by anthropogenically released chlorine and bromine compounds. In order to protect the O 3 layer, the production of halogenated source gases was banned by the Montreal Protocol of 1987 and its amendments. A slowdown of the O 3 decline and even an onset of a recovery have been inferred from observations (Newchurch et al., 2003;Yang et al., 2006;WMO, 2007). The recovery of the global stratospheric O 3 layer to values observed before the chlorofluorocarbon (CFC)-driven depletion is, however, not expected before the middle of this century (WMO, 2011).
Measurements of O 3 have a long tradition, and various trend studies have been performed using data from the last few decades (WMO, 2007). In recent years, several trend studies have investigated the impacts of O 3 -depleting substances (ODS) and dynamical forcings affecting the O 3 832 C. Gebhardt et al.: Stratospheric ozone trends and variability as seen by SCIAMACHY shield (WMO, 2007(WMO, , 2011. The majority of these trend assessments agreed that the present stage of stratospheric O 3 is somewhere between the rate of depletion slowing down and a subsequent turnaround phase, expected as a result of the measures implemented within the Montreal Protocol and its amendments. This manuscript contributes to our knowledge of stratospheric O 3 and its changes during the period following the turnaround in stratospheric halogens in the late 1990s, where the halogen load, still high, is slowly declining. Vertically resolved O 3 trends have been derived from observations by satellite instruments, ground-based lidars and microwave radiometers, and balloon-borne ozonesondes (Steinbrecht et al., 2006;Jones et al., 2009;Randel and Thompson, 2011;Mieruch et al., 2012). In Mieruch et al. (2012), O 3 trends from SCIAMACHY and those inferred from a range of instruments were compared within the time from 2002 to 2008 with the main focus on possible instrumental and algorithmic issues. Analysing a number of different satellite datasets, Jones et al. (2009) inferred O 3 trends from a multi-instrument mean between 1979 and 2008. SCIAMACHY, OSIRIS/Odin, and SMR/Odin contributed data starting in the years 2002 and 2001 respectively. Jones et al. (2009) showed that these satellite data agreed with each other to a high extent. After 1997, Jones et al. (2009) identified positive midlatitude O 3 trends between 35 and 45 km on the order of a few percent per decade. These findings are corroborated by Steinbrecht et al. (2006Steinbrecht et al. ( , 2009) by using ground-based measurements. As the upper stratosphere is highly sensitive to halogen chemistry (WMO, 1999), these results may be interpreted as signs of the onset of a turnaround in the O 3 abundance in response to declining halogens. Analysing a merged dataset from satellites and ozonesondes, Randel and Thompson (2011) identified negative O 3 trends in the lowermost tropical stratosphere. These were primarily attributed to dynamical factors, namely enhanced tropical upwelling.
In this study, a global O 3 dataset obtained from the SCIA-MACHY instrument on-board ENVISAT is investigated. SCIAMACHY was in routine operation from August 2002 until ENVISAT suddenly and unexpectedly failed on 8 April 2012. In its limb-viewing geometry, SCIAMACHY performed measurements of the scattered solar radiation. These limb measurements are used to retrieve vertical profiles of trace gases and aerosols and to detect clouds.
Rates of linear change or trends in the SCIAMACHY limb O 3 time series are determined in our investigation for the time range from August 2002 to April 2012. These trends are presented as functions of altitude and latitude. The resulting vertical trend profiles of O 3 are presented for latitudes between 60 • N and 60 • S. The trends from SCIAMACHY are compared with those derived from OSIRIS/Odin and SHADOZ ozonesondes in the tropics and with those from EOS MLS in the tropics and at midlatitudes. This paper is structured as follows. Section 2 describes briefly the SCIAMACHY limb O 3 retrieval. The multivari-ate linear regression, used in our study, is introduced and explained in Sect. 3. The O 3 time series from SCIAMACHY and their regression models are described in Sect. 4. In Sect. 5, the vertical profiles of the rate of linear change or trend of the O 3 retrieved from SCIAMACHY measurements are presented. O 3 trend profiles from SCIAMACHY are compared with those from EOS MLS and OSIRIS/Odin, other atmospheric limb sounders, in Sect. 6. In addition, comparisons are performed in the inner tropics including SHADOZ ozonesondes. Section 7 discusses atmospheric phenomena potentially responsible for the observed O 3 trends. In Sect. 8, sources of errors are discussed. Section 9 summarises the main results obtained in this study.

SCIAMACHY limb ozone
The European environmental research satellite ENVISAT hosted ten instruments, which were operational until the sudden loss of contact with ENVISAT on 8 April 2012. SCIA-MACHY observed electromagnetic radiation upwelling from the Earth's atmosphere in 3 measurement modes: occultation, nadir, and limb geometry. Detailed information on the instrument and mission objectives is provided by Burrows et al. (1995) and Bovensmann et al. (1999). In limbviewing geometry, the instrument scanned the horizon in 3.3 km steps from −3 to 92 km (0 to 92 km since October 2010). This vertical sampling and the instantaneous field of view (∼ 2.6 km in vertical direction at the tangent point) defined the vertical resolution, which was typically 3-4 km. ENVISAT was in a Sun-synchronous orbit with an inclination of 98 • . This resulted in global coverage for SCIA-MACHY limb measurements being achieved within 6 days at the Equator and less elsewhere.
The SCIAMACHY results presented here use the limb O 3 retrieval version 2.9 of IUP Bremen. As an intermediate retrieval result, monthly mean O 3 data are gridded horizontally into 5 • latitude × 15 • longitude and vertically into 1 km altitude. The overall altitude range of O 3 retrieval is 10 to 75 km. As the satellite retrieval often suffered from cloud interference below 15 km, only altitudes above 15 km are used in this study.
Comparing SCIAMACHY limb O 3 data to vertically resolved O 3 datasets from other instruments, Mieruch et al. (2012) showed that comparison results are not significantly changed using either collocated measurement pairs or zonal monthly means of gridded data. This highlights that zonal monthly means are appropriate for performing instrument comparisons. Our investigation uses zonal mean O 3 data for the latitudes defined by the gridding of data as well as for wider latitude bands. Determining the rate of linear change or trend for all altitudes between 15 and 50 km results in vertical profiles of O 3 trends.
The zonal monthly mean values are calculated as arithmetic means because the errors of single measurements are Atmos. Chem. Phys., 14, 831-846, 2014 www.atmos-chem-phys.net/14/831/2014/ expected to be normally distributed and no issue with outliers is known. Any of these zonal monthly means is typically composed of several hundreds of single measurements. Hence, the random errors of zonal monthly means can be deemed as negligible. Trend uncertainties used in the following are calculated from the covariance matrix of regression. The boundaries of 60 • N and 60 • S are chosen for this investigation to avoid any effects which are directly related to the polar vortex and because of gaps in the SCIAMACHY sampling during the high-latitude winter. In order to avoid possible influences of the South Atlantic Anomaly, measurements with tangent point coordinates between longitudes of 90 • W and 0 • W and latitudes south of 20 • S are excluded from this analysis (it is worth noting that a typical difference in latitude between instrument and tangent point coordinates of around 20 • is taken into account by the criteria used).

Multivariate linear regression
A multivariate linear regression is applied to the datasets with the autocorrelation of consecutive values accounted for by the Cochrane-Orcutt transformation (Cochrane and Orcutt, 1949). The observed values of autocorrelation are similar to those shown in Mieruch et al. (2012). The regression is unweighted. The following terms are used in the regression: where t is the time in months and μ, ω, and β 11 , β 12 , . . . , β 23 , β 24 are the fitting parameters. The harmonics with 12-, 6-, 4-, and 3-month periods, j = 1, 2, 3, 4, a sine and cosine term each, are used to represent seasonal changes. The harmonics of 12 and 6 months approximate the annual and semi-annual cycles. The combination of sine and cosine terms provides a flexible adjustment to any phase of the (semi-)annual variation. As discussed by Stiller et al. (2012), the terms having 4-and 3-month periods improve the fit quality because the oscillation patterns are possibly compressed or stretched with respect to a harmonic shape. In addition, a linear fit having a gradient or slope ω and an intercept μ is included in the regression analysis. The gradient represents the O 3 trend at a given altitude. The 1σ value defined by the covariance matrix of regression coefficients serves as trend uncertainty. It is also used to draw error bars of trends in subsequent figures. The trends and their uncertainties are presented in relative units of % dec −1 , i.e. percent per decade (relative with respect to the mean value of the underlying time series). The criterion for the trend being significant at the 95 % confidence level is that the absolute ratio of the trend to its uncertainty is larger than 2 (Tiao et al., 1990). As discussed later in Sect. 3.2, it is also advantageous to extend the linear regression (see Eq. 1) by additional terms representing the quasi-biennial oscillation (QBO), the solar cycle (SC), and the El Niño-Southern Oscillation (ENSO): At northern midlatitudes, Eq. (1) is optimised because of the strong interannual variability of the annual cycle of O 3 . In the 15-26 km range of the 60-50 • N latitude band, the cumulative eddy heat flux (based on ERA-Interim, 50 hPa eddy heat flux integrated from 45 • N to 75 • N, cumulation starts in October) is fitted instead of harmonic terms. The cumulative eddy heat flux is a proxy accounting for the wave-driven O 3 accumulation in winter and early spring. For 60-50 • N, the interannual variability in total O 3 is found to be well approximated by the cumulative eddy heat flux (Dhomse et al., 2006). Here, correlation tests show that throughout the altitude range of 15-26 km, the proxy being time lagged by 2 months is optimum.

QBO, solar cycle, and ENSO signatures in ozone
The quasi-biennial oscillation (QBO) is a quasi-periodic signal observed in the tropical stratospheric zonal wind speed. It alternates between east and west phase with the period varying between 2 and 3 yr. Previous studies report a maximum QBO response of tropical O 3 in the altitude ranges of 20-27 km and 30-38 km (Zawodny and McCormick, 1991;Chipperfield et al., 1994). In the tropics, the QBO response of O 3 is driven by effects related to the vertical transport (Butchart et al., 2003). The QBO-induced meridional circulation modulates the upwelling branch of the Brewer-Dobson circulation, leading to the largest changes of the O 3 volume mixing ratio (VMR) in the altitude range of large vertical gradients of the O 3 VMR, which are located below and above the maximum in the vertical O 3 VMR profiles. The result is a compacting (vertical squeezing) of the mean vertical O 3 VMR profile at westerly QBO phase and a lofting (vertical stretching) at easterly QBO phase. Around the maximum of the mean vertical profile of the O 3 VMR, where the vertical gradient changes sign, the phase change of the QBO response of O 3 takes place (Butchart et al., 2003). With the QBO phases modulating the Brewer-Dobson circulation, a QBO response is expected to be observed also in extratropical O 3 (Baldwin et al., 2001).
In the following, the 10 hPa and 30 hPa Singapore winds (Naujokat, 1986, available from http://www.geo.fu-berlin.de/en/met/ag/strat/produkte/qbo/) are used as fit proxies: with a and b being their fit coefficients. The indices "10" and "30" denote the pressure level of the proxy. This approach is applied both in the tropics and extratropics to account for any QBO effects. The monthly time series of the Singapore winds are smoothed by a 4-month running average before being used as QBO proxy. The same procedure is done with the Mg II index when used as a solar cycle proxy. Both proxies are not detrended. The 11 yr solar cycle, which leads to pronounced variations of the solar radiation in the UV spectral range, is known to influence stratospheric O 3 (Gray et al., 2010). It is widely accepted that the solar cycle response of O 3 occurs without any time lag and at positive correlation (e.g. Soukharev and Hood, 2006;Remsberg and Lingenfelser, 2010). The solar cycle response of O 3 may extend from the tropics into the midlatitudes (SPARC CCMVal, 2010). In the following, the solar cycle proxy is included in the multivariate linear regression for any latitude analysed. A multi-instrument monthly average Mg II index -which is mainly based on GOME, SCIAMACHY, and GOME-2 observations over the time span under study -serves as solar cycle proxy. The underlying data can be obtained from http://www.iup.uni-bremen. de/gome/solar/MgII_composite.dat (Snow et al., 2013;Weber et al., 2013).
In addition to the QBO and the solar cycle, the O 3 time series may also be influenced by the El Niño Southern Oscillation (ENSO) (Randel et al., 2009;Randel and Thompson, 2011). Employing an ENSO proxy based on the Niño 3.4 index, Oman et al. (2013) previously demonstrated its applicability to reveal ENSO signatures in vertically resolved O 3 time series from EOS MLS. With the ENSO response of O 3 propagating towards higher altitudes, time lags between ENSO signatures in O 3 and ENSO proxies may arise. For instance, Randel et al. (2009) and Randel and Thompson (2011) employed ENSO proxies at a time lag of 2 months.
Similar to Oman et al. (2013), an ENSO proxy based on anomalies of the Niño 3.4 index (available from http://www. cpc.ncep.noaa.gov/data/indices/) is used in our study. Time lags are accounted for by the combined fit of the proxy and its time derivative: where N 34 denotes the proxy and (t) the time lag. Accounting for the non-linearity of dN 34 (t) dt , the fit is repeated with the lagged proxy until the remaining time lag approaches zero. This iteration starts at a time lag of 2 months. The time lag of the proxy not settling at a certain value (including zero) or becoming unrealistic is the criterion for omitting ENSO terms at regression. In practice, the latter is the case at extratropical latitudes and in the tropics at altitudes above 25 km. In the following, ENSO is considered at regression only at tropical latitudes (any latitude band within 20 • N-20 • S) in the 15-25 km altitude range. If the criterion for omitting ENSO terms is still fulfilled, the underlying time series is excluded from analysis (resulting in the gaps in the vertical trend profiles shown Figs. 8 and 11 and in the QBO amplitude/solar cycle response shown in Figs. A1 and A2).

Time series of ozone
Figures 1-6 show O 3 time series at the altitudes of 21 km, 35 km, and 44 km at northern midlatitudes (60-50 • N) and in the tropics (20 • N-20 • S). The time series are overlaid by their fitting curves from the regression model, and the fit residuals are plotted below. The individual terms from the regression -i.e. the harmonic, QBO, solar cycle, and linear terms -are shown in corresponding panels below (in the tropics at 21 km, ENSO terms are included as well while, at northern midlatitudes at 21 km, eddy heat flux terms are shown instead of harmonic terms). The harmonic terms are shown comprising all periods used, i.e. 12, 6, 4, and 3 months, and also as combination of the terms with 4-and 3-month periods only. With the exception of 44 km at 60-50 • N, the latter terms of 4-and 3-month periods improve the closure between time series and fitting curve, but do not change the amplitude of the harmonic terms substantially. The harmonic terms are overlaid by the time series with all fit terms with the exception of the harmonics subtracted. Any other component of the fitting curve is overlaid by the time series with all other fit terms removed accordingly.
At all altitudes, good fit quality is obtained. The use of harmonics for approximating the (semi-)annual oscillation as well as the combination of the 10 hPa and 30 hPa Singapore winds to fit QBO appears to be a sufficiently complete approach. The fit residuals are well within ±5 % to ±10 % of the fitted time series both at midlatitudes and in the tropics.
At 21 and 35 km at midlatitudes, the annual cycle terms (green curve has an annual pattern) dominate in the fitting curve. At 21 km, O 3 has its annual maximum in late winter/early spring and declines throughout summer. At 35 km, the annual maximum of O 3 is reached in late spring/early summer, persisting for the summer months. At 44 km at midlatitudes, the harmonic terms have two maxima and minima per year. The annual minimum occurs in summer and a secondary minimum in winter. These patterns are explained as follows: at 21 km, the periodic variability of O 3 is dominated by transport effects in winter, leading to an accumulation of O 3 , and by catalytic O 3 depletion in summer which is responsible for the descending branch of the annual cycle. At 35 km and 44 km, the periodic variability is governed by photochemical O 3 production and by catalytic O 3 depletion respectively, both being at its maximum in summer (Perliski et al., 1989). At 35 km, the annual maximum in summer driven by photochemical production is flattened by catalytic depletion which is of minor consequence. At 44 km, catalytic depletion has gained control leading to the annual minimum in summer. Overlaid photochemical production of O 3 leads to a secondary minimum in winter.
At 21 km in the tropics, the annual cycle and the QBO make comparable contributions to the fitting curve. Discernible contributions are also made by the ENSO. The annual cycle has its maximum in the Northern Hemisphere summer. It resembles well the phase of the annual oscillation    of tropical O 3 in the lowermost tropical stratosphere which is driven by effects related to the tropical upwelling (Randel et al., 2007). At 35 km in the tropics, the QBO and semi-annual terms dominate the periodic variability of O 3 . As discussed before, the maximum QBO response of tropical O 3 is located within around 20-27 km and 30-38 km. At 44 km in the tropics, the semi-annual terms provide the most important contri-bution to the fitting curve. In the middle tropical stratosphere, the semi-annual oscillation of O 3 is driven by photochemical O 3 production, which reflects two Equator passages of the Sun per year (Perliski et al., 1989).   Figure 8 shows a latitude-altitude plot of the O 3 trends spanning the latitude range from 60 • N to 60 • S. The plot is based on zonally averaged O 3 data in 5 • latitude steps, defined by the data gridding described in Sect. 2. As a function of the latitude, the trends are shown for the 15 to 50 km altitude range. As evidenced before for the 20 • N-20 • S tropical latitude band, there are double-peaked positive trends below 30 km (significant at their maxima at around 21 km and between 25 and 30 km) and single-peaked negative trends between 30 and 38 km in the inner tropics. In the 30-35 km range, an "island" of significant negative O 3 trends of up to −20 % dec −1 now appears. At latitudes from 20 • N to 30 • N, significant positive trends are seen at altitudes between 20 and 25 km. Positive trends in the upper stratosphere (identified in Fig. 7 for 20 • N-20 • S between 39 and 47 km and also for 60-50 • N between 40 and 45 km) are observed over large parts of the latitude range from 60 • N to 60 • S. These trends are mostly significant at values of around 5 % dec −1 . The hemispheric asymmetry at midlatitudes between 25 and 35 km is clearly pronounced with significant negative trends in the Northern Hemisphere and significant positive trends in the Southern Hemisphere. At around 15 km, positive trends appear between 20 • S and 30 • S. These trends should, however, be treated with caution because of potential impact of clouds at altitudes below 20 km (Mieruch et al., 2012).

Atmos
A discussion of the latitude and altitude dependence of the QBO and solar cycle responses as seen by SCIAMACHY limb is presented in Appendix A.

OSIRIS/Odin
The Odin satellite, which carries the OSIRIS instrument (Llewellyn et al., 2004), follows a near-terminator orbit so that the illumination conditions specific to OSIRIS are always close to the solar terminator. The ascending node was between 6 and 7 p.m. over the time span considered here. Similar to SCIAMACHY, OSIRIS measures the scattered solar light in limb-viewing geometry. Seasonal variations of day-and nighttime interrupt the OSIRIS measurements in the winter hemisphere. As a result, OSIRIS provides continuous time series only in the tropics. The data might also be inhomogeneously distributed around the Equator, i.e. predominantly to the north from it during northern summer and vice versa. Thus, the comparison between the instruments is performed for the tropical 20 • N-20 • S latitude band only. The OSIRIS O 3 data, used in this study, are from the retrieval version 5.07 produced by the University of Saskatchewan, which has a modified filtering of outliers with respect to earlier versions. The OSIRIS data separate clearly into records at a.m. and p.m. local time. As pointed out by Huang et al. (2010), diurnal variations of O 3 are expected to be on the order of only a few percent in the stratosphere. The AM Equator crossing time of Odin is closer to the Equator crossing of ENVISAT at daylight. Accordingly, the a.m. data from OSIRIS, which are closer in local time to the data from SCIAMACHY, are used in the following (without any correction for diurnal variation).
The retrieval altitudes of OSIRIS and SCIAMACHY lie between one another. For each altitude of SCIAMACHY from 15 to 50 km, the O 3 values from the adjacent altitudes of OSIRIS were interpolated.
The instrument comparison between SCIAMACHY and OSIRIS (also between SCIAMACHY and EOS MLS as described in Sect. 6.2) uses zonal monthly mean data. As mentioned in Sect. 2, a comparison of zonal monthly means from gridded data is considered an equivalent alternative to the use of collocated data. As a further check, the trend profiles from SCIAMACHY and OSIRIS (comparison described in Sect. 6.4) were tested to remain practically unchanged when using collocated measurements.

EOS MLS
The MLS instrument aboard the EOS Aura satellite (Jarnot et al., 2006) measures the thermal radiation emitted by the Earth's atmosphere in the microwave spectral range on both the day and night side of the near-polar orbit. MLS went into routine operation in August 2004. In this comparison, we have used O 3 profiles retrieved from MLS daytime measurements. Aura crosses the Equator in ascending node at 13:45 local time. Its daytime measurements extend deep into both hemispheres similar to those from SCIAMACHY. Thus, the comparison between the instruments is also done at midlatitudes, represented by the latitude bands of 60-50 • N and 50-60 • S. The period used for the comparison is August 2004-April 2012.
In accordance with the recommendations from the MLS team, the MLS O 3 data used here are from the retrieval version 2.2. The pressure serves as the vertical coordinate in version 2.2. After conversion from pressure to geometric height, the MLS O 3 profiles are interpolated at the altitudes of the SCIAMACHY retrieval. Zonal monthly means are calculated between 15 and 50 km.

SHADOZ ozonesondes
Balloon-borne in situ measurements of ozonesondes are also used for a comparison with the O 3 trend from SCIAMACHY limb. The altitude range of the balloon sonde O 3 data is limited to 30-35 km by different factors, such as balloon burst and evaporation and/or freezing of the wet-chemical sensing solution used in the sondes. The data from ozonesondes have a vertical resolution on the order of 100 m, clearly finer than that from satellite instruments. For the comparison, the vertical O 3 profiles from the sondes are smoothed over intervals of ±2 km relative to altitudes defined by the O 3 retrieval from SCIAMACHY, i.e. from 15 to 30 km in steps of 1 km. In the following, these smoothed O 3 profiles are used to determine monthly averaged O 3 time series from the sondes.
SHADOZ is a network of more than 10 ozonesonde stations at (sub)tropical sites (e.g. Thompson, 2003). Since coming into being in 1998, the network has been systematically operational with typically several ozonesonde launches per month per station. The datasets of all stations are freely accessible via http://croc.gsfc.nasa.gov/shadoz/. Here, we select the inner-tropical stations at Ascension (8.0 • S, 14.4 • W), Kuala Lumpur (2.7 • N, 101.7 • E), Nairobi (1.3 • S, 36.8 • E), Natal (5.4 • S, 35.4 • W), and Paramaribo (5.8 • N, 55.2 • W). These stations are relatively equidistant in latitude (less importance is attached to their longitudes as tropical stratospheric O 3 is known to be highly zonally symmetric). The monthly averaged O 3 time series over all 5 stations are calculated. In order to account for possible offsets between the stations, the mean values of single-station time series are forced to be equal before averaging. The is done by shifting them so that a reference value is matched. This reference value is the mean value of one of the stations, in our case Paramaribo (arbitrarily selected). The resulting sonde trend profile is compared to the trend profile from SCIAMACHY, which is based on data selected at latitudes and longitudes around the sites of the ozonesonde stations and averaged accordingly.  MLS. There is also reasonable agreement at southern midlatitudes. The trend profiles from both instruments follow each other closely and have similar error bars. The agreement between SCIAMACHY and MLS holds up to 50 km, where both profiles become increasingly negative.

SCIAMACHY vs. OSIRIS and MLS
Tropical O 3 trends from SCIAMACHY, OSIRIS, and MLS are compared in Fig. 11. The tropics are represented by the latitude band of 20 • N-20 • S, and the time span from August 2004 to April 2012 is selected for the comparison. In qualitative agreement, SCIAMACHY, OSIRIS, and MLS show negative trends between 30 and 35 km. SCIAMACHY reaches around −10 % dec −1 , while the trends from OSIRIS are around −4 % dec −1 at maximum. The trends from MLS lie between those from SCIAMACHY and OSIRIS.
Between 25 and 30 km, the O 3 datasets from all three instruments show in good agreement positive trends. Agreement within the trend uncertainties is achieved down to around 20 km. In the UTLS region, the trends from SCIA-MACHY and OSIRIS change from moderately positive at 20 km to moderately negative at 15 km, while MLS increases to 10-20 % dec −1 (which is consistent with the results from SHADOZ; see following subsection). Between around 38 and 46 km, both SCIAMACHY and OSIRIS show positive trends. Trends up to around 6 % dec −1 are obtained from SCIAMACHY near 45-46 km and from OSIRIS around 40 km. Contrastingly, the trends from MLS are zero to negative above 40 km. Above 48 km, zero to slightly negative O 3 trends are shown by all compared instruments. Taken as a whole, the shape of the trend profile of tropical O 3 identified from SCIAMACHY in Sect. 5 is confirmed by both OSIRIS and MLS.  Figure 12 compares the SHADOZ sonde and SCIAMACHY O 3 trend profiles from 15 to 30 km. Though it is derived from data averaged over latitudes and longitudes matching the sites of ozonesonde stations, the trend profile from SCIA-MACHY resembles well the trend pattern derived from zonal averages at latitudes from around 5 • N to 5 • S in Fig. 8. This is a result of the high zonal symmetry of tropical O 3 (Thompson, 2003;Randel and Thompson, 2011). The trend profile from SCIAMACHY shows zero to positive trends between 15 and 30 km. Good agreement to the positive trends seen by the sondes is obtained between 18 and 29 km. Below 18 km, the trends from the sondes climb towards a maximum while the trends from SCIAMACHY decay towards zero. The differences between 15 and 20 km may arise from the satellite retrieval suffering from the impacts of high convective clouds; see e.g. Mieruch et al. (2012).

Discussion of factors potentially explaining the O 3 trends observed
The O 3 trend profiles from SCIAMACHY and the comparison instruments tend to have non-negative trends in the tropical 17-20 km range. Constraining our study of trends to the 2002 to 2012 decade, this behaviour is in contrast to the results obtained by Randel and Thompson (2011). Their O 3 trends were inferred from a dataset merged between SAGE II and SHADOZ spanning the time range of 1984-2009. Negative trends were identified in the lowermost tropical stratosphere and attributed to enhanced tropical upwelling.
In the tropical stratosphere below 30 km, both changing dynamics and chemical composition potentially explain the positive O 3 trends observed by SCIAMACHY. For example, changes in the rate of tropical upwelling, changes in the amount of halogens, or changes in the NO X amount are possibly responsible for changes in O 3 . Analysing O 3 partial columns between 18 and 25 km, Yang et al. (2006) found that halogens are of minor importance in this altitude range because CFCs do not release radicals immediately after their injection through the tropical tropopause. Since the vertical gradient of O 3 is positive, enhanced vertical advection has a negative O 3 response in the lower tropical stratosphere. Randel and Thompson (2011) suggested that an increase of the tropical upwelling has led to negative O 3 trends around 17 to 21 km observed over the period 1984-2009. As the O 3 trends seen by SCIAMACHY and the comparison instruments are non-negative at this altitude range and become clearly positive between 25 and 30 km, they must have another explanation. According to Nevison et al. (1999), an increase in NO X might be a reason for a positive trend in O 3 at these altitudes. This is because NO X is involved in buffering effects as a result of interactions with HO X and ClO X beside catalytic O 3 depletion. Nevison et al. (1999) have shown that an increase in NO X results in an O 3 enhancement extending up to around 30 km in the proximity of the Equator. As noted by Burrows et al. (1979), NO X may also catalytically produce O 3 in the lowermost stratosphere. The altitude range below 30 km coincides with the positive O 3 trends observed in our study. As also follows from Nevison et al. (1999), increasing NO X may lead to O 3 decreases above 30 km. Accordingly, increasing NO X may contribute to the negative O 3 trends seen by SCIAMACHY in the tropical 30-38 km range as well. However, the anthropogenic increase of NO X assumed in Nevison et al. (1999) results in a decrease of around 1 % per decade in tropical midstratospheric O 3 , which cannot explain the observed negative trends of −4 to −10 % dec −1 inferred in our study (cf. Fig. 11). Thus, the correlation between O 3 and NO X pointed out by Nevison et al. (1999) can explain the observed trends only if sources of NO X other than anthropogenic emissions exist. An investigation of the possible sources of NO X is beyond the scope of this study. Further research is needed on this issue.
Positive O 3 trends of approximately 5 % dec −1 are inferred from SCIAMACHY in the upper stratosphere over large parts of the tropics and midlatitudes. As O 3 in the upper stratosphere is highly sensitive to halogens (WMO, 1999), these trends are favoured to reflect declining halogens. Evidence for a turnaround in upper stratospheric O 3 has already been pointed out by Reinsel (2002). Some attribution to declining halogens has been provided by Newchurch et al. (2003). Their study is based on satellite data indicating at least a levelling-off of O 3 in the tropics and extratropics. Meanwhile the O 3 abundance continuing some levelling-off or slightly increasing is supported by various ground-based lidar and microwave radiometer stations (Steinbrecht et al., 2006(Steinbrecht et al., , 2009. A potential explanation for some of the O 3 trends obtained as a function of latitude and altitude is provided by Eckert et al. (2013) based on MIPAS/ENVISAT O 3 data covering the same observation period as SCIAMACHY. A hypothetical shift of subtropical mixing barriers is theorised to result in negative trends in the northern extratropics and positive trends in the southern extratropics. This pattern of trends is similar to the asymmetry between the northern and southern trend profiles from SCIAMACHY between 25 and 35 km. In addition, positive/negative O 3 trends in the lower/middle tropical stratosphere are theorised by Eckert et al. (2013), although the tropical trends from SCIAMACHY partly are more pronounced and somewhat different in altitude. This explanation, if correct, implies that the subtropical mixing barriers having undergone shifts during the last decade may contribute to the tropical O 3 trends observed in our study.

Potential instrumental and algorithmic issues
As discussed by von Savigny et al. (2005), pointing misalignment is one of the major error sources for limb-viewing instruments on ENVISAT. A robust method for determining tangent height offsets from solar occultation measurements of SCIAMACHY has been developed by Bramstedt et al. (2012). Using this method, tangent height offsets on the order of few hundred metres were identified. The overall behaviour of the pointing misalignment is mainly governed by constant and seasonally varying components. In addition, a continuous drift of the pointing misalignment on the order of 50 m throughout the first four years of the SCIAMACHY mission and no drift thereafter were identified. While any possible artefacts due to constant and seasonally varying components are neutralised by the corresponding regression terms and hence are not expected to produce trend errors, the influence of the continuous drift on the resulting trends needs to be further investigated. Assuming similar pointing drifts for both limb and occultation measurements, this is done by calculating the trends for drift-corrected time series. The weighting functions of the vertical profile of O 3 with respect to the pointing offset needed to perform the correction are calculated numerically using one representative profile for each season. The corrected time series yield up to 1 % dec −1 lower O 3 trends in the 40-50 km range and practically the same results as uncorrected elsewhere.
As described in Sect. 6.4, negative O 3 trends from SCIA-MACHY observed between 30 and 38 km in the tropics are qualitatively confirmed by both EOS MLS and OSIRIS/Odin, but the comparison instruments show less negative trends than SCIAMACHY. This altitude range is known to be possibly affected by reduced sensitivity in the SCIAMACHY retrieval resulting from a transition between the retrieval information coming from UV and visible wavelengths. Previously, positive offsets in SCIAMACHY O 3 number densities compared to the datasets from other instruments were observed in the tropics around 30 km and around 40 km by Mieruch et al. (2012). Similarly, compared to a variety of satellite instruments in Tegtmeier et al. (2013), SCIA-MACHY has 10-20 % higher O 3 mixing ratios in the tropics around 10 hPa, which is close to 30 km in altitude. In addition, the QBO amplitude seen by SCIAMACHY lies above the multi-instrument average in the middle tropical stratosphere. However, the QBO amplitude, if overestimated, is not necessarily a source of error for trends as QBO signals influence mainly trend uncertainties. As a result of the inclusion of QBO terms into the regression model used in this study, trend uncertainties of SCIAMACHY are clearly reduced in the middle and lower tropical stratosphere while trend values remain practically unchanged. Furthermore, analysing longterm time series, Tegtmeier et al. (2013) found no indications for temporal instability of the SCIAMACHY instrument.
As pointed out by Jones et al. (2009) and McLinden and Fioletov (2011), O 3 trends can differ dependent on the physical units used for ozone and the vertical coordinate (VMR versus number density, pressure versus geometric height) due to possible trends in the neutral density (temperature and pressure) that is needed for unit conversion. Our investigation shows that the O 3 trends in number density units, obtained from EOS MLS measurements using their own pressure and temperature information, agree better with SCIAMACHY in the tropical 30-40 km range, while no significant changes are observed below 30 km (not shown). However, converting number density into volume mixing ratio does not alter the ozone trends for OSIRIS/Odin and SCIAMACHY. This suggests that the neutral density trends derived from ECMWF data and those obtained from MLS are significantly different at these altitudes. Both SCIAMACHY and OSIRIS do not retrieve temperature (unlike MLS), and, therefore, ECMWF operational temperatures are used in the retrieval and unit conversion. As pointed out by Adams et al. (2013), drifts in the meteorological data can also affect ozone trends when units are converted. In our study, a significant disagreement above 35 km was identified when comparing the neutral density trends obtained from ECMWF operational and ERA-Interim datasets (not shown here).

Conclusions
Rates of linear change or trends were determined for stratospheric O 3 in the 15-50 km range for the time range August 2002 to April 2012. In the tropics (20 • N-20 • S), the O 3 trends from SCIAMACHY are found to be positive in the lower stratosphere and negative in the middle stratosphere. The positive trends maximise between 20 and 30 km, are double-peaked, and maximise at values close to 5 % dec −1 . The negative trends are single-peaked and maximise at 33-34 km at values of around −10 % dec −1 . The observed trends at this altitude are about twice the trends observed by OSIRIS (this paper), MIPAS (Eckert et al., 2013), and the combined SAGE II/GOMOS .
As suggested by Nevison et al. (1999), increasing NO X may contribute both to the positive O 3 trends observed below 30 km and the negative O 3 trends observed above 30 km in the tropics. In the upper stratosphere between around 40 and 47 km, positive O 3 trends of around 5 % dec −1 are seen by SCIAMACHY from the tropics to the midlatitudes of both hemispheres. These trends are believed to be in response to declining halogens as this region is highly sensitive to halogen chemistry (WMO, 1999). Similar O 3 trends have been previously reported by Steinbrecht et al. (2006), Jones et al. (2009), and WMO (2011 in the 35-45 km range.
For the latitude band of 20 • N-20 • S, representing the tropics, O 3 trends have been compared between SCIA-MACHY, OSIRIS/Odin, and EOS MLS. In addition, a trend comparison between SCIAMACHY and SHADOZ ozonesondes has been made within the inner tropics. For the latitude bands of 60-50 • N and 50-60 • S, representing the midlatitudes, trends have been compared between SCIA-MACHY and EOS MLS. The trends from SCIAMACHY and the comparison instruments are found to be in reasonable agreement down to altitudes around 20 km. The shape of tropical O 3 trends identified from SCIAMACHY is overall well confirmed by all comparison instruments. In the lowermost tropical stratosphere below 20 km, SCIAMACHY and the comparison instruments see non-negative trends but do not agree in detail. The other instruments show higher positive trends than SCIAMACHY.

Appendix A QBO and solar cycle variability as seen by SCIAMACHY
The QBO proxies, the 10 hPa and 30 hPa Singapore winds, are phase-shifted by approximately a quarter of a period. This combination can be roughly approximated by the sum of a sine and a cosine term with an effective frequency A sin (ωt + φ) = A cos (ωt) sin φ + A sin (ωt) cos φ = (A1) = a cos (ωt) + b sin (ωt).
The phase constant is given by φ = arctan ( a b ). For t = 0, this yields or A = a sin (arctan ( a b )) .
Determining the QBO amplitude by this approach, a and b are given by the fit coefficients of the QBO proxies (cf. Eq. 3). The inferred QBO amplitude from SCIAMACHY limb O 3 is shown in Fig. A1 as a function of altitude and latitude. At equatorial latitudes, a maximum amplitude is obtained at altitudes of 25-28 km and 30-36 km. This pattern of maximum amplitude is consistent with the vertical structure of the QBO response of tropical O 3 described in Sect. 3.2. Both the spatial distribution and the strength of the inferred amplitude are in close agreement with MIPAS/ENVISAT observations as shown by Eckert et al. (2013) who used a similar approach. The maximum of SCIAMACHY QBO amplitude of around 0.6 ppmv at around 33 km is larger than MIPAS values by a factor of around 1.5. This is consistent with findings by Tegtmeier et al. (2013). Comparing QBO signatures measured by a variety of instruments, it was shown that the QBO amplitude from SCIAMACHY lies above the multi-instrument  average, while MIPAS/ENVISAT exhibits a QBO amplitude lower than average. Some altitude dependence is also discernible in the solar cycle signal of O 3 , obtained from the use of the Mg II index as solar cycle proxy, shown in Fig. A2. For the Mg II index, a change of 0.01 between solar maximum and minimum is assumed (this is considered to be a realistic estimate of the difference between the maximum of solar cycle 24 and the minimum at the end of solar cycle 23). Scaling the maximum minus minimum change of the proxy by its regression coefficient yields the corresponding solar max minus min change in O 3 . Here, it is expressed in relative units (with respect to the mean value of the underlying time series). For large parts, the solar cycle signal is significant at the 2σ level at altitudes between 35 and 45 km and insignificant below and above. Analysing a merged satellite dataset from 1984-2011, the study of Kyrölä et al. (2013) inferred a largely similar distribution of significances for the solar cycle signal. The strength of their solar cycle signal being lower than ours by a factor up to 2 may be partly explained by the difference in the analysed time span and/or scaling the solar cycle signals in relative units with respect to the constant term of their regression (not the mean value of the underlying time series).