The representation of solar cycle signals in stratospheric ozone. Part II: Analysis of global models

. The impact of changes in incoming solar irradiance on stratospheric ozone abundances should be included in climate model simulations to fully capture the atmospheric response to solar variability. This study presents the ﬁrst systematic comparison of the solar-ozone response (SOR) during the 11 year solar cycle amongst different chemistry-climate models (CCMs) and ozone databases speciﬁed in climate models that do not include chemistry. We analyse the SOR in eight 5 CCMs from the WCRP/SPARC Chemistry-Climate Model Initiative (CCMI-1) and compare these with three ozone databases: the Bodeker Scientiﬁc database, the SPARC/AC&C database for CMIP5, and the SPARC/CCMI database for CMIP6. The results reveal substantial differences in the representation of the SOR between the CMIP5 and CMIP6 ozone databases. The peak amplitude of the SOR in the upper stratosphere (1-5hPa) decreases from 5% to 2% between the CMIP5 and CMIP6 10 databases. This difference is because the CMIP5 database was constructed from a regression model ﬁt to satellite observations, whereas the CMIP6 database is constructed from CCM simulations, which use a spectral solar irradiance (SSI) dataset with relatively weak UV forcing. The SOR in the CMIP6 ozone database is therefore implicitly more similar to the SOR in the CCMI-1 models than to the CMIP5 ozone database, which shows a greater resemblance in amplitude and structure to the 15 SOR in the Bodeker database. The latitudinal structure of the annual mean SOR in the CMIP6 ozone database and CCMI-1 models is considerably smoother than in the CMIP5 database, which shows strong gradients in the SOR across the midlatitudes owing to the paucity of observations at high latitudes. The SORs in the CMIP6 ozone database and in the CCMI-1 models show a strong seasonal dependence, including large meridional gradients at mid to high latitudes during winter; such sea- 20 sonal variations in the SOR are not included in the CMIP5 ozone database. Sensitivity experiments with a global atmospheric model without chemistry (ECHAM6.3) are performed to assess the impact of changes in the representation of the SOR and SSI forcing between CMIP5 and CMIP6. The experiments show that the smaller amplitude of the SOR in the CMIP6 ozone database compared to CMIP5 causes a decrease in the modelled tropical stratospheric temperature response over the solar 25 cycle of up to 0.6K, or around 50% of the total amplitude. The changes in the SOR explain most of the difference in the amplitude of the tropical stratospheric temperature response in the case with combined changes in SOR and SSI between CMIP5 and CMIP6. The results emphasise the importance of adequately representing the SOR in climate models to capture the impact of solar variability on the atmosphere. Since a number of limitations in the representation of the SOR in the CMIP5 30 ozone database have been identiﬁed, CMIP6 models without chemistry are encouraged to use the CMIP6 ozone database to capture the climate impacts of solar variability. (SSI) (SOR) and monthly mean percent ozone 280 anomalies at select pressure levels (1, 3, 5, 10, 30hPa) for the eight CCMI-1 models considered in this study. The anomalies are deﬁned relative to the period 1960-2009. These can be compared to Figures 2 and 8 in Maycock et al. (2016), which show equivalent timeseries for SAGE II and SBUV satellite ozone measurements. The CCMI-1 models show a long-term decline in stratospheric ozone, particularly in the mid and 285 upper stratosphere. This is consistent with the impact on ozone of increasing stratospheric inorganic chlorine and bromine abundances over this period (SPARC CCMVal (2010)). At 1hPa, the trend in ozone between 1979-1997 computed by linear regression ranges from -1.9 to -2.6%decade − 1 across the models. At 3hPa, the range in trends is -4.1 to -5.1%decade − 1 . These values are within the uncertainty bounds of satellite observed ozone trends over this period (Harris et al., 2015). 290

which use a spectral solar irradiance (SSI) dataset with relatively weak UV forcing. The SOR in the CMIP6 ozone database is therefore implicitly more similar to the SOR in the CCMI-1 models than to the CMIP5 ozone database, which shows a greater resemblance in amplitude and structure to the 15 SOR in the Bodeker database. The latitudinal structure of the annual mean SOR in the CMIP6 ozone database and CCMI-1 models is considerably smoother than in the CMIP5 database, which shows strong gradients in the SOR across the midlatitudes owing to the paucity of observations at high latitudes. The SORs in the CMIP6 ozone database and in the CCMI-1 models show a strong seasonal dependence, including large meridional gradients at mid to high latitudes during winter; such sea-20 sonal variations in the SOR are not included in the CMIP5 ozone database. Sensitivity experiments with a global atmospheric model without chemistry (ECHAM6.3) are performed to assess the impact of changes in the representation of the SOR and SSI forcing between CMIP5 and CMIP6. The experiments show that the smaller amplitude of the SOR in the CMIP6 ozone database compared to CMIP5 causes a decrease in the modelled tropical stratospheric temperature response over the solar 25 cycle of up to 0.6 K, or around 50% of the total amplitude. The changes in the SOR explain most of the difference in the amplitude of the tropical stratospheric temperature response in the case with combined changes in SOR and SSI between CMIP5 and CMIP6. The results emphasise the importance of adequately representing the SOR in climate models to capture the impact of solar variability on the atmosphere. Since a number of limitations in the representation of the SOR in the CMIP5 30 ozone database have been identified, CMIP6 models without chemistry are encouraged to use the CMIP6 ozone database to capture the climate impacts of solar variability.

Introduction
Stratospheric heating rates are enhanced between the minimum and maximum phases of the approximately 11 year solar cycle through two main effects: absorption of enhanced incoming ultraviolet 35 (UV) radiation and enhanced ozone concentrations (brought about by increased photochemical production) (e.g. Penner and Chang (1978); Brasseur and Simon (1981)). These radiative changes can drive feedbacks onto stratospheric dynamics, leading to amplified signals of solar cycle variability in regional surface climate via stratosphere-troposphere dynamical coupling (e.g. Kuroda and Kodera (2002)). To understand and model the impacts of solar cycle variability on the atmosphere and cli-40 mate it is therefore necessary to account for the characteristics of spectral solar irradiance (SSI) variability and the associated solar-ozone response (SOR) (e.g. Haigh (1994)). Maycock et al. (2016) examined the SOR in a number of recently updated satellite ozone datasets.
This study focuses on the representation of the SOR in global climate and chemistry-climate models.
At a minimum, models must include a sufficiently detailed representation of both SSI and the SOR Ozone Assessment Reports typically represent atmospheric ozone in one of two ways. Chemistryclimate models (CCMs) include interactive stratospheric chemistry and explicitly simulate a SOR that is consistent with their photolysis, radiation and transport schemes provided that SSI variations 50 are adequately (i.e. with sufficiently high spectral resolution) represented. A small, but growing, number of CCMs also include the chemical effects of galactic cosmic rays and solar energetic particles, though these effects are not explicitly considered in this study. Conversely, climate models do not routinely include interactive chemistry and must therefore prescribe a predefined ozone distribution to the radiation scheme taken from observations and/or models. Thus, if models without 55 chemistry are to capture the full atmospheric response to solar variability, they must prescribe an ozone dataset that includes a representation of the SOR.
The World Climate Research Programme (WCRP) fifth Coupled Model Intercomparison Project (CMIP5) included models with and without interactive stratospheric chemistry. All CMIP5 models were recommended to prescribe SSI using the Naval Research Laboratory Spectral Solar Ir-60 radiance (NRLSSI-1) dataset (Wang et al., 2005); those without chemistry were further recommended to prescribe ozone from the Stratosphere-troposphere Processes And their Role in Climate (SPARC)/Atmospheric Chemistry and Climate (AC&C; www.igacproject.org) ozone database (Cionni et al. (2011); hereafter referred to as CMIP5 ozone database). The CMIP5 CCMs that fully resolved the stratosphere show a large variation in the amplitude and structure of the modelled SOR 65 (Hood et al., 2015). This suggests that either the models implemented SSI differently, that there are large structural differences in the representation of chemical, dynamical or radiative processes between the models, and/or that the time series are too short to derive a robust SOR.
Differences in the representation of the SOR across CMIP5 models may have contributed to the large spread (∼0.3-1.2 K) in the peak tropical stratospheric temperature response between solar min-70 imum and maximum (Mitchell et al., 2015a). Other factors could include differences in the prescription of SSI and in the accuracy of the model radiation schemes (Nissen et al., 2007;Forster et al., 2011), but the quantitative importance of any one of these factors to explain the spread in modelled solar-climate responses is unclear. As was the case in CMIP5, CMIP6 will include a mixture of models with and without stratospheric chemistry. A new SPARC/CCMI ozone database has been created 75 for CMIP6 models without chemistry (hereafter referred to as CMIP6 ozone database). It is therefore important to compare the SOR in the recommended CMIP5 and CMIP6 ozone databases, since any differences may lead to changes in the modelled responses to solar forcing between CMIP5 and CMIP6 models.
In addition to analysis of CMIP5 models (Hood et al., 2015), comparisons of the SOR in CCMs 80 have been performed through the WCRP/SPARC Chemistry Climate Model Validation Exercises (CCMVal). The CCMVal-1 and CCMVal-2 models showed a positive annual mean SOR of up to ∼2.5% peaking in the tropics between ∼3-5 hPa and a maximum tropical mean temperature response in the upper stratosphere of ∼0.5-1.1 K (Austin et al., 2008;SPARC CCMVal, 2010). Vari-3 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-477, 2017 Manuscript under review for journal Atmos. Chem. Phys. Model Initiative (CCMI-1) experiments compared to previous versions, and it is therefore pertinent to evaluate the representation of the SOR in these new simulations.
Another potentially important factor to consider for modelling is the annual cycle in the SOR, which has been identified in available satellite observations (Maycock et al., 2016). Hood et al. (2015) found that the three CMIP5 CCMs with the largest horizontal gradients in the fractional 90 SOR in the upper stratosphere in early winter showed Northern hemisphere high latitude dynamical responses to the solar cycle that compared more closely with reanalysis data. The enhancement of the SOR at high latitudes is related to coupling between ozone and dynamics and may play a role in transferring the solar cycle signal from the upper stratosphere to the troposphere.
This study evaluates both the annual mean and annual cycle in the SOR in the CMIP5 and CMIP6 95 ozone databases and compares these with results from CCMI-1 models and satellite observations from Maycock et al. (2016). In addition to the CMIP ozone databases, we also analyse the recent Bodeker et al. (2013) ozone database for climate models (hereafter referred to as Bodeker ozone database). We further perform sensitivity experiments with a global atmospheric model to quantify the impact of changes in the SOR between the CMIP5 and CMIP6 ozone databases on the atmo-100 spheric response between the minimum and maximum phases of the 11 year solar cycle. Collectively these analyses provide a comprehensive overview of the current represention of the SOR in global models and the importance of this representation for modelling the response to the solar cycle. The outline of the manuscript is as follows: Section 2 describes the data and methods used to analyse the SOR, Section 3 presents the results, and Section 4 summarises our findings. Data are analysed from eight CCMI-1 models that were available from the British Atmospheric Data Centre archive at the time the study was being prepared, and which include the minimum re-110 quirements for capturing the SOR (i.e. a prescription of SSI variability in the chemistry scheme).
The models analysed are: CCSRNIES-MIROC3.2, CESM1(WACCM), CMAM, CNRM-CM5-3, EMAC(L90), LMDz-REPROBUS-CM5 (L39), MRI-ESM1r1, and SOCOL3 (see Table 1). A detailed description of the models is given by Morgenstern et al. (2017).   (Rayner et al., 2003), well-mixed greenhouse gases, volcanic aerosols, and the NRLSSI-1 SSI dataset that was also used in CMIP5 (Wang et al., 2005). CESM1(WACCM) uses 4 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-477, 2017 Manuscript under review for journal Atmos. Chem. Phys. The CMIP5 ozone database consists of monthly mean ozone mixing ratios on 24 pressure levels spanning 1000-1 hPa for the period 1850-2100. Data are provided on a regular 5/5 • longitude/latitude grid. Ozone values are provided as a 2-D (i.e. zonal mean) field in the stratosphere (at pressures less than 300 hPa) and as a 3-D field in the troposphere, with a blending across the tropopause. The tropospheric part of the database was constructed from CCM simulations. For the stratosphere, the 140 historical part of the database (1850-2009) was constructed from observations using an MLR model (that includes solar variability as one of the independent variables) fit to SAGE I and SAGE II version 6.2 satellite data and polar ozonesondes following Randel and Wu (2007). A SOR is therefore implicitly included in the historical part of the CMIP5 ozone database that will resemble the input observations fitted with the MLR model. However, owing to the paucity of long-term ozone 145 measurements at high latitudes, the SOR was only included between ±60 • latitude. This limitation led some CMIP5 modelling groups to make alterations to the CMIP5 ozone database, including extrapolation of the SOR coefficients at ±50 • latitude to the poles using a cosine latitude weighting. The CMIP5 models known to have employed this 'Extended CMIP5 ozone database' include HadGEM2-CC (Osprey et al., 2013), MPI-ESM (Schmidt et al., 2013) and CMCC-CC (Cagnazzo,  The CMIP5 ozone database is described in full by Cionni et al. (2011) and is available from: http://cmip-pcmdi.llnl.gov/cmip5/forcing.html. Documentation of the CMIP5 models that employed the CMIP5 ozone database is given by Eyring et al. (2013).

The CMIP6 ozone database
The CMIP6 ozone database for the historical period (1850-2014) consists of monthly mean ozone 165 mixing ratios on 66 pressure levels spanning 1000-0.0001 hPa. Data are provided as a 3-D field on a regular 2.5 × 1.9 • longitude/latitude grid. The database has been constructed using output from two CCMI-1 models (CESM1(WACCM) and CMAM), which have been weighted according to an evaluation of various performance metrics for ozone (M. Hegglin, pers. comms.). The CCMs followed the REF-C1 experiment protocol with prescribed observed SSTs, sea ice, well-mixed green-170 house gas concentrations and aerosols. Observed estimates of surface emissions of NO x and other tropospheric ozone precursor gases are prescribed. The two CCMs represent SSI in their radiation and chemical schemes. Only CESM1(WACCM) includes the chemical effects of energetic particles, which means the CMIP6 ozone database will only partly capture these effects. We analyse data from the CMIP6 ozone database over the period 1960-2011. The CMIP6 ozone database was accessed 175 from: https://esgf-node.llnl.gov/projects/input4mips.
As is the case for all the CCMI-1 models, the two CCMs used to create the CMIP6 ozone database were forced with the NRLSSI-1 dataset, whereas the CMIP6 models will be recommended to use a new merged SSI dataset described by Matthes et al. (2017). The change in UV forcing between solar cycle minimum and maximum is smaller in NRLSSI-1 than in the CMIP6 solar forcing dataset.

180
Specifically, the variability in the 200-400 nm band is around 30% smaller in NRLSSI-1 than in the CMIP6 SSI dataset (Matthes et al., 2017). Sensitivity experiments with two CCMs reveal that the weaker UV forcing in NRLSSI-1 reduces the amplitude of the tropical mean SOR in the stratosphere by up ∼0.3% compared to a case forced with CMIP6 solar forcing (see Figure 7(c) in (Matthes et al., 2017)). Therefore, there will be a small inconsistency between the amplitude of the SOR captured  2008)) fitted with an MLR model including terms for equivalent effective stratospheric chlorine (EESC), a linear trend, the QBO, the El Niño Southern Oscillation (ENSO), the solar cycle, and the Mt Pinatubo volcanic eruption. We note that since the BDBP contains SAGE II v6.2 mixing ratio 195 data, this is likely to provide a strong constraint on the SOR in the tropics and subtropics because SAGE II is a relatively long-term and stable ozone record. A single MLR fit is performed for all points on a given pressure surface to enable regression coefficients to be derived for latitudes where the observations are sparse (e.g. at high latitudes). We use the Tier 1.4 product from the Bodeker ozone database, which is a spatially filled field that includes contributions from all the MLR basis 200 functions.

The multiple linear regression (MLR) model
The SOR is analysed using an MLR model as described by Maycock et al. (2016). Briefly, the zonal mean ozone data are deseasonalised by removing the long-term monthly mean at each latitude and pressure level. As in past studies, we then perform an MLR analysis on the timeseries of monthly 205 mean anomalies at each location, O 3 (t), to diagnose the solar cycle component: where r(t) is a residual. The annual-mean SOR is calculated by regressing all months as a single timeseries. The monthly SOR is calculated by regressing interannual timeseries of each month separately. The monthly basis functions in Equation 1 are the F10.7cm radio solar flux, the CO 2 concen-210 tration at Mauna Loa, the equivalent effective stratospheric chlorine (EESC), the Nino 3.4 index to represent ENSO, and the volcanic aerosol surface area density (SAD V OLC ) averaged between ±30 • latitude and 15-35 km. For those CCMI-1 models and ozone databases that include QBO variability (see Table 1 which assumes the residuals r(t) have the form: 240 where a and b are constants and w(t) is a white noise process. This is identical to the approach employed in Maycock et al. (2016) and the recent SPARC SI 2 N analysis of ozone trends (Tummon et al., 2015;Harris et al., 2015). No autocorrelation term for the residuals is included in the analysis of the SOR annual cycle because the residuals for any given month are approximately uncorrelated from year-to-year. It is known that the ECHAM6.3 radiation code does not cover wavelengths below 200 nm and therefore the important Schumann-Runge bands and Lyman-alpha lines of ozone are not captured (Sukhodolov et al., 2014). This results in a too weak radiative response to the imposed solar forcing particularly in the mesosphere. Therefore we focus the analysis on the temperature response in the stratosphere where most of the absorption occurs at higher wavelengths and the performance is 260 comparable to models with a more comprehensive radiative code (Sukhodolov et al., 2014).
3 Results The CCMI-1 models show a long-term decline in stratospheric ozone, particularly in the mid and 285 upper stratosphere. This is consistent with the impact on ozone of increasing stratospheric inorganic chlorine and bromine abundances over this period (SPARC CCMVal (2010)). At 1 hPa, the trend in ozone between 1979-1997 computed by linear regression ranges from -1.9 to -2.6 % decade −1 across the models. At 3 hPa, the range in trends is -4.1 to -5.1 % decade −1 . These values are within the uncertainty bounds of satellite observed ozone trends over this period (Harris et al., 2015). The results in Figure 3 are broadly consistent with previous analyses of CCMs (Austin et al., 2008;SPARC CCMVal, 2010). The main exception to this is the absence in the multi-model mean of a significantly enhanced SOR in the tropical lower stratosphere. Figure 4(d) in Austin et al. (2008) shows a multi-model mean SOR for the CCMVal-1 models of around 5% per 130 SFU at ∼50 hPa, 305 as compared to around 1% in the CCMI-1 multi-model mean in Figure 3(i). However, there was large intermodel spread in this signal across the CCMVal-1 models and the multi-model mean SOR was dominated by strong responses in a few models that only ran for a short period  over which aliasing with the effects of volcanic aerosols can be significant (Chiodo et al., 2014). Since the CCMI-1 models are analysed for a longer period , this is a plausible explanation 310 for the differences in tropical lower stratospheric SOR between the CCMI-1 and CCMVal-1 model responses.

The SOR in CCMI-1 models
Outside of the tropics there are larger inter-model differences in the fractional SORs in Figure 3, with a range in the amplitude, sign and level of statistical significance of the diagnosed SOR in both hemispheres. One consistent feature across many of the models appears to be an enhanced SOR 315 in the Southern hemisphere high latitude lower stratosphere, which is evident in the multi-model mean. The annual cycles in the SOR in the individual models (see Supplementary Information) show that the strong gradients in the SOR at high latitudes found in some of the models tend to be more pronounced in the winter seasons. This behaviour, which is also seen in some satellite ozone datasets (Maycock et al., 2016), cannot be understood from photochemical processes alone 320 and must therefore be related to stratospheric circulation changes (Kuroda and Kodera, 2002). Such changes in ozone at high latitudes will be associated with a radiative perturbation that could lead to feedbacks onto circulation; however, the quantitative importance of such ozone-radiative feedbacks in the CMIP6 database; the latter being, as expected, similar to the long-term ozone trends in the CCMI-1 models shown in Figure 2. At 3 hPa, the CMIP5 ozone database shows a larger long-term trend by around a factor of two compared to the Bodeker and CMIP6 databases. Thus, the CMIP6 models that use the recommended CMIP6 ozone database might be expected to show a smaller cool-345 ing of the upper stratosphere over recent decades compared to an equivalent simulation using the CMIP5 database, owing to the smaller trend in ozone.  The SOR in the CMIP5 ozone database, Figure 5(b), shows a very similar structure to that found in SAGE v6.2 mixing ratios (see Figure 4(d) of Maycock et al. (2016)), consistent with those data 370 forming the backbone for the historical portion of the dataset (Cionni et al., 2011). Note that the MLR fitting was applied separately at each latitude band in the construction of the CMIP5 database, and this likely explains why the horizontal structure of the SOR is more heterogeneous than in the Bodeker ozone database. The sharp cut-offs in the SOR at ±60 • latitude are spurious and result from a lack of data points to constrain the SOR at high latitudes. As described in Section 2.1.2, the 375 Extended CMIP5 ozone database, Figure 5(c), applied a simple extrapolation to introduce a SOR in the extratropics. This structure, which shows a positive SOR in the northern extratropics and a negative SOR at pressures greater than ∼5 hPa polewards of 60 • S, is likely to be subject to considerable uncertainties owing both to the large uncertainties in the observed SOR at these latitudes (Maycock et al., 2016) and the fact that the high latitudes are filled using a simple extrapolation method.

380
Figure 5(d) shows the SOR from the CMIP6 ozone database. The amplitude of the SOR is around 1-2% in the upper stratosphere consistent with the CCMI-1 results in Figure 3. This is 2-3 times smaller, and is considerably smoother in latitude, than the SOR in the CMIP5 ozone database. In the lowermost tropical stratosphere, the CMIP6 database shows a positive SOR of up to ∼3% in the southern tropics. The Bodeker database, Figure 5(a), also shows a strong positive SOR above 385 the tropical tropopause although the structure is considerably less smooth in latitude. An enhanced SOR in the tropical lower stratosphere has been identified in satellite observations, albeit with large uncertainties (Gray et al., 2009;Austin et al., 2008;Soukharev and Hood, 2006;Maycock et al., 2016). It has been hypothesised that this feature may be dynamically forced by a weakening in the Brewer Dobson circulation between solar cycle minimum and maximum. However, some of 390 the CCMI-1 models in Figure 3 do not show an enhanced SOR in the tropical lower stratosphere, suggesting this feature is not captured consistently amongst models and ozone datasets.
To further compare the structure of the SOR in the tropics, Figure 6 shows vertical profiles of the annual and tropical (30 • S-30 • N) mean SOR in the CCMI-1 models and climate model ozone databases. The range in the best estimate SOR across the CCMI-1 models is shown in dark grey 395 shading, along with ±1 standard deviation of the intermodel spread. Observations from the SBU-VMOD VN8.6 (Frith et al., 2014) (black) and the SAGE-GOMOS 1 dataset (Kyrölä et al., 2015) (blue) are also shown (see Maycock et al. (2016) for details).
In the tropical lower stratosphere, the statistical uncertainties in the SOR are much larger than in the rest of the profile, and the best estimate SOR ranges from a small negative signal in the CMIP5 400 ozone database to 6% in the Bodeker ozone database. The SOR in the CMIP6 database shows a significant tropical mean SOR of 2% at 80 hPa, which is, as expected, within the range of the spread in the CCMI-1 model signals. There is therefore a distinct difference in the representation of the SOR in the tropical lower stratosphere in the CMIP5 and CMIP6 ozone databases, which may be important for the modelled atmospheric response to solar variability in CMIP5 and CMIP6 models 405 (see Section 3.4). Figure 6 further confirms that the two climate model ozone databases that include SAGE II v6.2 mixing ratio data (CMIP5 and Bodeker), show a substantially larger tropical mean SOR in the upper stratosphere. This is consistent with Maycock et al. (2016) who concluded that the SAGE II v6.2 mixing ratio data showed a considerably larger SOR in the tropical upper stratosphere compared to SAVE II v7.0 mixing ratios and SBUV based datasets.  Figure 7 shows the monthly mean SOR in the Extended CMIP5 ozone database and Figure 8 shows the same for the CMIP6 ozone database. The SOR in the CMIP5 database has a fixed structure and constant amplitude in all months; the small annual 415 cycle in the fractional SOR amplitude arises purely from the annual cycle in background ozone concentrations. There are well understood photochemical arguments for why the structure of the SOR is expected to track the position of the Sun through the year (Haigh, 1994). Furthermore, the coupling between ozone and stratospheric dynamics may lead to variations in the SOR at high latitudes in some months due to the formation in winter of the polar vortices and their subsequent 420 break-up in spring (Hood et al., 2015). For these reasons a complete absence of seasonal variation in the SOR is unrealistic. In contrast, the SOR in the CMIP6 ozone database, Figure 8, shows greater seasonal variation. Locally enhanced signals in the SOR are found in the southern high latitudes and in the northern high latitudes in winter, which may be linked to variations in the strength of the polar vortex (Kuroda and Kodera, 2002). Thus, the seasonal variability of the SOR in Figure   425 8 is likely to be more representative of the real atmosphere than the complete absence of seasonal variability in Figure 7. However, there are quantitative differences between the SOR annual cycle in the CMIP6 ozone database and that estimated from satellite observations (see Figure 13 of Maycock et al. (2016)). These differences may result from uncertainties in estimating the SOR from relatively short observational records, from errors in the representation of the SOR in the models used to 430 construct the CMIP6 ozone database, or a combination of factors. Thus there is a need for continued satellite measurements in order to reduce the large uncertainties in the observed SOR, particularly on seasonal timescales, and to provide a more stringent reference for ozone databases and models.

Atmospheric impact of change in SOR between CMIP5 and CMIP6 ozone databases
We now explore the atmospheric impacts of the differences between the SOR in the CMIP5 and 435 CMIP6 ozone databases using the ECHAM6.3 model sensitivity experiments described in Section 2.3. Figure 9 shows the tropical average annual mean temperature differences in the four solar cycle perturbation simulations with respect to the control simulation. Note that the tropospheric temperature responses in all simulations are small because the model includes fixed SSTs and therefore the troposphere does not fully adjust to the imposed solar forcing (e.g. Misios et al. (2016)).

440
The experiments performed to capture the total (i.e. SSI + SOR) solar cycle impact (dashed lines) show considerable differences in the tropical mean stratospheric temperature response between the recommended CMIP5 and CMIP6 forcings. In the CMIP5 case, the maximum temperature response is around 1.25 K near the stratopause, which can be compared to a much smaller response to the CMIP6 solar forcing inputs of 0.7 K. The SOR-only sensitivity experiments (solid lines) reveal that 445 much of the difference in the total temperature response can be attributed to the differences in the SOR between the CMIP5 and CMIP6 ozone databases. The SOR in the Extended CMIP5 ozone database induces a peak tropical temperature response of 0.9 K (solid red), which is three times larger than the maximum response to the SOR in the CMIP6 ozone database (solid blue). In addition to the marked differences in the maximum temperature response, there are also distinct differences in 450 vertical structure. In the CMIP5 case, there is a stronger vertical gradient in the temperature response, which can be attributed to the highly peaked structure of the SOR in the CMIP5 database at the stratopause compared to the smoother vertical structure of the SOR in the CMIP6 ozone database (cf. Figures 5(c) and 5(d)). The simulation forced with the SOR from the CMIP6 ozone database also shows a small secondary peak in tropical lower stratospheric temperature of ∼0.3 K due to the 455 presence of a locally enhanced SOR of ∼3%, which is not present in the CMIP5 ozone database. The results show that the change in the representation of the SOR between the recommended CMIP5 and CMIP6 ozone databases induces a much larger difference in the temperature response between solar cycle minimum and maximum than do changes in the recommended SSI forcing (see also Matthes et al. (2017)). models that used the CMIP5 ozone database exhibit a markedly larger temperature response near the tropical stratopause, with a stronger vertical gradient, compared to the models with interactive chemistry. One might therefore anticipate that the difference in the stratospheric temperature response between solar cycle minimum and maximum for models with and without interactive chemistry will be smaller in CMIP6 than was found in CMIP5 owing to the fact that the SOR in the 470 14 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-477, 2017 Manuscript under review for journal Atmos. Chem. Phys.

Conclusions
Changes in stratospheric ozone concentrations make a significant contribution to the atmospheric response to changes in incoming solar radiation over the 11 year solar cycle (e.g. Shibata and Kodera  We also analyse the SOR in three ozone databases that are prescribed in climate models without interactive chemistry: the Bodeker et al. (2013) Tier 1.4 ozone database and the CMIP5 ozone database (Cionni et al., 2011), which are both based on regression models fit to observations, and the CMIP6 ozone database, which is created from historical simulations from two CCMs (CESM1 (WACCM) and CMAM).

485
The CCMI-1 models simulate a SOR with a peak amplitude of 1-2% in the upper stratosphere (∼3-5 hPa). This is smaller than the SOR found in SAGE II v6.2 mixing ratio data and is more consistent with results from SAGE II v7.0 and SBUV satellite datasets (Maycock et al., 2016).
Some of the CCMs show larger fractional SORs in the high latitude winter stratosphere, which are strongly influenced by dynamical processes, although the amplitude and structure of these features 490 tend to be less consistent across the models than the response in the tropical upper stratosphere. In addition, some of the models, in particular CMAM, LMDz-REPROBUS-CM5, MRI-ESM1r1 and SOCOL3, show an enhanced SOR in the tropical lower stratosphere, which has been identified in some satellite ozone datasets (Maycock et al., 2016). As expected, the SOR in the CMIP6 ozone database generally resembles that in the CCMI-1 models, both in terms of its broad structure and 495 magnitude and the fact that it includes seasonal variability. We note that since the UV variability in the SSI forcing dataset used in the CCMI-1 models is relatively weak, the SOR in the CMIP6 ozone database is smaller than would be simulated in a CCM forced with the CMIP6 SSI dataset, which includes larger UV variability (Matthes et al., 2017).
There are stark differences between the SOR in the CMIP6 ozone database and those found in the 500 CMIP5 and Bodeker ozone databases. In particular, the peak amplitudes in the tropics are substantially larger (5%) in the latter databases compared to in the CMIP6 database (1.5%). This is because those databases are derived from observations that include SAGE II v6.2 mixing ratios, which as previously mentioned exhibit a larger SOR than found in other satellite ozone datasets (Maycock et al., 2016). In addition to differences in the peak magnitude of the SOR, there are also marked differences in the spatial structure of the SOR amongst the ozone databases. The CMIP5 database showed spurious large horizontal gradients in the SOR across the extratropics, which were reduced through implementation of a simple poleward extrapolation in the Extended CMIP5 ozone database (Schmidt et al., 2013;Osprey et al., 2013). Furthermore, while the CMIP6 database implicitly includes seasonal vari-510 ations in the SOR, as simulated by the CCMs used to construct the database, the CMIP5 database has a fixed annual mean SOR in all months, which is likely to be unrealistic. the recommended SOR on tropical stratospheric temperatures is many times larger than the separate impact (i.e. without ozone feedbacks) of changes in the recommended SSI forcing between CMIP5 and CMIP6. The results indicate that differences in the representation of the SOR amongst CMIP5 models is likely to be a major explanatory factor for the large spread in the stratospheric temperature responses to the solar cycle in CMIP5 models (Mitchell et al., 2015a). The broader relevance of dif-530 ferent representations of the SOR for atmospheric dynamics and regional surface climate responses to the solar cycle remains to be explored.
Substantial uncertainties remain in various factors related to understanding the SOR, which present challenges for including these effects in global models. Key issues include: outstanding large uncertainties in the SOR derived from observations (Maycock et al., 2016); outstanding uncertainties in 535 the characteristics of SSI variability (Ermolli et al., 2013;Haigh et al., 2010;Dhomse et al., 2016;Matthes et al., 2017); uncertainties in the ability of models to represent the effects of SSI variability on atmospheric radiation, photochemistry and dynamics (Forster et al., 2011;Sukhodolov et al., 2016;Hood et al., 2015;Matthes et al., 2017); and uncertainties in the magnitude of the observed temperature response to the solar cycle (Ramaswamy et al, 2001;Mitchell et al., 2015b). Despite 540 these various issues, information about the observed SOR has been used to exclude implausible scenarios for SSI variability (Ball et al., 2016) and this offers hope for further advances in understanding 16 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-477, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 31 May 2017 c Author(s) 2017. CC-BY 3.0 License. the SOR in the future. Improved physical understanding and constraints for model performance rely on long-term high quality observational datasets and it is therefore vitally important that satellite measurements of stratospheric ozone continue in the future.