SPARC Assessment of Chemistry Climate Models Chapter 8 ...reichler/publications/papers/...1 1 SPARC Assessment of Chemistry Climate Models 2 Chapter 8: Natural Variability of Stratospheric

1

SPARC Assessment of Chemistry Climate Models 1 Chapter 8: Natural Variability of Stratospheric Ozone 2 3 4 Lead Authors: Elisa Manzini, Katja Matthes 5 6 Co-Authors: 7 Christian Blume, Greg Bodeker, Chiara Cagnazzo, Natalia Calvo, Andrew Charlton-8 Perrez, Anne Douglass, Pier Giuseppe Fogli, Lesley Gray, Junsu Kim, Kuni Kodera, 9 Markus Kunze, Cristina Pena Ortiz, Bill Randel, Thomas Reichler, Gera Stenchikov, 10 Claudia Timmreck, Matt Toohey, and Shigeo Yoden 11 12 Executive Summary 13 14 Key Findings 15

16 Annual Cycle: 17

18 • The vertical and latitudinal distribution of the annual cycle in stratospheric zonal 19

monthly mean ozone is quite well represented in the models. The only outliner 20 found is a model with a top at pressure higher than 1 hPa (i.e., a “low top” model). 21 However, in the lower stratosphere a few models do not reproduce the 22 anthropogenic induced annual cycle (polar ozone depletion) that dominates the 23 annual ozone evolution in the SH, later winter and spring. 24

25 • The annual cycle in zonal mean column ozone is quite well represented in the 26

models. However, as noted above for the vertical and latitudinal distribution, few 27 models fail in representing the Antarctic ozone depletion typically emerging in the 28 climatological annual cycle of the last decades (since 1980 to present). 29

30 • In the comparisons with HALOE, the CCMVal-2 models show a larger spread in 31

their response to the annual cycle, in the NH and SH spring, upper troposphere 32 and lower stratosphere, than the CCMVal-1 models. 33

34 Interannual polar variability: 35

36 • The observed annual cycle in column ozone variability is well reproduced by all 37

models, in the sense that all show a minimum in variability in the respective 38 summer seasons. However, in the NH active period most of the models 39 underestimate the interannual polar ozone variability, suggestive of a common 40 bias. In the SH instead, the models both over- and under-estimate ozone 41 interannual variability. 42

43 • Most models capture the connections between the dynamical processes 44

responsible for the interannual polar ozone variations and the ozone response. 45 However, in the NH the low sensitivity of ozone to temperature in the majority of 46

2

models is in contrast with the assessment of the ozone sensitivity to wave drag 1 diagnosed from the heat fluxes. 2

3 Solar Cycle: 4 5

• The solar cycle in column ozone is well represented by the models although with 6 a large amplitude spread. While the direct solar response in temperature and 7 ozone in the upper stratosphere is well represented, the vertical structure in the 8 tropics for pressure higher than 10 hPa varies considerably among the models and 9 between models and observations. Large uncertainties exist in the middle to lower 10 stratosphere possibly due to non-linear interactions of the solar cycle with QBO, 11 ENSO and volcanic signals. Short observational records contribute to the 12 uncertainties. 13

14 • The latitudinal representation of the solar response in column ozone shows 15

improvements from earlier studies. A large spread occurs at mid to high latitudes 16 due to large interannual variability and limits the discussion of extratropical 17 signals. 18

19 • It is very difficult to exactly assign the performance of individual models to the 20

radiation, photolysis, or transport scheme of the models. 21 22

QBO: 23 24

• The modeling of the QBO in the CCMVal-2 models is judged to be still at a 25 primitive stage. A very few models are capable to spontaneously simulate the 26 salient features of QBO ozone variations. 27

28 • The nudging of the QBO induces substantial errors in QBO ozone variations, 29

notably in the amplitude of the column ozone. 30 31

ENSO: 32 33

• The ENSO signal in ozone that emerges from the CCMVal-2 models is consistent 34 with expectations from previous work. However, the uncertainty in the observed 35 signal prevents any firm conclusion on the model performances on this factor of 36 variability. 37

38 • In the NH spring, the relationship between temperature and ozone for the modeled 39

ENSO (warm events) is consistent with the interpretation that interannual 40 variability is the major cause that hampers the identification of a significant 41 ENSO signal in ozone. 42

43 Volcanoes: 44

45 • The volcanic signal in ozone differs considerably between models and depends on 46

3

the method how the direct effect of volcanic aerosols on the radiative transfer of 1 the stratosphere is represented. 2

3 • Post-eruption changes in total column ozone are well correlated with changes in 4

lower stratospheric ClO. It thus appears that while most models use a common 5 aerosol SAD data set to drive anomalous post-eruption chemistry, the models 6 display differing degrees of sensitivity to those aerosols, leading to differing 7 amounts of chlorine activation and associated ozone loss. 8

9 Recommendations 10 11 Annual cycle: 12 13

• Model tops above 1 hPa 14 15

• Assess the impact of the heterogeneous polar ozone chemistry leading to ozone 16 depletion on the annual cycle in ozone (distribution in the lower stratosphere and 17 column amount). 18

19 Interannual polar variability: 20 21

• Investigate the sensitivity of the SH large scale dynamics to the parameterization 22 of non-orographic gravity wave drag and its implication for the numerically 23 resolved interannual variability in ozone. 24

25 • Investigate which modeling factors may affect the temperature and ozone 26

responses to wave driving and induce biases in the consequent temperature - 27 ozone relationship. These factors can be purely technical (ozone and temperature 28 transport numerical schemes are usually different) and/or related to the 29 representation of chemical processes. 30

31 Solar Cycle: 32 33

• Models should use spectrally resolving solar radiation changes (and a suitable 34 radiation and photolysis code). More detailed intercomparison of radiation and 35 photolysis codes as well as sensitivity studies to understand the complex non-36 linear interactions of the solar signal with other natural variability signals are 37 needed. 38

39 QBO: 40 41

• Advance the development of the modeling of the QBO is called for. This 42 modeling problem is outstanding. It is an open issue that cannot be tackled by 43 assimilating selected properties of the observed QBO in the CCMs. 44 45 46

4

ENSO: 1 2

• Further research of the influence of ENSO on ozone variability is called for, 3 especially in the interpretation of the observed variations. 4

5 Volcanoes: 6 7

• Advance the parameterization of volcanic effects and understand observed 8 variations in response to volcanic eruptions in more detail. 9

10 11 8.0 Introduction 12 13 Stratospheric ozone is known to vary in response to a number of natural factors, such as 14 the seasonal and the 11-year cycles in solar irradiance, the QBO, ENSO, variations in 15 transport associated with large-scale circulations (i.e., Brewer Dobson circulation) and 16 dynamical variability associated with the annular modes. Aerosols from volcanic 17 eruptions can also affect stratospheric ozone, although their effects depend on the 18 background atmospheric composition. Ozone observations have demonstrated variations 19 on a large number of spatial and temporal scales. To quantify the impact of anthropogenic 20 perturbations of the ozone layer, and to make a reliable prediction of future ozone 21 abundances, it is therefore necessary to understand and to quantify the underlying natural 22 ozone variations. 23 24 The goal of this Chapter is to evaluate how well CCMs simulate natural stratospheric 25 ozone variability, based on our current knowledge about links between ozone variations 26 and natural forcings. The assessment of the modeled natural ozone variations is reported 27 here, with special attention paid to the CCMs’ overall performance. Fundamental 28 questions are: Do the models simulate realistic ozone variations? Which processes are 29 key in determining natural variability in stratospheric ozone? Do models which reproduce 30 natural variations in ozone, do so because these key processes are well simulated? The 31 response to these questions will serve to judge if the models simulate natural ozone 32 variations for the correct reasons. 33 34 The assessment of the modelled natural ozone variations from this chapter, together with 35 the modelled trends from Chapter 9, will feed into another fundamental question: To 36 what extent do we trust a trend predicted by a model that does not adequately represent 37 natural ozone variations? This question addresses the problem of overconfidence that 38 might be implied by models lacking natural ozone variations. 39 40 The relative importance of the different sources of natural variability in stratospheric 41 ozone is assessed by means of a multiple linear regression. When possible, the connection 42 between the sources of natural variability and ozone is addressed by analyzing the 43 processes that determine it. Systematic inter-comparisons of ozone as simulated by the 44 CCMs as well as individual model studies are assessed. The evaluation of the REF-B1 45 simulations of the ongoing CCMVal-2 project makes up the core of the assessment, while 46

5

the discussion of CCMVal-2 results with respect to the results obtained from CCMVal-1 1 simulations is carried out when possible. 2 3 This chapter aims at synthesizing the results of parts A and B with respect to natural 4 ozone variations. We consider trends related to anthropogenic ozone depletion in our 5 analysis in order to address the problem of how ozone natural variations are modeled but 6 leave the discussion of the effects of these trends to the following chapter (Chapter 9). 7 8 8.1 Data and Methodology 9 10 8.1.1 Data 11 12 In the following, a brief description is provided of the key ozone and temperature 13 observations employed to validate and assess the ability of the CCMs to simulate 14 observed variability. To account for the spread between available observational data sets 15 and individual estimates of measurement errors, several such data sets have been used. 16 17 Column ozone 18 19 The ground-based data set from Fioletov et al. (2002) (ftp://ftp.tor.ec.gc.ca/Projects-20 Campaigns/ZonalMeans/) was combined with the merged satellite total ozone data set 21 (TOMS/SBUV) from the Total Ozone Mapping Spectrometer (TOMS) and Solar 22 Backscatter Ultraviolet 2 (SBUV/2) instruments (Stolarksi and Frith 2006; http://acdb-23 ext.gsfc.nasa.gov/Data_services/merged/) to obtain a long-term data set for the period 24 1964-2008. Satellite data were employed where available and ground-based data were 25 used to fill the gaps. This data set is referred to as “TOMS+gb”. The NIWA combined 26 total column ozone data base for the shorter period 1979-2008 (updated from Bodeker et 27 al., 2005) is also employed, hereafter referred to as NIWA-column. 28 29 Ozone profiles 30 31 Several ozone profile data sets are employed. The Randel&Wu-3D data set (Randel and 32 Wu, 2007) is based on output from a regression model applied to ozone anomalies. The 33 regression model includes a decadal trend (EESC), the QBO, 11-year solar cycle and an 34 ENSO basis function, which are fitted to SAGE I and II satellite ozone anomalies. The 35 regression output is added to a seasonal mean, zonal mean, vertically resolved ozone 36 climatology (Fortuin and Kelder, 1998). The NIWA-3D data set is based on satellite 37 (SAGE I and II, POAM II, and III, HALOE) and ozonesonde profiles where regression 38 constrained interpolation has been used to produce a gap free data set (Hassler et al., 39 2009). 40 41 For the seasonal cycle studies of ozone, the Microwave Limb Spectrometer (MLS) data 42 from the NASA’s Aura satellite (Waters et al., 2006; Froidevaux et al., 2008) is also 43 employed. The MLS instrument has made global measurements nearly every day since 44 August 2004 and is therefore ideal for examining the seasonal cycle at various pressure 45 levels. Monthly averaged values of MLS ozone are computed for 6-degree latitude bins. 46

6

The ozone climatology for the period 1991-2002 from the Halogen Occultation 1 Experiment (HALOE) on board the Upper Atmosphere Research Satellite (UARS; 2 Russell et al., 1993) is also used. Data after September 2002 have not been included 3 because of the unusual major warming in the Antarctic in 2002 and because the 4 observations have been less frequent since 2002 (Grooß and Russell, 2005). 5 6 Temperature 7 8 We use different temperature data sets to compare the temperature signals in the CCMs 9 with available observations: 1) SSU (Stratospheric Sounding Unit) temperature data for 10 the middle and upper stratosphere (Randel et al., 2009), 2) the Radiosonde Innovation 11 Composite Homogenization (RICH) data set that uses the ERA40 reanalysis to identify 12 break points, which are then adjusted using neighboring radiosonde observations in the 13 lower stratosphere and troposphere. (Haimberger et al. 2008; 14 http://www.sparc.sunysb.edu/html/updated_temp.html), and 3) the ERA40 reanalysis 15 temperature data (Uppala et al., 2004). We use the reanalyses to allow comparison of 16 similar time and spatial coverage as in the CCMs, keeping in mind the uncertainties 17 related to possible spurious trends in this data set (for a discussion see e.g., Randel et al., 18 2009). 19 20 8.1.2 Multiple Linear Regression Analysis 21 22 Multiple linear regression analyses (MLR) is a commonly used method to assess the 23 relative contributions of different drivers of variability in geophysical time series e.g. 24 near global total column ozone (WMO, 2006). Here we compare results from a MLR 25 analysis applied to monthly ozone and temperature fields from the REF-B1 simulations 26 of CCMVal-2 with results from an identical analysis of the appropriate observational 27 datasets described above. Although the focus is on sources of natural variability (annual 28 and semiannual cycle, solar cycle, QBO, ENSO, and volcanoes), a secular term (EESC; 29 equivalent effective stratospheric chlorine; Daniel et al. [1995] for ozone and a linear 30 trend term for temperature) is also required to account for the substantial trend in ozone 31 and temperature over the period examined. 32 33 The MLR analysis is based on the method described in Bodeker et al. (2001) to model a 34 time dependent variable, e.g. ozone: 35 36

37 38 39 The first term in the regression model (βoffs coefficient times the offset basis function) 40 represents a constant offset and, when expanded in a Fourier expansion, represents the 41 mean annual cycle. In this case, with three Fourier pairs (N=3 in the equation above), the 42 annual cycle is modeled as a summation of 12, 6 and 4 monthly cycles each of variable 43

7

phase. All basis functions are detrended except for the EESC and volcano basis functions 1 and the offset is removed except for the volcano basis function. The sensitivity of the 2 MLR basis functions to different numbers of Fourier pairs was tested carefully. The three 3 Fourier pair expansion for the EESC basis function was chosen to account for the strong 4 seasonal cycle in the effect of EESC on ozone, particularly in the polar regions. For all 5 other basis functions the results are not significantly influenced by changing the number 6 of Fourier expansions. 7 8 The EESC basis function represents the total halogen loading of the stratosphere effective 9 in ozone depletion, appropriately weighted by the mean age of air (Age 3.0 years Width 10 1.5 years). For most of the CCMs, the EESC has been calculated using the formula 11 suggested by Newman et al. (2007): Cly + 60 * Bry and the global monthly mean values 12 at 50hPa, and is referred to as ESC in Eyring et al. (2007). Some CCMs do not provide 13 Cly and/or Bry and therefore for these CCMs the observed EESC is used (CNRM-ACM, 14 E39C, Niwa-SOCOL, UMUKCA-METO, and UMUKCA-UCAM). The EESC term 15 (βEESC coefficient) represents the anthropogenic part of the signal and is not discussed 16 until the next chapter (Chapter 9). Note that we did not include an additional linear trend 17 term for the ozone regression as all long-term secular changes are assumed to be captured 18 by the EESC basis function. Note also that we exchange the EESC term with a linear 19 trend term for the temperature regression. 20 21 The QBO basis function is specified by the monthly mean 50 hPa zonal wind (except for 22 AMTRAC3 10 hPa and UMSLIMCAT 30 hPa is used) for each individual model 23 realization. Since the phase of the QBO varies with latitude and altitude a second QBO 24 basis function is included, which is orthogonal to the first, as described by Austin et al. 25 [2008]. For the CCMs in Group A of Table 8.3.1, the QBO basis function is neglected, 26 given their lack of interannual variability in the tropics (see Figure 8.3.1). 27 28 The observed Nino 3+4 SST index is used for the ENSO basis function without time 29 shift. The F10.7cm radio flux is employed for the 11-year solar cycle basis function. The 30 volcanic aerosol basis functions for Agung, El Chicon and Pinatubo are taken from 31 Bodeker et al. (2001). R is the residual. To account for the autocorrelation in the 32 residuals, an autoregressive model has been used: first a fit to measurements has been 33 performed with the calculation of the residual. Then the autocorrelation coefficient was 34 calculated after equation 6 in Bodeker et al. (1998) and used to transform the basis 35 functions. Afterwards the MLR analysis is applied a second time to the original data 36 series, the fit now includes the autocorrelation of the basis functions. 37 In summary, only the QBO and the EESC basis functions are dependent on the model 38 simulations. All the other basis functions are common to the MLR analyses of both the 39 CCM and observational time series. 40 41 In Figure 8.1 the contribution of the various natural as well as anthropogenic contribution 42 to global (60S-60N) column ozone variations is shown for the combined TOMS+gb data 43 set. The RHS of Figure 8.1 shows that the observed decrease in column ozone is almost 44 completely explained by the increase in the atmospheric chlorine loading. Natural 45 variability represents only a small but non-negligible variability component (LHS Figure 46

8

8.1). The annual cycle dominates the natural variability with an amplitude of ~12 DU, 1 followed by the 11-year solar cycle with ~6 DU between solar maximum and solar 2 minimum, the QBO with ~4 DU between maximum QBO easterlies and westerlies, a 3 small component associated with ENSO of ~1 DU, and the volcanic contribution which 4 has distinct and unevenly distributed contributions of up to 6 DU. Note that the residual, 5 especially before the satellite era is relatively large (up to +/- 5 DU) and we can only 6 speculate that this has to do with the data quality. Also we would like to emphasize that 7 we use a linear regression model for highly non-linear processes in the atmosphere. In 8 other words the residual represents to some part also the failure of a linear regression 9 analysis to account for non-linear processes in the atmosphere. 10 11 The results of the MLR analysis are presented in the relevant following subsections (8.2-12 8.7), together with process oriented studies. For most CCMVal-2 models the whole time 13 series from 1960 to 2004 is considered (although some only provide data up to 2000, e.g. 14 EMAC). Comparisons with observations are also described, employing data for the same 15 time period (1960-2004) or only data from the satellite era (1979-2004) are employed, as 16 appropriate. In these cases, the sensitivity of the MLR analysis to the employed time 17 period has been tested (but is not shown); unless otherwise stated, the essential results are 18 not substantially affected by the shortened period, although the amplitude of the signal is 19 usually larger. 20 21 8.2 Annual Cycle in Ozone 22 23 Pronounced variations in stratospheric ozone are caused by annual variations in transport 24 and photochemistry. The transport variations are driven by dynamical processes (Chapter 25 4 and Chapter 5) and can affect ozone either directly or indirectly (through changed 26 transport of ozone-depleting substances). Photochemical production of ozone depends on 27 annual variations in the solar irradiance (Chapter 3 and Chapter 6). The resulting annual 28 cycle in column ozone is characterized by (a) low amounts in the tropics year-round, (b) 29 maxima in the spring of Northern hemisphere high latitudes and Southern hemisphere 30 middle latitudes, and (c) larger hemispheric-mean amounts in the Northern versus the 31 Southern hemisphere. This annual evolution of column ozone reflects the dominant 32 influence of transport processes on lower stratospheric ozone. This section focuses on 33 how the annual cycle in stratospheric ozone and column ozone is represented in the 34 CCMs, and also touches on the evaluation of global ozone biases. 35 36 8.2.1 Annual cycle at selected locations in the stratosphere 37 38 The photochemical timescale for ozone varies seasonally as a function of latitude and 39 pressure. In the lower stratosphere middle latitudes, the timescale is long and the seasonal 40 cycle is largely controlled by transport. In the upper stratosphere, the timescale is short 41 and the ozone mixing ratio reflects a near balance between production and loss. Since the 42 timescales for transport and for photochemical processes both vary seasonally, in some 43 parts of the stratosphere both types of process contribute to the stratospheric 44 concentration of ozone. For example, in the northern latitude winter stratosphere transport 45 processes control the seasonal build-up of ozone through descent at the edge of the vortex 46

9

and this is then moderated at high latitudes by chemical loss associated with polar 1 processes. In summer transport effects are minimal and the photochemical time scale 2 decreases from several years to 30 days or less, producing a summer minimum that varies 3 little from year to year. 4 5 In Figure 8.2.1 the annual cycle in ozone mixing ratios simulated by 15 CCMs is 6 compared with MLS observations. At 1 hPa the time evolution of monthly-mean, zonal-7 mean ozone are shown at 40oS, the Equator and 40oN. At 46 hPa corresponding plots are 8 shown for 72oS, the Equator, and 72oN. Four separate years of MLS observations are 9 shown in the Southern Hemisphere and Equatorial plots (January 2005-December 2008) 10 and three years (July 2005-June 2008) are shown in the Northern Hemisphere; the 11 Northern Hemisphere observations are phase shifted to align the seasons with those of the 12 Southern Hemisphere observations. From the models, only a single year’s annual cycle is 13 shown, taken from the early 2000s, for consistency with the data. Examination of up to 14 an additional 10 years per model (not shown) has demonstrated that the comparisons are 15 representative. The annual mean has been subtracted in all figures to emphasize the 16 seasonal variation in both observations and simulations. 17 18 Although the ozone column is dominated by mixing ratios in the lower stratosphere and 19 hence its annual cycle is barely affected by the evolution of upper stratosphere mixing 20 ratios, a comparison at 1 hPa provides a simple check on the performance of the 21 photochemical schemes implemented in the various models (compare also the more 22 detailed comparison of photochemical schemes in Chapter 6). The simulated annual cycle 23 at both 40oS and 40oN generally approximates the MLS data. The simulated annual cycle 24 in temperature also agrees with observations (Fig. 8.2.S1 in the supplementary material), 25 so this comparison verifies the simulated sensitivity to temperature. A positive anomaly 26 in the Southern hemisphere during May and June in the MLS is not reproduced by any 27 model. A similar (negative) feature in temperature mirrors this anomaly. In the tropics, a 28 semi-annual oscillation is also seen in the observations. Many of the models also 29 reproduce this semi-annual variation but with differences in the timing. This phase 30 difference between models and observations is also seen in the temperature variations 31 (Fig. 8.2.S1) and therefore explains the mismatch in ozone. To summarize, the models 32 exhibit the appropriate sensitivity to temperature, so, when the simulation reproduces (or 33 does not reproduce) the temperature variation, the ozone variation matches (does not 34 match) that observed. 35

36 At 46 hPa during winter and spring in the high latitude Southern hemisphere, the ozone 37 mixing ratio anomaly is dominated by polar ozone loss. Figure 8.2.1 (bottom) shows that 38 the models generally reproduce this variation, except for UMUKCA-METO and 39 UMUKCA-UCAM. For both of these simulations, there is polar ozone loss, but it does 40 not extend to 72oS. While observations show a peak ozone loss in September, the CCMs 41 response is shifted by one to two months. At the Equator the MLS data show a small 42 seasonal variation that depends on the phase of the QBO and is not fully captured by the 43 models (see section 8.3). In the Northern hemisphere, transport and polar ozone 44 destruction processes control the evolution during winter/spring. Both contribute to the 45 substantial observed variability in ozone during these seasons. This interannual variability 46

10

is so large (see section 8.4) that differences between the observations and simulations are 1 not significant. During the summer the photochemical timescale decreases to 30-60 days, 2 the circulation is near zonal with little horizontal or vertical mixing and the ozone mixing 3 ratio is close to photochemical balance. The interannual variability of the observed ozone 4 mixing ratio during this period is minimal. In the models, there is a relatively large spread 5 of the models during this period (and also during January-February-March in the 6 Southern Hemisphere), which likely reflects the spread of temperatures between the 7 models. Nevertheless, the simulated ozone mixing ratios return to values that are the 8 same each year within a few percent in each of the last 10 years of the integrations (not 9 shown), in agreement with the observations, thus demonstrating that models make a 10 reasonable transition to photochemical control in summer. This variation decreases with 11 increasing pressure; at 70 hPa the models reproduce the observed small annual variation 12 in ozone mixing ratio (not shown). 13 14 8.2.2 Ozone in March and October 15 16 Figure 8.2.2 compares climatological mean vertical profiles and latitudinal cross sections 17 of ozone derived from the CCMVal-2 models and HALOE observations (similar to 18 Figure 13 from Eyring et al. 2006 for the CCMVal-1 models). At the Equator, most 19 models agree well with HALOE observations and lie within one standard deviation of the 20 HALOE mean, except for the CCSRNIES model that shows unusually large ozone peak 21 values at 10hPa (consistent with low CH4? Check with chemistry chapter). During 22 Northern hemisphere and Southern hemisphere spring there is a larger spread between the 23 models and only a few lie within one standard deviation of the HALOE mean. This is 24 especially true in the lower stratosphere/upper troposphere where the CCMVal-1 25 simulations showed very good agreement with observations but CCMVal-2 simulations 26 shows a much larger spread (see also Chapter 7 for a detailed discussion on UTLS 27 performance of each model). 28 29 In the Southern hemisphere, the vertical profiles of CCSRNIES, CAM3.5, EMAC, 30 UMUKCA-METO, and UMUKCA-UCAM are biased high, while LMDZrepro is biased 31 low. In the Northern hemisphere again CCSRNIES and CAM3.5 are biased high while 32 this time SOCOL is biased low. For the CCSRNIES model, the overestimation of peak 33 ozone values in the tropics and polar regions was already evident in CCMVal-1 and is 34 likely related to overestimation of O2 photolysis rates at this altitude (check with 35 chemistry chapter!). The pronounced ozone bias that was evident in LMDZrepro in 36 CCMVal-1 has been improved (because temperature bias has improved?) but this model 37 is still biased low. 38 39 The lower section of figure 8.2.2 shows that the latitudinal representation of ozone in the 40 lower stratosphere in spring-time of each hemisphere has been improved since CCMVal-41 1. Between 60°S and 60°N most models lie in the one standard deviation range of the 42 HALOE data. The CNRM-ACM is a clear outlier, which substantially underestimates the 43 values. At polar latitudes more than half of the CCMs significantly overestimate the 44 ozone values from HALOE (check with chemistry chapter why!). SOCOL, Niwa-45 SOCOL, AMTRAC3, UMSLIMCAT, and WACCM agree best with observations at 46

11

northern high latitudes in March, while at southern high latitudes additionally, CNRM-1 ACM and LMDZrepro are equally good compared to observations. 2 3 8.2.3 Metrics 4 5 In this section, the differences between the modeled and observed annual cycle in ozone 6 previously discussed are extended and quantified by means of normalized Taylor 7 diagrams (Taylor, 2001). The usefulness of the Taylor diagrams is their compact 8 representation of pattern statistics (correlations and centered root mean square errors) 9 between two fields, thus providing a straightforward methodology to quantify and 10 compare the pattern-similarity of a large number of fields (model diagnostics) with 11 respect to a reference field (observations). 12

13 To provide a quantification of how well the background ozone is simulated, Figure 8.2.3 14 shows the normalized Taylor diagram of the zonal mean, annual mean ozone component 15 from the MLR analysis performed on the model results (period: 1960-2004) and the 16 NIWA-3D observations (period: 1979-2007). The zonal mean, annual mean ozone 17 distribution is characterized by a maximum of about 9 ppmv at 10 hPa in the deep tropics 18 in the NIWA-3D observations. Since the focus of the evaluation is stratospheric ozone, 19 the pattern statistics are computed for the latitude-pressure sections ranging respectively 20 from the South to the North poles, and from 500 to 1 hPa pressures. The pattern statistics 21 calculation includes area weights, but no weighting in pressure. Figure 8.2.3 shows that 22 this zonal mean, annual mean field is extremely well simulated by all the models: The 23 correlations are about 0.99 or higher, and the NRMS (Normalized Root Mean Square) 24 errors are rather small, as demonstrated by the close clustering of the model signatures 25 around the black solid point on the abscissa, which is the reference observation. 26

27 Figure 8.2.4 shows the normalized Taylor diagram for the annual cycle and the 28 semiannual cycle components in ozone from the MLR analysis. Also in this case, the 29 pattern statistics are computed for the latitude-pressure sections ranging respectively from 30 the South to the North poles, and from 500 to 1 hPa pressures; and the pattern statistics 31 calculation includes area weights, but no weighting in pressure. The corresponding fields, 32 from which the correlations and NRMS errors between each model and the observations 33 are computed, are shown in the supplementary material (Figs. (8.2.S2, 8.2.S3). The major 34 feature of the annual cycle in the NIWA-3D observations is the anti-symmetric pattern 35 centered at the equator, between 10 and 30 hPa. The visual inspection of each model 36 panel shows that this pattern is captured well by all models, although the magnitude of 37 the oscillation and the various additional features present in the annual cycle field are less 38 well simulated. The Taylor diagram hence shows a substantial correlation (between 0.8 39 and 0.95) for all the models but for one exception, the CAM3.5 model that is an outlier, 40 with a correlation of only 0.6, possibly because of its low top. The NRMS errors are 41 approximately half the size of the spatial standard deviation of the reference observation 42 field. The exception is again the CAM3.5 model, with an nrm error of about 1. For the 43 semiannual cycle, the quality of the model results is slightly degraded, showing the 44 difficulties of modeling relatively smaller scale features. 45

46

12

8.2.3.1 Column ozone 1 2

The evaluation of the annual cycle in column ozone is performed on the zonal monthly 3 mean model data and the observations. For the models, only data from 1980 to 2000 (or 4 up to the end of the simulation) are considered, to match the period of the column ozone 5 NIWA dataset (1979-2007, Bodeker et al., 2005, updated) used as the reference field. The 6 normalized Taylor diagram for the zonal mean column ozone is shown in Figure 8.2.5, 7 and the corresponding fields in the supplementary material (Fig. 8.2.S4). Also in this 8 case, the Taylor diagram demonstrates that the models capture the phase of the annual 9 cycle and the latitudinal distribution of the total ozone quite well (correlations ranging 10 from 0.9 and 0.97). Also the amplitude of the annual cycle is captured by the models, 11 with NRMS errors of about (or less of) half the size of the spatial standard deviation of 12 the reference observation field. 13

14 In the computation of the Taylor diagram, the error in the mean bias is excluded. Given 15 that it is important to report also about this quantity for a complete evaluation, the mean 16 bias is shown for the global mean total ozone in Table 8.2.1. With the exception of a few 17 outliers (E39C and MRI), the mean global bias is rather small (below 10%). 18 19 8.3 Interannual polar ozone variability 20

21 In the Extratropics, interannual natural variations in stratospheric ozone are largest in the 22 polar regions, and tend to maximize during the spring season. To diagnose the annual 23 cycle in interannual variability in the models, Figure 8.3.1a shows the monthly 24 interannual standard deviation of column ozone averaged over the polar caps (60N-90N, 25 LHS and 60S-90S RHS), from the models and the observations (i.e., the NIWA column 26 ozone dataset). The corresponding annual cycle in total ozone is shown in Figure 8.3.1b. 27 Prior to the calculation of Figure 8.3.1a, decadal scale trends were removed from the 28 data. This was accomplished by calculating a low-pass filtered version of the data (using 29 Gaussian-weighted running means with a full width of half maximum of 9 years) and by 30 removing it from the original. The resulting time series has the same interannual 31 variability as the original data but with decadal scale trends removed. The results shown 32 in Figure 8.3.1 are computed for two periods, respectively since 1960 (upper panels) and 33 since 1980 (lower panels) 34

35 Figure 8.3.1a shows that the interannual variability of the NIWA column ozone exhibits a 36 pronounced annual cycle and maximizes during the dynamically active late winter and 37 early spring periods of each hemisphere, which is February-April in the Northern 38 hemisphere and September-November in the Southern hemisphere. The two black lines 39 representing the NIWA observations are for two different time periods, excluding 40 (dashed) or including (solid) the exceptional vortex split event in September 2002 over 41 the Southern hemisphere. 42 43 The simulation of the observed seasonality in the variability of total ozone represents an 44 important model benchmark. Figure 8.3.1a demonstrates that the observed annual cycle 45 in column ozone variability is well reproduced by all models, in the sense that all show a 46

13

minimum in variability in the respective summer seasons. During the Southern 1 hemisphere active period, the model-range results tend to surround the observations. 2 Models with particularly low variability are GEOSCCM (with virtually no increase in 3 variability from the summer to the dynamically active period) and UMUKCA-UCAM (in 4 November). Models with particularly high variability suggesting an early start of the 5 active period are CAM3.5 and CCSRNIES, while CMAM has excessive variability in 6 November. During the Northern hemisphere active period most of the models 7 underestimate the observed total ozone variability, up to almost a factor of two in a few 8 cases (CMAM and UMSLIMCAT), indicating a systematic bias. One notable exception 9 is the MRI model, which exhibits very large variability (more than a factor of 2, in 10 March). The results for individual ensemble members of MRI (not shown) are very 11 similar, indicating that this high variability is not due to sampling uncertainty. Another 12 outlier is the WACCM model, with prolonged period of high variability, extending into 13 June over the Northern hemisphere and into January over the Southern hemisphere. The 14 differences in the interannual variability computed over the period since 1980 and since 15 1960 are generally small and consistent among the models, with the exceptions of the 16 UMUKCA-UCAM model in February-March, Northern hemisphere (compare upper and 17 lower panels of Figure 8.3.1a). 18

19 The annual cycle of the total ozone averaged over the polar caps (Figure 8.3.1b) shows 20 that in the Northern hemisphere the months with high variability are characterized by 21 larger column ozone amounts and the model-range surrounds the observations. In the 22 Southern hemisphere, the situation is complicated by the presence of the ozone hole, an 23 anthropogenic modification of the annual cycle. 5 models (CAM3.5, E39CA, EMAC, 24 UMUKCA-METO, UMUKCA-UCAM) do not reproduce the dip in ozone between 25 August and November and hence fail to model the impact of the ozone hole depletion on 26 the annual cycle. 27

28 The winter and spring evolution of column ozone is associated with the time-variation of 29 planetary-scale and gravity wave activity and its influence on the strength of polar 30 descent in the Brewer-Dobson circulation (Fusco and Salby, 1999, Randel et al., 2002). 31 When the wave activity is high, adiabatic descent at high latitudes is strengthened, 32 leading to increased transport of ozone-rich air from the tropical middle stratosphere 33 (where ozone is photochemically produced) to the polar lower stratosphere. In addition, 34 increased wave activity leads to a more disturbed polar vortex and hence to warmer polar 35 temperatures, creating unfavourable conditions for heterogeneous chemical depletion of 36 ozone due to heterogeneous processes. Hence, a positive correlation between column 37 ozone amounts and wintertime wave activity is expected. To evaluate the relationship 38 between the wintertime ozone variability shown in Figure 8.4.1 and dynamical variability 39 in the models, we diagnose relationships between heat flux and total ozone, temperature 40 and total ozone, and the stratospheric annular mode and total ozone. 41

42 8.3.1 Heat flux and total ozone 43

44 Weber et al. (2003) show a compact relationship between the spring-to-fall ozone ratio in 45 each hemisphere and the wintertime mean heat flux. The presence of a similar 46

14

relationship is investigated in the REF-B1 integrations of twelve of the CCMVal-2 1 models. In an extension of the Weber et al. study, the CCMVal-2 models are compared to 2 observations using wintertime mean 100 hPa meridional heat flux derived from the ERA 3 interim dataset and the fall to spring ratio in total column ozone derived from the NIWA 4 assimilated total ozone dataset. Total column ozone ratios are calculated for the March 5 over September in the Northern hemisphere and September over March in the Southern 6 hemisphere, using area weighted averages between 50° and the pole. Heat fluxes are 7 averaged between 45° and 75°, using extended winter means (September-March in the 8 NH and March-September in the SH). Southern hemisphere data are detrended as done 9 for Figure 8.3.1a. In order for the ratio of spring-to-fall ozone ratio to be calculated it is 10 necessary to add a climatological ozone concentration to the filtered time series: a 10 year 11 mean, monthly ozone concentration (1990-2000) was employed for this, since it is a 12 period common to both the data and models. 13 14 Results from the individual scatter plots (see supplementary material Figs. 8.3.S1 and 15 8.3.S2) are summarized in a single plot as shown in Figure 8.3.2, where the slope of each 16 scatter plot is plotted against the mean spring-to-fall ozone ratio for each model or 17 dataset, along with 95% confidence intervals for each parameter. The slope of the scatter 18 plot (plotted on the y-axis) describes the typical response of the spring-to-fall ozone ratio 19 to a one-unit increase in the absolute value of 100 hPa meridional heat flux. Since the 20 absolute value of the heat flux is a proxy for the upward component of the Eliassen-Palm 21 flux, the slope diagnoses the response of ozone at each polar cap to changes in the 22 amount of planetary wave activity entering the lower stratosphere. The mean ratio of the 23 spring-to-fall ozone concentration diagnoses the average seasonality in ozone 24 concentrations present in each model. This is a more useful measure of the position of the 25 model on each scatter plot than the intercept of the regression line, which is usually a 26 large distance from the centre of the cloud of points for each model. 27

28 In general, models seem to reproduce the relationship between dynamical variability and 29 ozone well, with all simulating a positive relationship clustered around the observations. 30 In the SH (right panel), the slope (representing the relationship between spring-to-autumn 31 ozone ratio and meridional heat flux), compares well with the observed relationship 32 observed between the ERA-interim and NIWA datasets. However, there is also a 33 relatively large and obvious spread in the mean ratio of September-to-March total ozone 34 between the models. About half of the models have a smaller ratio than observed. This 35 suggests that the September-to-March ratio is less than 1.0 because of polar ozone 36 depletion, i.e. the ratio is influenced by chemistry and not just dynamics. Hence, biases in 37 the modeling of chemical processes can contribute to the model spread. Possibly, it may 38 also suggest that the advection of ozone rich air into the polar cap, which would tend to 39 produce a September-to-March ratio above 1, is weaker in most models than in the 40 reanalysis. However, analysis of some of the same models in Chapter 4 did not suggest 41 that the strength of their Brewer-Dobson circulation was too weak. There is some 42 indication in Chapter 5 that models with a much lower ratio of September to March ozone 43 in the southern hemisphere perform poorly in diagnostics of their polar isolation 44 (LMDZrepro, MRI, Niwa-SOCOL, SOCOL). However, it is also true that some models 45 with good transport diagnostics also show low spring to fall ozone ratios here. 46

15

1 In the NH (left panel of Figure 8.3.2), similar comments apply. No model has a 2 significantly different relationship between spring-to-fall ozone ratio and meridional heat 3 flux to that observed, although the ULAQ model has a much weaker relationship between 4 heat flux and March-to-September ozone ratio. However, in contrast to the SH, although 5 there is a large spread in the mean fall-to-spring ozone ratio in the models, the reanalysis 6 data falls approximately in the middle of the model range. In this case, very weak 7 transport of ozone into the vortex is implied for the Niwa_SOCOL and SOCOL models 8 and very strong transport of ozone into the vortex is implied in the GEOSCCM, ULAQ 9 and UMUKCA-METO models. 10 11 8.3.2 Temperature and total ozone 12

13 The analysis between heat flux and total ozone presented in Section 8.3.1 could only be 14 performed for a subset of CCMVal-2 models, since many models did not provide the 15 necessary data. We now continue with a somewhat similar analysis, making use of the 16 tight relationship between heat flux and temperature (Chapter 4) and evaluate the 17 relationship between total ozone and lower stratospheric temperatures. The existence of 18 such a relationship has previously been identified from radiosonde observations by 19 Fortuin and Kelder (1996). 20 21 Polar cap averaged (60-90°) monthly temperature at 50 hPa are used as a proxy for heat 22 flux and are compared against polar cap averaged (60-90°) total column ozone. The 23 analysis is focused on spring (March for the Northern Hemisphere and November for the 24 Southern Hemisphere), which is the time when the cumulative effects of wave activity 25 during winter on temperature are most pronounced. The analysis is performed for every 26 year of model data since 1960 and for NIWA/reanalysis since 1980. 27 28 Instead of showing scatter plots of the results, we use a similar technique as in Section 29 8.3.1 and calculate linear fits between ozone and temperature. Figure 8.3.3 displays for 30 each model and for the observations the parameters of the fits and their 95% confidence 31 intervals. The slope of the scatter plot, shown on the y-axis, indicates how sensitive total 32 ozone is to a given temperature perturbation. The x-axis represents the ozone amount of 33 the linear fit at a temperature of 200 K, which is used as second parameter to describe the 34 goodness of the fits. For the observations (black signatures), we show results from NIWA 35 total ozone and temperatures derived from either (solid) NCEP/NCAR or (broken) ERA-36 40 reanalysis. 37 38 The results shown in Figure 8.3.3 indicate that the models over both hemispheres 39 reproduce the expected positive relationship seen in the observations, in the sence that all 40 slopes are positive and total ozone increases when temperatures are anomalously warm. 41 However, for the Northern Hemisphere (March, left panel), only 5 models (AMTRAC3, 42 CCSRNIES, MRI, Niwa-SOCOL, SOCOL) reproduce the observed relationship well. 43 One model (CNRM-ACM) considerably overestimates the observed slope. Most models 44 instead underestimate the slope by almost a factor of two (6 versus 3 DU/K), indicating 45 that for most models the simulated ozone is less sensitive to a given temperature 46

16

perturbation than the observations. The x-axis values, the amount of ozone at a 1 temperature of 200 K, are also larger for most models. 2 3 In November (Southern hemisphere, right panel of Figure 8.3.3), the number of models 4 that either over- or underestimate the observed slope is quite evenly distributed around 5 the observations. Most models tend to overestimate the amount of ozone associated with 6 T = 200 K, which is consistent with the results shown in the bottom right panel of Figure 7 8.3.1b, i.e., most models overestimate the amount of spring ozone over the Southern 8 Hemisphere. The positive ozone bias is particularly large for the UMUKCA-UCAM 9 model, and consequently this model stands out in the plots of Figure 8.3.3 over both 10 hemispheres. It should be noted that these positive biases are mostly a consequence from 11 using a longer averaging period (1960-) for the models than for the observations (1980-), 12 and thus somewhat discounting the effects of the ozone hole. This can be confirmed by 13 comparing with the upper right panel of Figure 8.3.1b, where the bias of the multi-model 14 mean (gray line) is close to zero. 15 16 8.3.3 Stratospheric annular mode and total ozone 17

18 High wave activity leads to a relatively weak and warm stratospheric polar vortex and a 19 negative annular mode (AM). On interannual timescales the strength of the AM in the 20 lower stratosphere and the heat fluxes at 100 hPa are closely connected (Hu and Tung, 21 2002), hence a relationship should also exist between total ozone variation and the AM. 22 This possibility is investigated by regressing the monthly mean total ozone time series on 23 the AM index at 50 hPa. The 50 hPa level is chosen because total ozone is mostly 24 affected by variations in the lower stratosphere. Figure 8.3.4 shows the regression 25 coefficients between local variations of total ozone and an AM index. Total ozone 26 amounts are from the NIWA dataset and the AM index is derived from the ERA40 or 27 NNR (as indicated on top of each panel). A relatively simple definition for the AM index 28 is employed, which uses polar cap averages (60-90°) of monthly mean zonal mean 29 geopotential height anomalies at 50 hPa. Prior to the analysis, all data are detrended as 30 explained before. The resulting index has the opposite polarity from the usual EOF-based 31 AM index and it carries units of gpm. This AM definition not only has the advantage that 32 it is easier to calculate, it also represents an absolute measure and thus avoids possible 33 ambiguities associated with the polarity and magnitude of the traditional EOF-based 34 approach. Comparisons between the EOF-based AM mode and the simpler version used 35 here show virtually identical outcomes (not shown). As expected, using the simple AM 36 leads to positive regressions over the northern polar region. Total ozone is high when the 37 AM is positive, i.e., when the geopotential height anomalies over the pole are positive, 38 indicative of a warm and weak vortex, increased wave activity, and an anomalously 39 strong descending arm of the Brewer Dobson circulation at polar latitudes. As in Section 40 8.3.2, we focus on the spring months March and November, a time when this relationship 41 is expected to be most robust. 42

43 The left panel of Figure 8.3.4 is for the Northern hemisphere, March. The results for the 44 observations, which are based on total ozone from NIWA and 50 hPa geopotential 45 heights derived from either ERA40 or, NCEP/NCAR reanalysis are shown on the upper 46

17

left. As expected, using the simple AM leads to positive regressions over the northern 1 polar region. Comparing the models with the observations, it becomes clear that over the 2 Northern Hemisphere most CCMVal-2 models reproduce the structure of the observed 3 regression pattern quite well: 12 models (AMTRAC3, CCSRNIES, CMAM EMAC, 4 LMDZrepro, MRI, Niwa_SOCOL, SOCOL, UMSLIMCAT, UMUKCA_METO, 5 UMUKCA-UCAM and WACCM) out of the 17 models exhibit correlations of at least 6 80% with the observations. The outliers are CAM3.5 (65%), CNRM-ACM (62%), 7 E39CA (75%), GEOSCCM (67%), and ULAQ (79%). Models generally underestimate 8 the strength of observed regressions, in particular AMTRAC3, CAM3.5, CMAM, and 9 E39CA. The right panel of Figure 8.3.4 presents the outcome of the regression analysis 10 for the Southern hemisphere, November. There is some tendency of the models to better 11 simulate the observed pattern over the Southern hemisphere, except for the ULAQ, and 12 WACCM models. We also note that individual models show similar performances over 13 both hemispheres. 14

15 8.4 Solar cycle 16 17 The 11-year solar cycle has a direct impact on ozone via radiation and chemistry in the 18 upper stratosphere and indirect effects on dynamics, transport and chemistry throughout 19 the stratosphere (e.g., review by Gray et al., 2009). The direct effect in the upper 20 stratosphere depends on a good representation of solar radiation processes in the radiation 21 as well as in the photochemistry codes (see Chapter 3 for a detailed comparison of 22 radiation codes and Chapter 6 for a detailed comparison of the photochemical schemes) 23 and were reasonably well simulated by the CCMVal-1 models (Austin et al., 2008). 24 However, the indirect dynamical effects in the tropical lower stratosphere and extra-25 tropical stratosphere and the extension of the signal into the troposphere (see e.g., Haigh 26 et al., 1999; Kodera and Kuroda, 2002; Haigh et al., 2005; Kodera et al., 2006) are more 27 challenging to reproduce. Matthes et al. (2003) suggested that a realistic representation of 28 the model’s climatology is an important pre-requisite for the indirect dynamical effects. 29 Other suggested important “ingredients” are a QBO, a time-varying solar cycle, and 30 realistic interannual variability in the SSTs. One remaining challenging task is to 31 understand the observed interaction of the solar signal with the tropical oscillations (QBO 32 and SAO) in the equatorial as well as in the high-latitude stratosphere (see e.g., Labitzke 33 and van Loon, 1988; Gray et al., 2001). This interaction is still difficult to investigate 34 since the number of observed events when separated into solar and QBO phases is small 35 and only some of the CCMs reproduce an internally generated QBO to study the full 36 solar/QBO interaction (see section 8.3). On the other hand there is still considerable 37 spread in the observed solar cycle signal, so an understanding of the modeled responses 38 might help to understand the observed response. In the following section the solar cycle 39 response is examined, but only in the stratosphere without touching in detail the 40 complicated tropospheric responses and extra-tropical interactions, which are beyond the 41 scope of the current report. 42 43 Figure 8.4.1 shows a comparison of the solar regression coefficients in total column 44 ozone for the CCMVal-2 CCMs relative to the observed solar regression coefficient from 45 TOMS+gb data shown in Figure 8.1. There are two model groups. CCMs in the first 46

18

group do not prescribe a solar cycle in irradiances (GEOSCCM, ULAQ, UMUKCA-1 METO, UMUKCA-UCAM; see Chapter 2 Table 2.11; referred to as non-sc group), while 2 all the other CCMs impose a solar cycle (sc group). Consistently, Figure 8.4.1 shows 3 values of the solar coefficient in total ozone around 0 for the non-sc group models (note 4 that UMUKCA-UCAM does not provide total column ozone and is therefore not shown). 5 Most of the sc group models show a solar regression coefficient that is approximately 6 half of the observed value. Note that there are substantial differences in the magnitude of 7 the solar regression coefficients between different ozone data sets and that the TOMS+gb 8 data set is biased high (Randel and Wu, 2007). CMAM and E39C show the best 9 agreement with the TOMS+gb observed values. CMAM is one of the models that 10 perform especially well in the solar cycle radiation intercomparion (section 3.6). 11 Surprising is the difference between the two low-top CCMs E39C and CAM3.5 that do 12 not include the whole stratosphere. While CAM3.5 shows a very small correspondence 13 with observations something that would be expected from a low top model, E39C does 14 perform very well. Differences in the radiation schemes and the input data (either 15 spectrally resolved solar UV data and/or total solar irradiance (TSI) data) are discussed in 16 chapter 3 section 3.6. 17 18 8.4.1 Vertical structure of temperature and ozone signal in the tropics 19 20 Considerable discrepancies exist in the vertical structure of the tropical solar signal 21 between the various observational datasets (Gray et al., 2009) as well as between 22 observations and CCMs (WMO, 2006). In the upper stratosphere where the observed 23 temperature and ozone signals are greatest, there is good agreement amongst the CCMs 24 and with the observations, within uncertainty ranges. Below 10 hPa, however, there is 25 less good agreement. Austin et al. (2007, 2008) showed that more recent model studies 26 have achieved an improved vertical structure in this region and speculated that it may be 27 related to (a) the introduction of time-varying solar cycle simulations instead of the 28 constant solar min/max simulations that had previously been performed because of 29 limited computer resources or (b) an aliasing effect of the SSTs with the solar cycle. 30 Marsh and Garcia (2007) discuss the inability of the MLR technique to take into account 31 autocorrelation between e.g., the solar and the ENSO signal, although the MLR analysis 32 employed here should be able to handle this since the autocorrelation in the residual is 33 taken into account (as did Crooks and Gray, 2005). Nevertheless, the real atmosphere is 34 highly non-linear and it may be difficult to capture the solar signal completely with the 35 linear method used here. Another factor that complicates the solar signal is the QBO. Lee 36 and Smith (2003) and Smith and Matthes (2008) discuss an aliasing effect of the QBO 37 (and volcanoes) with the solar cycle. Recently, Matthes et al. (2009 a,b) showed that they 38 can reproduce the observed vertical structure in the tropical solar ozone and temperature 39 signal in the middle and lower stratosphere only when a QBO is present in the CCM. 40 41 Figure 8.4.2 shows the annual mean of the tropical vertical solar signal in temperature 42 and ozone. The CCMs without solar cycle (top panels) show consistently a response 43 around zero (GEOSCCM, UMUKCA-METO, UMUKCA-UCAM, and ULAQ). All other 44 models (bottom panels) and observations show the largest temperature and ozone solar 45 response in the upper stratosphere at 1 and 3 hPa, respectively. This is the direct solar 46

19

effect due to enhanced UV absorption during solar maxima that leads to higher 1 temperatures and greater ozone production, that in turn positively feedback to each other. 2 The range in temperature response around the stratopause varies from up to 0.8 K in 3 WACCM, CMAM, AMTRAC, and UMSLIMCAT down to ~0.2 K in LMDZrepro and 4 CNRM-ACM. Note that UMSLIMCAT shows a larger warming of about 1K higher up at 5 55km. 6 7 The temperature signals are consistent with the shortwave heating rate differences shown 8 in the supplementary material (Fig. 8.4.S2) and the offline solar radiation calculation 9 results1 in chapter 3 for the subset of CCMs that did provide data (section 3.6). The solar 10 signal in shortwave heating rates is largest for EMAC, WACCM, MRI, CMAM, and 11 CCSRNIES and almost half that magnitude for LMDZrepro. It is close to zero for the 12 UMUKCA-UCAM model that did not impose a solar cycle. CCMs that only prescribe 13 total solar irradiance changes, i.e. LMDZrepro, underestimate shortwave heating and 14 therefore solar induced temperature signals. Some of the CCMs with large shortwave 15 heating response (MRI and EMAC) show a smaller temperature signals than e.g., 16 WACCM, CMAM and CCSRNIES and other CCMs with smaller heating rate responses 17 show however a larger temperature signal. This is related to the solar induced ozone 18 signal and therefore the performance of the photochemical schemes (compare chapter 5 19 for a detailed comparison). In other words the solar induced temperature signals in Fig. 20 8.4.2 are a combination of solar UV radiation and solar induced ozone changes and are 21 dependent on the prescription of spectrally resolved or total solar irradiance changes in 22 the radiation and the photochemical schemes and their individual performances (compare 23 chapters 3 and 6). 24 25 The temperature and ozone responses are largely consistent in the upper stratosphere 26 between the CCMs and compared to observations. The CNRM-ACM features the largest 27 ozone signal with the largest uncertainties (Fig. 8.4.S1) in the upper stratosphere. This 28 might be related to its photochemistry (compare chapter 6). The solar temperature 29 response in the SSU and the ERA-40 data lies in the middle of the CCM responses with 30 large uncertainties of about 1.4 K especially for the ERA-40 data (supplementary material 31 Fig. 8.4.S1). The RICH radiosonde data indicate a relative maximum of 0.5K in the lower 32 stratosphere. The solar ozone response in the CCMs compares well with the observational 33 response in the Randel and Wu data set. The NIWA 3D data set produces a greater solar 34 ozone signal comparable to the CNRM-ACM. Note that the MLR presented here 35 produces slightly smaller solar signals then previously published for SSU and MSU4 (we 36 only use SSU) (Randel et al., 2009) and SAGE ozone data. This is certainly related to the 37 different analysis technique and differing basis functions. 38 39

1 We note that the offline radiation calculations for each of the CCMs shown is not necessarily in

accordance with what have been used for the REF-B1 simulations. E.g., UMUKCA-UCAM does not have a solar cycle in the REF-B1 simulation but shows shortwave heating rate differences from the offline radiation calculations in section 3.6 that are related to solar incuded ozone changes in the offline calculations only.

20

Discrepancy between the CCMs themselves and the observations increases below 10hPa 1 consistent with the increase of the uncertainties (compare Fig. 8.4.S1 in the 2 supplementary material). Most CCMs show a positive solar temperature signal that 3 increases with increasing height in good agreement with the SSU data, except 4 CCSRNIES and EMAC that show a relative minimum like the ERA-40 data although the 5 height differs. Some CCMs (AMTRAC3, CMAM, CCSRNIES, UMSLIMCAT and 6 WACCM), show an indication of a secondary temperature maximum (apparent in RICH 7 radiosonde data and ERA40) in the lower stratosphere. But as noted above these changes 8 are statistically not significant. 9 10 The vertical structure of the solar signal in ozone is much better represented between the 11 CCMs and compared to observations. The CCMs compare well with the Randel&Wu 12 ozone data in the middle and upper stratosphere while the agreement with the NIWA 3D 13 data set is better in the lower stratosphere. A secondary peak in ozone in the lower 14 stratosphere is simulated by AMTRAC3, CNRM-ACM, CCSRNIES, MRI and WACCM. 15 Except for CNRM-ACM all these CCMs have variability related to a (prescribed) or 16 internally generated QBO-like oscillation. The response in the CNRM-ACM is larger 17 than in the other CCMs and could be related to the differences in the transport scheme 18 (compare chapter 5). Note that both low-top CCMs (CAM3.5 and E39CA) produce a 19 small solar signal in temperature and CAM3.5 also in ozone since they do not include the 20 stratopause region where the initial solar signal appears, whereas E39C shows a relative 21 large solar ozone signal consistent with the largest signal in total column ozone in Figure 22 8.4.1. 23 24 The uncertainty in the observed signals and the wide spread of model responses is 25 exemplarily summarized in the Taylor diagram (Figure 8.4.3) for ozone only. The four 26 models (GEOSCCM, UMUKCA-METO, UMUKCA-UCAM, and ULAQ) without solar 27 cycle are excluded. All the models that impose the 11-year solar cycle show correlations 28 below 0.8 with the NIWA data. CAM3.5 and CNRM-ACM stand out of the cloud of 29 model signatures, respectively with the smallest and largest standard deviation. 30 31 8.4.2 Latitudinal structure of the solar signal in temperature and ozone 32 33 The latitudinal structure of the amplitude of the solar cycle in temperature and ozone is 34 shown in Figure 8.4.4 at 1 and 3 hPa, respectively. Apparent is the large spread of model 35 results as well as the differences in the observational data sets itself. The solar signal in 36 ozone between the models and also between the Randel&Wu and the NIWA 3D ozone 37 data sets is in good agreement in the tropics and mid latitudes while large differences 38 occur at northern and southern high latitudes due to large interannual variability (compare 39 section 8.3). EMAC and WACCM show the largest latitudinal variations. 40 41 The solar temperature signal shows less variability between the CCMs than the ozone 42 signal and is in good agreement with the SSU data. The ERA-40 response, on the other 43 hand, shows a positive response at equatorial latitudes and a negative response at higher 44 latitudes. As with the ozone datasets, there is significant variation between the different 45 observational datasets (Gray et al., 2009). However, neither observational dataset 46

21

demonstrates statistical significance poleward of ~30o, so validation of the models at 1 these latitudes is difficult. The difficulty of reproducing the latitudinal structure of the 2 solar signal is also featured in the latitudinal structure of the annual-mean solar regression 3 coefficient for total column ozone (see supplementary material, Fig. 8.4.S3). The spread 4 in model responses is especially large at high northern latitudes due to dynamical 5 interactions. Very large deviations are seen for EMAC at southern high latitudes and 6 WACCM at northern high latitudes. The differences that we see might be related to 7 differences in the transport schemes since total column ozone variations are dominated by 8 transport effects of lower stratospheric ozone (compare chapter 5). 9 10 Since the spread in both the modeled and observed solar cycle signal is so large, 11 especially at high latitudes, no further diagnostics are presented to investigate dynamical 12 feedback mechanisms (Kodera and Kuroda, 2002), such as those shown by Matthes et al. 13 (2003) who investigated GCMs in which the ozone solar signal was imposed. Recent 14 model studies (e.g., Matthes et al., 2006; Gray et al., 2006; Matthes et al., 2009a,b; Ito et 15 al., 2009) suggest that these dynamical feedback mechanisms are particularly difficult to 16 reproduce, because of possible non-linear interaction with the QBO, and are currently 17 best investigated in more idealized model studies in which the various influences can be 18 separately examined. 19 20 Several studies have highlighted the limitations of the MLR analysis with respect to the 21 time period chosen and the difficulty of separating auto-correlated signals such as the 22 solar and the QBO, volcanic or ENSO signal in the equatorial lower stratosphere (e.g., 23 Smith and Matthes, 2008; Marsh and Garcia, 2007; Austin et al., 2008). The sensitivity of 24 the MLR analysis presented here has been tested using different time periods, i.e. 1960-25 2004 and 1979-2004. The results are not very sensitive to the period chosen, apart from 26 the magnitude of the response changes, which is larger for the shorter time period. This 27 allows confidence in the performance of the MLR method, provided careful 28 representation is made of all possible basis functions as well as an autocorrelation of the 29 residual. 30 31 8.5 QBO in Ozone 32 33 In the tropical stratosphere, the QBO in zonal wind is a major driver of ozone variability 34 (see Baldwin et al 2001 for a review). Typically, however, general circulation models of 35 the atmosphere have difficulties in spontaneously simulating the QBO. In order to 36 simulate a realistic QBO, a model should be able to support a realistic spectrum (temporal 37 and spatial) of upward propagating waves in the tropics. This is a major challenge, 38 because this spectrum of waves depends on many technical aspects of an atmospheric 39 general circulation model, such as tropical convection parameterization, stability of the 40 troposphere, sea surface temperatures, vertical and horizontal resolutions and atmospheric 41 gravity wave parameterizations (e.g., Scaife et al., 2000; Giorgetta et al., 2002, 2006; 42 Shibata and Deushi, 2005). 43 44 A model that does not appropriately simulate the QBO in zonal wind, also severely 45 misrepresents the natural ozone variations associated with the QBO (Punge and 46

22

Giorgetta, 2008). Therefore some modeling groups have decided to impose the QBO by 1 assimilation techniques (i.e., nudging, see Chapter 2) of either the equatorial zonal winds 2 or the vorticity. The CCMs that assimilate the QBO in the REF-B1 simulation are 3 reported in Table 2.9 of Chapter 2, and are the Group C Models of Table 8.5.1. Although 4 the QBO assimilation should alleviate the biases in the ozone distribution associated with 5 the problem of properly representing the QBO, the assimilation technique removes the 6 predictive capability of a model. It is therefore possible to evaluate the response of ozone 7 to a prescribed QBO forcing, but a prediction of future ozone behavior related to the 8 QBO is impossible with this methodology. 9 10 8.5.1 Equatorial Variability and the QBO signal in the stratosphere 11 12 Figure 8.5.1 shows the vertical profile of the variability of zonal mean zonal wind (left) 13 and ozone in DU/km (right) at the Equator (average 5°S-5°N) computed as the standard 14 deviation of the monthly values for the period 1960-1999. These results are compared 15 with observations (black lines), respectively the ERA-40 zonal mean wind and the 16 Randel&Wu-3D ozone datasets. In both model and observational data, the linear trend 17 and the annual cycle have been removed. In addition, a band pass filter has been applied 18 to the time series to extract only those oscillations with periods between 9-48 months. 19 20 The upper panels of Figure 8.5.1 include only the models with nudged QBO (the models 21 in Group C of Table 8.5.1, see also Table 2.9 in Chapter 2), while the bottom panels 22 include the rest of the models (both Groups A and B of Table 8.5.1). 23 24 The models with in Group C are characterized by substantial variability of up to ~18 ms-1 25 in zonal mean zonal wind and in the range 0.7 to 1.5 DU/km in ozone, as expected 26 because of the assimilation. In addition to the main peak near 20 hPa in zonal mean zonal 27 wind, some models (Niwa_SOCOL, SOCOL, and to a lesser extent WACCM) show a 28 secondary peak in zonal wind variability near 1 hPa. This variability could be an 29 excessive QBO modulation of SAO at these altitudes, possibly a side effect of the applied 30 nudging. 31 32 In the models that did not assimilate the QBO (Figure 8.5.1, lower panels), the zonal 33 wind variability clusters instead into two groups: 4 models (GEOSCCM, LMDZrepro, 34 CNRM-ACM, and CMAM) have variability less than 5 m/s (Group A); and 5 models 35 (AMTRAC3, MRI, UMUKCA-METO, UMUKCA-UCAM, and UMSLIMCAT) have 36 variability in the range 7 to 20 m/s (Group B). Group A severely underestimates the zonal 37 wind variability, leading to the conclusion that the QBO in zonal wind is not internally 38 generated to a sufficient degree in these models. Consistently, for these models the QBO 39 basis functions in the MLR analysis are set to zero (see Table 8.5.1). The variability in 40 Group B is much more realistic when compared with ERA-40 data, although the 41 maximum amplitude is generally underestimated and is located at lower pressure (i.e. 42 higher in the atmosphere) than observed. 43 44 The observed ozone interannual variability (Figure 8.5.1, right panels) shows two 45 maxima (10 and 30 hPa). These maxima are due to, respectively, the modulation of the 46

23

ozone chemistry in the middle stratosphere (see Chapter 6) and the advection of ozone by 1 the secondary meridional circulation (see Chapter 5) in the lower stratosphere (Gray and 2 Chipperfield 1990). The models with nudged QBO (Group C, upper right panel) show the 3 clear double peak structure, in phase with the Randel&Wu-3D observations, although 4 with a wide range of magnitudes. The models without nudging (lower panel) that showed 5 little variance in wind at the Equator also simulate little variance in ozone (Group A). The 6 exception in Group A is the CNRM-ACM model, with a 0.6 DU/km peak in ozone 7 variability at 30 hPa. The time series of the ozone vertical distribution shown in the 8 supplementary material for CNRM-ACM shown that these variations are not downward 9 propagating, consistently with the fact that this models does not simulate the QBO. 10 Possibly, these variations are associated with ENSO (that could still be present in the 11 band pass used). A similar behaviour was previously reported for a CCMVal-1 model 12 (Punge and Giorgetta, 2008). With the exception of UMUKCA-METO, the models in 13 Group B show the double peak in ozone variability, each of them to a different degree. 14 However, aside from this very broad comparison, there does not seem to be a linear 15 relationship between the variability in zonal winds and ozone in Groups B and C, 16 suggesting a range of sensitivity of the ozone to the zonal wind QBO, which is 17 independent of whether it is imposed or internally generated. In particular, in Group B the 18 UMSLIMCAT model appears to be characterized by low ozone sensitivity, with higher 19 than observed wind variability but half than observed ozone variability at 30 hPa. In 20 Group C, the ULAQ and WACCM models appear to have a higher than observed ozone 21 sensitivity, while the Niwa_SOCOL and SOCOL models lower than observed. Note that 22 the two SOCOL models and WACCM are very close to observations in their zonal wind 23 variability at 30 hPa, while they differ of a factor two in their ozone variability. 24 25 An alternative measure of the CCMs’ representation of the ozone QBO is the vertical 26 distribution of the annual mean equatorial (5°S-5°N) QBO regression coefficients from 27 the MLR analysis (which is represented in terms of ozone mixing ratios for all CCMs 28 and NIWA-3D observations and DU/km for the Randel&Wu ozone). These coefficients 29 are shown in Figure 8.5.2a for the models in Group B (internal QBO-like oscillation) and 30 in Figure 8.5.2b for models with nudged QBO (Group C), together with the 31 corresponding coefficients from the analysis of observations. For a better comparison of 32 the QBO signal between the CCMs and CCMs and observations, the sum of the two QBO 33 regression coefficients has been multiplied by typical mean QBO amplitude of 30 m/s. 34 All CCMs in Group C (right panel), but WACCM, capture well the vertical structure of 35 the QBO signal. However, all the Group C models show a much larger effect on ozone in 36 the lowermost stratosphere than observations, while they reasonably match the 37 observations at 30 hPa. The overestimation in the lowermost stratosphere (around 100 38 hPa is found for UMUKCA-METO only, in Group B. The AMTRAC3 model has indeed 39 little signal, consistently with the small standard deviations in zonal mean wind and 40 ozone, largely underestimated (compare Figure 8.5.1). 41 42 8.5.2 QBO signal in column ozone 43 44 The latitudinal distribution of the combined annual-mean QBO coefficients from the 45 MLR analysis of column ozone amounts is presented in Figure 8.5.3, Group B models 46

24

(UMUKCA-UCAM missing because column ozone not provided) are shown in panel (a) 1 and Group C models in panel (b). The corresponding signal from the TOMS+gb data 2 analysis for the 1964-2004 period is also shown. In Group B (panel a), the latitudinal 3 structure, characterized by phase changes in the subtropics (15-20degrees) and at 4 middle/high latitudes, is very well represented for UMUKCA-METO and MRI (except 5 for southern high latitudes), while AMTRAC3 and UMSLIMCAT features a flat 6 latitudinal contribution of the QBO-like oscillation on total ozone. The CCMs that nudge 7 a QBO (panel b) show a reasonable agreement with the latitudinal structure of the QBO 8 in total ozone, especially in the subtropics. The ULAQ model shows rather larger 9 amplitude variations than any other model, perhaps associated with the overestimation of 10 the QBO contribution from ~10 hPa evident in Figure 8.5.1 (top, right) and therefore 11 indicating problems with its nudging technique. The SOCOL and Niwa-SOCOL model 12 tend to underestimate the observations away from the subtropics, while WACCM, 13 CAM3.5, and EMAC tend to overestimate them. CCSRNIES and E39CA show excellent 14 agreement with observations for these diagnostics. 15 16 Figure 8.5.4 shows the temporal evolution of the combination of the two QBO regression 17 coefficients of the MLR analysis for equatorial column ozone (averaged from 5°S-5°N) 18 for the CCMVal-2 models and TOMS+bg data. As expected, models that nudge the QBO 19 closely follow the observed QBO in column ozone, although they largely overestimate 20 the amplitude: the observations show a ±5 DU variation, while the CCMs show more 21 than ±10 DU. The largest overestimations occurs for ULAQ, E39C, and WACCM, 22 consistent with the largest standard deviation in ozone in Figure 8.5.1. Since the 23 equatorial column ozone QBO is primarily a result of QBO-induced vertical motion in 24 the lower to mid stratosphere, this discrepancy may arise from a number of sources, 25 including an error in the magnitude of the induced vertical motions, in the vertical 26 gradient of ozone in the vicinity of the induced motions or in the factors that determine 27 whether the QBO ozone response is controlled by dynamical or radiative/chemical 28 processes. 29 30 We do not expect the group B models to be in phase with observations, but it is apparent 31 that that the observed amplitude of the QBO signal in ozone is much better captured by 32 the models that internally generate a QBO (Fig. 8.5.4b). This is especially true for the 33 MRI and the UMSLIMCAT model. This may be because these models generally 34 underestimate the zonal wind variability. The AMTRAC3 model shows very little 35 tropical variations, consistent with the very weak “QBO-like” zonal wind oscillation in 36 the lower stratosphere, which is the dominant influence on column ozone variations. 37 38 8.6 ENSO signal in ozone 39 40 The El Niño Southern Oscillation (ENSO) is a tropical atmosphere-ocean phenomenon 41 and a source of large-scale climate variability for the atmosphere–ocean system. Its 42 influence on the stratosphere has been increasingly recognized, with the advent of 43 ensemble modeling and with the availability of longer observational datasets. Most of the 44 published work focused on the ENSO signal in the polar lower stratosphere, because of 45 the established teleconnections between the warm phases of ENSO and the mid latitude 46

25

North Pacific region (e.g. Hoerling et al., 1997) which can favor the enhancement of mid-1 latitude planetary waves and their upward propagation into the stratosphere. Due to this 2 increase in extra-tropical stratospheric planetary wave activity, warm ENSO events have 3 been found to be associated with anomalous warming and anomalously high geopotential 4 height in the polar stratosphere, both from observations (van Loon and Labitzke, 1987; 5 Hamilton, 1993; Brönnimann et al., 2004; Camp and Tung, 2007; Garfinkel and 6 Hartmann 2007) and comprehensive modeling of the troposphere-stratosphere system 7 (Sassi et al., 2004; Taguchi and Hartmann, 2006; Manzini et al., 2006; Garcia-Herrera et 8 al., 2006; Bell et al. 2009). These signals in temperature have been found consistent with 9 signals in ozone during ENSO (Fischer et al., 2008; Steinbrecht et al., 2006; Brönnimann 10 et al., 2006) 11 12 The polar warming and enhanced ozone associated with warm ENSO events are a 13 manifestation of a stronger Brewer-Dobson circulation during ENSO and a signal in both 14 temperature and ozone is therefore also expected in the tropics (Cagnazzo et al 2009; 15 Free and Seidel, 2009; Randel et al 2009). 16 17 The ENSO tropical signals in zonal monthly temperature and ozone from the MLRA 18 analysis are shown in Figure 8.6.1. In the lowermost stratosphere (~70 hPa) the 19 temperature ENSO signal is mostly consistent between the CCMs: most models show a 20 cooling that surround the observed cooling of ~1 K. In the troposphere, the typical ENSO 21 warming is about 0.6 K, generally lower than that estimated by the models (as well as 22 other observations, Free and Seidel, 2009). The modeled ENSO tropical signal in ozone 23 show a large spread, although the models that show a signal, tend to have it peaking in a 24 narrow layer between 100 and 0 hPa. In this case also the uncertainty in the observations 25 appears to be large (compare the NIWA and the Randel&Wu datasets). For the Northern 26 polar cap (Figure 8.6.2), the observations show the typical warming in the lower 27 stratosphere, and the model results envelop the observations, mostly showing a warming, 28 as expected. The ozone ENSO signal in the model also tends to be positive, although the 29 spread of the model responses is clearly large. 30 31 An alternative measure of the CCMs representation of the ENSO signal in ozone is to 32 follow the methodology of Cagnazzo et al (2009) and recompute their Figure 5 (based on 33 he CCMVal-1 models) CCMVal2 models (Figure 8.6.3), which shows the relationship 34 between the north polar cap ENSO response in the temperature and column ozone fields 35 in February-March. As in Cagnazzo et al. (2009) the ENSO signal has been extracted by 36 calculating difference fields between composites of warm ENSO and NEUTRAL years. 37 Warm ENSO years are defined as the four largest events in the period 1980-1999 and 38 NEUTRAL years are the remaining years when both the four largest warm and cold 39 ENSO events have been excluded. During the period (1980-1999), the cold ENSO events 40 are smaller in magnitudes and have not been found to significantly affect the stratosphere 41 (Manzini et al., 2006). 42 43 In agreement with Cagnazzo et al. (2009), a clear positive correlation is found between 44 the modeled colum ozone and temperature anomalies at high latitude (0.87, significant at 45 more than 99.9%) supporting the idea that anomalies in temperature and column ozone 46

26

are influenced by the same (Brewer-Dobson circulation) mechanism. However, in Figure 1 8.6.2 there is a less distinct dominance of cases in the upper right quadrant, indicating 2 positive temperature anomalies and increased ozone during ENSO, than for the CCMVal-3 1 models discussed by Cagnazzo et al (2009). Given that a positive relationship is 4 expected (see also section 8.4), these results suggest that the spread of the CCMVal-2 5 modeled response is influenced more by internal variability than that of the CCMVal-1 6 models. Possible reasons for this change are the sea surface temperature used and/or the 7 updates to the models since CCCMVal-1, particularly those related to the inclusion of 8 additional forcing that may interfere with the ENSO signal, such as the QBO and the 9 aerosols from volcanic eruptions. 10 11 8.7 Volcanoes and Stratospheric Aerosols 12 13 The volcanic impact on stratospheric ozone depends on (a) the amount of injected 14 material, in particular sulfur, (b) the geographical distribution of the aerosols, and (c) the 15 levels of EESC loading. The ozone changes are related to the effect of aerosols on the 16 chemical and radiative conditions of the lower stratosphere. Aerosols provide surfaces for 17 heterogeneous reactions to occur, while also absorbing, reflecting and scattering incident 18 radiation. Dynamical and chemical ozone changes are not independent of each other. 19 Dynamical changes resulting from major volcanic eruptions contribute to ozone changes, 20 while heterogeneous chemistry occurring on aerosol surfaces affects ozone 21 concentrations, producing an additional indirect radiative impact. 22

23 Observed column ozone reduction after the Mt. Pinatubo and the El Chichón eruption 24 range from about 2% in the tropics to about 5% (Pinatubo) and 2-3% (El Chichón) in mid 25 latitudes (Angell, 1997; Solomon et al., 1998). Very large ozone losses were observed 26 after the Mt. Pinatubo eruption at high northern latitudes in February and in March, for 27 example Randel et al. (1995) found losses of 10% in ozone total column in 1992 28 northward of 60N and 10-12% in 1993. Ozone sonde profiles after the Mt. Pinatubo 29 eruption show that the concentration did not decrease uniformly at all altitudes (Hofmann 30 et al., 1993; Grant et al., 1994). After the Agung eruption in 1963 a slight increase in 31 global total column ozone was found (Angell, 1997), possibly due to the suppression of 32 nitrogen oxides in the low-chlorine conditions (Tie and Brasseur, 1995). 33

34 The methods used to simulate the volcanic impact in the CCMs have been introduced in 35 detail in Chapter 2. Heterogeneous chemical reactions on the volcanic aerosol surfaces 36 are calculated using a prescribed zonal mean aerosol surface area density (SAD) time 37 series. In the CCMVal-2 model runs, most models (all but AMTRAC3, E39CA, EMAC, 38 and GEOSCCM) have prescribed SADs using the dataset compiled and made available 39 through the SPARC Assessment of Stratospheric Aerosol Properties (Thomason, 2006). 40 The radiative effects of volcanic aerosols have been incorporated into the model in a 41 number of different ways or, in some cases, completely neglected. Table 2.10 42 summarizes the different methods used by the different models, which include (1) no 43 simulation of direct radiative effects, (2) prescribed heating rate anomalies based on off-44 line radiative calculations, (3) on-line radiative calculations using aerosol properties 45 estimated from observations, (4) on-line radiative calculations using optical depths 46

27

derived from the SPARC SAD data setand (5) full microphysical modeling of volcanic 1 aerosols based on prescribed stratospheric influx of volcanic SO2. 2

3 The result of volcanic forcing on stratospheric temperatures can be seen most simply 4 through inspection of global-mean, annual-mean temperature timeseries. These are shown 5 at 50 hPa in Figure 8.7.1 (top panel), as anomalies from pre-volcanic conditions for the 6 three eruptions of the 1960-2000 time period: Agung (1963), El Chichón (1982) and Mt. 7 Pinatubo (1991). The anomalies are calculated as deviations from the mean of the 5 years 8 (3 years for Agung) preceding the year of the eruption. There is a considerable spread in 9 the post-volcanic eruption temperatures in the models. For example, in 1992 after the 10 Pinatubo eruption, the changes in 50 hPa temperature range from +9 to -1 K, while the 11 observations show a +0.5-1 K change. CNRM-ACM appears as an outlier in this 12 diagnostic, with temperature increases much larger than the other models or the 13 observations. The cause of the temperature increases is related to how the radiative 14 scheme responds to the volcanic aerosols. Subsequent runs of the CNRM-ACM model, in 15 which the aerosol properties have been modified to exhibit less absorption, have shown 16 temperature evolution in the range of that of the CCMVal-2 CCMs (Martine Michou, 17 personal communication, 2009). The temperature behavior of the models is strongly 18 dependent on the parameterization method employed to simulate the direct radiative 19 effects of volcanic aerosol loading. In the lower panel of Figure 8.7.1 the same 20 temperature anomalies have been replotted, but color-coded by parameterization method. 21 This plot shows that using aerosol optical depths derived from the SPARC SADs leads, at 22 least in the Pinatubo and Agung eruptions, to anomalously large temperature 23 perturbations compared to those estimated from the ERA-40 dataset. Those models that 24 did not include radiative effects of volcanic aerosols show little change in 50 hPa 25 temperature, although two models show slight decreases after the Pinatubo eruption, as 26 might be expected due to chemical induced ozone decreases. Finally, the models which 27 employ optical depth estimates from GISS, and those which use prescribed heating rates 28 show (for some models) quite good agreement with the observations but also a large 29 spread. 30

31

Inspection of the vertical structure of the temperature anomalies can help evaluate the 32 form of the discrepancies between models. Figure 8.7.2 (LHS) shows the annual mean 33 regression coefficients for temperature in the tropics, where the temperature increases are 34 largest, and can have an impact on the general circulation. The structure of the 35 temperature perturbations is generally consistent between the models, with maximum 36 heating at ~50 hPa (20 km), in good agreement with observations. There is excellent 37 agreement between the models which show the largest regression coefficient values in the 38 region of maximum heating in Figure 8.7.2 and those that show the largest temperature 39 changes in Figure 8.7.1. A number of outliers in Figure 8.7.1 also show deviations from 40 the general vertical structure of the temperature regression coefficients. For example, 41 CCSRNIES, which showed post-volcanic cooling at 50 hPa is seen in Figure 8.7.2 to 42 show positive regression coefficients only at heights above 40 hPa, and negative ones 43 between 50 and 100 hPa. On the other hand, post-eruption heating in the EMAC model is 44 restricted to heights below 50 hPa, which helps explain why the EMAC 50 hPa 45

28

temperature anomalies of Figure 8.7.1 are different from the other models using 1 prescribed heating rates. 2

3 Figure 8.7.3 shows global-mean, annual-mean total-column ozone anomalies compared 4 with pre-volcanic conditions2. Local minima in the years just after the El Chichón and 5 Pinatubo eruptions are associated with the effects of the volcanic aerosols. The observed 6 anomalies after the El Chichón and Pinatubo eruptions were of the order of 10 DU. There 7 is a large degree of scatter in the model results, ranging from some models showing post-8 volcanic decreases of up to 15-20 DU (CCSRNIES, MRI, ULAQ) and, for El Chichón, 9 small post-eruption increases (EMAC,UMUKCA-METO). For the Agung eruption, some 10 models seem to show a slight increase in total ozone in the year of the eruption, however, 11 it is impossible to attribute any ozone changes to the volcanic effects, as the spread in 12 modeled values stays relatively constant over the time span shown. The vertical 13 distribution of tropical volcanic ozone loss is shown in Figure 8.7.2 (right-hand column), 14 in terms of the annual-mean tropical regression coefficients for the different models. 15 They show a structure characteristic of post-eruption tropical ozone behavior, with ozone 16 loss in the lower stratosphere, and ozone increases above (as observed for the Pinatubo 17 eruption by Grant et al. 1992, 1994). This structure can be explained by enhanced vertical 18 transport due to the lower stratospheric heating produced by aerosol heating. While the 19 enhancement of ozone creation and loss are comparable in terms of mixing ratio, the 20 increased atmospheric density in the lower stratosphere means that the net effect of 21 enhanced vertical transport leads to a reduction of the total column. Comparing the left 22 and right-hand columns of Figure 8.7.2, the models that show the largest temperature 23 changes (CNRM-ACM, WACCM, SOCOL, Niwa-SOCOL, MRI) also show the largest 24 ozone perturbations in the tropical lower stratosphere. 25

26 Slight differences in the vertical structure of the regression coefficients can help shed 27 light on why the global-mean total ozone time-series in Figure 8.7.3 differ. The models 28 generally show the largest ozone loss at 30 hPa (25 km). After Pinatubo and El Chichón, 29 two models (CCSRNIES and ULAQ) show negative regression coefficients at lower 30 heights than the other models and these two models also have the largest total ozone 31 losses in Figure 8.7.3. 32

33

Since a large amount of volcano-related ozone loss is related to heterogeneous chemistry, 34 one would expect the models with largest ozone loss to have the largest amounts of 35 chlorine activation. Figure 8.7.4 confirms this, showing the ozone anomaly in the year 36 following each eruption as a function of the anomaly in ClO at 50 hPa. For each eruption, 37 one sees a relatively linear relationship between ozone loss and chlorine activation. Note 38 that by choosing to look only at the year after each eruption, the relationship between 39 ClO and ozone for CNRM-ACM is not well represented by these plots, since this model 40 displays maximum ClO and ozone anomalies three years after each eruption, and in fact 41 shows negative ClO anomalies for the first year after the Mt. Pinatubo eruption (with 42 large increases afterwards). Latitude-time plots of ClO and total ozone abundances (not 43

2 Note that in these plots the anomalies are the result of a number of factors including volcanic effects but also the EESC related trend and the QBO.

29

shown) confirm that the models with largest total ozone loss, including CCSNRIES and 1 ULAQ, are characterized by chlorine activation and ozone loss extending from the tropics 2 to the high latitudes. Thus, the cause of the anomalous ozone loss in these models is the 3 anomalous chlorine activation, which may itself be due to biases in quantities such as the 4 total chlorine or lower stratospheric temperature. 5 6 7 8.8 Summary 8 9 Although the MRL analysis is a powerful tool for synthesizing the relative influence of 10 the variability sources on natural ozone variation, it cannot take into account the fact that 11 the net effect of the natural variations on ozone is usually a non-linear combination of the 12 single contributions of variability factors. Nonlinearities have been reported for the 13 combined ENSO and QBO signal (Calvo et al., 2009), the solar-QBO and volcanic 14 signals (Lee and Smith, 2003), solar-QBO signals (Smith and Matthes, 2008; Matthes et 15 al., 2009a,b, Camp and Tung 2008), the solar-SST signal (Marsh and Garcia, 2007; 16 Austin et al., 2008), and ENSO, QBO, and solar interconnections (e.g., Kryjov and Park, 17 2007; Kuroda, 2007; Kodera et al., 2007). Many of these interconnections of the natural 18 variability sources are objectives of current research and we can therefore only assess the 19 MLR results shown in this chapter. 20 21 Another limitation of the assessment in this chapter is the relatively short observational 22 record which limits the statistical significance of many of the shown responses to 23 individual components. This is especially true for the 11-year solar cycle where only data 24 for two and a half cycles are available. Additionally, large volcanic eruptions took place 25 during solar maximum phases of the solar cycle and may impact the solar response. And 26 for a composite analysis of solar cycle, QBO, and ENSO a limited amount of years is 27 available. Additional limitations occur due to the MLR used. We get consistently lower 28 signals than Randel and Wu (2007) and Randel et al. (2009) and conclude that this is 29 related to the different number and also the treatment of the basis functions. 30 31 However, more reliable data are available for the annual cycle and we have therefore 32 used some selected diagnostics to quantify and summarize the models performance to 33 natural forcings with a skill function. An exception is the QBO since the modeling of this 34 phenomenon in CCMs is in a too primitive stage to apply performance metrics. 35 36 The skill function introduced by Taylor (2001; equation 4) is applied to the Taylor 37 diagrams shown in sections 8.2 and 8.4. The results are summarized in matrix form in 38 Figure 8.8.1. The models behind the numbers are reported in Table 8.8.1. Models are best 39 performing for global diagnostics (see O3-MEAN from Figure 8.2.3, and 40 TOZ_ANNUAL from Figure 8.2.5). Slightly less but still very high model skills are 41 found for the amplitudes of the annual and semiannual cycles (see O3-ANNUAL from 42 Figure 8.2.4a and O3-SemiA from Figure 8.2.4b). To capture the vertical structure of the 43 solar cycle signal is instead more difficult (Figure 8.4.2), although a number of models 44 shows skills between 0.7-0.9. 45 46

30

In order to approach the question of model performance per individual natural variability 1 component and the importance of the different factors for the evolution and prediction of 2 stratospheric ozone, Fig. 8.8.2 shows the MLR regression coefficients for total column 3 ozone for CCMVal-2 models relative to the observed regression coefficients from the 4 TOMS+gb data in Figure 8.1 in the order of their relative importance from top to bottom, 5 i.e., the annual cycle, the solar cycle, the QBO, ENSO, volcanos, and at the end also the 6 anthropogenic long-term trend related to atmospheric chlorine loading (EESC). 7 8 The results of the annual cycle were already presented in Fig. 8.8.1, and the results for the 9 solar cycle are also shown in Figure 8.4.1. It is apparent that the performance of the 10 individual models for the annual cycle, the solar cycle and the long-term secular trend 11 due to increased atmospheric chlorine loading is high. All other natural variability 12 components are less well reproduced. The QBO is very difficult to internally generate, so 13 models either do not have a QBO (marked with dots those excluded from the MLR 14 analysis), or they nudge the QBO and overestimate total column ozone variations or 15 internally generate a QBO and underestimate total column ozone variations. Uncertainties 16 in the observations and large interannual variability limit the assessment of the model 17 performance for ENSO and volcanoes indicated by low values in Figure 8.8.2. 18 19 Summary on annual cycle 20 21 The assessment of the performance of the models for the annual cycle in ozone that is 22 summarized in Figure 8.8.1, shows that the vertical and latitudinal distribution of the 23 annual cycle in stratospheric zonal monthly mean ozone is quite well represented in the 24 models. The only outlier found is the CAM3.5 model, possibly because of its low top 25 (note that data were not available for E39C, another low top model, for this diagnostic, at 26 the time of writing). 27 28 The comparison with the MLS data shows that the processes leading to the annual cycle 29 in the upper stratosphere are well captured by the models: The anti-correlations between 30 temperature and ozone at 1 hPa are broadly captured (simple check of photolysis 31 scheme). The magnitude of the annual cycle and the transition to summer conditions are 32 also well reproduced by the models. However, in the lower stratosphere a few models 33 (UMUKCA-METO and UMUKCA-UCAM) do not reproduce the anthropogenic 34 deviation of the annual cycle (polar ozone depletion) that dominates the late winter and 35 spring in the SH. 36 37 The amplitude of the annual cycle in zonal mean column ozone is quite well represented 38 in the models, see the skill of TOZ_ANNUAL in Figure 8.8.1. However, as noted above 39 for the vertical and latitudinal distribution, a few models fail to produce the seasonal 40 Antarctic ozone depletion (CAM3.5, E39C, EMAC, UMUKCA-METO, and UMUKCA-41 UCAM). 42 43 In comparison with HALOE, the CCMVal-2 models show a larger spread in their 44 response to the annual cycle, in the NH and SH spring, upper troposphere and lower 45 stratosphere, than the CCMVal-1 models. CCMVal-2 model outliers are: CCSRNIES, 46

31

which shows unusually large ozone values at 10 hPa in March at the Equator and in 1 October at 80S; and CNRM-ACM at 50 hPa, in March. 2 3 Summary on interannual polar variability 4 5 The observed annual cycle in column ozone variability is well reproduced by all models, 6 in the sense that all show a minimum in variability in the respective summer seasons. In 7 the NH dynamically active period, most of the models underestimate the interannual 8 polar variability, indicating a common bias. Models that are biased low: CMAM and 9 UMSLIMCAT. Models that are biased high: MRI and WACCM (late spring). During the 10 SH dynamically active period, the model spread envelops the observations, suggesting a 11 wide variety of causes for the model biases. Models that are biased low: GEOSCCM and 12 UMUKCA-UCAM. Models that are biased high: CAM3.5, CCSRNIES (both more in 13 late summer) and CMAM. A plausible explanation for this hemispheric difference in the 14 interannual variability model biases is the known sensitivity of the SH large scale 15 dynamics to the parameterization of non-orographic gravity wave drag. 16 17 The majority of models reproduce quite well the relationship between winter mean heat 18 flux and spring-to-fall ozone ratio in both the NH and SH. This result indicates that the 19 sensitivity of ozone to the heat fluxes is realistic. The only obvious outlier is the ULAQ 20 model, which appears to severely underestimate the relationship in the NH. 21 22 The models reproduce the observed ozone-temperature relationship quite well, although 23 for the Northern Hemisphere for a number of models the ozone is less responsive to 24 temperature perturbations than in the observations. Among the models with low 25 sensitivity are CMAM and UMSLIMCAT, models with also particularly low ozone 26 standard deviation. The obvious outlier is CNRM-ACM, substantially overestimating the 27 relationship. In the SH, the spread of the models surrounds the observations, as in the 28 case of the ozone standard deviation, but with correspondence does not hold by model: 29 for instance GEOSCCM overestimate the slope but has little variability, while CMAM 30 has a realist slope but overestimates the variability. 31 32 The regression of the column ozone on the simplified AM index further confirms that the 33 modeled interannual polar ozone variations are due to the known dynamical processes 34 affecting the variability of the stratospheric vortex and that these processes and their 35 connection to ozone are generally well simulated. However, in the NH the low 36 sensitivity of ozone to temperature in the majority of models is in contrast with the 37 assessment of the ozone sensitivity to wave drag diagnosed from the heat fluxes. This 38 contrasting result suggests that either the temperature and ozone responses to wave 39 driving are not fully consistent in the models (ozone and temperature transport numerical 40 schemes are usually different, see Chapter 2), or that chemical processes alter the ozone-41 temperature relationship expected from dynamical processes, in different ways (because 42 of modeling biases) in the models and the observations. 43 44 45 46

32

Summary on 11-year solar cycle 1 2 Most CCMs imposed a solar cycle in the CCMVal2 REF-B1 simulations (sc group), four 3 CCMs did not (no-sc group, i.e. GEOSCCM, ULAQ, UMUKCA-METO, UMUKCA-4 UCAM). The solar cycle in total column ozone is well represented in the CCMs of the sc 5 group, although with a large amplitude spread. Most models reproduce half of the 6 observed solar total column ozone variations from 60S to 60N, but those are biased high 7 (Randel and Wu, 2007). CMAM and E39C show best agreement with observations, 8 CAM3.5 lowest agreement. CMAM is one of the models that perform well in the solar 9 radiation intercomparison. The vertical structure of the tropical solar signal in ozone and 10 temperature is more difficult to model. While the direct solar response in temperature and 11 ozone in the upper stratosphere is good represented (best for WACCM, CMAM, 12 AMTRAC3, and UMSLIMCAT, lowest for LMDZrepro) the vertical structure in the 13 tropics below 10 hPa varies a lot among the CCMs but also among different 14 observational data sets. Especially in the lower stratosphere uncertainties are large and 15 might be related to the non-linear interactions of a number of signals (solar, QBO, ENSO, 16 volcanos) that might be not handled correctly in a MLR as discussed earlier. The large 17 spread in the vertical solar ozone response in the tropics is also shown in a Taylor 18 diagram (Fig. 8.4.3). Another limiting factor might be the fact that we only used one 19 ensemble of each CCM, an ensemble mean for the models that delivered several 20 ensembles might limit the large uncertainties in the middle and lower stratosphere as 21 shown by Austin et al. (2008). In general the agreement between the CCMs and between 22 the models and observations is better for ozone than for temperature. The latitudinal 23 representation of the solar response in total column ozone shows improved representation 24 to CCMVal-1 but a large spread especially at mid to high latitudes due to large 25 interannual variability. 26 27 Compared to CCMVal-1 the representation of the solar cycle in radiation and chemistry 28 has been improved since the scaling with the F10.7cm solar radio flux has been 29 exchanged by prescribing daily varying spectrally resolved irradiance data from the 30 SOLARIS project (Matthes et al., 2007). Nevertheless, the response in the CCMs shows 31 large differences related to either the performance of the radiation schemes (compare 32 chapter 3, section 3.6) and the prescription of total solar irradiance or spectrally resolved 33 data, the photolysis scheme (compare chapter 6) as well as dynamical and transport 34 differences that are very difficult to separate. 35 36 Summary on QBO 37 38 Metrics are not computed for the QBO signal in ozone because the status of the 39 modelling of the QBO in CCM is judged to be still at a primitive stage. Some AGCM in 40 recent years have been able to simulate a quite realistic QBO in zonal winds and related 41 dynamical quantities, but it does not seem that this expertise has passed to the CCMs, 42 possibly also because of the computational and/or developmental constrain of the 43 additional chemical modelling. We remark that the widespread technique (8 out of 17 or 44 18 models) of nudging the QBO induces substantial errors, in addition to be conceptually 45 inconsistent with the objective of a predictive dynamical model. These induced errors are 46

33

in cases as large as the QBO signal in ozone. We conclude that QBO modelling in the 1 CCMs as implemented for CCMval-2 is an outstanding problem. 2 3 In summary, there are three groups of models, listed in Table 8.5.1: Group A with 4 negligible tropical variability, Group B with intermediate to large tropical variability, and 5 the Group C models, with externally imposed tropical variability. It is concluded that the 6 QBO is not simulated in all the models of Group A and in the AMTRAC3 model of in 7 Group B. While for the MRI, UMSLIMCAT, and UMUKCA-METO models of Group B, 8 the QBO is reasonably well represented. We do not judge UMUKCA-UCAM, because no 9 column ozone was provided, and the analysis is limited. 10 11 From the MLR analysis for the QBO signal, large overestimations of the amplitude of 12 column ozone variations for the models with nudged QBO are reported. The amplitude is 13 much better represented for the MRI, UMSLIMCAT, and UMUKCA-METO models, 14 with a spontaneous (i.e., internally generated) QBO. Possibly because the consistency 15 provided by a spontaneous QBO improves the simulation of the secondary meridional 16 residual circulation associated with the QBO and responsible for the QBO signal in 17 column ozone. Moreover, it is found that the latitudinal representation of colum ozone 18 signal for nudged and spontaneous QBO is captured well, consistently with the broad 19 agreement at 20-30 hPa, but for the UMSLIMCAT model. 20 21 Summary on ENSO 22 23 In the case of the ENSO signal in the ozone vertical profiles, it is hard to judge if the 24 modelled ozone variations are consistent with the observations, because of the large 25 uncertainty in the observations. The ENSO signal in temperature is more clearly 26 emerging from the models. Especially in the tropical lower stratosphere, where most of 27 the models show a cooling in agreement with observations, both in the location and in the 28 strength. 29 30 By looking at column ozone a clear picture emerge, with the spread of the model 31 responses explained by interannual variability. Note indeed that the slope (from the 32 ensemble of models) deduced by Figure 8.6.3 is consistent with the slope estimated by 33 observations in Figure 8.3.3. 34 35 Because of the large role of interannual variability and the uncertainty in the 36 observations, it is not possible to measure the model performance in the simulation of the 37 ENSO signal in ozone. 38 39 Summary on volcanic aerosols 40 41 CCMVal-2 REF-B1 runs show a considerable spread in their simulated response to 42 volcanic eruptions as already discussed in Fig. 8.8.2, and seen in examination of modeled 43 temperature and ozone. The fact that many fundamentally different methods have been 44 employed to parameterize the direct effect of volcanic aerosols on the radiative transfer of 45 the stratosphere (Fig. 8.7.1b) helps explain, at least in part, the wide range of post-46

34

eruption temperature anomalies seen in the different models. For example, models that 1 estimate aerosol optical depth from the SAD data set of Thomason (2006) consistently 2 overestimate lower stratospheric temperatures after the Mt. Pinatubo eruption compared 3 to the ERA-40 data set. On the other hand, models which use the GISS aerosol optical 4 depth data set lead to wide ranging estimates of lower stratospheric heating. Post-eruption 5 changes in total column ozone are well correlated with changes in lower stratospheric 6 ClO. It thus appears that while most models use a common aerosol SAD data set to drive 7 anomalous post-eruption chemistry, the models display differing degrees of sensitivity to 8 those aerosols, leading to differing amounts of chlorine activation and associated ozone 9 loss. 10 11 12

13

35

References 1 2 Ammann, C., G. Meehl, W. Washington, and C. Zender, 2003. A monthly and latitudinally varying forcing 3 data set in simulations of 20th century climate, Geophys. Res. Lett., 30(12), 1657, 4 doi:10.1029/2003GL016875. 5 6 Angell, J. K., 1997. Estimated impact of Agung, El Chichón and Pinatubo volcanic eruptions on global and 7 regional total ozone after adjustment for the QBO, Geophys. Res. Lett., 24, 647-650. 8 9 Austin, J., K. Tourpali, E. Rozanov, H. Akiyoshi, S. Bekki, G. Bodeker, C. Brhl, N. Butchart, M. 10 Chipperfield, M. Deushi, V.I. Fomichev, M.A. Giorgetta, L. Gray, K. Kodera, D. Kinnison, E. Manzini, D. 11 Marsh, K. Matthes, T. Nagashima, K. Shibata, R.S. Stolarski, H. Struthers, and W. Tian, 2008. Coupled 12 chemistry climate model simulations of the solar cycle in ozone and temperature, J. Geophys. Res. 113, 13 D11306, doi:10.1029/2007JD009391. 14 15 Baldwin, M. P., and L. J. Gray, 2005. Tropical stratospheric zonal winds in ECMWF ERA-40 reanalysis, 16 rocketsonde data, and rawinsonde data, Geophys. Res. Letts., 32, doi:10.1029/2004GL022328. 17 18 Baldwin, M. P., L. J. Gray, T. J. Dunkerton, K. Hamilton, P. H. Haynes, W. J. Randel, J. R. Holton, M. J. 19 Alexander, I. Hirota, T. Horinouchi, D. B. A. Jones, J. S. Kinnersley, C. Marquardt, K. Sato, and M. 20 Takahashi, 2001. The Quasi-Biennial Oscillation, Rev. Geophys., 39, 179-229. 21 22 Bodeker, G.E, J. C. Scott, K. Kreher, and R. L. McKenzie, 2001. Global ozone trends in potential vorticity 23 coordinates using TOMS and GOME intercompared against the Dobson network: 1978-1998, J. Geophys. 24 Res., 106, 23,029-23,042. 25 26 Bodeker, G.E., I. S. Boyd and W. A. Matthews, 1998. Trends and variability in vertical ozone and 27 temperature profiles measured by ozonesondes at Lauder, New Zealand' 1986-1996, J. Geophys. Res., 103, 28 28,661-28,681. 29 30 Bodeker, G. E., Shiona, H., and Eskes, H., 2005. Indicators of Antarctic ozone depletion, Atmos. Chem. 31 Phys., 5, 2603–2615. 32 33 Brönnimann, S., J. Luterbacher, J. Staehelin, T. M. Svendby, G. Hansen, and T. Svenoe, 2004. Extreme 34 climate of the global troposphere and stratosphere in 1940 – 42 related to El Nino, Nature, 431, 971 – 974. 35 36 Brönnimann, S., Schraner, M., Müller, B., Fischer, A., Brunner, D., Rozanov, E., and Egorova, T.: The 37 1986-1989 ENSO cycle in a chemical climate model, Atmos. Chem. Phys., 6, 4669–4685, 2006. 38 39 Cagnazzo et al., 2009. Northern winter stratospheric temperature and ozone responses to ENSO inferred 40 from an ensemble of Chemistry Climate Models. Atmos. Chem. Phys. Discuss., 9, 12141–12170, 41 42 Calvo, N., M. A. Giorgetta, R. Garcia-Herrera, and E. Manzini (2009), Nonlinearity of the combined warm 43 ENSO and QBO effects on the Northern Hemisphere polar vortex in MAECHAM5 simulations, J. 44 Geophys. Res., 114, D13109, doi:10.1029/ 2008JD011445. 45 46 Camp, C. D., and K.-K. Tung, 2007. Stratospheric polar warming by ENSO in winter: A statistical study, 47 Geophys. Res. Lett., 34, L04809, doi:10.1029/2006GL028521. 48 49 Fioletov, V. E., G. E. Bodeker, A. J. Miller, R. D. McPeters and R. Stolarski, 2002. Global and zonal total 50 ozone variations estimated from ground-based and satellite measurements: 1964-2000,.J. Geophys Res,. 51 107 ,(D22). 52 53 Fischer, A. M., D. T. Shindell, B. Winter, M. S. Bourqui, G. Faluvegi, E. Rozanov, M. Schraner, and S. 54 Brönnimann, 2008. Stratospheric winter climate response to ENSO in three chemistry-climate models, 55

36

Geophys. Res. Lett., 35, L13819, doi:10.1029/2008GL034289. 1 2 Fortuin, J.P.F., and H. Kelder, 1996. Possible links between ozone and temperature profiles, Geophys. Res. 3 Lett., 23, 1517-1520. 4 5 Fortuin, J.P.F., and H. Kelder, 1998. An ozone climatology based on ozonesonde and satellite 6 measurements, Journal of Geophysical Research, 103 (D24), 31709-31734. 7 8 Free, M. and D. Seidel, 2009. The observed ENSO temperature signal in the stratosphere. (accepted J. 9 Geophys. Res.) 10 11 Froidevaux, L., et al., 2008. Validation of Aura Microwave Limb Sounder stratospheric ozone 12 measurements, J. Geophys. Res., 113, D15S20, doi:10.1029/2007JD008771. 13 14 Fusco, A. C., and Salby, M. L., 1999. Interannual variations of total ozone and their relationship to 15 variations of planetary wave activity, J. Climate, 12, 1619 – 1629. 16 17 Garcia-Herrera, R., Calvo, N., Garcia, R. R., and Giorgetta, M. A., 2006. Propagation of ENSO temperature 18 signals into the middle atmosphere: A comparison of two general circulation models and ERA-40 19 reanalysis data, J. Geophys. Res., 111, D06101 doi:10.1029/2005JD006061. 20 21 Garfinkel, C. I., and Hartmann, D. L., 2007. Effects of El Nino -Southern Oscillation and the Quasi- 22 Biennial Oscillation on polar temperatures in the stratosphere, J. Geophys. Res., 112, D19112, 23 doi:10.1029/2007JD008481. 24 25 Giorgetta, M. A., E. Manzini, and E. Roeckner, 2002. Forcing of the quasi-biennial oscillation from a broad 26 spectrum of atmospheric waves, Geophys. Res. Lett., 29, doi:10.1029/2002GL14756. 27 28 Giorgetta, M.A., E. Manzini, E. Roeckner, M. Esch, and L. Bengtsson, 2006. Climatology and Forcing of 29 the Quasi-Biennial Oscillation in the MAECHAM5 Model, J. Climate, 19, 882 3901, 30 doi:10.1175/JCLI3830.1. 31 32 Grant, W. B., E. V. Browell, J. Fishman. V. G. Brackett. R. E. Veiga, D. Nganga. A. Minga, B. Cros. C. F. 33 Butler. M. A. Fenn, C. S. Long, and L. L Stowe, 1994. Aerosol-associated changes in tropical stratospheric 34 ozone following the eruption of Mount Pinatubo, J. Geophys. Res., 99, 8197-8211. 35 36 Gray, L.J., J. Beer, M. Geller, J.D. Haigh, M. Lockwood, K. Matthes, U. Cubasch, D. Fleitmann, G. 37 Harrison, L. Hood, J. Luterbacher, G. A. Meehl, D. Shindell, B. van Geel, and W. White, 2009. Solar 38 Influences on Climate, Reviews of Geophysics, in press. 39 40 Gray, L.J. and M.P. Chipperfield, 1990. On the interannual variability of trace gases in the middle 41 atmosphere, Geophys. Res. Lett., 17, 933-936. 42 43 Gray, L. J., S. J. Phipps, T. J. Dunkerton, M. P. Baldwin, E. F. Drysdale, and M. R. Allen, 2001. A data 44 study of the influence of the equatorial upper stratosphere on northern hemisphere stratospheric sudden 45 warmings, Q. J. R. Meteorol.Soc., 127, 1985-2003. 46 47 Gray, L.J, S.A. Crooks, M.A. Palmer, C.L. Pascow, and S. Sparrow, 2006. A possible transfer mechanism 48 for the 11-year solar cycle to the lower stratosphere, Space Sci. Rev., 125, .357-370. 49 50 Grooß, J.-U., and J. M. Russell III, 2005. Technical note: A stratospheric climatology for O3, H2O, CH4, 51 NOx, HCl and HF derived from HALOE measurements, Atmos. Chem. Phys., 5, 2797– 2807. 52 53 Haigh, J.D., M. Blackburn, and R. Day, 2005. The Response of Tropospheric Circulation to Perturbations 54 in Lower-Stratospheric Temperature, J. Climate, 18, 3672-3685. 55 56

37

Haimberger, L., C. Tavolato, S. Sperka, 2008. Towards elimination of the warm bias in historic radiosonde 1 temperature records — some new results from a comprehensive intercomparison of upper air data. J. 2 Climate, 21, 4587-4606. 3 4 Hamilton, K., 1993. An examination of observed Southern Oscillation effects in the Northern Hemisphere 5 stratosphere. J. Atmos. Sci., 50, 3468–3473. 6 7 Hassler, B., G. E. Bodeker, and M. Dameris, 2008. Technical Note: A new global database of trace gases 8 and aerosols from multiple sources of high vertical resolution measurements, Atmos. Chem. Phys. Discuss., 9 8, 7657-7702. 10 11 Hassler, B., G.E. Bodeker, I. Cionni, and M. Dameris, 2009. A vertically resolved, monthly mean ozone 12 database from 1979 to 2100 for constraining global climate model simulations, International Journal of 13 Remote Sensing, 30 (15-16), 4009–4018. 14 15 Hoerling, M. P., Kumar, A. and Zhong, M., 1997. El Niño, La Niña, and the nonlinearity of their 16 teleconnections. J. Climate, 10, 1769–1786. 17 18 Hofmann, D. J., S.J. Oltmans, J.M. Harris, W.D. Komhyr, J.A. Lathrop, T. DeFoor, and D. Kuniyuki, 1993. 19 Ozonesonde measurements at Hilo, Hawaii following the eruption of Pinatubo, Geophys. Res. Lett., 20, 20 1555-1558. 21 22 Hu, Y. and K.K.Tung, 2002. Interannual and Decadal Variations of Planetary Wave Activity, Stratospheric 23 Cooling, and Northern Hemisphere Annular Mode. J. Climate, 15, 1659-1673. 24 25 Ito, K., Y. Naito, and S. Yoden, 2009. Combined effects of QBO and 11-year solar cycle on the winter 26 hemisphere in a stratosphere-troposphere coupled system, Geophys. Res. Lett., 36, L11804, 27 doi:10.1029/2008GL037117. 28 29 Kodera, K., and Y. Kuroda, 2002. Dynamical response to the solar cycle, J. Geophys. Res., 30 107(D24), 4749, doi:10.1029/2002JD002224. 31 32 Kryjov, V.N. and C.-K. Park (2007), Solar Modulation of the El Nino/Southern Oscillation 33 impact on the Northern Hemisphere Annular Mode. Geophys. Res. Lett., 34, L10701, doi: 34 10.1029/2006GL028015. 35 36 Kuroda 2007. Effect of QBO and ENSO on the Solar Cycle modulation of winter North Atlantic 37 Oscillation. J. Met. Soc. Japan. 85,6,889-898. 38 39 Labitzke, K., 1987. Sunspots, the QBO and the stratospheric temperature in the north polar 40 region, Geophys. Res. Lett., 14, 535-537. 41 42 Labitzke, K., and H. v. Loon, 1988. Associations between the 11-year solar cycle, the QBO and the 43 atmosphere, Part I: The troposphere and stratosphere in the northern hemisphere in winter, J. Atmos. Terr. 44 Phys., 50, 197-206. 45 46 Lee, H. and A. K. Smith, 2003. Simulation of the combined effects of solar cycle, quasi-biennial 47 oscillation, and volcanic forcing on stratospheric ozone changes in recent decades, J. Geophys. Res., 108, 48 4049, doi:10.1029/2001JD001503. 49 50 Manzini et al., 2006. The influence of sea surface temperatures on the Northern winter stratosphere: 51 Ensemble simulations with the MAECHAM5 model, J. Climate, 19, 3863-3881. 52 53 Marsh, D. R., and R. R. Garcia, 2007. Attribution of decadal variability in lower-stratospheric tropical 54 ozone, Geophys. Res. Lett., 34, L21807, doi:10.1029/2007GL030935. 55

38

1 Matthes, K., R.R. Garcia, Y. Kuroda, D.R. Marsh, and K. Kodera, 2009b. The Role of the QBO in 2 Modeling the Influence of the 11-Year Solar Cycle on the Atmosphere Using Time-Variable Forcings, to 3 be submitted to J. Geophys. Res. 4 5 Matthes, K., Y. Kuroda, K. Kodera, and U. Langematz, 2006. Transfer of the Solar Signal from the 6 Stratosphere to the Troposphere: Northern Winter, J. Geophys. Res., 111, D06108, doi:10.1029/ 7 2005JD006283. 8 9 Matthes, K., K. Kodera, L.J. Gray, et al., 2007: Report on the first Solar Influence for SPARC (SOLARIS) 10 workshop in Boulder/CO October 3-6 2006, SPARC Newsletter 28. 11 12 Matthes, K., U. Langematz, L.J. Gray, K. Kodera, and K. Labitzke, 2004. Improved 11-Year Solar Signal 13 in the Freie Universität Berlin Climate Middle Atmosphere Model (FUB-CMAM), J. Geophys. Res., 109, 14 D06101, doi:10.1029/2003JD004012. 15 16 Matthes, K., D.R. Marsh, R.R. Garcia, D. Kinnison, F. Sassi, and S. Walters, 2009a. The Role of the QBO 17 in Modeling the Influence of the 11-Year Solar Cycle on the Atmosphere Using Constant Forcings, 18 submitted to J. Geophys. Res. 19 20 Matthes, K., K. Kodera, J.D. Haigh, D.T. Shindell, K. Shibata, U. Langematz, E. Rozanov, and Y. Kuroda, 21 2003. GRIPS solar experiments intercomparison project: Initial results, Pap. Meteorol. Geophys., 54, 71-22 90. 23 24 Newman, P.A., J.S. Daniel, D.W. Waugh, and E.R. Nash, 2007. A new formulation of equivalent effective 25 stratospheric chlorine (EESC), Atmos. Chem. Phys., 7 4537-4552. 26 27 Punge, H.J., and M. A. Giorgetta, 2008. Net effect of the QBO in a chemistry-climate model, Atmos. Chem. 28 Phys., 8, 6505–6525. 29 30 Randel, W. J., Wu, F., and Stolarski, R., 2002. Changes in column ozone correlated with the stratospheric 31 EP flux, J. Meteorol. Soc. Japan, 80, 849–862. 32 33 Randel, W.J. and F. Wu, 2007. A stratospheric ozone profile data set for 1979-2005: variability, trends, and 34 comparisons with column ozone data, J. Geophys. Res., 112, D06313, doi:10.1029/2006JD007339. 35 36 Randel, W. J., et al., 2009. An update of observed stratospheric temperature trends, J. Geophys. Res., 114, 37 D02107, doi:10.1029/2008JD010421. 38 39 Randel, W. J., R. R. Garcia, N. Calvo, and D. Marsh, 2009. ENSO influence on zonal mean temperature 40 and ozone in the tropical lower stratosphere, Geophys. Res. Lett., in press. 41 42 Randel, W. J, F. Wu, J. M. Russell III, J. W. Waters, and L. Froidevaux, 1995. Ozone and temperature 43 changes in the stratosphere following the eruption of Mount Pinatubo, J. Geophys. Res., 100, 16753-44 16764. 45 46 Russell, J. M., III, et al., 1993. The Halogen Occultation Experiment, J. Geophys. Res., 98(D6), 10,777– 47 10,797. 48 49 Sassi, F., D. Kinnison, B. A. Boville, R. R. Garcia, and R. Roble, 2004. Effect of El Niño–Southern 50 Oscillation on the dynami cal, thermal, and chemical structure of the middle atmophere, J. Geophys . Res., 51 109, D17108, doi:10.1029/2003JD004434. 52 53 Sato, M., J.E. Hansen, M. P. McCormick, and J. B. Pollack, 1993. Stratospheric aerosol optical depths 54 1850-1990, J. Geophys. Res., 98, 22,987-22994. 55 56

39

Scaife, A., N, Butchart, C. Warner, D. Staniforth, W. Norton, and J. Austin, 2000. Realistic Quasi-Biennial 1 Oscillation in a simulation of the global climate, Geophys. Res. Lett., 27, 2 3481-3484. 3 4 Schmidt, G.A., R. Ruedy, J.E. Hansen, I. Aleinov, N. Bell, M. Bauer, S. Bauer, B. Cairns, V. Canuto, Y. 5 Cheng, A. Del Genio, G. Faluvegi, A.D. Friend, T.M. Hall, Y. Hu, M. Kelley, N.Y. Kiang, D. Koch, A.A. 6 Lacis, J. Lerner, K.K. Lo, R.L. Miller, L. Nazarenko, V. Oinas, Ja. Perlwitz, Ju. Perlwitz, D. Rind, A. 7 Romanou, G.L. Russell, Mki. Sato, D.T. 8 9 Shindell, P.H. Stone, S. Sun, N. Tausnev, D. Thresher, and M.-S. Yao, 2006. Present day atmospheric 10 simulations using GISS ModelE: Comparison to in-situ, satellite and reanalysis data. J. Climate, 19, 153-11 192, doi:10.1175/JCLI3612.1. 12 13 Shibata, K., and M. Deushi, 2005. Partitioning between resolved wave forcing and unresolved gravity wave 14 forcing to the quasi-biennial oscillation as revealed with a coupled chemistry-climate model, Geophys. Res. 15 Lett., 32, L12820, doi:10.1029/2005GL022885. 16 17 Solomon, S., R. W. Portmann, R. R. Garcia, W. Randel, F. Wu, R. Nagatani, J. Gleason, L. Thomason, L. 18 R. Poole, and M. P. McCormick, 1998. Ozone depletion at mid-latitudes: coupling of volcanic aerosols and 19 temperature variability to anthropogenic chlorine, Geophys. Res. Lett., 25, 1871-1874. 20 21 Smith, A.K., and K. Matthes, 2008. Decadal-Scale Periodicities in the Stratosphere Associated with the 11-22 Year Solar Cycle and the QBO, J. Geophys. Res., 113, doi:10.1029/2007JD009051. 23 24 Steinbrecht, W., Hassler, B., Brühl, C., Dameris, M., Giorgetta, M. A., Grewe, V., Manzini, E., Matthes, S., 25 Schnadt, C., Steil, B., and Winkler, P., 2006. Interannual variation patterns of total ozone and temperature 26 in observations and model simulations, Atmos. Chem. Phys., 6, 349– 374. 27 28 Stenchikov, G. , A. Robock, V. Ramaswamy, M. D. Schwarzkopf, K. Hamilton, and S. Ramachandran, 29 2002. Arctic Oscillation response to the 1991 Mount Pinatubo eruption: Effects of volcanic aerosols and 30 ozone depletion, J. Geophys. Res. , 107( D24) , 4803, doi: 10. 1029/2002JD002090. 31 32 Stenchikov, G., K. Hamilton, A. Robock, V. Ramaswamy, and M. D. Schwarzkopf, 2004. Arctic 33 Oscillation response to the 1991 Pinatubo Eruption in the SKYHI GCM with a realistic Quasi-Biennial 34 Oscillation, J. Geophys. Res., 109, D03112, doi:10.1029/2003JD003699. 35 36 Stolarski, R.S. and S. Frith, 2006. Search for evidence of trend slow-down in the long-term TOMS/SBUV 37 total ozone data record: the importance of instrument drift uncertainty and fingerprint detection, Atmos. 38 Chem. Phys., 6, 3883-3912. 39 40 Stolarski, R.S., A.R. Douglass, S. Steenrod, and S. Pawson, 2006. Trends in stratospheric ozone: Lessons 41 learned from a 3D Chemical Transport Model, J. Atmos. Sci., 63 (3), 1028-1041. 42 43 Taguchi, M., and D. L. Hartmann, 2006. Increased occurrence of stratospheric sudden warmings during El 44 Nino as simulated by WACCM, J. Climate, 19(3), 324 – 332. 45 46 Taylor, K. E., 2001. Summarizing multiple aspects of model performance in a single diagram, J. Geophys. 47 Res., 106, 7183-7192. 48 49 Textor, C., Graf, H.-F., Herzog, M., and J. M. Oberhuber, 2003. Injection of Gases into the Stratosphere by 50 Explosive Volcanic Eruptions, J. Geophys. Res., 108 (D19), 4606, doi:10.1029/2002JD002987. 51 52 Tie, X.X., and G.P. Brasseur, 1995. The response of stratospheric ozone to volcanic eruptions: Sensitivity 53 to atmospheric chlorine loading, Geophys. Res. Lett., 22 (22), 3035-3038. 54 55 Uppala, S., and Coauthors, 2004. ERA-40: ECMWF 45-year reanalysis of the global atmosphere and 56

40

surface conditions 1957–2002: ECMWF Newsletter, Vol. 101, ECMWF, Reading, United Kingdom, 2–21. 1 2 Van Loon, H., and K. Labitzke, 1987. The Southern Oscillation. Part V: The anomalies in the lower 3 stratosphere of the Northern Hemisphere in winter and a comparison with the quasi-biennial oscillation. 4 Mon. Wea. Rev., 115, 357–369. 5

6 Waters, J.W., et al., 2006. The Earth Observing System Microwave Limb Sounder (EOS MLS) on the Aura 7 Satellite, IEEE Trans. Geosci. Remote Sens., 44, 1075-1092. 8 9 Weber, M., S. Dhomse, F. Wittrock, A Richter, B.-M. Sinnhuber and J. P. Burrows, 2003. Dynamical 10 control of NH and SH winter/spring total ozone from GOME observations in 1995-2002, Geophys. Res. 11 Lett., 30, 1583, doi:10.1029/2002GL016799. 12 13 World Meteorological Organisation, 2007. Scientific Assessement of Ozone Depletion: 2006, Global 14 Ozone Res. and Monit. Proj. Rep. 50, World Meteorolog. Organ. Ozone Secr., Geneva, Switzerland. 15 16 17

18

41

1 Table 8.2.1: Global total ozone model bias in %. Data from Figure 8.2.8 2 3 Model

TOZ Bias (%)

AMTRAC -1.3

CAM3.5 -1.2

CCRNIES 7.6

CMAM -0.7

CNRM-ACM -1.3

E39C 17.0

EMAC 3.8

GEOSCCM 7.5

LMDZrepro 6.3

MRI 12

Niwa_SOCOL 1.1

SOCOL -0.8

ULAQ 5.0

UMSLIMCAT -7.8

UMUKCA-METO 9.3

UMUKCA-UCAMN/A

WACCM -0.2

4 5 6 7 8 9 10 11 12 13 14 15 16

42

1 Table 8.5.1: Tropical variabillity in the CCMVal-2 models. Models in Group A and 2 Group B do not assimilate the QBO. Models in Group C assimilate the QBO (via nudging 3 of the zonal winds or vorticity). Group A models have basis functions in the MLR 4 analysis set to zero. Models in Group B and C are included in the MLR analysis. 5 6 7 GROUP A

GROUP B

GROUP C

CMAM AMTRAC3 CAM3.5

CNRM-ACM

MRI

CCSRNIES

GEOSCCM

UMUKACA-METO

E39CA

LMDZrepro

UMUKCA-UCAM

EMAC

UMSLIMCAT

Niwa_SOCOL

SOCOL

ULAQ

WACCM

8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

43

1 Table 8.8.1: CCMVal-2 Model list for Figure 8.8.1 2 3 #

model name

1.

AMTRAC

2.

CAM3.5

3.

CCRNIES

4.

CMAM

5.

CNRM-ACM

6.

E39C

7.

EMAC

8.

GEOSCCM

9.

LMDZrepro

10.

MRI

11.

Niwa_SOCOL

12.

SOCOL

13.

ULAQ

14.

UMSLIMCAT

15.

UMUKCA-METO

16.

UMUKCA-UCAM

17.

WACCM

18.

MULTI_MODEL

4 5 6

Figure 8.1: Ozone variations for 60°S-60°N estimated from a combined ground-based and satellite data set (Dobson, TOMS+SBUV) and individual components that comprise ozone variations. Top: Original data (red) and fitted with a multiple linear regression (MLR) model (black) from 1964 to p g ( ) ( )2008, the residual is the difference between the original and the fitted time series. See text for details on the MLR.Bottom: Deseasonalized, area-weighted total ozone deviations estimated from ground-based+satellite data adjusted for solar, volcanic, QBO and residual effects, for 60°S-60°N. The red line represents the EESCcurve scaled to fit the data from 1964-2008.

(a)

(b)

Figure 8.2.1: (a) Monthly mean ozone mixing ratios (ppmv) at 1hPa throughout the year, 40oS (left), Equator (middle) and 40oN (right) from several years of MLS observations (black lines) and for the CCMVal-2 CCMs (one’s year zonal mean ozoneobservations (black lines) and for the CCMVal-2 CCMs (one s year zonal mean ozone in the early 2000s). MLS data are averaged for a six degree latitude band centered on the selected latitudes. (b) Same as 8.2.1a but at 46 hPa, 72oS (left), Equator (middle) and 72oN (right)Note that CAM3.5 and E39CA are not shown.

Figure 8.2.2: Climatological zonal mean O3 mixing ratios from the CCMVal-2 CCMs and HALOE in ppmv. Vertical profiles at (a) 80°N in March, (b) 0° in March, and (c) 80°S in October. Latitudinal profiles at 50 hPa in (d) March and (e) October. The grey area shows HALOE plus and minus 1 standard deviation (s) about the climatological zonal mean.Same as Figure 13 for CCMVal-1 CCMs in Eyring et al. (2006).

Figure 8.2.3: Normalized Taylor diagram of the annual mean climatological ozone t f th MLR l i P tt t ti ti b t th CCMV l 2 d lcomponent from the MLR analysis. Pattern statistics between the CCMVal-2 models

(1960-2004) and the NIWA data (1979-2007). The pattern statistics have been computed for the 1-500 hPa range, 90°S-90°N, using equal weights (weight=1) in log-p.

Figure 8.2.4: Normalized Taylor diagram of the amplitude of the annual cycle (at left) d th lit d f th i l l ( t i ht) t f th MLRand the amplitude of the semiannual cycle (at right) ozone components from the MLR

analysis. Pattern statistics between the CCMVal-2 models (1960-2004) and the NIWA data (1979-2004). The pattern statistics have been computed for the 1-500 hPa range, 90°S-90°N, using equal weights (weight=1) in log-p. The corresponding fields are shown in the supplementary material Figures 8.2.S2 and 8.2.S3, respectively.

Figure 8.2.5: Normalized Taylor diagram of the annual mean in zonal mean total P tt t ti ti b t th CCMV l 2 d l (1979 2004) d th NIWAozone. Pattern statistics between the CCMVal-2 models (1979-2004) and the NIWA

data (1979-2004). The corresponding fields are shown in the supplementary material (Figure 8.2.S4).

Figure 8.3.1a: Interannual variability of polar cap averaged (60-90°) total ozone (DU) by month of the year calculated for the (left) NH and (right) SH over the period onward of (top) 1980 and (bottom) 1960. The mean of all available ensemble members for the same model is used. NIWA/ERA40 include all years except the SH vortex split event, the NIWA/NNR includes everything.everything.

Figure 8.3.1b: Mean polar cap averaged total ozone (DU) by month of the year calculated for the (left) NH and (right) SH over the period onward of (top) 1980 and (bottom) 1960. The mean of all available ensemble members for the same model is usedmembers for the same model is used.

Fi 8 3 2 Li i t f tt l t f S i /A tFigure 8.3.2: Linear regression parameter of scatter plots of Spring/Autumn ozone ratio and winter mean heatflux for CCMVal REF-B1 integrations plotted against mean Spring/Autumn ozone ratio for each model. Left panel shows results for NH, right panel shows results for SH. Black symbols and dotted lines show results from ERA-interim re-analysis and NIWA ozone dataset. Each model is plotted with a single coloured dot or square, 95% confidence intervals for parameters are shown in solid lines.

Figure 8.3.3: Parameters of linear fit to scatter plot of polar cap averaged total ozone (DU) vs. polar cap averaged 50 hPa temperatures. Shown is the ozone value of the linear fit at T = 200 K (x-axis) plotted against the slope of the regression line (y-axis). The thin lines indicate the confidence limits derived from the width of the 95% distributions using boot strapping. This concept is similar to the approach by Newman et al. (2001). All available model data onward of 1960 are included in the calculationsmodel data onward of 1960 are included in the calculations.

Figure 8 3 4: Regression of total ozone on a simple definition of the (a)Figure 8.3.4: Regression of total ozone on a simple definition of the (a) NAM during March (Feb-Mar) and (b) SAM during November (Oct-Nov). Contour interval is 0.04 DU/gpm. The numbers on top of each map represent (left) pattern correlations (x100) and (right) nrms-errors (x100) between model results and the observations (NIWA/NNR). Numbers in parenthesis after model names indicate period of years included in the calculations.

Figure 8.4.1: Total column ozone solar regression coefficients for CCMVal-2 CCMs (x) relative to the observed solar regression coefficient (x0) from the TOMS+gb data in Figure 8.1.

a) Temperature b) Ozone

Figure 8.4.2: Annual mean tropical (25°S-25°N) solar regression coefficients for (a) t t i K l i 100 it f th F10 7 di fl d (b) i %/100temperature in Kelvin per 100 units of the F10.7cm radio flux, and (b) ozone in %/100 F10.7cm units from CCMVal-2 CCMs (1960-2004) and observations (NIWA ozone (1979-2004), Randel and Wu ozone (1979-2005), RICH radiosonde data (1960-2004), SSU data (1979-2005), and ERA40 (1979-2001)) from 100 to 0.1 hPa. Top: CCMs without imposed solar cycle (non-sc group), bottom: CCMs with imposed solar cycle (sc group).

Figure 8.4.3: Normalized Taylor diagram of the annual mean tropical (25°S-25°N) solar ozone component from the MLR analysis Pattern statistics between thesolar ozone component from the MLR analysis. Pattern statistics between the CCMVal-2 models (1960-2004) and the NIWA data (1979-2007). The pattern statistics have been computed for the 1-100 hPa range, 25°S-25°N, using equal weights (weight=1) in log-p. The corresponding fields are shown in Figure 8.4.2 (right).

a) Ozone

b) Temperature

Figure 8.4.4: Amplitude of the solar cycle in the upper stratosphere over latitude for 3 hP i % ( ) d 1 hP i K (b )ozone at 3 hPa in % (top) and temperature at 1 hPa in K (bottom).

Figure 8.5.1: Monthly zonal mean standard deviation of zonal mean wind (left, m/s) and ozone (right, DU/km) averaged from 5S to 5N. Results from the CCMVal-2 simulations (in colour), ERA40 (left, black), and SAGE data (right, black). From detrended and deseasonalized time series. Top panels: group C CCMs (Tab. 8.3.1), bottom panels: group B CCMs.

(b)(a)

Figure 8.5.2: Annual mean equatorial (5S-5N) regression coefficients for ozone from CCMVal-2 CCMs (1960-2004) and observations (NIWA ozone, 1979-2004) for the combination of the two QBO coefficients in %. (a) Group B CCMs from Tab. 8.3.1, (b) Group C CCMs. The combined QBO coefficient has been ( ) pmultiplied by 30, see text for details.

(a)

(b)(b)

Figure 8.5.3: Latitude distribution of the total ozone QBO regression coefficients for CCMVal-2 CCMs (1960-2004) and observations (TOMS+gb1964-2004) for the combination of the two QBO coefficients in DU (a) Group B1964 2004) for the combination of the two QBO coefficients in DU. (a) Group B CCMs, (b) Group C CCMs.

(a)

(b)

Figure 8.5.4: Combination of the QBO regression coefficients for the monthly zonal mean total ozone averaged from 5°S to 5°N. Results from the CCMVal-2 CCMs (in colour; 1960-2004) and the TOMS+gb data set (black; 1964-2004). (a) Group C CCMs, and (b) Group B CCMs.

a) Temperature b) Ozone

Figure 8.6.1: Annual mean tropical (25S-25N) regression coefficients for (a) temperature in Kelvin and (b) ozone in percent from CCMVal-2 CCMs (1960-2004) and observations (NIWA ozone (1979-2004); Randel and Wu ozone (1979-2005)) for the ENSO coefficient in K from 1000 to 1 hPa. The ENSO coefficient has been multiplied by 2.5 see text for details.

a) Temperature b) Ozone

Figure 8.6.2: Annual mean polarcap (70N-90N) regression coefficients for (a) temperature in Kelvin and (b) ozone in percent from CCMVal-2 CCMs (1960-2004) and observations (NIWA ozone (1979-2004)) for the ENSO coefficient in K from 1000 to 1 hPa. The ENSO coefficient has been multiplied by 2.5 see text for details.

Figure 8.6.3: Scatter plot of the February-March polar cap average of the total ozone ENSO anomaly (DU) versus the temperature ENSO anomaly (30-70 hPa average, K). Black star: NIWA versus ERA40 signature. Color: CCMVal-2 simulations. Polar cap average computed over 70N-90N.

Figure 8.7.1: Annual mean global mean 50 hPa temperature anomalies from pre-volcanicconditions for the Agung, El Chichón and Pinatubo eruptions. Top: model results color-coded by model, bottom: results color-coded by type of volcanic heating parameterization used, including: optical properties derived from SADs (red), optical properties from SAGE/GISS data set (green) optical properties from Ammann et al (2003) (cyan)SAGE/GISS data set (green), optical properties from Ammann et al. (2003) (cyan),prescribed heating rate anomalies (yellow),none (blue), full aerosol microphysics (purple). ERA-40 50 hPa temperature anomalies are shown in both plots in black.

a) Temperature b) Ozone

Agu

ngE

l Chi

con

Pin

atub

o

Figure 8.7.2: Annual mean tropical (25S-25N) regression coefficients from CCMVal-2 CCMs (1960-2004) and observations for the volcanic coefficients from 1000 to 1 hPa. (a) temperature in K; (ERA40 and RICH), (b) ozone in %, no observations are shown due to large uncertainties.

Figure 8.7.3: Annual mean global mean total column ozone anomalies from pre-volcanic conditions for the Agung, El Chichón and Pinatubo eruptions. Ozone anomalies from the TOMS+gb data set are shown in black.

Figure 8.7.4: Post volcanic eruption annual mean global mean anomalies of total column ozone as a function of similarly calculated anomalies in ClO at 50 hPa, for the models that have reported ClO mixing ratios.

Figure 8.8.1: Matrix displaying the skill (see color bar), following eq 4 of Taylor 2001, foreach of the Taylor diagrams presented in the Chapter. see text for details.

Figure 8.8.2: Total ozone regression coefficients for CCMVal-2 CCMs (x) relative to the observed regression coefficient (x0) from the TOMS-SBUB-gb data in Figure 8.1. See text for details.

SupplementaryFigures

Figure 8.2.S1: Temperature signal corresponding to the seasonal ozone cycle shown inFigure 8.2.S1: Temperature signal corresponding to the seasonal ozone cycle shown in Figures 8.2.1 a,b.

Figure 8.2.S2: Annual cycle components latitude-height sections from MLR for the CCMVal-2 CCMs (1960-2004) and NIWA data (1979-2007).

Figure 8.2.S2: Semiannual cycle components latitude-height sections from MLR for the CCMVal-2 CCMs (1960-2004) and NIWA data (1979-2007).

Figure 8.2.S4: Annual cycle of the zonal mean total ozone for the NIWA dataset and the CCMVal-2 models.

Figure 8.3.S1: Scatter plot of Sep/Mar ozone ratio and winter mean heat flux for the southern hemisphere. See text for description of analysis procedure. Dots represent a single year of data from reanalysis or model. Solid lines represent a least-squares fit to the reanalysis or model in question.

Fi 8 3 S2 S tt l t f M /S ti d i t h tfl f thFigure 8.3.S2:Scatter plot of Mar/Sep ozone ratio and winter mean heatflux for the northern hemisphere. See text for description of analysis procedure. Dots represent a single year of data from reanalysis or model. Solid lines represent a least-squares fit to the reanalysis or model in question.

Figure 8.4.S1: Uncertainties of the solar regression coefficient for Figure 8.4.2 for temperature in K (left) and ozone in % (right). Top: non-sc group CCMs, bottom: sc gro p CCMsgroup CCMs.

Figure 8.4.S2: Solar regression coefficient for the shortwave heating rates in K/d per 100 units of the F10.7cm radio flux (left) and the uncertainties (right) for those CCMs that pro ided dataCCMs that provided data.

Figure 8.4.S3: Annual mean latitude distribution of the solar coefficient (%/100 units ofFigure 8.4.S3: Annual mean latitude distribution of the solar coefficient (%/100 units of the F10.7cm radio flux) for total ozone for the CCMVal-2 CCMs and TOMS+gb observations. Top: non-sc group CCMs bottom: sc group CCMs.

Fi 8 5 S1 L tit d ti i f th t t l i l f 1993 t 1996 f hFigure 8.5.S1: Latitude-time series of the total ozone signal from 1993 to 1996 for each of the CCMs .

Figure 8.5.S2: Latitude-time series of the total ozone signal from 1993 to 1996 for each of the CCMs .

Figure 8.5.S3: Uncertainties of the ozone QBO regression coefficient in percent for Figure 8.3.2 for CCMs in group B (left) and CCMs in group C (right).

Figure 8.8.S1: Residual (difference between original and fitted data) of total ozone forTOMS+gb observations similar to Figure 8.1 compared to all CCMs.