The evolution of stratospheric ozone from 1960 to 2100 is examined in simulations from 14 chemistry-climate models, driven by prescribed levels of halogens and greenhouse gases. There is general agreement among the models that total column ozone reached a minimum around year 2000 at all latitudes, projected to be followed by an increase over the first half of the 21st century. In the second half of the 21st century, ozone is projected to continue increasing, level off, or even decrease depending on the latitude. Separation into partial columns above and below 20 hPa reveals that these latitudinal differences are almost completely caused by differences in the model projections of ozone in the lower stratosphere. At all latitudes, upper stratospheric ozone increases throughout the 21st century and is projected to return to 1960 levels well before the end of the century, although there is a spread among models in the dates that ozone returns to specific historical values. We find decreasing halogens and declining upper atmospheric temperatures, driven by increasing greenhouse gases, contribute almost equally to increases in upper stratospheric ozone. In the tropical lower stratosphere, an increase in upwelling causes a steady decrease in ozone through the 21st century, and total column ozone does not return to 1960 levels in most of the models. In contrast, lower stratospheric and total column ozone in middle and high latitudes increases during the 21st century, returning to 1960 levels well before the end of the century in most models.
 Projecting the evolution of ozone in the 21st century is a critical issue. While changes in ozone are presently controlled primarily by declines in halogen concentrations, variations in temperature, circulation, and oxides of nitrogen and hydrogen also affect ozone [World Meteorological Organization (WMO), 2003, 2007]. Throughout the stratosphere, there will be long-term changes in various processes as well as the relative importance of these processes on ozone. It is difficult to find a single approach to identify the contributions of different mechanisms affecting ozone levels throughout the stratosphere.
Eyring et al.  laid the groundwork for the Chemistry-Climate Model Validation (CCMVal) activity in which numerous chemistry-climate models (CCMs) were evaluated to increase our confidence in projecting future stratospheric ozone change. The results of the CCMVal-1 simulations of past ozone changes conducted for this activity were presented by Eyring et al. , who evaluated processes important in determining the distribution of ozone. In a follow-up study, Eyring et al.  presented the projections of stratospheric ozone change over the 21st century simulated by these CCMs and discussed how quantities that can impact ozone are projected to change.
 CCMVal-1 was the first major coordinated activity in which an ensemble of CCMs performed simulations with similar external forcings to assess ozone evolution [Eyring et al., 2007]. The CCMVal-2 activity [Eyring et al., 2008] included more models (14 are used here), and, in addition, more simulations cover the entire period of interest (1960–2100) than in CCMVal-1 (three models). Here we make use of CCMs that participated in the CCMVal-2 activity and that contributed projections of stratospheric ozone evolution until the end of the 21st century. These simulations are based on observed GHG and halogen concentrations in the past and on one projected scenario for the future.
Austin et al. [2010a] examined the evolution of total column ozone and compared the dates ozone and Cly return to a particular historical level in the CCMVal-2 models. Here we (1) contrast the evolution of ozone in the upper and lower stratosphere and (2) examine the causes of the ozone changes (and differences among the models). Section 2 gives an overview of the models and model simulations. The results are discussed for extrapolar ozone in section 3 and polar ozone in section 4. Variations in ozone recovery by region are discussed in section 5, and a summary of the conclusions is given in section 6.
2. Models and Method
2.1. Model Simulations
 CCMVal-2 model simulations were conducted to improve the understanding of models through process-oriented evaluation along with discussion and coordinated analysis [SPARC CCMVal, 2010]. Information about individual model simulations and references for each are summarized in Table 1. In addition, Morgenstern et al.  presented a much more detailed overview of the models that participated in this activity. Here we consider simulations from the CCMs that cover the years 1960 to 2099 (except Unified Model/United Kingdom Chemistry Aerosol Community Model–Met Office (UMUKCA-METO), which ends in 2083), with domains that include the upper stratosphere. Only these can be used to contrast the ozone evolution in the lower and upper stratosphere.
HadGEM, Hadley Centre Global Environment Model; CAM, Community Atmosphere Model; MJ98, MRI/Japan Meteorological Agency 98; CCSM, Community Climate System Model; MIROC, Model for Interdisciplinary Research on Climate; MAECHAM4, Middle Atmosphere European Center/Hamburg version 4.
 Reference simulation REF-B2 uses the A1B GHG scenario from the Intergovernmental Panel on Climate Change (IPCC)  and the revised A1 halogen scenario from WMO [2007; SPARC CCMVal, 2010]. The revised A1 halogen scenario includes the earlier phase out of hydrochlorofluorocarbons mandated by recent adjustments to the Montreal Protocol. For the Goddard Earth Observing System–Chemistry-Climate Model (GEOSCCM), a combination of REF-B1 (1960–2000) and REF-B2 (2001–2099) was used. REF-B1 differs only in that observed Hadley sea surface temperatures (SSTs) and sea ice data [Rayner et al., 2003] were used instead of modeled SSTs. All model runs with the exception of the Canadian Middle Atmosphere Model (CMAM) use prescribed sea surface temperatures and sea ice extent. These SSTs and sea ice extent are based on fully coupled atmosphere-ocean runs for the A1B scenario, except for GEOSCCM, which uses observed SSTs and sea ice for 1960–2000 (see Table 1). CMAM includes a fully coupled ocean model in its simulations. All models use comprehensive stratospheric chemistry schemes except Atmospheric Model with Transport and Chemistry 3 (AMTRAC3), in which halogen-containing source gases are not explicitly modeled but, rather, Cly and Bry are parameterized [Austin and Wilson, 2010; Morgenstern et al., 2010]. REF-B2 simulations use background nonvolcanic aerosol loading, and there are no imposed solar cycle variations. Although several modeling groups submitted an ensemble of simulations, only the first member from each group was used in this analysis because as shown from CCMVal-1 simulations, the intermodel differences from these simulations are generally much larger than the ensemble spreads [Eyring et al., 2007]. The annually averaged zonal mean is computed from the monthly zonal mean model output and used in this study. The partial and total column ozone amounts are integrated over the standard CCMVal-2 pressure levels.
 To view all 14 models on a single plot, filtering is necessary to remove short-term (e.g., high-frequency) variations. In all cases (except where noted), a 1:2:1 filter is used iteratively 30 times as described by Eyring et al.  to smooth the model output displayed in Figures 1, 2, 3, 7, 8, 9, and 10. This filter is only applied for graphics; all analyses use unfiltered data. The filter has a half-amplitude response of 21 years applied to annually averaged model output. While some differences can be seen in individual models between this filter and the time series additive-model analysis [Scinocca et al., 2010], they are typically very small and do not impact our conclusions. The time series additive-model is a nonparametric additive model in which the trend estimate produces residuals that are independent, normally distributed random variables [SPARC CCMVal, 2010].
2.2. Multiple Linear Regression Analysis
 One of the primary methods we use to estimate the contribution of different mechanisms to the simulated changes in middle and upper stratospheric ozone is multiple linear regression (MLR). The method is explained in detail by Oman et al. , but we repeat some of the basic details here. For a given location, MLR is applied to determine the coefficients mX such that
where ΔO3(t) are the annual mean ozone values, Xj are the different annual means of the quantities that could influence ozone, the coefficients mX are the sensitivity of ozone to the quantities X(i.e., mX = ∂O3/∂X), and ɛ is the residual in the fit. To do this, four explanatory variables (Xj) are used in (1): (1) Cly+αBry, (2) reactive nitrogen (NOy = NO + NO2 + NO3 + 2*(N2O5) + HNO3 + HO2NO2 + ClONO2 + BrONO2), (3) reactive hydrogen (HOx = OH+HO2), and (4) temperature. Examples of the predictor time series can be seen in Figure 2 and have been smoothed for plotting purposes. Each term on the right side of equation (1) gives the “contribution” of the response in ozone caused by a change in X. We use α = 5 in the definition as Daniel et al.  showed this is an appropriate value for the upper stratosphere, which is the chemically driven region for which the MLR analysis is most appropriate. A test of this method using α = 60 did not noticeably impact any of the individual contributions calculated in this study. We use α = 60 in section 4, when examining polar lower stratospheric ozone changes, since a higher value of α is needed to represent the increased importance of bromine chemistry to polar ozone depletion. This value of 60 is based on the WMO  recommendation for global ozone destruction and is similar to the value of 65 for polar ozone destruction given in WMO .
 There are several limitations with the above linear regression approach, as discussed by Oman et al. , that are summarized below. First, other mechanisms that are not considered in the regression (e.g., transport) could play a role. The role of transport on ozone is especially important in the lower stratosphere, and details of this complication are discussed in section 3.4. Second, significant correlations can exist between the temporal variations of the quantities, i.e., the quantities are not necessarily independent. Third, a high correlation between ozone and a quantity does not show causality, as ozone could be causing the quantity to change, or changes in another quantity could be causing both ozone and the quantity of interest to change in a correlated way. Temperature and ozone in the upper stratosphere are an example of this third complication, e.g., changes in ozone cause variations in temperature through changes in short-wave heating [Shine et al., 2003]. At the same time, the local ozone concentration responds to variations in temperature, because the chemical reaction rates are temperature dependent. Also, the relationship between the regression variables and ozone may not be linear. For example, the ClO/Cly ratio varies as a function of temperature in a highly nonlinear manner. One could use a regression based on the rates of odd-oxygen loss by nitrogen, chlorine, and bromine radical abundances as regressor variables, as was done by Yang et al. . However, this treatment is beyond the scope of the present analysis, as the information archived by each model makes the use of radical abundances challenging to implement. Because of the above limitations, caution must be applied when interpreting the MLR results presented below.
 UMUKCA-UCAM and AMTRAC3 did not provide HOx, so for these models the MLR consists of three explanatory variables. This analysis has been repeated without HOx for all models, and while there is some decrease in explained variance from the MLR, the main findings and conclusions are not impacted.
3. Extrapolar Ozone
3.1. Total Column Ozone
Austin et al. [2010a] examined the changes in total column ozone in the CCMVal-2 simulations. Here we start with total column ozone and then break down the ozone amounts into two partial columns for the extrapolar region. Figures 1a–1c show the evolution of total column ozone amounts over the tropics (25°S–25°N) and midlatitudes of each hemisphere (35°–60°S and 35°–60°N), for each of the CCMVal-2 models. The evolution is shown with respect to 1960 levels and has been smoothed as described above. Also shown is the evolution of ground-based total column ozone measurements (black squares; updated from Fioletov et al. ) with respect to the smoothed 1964 value, the first year of the data set. The difference between 1960 and 1964 ozone values is not expected to be significant. Total and partial column ozone measurements are from the Total and Profile Merged Ozone Data Set (MOD) [Stolarski and Frith, 2006; Ziemke et al., 2005] (http://acdb-ext.gsfc.nasa.gov/Data_services/merged/index.html). The Total MOD includes observations from the Nimbus 7 and Earth Probe Total Ozone Mapping Spectrometer instruments, from the solar backscatter ultraviolet (SBUV and SBUV2) series of instruments, and the Aura Ozone Monitoring Instrument. The Profile MOD, hereafter referred to as SBUV MOD, includes profile data from the SBUV instruments. The upper partial column is constructed from integrated SBUV MOD data from 20 to 1 hPa. The lower partial column is computed by subtracting the upper partial column from the Total MOD observations. Total Ozone Mapping Spectrometer, SBUV, and Ozone Monitoring Instrument are all nadir-viewing instruments and only make measurements over sunlit portions of the globe. No measurements are available during polar night. One issue is that using 1979 as a reference year may not be the best comparison to models referenced to 1960. However, for the extrapolar regions plotted in Figure 1, we note that total column ozone amounts between this and the ground-based data set agree to within 2 Dobson units (DU) in 1979; this was also found by Fioletov et al. . We reference the models to 1960 because important differences have emerged by 1980, because significant increases in halogens have occurred by 1980.
 There is qualitative agreement in the evolution of ozone among regions and models at the broadest level, with ozone decreasing from 1960 to around 2000, and then increasing in the first half of the 21st century [Austin et al., 2010a]. All models consistently indicate tropical total column ozone at the end of the 21st century is less than it was in 1960 (Figure 1b). In contrast, total column ozone over the midlatitudes (Figures 1a and 1c) in 2100 exceeds 1960 values in nearly all the models, with a generally larger increase in Northern Hemisphere (NH) midlatitudes.
 There is a large spread in the magnitude of the peak ozone loss (around 2000) and the date that column ozone is projected to return to 1960 levels. For example, in Southern Hemisphere (SH) midlatitudes, the simulated decrease in ozone from 1960 to 2000 varies from around 10 DU to over 50 DU, while the date of return to 1960 levels for the same region varies from around 2030 to after 2100. In all three latitude bands, Centre National Recherche Meteorologique–ARPEGE-Climat coupled MOCAGE (CNRM-ACM) and Meteorological Research Institute (MRI) show the largest loss in total column ozone by the year 2000 and also tend to have later return dates. UMUKCA-METO has the smallest ozone loss in the tropics and SH and one of the smallest decreases in NH midlatitudes. The reasons for these model differences are examined in section 3.2. Also, there is a large difference in the ozone decline in the SH midlatitudes prior to 1975 found by the various models (declines ranging from 0 to 20 DU), with no change over this time period apparent in the ground-based ozone observations. Ground-based total column ozone observations indicate about a 20 DU decrease by 2000 in the SH midlatitudes and about a 15 DU decrease in the NH midlatitudes with similar decreases in the SBUV MOD observations. In the tropics, the small observed column ozone changes are similar to those produced by UMUKCA-METO but are less than the ozone changes simulated by all other models.
3.2. Partial Column Ozone
 It is useful to look at changes in the partial columns [e.g., Yang et al., 2006; Li et al., 2009] to understand the relative role of changes in the upper and lower portions of the stratosphere in the differences described above. As described by Oman et al. , we split the stratosphere at 20 hPa because the ozone changes at 20 hPa are generally small and 20 hPa separates the photochemically controlled ozone region above from the transport and chemically driven region below. Other studies have used 15 hPa [e.g., Li et al., 2009] or 25 km altitude (∼25 hPa) [e.g., Yang et al., 2006], but we prefer 20 hPa, as used by Oman et al. . The results presented are not sensitive to the precise specification of the pressure chosen to distinguish the upper and lower partial columns within the range of these previous studies. CAM3.5 has a model top at 3.5 hPa, which could impact the model's upper stratospheric ozone projection.
Figures 1d–1f and 1g–1i show the evolution of the partial column ozone amounts for the upper (20–0.1 hPa) and lower (500–20 hPa) portions of the column. The evolution in the upper partial column ozone is similar among the three regions, but there are significant differences in the lower partial column. It is the differences in the lower stratosphere that cause differences in the evolution of column ozone among regions. In particular, the decrease in tropical column ozone over the latter half of the 21st century is due to decreases in the lower stratosphere (Figure 1h), which are larger than increases in upper stratospheric ozone (Figure 1e).
 Models with larger (smaller) total column ozone changes around year 2000 generally have larger (smaller) changes in both lower and upper partial columns. There are quantitative differences in the upper columns, but these are generally smaller than the differences in the lower columns, especially in midlatitudes (note the different scales for upper and lower column plots in Figure 1). Peak losses occur around the year 2000, with a range of 3–7 DU over the tropics and midlatitudes for the upper partial column of most models, although MRI and CNRM-ACM have larger losses of around 10 DU. Observations over these three latitude bands indicate a decrease of 5–7 DU from 1979. In the lower partial column, there are much greater differences in the ozone change over this range of latitudes. Midlatitude ozone losses around 2000 are largest in the SH, as expected because of the effects of the Antarctic ozone hole [Atkinson et al., 1989]. Ozone losses range from 10 to 25 DU over most models with less ozone loss in UMUKCA-METO and significantly more in MRI and CNRM-ACM. Tropical latitudes show the largest loss of ozone by the end of the 21st century, with a range of 7–18 DU, with most models showing between 2 to 5 DU loss by 2010 from the 1960 reference. SBUV MOD observations do not show a trend in the partial lower stratosphere column ozone over the 1979–2009 time period. Randel and Wu  reported negative trends in tropical lower stratospheric ozone using Stratospheric Aerosol and Gas Experiment (SAGE) I and SAGE II data over a similar time period, suggesting further study is needed to reconcile these differences [WMO, 2007]. The tropical lower partial column is the only region where ozone is projected to continue to decline until the end of the 21st century, because of increased strength of the Brewer-Dobson circulation in the models that is driven by rising GHG concentrations [Rind et al., 1998; Butchart and Scaife, 2001; Butchart et al., 2006; Shepherd, 2008; Li et al., 2009; Garcia and Randel ; Akiyoshi et al., 2010]. An increase in the circulation of the stratosphere reduces the transit time of an air parcel through the tropical lower stratosphere, resulting in lower values of ozone because of less time for accumulation of the ozone that is produced in this region [Avallone and Prather, 1996].
3.3. Upper Stratosphere
 As discussed above, and shown in Figures 1d–1f, the evolution of ozone in the upper portion of the stratosphere is qualitatively similar among most models. Also, the overall behavior of ozone in the tropics and midlatitudes (both hemispheres) is similar. Nonetheless, there are quantitative differences among the models. We use the MLR analysis described in section 2 to examine which explanatory variables are causing the ozone changes and the differences among the models.
 We consider first the evolution of ozone at 5 hPa in the tropics (25°S–25°N). Figure 2a shows the evolution of the ozone mixing ratio, while Figure 2b shows the change in ozone with respect to 1960 levels. There is a large spread in the time-mean values of ozone among the models (from around 8 ppmv to around 11 ppmv) and also some differences in changes relative to 1960 levels. All models show a peak ozone loss around 2000, with magnitudes of 0.5 and 0.7 ppm for most models. This peak ozone loss coincides with the peak chlorine loading, which is specified by the boundary conditions. The MRI and CNRM-ACM models show the largest peak ozone loss (greater than 1.0 ppm), whereas AMTRAC3 and Laboratoire de Météorologie Dynamique Zoom–REPROBUS (LMDZrepro) have the smallest loss (less than 0.4 ppm). All models also show 5 hPa ozone returning to 1960 levels before the end of the 21st century, but there is a large spread in dates when this occurs (from early 2020s to 2060s).
 As presented in the Introduction, several mechanisms can cause changes in ozone, including changes in halogens, temperature, reactive nitrogen and hydrogen, and transport. Figures 2c, 2e, 2g, and 2i show the evolution of Cly + αBry, temperature, NOy, and HOx, respectively. The evolution of Cly + αBry and variation among models (Figure 2c) is somewhat similar to that of ozone (Figure 2a), suggesting that differences in Cly + αBry can explain much of the ozone evolution. However, there is not a simple one-to-one relationship between changes in ozone and those in Cly + αBry. For instance, ozone exceeds 1960s values in the later part of the 21st century, while Cly + αBry is still above 1960 levels. Also, models with larger values of Cly + αBry (Figure 2c) do not necessarily have larger ozone loss, e.g., MRI and CNRM-ACM have the largest peaks in ozone loss but do not have the largest changes in Cly + αBry.
 One possible cause for the lack of a simple relationship between ozone and halogen changes is differences in temperature trends. As shown in Figure 2e, all models show cooling throughout the 21st century, largely from increasing CO2 concentrations [Shindell et al., 1998], and this is expected to increase ozone (by slowing down the gas phase rate of reactions that destroy ozone). However, it should be noted that over the recent past about 50% of the cooling is associated with the ozone loss from Cly + αBry [Shepherd and Jonsson, 2008]. The simulated future cooling, mostly from increasing CO2 concentrations, explains the increase in ozone above 1960 levels before Cly + αBry returns to 1960 values. Most models cool by about 8–9 K below 1960 levels by the end of century, but a few models have cooling exceeding 10 K (Center for Climate System Research/National Institute for Environmental Studies (CCSRNIES), Unified Model–SLIMCAT (UMSLIMCAT), LMDZrepro, and CMAM). These four models have among the largest ozone increases by 2100, which is consistent with an expected increase in ozone attributed to cooling, as shown in Figure 2b.
 To quantify the relative contributions of the above mechanisms to ozone changes, we now apply an MLR analysis to model outputs. As described in section 2, the MLR analysis is used to quantify the contributions of changes in Cly + αBry, temperature, NOy, and HOx to the ozone changes. An example of the correlation coefficient squared, R2, between the modeled ozone time series and the MLR is shown in the study by Oman et al. [2010; their Figure 4]. Typically, the regression explains more than 90% of the variance (R2 > 0.9) throughout much of the upper stratosphere (above 10 hPa). Poleward of 60° latitude, and in the lower stratosphere (100–10 hPa), the regression typically explains less than 80% of the variance. In these regions, regression results are less reliable. As previously discussed in section 2, there are a number of caveats with this type of analysis, and care should be taken in interpreting the results. The MLR analysis is performed on the annual average model output, which takes advantage of both short-term (interannual) variability and long-term trends in separating the relative contributions. This MLR analysis has been compared and agrees well with chemical box model calculations in the upper stratosphere for the GEOSCCM [Oman et al., 2010].
 The contributions from Cly + αBry, temperature, NOy, and HOx are shown in Figures 2d, 2f, 2h, and 2j, respectively. Differences among models in the contribution from Cly + αBry are generally similar to relative variations in Cly + αBry (compare Figures 2d and 2c), with some exceptions. Most notably, MRI and CNRM-ACM have the largest ozone loss at this level from 1960 to 2000 despite not having unusually large amounts of Cly + αBry. This is related to larger sensitivities to Cly + αBry, and we quantify this sensitivity below.
 Temperature trends play an important role in the return of ozone to above historical levels: the four models with the largest cooling trend (LMDZrepro, CCSRNIES, CMAM, and UMSLIMCAT) (Figure 2e) have the largest increases in ozone relative to 1960 levels (see Figure 2b). The differences in cooling among models explain most of the variation in projected ozone by 2100, although, again, we note that there is not a simple one-to-one relationship between ozone and temperature.
 The contributions of NOy and HOx to changes in ozone are significantly smaller in most cases. Figures 2h and 2j show the contributions of NOy and HOx to ozone changes. For NOy, most models show losses of 0.2–0.3 ppm by 2100 with a few models showing somewhat larger loss, up to a loss of 0.7 ppm in AMTRAC3. While 5 hPa is typically one of the largest loss regions from NOy changes in volume mixing ratio, HOx has a much smaller impact at this level and typically is more important at pressures lower than 5 hPa [Jackman et al., 1986]. All models show a relatively small contribution (less than 0.1 ppm) from HOx except Universitá degli Studi dell'Aquila (ULAQ), which has the largest HOx trend (Figure 2i). ULAQ has one of the largest temperature trends at the tropical tropopause, so stratospheric water vapor increases more in their model than the others, which could be related to the coarser model resolution. In general, for the A1B GHG scenario considered in these CCMVal-2 simulations, NOy and HOx contributions are much smaller than the contributions from halogen recovery and stratospheric cooling over the 21st century. However, this may not be the case for other possible future scenarios for N2O and CH4 [e.g., Oman et al., 2010].
 The above analysis indicates, for the GHG and halogen scenario considered, that changes in Cly + αBry and temperature dominate tropical ozone evolution at 5 hPa, and differences in the simulated changes in these quantities explain most (but not all) of the variations among the models. However, by 2100 at 5 hPa, temperature and NOy generally show the largest contribution to the change in ozone with respect to 1960 values. Table 2 shows, for all 14 models, the numerical values of the individual sensitivity coefficients, mX, and the 3σ uncertainty of the fit for 25°S–25°N at 5 hPa. This was shown in tropical profile form by Oman et al.  and agrees well, in general, with the results presented here. HOx is not a significant driver of ozone variability at 5 hPa, which is consistent with these results as the sensitivity for about half of the models is not statistically significantly different from zero. NOy coefficients, with the exception of that from MRI, are all significantly different from zero but do have more uncertainty than the remaining explanatory variables. To see how representative the changes at 5 hPa are over a range of pressures, Figure 3 shows vertical profiles of changes between 2000 and 2100 for 25°S–25°N.
Table 2. Sensitivity Coefficients and Their 3σ Uncertainty of the Fit for 25°S–25°N, 5 hPa
Cly + αBry (ppm/ppb)
−0.36 ± 0.09
−0.14 ± 0.04
−0.31 ± 0.16
−0.37 ± 0.01
−0.15 ± 0.01
−0.27 ± 0.02
0.013 ± 0.004
−0.37 ± 0.01
−0.15 ± 0.01
−0.21 ± 0.03
0.001 ± 0.003
−0.31 ± 0.01
−0.12 ± 0.01
−0.19 ± 0.02
0.003 ± 0.002
−0.48 ± 0.02
−0.12 ± 0.02
−0.16 ± 0.07
0.006 ± 0.014
−0.38 ± 0.01
−0.13 ± 0.01
−0.19 ± 0.01
0.005 ± 0.001
−0.33 ± 0.01
−0.13 ± 0.01
−0.20 ± 0.02
0.001 ± 0.003
−0.65 ± 0.03
−0.10 ± 0.02
−0.02 ± 0.07
0.008 ± 0.019
−0.35 ± 0.01
−0.12 ± 0.01
−0.14 ± 0.04
0.001 ± 0.004
−0.34 ± 0.02
−0.15 ± 0.02
−0.17 ± 0.03
−0.010 ± 0.006
−0.33 ± 0.02
−0.13 ± 0.01
−0.23 ± 0.05
0.011 ± 0.007
−0.32 ± 0.01
−0.14 ± 0.01
−0.20 ± 0.03
0.005 ± 0.009
−0.31 ± 0.02
−0.13 ± 0.01
−0.18 ± 0.04
−0.37 ± 0.01
−0.14 ± 0.01
−0.22 ± 0.03
0.008 ± 0.004
Figure 3a shows each model's reference ozone profile for the year 2000, while changes by 2100 appear in Figure 3b. Also plotted on Figure 3a are Microwave Limb Sounder (MLS) satellite measurements (black squares) averaged from 2005 to 2009. The peak ozone increase is typically at 3 hPa (Figure 3b) with most models centered at around 1.5 ppm. Figure 3c shows that in the upper stratosphere, most models have a Cly + αBry decrease of 2 ppb over this time period. CCSRNIES and Solar-Climate-Ozone Links (SOCOL) have larger than average decreases in Cly + αBry, and AMTRAC3 has a smaller than average Cly + αBry change with a different vertical structure that is not consistent with observations, which is due to how AMTRAC3 treats the breakdown of halogens. Again, we see that MRI and CNRM-ACM have the largest ozone changes caused by Cly + αBry (Figure 3d), with AMTRAC3 having the smallest increase over the upper stratosphere. Most models show about equal contributions from decreases in Cly + αBry and decreases in temperature to the ozone increases over the 21st century. Observations from the MLS [Santee et al., 2008a] for HCl (data not shown) suggest that while a few models are in good agreement with MLS in the upper stratosphere, many appear to be on the low end with only CCSRNIES on the high end of observations. Possible causes for these differences include differences in the number of Cl species included in the models, as well as conservation issues in a few models [SPARC CCMVal, 2010].
Figure 3 shows that the key features seen at 5 hPa hold throughout the upper stratosphere (10–1 hPa). Models that have high (low) ozone changes at 5 hPa have similar characteristics throughout the upper stratosphere. Also, differences in Cly + αBry can explain most of the variations among the models with additional contributions from temperature. Figure 3h shows the contribution from NOy to changes in ozone: it is typically much smaller, and opposite in sign, than changes caused by Cly + αBry or temperature. In general, models show NOy-related ozone loss of 0.2 ppm around 3–10 hPa, with only AMTRAC3 exhibiting a larger loss. As presented in the Introduction, stratospheric cooling increases loss of NOy [Rosenfield and Douglass, 1998], causing a smaller increase (and even some decreases) (Figure 3g) in NOy over the 21st century than would be expected from N2O increases in the absence of climate change. The interpretation of changes in lower stratospheric ozone is complicated by the increasingly important impact of transport, which is not explicitly included in the MLR analysis and will be discussed in more detail in section 3.4.
 As discussed above, there are not always one-to-one relationships between changes in Cly + αBry and temperature and their contribution to ozone changes. One possible reason for this is the different sensitivities of ozone to Cly + αBry and temperature. That is, a unit of change in the explanatory variable can lead to a slightly different ozone change in each model. Figure 4 shows the vertical profiles of the sensitivity of tropical ozone (25°S–25°N) to Cly + αBry, temperature, and NOy calculated over the period 1960–2100. Overall, there is good agreement between the sensitivities of the various models. All models show a peak Cly + αBry sensitivity at 3 hPa, mostly around −0.35 ppm/ppb. However, two models stand out with much larger sensitivities, CNRM-ACM (−0.6 ppm/ppb) and MRI (−0.8 ppm/ppb) at 3 hPa. Also plotted on Figure 4 (black Xs) are the chemical box model calculations for July at 2°N shown and described by Oman et al. . The differences are very similar to those seen by Oman et al. , with the best agreement seen from 0.9 to 2.9 hPa and with slightly higher sensitivities seen at 5.6 and 6.9 hPa in the box model calculations compared to most models.
 Further insight into the sensitivity of modeled ozone to Cly + αBry can be gained by examining the abundance of ClO and BrO archived by the various CCMs. Here we provide a snapshot of the analysis presented in chapter 6 of SPARC CCMVal . Figure 5 shows profiles of the ClO/Cly ratio archived by four of the CCMs at 35°N for September 1993 and at 22°N for February 1996 (colored lines). The black lines represent the calculated ratio of ClO/Cly found using a photochemical steady state (PSS) box model [e.g., Canty et al., 2006, and references therein) constrained by fields of O3, H2O, CH4, Cly, NOy, aerosol surface area, and other constituents archived by each CCM group. Results are shown for four models: the two in question (MRI and CNRM-ACM) and two that archive ClO values that are well explained by the PSS simulation (CMAM and Whole-Atmosphere Chemistry-Climate Model (WACCM)). The numerical value on each panel is the “metric” assigned to each comparison, with a value of unity indicating perfect agreement with the PSS simulation.
 The profiles of ClO archived by CMAM and WACCM are in excellent agreement with the PSS simulation, indicating that implementation of the chemical mechanism within these models agrees well with the implementation of the mechanism within the PSS model. However, the MRI ClO profile is much higher (Figure 5, green dashed line) than the profile found using the PSS model. As explained in the SPARC CCMVal  study, the MRI chemical mechanism neglected the channel for the loss of ClO by the reaction of ClO + OH to yield HCl + O2. This leads to much higher values of ClO than are reported by other models or observed in the atmosphere [Santee et al., 2008b]. Even when we neglect this reaction channel in our PSS model (Figure 5, black dotted curves), we still cannot account for the high values of ClO archived by the MRI group. We conclude that the high sensitivity of O3 to Cly + αBry exhibited by the MRI model is due to the high concentrations of ClO in this model, attributed in part to the neglect of the ClO + OH → HCl + O2 channel. Including this reaction has been shown to produce better agreement with various observations [Chandra et al., 1995].
 The reason for the high sensitivity of O3 to Cly + αBry within the CNRM-ACM model is not known. For most altitudes, ClO within CNRM-ACM is simulated well by the PSS model. CNRM-ACM does simulate much higher values of ClO just above the tropopause than are accounted for by the PSS simulation; within CNRM-ACM, it appears that chlorine activation is occurring for warmer conditions than suggested by the PSS comparison. The high ClO just above the tropopause cannot explain the large sensitivity of O3 within this model to Cly + αBry shown in Figures 2d, 3d, and 4a. We have examined profiles of BrO within the CNRM-ACM and the PSS model (data not shown); we do not believe, based on this comparison, that the high sensitivity of O3 to halogens exhibited by CNRM-ACM is due to the calculation of BrO within this CCM. It is important to note that the PSS comparisons were done for the REF-B1 simulation (1960–2000, observed aerosols), whereas the bulk of this analysis is based on the REF-B2 simulation (1960–2100, background aerosols). Finally, our comparisons cannot reveal possible errors in the coding of a rate constant that would affect the rate of ozone loss by ClO or BrO but not affect the abundance of ClO or BrO (i.e., the reaction of ClO + O). The reason for the high sensitivity of O3 to Cly + αBry within the CNRM-ACM model remains unexplained.
 We performed an analysis of the statistical uncertainty from the MLR for the regression variable sensitivities as described by Oman et al.  and found similar results (data not shown). In addition, we have added the impact of autocorrelation of the residual on the uncertainty of the individual sensitivities [Tiao et al., 1990, their Appendix A]. In the middle to upper stratosphere, there is relatively more uncertainty in the sensitivity of ozone to NOy compared to much smaller uncertainty associated with sensitivity of ozone to Cly + αBry and temperature (also shown in Table 2). The larger uncertainty in the sensitivity to NOy could be due to NOy being less correlated with ozone variations than for the other explanatory variables. In general, in the lower stratosphere, from 100 to 60 hPa, the sensitivities are not statistically significantly different from zero. The most statistically significant sensitivities typically are found from 10 to 1 hPa for all explanatory variables except HOx. For HOx, the sensitivity is low (statistically indistinguishable from zero) at all altitudes except near the stratopause for most models.
 The evolution of upper stratospheric ozone and of the different contributing ozone loss mechanisms at midlatitudes is very similar to that in the tropics (data not shown). There are reasonably balanced, nearly equal contributions of Cly + αBry and temperature changes in the midlatitude upper stratosphere as was found for the tropics. Again, MRI and CNRM-ACM stand out with much larger contributions from Cly + αBry to ozone changes.
3.4. Lower Stratosphere
 As shown in Figure 1, the evolution of ozone in the tropical lower stratosphere (p > 20 hPa) differs from that in the upper portion. This can also be seen in the profiles shown in Figure 3b.
 Although an MLR analysis has been applied to the lower stratosphere, interpretation of these results is complicated by the large role of transport changes and the tight coupling of upwelling and temperature. In the tropical lower stratosphere, the loss that the MLR attributes to a temperature change is actually largely a response to increased upwelling, which acts to decrease ozone. This is consistent with Avallone and Prather , who showed that changes in upwelling are dominant in controlling tropical lower stratospheric ozone. Temperatures can be reduced by both increases in upwelling and decreased ozone. Previous studies have shown that an increase in the tropical upwelling is the principal cause of a decrease in tropical ozone [Shepherd, 2008; Li et al., 2009] with an additional possibly smaller contribution from “reverse self-healing” [Rosenfield et al., 2002]. Reverse self-healing occurs as upper stratospheric ozone rises above historical levels causing less ultraviolet radiation to penetrate into the lower stratosphere resulting in the decreased production of ozone.
 Since we do not have a transport proxy in the MLR analysis and there is significant temperature-upwelling coupling, we do not perform a detailed analysis of the MLR results in the lower stratosphere. Instead, we look at the correlation between upwelling and ozone changes in the CCMVal-2 models. Figure 6 compares the change in tropical ozone (25°S–25°N) at 50 hPa between 1960 and 2100 to the corresponding change in the residual vertical velocity (w*) at 70 hPa, for the 12 models that provided w* fields. We used the w* from the level below that of ozone, since ozone is impacted not only by local changes in upwelling but also by changes below impacting both advection and production. All models show an increase in tropical upwelling and a decrease in ozone over this time period. However, there is a large spread among the models in the change in both quantities. There is a fairly compact relationship between these two quantities, and a linear fit (black line) nearly intersects the origin, indicating that increases in upwelling are the dominant contributor to ozone decreases at this level. Most models indicate increases in upwelling of 0.04–0.10 mm/s and ozone decreases of 0.15–0.35 ppm over this time period. SOCOL appears to stand out from this group with significantly larger increases in upwelling and ozone decreases, which act in concert to produce the very large temperature reductions seen in Figure 3e. The impact of this larger change in upwelling in SOCOL can also clearly be seen in both total column ozone change (Figure 1b) and partial lower stratospheric ozone change (Figure 1h). In general, differences in the change in upwelling explain differences in lower stratosphere tropical ozone in the CCMVal-2 models. Also, this linear relationship exists when using the same level for ozone and w* (both at 50 and 70 hPa). It would be useful in future studies to further explore the impact of changes in transport and ultraviolet radiation on tropical ozone through additional CCM simulations. For example, a series of simulations in which the various controlling parameters are varied individually, and differences in the model output are examined to better isolate cause and effect as well as nonlinearities, would be especially helpful.
 In the midlatitude lower stratosphere, transport is also thought to play an important role for ozone changes, which are generally positive over the 21st century; however, the exact role is not quantified in the present analysis (see further discussion below). In the midlatitudes, the MLR analysis was performed, and results are shown in the SPARC CCMVal  study. Given the similar response in the upper stratosphere compared to the tropics, the results are not repeated here.
4. Polar Ozone
 We now consider changes in polar ozone. Austin et al. [2010b] evaluated the CCMVal-2 model simulations of the Antarctic ozone hole in detail and described both the processes that are well simulated and the areas that need additional work. More information on model biases, especially those involving temperature thresholds for polar stratospheric cloud (PSC) formation, are provided by Austin et al. [2010b].
 We will first focus on annual average changes and then look at a more detailed analysis of spring when the largest ozone depletion is observed. Figure 7 shows the evolution of annual average southern and northern polar ozone for total column ozone (a and b, top), upper partial column (c and d, middle), and lower partial column (e and f, bottom). We show ground-based total column ozone measurements (black squares, updated from Fioletov et al. ) with respect to 1964 in Figures 7a and 7b. SBUV MOD observations (data not shown) are only available for the sunlit portion of the year over the polar latitudes, which could lead to differences compared to the annual average model output shown in Figure 7.
 Evolution of the upper partial columns is very similar to that of midlatitudes (Figures 1d and 1f). However, the projections of the lower and total columns differ. All models show larger peak ozone loss around 2000 in the southern polar region than in the northern polar region. This is expected and seen in observations, as there is greater PSC formation in the southern polar region and a more stable polar vortex that lasts into spring each year. In general, there are larger increases in ozone amounts by 2100 compared to 1960 in the northern polar region. It is difficult to say quantitatively with the existing simulations how much of the differences seen by 2100 in the total and partial lower stratospheric column ozone are due to variations in polar chemistry, circulation changes, or a combination of the two. Additional simulations with fixed low chlorine concentrations or fixed GHG concentrations could help to quantify the relative impacts.
 The largest ozone losses occur in the spring in the lower portion of the stratospheric column, so we focus on this time and region in more detail. Figure 8a shows the Antarctic (60°–90°S) lower stratospheric column ozone for October. Peak ozone loss averages about 100 DU around 2000 with a range from models between 70 and 170 DU. SBUV MOD partial column observations (black triangles) show an ozone loss of about 80–90 DU between 1979 and ∼2000. The limitation on the observational reference year could be impacting the comparison as most models show between 20 and 40 DU ozone loss by 1979. We applied a linear regression analysis to the lower stratospheric column ozone amounts to calculate the contribution from Cly + αBry using α = 60.
Figure 8b shows the contribution of Cly + αBry and the residual (Figure 8c), which is much smaller and, for most models, typically less than ±20 DU. Cly + αBry dominates the partial column ozone until the end of the 21st century when the residual is, in most cases, of comparable magnitude. As a proxy for changes in the circulation of the stratosphere, we can look at the mean age of stratospheric air. The decrease in mean age of air (i.e., stronger circulation) (Figure 8e) by itself and in combination with upper stratospheric ozone returning to above historical levels acted to cause increased ozone change (i.e, positive residual). However, this could be tempered by the cooling of the polar lower stratospheric temperatures (Figure 8f shows 50 hPa temperatures), which occurs in nearly all models.
Figure 9a shows the Arctic (60°–90°N) lower stratospheric column ozone for March. We also note that this area is typically bigger than the size of the polar vortex, so this average also includes mixing effects from the midlatitudes (also to a lesser extent in the Southern Hemisphere during October). Peak ozone loss averages at 30 DU around 2000 with a range between 10 and 60 DU. SBUV MOD partial column observations (black triangles) show between 30 and 50 DU loss. Again, we linearly regressed the Cly + αBry against the partial lower stratospheric column ozone. Figure 9b shows the contribution of Cly + αBry and the residual (Figure 9c), which in this case is generally larger than in the Antarctic with a consistent 20–40 DU residual by 2100 for most models. This seems to indicate that in the NH polar regions, circulation changes cause a larger increase in ozone than seen in the SH by 2100.
5. Ozone Recovery
 The above differences in ozone return dates between regions are summarized in Figure 10, which shows the evolution of the multimodel mean (MMM) total and partial column ozone, with respect to 1960 for high latitudes (solid curves) and midlatitudes (dashed curves). Total column ozone for the MMM (Figure 10a) shows a return to 1960 values by the 2030s in the NH in both middle and high latitudes. In the SH, the total column ozone MMM exhibits a return to 1960 values by the 2060s in the midlatitudes and by 2080 at high latitudes. Nearly all the differences in total column ozone evolution are from the lower portion of the column (500–20 hPa) (Figure 10c) with very similar changes in both hemispheres occurring in the upper portion (Figure 10b). For a more in-depth discussion of return dates, especially with respect to 1980 values, see the study by Austin et al. [2010a].
 We see a significant difference in the dates ozone and halogens return to historical levels (i.e., 1960 levels). The MMM Cly + αBry does not return to 1960 values by 2100. This is later than the return of ozone (for extratropical regions). The cause of the earlier return of ozone in midlatitudes [Austin et al., 2010a, Figure 16] in both hemispheres is likely from an increased circulation [Austin and Wilson, 2006; Waugh et al. 2009; Li et al., 2009] along with increased upper stratospheric ozone from stratospheric cooling. The impact of circulation changes on the midlatitude ozone return date is discussed in more detail by Austin et al. [2010a].
 There is a noticeable difference in the dates when Northern and Southern Hemisphere ozone returns to 1960 values (with no such difference in Cly + αBry; data not shown). This interhemispheric difference is likely caused by asymmetries in transport between the two hemispheres. Some of the delay in the SH return is likely from asymmetries in polar ozone loss since there is significant polar ozone loss occurring in the SH in the early 21st century when NH ozone is returning to 1980 levels (not shown with reference to 1980 levels), causing less ozone to be transported into SH midlatitudes when the vortex breaks down [Atkinson et al., 1989]. The impact of polar ozone losses being transported into the midlatitudes likely plays the dominant role in the later return to historical values seen in models in the SH midlatitudes during first half of the 21st century with any asymmetries in the strengthening of the stratospheric circulation likely having a larger impact toward the later half of the 21st century.
 Simulations of the past and future were performed using 14 chemistry-climate models for the CCMVal-2 activity. Austin et al. [2010a] examined the evolution of total column ozone and compared the recovery of column ozone and Cly in these models. Here we have contrasted the evolution in the upper and lower stratosphere and used MLR in the tropics to attribute ozone changes to the predictors with a particular focus on the 21st century.
 The simulations presented here and by Austin et al. [2010a] show that there is some general agreement in the ozone evolution among the models, with all showing column ozone decreasing from 1960 to around 2000 and then increasing over the first half of the 21st century. Models also show that the column ozone evolution varies with latitude, especially in the latter half of the 21st century. There are clearly some quantitative differences in ozone evolution across the CCMVal-2 simulations, e.g., a large spread in simulated return of ozone to historical values [Austin et al., 2010a].
 Separation into partial columns above and below 20 hPa reveals that the latitudinal differences in the evolution of the column ozone are almost completely attributed to differences in the lower stratosphere (the region of the atmosphere below 20 hPa). In all models, there are only weak latitudinal variations in the evolution of upper stratospheric ozone, and at all latitudes, upper stratospheric ozone increases throughout the 21st century and returns to 1960 levels well before the end of the century. There is, however, a large spread in dates of return to historical values, with, for example, dates of return to 1960 levels at 5 hPa varying from 2020s to 2060s. Over the 21st century, a multiple linear regression analysis indicates that the upper stratospheric ozone increase comes from almost equal contributions of decreases in halogens (Cly + αBry) and cooling from increased greenhouse gas concentrations (for the A1B greenhouse gas scenario considered in these simulations), with only small contributions from NOy and HOx. This result depends on the greenhouse gas scenario. Larger changes in N2O and CH4 (source gases of NOy and HOx, respectively) than considered here could impact the relative contributions from NOy and HOx, as shown by Oman et al. . It is also important to keep in mind the limitations of the MLR analysis discussed in section 2.2 when considering these results.
 In the tropical lower stratosphere, there is a steady decrease in ozone through the 21st century in all models, whereas ozone in middle and high latitudes increases during the 21st century. As a consequence of increased tropical upwelling induced by rising levels of GHGs, modeled tropical column ozone does not return to 1960 levels even by the end of the century, whereas extratropical ozone returns or is close to 1960 levels well before the end of the century. In the model simulations, while the largest decreases in partial lower stratospheric ozone columns occur during the second half of the 21st century, some smaller decreases are already evident over the recent past. However, ground-based and SBUV satellite observations of total and partial tropical lower stratospheric ozone columns do not show this decrease. Randel and Wu  reported negative trends in tropical lower stratospheric ozone using SAGE I and SAGE II data. More work should be done to understand the actual time history of tropical, lower stratospheric ozone to enable a more meaningful comparison to modeled trends. Models with larger future ozone decreases have larger increases in tropical upwelling. Changes in transport also play an important role in the evolution of midlatitude lower stratospheric ozone and contribute to the earlier return of ozone to specific historical values than is found for Cly + αBry.
 There are quantitative differences between the hemispheres in the simulated ozone evolution. In particular, ozone returns to historical values earlier in the Northern Hemisphere. This difference in hemispheric return dates varies between 20 and 50 years depending on the latitude of interest for the 1960 reference year chosen, and it is almost completely related to differences in the lower stratosphere. Additional simulations would need to be completed to quantitatively attribute causes for hemispheric differences. However, based on existing simulations, it appears that the larger ozone loss from polar chemistry in the SH and the transport of this ozone-depleted air into the midlatitudes as the vortex breaks down is one cause, especially during the first half of the 21st century. Changes in the stratospheric circulation could also play a role in these differences.
 At high latitudes, especially in the SH (Figure 8), Cly + αBry dominates long-term trends in the lower portion of the column ozone (and total column ozone) with a small (generally positive) residual. This residual is likely caused by circulation changes and upper stratospheric return to above-historical levels.
 These results reinforce the conclusions of Eyring et al.  using simulations from the CCMVal-1 activity: decreasing levels of halogens, continued stratospheric cooling, and changes in circulation are the major factors driving 21st century ozone trends. It is important to keep in mind that all of these simulations are based on a single GHG and halogen scenario, with only one model including an interactive ocean. In addition, the use of mixing ratio-based halogen boundary conditions rather than emission-based constrains the model's response [Douglass et al., 2008]. Future work should be done to explore these issues and their impact on ozone evolution in more detail.
 This research was supported by the NASA MAP, ACMAP, and Aura programs and the NSF Large-scale Climate Dynamics program. We thank Susan Strahan for very helpful comments on this manuscript, and we thank three anonymous reviewers for thoughtful, constructive reviews of the submitted paper. We acknowledge the Chemistry-Climate Model Validation (CCMVal) Activity of the WCRP SPARC project for organizing and coordinating the model data analysis activity and the British Atmospheric Data Centre for collecting and archiving the CCMVal model outputs. CCSRNIES research was supported by the Global Environmental Research Fund of the Ministry of the Environment of Japan (A-071 and A-0903), and simulations were completed with the supercomputer at CGER, NIES. The MRI simulation was made with the supercomputer at the National Institute for Environmental Studies, Japan. The contribution from the Met Office Hadley Centre was supported by the Joint DECC and Defra Integrated Climate Programme-DECC/Defra (GA01101). The contribution from the LATMOS-IPSL was supported by the European Commission through the funding of the RECONCILE and GEOMON projects. The National Center for Atmospheric Research is operated by the University Corporation for Atmospheric Research under sponsorship of the National Science Foundation. Any opinions, findings, and conclusions or recommendations expressed in the publication are those of the authors and do not necessarily reflect the views of the National Science Foundation.