This paper uses an unsupervised classification technique, Gaussian Mixture Modelling (GMM), to classify ozone profiles from historic and future climate simulations of the UKESM1 model. They find 6 classes of profiles, which have coherent distributions.  The study then investigates how the classes change between the historic and future climate simulations, finding that the tropical classes expand in the future projections.  The paper suggests that this GMM method could be useful for inter-model or model-data comparisons by enabling profiles to be grouped by coherent structures rather than set latitudes.  The use of machine learning to classify ozonesondes for model comparisons is a promising approach, and the paper shows that the resulting classifications are consistent with physical interpretations of ozone variability.  However, the paper could be improved by more clarity or examples on what specific advances are provided by this methodology compared to other approaches.  In addition, stronger justification for some of the methodological choices is needed, as described in the comments below.

General comments:

1. Since the chemical and dynamical processes controlling ozone concentrations in the stratosphere are quite different from those controlling near-surface ozone, what is the rationale for performing the GMM classification on the entire ozone profile? Could you instead cluster different vertical regions, such as stratosphere or troposphere, separately?  It seems like the results might be easier to interpret and the clusters more applicable to model comparisons of specific features like surface concentration if signals from near-surface processes weren’t mixed together with signals from stratospheric circulation in the creation of the clusters.
2. What is the advantage of using the GMM clustering method over just grouping profiles by e.g. tropopause height or altitude of peak ozone, since these seem to be prominent features distinguishing the derived classifications? It is encouraging to see that the GMM analysis leads to results that are consistent with known sources of variability, but to justify the complexity of this GMM approach, it would be helpful to also highlight specific cases where the GMM creates a more meaningful classification than could be obtained with a single variable such as tropopause height.

Specific comments:

1. Lines 46-49: Please provide a reference.
2. Lines 59-60: The discussion of previous work on ozone clustering could be expanded.
3. Lines 97-98: The requirement of surface pressure reaching 1000 hPa seems like a significant limitation. Would the results be much different (and the coverage increase) if you used something like 900 hPa instead?
4. Line 130: How does the pressure level standardization affect the relative importance of the stratospheric versus the tropospheric portions of the profile in determining the clusters?
5. Line 149: Define BIC and refer the reader to the description in the appendix
6. Lines 194-201: Do the higher tropopause and higher surface values both contribute to the definition of this cluster, or is it just that the clusters vary strongly with latitude (as shown in Fig. 4) and many other features also co-vary with latitude?
7. Line 216: is mPa the right unit here?
8. Line 219: Replace “reasonable” with something more quantitative
9. Fig 4 (and 5) and Fig 4 caption: Does “median” make sense with respect to classifications here? Are the classes quantitatively ordered such that class 3 is in between classes 2 and 4?  Also, is there much temporal variability (within the decade) in what class a particular grid box falls in?  If so, it would be nice to show that since it could help clarify how the GMM classification differs from a purely latitude-based classification.
10. Line 249: Does this mean the fact that class 1 has the lowest ozone, or the fact that the class 1 ozone is lower in the historic run, is consistent with the reduction with precursors?
11. Lines 272-275: Is this explanation proven by your analysis or just consistent with your results?
12. Lines 284-285: Please explain how this conclusion is reached from Figs 4-5
13. Lines 295-297: Is it possible to relate this quantitatively to the extent of the model’s Hadley cell?
14. Lines 298-304: Are these results different from what would be inferred with latitudinal averages?
15. Line 321: This statement needs more support. Relate to Table 4?

Technical corrections:

1. Line 220: Please reword to clarify “Biomass burning in Africa produces…” or similar
2. Line 249: “consistent” not “in consistency”
3. Line 253: Should this say “stratosphere” or “atmosphere”?
4. Lines 270-271: Please reword this sentence for clarity

The authors have developed an interesting concept for exploring the shift in atmospheric regimes with similar ozone profiles. I have several concerns regarding the methodology which need to be addressed before I can recommend the paper for publication.

Major comments:

1) This analysis uses annual mean ozone profiles, which provide no information on ozone’s high day-to-day variability in the troposphere. The method is also applied to full profiles that include the troposphere and stratosphere. The result is a very smooth ozone field that is dominated by the stratosphere.  As I describe below, the resulting clusters seem to be insensitive to prominent tropospheric ozone features.  For this reason, I think the analysis needs to be applied to the troposphere and the stratosphere separately.

2) What is the impact (or limitation) of using annual averages?  Ozone concentrations vary widely from summer to winter in both the troposphere and the stratosphere.  How different are the clusters if the analysis is applied separately to summer and winter months?  Another problem with the annual average is that mid-latitudes are heavily influenced by polar air masses in winter (low tropopause), and by tropical air masses in summer (high tropopause).  So the annual average is just an unrealistic homogenization of very different air masses, and does not reflect the typical ozone profiles one might find in any given month or season. 

3) The analysis makes no use of observations, and with no evaluation against real-world data we are unable to understand the accuracy of the method.  Stauffer et al. (2018) clustered ozone profiles at more than 2 dozen ozonesonde stations worldwide.  I realize the authors can’t use sparse observations as the basis for this global-scale analysis, but they can certainly evaluate the results against observations.  The authors should examine the observed profiles above the ozonesonde stations that lie within each of the clustered regions.  Do the profiles within each region have similar characteristics?  If so, then the method is applicable to the real-world; if not, then the usefulness of the method is questionable.  What is the result when the observations are then examined by season?  Are the observations within each cluster similar to each other in summer, and also in winter?  Or does everything break down (see my comment above about seasonal variability in the mid-latitudes). 

Other comments:

Line 59

When reviewing clustering techniques as applied to ozone profiles, the authors should include Stauffer et al. (2016, 2018).

Line 61

The perceived methodology and aim of Chang et al. 2017, as stated in the manuscript, is not correct.  Chang et al. 2017 are not seeking to cluster similar ozone monitoring sites. Rather they are trying to quantify the regional-scale, long-term trend of ozone, while accounting for the spatial distribution of the sites and the correlation between sites.  This method accounts for the uneven distribution of sites and prevents any heavily-sampled sub-region from exerting an out-sized influence on the trend.

Line 95-98

I don’t understand why the study is limited to 1-1000 hPa. This omits a large section of the globe, i.e. land regions more than 100-200 m above sea level.  I realize the method cannot tolerate missing values, but why not conduct the study for profiles in the range 1-950 hPa; this way you retain most of the land areas.

Line 110

This statement is problematic:

“The motivation behind withholding the geographical information is that there is no reason for the vertical ozone structure of the profile to be unique to a given region (Maze et al., 2017).”  Using a paper that deals with ocean temperature, the authors seem to suggest that there is no discernable structure in the global ozone distribution and that one region is no different from another.  Yet, plenty of observation-based studies identify clear structure in the global ozone distribution that varies with season [Kley et al., 1996; Thouret et al., 1998; Oltmans et al., 1996, 2004; Thompson et al., 2003; Cooper et al., 2007; Gaudel et al., 2020;].  Therefore, certain profile types are more likely to occur in some regions than in other regions.  This statement needs to be revised.

Line 173

To say that the tropopause is around 300 hPa is a gross over-simplification. As can be seen in Figure 2, there are plenty of profiles in which the tropopause is around 150 hPa, which is common in the tropics.

Line 175

The statement that ozone increases near the surface is problematic because ozone is plotted in units of mPa.  If ozone is plotted in units of ppbv (the typical unit for evaluating air pollution levels in the troposphere) then we would see that the average ozone profile has more ozone in the upper troposphere, especially at mid- and high latitudes (see the ozone profile papers that I cited above). Furthermore, Jaffe and Wigder (2012) is not a sufficient reference because they only discuss ozone at the surface and do not mention the vertical distribution of ozone.

Figure 10 of Gaudel et al. (2018) compares the tropospheric ozone distribution from five satellite products.  In the tropics two consistent features are the very low ozone above Indonesia and the very high ozone in the tropical South Atlantic.   In situ observations have confirmed these features (Kley et al. 1996; Thompson et al., 2021) and the ozone enhancement in the South Atlantic is far greater than any other tropical region (Bourgeois et al., 2021).  Figure 4 shows that the clustering routine completely misses these prominent features, which means, 1) the model does not simulate these features, which calls into question the usefulness of the model; 2) the clustering routine is completely dominated by the stratosphere and has no sensitivity to the troposphere; if this is the case, then the clustering should be applied to the troposphere and stratosphere separately.

Line 216

Why is the high surface ozone only attributed to biomass burning?  This cluster spans the major fossil fuel combustions regions of the northern hemisphere, which are known to drive ozone production across the region. 

Line 265

The statement that ozone precursor emissions generally increase under SSP5-8.5 isn’t really correct as emissions continue to decrease in developed nations, but increase in the developing world. This discussion should also consider the findings of Zanis et al., 2022.

References: 

Bourgeois, I., Peischl, J., Thompson, C. R., Aikin, K. C., Campos, T., Clark, H., Commane, R., Daube, B., Diskin, G. W., Elkins, J. W., Gao, R.-S., Gaudel, A., Hintsa, E. J., Johnson, B. J., Kivi, R., McKain, K., Moore, F. L., Parrish, D. D., Querel, R., Ray, E., Sánchez, R., Sweeney, C., Tarasick, D. W., Thompson, A. M., Thouret, V., Witte, J. C., Wofsy, S. C., and Ryerson, T. B.: Global-scale distribution of ozone in the remote troposphere from the ATom and HIPPO airborne field missions, Atmos. Chem. Phys., 20, 10611–10635, https://doi.org/10.5194/acp-20-10611-2020, 2020.

Cooper, O. R., M. Trainer, A. M. Thompson, S. J. Oltmans, D. W. Tarasick, J. C. Witte, A. Stohl, S. Eckhardt, J. Lelieveld, M. J. Newchurch, B. J. Johnson, R. W. Portmann, L. Kalnajs, M. K. Dubey, T. Leblanc, I. S. McDermid, G. Forbes, D. Wolfe, T. Carey-Smith, G. A. Morris, B. Lefer, B. Rappenglück, E. Joseph, F. Schmidlin, J. Meagher, F. C. Fehsenfeld, T. J. Keating, R. A. Van Curen and K. Minschwaner (2007),  Evidence for a recurring eastern North America upper tropospheric ozone maximum during summer, J. Geophys. Res., 112, D23304, doi:10.1029/2007JD008710.

Gaudel, A., et al. (2018), Tropospheric Ozone Assessment Report:  Present-day distribution and trends of tropospheric ozone relevant to climate and global atmospheric chemistry model evaluation, Elem. Sci. Anth., 6(1):39, DOI: https://doi.org/10.1525/elementa.291

Gaudel, A., O. R. Cooper, K.-L. Chang, I. Bourgeois, J. R. Ziemke, S. A. Strode, L. D. Oman, P. Sellitto, P. Nédélec, R. Blot, V. Thouret, C. Granier (2020), Aircraft observations since the 1990s reveal increases of tropospheric ozone at multiple locations across the Northern Hemisphere. Sci. Adv. 6, eaba8272, DOI: 10.1126/sciadv.aba8272

Kley, D., PJ Crutzen, HGJ Smit, H Vömel, SJ Oltmans, H Grassl (1996), Observations of near-zero ozone concentrations over the convective Pacific: Effects on air chemistry, Science 274 (5285), 230-233

Oltmans, S. J., et al. (1996), Summer and spring ozone profiles over the North Atlantic from ozonesonde measurements, J. Geophys. Res., 101( D22), 29179– 29200, doi:10.1029/96JD01713.

Oltmans, S. J., et al. (2004), Tropospheric ozone over the North Pacific from ozonesonde observations, J. Geophys. Res., 109, D15S01, doi:10.1029/2003JD003466.

Stauffer, R.M., Thompson, A.M. and Young, G.S., 2016. Tropospheric ozonesonde profiles at long‐term US monitoring sites: 1. A climatology based on self‐organizing maps. Journal of Geophysical Research: Atmospheres, 121(3), pp.1320-1339.

Stauffer, R.M., Thompson, A.M. and Witte, J.C., 2018. Characterizing global ozonesonde profile variability from surface to the UT/LS with a clustering technique and MERRA‐2 reanalysis. Journal of Geophysical Research: Atmospheres, 123(11), pp.6213-6229.

Thompson, A. M., et al. (2003), Southern Hemisphere Additional Ozonesondes (SHADOZ) 1998–2000 tropical ozone climatology 1. Comparison with Total Ozone Mapping Spectrometer (TOMS) and ground-based measurements, J. Geophys. Res., 108, 8238, doi:10.1029/2001JD000967, D2.

Thompson, A. M., and et al., (2003), Southern Hemisphere Additional Ozonesondes (SHADOZ) 1998–2000 tropical ozone climatology 2. Tropospheric variability and the zonal wave-one, J. Geophys. Res., 108, 8241, doi:10.1029/2002JD002241, D2.

Thompson, A.M., Stauffer, R.M., Wargan, K., Witte, J.C., Kollonige, D.E. and Ziemke, J.R., 2021. Regional and Seasonal trends in tropical ozone from SHADOZ profiles: Reference for models and satellite products. Journal of Geophysical Research: Atmospheres, 126(22), p.e2021JD034691.

Thouret, V., Marenco, A., Nédélec, P. and Grouhel, C., 1998. Ozone climatologies at 9–12 km altitude as seen by the MOZAIC airborne program between September 1994 and August 1996. Journal of Geophysical Research: Atmospheres, 103(D19), pp.25653-25679.

Zanis, P.; Akritidis, D.; Turnock, S. et al. Climate Change Penalty and Benefit on Surface Ozone: A Global Perspective Based on CMIP6 Earth System Models. Environ. Res. Lett. 2022, 17 (2), 024014.  https://doi.org/10.1088/1748-9326/ac4a34.