In this study, the authors improve and apply an aerosol regime analysis published previously (Li et al. 2022, L22) to past and future simulations of aerosols in the ECHAM-MESSY climate model. They find that cleaner regimes dominated in 1750 and 1850 and may dominate again in the future depending on the emission scenario.

The paper is well written, and figures illustrate the discussion well. However, the study suffers from only quantifying the emission-driven changes in aerosol regimes, not looking at climate-driven changes. This limitation has two consequences that severely reduces the insights gained by the study:

• First, there is little point in performing a regime analysis, or even in running a climate model. All the information is already in the emissions. This is particularly obvious in section 3.2, which reaches the same conclusions as the preceding sections but based on emissions only.
• Second, focusing on emissions leads to the probably very misleading conclusion that some future scenarios lead back to preindustrial conditions. That is unlikely: oceans will have warmed, soils and forests will have changed, bare soils will have expanded or shrunk, droughts and other weather extremes will be more frequent. All those changes will affect emissions from land- and ocean-based biogenic aerosols, biomass-burning aerosols, mineral dust, thus affecting the aerosol regimes, as noted in L22 and lines 136.

Still, there are insights to be gained from an emission-only focus, which is why I recommend major revisions. A possible way to address the criticisms above could be to:

• Connect changes in emissions in a given region to changes in aerosol regimes over a wider area. That would involve merging section 3.2 with the rest of the analysis.
• Clearly acknowledge and flag the key limitation of the study in the abstract and add a long discussion to section 4, to elaborate on how climate feedbacks might alter the conclusions. There is an increasing body of work on climate-driven aerosol feedbacks. Chapters 6 (and perhaps 7) of the IPCC AR6 is a good starting point.

Another point of concern comes from Section 2.3. I can just about understand why present-day clusters cannot be used to analyse regimes at other points in time, but there are two aspects that I do not understand:

• First, how does applying a random forest helps? As stated in the introduction, there are few pristine regions in the present day, and they are limited to the aerosols that happen to be emitted in those regions. Don’t you necessarily end up extrapolating out of your training dataset when applying the learning to other times?
• Second, why did you choose present day as the reference? As stated in the introduction, the reference is normally preindustrial, with some variations as to which year or period is used in practice. And, linking to the previous point, how much does the period used for training matter in terms of regime identification?

Other comments:

Lines 114-115: Could note that the IPCC AR6 uses 1750 as a preindustrial reference to assess radiative forcing, but uses 1850-1900 for other aspects, like surface temperature change.

Line 126: “proven” is too strong a word since you do not define what “properly” means. Model skill depends on the level of detail of the comparison, and on the purpose.

Line 132: Would be useful to say here that the time slices are 10-year long. That information only appears on line 149, which is a bit late.

Line 149: So learning is done on annual means only? Wouldn’t you get more information when using seasonal or monthly means, given the large seasonality of many aerosol types?

Line 208: It would be useful to summarise here the regimes according to L22, and especially what “level” means in, for example, “dust dominated level 1”. That information is partly given is section 3.1, which is late.

Lines 315-317: It would be good to remind the reader that the impact of climate change is not considered here, because this kind of conclusion would probably change if climate feedbacks onto aerosol emissions were included.

Line 361: What is the meaning of having two different regimes for the Arctic and Antarctic?  It seems to be purely related to magnitude, rather than changes in composition.

Line 490-494: That section needs to be more critical of the emission datasets. This is especially true of biomass-burning and biogenic emissions. We do not know what they were in preindustrial conditions (see for example Marlon et al. 2016) and the future climate-driven changes are unlikely to be properly represented in the CMIP emission datasets. Those uncertainties are crucial to some of the conclusions of the paper, for example paragraph 284-299.

 Technical comments:

• Line 161: evaluating -> evaluation