The manuscript of Tsimpidi et al. “Aerosol Composition Trends during 2000 -2020: In depth insights from model predictions and multiple worldwide observation datasets” compares measured and modeled aerosol pollutants between 2000-2020 in a global scale. The paper includes a large experimental data set, and it examines extensively the pollutants trend for each area. This is an important study that should be published, after modifications.

Major comments:

My main concern is that due to the large amount of the compared data (time and space) the paper is quite long (50 pages without including references), and the reader gets tired fast. From my point of view, it is difficult to digest all this information. So, it should be somehow more concentrated and shortened. In addition, there is a lot of statistical information but in general there is little connection between all these results. Moreover, the reasons for any discrepancies between measurements and simulations should be discussed in more detail and propose modification/addition in the model to capture more accurately the measurements.

Section 4 (In depth model Evaluation) describes the comparison between measured and modeled mass concentrations. In each sub-section (4.1-4.4) the authors compare the average mass concentration (sulfate, nitrate, ammonium, OA, SOA and POA) from each campaign to the average mass concentration that the model predicts for the specific site and time. What is the time resolution of the EMAC? Hourly/daily/monthly? This should be explained in the text. The comparison between the model and the measurements should be made throughout the whole campaign and not taking on only one value from each campaign. This could be misleading for the model’s performance as for example some days (if the model resolution is every 24 hours) could be very badly simulated, leading to a high overall discrepancy.

On the contrary section 5 (Aerosol Trends) is much more meaningful for measurements and model comparison. The core of the results should be this part. Section 4 should be complementary to section 5, and I suggest that the authors should incorporate selected parts of section 4 to section 5 accordingly. The discussion should be done by area (i.e. Europe, N. America, E. Asia) so that the reader reads the “story” of each area.

Minor comments:

There are several mistakes using the words “best” and “worst” in the text. For example, lines, 618, 620, 649, 651, 810 etc. Please check the whole manuscript and make the appropriate changes. Also check the usage of the word “highest” (e.g., lines, 286, 297, 411 etc.).

Line 70: Please replace “form” with “formed”.

Line 228: “high OA fractions with regional means” please rephrase.

Line 414: PM_2.5 please use subscript for ‘2.5’.

Line 624: Please replace “underestimating” with “it underestimates”.

Line 626: Please add “it” after therefore.

Line 637: Please replace “show” with “shows”.

Line 653: Please replace “resolved” with “simulated”.

Line 678: Please delete “the”.

Line 681: Please replace “lie” with “lies”.

Line 713-714: “Ammonium tends to be overestimated during autumn and underestimated during the rest of the seasons; especially during the summer” Is there any explanation for this tendency?

Line 719: “While the good model performance” please rephrase.

Lines 727-728: “On the other hand, ammonium is overpredicted close to the deserts of Inland.

728 China (e.g., over Tibet) and over South Korea” Do you have any explanation about this behavior?

Line 750: “EMAC tends to overpredict some low OA concentrations measured by AMS” this sentence is not very clear, please rephrase.

Line 785: “.. evaporation of organic compounds upon emission…” So, vaporization is not considered by the model? Please explain.

Line 803: Please replace “are in a good” with “are in a good”.

Figure 19: There are no g and h subplots, so please make the appropriate changes in the figure caption.

Figure 21: There are no g and h subplots, so please make the appropriate changes in the figure caption.

Line 1065: PM₁ please use subscript for ‘1’.

Line 1080: Since there is little discussion about EC in the text, it should not be mentioned in the conclusions.

Figures 17, 20 and 22 should be moved to the SI, they are just the average of Figures 16, 18, 19 and 21 and they don’t add value to the paper.

This manuscript describes the trends in global surface AMS/ACSM PM1 and PM2.5 measurements, along with a global model simulation (EMAC) for the first two decades of the 21^st century. The focus of the analysis is the aerosol composition and how it changes over time. The results are not particularly novel, but the analysis is comprehensive, the paper is clearly written, and I think this will be a useful reference for the community. I also found the discussion to be well-balanced and well-supported. I have a few comments that I believe should be addressed prior to publication. I also include a long list of more minor corrections/suggestions.

1. Two important aspects of terminology:
   1. Surface: the manuscript focuses exclusively on surface concentrations. This should be made clear in the title, the abstract (line 17), and to varying degree throughout in the text.
   2. Non-refractory PM: Throughout the manuscript, the aerosol speciation is given for “fine aerosol” but particularly when discussing the AMS/ACSM observations, what is reported is “fine non-refractory aerosol”. This should be added throughout (e.g.: line 20, line 263, 265, 272, 304, etc) so that the reader is clear that this is the total excluding BC, dust, SS, etc.
2. In Section 2: More detail is needed on the measurement uncertainties, detection limits, collection efficiency (in the case of the AMS/ACSM) and the variability in instrument or operation. All of these could be relevant to the comparison across observations and with the model.
3. Section 5: The model trends, which in many cases do not match the observations, are largely be driven by emissions. Thus, a more detailed discussion (and plots) of the trends in emissions are needed. Can the authors compare the trends in emissions with other inventories? To what degree are anthropogenic emissions consistent with CEDS or regional inventories? The paper is generally lacking substance on the “why” for model failures and a more thorough discussion and comparison of emissions could provide much needed insight here.

Minor Comments

1. Line 29: “with varying accuracy in model predictions” is rather vague. It would be useful to provide more concrete conclusions in the abstract. For example, it’s clear that the model underestimates the trends in inorganics in North America and Europe.
2. Line 66: SO2 also comes from DMS
3. Line 67-68: This describes the source of HNO3, but given that not all HNO3 forms particulate nitrate, there should be some mention here of the role of thermodynamic partitioning
4. Line 72: Fires and VCPs are also sources of VOCs
5. Lines 73-103: This paragraph is rather uneven. Details on policy and trends are provided for Europe, but no trends are given for North America or Asia. Can more detail on these regional trends be added? It would also be useful to insert a couple of sentences on the policy or trends context for the rest of the world.
6. Lines 115-129: I would suggest that the authors might want to invert the paragraph to start with the observations to be consistent with the flow of the manuscript.
7. Section 2: I believe that the authors are using dry PM1 and PM2.5 throughout (both in the model and observations) – this should be specified in this section.
8. Section 3: How is the model sampled for the comparison with observations that follow? Daily averages, 3-day averages where relevant, etc.?
9. Line 182: in principle POA can be “fresh” (lower O:C) or “aged” (higher O:C). I believe that aged POA is characterized as SOA in EMAC and as OOA in the observations, which is perhaps why the authors are focusing on “fresh” POA here, but the sentence should be clarified. It would also be worth clarifying that you mean only fresh POA on line 777.
10. Line 259: suggest text be modified to read “well represented at one site over Africa”
11. Figure 4: can the authors use colours to show which regions in Figure 3 correspond to the regions in the barplot?
12. Figure 4: what fraction of the data falls below the detection limits?
13. Lines 279-284: It is odd that the authors here focus on the sulfate instead of the nitrate. The sulfate concentrations are similar in Europe and North America, it is the nitrate that is considerably lower in NA. I recommend that the authors alter the text to focus on the “surprisingly low nitrate in North America”, and also ask whether this is surprising or rather consistent with the NOx emissions over Europe vs North America
14. Line 380: how many SPARTAN sites are used here?
15. Line 395: The text refers to OA, but Figure 7 shows OC. Please be consistent and if/when OA is used, describe the application of the OM:OC.
16. Line 418: what about crustal? It appears to be quite important in the Middle East and Africa so it seems odd that the other species add up to 100%
17. Line 511: which AVOC and BVOC species are included as SOA precursors?
18. Line 515-516: does this imply that there is no loss of carbon to the gas phase via fragmentation (i.e. HCHO)?
19. Section 3.4: It would be helpful to specify which (all?) emissions inventories are year-varying
20. Figure 9: The shades of green are quite difficult to distinguish; I suggest using more distinct shades. Also: rainbow colour bars are to be avoided, so I would strongly suggest using a different colour bar for the map (https://thenextweb.com/news/stop-using-rainbow-maps-doesnt-do-data-justice-syndication)
21. Lines 606-607: Why not include total dust?
22. Section 4: It would be helpful if the statistics include R2 to summarize the model skill in capturing the variability. This could be discussed in the text and should be included in all the tables.
23. Scatterplots (Figures 10, 11, 12, 13): Most of these are difficult to see (lots of wasted white space). The log-log axes are not very helpful. I would strongly recommend that the authors consider using linear scales.
24. Line 765: Replace “every” with “many”. The Tsigaridis et al. 2014 comparison only shows NMB, and is therefore not very relevant for showing “high concentrations”. This reference is also over 10 years old now, so it may be too strong to say that all models still perform similarly.
25. Figure 15: Are these differences statistically significant?
26. Figure 15: Could you also add BC, dust, and SS?
27. Figure 17: The authors might consider annotating the x-axis. There is a lot of detail in the caption that the reader has to hunt for.
28. Lines 1061-1063: Given that model does not reproduce the observed trends in many species, regions, it would not be a very faithful representation of the “global trends in atmospheric aerosol composition”. I suggest that the authors modify this paragraph to focus on the observed trends and then how these trends were used to test the model simulation.
29. Lines 1085-1086: What about dust?
30. Line 1107: Please correct the text: the model does not show “a major decline”