Overview:

The paper is an evaluation of biases, variability and multi-year trends of the CAMSRA and MERRA-2 reanalysis data surface sets for O3, NO2, CO, SO2, PM10 and PM2.5 for the 2003-2020 period over Europe. CAMSRA shows better agreement with surface observations than MERRA-2 with the exception of SO2.

General comment:

The evaluation procedure applied in the paper is sound and provides valuable insight for the user of the data sets. 

My main concern of the paper is the selection of the stations for the evaluation. Urban stations, which are probably the largest fraction of the European AQ networks, were considered as “background” stations for the paper and combined with the rural stations.  I think it is not common practice to include urban stations in the group of background stations. In any case, a stratification between “rural” and “urban” stations is strongly recommended as the scale represented by urban observations is often much smaller as the grid-point resolution of the evaluated models. I think the authors need to quantify the errors against rural stations and urban stations separately, and clarify if the presented results are dominated by the comparison with urban stations or not.  

The paper can further be improved by adding more information about the speciation of the PM10 and 2.5 values. It would be interesting to know how large the sulphate (given the differences in SO2) contributions were, and to what extend dust and sea salt contributed to PM in both data sets.

Specific comments.

Please include more references in table 1 and discuss their findings. For example, 

Huijnen, V., Miyazaki, K., Flemming, J., Inness, A., Sekiya, T., and Schultz, M. G.: An intercomparison of tropospheric ozone reanalysis products from CAMS, CAMS interim, TCR-1, and TCR-2, Geosci. Model Dev., 13, 1513–1544, https://doi.org/10.5194/gmd-13-1513-2020, 2020.

Ukhov,S. Mostamandi,N. Krotkov,J. Flemming,A. da Silva,C. Li,V. Fioletov,C. McLinden,A. Anisimov,Y. M. Alshehri,G. Stenchikov: Study of SO2 Pollution in the Middle East Using MERRA-2, CAMS Data Assimilation Products, and High-Resolution WRF-Chem Simulations

Arfan Ali, Muhammad Bilal, Yu Wang, Janet E. Nichol, Alaa Mhawish, Zhongfeng Qiu, Gerrit de Leeuw, Yuanzhi Zhang, Yating Zhan, Kuo Liao, Mansour Almazroui, Ramzah Dambul, Shamsuddin Shahid, M. Nazrul Islam, Accuracy assessment of CAMS and MERRA-2 reanalysis PM2.5 and PM10 concentrations over China, Atmospheric Environment, Volume 288, 2022, 119297, ISSN 1352-2310, https://doi.org/10.1016/j.atmosenv.2022.119297.

L 75: add references (links or DOI) of the observational data sets

L 93: Atmospheric composition fields are only available in grid-point space. 

L 101: SOA is included in the two OM aerosols variables in CAMSRA.

L120: Please mention here (and also for CAMSRA) the vertical extent of the model level at the surface, which may be important for the interpretation of the results.

L125: Please discuss the formulae applied to derive PM10/2.5 for CAMS RA and compare it to the approach used for MERRA-2

L152: To consider “urban” stations as “background” stations seems far-fetched. Excluding the urban regime, at least as option seems necessary for the study.  (see my general comments)

L 153 Provide reference for the classification 

L 162 Provide information about the typical grid-box variability, i.e. the deviation (variance) of the observations from the mean.  Please provide information how many of the “observation grid boxes” are actual composed of multi-station means.  For example, a pdf of the number of stations per grid box would be informative.

L175 Please, clarify if PCC represents a spatial or a temporal correlation.

L 180 Please, clarify the accumulation index I, i.e. is it over time per station, or over stations per time instance)

Fig 4 Why is there no PCC for Tr of Merra-2

L 420 Please, add some information about speciation for PM10 & PM 2.5

The authors present a thorough evaluation of two global reanalysis products with focus on air quality aspects over Europe, in terms of quality of the climatology, and its trends.

For this they use observational data from the EEA AIRBASE and AQ eReporting databases, for which a thorough filtering and quality assurance has been setup. Based on this they find that the CAMS Reanalysis overall outperforms MERRA, although significant biases remain, for which a further bias-correction methods should be developed to make the data most useful.

This manuscript is well written, and well fitted for publication in GMD, after consideration of a few comments:

• Table 1. The study by Huijnen et al. (GMD 2020) is missing in this list, even though they report to some extent on the performance of surface ozone in the CAMSRA, even with focus over Europe. Is there any reason why it is not needed to include it here, or is it an oversight?
• l. 85 Please include a reference to the aerosol scheme used in the CAMS Reanalysis (Morcrette et al., I believe)
• l. 90: “Meteorological observations and fields are taken from ERA5” while this is probably very close, please note that the CAMS Reanalysis applies its own assimilation of meteorological variables, i.e. it is not identical to ERA5. 
• Being to some extent involved in the generation of the CAMS Reanalysis, I have some small comments to better refer to the configuration specifications as described in Sec. 2.1.1, see also below
• Appendix B: This is useful information, which could be taken over by others in the community. It might be useful if you can additionally better specify flags 110-111: How are these ‘scientifically feasible values’ defined exactly?
• Line 153: “urban”: While the global reanalyses are known to have difficulties to represent urban-type conditions, would it make a difference to exclude the “urban” sector from your evaluation? I wonder if particularly for NO2 and O3 (but possibly also for PM) this could still result in significantly different, and more relevant, performance statistics? (although I can see that for your future applications you exactly require knowledge about the bias correction wrt urban stations, probably).
• Along these lines, it could be useful if the authors include (if not now, possibly in future) a correction factor for interference in the in-situ observations of NO2 for PAN and HNO3, see also a discussion of this impact in Poraicu et al. (GMDD 2022), and references therein.
• Figure 1. There is a considerable change in the observational network over time. Do you have any indications to what extent this has impact on the computed trends? Furthermore, a suspicious peak in the number of observations appear around the end of 2012 for various compounds. Do you have any understanding what has caused this?
• Sec. 3.1 the authors now focus on monthly mean values. It would be nice and interesting to try also other metrics (e.g. biases, and  trends in summertime, daytime ozone), particularly as this is more relevant for health-related applications. But I can understand if that is out of scope of the current work.

Technical comments

• The first couple of sentences of the abstract are of course true; on the other hand to my taste this is possibly a bit over-the-top, and high-level to motivate the work that you present here, that is not needed. It would work if you are a bit more modest here.
• l6.: “…they do not integrate surface measurements…” :I’d suggest to include the word ‘generally’: it is true that current, specified global reanalyses do not include surface obs, but it is not a rule, I’d say.
• l.71-72: I understand it’s easier to simply talk about concentrations, but when reporting O3 in units ppbv this is really ‘volume mixing ratio’ - some further clarification at this point might be good 
• l. 94: “The bias…” I think it is more accurate to write something like: “The biases present in the different AC satellite retrieval datasets..”
• l 97. “separate chemical compounds” better write “separate aerosol compounds”?
• Table 2, “Assimilation system: IFS Cycle 42r1 4D-Var ; Meteorology: ERA5” better write here: “Assimilation system: 4D-Var ; Meteorology: IFS Cycle 42r1”

References

Poraicu, C., Müller, J.-F., Stavrakou, T., Fonteyn, D., Tack, F., Deutsch, F., Laffineur, Q., Van Malderen, R., and Veldeman, N.: Cross-evaluating WRF-Chem v4.1.2, TROPOMI, APEX and in situ NO2 measurements over Antwerp, Belgium, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2022-882, 2022.

Huijnen, V., Miyazaki, K., Flemming, J., Inness, A., Sekiya, T., and Schultz, M. G.: An intercomparison of tropospheric ozone reanalysis products from CAMS, CAMS interim, TCR-1, and TCR-2, Geosci. Model Dev., 13, 1513–1544, https://doi.org/10.5194/gmd-13-1513-2020, 2020.

This paper evaluates and compares atmospheric composition reanalyses (CAMSRA and MERRA-2) over Europe. Their approach and results are straightforward and seems appropriate to the scope of GMD journal, though I don’t see novelty and originality. However, I think the results of this paper will serve as a good reference for long-term atmospheric composition data over Europe. 

I have a few questions and suggestions.

1) I wonder how their surface observation data (AIRBASE and AQ_eReporting from EEA) are different from observational data that were used in other previous studies. That is, did the authors use the same observation data, or are there any differences in surface observation data? For example, they said that Wagner et al. (2021) used EMEP observations. More detailed and specific description of how their data are different or the same would be helpful for readers.

2) It would be good and interesting to compare the performance of the two reanalyses for different settings, e.g., urban or highly polluted regions vs. remote regions, and so on. 

3) The authors provided spatial maps of pollutants from CAMSRA and MERRA-2. How about showing the same thing (spatial map) for the surface observations? I understand that they gridded the surface observation data, and there might be missing grids because of a lack of observation data. But, spatial maps from observations would provide an intuitive insight into the atmospheric composition over Europe.

4) Figure 5. I doubt if the authors correctly presented the results for SO2 in Fig. 5. In Fig. 5a, MERRA-2 SO2 concentration (in blue) is significantly higher than observed (black) and CAMSRA (green) SO2 during the study period. However, SO2 in Fig. 5c, which represents the spatial averaged SO2 from CAMSRA is apparently higher than that from MERRA-2 (Fig. 5d). I strongly suggest that the authors should double check if all the data and analysis are correct.

5) line 304. Why does MERRA-2 show a large increase in CO concentration in 2020?

6) Some figures (e.g., Fig. 1, Fig. 6, Fig. 7) do not show long-term trend lines for MERRA-2. 

7) line 35. Remove “to”

8) line 38. Add “and” before health studies

9) line 74. Data comes -> data come, or revise the first several words.

10) line 146. What is “BSC”?