This manuscript provides a thorough accounting of the procedure, and changes in performance, of the CAMS NRT system when including TROPOMI observations of carbon monoxide (CO). The authors show that by including the TROPOMI CO observations the CAMS CO columns increase by on average 8%, which result in a much improved match compared to independent observations, as compared to a version of the CAMS system where only MOPITT and IASI observations are assimilated.

The authors furthermore assess more closely the impact of TROPOMI assimilation for fire plumes. While the total column can well be constrained, the TROPOMI data does not provide further constraints on individual plumes that are transported across continents, at altitudes above 500 hPa. This suggests (even though not written explicitly) that modeling aspects, such as accurate emission estimates and transport, remain essential to be able to capture such events.

The manuscript is well written, and provides a good description of the current state, and technical details, on the TROPOMI CO assimilation configuration in CAMS.

However I find it rather an accounting of the technical aspects, which would indeed be valuable as a technical report, but less so as a scientific contribution as could be expected in ACP. For that, a more critical assessment of various aspects would be expected. The authors should be more rigorous and selective in the figures that they wish to present, to make their points.

For example, the manuscript contains a long section on the monitoring of TROPOMI data in the CAMS system. While this is valuable, and well-understood from an operational perspective, the value for the reader is not clear, and does not justify a lengthy reporting. I suggest to condense, or even remove this section, or alternatively move this into supplementary material, also because reported performance statistics over the given time period are difficult to interpret/reproduce, and not generally interesting, given the changes in the CAMS system and observation version.

Also various figures do not contain any novel information, and could possibly be moved into supplementary material (or removed completely), e.g. Fig. 5, while Fig. 6 and Fig 10 contain also very similar information;  as well as Figure 12, 17, 19, 20, and redundancy in Figures 22-25).

Also, discrepancies, e.g. between TROPOMI and MOPITT and IASI,  as well as between TROPOMI observations and GFAS fire estimates are indicated, but the implications of these discrepancies  are so far essentially disregarded. Such implications are however very important for generating long, consistent time series of TCCO in future reanalyses, and for the accuracy assessment of GFAS fire CO emission estimates, e.g. in terms of its timing, and magnitude. I think exactly such important messages cab be drawn from this study, but appear lacking. 

So, In an updated version of this manuscript I challenge the authors to provide a more critical  of such aspects, and reduced reporting particularly on the monitoring aspects.

These elements should ideally also be reflected in an updated version of the abstract, which could well be shortened and made more to-the-point, and the conclusions section.

More detailed comments

pp5, line 19: You describe why the IASI data is bias-corrected, and not MOPITT. (for historical reasons). Given that IASI appears more consistent with TROPOMI, wouldn’t  this study be an argument to change this configuration, and apply bias correction to MOPITT instead?

pp5 line 30 and 34: 150 days were used for generating the BGE, both in the old and new configuration. Given the different lifetime, and specifics of CO for different seasons over the globe, it seems to make a difference for which period the 150 days were chosen (summer or winter). 

pp6, line 2: “constant in time, globally averaged (…) profile” This seems a gross simplification of the BGE information, given that the tropopause height varies strongly over the globe, and that CO emissions and lifetime features different uncertainties depending on season and region. Could you comment on this? To what extend is such a simplification applied here expected to contribute to overall errors in resulting CO analyses?

pp10-11: The authors provide a thorough, and useful accounting of the quality of the various TROPOMI CO data streams. My take on this is that the TROPOMI CO data, as used in the CAMS system, is actually biased high by a value of 6.5% (NDACC) to 9% (TCCON) (pp 11, line 9), and even larger positive biases over Antarctica. Is this correct? If so, how does this relate to percentual differences of monitoring experiments as reported later on in the manuscript (pp 13, line 14-15: departures of 9-11%). This also to provide further justification of the statement given on pp 17, line 1-2 (“this suggests that (…) CAMS is biased low and not that TROPOMI CO is biased high”) and pp 18, line 1-2

(“In ASSIM TROPOMI TCCO is used without bias correction”).  It seems that the definite justification for all this is only provided in sec. 4.2.3, when independent evaluation is introduced. 

Sec. 3: Please consider to condense this section, or move this to an appendix or so. The only relevant figure is maybe Fig. 7, except that it might be better to present the data for a more consistent model configuration and time period (e.g. only for the full year of 2021). Then also the changes to the CAMS model system (incl. Table 2) could be moved there.

pp20, line 2-3: “MOPITT departures are increased in ASSIM”. I think this is a very important finding that deserves more attention than currently given. Despite the arguments given earlier about the different sensitivities of MOPITT and TROPOMI, I find it worrying that the math with MOPITT decreases, given that I expect that MOPITT TIR retrievals have been thoroughly validated, and improved, over many years (what is their reported uncertainty statistics?). Does this study, and particularly Figure 11, imply that MOPITT TIR retrievals  of TCCO are biased low? What are implications for CAMS when developing a new reanalysis product? Your judgement as part of the discussion and conclusions section would be much appreciated.

pp 24, line 7: “leads to changes in surface ozone”: Please remove this statement if not further substantiated. Even though O3 is undoubtedly modified, It is in my view much more likely that the changes seen are a direct effect of CO assimilation.

Furthermore, when presenting evaluation of CO against surface observation networks, please provide information on the quality and representativity of these observations - this is to my knowledge not straight-forward, which makes interpretation of results in Figure 18 difficult, if not impossible.

Sec. 4.2.4 (Boreal Wildfires), particularly Figures 19-20:  I believe these figures, and their discussion, even though interesting as an assessment on its own, are out of scope wrt the topic of  this manuscript. When presenting these results from the GFAS system in a manuscript, I rather expect a scientific judgement, e.g. with respect to their quality, and/or impact in the CAMS system, particularly as TROPOMI CO data really enables such quantification. This connection is currently essentially missing, which therefore does not justify the figures shown here. To the least the authors should be more selective. 

pp 27, line 5: “These high values are better captured in ASSIM than CTRL”. What are the implications of this finding for the GFAS emission amounts reported above?

pp 29: “horizontal mismatch”. Considering that the modeled plume is present one day too late, could it possibly also be a mismatch in time of the GFAS emissions in this configuration?

pp 29, l16 “can be considered a success”: I quite do not agree with this statement. Of course TROPOMI CO is very successful in correcting the total column, much better than was the case in the CTRL configuration, but from the given evidence it appears simply not able to significantly improve the placement of the plumes in the right altitude, which was the topic of exactly this analysis. 

pp32, line 1 (46 Tg, 17 Tg, 9 Tg) - please consider to remove the statement on quantification of CO emissions here. The current evaluations actually suggest that these are under-estimates (but unclear by how much)

Conclusions section: Please provide an overview of open issues, and current limitations.

Technical corrections

pp1 line 28: “already … already” please rewrite removing one ‘already’.

pp3 line 16: “published” change to “reported”, remove “but not in a peer reviewed publication”.

line 18 “additional work”, please specify “work” here, ore rewrite this sentence. “Additional work” does not justify a manuscript per se. 

line 39: “CAMS-GLOB-ANT”

pp4, line 1 “CAMS-GLOB-BIO”

pp6, line 6/7: please provide a more generic reference, this is not really readable. Alternatively consider to move this technical information, which is not essential to the subject of TROPOMI CO assimilation, towards an appendix or so (see also generic comments above)

pp 7, line 8: IAGOS: Provide here the explanation of this acronym, and not in line 12.

line 15 “in this paper” -> “here”

line 28: “station”->”stations”

pp 9, Table 3: Check ordering of rows: one date stamp (2019-08-06) should come after 2019-07-03?

pp 13, line 15: “medium”->”median”

General Comments:

This manuscript presents the assimilation results of NRT CAMS by assimilating TROPOMI CO observations with the following contents: 1) assimilation result of CAMS by assimilating MOPITT and IASI CO; 2) difference between TROPOMI CO and CAMS; 3) difference between CAMS-TROPOMI with control run; 4) validation with respect to independent observations; 5) impact of boreal fires. The experiment results, shown in this manuscript, are helpful for better assimilations of tropospheric CO. I recommend the paper for publication after consideration of the points below.

In general comments, the detail of this manuscript should be improved. There are different experiment configurations with complicated names such as “CY47R3”; The titles of subsections need to be improved, for example, “4.2.1 Difference plots ASSIM minus CTRL experiments”; the fonts of some figures are too small to see; the titles of some figures seem from original experiment records and should be re-designed; the term of “departure” is widely used in figure captions and should be clearly defined: whether they represent model minus observation or observation minus model. While the assimilation result is worthy of publication, the style of the current version is a little different from common research articles.

Specific Comments:

1. Page 4, Line 20: What is the purpose of “randomly selecting one observation in each grid cell”? Is it better to use the average of observations within each grid cell to reduce the effect of random errors in satellite observations?
2. Page5, Lines 11-20: I understand there could be some “historical reasons”. However, the description here is uncommon in research articles.
3. Page11, Line29: The direct averaging of satellite data to the model grid may not be a strict "super-observation" because the effect of observation error at each observation point is not considered, for example, Section 2.3.3 in Miyazaki et al. 2012. In addition, I don’t think the "super-observation" can reduce random errors. Miyazaki, K., Eskes, H. J., and Sudo, K.: Global NOx emission estimates derived from an assimilation of OMI tropospheric NO2 columns, Atmos. Chem. Phys., 12, 2263–2288, https://doi.org/10.5194/acp-12-2263-2012, 2012.
4. Page 17, Lines 26-27: what is the difference between CTRL and CAMS discussed in the previous section?
5. As shown in Figure 11, the improvement by assimilating TROPOMI CO with respect to the original MOPITT+IASI is mainly due to the underestimated CO columns in MOPITT TIR data. However, why the newer multi-spectral TIR+NIR MOPITT data were not considered? In addition, the a priori biases are comparable between MOPITT (0.6) and IASI (1.5), then why the a posteriori bias is much larger in MOPITT (-3.3) than IASI (1.2)?
6. Page 25, Figure 18: It is interesting to see the large difference between Europe/North America and China. What could be the possible explanations? A recent study found a similar small influence of MOPITT CO assimilations on surface CO concentrations in China. In addition, what is the meaning of “Marco Polo” in the title of Figure 18c? Tang, Z., Chen, J., and Jiang, Z.: Discrepancy in assimilated atmospheric CO over East Asia in 2015–2020 by assimilating satellite and surface CO measurements, Atmos. Chem. Phys., 22, 7815–7826, https://doi.org/10.5194/acp-22-7815-2022, 2022.

Technical Comments:

1. Page 1, Line 22: the abbreviation of “NRT” should be defined.
2. Page 5, Line24: change “2021-07-09” to “2019-07-09”
3. Page 17, Line8: change “3 10 18” to “4.3 x10 18”
4. Page 21, Figure 12 and Figure 13: the definitions for the abbreviations of “FgDep” and “AnDep” should be provided.
5. Page 22, Figure 14: I assume the left and right columns are relative biases for the assimilated and control runs, respectively. But it should be clarified in the figure caption.

In their paper, Inness et al descibe the assimilation of TROPOMI CO in the world-leading CAMS analysis system. The combined assimilation of TROPOMI, MOPITT and IASI CO is described in detail, and major advances are reported by using TROPOMI. The paper is well written, is presenting important results, and is complemented by validation results with independent observations. I am much in favour of publishing these results. Several questions came up when reading the paper, and I would ask the authors to answer those before the paper is published.

IAGOS data is used prominently in the paper so it seems reasonable to invite an IAGOS team member as co-author, or at least mention the programme explicitly in the acknowledgement.

The paragraph starting with “The differences between the observations and the model equivalents are called departures" (page 3) is general and rather long to my taste. The sensitivity of the departures to detect small changes is an important point to make I agree.

"Initial work about the use of early TROPOMI CO data in the CAMS system was published in an ECMWF technical memorandum (Inness et al., 2019c) but not in a peer reviewed publication." 

The early paper of Borsdorff (a co-author) from 2018 already showed striking good agreement between early TROPOMI retrievals and the CAMS system. Reading this paper it is a bit surprising that "a more mature retrieval version" seems to be needed for the assimilation to be successful, as is suggested by the text of the introduction? The good initial comparisons are discussed on page 10.

"contains a lot of additional work," (page 3, line 18) If this is relevant, it should be specified what the additional work consists of.

Could you please comment why data with LAT < 60S is blacklisted?

Equation 2, 3: I would expect to see error terms here to reflect measurement and retrieval uncertainties, and forecast errors, especially since the "true" atmosphere is introduced here. The equations suggest that the true state can be quantified which is not the case. I would recommend that the error terms are explicitly added to these equations to avoid confusion. "d" is resulting from these errors.

"we remove the influence of the a-priori profile" (page 5, line 9) In general the a-priori does not only appear explicitly in these equations, but also implicitly through the dependence of the retrieval "y" and retrieval error on the a-priori, since a badly chosen a-priori will generally lead to larger retrieval errors and larger d values. 

"the thickness of the 1000-300 hPa layer," is a bias predictor (page 5, line 15) What is the physical relation between the bias and this quantity?

NEWBGE and OLDBGE. In many data assimilation applications the background error is scaled using chi-square statistics, which basically requires that the background errors are consistent with the computed "d" values and observation errors. Is something similar done in the CAMS system? Phrased differently: is there statistical evidence that the NEWBGE is more realistic than OLDBGE?

Also for equation 4 (page 10) I would suggest to explicitly add the error terms.

"with a mean difference between the data sets of 3.2±5.5 %" (page 10, line 21) Is CAMS higher or lower?

"9.22 ± 3.45% against standard TCCON XCO data and 2.45±3.38% against TCCON unscaled XCO "  (page 10, line 27) Please explain why there are two numbers and which one is reflecting the actual bias more.

From the abstract I got the (maybe wrong) impression (I realise it is not explicitly mentioned in this way) that the retrieval upgrade to version 02.02.00 was quite important for the success. From section 2.2, however, it is not so clear how much the retrieval versions 1 and 2 differ. The validation results are discussed, but the versions are not separately reported. Is this information available? Is there a clear indication of changes/improvements in version(s) 2 compared to retrievals before Summer 2021? It would be good to have some quantitative information of the differences before the comparisons with CAMS are presented. 

"It is not clear yet if this is the direct result of the assimilation of TROPOMI data leading to increased CO in the upper troposphere, or it could be the result of convective transport. " (page 18) I guess this could be easily studied by looking at the vertical profile of the analysis increments.

"illustrating that the assimilation of TROPOMI CO improves the fit of the CAMS analysis to the IASI-BC data globally." (page 20 top) A really nice result!

The results for West Africa in Fig 16 could be linked to panel b in Fig. 10, and provides evidence I guess that the increments around 200 hPa are a good result. Please add a comment in the text.

Figure 16 caption: IASI should be IAGOS

What is the relation between "Marco Polo" and China surface observations in Fig 18? Should this be CNEMC?

The results suggest that the low biases in CAMS are linked to a low bias in emissions. Are there future approaches developed to improve this situation, e.g. a bias correction applied to the emissions or emission inversions? Please comment at the end of the conclusions section.