We have made more than 1000 modifications to the manuscript in response to reviewer and editorial comments.  None of these change the main conclusions.  Most of them aim to improve readability or serve as typographical corrections; a small number deal with technical oversights, which we have corrected.  We thank all three reviewers for their time.  We have added a line to the Acknowledgments: “MJM also thanks the three reviewers, in particular John Miller, for insightful comments which improved the quality of the final manuscript.”

Throughout our responses, we refer to line numbers in the original manuscript with a lowercase “l”, while line numbers in the final submitted manuscript (without track changes, i.e., the clean version) begin with a capital “L”.

We have taken seriously comments by two reviewers on the length and balance between the different parts.  We summarize our changes in the following table (bold values indicate major changes):

We have heard and responded to reviewer concerns about length and balance between fossil and land fluxes.  We initially removed two figures and almost 100 lines of text (placing one in the Appendix).  Despite this, the length of the manuscript and the major sections were not very different, due to additional clarifications requested by reviewers.  We therefore merged six other figures together into three, moving one of the original six to the Appendix as it illustrated a point we refer to in the main text, and combined the sections with those figures (Cropland and Grassland became a single section, as did bottom-up and top-down comparisons to LULUCF NGHGIs).  This resulted in modest gains in the Forest land section, significant gains in the new combined Grassland/Cropland (25%, targeted for reduction due to the emissions being considerably lower than those on Forest land), and significant gains in the new combined LULUCF BU/TD (25%, while avoiding further reductions as this section is relatively more important to the overall picture).  The Results section is now more balanced between fossil and land fluxes (3.8:1 ratio of land to other, down from 6.1:1 in the initial version).  This is all shown in the attached table.

In addition to changes made to reviewer comments, we have made the following modifications.

The author “Fortems, A.” was changed to “Fortems-Cheiney, A.”.

The affiliation “UniSystems Company, Milan, Italy” has been added for authors M. Crippa and E. Solazzo.  The JRC affiliation has been retained for M. Crippa, but dropped for E. Solazzo.  The second affiliation “Universita degli Studi di Milano, Milano, Italy” has been added for M. Vizzarri.

Some minor typographical and formatting errors were corrected while responding to comments below.

We added some subsection numbers into the appendix, as the length made it difficult to quickly reference just by the name of the subsection.  We also harmonized cross-referencing and section/subsection title formats, using the styles "Heading 1", "Heading 2", "Heading 3", and "Heading 4", and avoiding all colored font.

A new version of the data has been uploaded to Zenodo (https://doi.org/10.5281/zenodo.8148461).  The primary change was the uncertainty of the NGHGI GL estimates, which was greatly reduced after an error was discovered.  None of the conclusions were altered.  We have replaced the old DOI address in the paper by this one.

Text in the appendix related to an inland water dataset was removed, as the dataset was neither updated nor used in this work (a related, but different dataset is described in the section on lateral fluxes).

In addition, five modifications were requested by ESSD upon initial manuscript submission.  We have included those changes here:

1. The citation (McGrath et al., 2023) was added to the DOI https://doi.org/10.5281/zenodo.8148461 in the abstract (note the update for the dataset version and year, as well)
2. The Grassi et al (2022b) reference has finished review and been published. We have updated the citation to reflect this, and updated the citation to Grassi et al. (2023) due to the change in year. No other works are listed “in review”.  We have removed two references listed as “in prep”.
3. We have mentioned that one of the co-authors is a member of the editorial board of ESSD under the headline "Competing interests".
4. We removed blue text in multiple places due to automatic hyperlinking and sectioning (changes not always tracked). 
5. “Sweden” was added to affiliation #15.

Note that figure and section numbers have been modified as a result of efforts to shorten the main text.

We fully agree on the importance of recognizing data providers when publishing complex scientific analysis.  We have already made a significant improvement in this direction compared to previous studies through extensive engagement with data providers whose products appear in all figures and analysis used in this work, including those from publicly-available model intercomparison projects like TRENDY and the GCP.  This has resulted in a manuscript with 65 co-authors, 49 pages of supplementary information (including more detailed model and data descriptions), and around 180 references.

In regards to the SOCAT product, we provided a reference of Bakker et al., 2016 (l2188).  No DOI exists for the version of SOCAT used in this work.  However, we have provided the DOI to the most recent version at the same line, and we urge data providers to include DOIs with all versions of their work released to the public to enable accurate and transparent citations.  The following has been added to the References (complete author list included in the paper, but omitted here):

“Bakker, D. C. E., …: Surface Ocean CO₂ Atlas Database Version 2022 (SOCATv2022) (NCEI Accession 0253659). Subset v2021. NOAA National Centers for Environmental Information. Dataset. https://doi.org/10.25921/1h9f-nb73. Accessed 01 July 2021.”

As noted by the commenter, some datasets are essential to atmospheric inversions, in particular the station-level observations of atmospheric carbon dioxide concentrations as a function of time.  For the EUROCOM simulations, Monteil et al. (2020) and Thompson et al. (2020), already cited in our manuscript, provide good traceability for the particular observational dataset used, with several paragraphs of text and acknowledgements.  We have added the citation of the atmospheric mole fraction with the following sentence at L2096: “The observational dataset used for the EUROCOM drought ensemble is accessible on the ICOS Carbon Portal (Drought 2018 Team and ICOS Atmosphere Thematic Centre, 2020).”  We have also added the reference in the References:

“Drought 2018 Team, ICOS Atmosphere Thematic Centre: Drought-2018 atmospheric CO₂ Mole Fraction product for 48 stations (96 sample heights), Integrated Carbon Observation System [data set], https://doi.org/10.18160/ERE9-9D85, 2020.”

We did something similar for the inversions used in the Global Carbon Budget, whose information is reported by Friedlingstein et al. (2022) in Table A4, adding explicit data references in our section GCP 2021.

L2080: “For details see Friedlingstein et al. (2022), in particular Table A4.  Atmospheric observations for most model systems are taken from Cox et al. (2021) and Di Sarra et al. (2021).  Note that one of the ensemble”

with the following references (full author list included in the manuscript):

“Cox, A… Multi-laboratory compilation of atmospheric carbon dioxide data for the period 1957–2019; obspack_CO2_1_GLOBALVIEWplus_v6.1_2021-03-01, NOAA Global Monitoring Laboratory [data set], https://doi.org/10.25925/20201204, 2021. 

“Di Sarra, A. G.,…Multi-laboratory compilation of atmospheric carbon dioxide data for the years 2020–2021; obspack_CO2_1_NRT_v6.1.1_2021-05-17, NOAA Global Monitoring Laboratory [data set], https://doi.org/10.25925/20210517, 2021.”

In addition, we have made explicit data usage for inversions performed for the VERIFY project and not reported elsewhere by incorporating text and references as possible into the Appendix on regional inversions:

L1923: “For the regional inversions, atmospheric observations of CO₂ were taken from multiple sources.  For CarboScopeRegional, atmospheric observations were taken from the ICOS 2021.1 ATC (ICOS RI, 2021) and the GlobalViewPlus 6.1 product (Schuldt et al., 2021a).  For the CIF-CHIMERE inversions, atmospheric observations of CO₂ for the period 2005-2020 were taken from the ICOS 2021.1 ATC (ICOS RI, 2021) and SNO_SIFA L2 (SNO-IFA, 2023) releases, along with data distributed through the GlobalViewPlus 6.1 product (Schuldt et al., 2021a).  For LUMIA inversions, atmospheric observations of CO₂ for the period 2006-2018 were taken from the dataset prepared for the 2018 drought task force initiative (Thompson et al., 2020). For the more recent years, data were used from the ICOS 2021.1 ATC release (ICOS RI, 2021), along with data distributed through the GlobalViewPlus 7.0 product (Schuldt et al., 2021b), and, for four sites, data distributed through the World Data Center for Greenhouse Gases. “

The situation for bottom-up models is slightly different.  We do not report results from machine-learning upscaled Fluxnet products (such as Fluxcom) in the current work.  While evaluation against eddy covariance sites (such as FLUXNET and ICOS) is likely the norm for dynamic global vegetation models used in the TRENDY ensemble, it’s unclear how many models rely on such results for parameterization of various processes, nor how important these processes are for the overall model behavior compared to those which use other data streams.  In other words, even if evaluations are run, it’s unclear to what extent models are changed in response to results.  As such, we prefer to trust original citations of the models to resolve these questions, and we defer to editorial policies of the journal for additional citation standards.

Thank you for the positive comments.

For the general comments, we take note of the length of the text and have made modifications to make it more accessible for new readers to the field. 

            1) We have added Table A1 in the Appendix summarizing terminology and acronyms.  We have included a reference to this table at several places in the text, also responding to a comment from John Miller.

            2) l175-188 (the final paragraph in the Introduction) provide a general overview of the focus of the paper.  We have added a new paragraph following this one which contains a brief description of targeted users, potential data usage/application, and a list of key take-home messages in order to help the reader narrow down sections relevant to their interests.

L201: “The dataset assembled in this paper (McGrath et al., 2023) provides annual values of carbon dioxide emissions and sinks in fossil and LULUCF sectors for the EU27+UK across a range of data products based on different methodologies.  This enables, for example, researchers producing datasets based on new methods a source of evaluation in the form of a best-estimate range of values.  Decision makers may also find the results useful for targeting mitigation efforts in the EU27+UK by providing a more complete subsectorial breakdown.  While NGHGIs already provide detailed data-based disaggregation based on activities, the dataset here adds additional constraints from independent data and models used outside of the inventory community. In addition, this paper outlines a methodology by which users of country-level CO₂ emission data can compare datasets against NGHGIs and identify where agreement occurs for the right (and wrong) reasons.  Section 3.1 highlights the extreme difference between current fossil emissions and uptake by the land surface.  Section 3.2 looks at an ensemble of bottom-up estimates of fossil CO₂ emissions, in addition to a preliminary inversion using atmospheric NO₂ observations as a constraint.  Sections 3.3.2 and 3.3.3 show that better agreement between the NGHGI and other models occurs when the models are driven strongly be category-specific data in forestry, grasslands, and croplands, as opposed to more generalized models created to couple to atmospheric models in global climate projections.  Section 3.3.4 highlights the challenges currently facing comparison of atmospheric inversion models with NGHGIs, while simultaneously showing improvement by accounting for net emissions for lateral transfer of carbon between countries.  Section 3.4 provides more discussion around uncertainties in both top-down and bottom-up estimates.”

            3) In an effort to both reduce the length of the results section and to balance discussion around CO₂ fossil and LULUCF emissions, we have removed text from Section 3.3 (LULUCF results), in particular a large section of 3.3.4 which is summarized elsewhere in the literature, and Section 3.3.2 (in response as well to concerns over the understandability of Figure 3). We have kept Figure 3 in the Appendix in case it is useful for readers familiar with the previous paper, and we reference it once in the main text.  This eliminates almost 100 lines and one figure from the LULUCF Results section.  Comments from John Miller led to additional shortening, which served to further balance presentation of CO₂ fossil and LULUCF fluxes.

For the specific comments,

l148-163: Modifications made in response to the general comments above have reduced the number of detailed comparisons in the LULUCF results, and consequently address the comment here. 

l390-395: We have added references to specific figures, taking into account additional text in response to John Miller:

L481: “An overview of all CO₂ fossil and land datasets in this work (Fig. 1) leads to a series of conclusions: 1) Regardless of the method used (NGHGI, bottom-up models, top-down models), the timeseries of annual fluxes from fossil CO₂ emissions rest almost one order of magnitude higher than removals from CO₂ uptake/removal by the land surface and well outside uncertainty estimates (Figs. 1a-c); 2) Uncertainties are much larger in the LULUCF estimates than in the fossil CO₂ estimates, regardless if one represents uncertainty by internal random error (i.e., the NGHGI totals in Fig. 1a, and the sub-sector LULUCF fluxes in Fig. 1d) or ensemble spread (i.e., bottom-up models in Fig. 1b, and the sub-sector LULUCF fluxes in Fig. 1e); 3) Interannual variability (IAV) is much more present in non-NGHGI LULUCF datasets (colored lines in Figs. 1b,c,e) than in NGHGI LULUCF datasets (Figs. 1a,d) or any of the fossil datasets (black lines in all subplots)."

l413-415: The caption text reads, “Panels (d) and (e) include a breakdown of the LULUCF flux”.  Top-down fluxes are not included in panels (d) and (e).  For clarity, we have changed this text to explicitly mention bottom-up fluxes: “include a breakdown of the [bottom-up] LULUCF flux”.  In addition, the “CO2” in l414 has been changed so that the “2” is a subscript.

l484-1024:  We have modified the beginning of Section 3.3 (l485-487) to help interpretation of the differences between the results in this work and those from Petrescu et al. (2021b).  In particular, large changes can be seen when significant processes are added to a model (e.g., with ORCHIDEE), as this case is no longer simply a result of adding additional years of forcing data and re-running the previous model.  We have also added this information to Table 2.  Note that the actual graphs showing the "old" results have been removed, but we still think this text is relevant.

L594: “The following graphs occasionally show large differences compared to previously-reported values.  This may happen when the model has undergone substantial changes since the work of Petrescu et al. (2021b), such as the case with ORCHIDEE and the addition of a dynamic nitrogen cycle coupled to the carbon cycle.  Such cases are both identified in the text as appropriate and as well as in Table 2.”

l552: In response to the concerns over length; the LULUCF focus of the manuscript; and the understandability of Figure 3, we have removed Figure 3 and Section 3.3.2 as it was not essential and focused only on NGHGIs without connecting solidly to the other datasets.  We have placed the figure in the Appendix with a reference from the text in the section on total bottom-up and top-down LULUCF estimates.

l1025: We have made this an independent section (3.4).

l1169: We have added the caveat in the conclusion suggested by the referee, taking into account a comment by John Miller:

L1086: “Fossil CO2 emissions are more straightforward to estimate than ecosystem fluxes due to extensive data collection around fuel production and trade, assuming that fuel statistics and accurate emission factors are available.”

For the technical corrections, we have: 1) corrected the font of “the notations” at l307; 2) put in bold “Forest land” at l591 to be consistent with following subsections (and removed the bold at l592, also for consistency); 3) removed the "old" results, and thus the faded panels, from all plots in response to this and a comment from John Miller; 4) doubled-checked the formatting of all section titles to be consistent.

Note that Figure 11 was removed in response to comments by John Miller, and thus this reviewer’s comment about the alignment of the caption is moot.

We are happy that the reviewer finds our work to be “hugely impressive”, and that he has taken the time to go through the main text and make significant comments.  It’s clear the reviewer has made an effort to understand the entire text, which helps improve the overall readability, accuracy, and consistency of our manuscript.  We detail our responses below.  We first respond to those comments requiring substantial amounts of work/reflection, which are open to interpretation, or to which we hold an alternative view.  For straightforward comments (33 out of around 170 comments), we have made the proposed changes and noted them at the end of our response.  Some of the comments in the text are covered by the general reviewer comments (1-5), and in those cases we group them together.

We apologize for any comments we may have missed.

Due to the length of this comment (32 pages), we have attached the entirety of the comment as a .pdf, in order to keep the formatting.  We feel this helps clarity.