This study represents a step-forward based on previous works by Grassi et al. and the 2021 Global Carbon Budget update. It extends the reconciliation between NGHGIs and bookkeeping models on carbon fluxes over ‘managed land’ to cover different categories of land use (and land use change) and to cover different nations and regions. It further establishes a framework that allows such reconciliation in the future bearing in mind both methods could evolve. Then future directions for improving the method consistency and confidence in quantifying national achievements in managed land carbon sink are provided.

I appreciate having the opportunity to review this work. This can definitely clarify confusions for those who are not so familiar with this field. I suggest its publication after addressing my comments. Most of my comments are to help enhance clarity. Some comments are rather a little philosophical and the authors can consider them as suggestive to enhance their discussions if they find it useful, but not that they must be addressed in a hard way.

General comments:

• In the methods section: need to explain that DGVMs in S2 simulations did not explicitly simulate secondary forest, but their simulated sinks driven by environmental changes over the domain of secondary forests from another map was used, to avoid confusions. The underlying assumption is that environmental responses of carbon sink over primary and secondary forests are the same, which is not necessarily true (Lines 366-367 describes something relevant but the results there is just a coincidence I believe. DGVMs do not simulate secondary forest so any difference, if found, can only emerge from segregation of spatial grids. I suggest removing results in lines 366-367 because it makes no real sense).
• The use of Potapov et al. (2017) to approximate unmanaged forest needs to be justified. The fact that it has been used in previous studies (Deng et al., 2022) does not justify its appropriateness. The most important point is, under the very loose definition of ‘managed land’ in the current IPCC guidelines, the domain of managed land is almost completely up to the nations and completely loses the objectivity with almost no relation to the actual state of the land whether it’s managed or not.  Hence the comparisons in the first two columns of Supplement Table 1 can potentially be out of pure coincidence. Note that the nations have incentives to expand the areas of managed land, whatever their real status, if they have confidence that the concerned pieces of land will be a carbon sink for a reasonably long time in the future (hence the consistency in reporting managed land is a key in NGHGIs). I don't see the advantage of using Potapov et al. compared to an alternative approach where one just uses the fraction of managed vs. unmanaged land to simply distribute the total simulated indirect carbon sink to managed land. Second, there is the conceptual inconsistency between primary forest and the IFLs in Potapov et al. as the authors explained in their paper, and the limit of the 500 km2 minimum size, but this is rather a minor point.
• The separation between forest carbon sink and deforestation fluxes is nice but also comes with uncertainty. The key is fluxes relate to gross deforestation might have not been reported by nations and the area undergoing gross deforestation depend on the spatial resolution of the land cover data used. The discussions and associated uncertainties in this respect need to be strengthened.
• I also noted the lively discussions between Malte Meinshausen and Sandro Federici. While there is perhaps no need to provide the corrections due to indirect effects because readers who understand this paper can easily obtain them by using the supplementary Table 1. But I suggest authors enhance the discussions regarding the potential leakages in current IPCC guidelines. That is, when the pervasive indirect effects of increasing CO2 show as a sink-enhancing term over actually unmanaged land, the nations have incentives to claim these lands as managed lands and the associated carbon sinks as the national contributions to mitigate climate change. But when the same indirect effects show as a carbon source term in forms of growing wildfires or large-scale forest dieback, the nations will have incentives to say these are natural disturbances and are not national liabilities.

Below are some minor comments:

Lines 86-88: we know that these differences reside in their respective concepts and are unlikely resolved by ‘new observations and platforms.’ Please rephrase.

Line 933-934: “direct human-induced effects” should be “direct land use change effects”, land-use change includes shifting cultivation, harvest and regrowth due to abandonment or afforestation/reforestation. Need to rephrase here.

Figure 1: strictly speaking there are only unmanaged lands if trendy S2 simulations are meant here. Need to also state in the methods that environmental effects on managed land are treated equal as those on unmanaged lands, which is a critical assumption in the paper but not necessarily true.

Line 112: “global models”: do you mean bookkeeping models? As trendy S3 is not used throughout the whole study, it’s better to be specific.

Line 125: this gives a sense that DGVMs simulate secondary forest. But in Grassi et al. (2018), only the spatial map of secondary forests (from LUH2?) were used and applied on DGVM simulated carbon sinks of intact land, no? Please correct to provide a more accurate description.

Line 128: “proxy for managed” => should be proxy for environmental effects on management forest.

Line 178: “natural land cover changes (Sitch et al., 2015)”, I guess it should be “natural vegetation dynamics”. The authors should mean land cover changes that are driven by environmental changes rather than human-induced land use.

Line 179: ‘anthropogenic fluxes’ and line 192 ‘anthropogenic CO2 emissions’. These should be both ‘land use change emissions’, not to be confused with ‘anthropogenic CO2 emissions’ in lines 93 and 95.

Line 191: what do you mean by ‘forest thinning from below’?

Line 283-286: Note that simulated natural carbon sinks over non-forest land account for ~ 1/3 by DGVMs (Table A8, 2022, ESSD; also shown in Fig. 10 in this paper). According to the approach used here, I guess some of this non-forest natural sink is ascribed to secondary forests. This might need clarification. This points needs to be considered also relevant with lines 310-315.

Lines 374-377: the discussions here comparing the adjustments from a single model and from an ensemble of models seem not having a lot of insights for me. The agreement might be just out of coincidence. I suggest removing it.

Lines 384-391: The authors can just explain that the simulated natural sink per forest area was used from DGVMs so that the effects of hypothetical forest areas at preindustrial conditions were filter out, which would be easier to understand.

Line 460-461: I guess there is also the difference in gross vs. net land use change between those accounted for in BMs and by GHGIs, not limited to shifting cultivation.

Line 566: “NGHGIs typically report estimates of gross deforestation” => Are there evidences supporting this? At least this seems not the case for China and India because they do not show any deforestation flux. Reporting gross vs. net changes in forest area in the category of deforestation could be an important source of uncertainty, because gross forest change typically depends on the spatial resolution of the underlying land cover data that are used to derive land cover change. (https://www.nature.com/articles/s41467-021-22702-2)

Line 559-560: Could it be an option for the land use change segregations used in BMs to move close to IPCC guidelines? A greater disaggregation in BMs will for sure better, but does not necessarily lead to better comparison with GHGIs and allow avoiding the cross-walking like in Table 1. The large discrepancy in the category of ‘other lands’ (Fig. 9) probably points to potential underlying mismatch in land use change category between these two approaches.

Line 586-587: with climate warming as a directional change, the assumption that climate-driven changes or variations in disturbance will cancel to a net-zero effect over time will unlikely hold. This comes back to the discussions between Malte Meinshausen and Sandro Federici (cf. the open discussion online: https://essd.copernicus.org/preprints/essd-2022-245/). Both being as directional changes, carbon sink enhanced by CO2 concentration increase are accounted as national contributions to carbon sink, but carbon sources enhanced by climate change (e.g. wildfires) were left out.

Line 590-592: completely agree with this.

The paper provides an interesting advancement of the previous analysis of the same authors on the mismatch between national inventories and global models used for estimating the emissions of the land sector. The paper is well structured and accurate in the analysis and clarifies many of the possible reasons of mismatch between statistics and estimations, providing also perspectives for the reconciliation of these differences. Although the comparison is very detailed and accurate, there is no information on the different pools that are considered in the different sources considered (NGHGI, BM and DGVM) and whether this can be a source of mismatch between estimations. This should be clarified in the method section as it is a potential source of discrepancy (potentially also adding this information in Table 1).

Overall the assessment is very positive, I therefore recommend the publication with minor revisions.

Specific comments:

-Lines 94- 95 Missing quotation to the sentence “BMs estimate that land use is a net source of CO2 globally, mainly due to deforestation, equal to around 12% of total global 95 anthropogenic CO2 emissions.”

Line 108 Suggest to add the reference to Figure 1 at the end of the para

Line 112 “The main conceptual difference is that global models consider those forests as managed that were subject to recent harvest and have not yet regrown to pre-harvest stock level….” With global models are you referring to both DGVM and BM? Otherwise please specify.

Lines 322-25:  The sentence is not clear, I suggest to reformulate it.

Line 374-377: This paragraph refers to an “adjustment” but I think it is more correct to refer to a “difference” or “gap” as the number represent a difference between estimations, while the adjustment is done when you want to make one of the two comparable to the other (e.g. line 378)

468 – missing space between “thatBMs”

595 – not clear why a the inclusions to natural terrestrial sink would lead to a “double-counting” perhaps to a wrong attribution of the natural fluxes to anthropogenic causes?

Table 1: add a reference to the pools included in the different sources. On Organic soils: Oscar uses the same dataset as BLUE, I guess it is not the case of H&N. I suggest to include the source of the datasets for the three BMs. In the row LULUCF net, last three cells (related to BMs): instead of repeating the same information, the information could be included once, merging the three cells (also in other cases above where the same approach is used by different models)

Review comments: Mapping land-use fluxes for 2001-2020 from global models to national inventories

I have been following this work for some years, and I am glad to see an article which describes the methods in more detail. Though, more detail leads to more questions! Overall, I think this is a well written article that makes the necessary data descriptions (with some potential modifications), and I do not see any major barriers to publication. The detailed method description does make me reflect on various historical assumptions (not made by the authors, but further locked in by the authors). In a sense, the authors mechanically go through a method to bridge the gap between different definitions, ok for ESSD I guess, but I think there could be a little more reflection on the appropriateness of past decisions and therefore potential pathways forward to avoid the ‘gap’ (the authors do some of this already in the discussion).

In a sense, the paper provides an ad-hoc “fix” to a problem that sort of should not exist. Of course, there may be good reasons to have different definitions, and therefore a method to bridge those definitions. Though, at some stage, one may start to question the definitions and whether they are appropriate for the times. Science evolves, as does policy. In both cases, BM and UNFCCC, are wedded to decisions made in the past. The authors lock this in with comments like “unlikely countries will change”. We tend to hold on to those decisions as hard as we can, even if the justification is rather weak! Key questions that come to mind: Is it appropriate to use BMs in the carbon budget? How can the Loss of Additional Sink Capacity not be included in the carbon budget? Why can countries continue to have such inconsistent definitions of managed land? Is managed land still a relevant proxy for anthropogenic? If BMs included indirect effects (such as in DGVMs) and UNFCCC had data based definitions of ‘managed’ land, the gap may diminishe significantly? Ok, I don’t expect the authors to solve these issues, but their method and dataset puts them in a unique position to comment on these issues. The UNFCCC is based on IPCC guidelines, and IPCC guidelines are also informed by IPCC ARs, so if we have better science, or can do things differently and better, that should be communicated, not just assume BMs or UNFCCC will never evolve. 

A major comment on the method (description). I do not think the method has to be changed, but I think we need a full description of TRENDY. Is there a TRENDY paper that describes the different runs, consequences, etc? We need a list of S0, S1, S2, S3, etc, and a description of what they mean. A figure may be nice. It seems to be so many issues arise because of the way the carbon budget has defined things. What is the land sink in terms of assumptions (S2)? How do the BMs match to the Sx conventions? What are the DGVMs not used for net LUC? How does the LASC fit in? And what does all this mean for the ‘gap’? Ok, I can imagine a response from the authors would be that this should appear elsewhere, it should be in the global carbon budget paper, or in a TRENDY specific paper, etc. But, unless one has a 10-year experience with TRENDY and its protocol, understanding some choices made in the paper, and the consequences thereof, is difficult.

Another major comment, which I did not notice to the end, is that the data is just a table? I was expecting something more comprehensive then this. Basically, the paper is providing a summary of key results, not the data used in the paper and for the analysis?

I have some more specific comments, building on my points above, in order they appear in the article:

1. General, lines 67+. What is the dataset? It is a dataset which reconciles the different between two datasets? Or is it an additional disaggregation of the TRENDY dataset? Perhaps something like this is needed: “Here we provide additional disaggregation to existing models runs from DGVMs to allow a reconciliation between…”. I have not looked at the actual dataset, but I am sort of curious of what it actually is. Is it a better version of the TRENDY dataset?
2. General, dataset: Ok, I have now looked at the dataset, and it is Table S1? Is that it? I was expecting country level estimates at the same level of detail as in the figures. I can’t really do much with this dataset, it is only a table with summary statistics?
3. Line 181+: Sland is S2? This assumes the land areas in PI? If the land areas were allowed to evolve over time, as they did in reality, this leads to the LASC but also includes LUC? But the LUC estimate, based on BMs, does not include any indirect effects. The total land sink is S2+BM+LASC? Alternatively, this is S3? LUC according to DGVMs would be S3-S2? To my general comment above, it really needs a figure to explain this, and put a magnitude on some of these effects. According to Friedlingstein et al 2022, “The resulting loss of additional sink capacity amounts to 0.9GtC/yr”, like this is ~10% of the total emissions? This is not trivial! DGVMs have the ability to consider evolving land areas and indirect effects on LUC, but those results are ignored? In the context of comparing to NGHGIs, this seems puzzling! DGVMs are uniquely positioned to bridge the gaps, and this is what the authors do, and so what are the implications if BMs are not the starting point and instead DGVMs are?
4. Line 181+: Building on the previous comment. What is the justification for starting with the BMs? One could start from the LUC estimates from DGVMs? That would be more consistent? It would be useful to explain this more, and potentially show how the BMs differ to the DGVMs for LUC? I presume through the Sx runs, it is possible to make a self-consistent definition of the total net land sink, and then disaggregate into LUC and land sink components that are most useful. If the DGVMs are used for LUC, does the gap between UNFCCC and DGVM LUC differ as much as if BMs were used?
5. Line 205: This text sounds like the country-based data are the national submissions? Perhaps “We used the most up-to-date and complete compilation of country-level LULUCF estimates (Grassi et al), … This dataset builds on … [and mention submissions]”. Since “submissions” is used before Grassi et al, it reads as though Grassi et al is the UNFCCC country emissions. Some rewording will avoid confusion.
6. Line 275: The “but exclude land-use change”. Technically, the process is included in the DGVMs, but the area that underwent LUC is excluded (basically, help at PI). Something like “but the land area is held at PI values as a proxy to exclude LUC”? If correct?
7. Line 275+: The steps “results were first…and then” could be better explained. The “first” step, with Hansen et al, is to ensure consistent definitions of forests? The “then” or “second” step, is to mask managed land by assuming intact forests? The main point is to explain why you do the “first step”, as you don’t say why (I don’t think). There is also an issue of how the PI area maps to the current intact area? Are these all subsets (PI > Han > Intact)?
8. Line 280+. Somewhere, perhaps in a separate paragraph, can you define what “intact” forests are, and why they are a good proxy for “managed land”. Not obvious, even if the statistics look rather good in the figures.
9. Line 316: Just clarifying. The managed land mask is frozen at 2013 values, it has no trend? Any thoughts on the potential implications? The 2% is based on the total area, since LULUCF is a change, a 2% change to forest area is rather significant? Basically, it is the changes that are relevant? And 2% is a big change?
10. Line 321: The LASC sits on the LUC side? Or you saying that the intact mask has an area less than the PI forest area, so any LASC are masked out? I don’t really understand this paragraph, but it is obviously worth mentioning. It probably needs an explanation for people that are not experts on TRENDY protocols and LASC!
11. Line 340: What is deforestation here? Is this the process of forest to non-forest, or is this a positive (gross) flux defined in BMs?
12. Line 343+: It is sticks out that H&N is quite different to the others. Can you provide some more explanation on this?
13. Line 383+: Again, this LASC seems to be something the reader has to have a good handle on to interpret some of these results. You say it is taken care of, which I can trust, but in the methods I think the reader has to have a much better of understanding of LASC, etc, to be able to pass the comments here.
14. Line 400+: Why is the method worse in Canada, China, India, etc. You touch on this later, but perhaps just add some pointers that you dig into these differences later. As a reader, I am interested in in why things differ, not just stating they differ.
15. Line 534: After reading the size of the LASC, can you be more specific on the “is likely an overestimate”. The LASC seems to be not insignificant! This is a constant theme throughout the paper. LASC seems like a rather significant issue, but it is randomly brought up (it is as if a diligent author noticed “oh, don’t forget LASC, we should mention that here”). Does the article need a more systematic discussion of LASC?
16. Line 550+: A “short-term and pragmatic fix”. The discussion is really how to marginally modify the status quo, what can BMs tweak and what can UNFCCC tweak? I sort of see more fundamental questions rising. Are BMs, which ignore indirect effects, too outdated now? They were good when Houghton first did it, but hey, we can do better now? Line 558 mentions are fairly fundamental issue of land management and demographic models, but then Line 559 says “a greater disaggregation” of BMs. What is the point of disaggregating BMs if demographic models are needed? One could even think in terms of observational constraints. UNFCCC inventories built on forest inventories is close to an observable, but BMs don’t include indirect effects? DGVMs may be more similar to a UNFCCC inventory in terms of processes included (indirect effects), give or take the area issue and annual variability, so could DGVMs be better overall for the LUC (vs BMs)? Or are DGVMs too uncertain? Basically, could a “short-term and pragmatic fix” be to drop the BMs? (sorry to those running BMs). BMs seems like a simple climate model that ignores the carbon cycle? BMs are not really inputs into climate models, that is more land transitions? To rephase your comment, “what is the long-term and non-pragmatic fix”?
17. Line 574+: A good point to bring out is how much of the gap is due to differences in areas and how much is due to indirect effects? Quantifying that would be a very valuable exercise, and gets closer to original notion of separating anthropogenic effects? Historically, the UNFCCC approach of managed land is a proxy for anthropogenic. Do we have better science now that we can use a different proxy? Does the data in the paper give a method to better define anthropogenic?
18. Line 618: “it is unlikely that countries will change”. Bleh. What time frame, short to medium term? Countries routinely make changes, revisions to LULUCF can be rather significant too. I disagree with this starting point. If we (the scientists) but forward good reasons to change, in IPCC reports, maybe they will change. That doesn’t mean they change to BMs or DGVMs, but they change. The managed land issue was a poor proxy for anthropogenic, with many issues. If the science has evolved to do better, just say that, and don’t prejudge whether countries are rigid or not. Same goes for the science community too… The budget once had H&N, then included BLUE, and now OSCAR also. It used to have a residual sink, now it has DGVMs. We have to think in terms of evolution, not holding onto old ideas.
19. Line 651: I don’t like this sentence. It implies that the science community has done it wrong, and when corrected, land is a net sink overall. One could also write: “When countries include the natural land sink together with source from LUC, they can report a sink”. What are you trying to say, perhaps: “When results from DGVMs and BMs are redefined to include an expanded definition of forests and indirect effects as a part of the anthropogenic sink, our analysis confirms that sink estimated in NGHGIs”. It is difficult to get the language right, but I think you want to avoid implicitly assuming NGHGIs are superior.
20. Conclusions: An important point to emphasis somewhere here is the distinction between climate modelling and inventories. The NGHGIs cannot be fed into a climate model (or integrated assessment model), without those models making potentially significant modifications. Sure, let countries get excited about the sink in their NGHGI, but this has to balanced with a smaller carbon budget, potentially significantly for some countries, and the ability to maintain a sink under climate change, may not be trivial. A major point here is the NGHGI approach is really a poisoned chalice (in my opinion), countries and some scientists have yet to realise that!
21. Figure 1: Can you make a similar figure, but for S0, S1, S2, S3, LASC, BMs, NGHGIs? Not sure if possible, but consider it a challenge!
22. Figure 3: I guess the average as a median would just be the middle model? In any case, the “gap” to reconcile is really a heavy function of the BM. If OSCAR was correct the gap would be much smaller, H&N much larger. What are these BMs doing differently?
23. Figure 5: I guess a is Sland, disaggregated? For b, at what stage do you filter out models. Is there reason to believe the model at -2 and constant? Is there a trend in Sland, and is that trend the same / different as the trend in non-intact? How does the LASC fit into this figure?
24. Figure 6: Why is H&N doing so well?
25. Figure 10: Potentially useful figure. It would be even better if the LASC could be included? That would be a nice visual of the size of this effect and its potential importance. In the same vein, perhaps including LUC from the DGVMs?