This paper discusses a group of transient simulations of the early last deglaciation. It is a useful summary of the major features of the simulations. I have some minor comments on the paper.

The subtitle meltwater paradox is buried in the much more diverse discussions of other features of the simulations. Or it is only a very small part of the paper. Therefore, it should not be there.

The value of this paper is mainly recording these features as a reference. 

88 “….whilst more recent modelling studies show a deep and strong ocean circulation …..”

This part of discussion should include a more recent proposal that, more likely, AMOC strength at the LGM is not too different from that at PI, in spite of the robust shallowing structure (Gu et al., 2020; Zhu et al, 2021).

189: “….This creates a meltwater paradox, where the freshwater forcing required by models to produce recorded climate change is broadly in opposition to the meltwater history reconstructed from ice sheet and sea level records”.

It should be pointed out that part of this seemingly inconsistent result for BA onset/M1A may be reconciled, partly, if the M1A meltwater flux is mostly injected into the Southern Ocean, as implied by the reconstruction of sea level rise patterns.  

215--: The reason most models do not produce the abrupt BA onset under a smooth meltwater forcing is because the AMOC in those models are monostable. This stability, however, may be a bias, as discussed in previous works (e.g. Liu et al., 2014).

“4.1: timing of deglaciation” and Fig.2a-d: Greenland temperature. The smooth-looking Buizert curve at 19-18ka may be due to the muted d18O response by the expanding sea ice cover, instead of declining temperature (He et al., 2021)

626:” The AMOC for each of the HadCM3, MPI, and iLOVECLIM simulations is impacted by the chosen meltwater scenario during the deglaciation….”. For this section of discussion, it may be useful to diagnose the “effective” freshwater forcing on deep convection site, as discussed in He et al., 2020.

Fig.4, Fig.S4: Shakun et al 2012 is mostly SST, not surface air temperature!

Fig.8: caption. shouldn’t “Spatial correlation…” be “Spatial distribution of the temporal correlation …”?

744: “4.5 Meltwater paradox…”…

The paradox is an obvious problem. Here, there is little discussion, or even speculation on the nature of the problem. Some discussions will be useful.

More specifically, it concerns with 

• the AMOC response to climate forcing,

and

• the climate response to AMOC change.

A climate model can be good for 2, but not for 1, and vice versa. So, it depends on the purpose of the modeling work. If one is more interested in the response of global climate to AMOC change, the meltwater can be adjusted to make AMOC like reconstruction, and vice versa. One may speculate this partly related to the model bias of the seemingly over-stable AMOC in current CGCMs. But, this problem seems to be present in EMICs too. If the AMOCs in these EMICs are bistable AMOC, this paradox would involve additional factors.   

Gu, S., Z. Liu, D. W Oppo, J. Lynch-Stieglitz, A. Jahn; J. Zhang and L. Wu, 2020: Assessing the potential capability of reconstructing glacial Atlantic water masses and AMOC using multiple proxies in CESM. Earth & Planetary Sci. Lett. 541, 10.1016/j.epsl.2020.116294

He, C., Z. Liu, J. Zhu, J. Zhang, S. Gu, B. L. Otto-Bliesner, B. Esther,C. Zhu, Y. Jin and J. Sun, 2020: North Atlantic subsurface temperature response controlled by effective freshwater input in "Heinrich" events. Earth & Planetary Sci. Lett., 10.1016/j.epsl.2020.116247.

He, C., Z. Liu, B. L. Otto-Bliesner, E.C. Brady, C. Zhu, R. Tomas, C. Buizert and J. Severinghaus, 2021: Abrupt Heinrich Stadial 1 Cooling Missing in Greenland Oxygen Isotopes. Sci. Adv., 10.1126/sciadv.abh1007

Liu, W., Z. Liu and E. Brady, 2014: Why is AMOC mono-stable in coupled general circulation models?  J. Clim. 27, 2427-2443.

Zhu, C. Z. Liu, S. Zhang and L. Wu, 2021: Global oceanic overturning circulation forced by the competition between greenhouse gases and continental ice sheets during last deglacial. J. Clim., 10.1175/JCLI-D-21-0125.1

The paper presents a study of a multi-model climate ensemble of the early last deglaciation. The models differ in their forcings including the effect of greenhouse gasses and meltwater input. The models are stratified according to the implementation of the distribution of meltwater - as this seems to be the dominating factor: melt-uniform, melt-routed, TraCe-like,  and 'bespoke'.  The paper discusses how the surface temperature and the  AMOC develops in these models -- for example, the warming begins at high latitudes while it is delayed in the tropics.



I am not an expert in the field, but I find the paper interesting, although perhaps too long. In particular, I think that the Introduction could be  shortened (it is now 7 pages).



Specific comments:

Title: I had to look up 'reigns supreme' to get the exact meaning.  Also, reading the paper I don't find a lot of results about the paradox, with section 4.5 being rather inconclusive. Perhaps it would be better just to emphasize the large role of meltwater in the model results.



Figure 1: Panel c is not mentioned in the caption. Does the legend right of panel b also covers panel c? What is the red curve in panel  a? Perhaps you should not include references in the caption; it makes it hard to read.



l398: I guess that what is important here (Fig. 4) is if the model mean is significantly different from 0. The measure used in the paper - 70 % agreement in sign - depends on the number of models and corresponds here to 12 models of one sign and 5 of the opposite. It is not clear what the probability is that this will happen by chance.



l404:  The subsection should be 3.1. The same for other subsections in section 3.



l467: I don't understand how the year of significant warming (Fig. 5)  is determined. The description in the text should be improved. Are you, for each model, comparing 100 years centred about a t_0 with the 500 years in 20-19.5 ka BP, calculating the significance of this difference, and varying t_0 until p < 0.01?  If this is the case, are you assuming all 100 years are independent? If they are not independent, the degrees of freedom in the t-test should be smaller and the differences will be less significant.



Paragraph beginning at l520: I wonder if the comparison with the proxy data could be more detailed. It would be interesting also to see the comparison to individual models and not just the model mean.



l548: There seems to be something missing from the text here. Also, Fig. 6 is not discussed much in the text. For example, what is the reason for the peculiar shape of the curves in the TraCe-like experiments?



Figs 7 and 8: What are the units of the slopes? Are they Temp/AMOC or  AMOC/Temp (Fig. 7)?  Instead of looking at the influences of AMOC and  CO2 on the temperature individually, the authors could try a multiple linear regression. As it is now, the analysis could be influenced by the correlations between CO2 and AMOC.



l667: Can the large abrupt changes in MPI_Routed_glac (panel b in Fig 9) be related to the forcing in Fig. 1?

Review of Climate of the past 2023 1802

Summary

The author team investigates the early last deglaciation using a multi-model ensemble of 17 simulations. The authors take advantage form the fact that different modelling groups handle external forcings differently. Some even tested this handling using the same model. The authors identify that how meltwater is introduced to the North Atlantic strongly impacts the climate evolution of the deglaciation. The authors further found that the climate response to freshwater input is model dependent but also dependent on other forcings such as CO2 and ice sheet configuration.

General

The paper presents a classical model intercomparison approach. It is overall well structured and well wr9itte, also though it is in some parts a bit lengthy. So, I encourage the author team to shorten the manuscript. Besides I have some minor to major recommendations prior to a possible publication in Climate of the Past.

Comments

1. Title: Given that the meltwater paradox is only weakly touched in the manuscript it gets too much weight being mentioned in the title and the abstract.
2. The abstract is rather long, please shorten it.
3. L52-55: I do not understand this sentence. Maybe this could be removed.
4. L55-57: This is a rather general statement and I think can be removed.
5. L87: please change to “suggested a weaker”
6. L88: please change to “studies showed a deep”
7. L92: please change to “showed”
8. L96: please change to "demonstrated”
9. L99-100: This is a rather long sentence, so I suggest splitting it here: “… Ng et al. 2018). Modelling studies … suggested …”
10. L132: CCm2 is not explained.
11. L143-…: I think there is also some papers discussing how freshwater is implemented to the North Atlantic and how this affects the response of the AMOC. For example, Stocker et al (2007) but also references in there.

Stocker, T.F., A. Timmermann, M. Renold, O. Timm, 2007, Effects of salt compensation on the climate model response in simulations of large changes of the Atlantic meridional overturning circulation, J. Climate 20, 5912-5928

1. L151-156: The sentence is awkward, something is missing, and it is rather long, so please clarify.
2. L152: Please also check Yoshimori et al. 2010, which also show oscillations of the AMOC under cold conditions.

Yoshimori, M., M. Renold, C.C. Raible, T.F. Stocker, 2010, Simulated decadal oscillations of the Atlantic meridional overturning circulation in a cold climate state. Clim. Dyn. 34, 101-121

1. L164: “have performed” to “performed” Please check in the entire manuscript the tense.
2. L210: conducted
3. L233: demonstrated
4. L235-238: The sentence does not read good, please change.
5. L239: Suggestion: “Sun et al. (2022) showed the effect of these forcings on the sensitivity of the AMOC using multiple …”
6. L247-279: I think these two paragraphs are not necessary. PMIP could be mentioned later, see next comment 19.
7. L282: Suggestion: “… simulations available from PIMIP4 (add references here) to better understand …”
8. L294: “The comparison is based on …”
9. L452: What is meant by “This meltwater forcing presents itself in a higher variability”?
10. L536-539: This sentence is awkward, please clarify.
11. L551: Which figure do you refer here, Fig 7 or Fig 8. My guess is Fig 7.
12. L581: I suggest to also reference Fig 7 here, so: “… temperature (Fig. 7) for HadCM3_TraCE, …”
13. L620: “In this study we include multiple …”
14. L627: “see section 4.1). However, …”
15. L670: showed
16. L675: “Ensemble absolute surface air temperature in the North Atlantic“ makes no sense so please remove “Ensemble”
17. L697: “the models contribute to the “
18. L715: Missing comma before respectively.
19. L739: No line break as the next paragraph is only one sentence.
20. L752-758: The sentences are a bit strange, so please reformulate.
21. L792: How can an “impact” define a “result”? This sentence needs revisions.
22. L798: Please change “resultant impact” to “impact”.
23. L817: “... project compares simulations …”
24. L818: “… however it poses the …”

Summary:

This study presents a climate model intercomparison of the last deglaciation, which includes results from nine climate models using a range of boundary conditions and forcings. The introduction provides a thorough background on previous research linking freshwater fluxes, ice sheet geometry, AMOC, and abrupt climate events that motivates the author’s particular focus on the impact of the quantity, timing, and distribution of freshwater forcings in models. The authors present a detailed discussion of the ensemble from 20-15 ka BP and the results of their analyses, which is compelling (except for one analysis) but would benefit from summaries in each section of the main findings. The warming detection analysis from Roche et al. (2011), I believe, is incorrectly implemented. Overall, this study is an important contribution and lays the groundwork for future intercomparison studies aimed at resolving the deglacial meltwater paradox. The study would benefit, however, from more details on how the results should inform future ensembles of simulations of the last deglaciation.

Major Comments:

(1) The analysis in section 4.1 (lines 567-470 and Figure 5) to detect the start of the warming out of the LGM, which is essentially a change detection analysis, incorrectly uses a baseline that overlaps with the analysis period. The authors follow the method used in Roche et al. (2011) with the exception that they use 20-19.5 ka BP as the reference LGM climate instead of an earlier time period or control simulation. It is my understanding that the authors then conduct a one-sided Student’s t-test between this reference period and 100-year samples starting in 20 ka BP. The timing of the first 100-year sample is not explicitly stated in the text, but can instead be inferred from Figure 5 which shows the start of LGM warming occurring as early as 20 ka BP. It is the overlap of the analysis samples and the reference period that can give an incorrect result for the start of the LGM warming. Here are a few examples of this:

a) If the reference period has a cooling trend which is followed by a warming trend beginning in 19.5 ka BP, then this analysis may find that the warming started in the 20-19.9 ka BP sample simply because this part of the reference period is the warmest.

b) If entire reference period (20-15 ka BP) has a warming trend, then this analysis will likely find that the warming started near the end of the reference period or sometime afterwards. Such a result should be accompanied by an explanation that this is the latest possible time the warming could have started, and that it may have begun earlier.

c) The two examples above also show that this analysis will indicate that example (a) started warming before example (b) when in fact the opposite is true.

It may be that some of the model simulations do not suffer from these issues, but the fact that Figure 5 shows the start of warming at 20 ka BP for some locations and some models suggests that something akin to example (a) is happening. I cannot think of a scenario for which this analysis would correctly identify warming starting in 20 ka BP.

To correct this analysis, one option would be to search for the start of the warming beginning after the reference period (i.e., beginning with the 19.5-19.4 ka BP sample). As for example (b), this may mean that the start of the warming is detected after the real start. Such caveats and their implications should be stated clearly in the text.

(2) The Results/Discussion subsections provide a lot of interesting details, but the main findings in each section are not apparent. It would be helpful to have a summary of the main points at the end of each section.

(3) In the Conclusion (lines 822-826), the authors mention a protocol to “assist with narrowing down the uncertainties regarding the meltwater paradox”, but it’s not clear from the rest of the paragraph what this protocol would be. Since it seems to be a key contribution of this study to assist in designing such protocols, this deserves more discussion. At the end, the authors explain that additional experiments testing out meltwater scenarios would be beneficial. It would be helpful to know specifically how this study would inform the design of such experiments.

Minor Comments:

• Line 49: Define the AMOC acronym.
• Line 52: “demonstrate” --> “demonstrates”
• Line 86: “preceding” seems like the incorrect word here. Perhaps “subsequent”?
• Line 91: “data assimilation modelling studies” is not an accurate portrayal of these studies. “data-model comparison studies” would be more accurate.
• Line 103: It’s not clear that any feedbacks are discussed in the rest of this paragraph.
• Line 404: The section numbering is off for sections 4.1-4.5.
• Line 434-443: In this paragraph the authors conclude that freshwater forcing is the dominant driver of the abrupt temperature changes in the HadCM3_TraCE and TraCE-21ka simulations. Though I agree that this is likely the case, it is because simulations with different freshwater forcings yet similar other forcings and boundary conditions do not show this abrupt temperature decrease. This is the exact opposite reasoning that the authors use in this paragraph. The authors claim it is the fact that these two simulations are different in all respects except for the freshwater forcing that allows for the conclusion that freshwater forcing causes the temperature decrease. If many variables are different between the simulations, how can I attribute similarities or differences to any one aspect of the model?
• Lines 551, 600, 669, and 703: Fix figure citations.
• Lines 562-564: I find this sentence to be misleading. Isn’t it only the MPI model that shows this difference? The other option is the iLOVECLIM model, which doesn’t show this difference. Please expand on why the MPI models show this difference.
• Lines 583-586: The impact of CO2 could be weakened or postponed, but it could also be that the CO2-caused warming is just as strong but is masked by the larger signal of the response to the freshwater forcing. Can this be ruled out by the results?
• Line 619 & 624: This section discusses both climate and ice sheet forcings and boundary conditions. The section title and introduction to the section should reflect this.
• Line 629: Also reference a figure that shows the AMOC results, like Figure 2.
• Line 705: By “stronger impact” do the authors mean the model may be more sensitive to the orbital forcing?

Summary

Snoll and co-authors have conducted and analysed a multi-model ensemble of simulations of part of the last deglaciation. They highlight areas  of agreement and disagreement between the simulations, focussing specifically on the impact of the use of different icesheet reconstructions and associated freshweater flux boundary conditions used in each. The comparison appears to have been conducted carefully and is generally well presented, although the paper does feel over-long in places. I would recommend publications after consideration of minor issues listed below - and those raised by the other reviewers, of course.

General

I'm coming late to this paper and I can see it has a number of reviews already, so I'll try not repeat too much of what the others have already noted.

Complex climate model simulations of paleoclimate are a perennial topic and the results can often be very model and set-up dependent. This being the case I think the authors have done a good job in pulling together the range of setups and results they're working with.  The specific focus on the icesheet and freshwater boundary conditions makes for a good frame in my opinion and they've used it to produce an analysis that is more than the sum of its parts and should be useful for other researchers in this area - even if the fundamental conclusion is that well-recognised issues and internal inconsistencies between the boundary conditions and desired model behaviour are a major feature and that nothing here helps to resolve those.

Aside from some minor comments to follow, my major recommendation would be to tighten up areas that could be made more concise and to the    point - the abstract and the Introduction in particular - and to generally proofread for grammar (tense and number agreement mostly) and  context (ie give dates as well as names when referring to paleo events) to aid the understanding of less-expert readers who may not be familiar with the periods or PMIP conventions.

Specific Comments

line 1: "PMIP LDv1" may be precise, but it's not helpful for those not already familiar with the topic - so not a great candidate for inclusion in a title - and is never actually explained in the text. I'm not a fan of "The meltwater paradox reigns supreme" either - if this phrase's meaning and its implications were explained clearly in the abstract and made a major feature early in the Introduction then it could be justified, but as it is this is another rather confusing feature of the title for a non-expert.

line 42 (and on): there's a lot of detail reported in this part of the abstract that I don't think helps a reader to see the key messages of the paper.

line 64: what is the reference year for "Present" here? 1950, 2000?

line 72: "are" -> "were"

line 85 (and on): this is a very long Introduction, and reads more like a general literature review of the field rather than focusing on previous findings of specific relevance.

line 152: the clause in parentheses is very long and could be sentence or two in its own right if the result is worth saying.

line 196: I don't think the author's italics are necessary.

Figure 1: The FAMOUS line in panel b is very unclear, if it is there at all? Is panel c actually referred to at any point in the text?

line 366: Although often downplayed, it's not true to say that the UVic results are omitted from further discussion in this study is it?

line 478: "would be" should be "was"?

line 530 (and on): this phrasing might prompt a simplistic question that could be clarified: if the NAtl is the region with most constraints  and also the area with most variaiton across the simulations, why can we not just say that the simulation that's closest to the constraints must be right?

line 551 (and elsewhere): the two "Fig" references are run into each other.

line 554, 559: Two regression analyses have been done, temperature vs AMOC and temperature vs CO2. These sentences talk about R^2 values       without being very clear about which of these analyses they are refering to.

line 669: Fig references run into each other

Figure 7: the HadCM3_trace and FAMOUS correlations have oddly patterned, very strong correlations in places over Antarctica. What's going on there?

line 685: long sentence with confusing clause structure.

line 705: "is" -> "are"

line 750 (and on): a reminder of the dates for these events would be useful in this paragraph for non experts.

line 765: this long summary of Obase and Abe-Ouchi (2019) seems to go into more detail than is necessary to convey the relevant feature of their simulation.

line 768: why was this cut-off date chosen for this study? It would seem that including the Bolling Warming and the end of Heinrich1 that  occurs just beyond the cut-off would be a very useful feature to discuss in a paper on the topic of deglacial AMOC and freshwater forcing.

line 836: the Code and Data availability statements are clearly currently unverifiable.