Review of  Wu et al: Mixed layer depth in the PMIP4 midHolocene simulations: comparison to proxy data in North Atlantic deep convection regions

Wu and colleagues present the first PMIP4 intercomparison for mid-Holocene mixed layer depths and compare it with the first available MLD reconstruction data set. There is a large spread of the MLD response to MH forcing in the PMIP4 models. The spread is overall larger than the signal of the multi-model mean (MMM). There is moderate agreement of the MMM with reconstructions in the Nordic Seas, but there is very little agreement between the MMM and the reconstructions in the Labrador Sea and the greater subpolar gyre region. Only one of the 15 models analysed shows significant agreement with the reconstructions based on a Cohen's kappa test. The authors relate the large model spread to differences in the simulated sea ice and the general disagreement between the models and the reconstructions to the missing effect of glacial meltwater in the Labrador Sea. They further highlight the importance of including the effect of glacial meltwater in simulations of future simulations to capture future MLD changes more accurately.

Understanding the response of deep convection in the North Atlantic to different climate forcings is important and highly relevant for the greater climate community. The arguments with sea ice and meltwater are convincing and mostly well presented. Before accepting the study for publication, I would ask the authors to include more information on the greater context, details about the sea-ice reconstruction and a more thorough discussion of the proxy-related uncertainties (see main comments below). The results are in parts well presented, but could be made much more accessible by improving some of the figures (see suggestions below).

Main

--------------------------------------------

Intro/ literature context: The Introduction is very short and remains quite superficial. In particular, the context of previous literature is missing. I understand that this is the first model-proxy comparison for MH MLDs, and also the first PMIP4 MLD model intercomparison. However, previous studies have looked at changes in deep convection or MLDs in individual models or proxies (e.g. Thornally et al, 2013, https://doi.org/10.5194/cp-9-2073-2013; Otto-Bliesner et al, 2020, https://doi.org/10.1029/2020PA003957). What have they found and what open questions did they leave? Also, the MLD reconstruction has already been published, so a short summary of your previous study could also be helpful in the introduction to set the scene.

Calendar adjustment: Because of the different orbital configuration, the seasons had different lengths during the MH than PI. If monthly or seasonal variables are considered, I understand that a calendar adjustment should be performed to account for that (Bartlein & Shafer, 2019, https://doi.org/10.5194/gmd-12-3889-2019). This was also done, e.g. in Brierly et al 2020. I would expect this to be relevant also for seasonal variables like sea-ice and MLD. Did you also perform a calendar adjustment? And if not, could you shortly justify why this is not necessary here?

Focus on North Atlantic only: I recommend to exclude the Southern Ocean, both from the text and the figures. The main focus of the study is the North Atlantic, so the short passages on the Southern Ocean are only disrupting the flow of the story. Excluding the SO also from the figures will greatly improve readability of the figures (see comments on Fig 1, S2 and S3).

Sea-ice reconstruction: There are no details given for the sea-ice reconstruction. What is the reconstruction based on? It seems as if it was based on the same cores, but on which proxies. Did you perform it yourself or was it previously published?  Have you performed the similarity test also for sea ice? And would the model with the highest similarity score also have the best sea-ice agreement?

Discussion of uncertainties: Please add some discussion of the proxy-related uncertainties. What are the proxy-related uncertainties? How confident are you in both the MLD and the sea-ice reconstruction? In the discussion you only state "The model biases are challenging to analyse here as there is uncertainty in the reconstructed MLD data as well (see Wu et al.,

2025)." This is not sufficient. Did you take the uncertainty into account? Can you quantify it? Is it comparable to the uncertainty of the multi-model mean (MMM)?

Minor

--------------------------------------------

l.36-37: Shortly elaborate which parts of the global climate are simulated well.

l.53-59: Please shortly explain what the modern analogue technique does. Also from reading Wu et al 2025, I understood that WOA18 is the best choice for the calibration? Why switch to Boyer-Montegut and are the results very sensitive to the calibration data set?

Fig.1: Limit the extent of the map to the North Atl., perhaps replace each global map with a smaller map as in Fig.3. Also include the proxies in each of the maps? Also please use a discrete colour-bar similar to Fig.3.

Fig.2: I find the line for the "modern data" in the MH panels somewhat misleading. Consider removing it.

l.109-111: "recurring pattern", please be specific and say which pattern is captured by the MMM.

l.130-131 & l.139-148: How do these two parts go together? It is very difficult to see from Fig.1 and S2, whether the general statement in l.130-131 holds (see comments on suggestions for Fig.1, S2 and S3). From comparing Fig.3a and Fig.S4, the general statement does also not really hold everywhere for the MMM. And, the individual model descriptions in l.139-148 also seem to show that the sea-ice extent and the MLD are unrelated in many models.

l.175-181: How confident are you in the sea-ice reconstructions (see also main comments on sea-ice reconstruction and proxy-related uncertainties)?

l.180: Fig.S4 only shows anomalies not the absolute values.

l.215-217: Would higher-frequency output be beneficial for the model-proxy comparison? Can the proxies resolve extreme events?

l.123: Not sure, that Fig.2 supports this statement of a "realistic" seasonal cycle. All models have a seasonal cycle in MLD, but most models either strongly under- or overestimate the seasonal cycle. Is that realistic?

Fig.S2 and S3: same comment as to Fig.1. Limit the map to the North Atlantic. Consider adding the proxies also to Fig.2, and again use a discrete colour-bar. And please reverse the colour-bar to have red indicating a sea-ice increase. I think I understand why red was chosen for a sea-ice decrease, but I automatically read "red" as a positive anomaly and had to re-think multiple times while looking at Figures S2 and S4.

I would suggest to include Fig S4 in the main manuscript. Please reverse the colour-bar also here to have red indicating a sea-ice increase.

Throughout the text, "observed" or "observation" is often used for a statement that is based on simulated results. I suggest to change the wording, as "observations" are immediately associated with actual measurements on the actual world. 

Summary 

Wu et al. analyze changes in mixed layer depth (MLD) during the mid-Holocene relative to the pre-industrial control (piControl) using the latest PMIP4 models. They identify substantial inter-model discrepancies in the simulated MLD response within the Labrador Sea, noting that most models fail to reproduce the shoaling indicated by proxy records. The authors suggest that the absence of meltwater forcing in the experimental design may explain this proxy-model mismatch. In contrast, the models show a relatively consistent MLD response in the Norwegian Sea, which aligns broadly with proxy data. The authors attribute this agreement to the northward retreat of the sea-ice edge driven by enhanced boreal summer insolation during the mid-Holocene.

I believe this is an important study that improves our understanding of MLD changes during the mid-Holocene. The comparison between the latest proxy reconstructions and PMIP4 simulations is novel and will likely attract significant interest from the readership of Climate of the Past. However, for the reasons listed below, I recommend major revisions. I look forward to seeing a revised version of the manuscript.

Main Comments

1. Inclusion of PMIP3 and deglaciation simulations

As this appears to be the first study conducting a proxy–model comparison of MLD for the mid-Holocene, it would be highly beneficial to include simulations from PMIP3 and available deglaciation runs in the analysis. Incorporating these outputs would allow for a more comprehensive assessment of model robustness and inter-model spread. Furthermore, including deglaciation simulations would provide the necessary data to validate the authors’ hypothesis (L182) that freshwater forcing is a key driver of MLD shoaling in the Labrador Sea.

2. Discussion of proxy and modeling uncertainties 

I am concerned about the sensitivity of the results presented in L112–121 to uncertainties in the proxy data and the specific definition of MLD used. Based on the current analysis, it appears that models with the smallest MLD changes yield smaller RMSE values (e.g., Fig. 1). If this is the case, how should the RMSE be interpreted? The authors should carefully reconcile this by discussing the impacts of proxy uncertainty and MLD definitions on their findings.

In this regard, please provide the uncertainty ranges for the reconstructed MLD for both the Mid-Holocene (MH) and Pre-Industrial (PI) periods (L56–58). One approach would be to show the maximum/minimum MLD over the 5,500–6,500 BP time domain. Since the PMIP4 models lack freshwater forcing, it may be more reasonable to compare the models against the maximum MLD of the proxy records, as these likely represent periods with minimal impact from ice discharge.

Specific Comments

L36: Could you elaborate further on this sentence (e.g., providing both global and Arctic perspectives)?

L82–83: Could the model spread be related to the magnitude of Arctic warming? For instance, do IPSL and NESM3 exhibit weaker Arctic warming compared to the others?

L155–161: Does this imply that the vegetation feedback enhances the AMOC, thereby causing the retreat of Arctic/Labrador sea ice? The spatial map of sea-ice anomalies in Fig. S2 seems consistent with this view (showing reduced sea ice in the North Atlantic and increased sea ice in the Southern Ocean). Additionally, vegetation feedback can affect Arctic sea ice by modifying planetary albedo (e.g., O’ishi and Abe-Ouchi 2011 https://doi.org/10.1029/2011GL048001; O’ishi et al. 2021 https://doi.org/10.5194/cp-17-21-2021); these mechanisms should be explicitly mentioned.

L162–163 and L172: Please clarify these points. Do the authors mean that the differences between EC-Earth3 and IPSL-CM6A-LR reflect opposing AMOC responses?

Fig. 1: I agree with Referee 1 regarding the removal of MLD maps over the Southern Ocean to keep the focus on the North Atlantic/Arctic.