Review of "Dry snow initialization and densification over the Greenland and Antarctic ice sheets in the ORCHIDEE land surface model" by Philippe Conesa et al.

With interest I have read and reviewed this manuscript. It describes improvements in the description of the surface snow and ice of ice sheets in ORCHIDEE, part of the ISPL model.

In my view this manuscript is, after some revisions, very suitable for publication in the cryosphere. The largest proposed changes are, firstly, in the model description (sections 2 and 3), as I got (initially) lost in it and essential details are missing, see below. Secondly, the discussion and figures in sections 5.2 and 5.3, need improvements, which is also discussed below in detail. At the other hand, the proposed methods make sense to me, and the choices made on what is discussed, and what not, are good, so in my opinion no significant changes are needed on that point.

Major:

A) The model and experiment setup descriptions are unclear at points and too brief. In part, this is due to the chosen order to give first a brief overview of things in section 2, and dive further into it all in section 3. The current set-up implicitly assumes that readers more-or-less do know how ORCHIDEE works, but that is not necessarily the case (at least not for me). Therefore, I propose to restructure these two sections, namely

First a description of the (forward) ORCHIDEE model, including the description of the densification (current section 3.3.1) and surface snow density (section 3.2 - if this is part of the forward model). It is logic to me to start with brief descriptions (in words) of the parameterizations that has not been adjusted compared published versions (like heat diffusion, water retention, snow aging) and continue (in next subsections) then which elements that have been adjusted or needs to be described in more detail.

Second a description of the model/methodology to initialize ORCHIDEE (current section 3.1, and if the surface snow density estimation (current section 3.2) is not used in the forward model, also this.

Third a description of the calibration strategy, current sections 2.4, 3.3.2, and 3.3.3.

Fourth a description of the experimental setup and forcing, current section 2.3. Here, it remained unclear to me how the forcing with MAR data is carried out, remains unclear to me. It is stated that ORCHIDEE calculates a SEB, but I don't see a description how this is carried out in offline mode, where the atmosphere model state and ORCHIDEE surface state does not align, especially when MAR hasn't been run with ORCHIDEE model. How are the surface energy fluxes are adjusted to close the SEB while following the atmospheric forcing? Please describe this in the revised manuscript.

B) ORCHIDEE is part of the ESM IPSL (state that more clearly in the introduction and model description, I hardly noticed it). Therefore, I understand the rationale to use climatological relations between the surface snow density and driving processes for the snowpack initialisation. However, I would say that during dynamic simulations for climates different than the "current one", one would like to use "non-present day" data as weather and climate during a simulation will/can be different than as foreseen during the initialisation. How could that be resolved? Please comment on this and this choice in the manuscript.

Minor:

29: "Increase in ice concentration". Please rephrase as I don't understand what you mean with it. Do you mean, more bare ice at the surface, or increase ice lens formation, or even something else?

54: "years to centuries". As far as I know does densification take at least 10 years, even for extreme accumulation sites. Ice can form faster, but that is due to refreezing, and that is formally not densification.

L111: At this point in the manuscript, it is not clear why it is relevant that firn cores reach a depth of 10 m - for "full" densification models, one wish to have a profile ideally to a density of 830 kg/m3. So, mention why this 10 m is relevant for this study.

L129: Please mention the typical depth of the snowpack for accumulation conditions. I guess it is 10-20 m, given the other numbers.

L139: The order of retention and percolation does not matter that much if the time step is not too long, namely that the meltwater production during a time step is less than the retention capacity of layers. I'm surprised that switching the order makes such a difference. However, most importantly, this is discussed in detail later in the study. So, refrain here from conclusions but simply state here the code change and that the impact is discussed later in the manuscript.

L147: "Energy balance": That is commonly used for the surface energy balance, but it is uncommon for a subsurface layer. More common is, IMHO, "cold content".

L155-157: It looks like information got here twice, remove duplications.

L220: Please note that the relations of Ligtenberg et al (2011) are updated in forthcoming studies and that their exact values depend on thinks like "surface climate biases".

L230: As the surface and 10m snow temperature are two different things, this 0.8 C difference is not a bias but simply a difference. It is a different situation if the modelled 10m temperature is compared with observed 10m temperatures, however, that is premature to present in Figure 2 (if the authors wish to present that too).

L236: Please describe too what is done for locations where net ablation is expected. Is then also a 10 m snowpack described?

L384: Please motivate why 40-year average surface snow density is used. I'm not sure if (each of) the observations can be linked to a certain date (it should be, though), and if that is possible it would be more proper to take the surface snow density as modelled for that location and specific date (both by approximation, if needed). The surface snow density exhibits a seasonal cycle, especially when melting occurs, and now this cycle is ignored. This exact-time-fixing might also remove the need to "partially exclude" melting locations. It will not change that the new version is much better than the old one.

L405: You might to consider to test the IMAU-FDM fresh snow estimate for Greenland as well (Brils et al, https://doi.org/10.5194/gmd-15-7121-2022) to strengthen the point that the single formulation suitable for two ice sheets is hard to get.

Figure 6 caption: "Superior to" -> "larger than"

L420: State directly that this -2.3 kg m-3 relates to figure 7a-left and 5 m snow density, and not in the second sentence.

L437: Formally, the descriptions of the experiments (L439-441+Table 5) should be part of the experiment design, the fourth part of the new model & experiments section.

Figure 8, panels d-f: this is not an anomaly but difference.

L465: this statement about thermal conductivity description should be not new, but already mentioned in the model description - and repeated (possibly shortened) here.

L480: Again, model setup details should not be introduced in the discussion of results. Please discuss OPT12L in sufficient detail in the model description section and focus in this section on the model results.

L485-514: The analysis of the authors is correct, but I find the provided figures not very helpful to showcase this. I would propose that they instead show in Figure 10 maps of modelled melt and refreezing for the old and new version (like panel 10a), as well as a map of modelled refreezing by MAR. Next, when the authors focus on 2012, they could better show profile timeseries of snow density, snow temperature and water content (and 1D timeseries of modelled melt and vertically integrated refreezing) for the period that refreezing is buffering melt water, to demonstrate why the new version can refreeze more water. See for example https://tc.copernicus.org/articles/14/3785/2020/tc-14-3785-2020.html, Figure 2, 4, 6 ... (but then not covering years but weeks). Add to the revised caption of which period panel 10a (and the new panels) are  the average. If the authors do not adopt this change (which I would regret) please reconsider panels 10d and 10e, as the 3-hourly frequency leads to "brown area's" instead of clear difference between two simulations.

Section 5.3: The definition of ablation area is the area where (ice) mass loss exceeds mass gain, which is not equal that snow melt exceeds accumulation. Due to this IMHO incorrect definition of 'ablation area', the discussed points are in the lower percolation zone. Please adjust the naming of the region and not the discussed points (as snow densification in the true ablation zone is very dull - its only about seasonal snow). The discussion of results in this paragraph is not very convincing. Yes, for a chance for modelling realistic firn profiles, correct surface conditions (melt+temperature+accumulation) are needed, but the authors do not show/discuss if ORCHIDEE got realistic values or not - please add this. Still, if correct surface conditions are provided, models can go easily err in the lower percolation zone, as the eventual density profile depends on the long-term history (melt is increasing over Greenland, so these firn columns are not in long term equilibrium) and the efficiency of water percolation, refreezing and runoff. If the authors really would like to show how well ORCHIDEE can model the firn layer in the percolation zone, the authors should mimic (in condensed form) the analysis discussion of the 'reference paper on this matter', https://tc.copernicus.org/articles/14/3785/2020/tc-14-3785-2020.html. Given that only one model needs to be discussed, a few panels would do. I like that the authos have one figure per section, retain that, but the proposed suggestions require some choices on what to show in Figure 11 - and what needs to be shown in the supplementary materials. Lastly, relate your findings on 'what is needed to model the firn layer in the percolation zone right' to findings in existing literature, as the discussion is and will be too short for firm conclusions (which is fine).

Figure 11, S1, S2, and S4: please add dots and error bars to show the estimated 5 and 10 snow depth and the uncertainty as defined in Eq. 8-9. Formally, these uncertainties (thus error bars) should also be added in Figures 2, 3(?), 5a, 6, 7a, and 11a.

L536: "We firstly": start a new paragraph here, like you do in L541.

L541: Make clearer that these changes are within the forward model of ORCHIDEE, not in the initialization procedure.

L559-561: Such statements cannot be made without a proper introduction of the parameterization in the model description - so ensure the revised model description is detailed enough to allow statements like these. Now, something new is introduced in the 'conclusion', which should be avoided. Besides that, IMHO deeper refreezing in the lower percolation zones with ice lenses and ice slabs is more affected by the (in)ability of percolation water to reach all firn in a layer - ice lenses and ice slabs tend to make the horizontal water distribution very inhomogeneous.