Review of the manuscript “Improvements of the land surface configuration to better simulate seasonal snow cover in the European Alps with the CNRM-AROME (cycle 46) convection-permitting regional climate model” by D. Monteiro and co-authors submitted to GMD  “egusphere-2024-249”.

The study presents an improved land surface snow representation in the CNRM-AROME model evaluated over the European Alps at convection-permitting resolution (2.5km). The snow simulations are evaluated against a large sample of in-situ snow depth observations and remote sensing snow cover. The manuscript is well written with good quality graphics and with a very detailed and interesting process-oriented discussion of the results. In my opinion, this manuscript is of interest to the community and fits well with GMD scope. However, there are several decisions on the manuscript organization and results presentation that, in my opinion, limit the main message of the study, and my suggestion would be for the authors to consider some re-organization of the results presentation and discussion, along with a few minor clarifications listed bellow. 

ES.DIF should not be an experiment but a sensitivity test, It is mentioned in section 2.3.2: “However, note that only 240 one patch is used herein for the NATURE tile, which is not the way the configuration is implemented for the coupled systems CNRM-CM6 and CNRM-ALADIN using 12 patches, and HARMONIE-Climate using 2 patches.” and also in the discussion “We also note that there are little to no cases where the ES-DIF configuration, with only one patch, has been implemented in an offline or coupled modeling system.” 

I would suggest to re-organize partially the results/discussion  having two configurations : D95-3L & ES-DIF-OPT, where ES-DIF-OPT is presented and D95-3L used as benchmark, while ES-DIF, 3-PATCHS, GFLUX, WSN-1 are sensitivity experiment to allow a process-oriented discussion. For example, in my opinion, Figures B1&B2 are the most interesting results of this study, but are presented in the appendix.  

The 3-PATCHS:  The decision to use 3-patches seems a  bit unclear. Why not 2 or 4 or 5 ? Just computational cost ? I understand that this is likely driven also by expert knowledge that is difficult to justify. However, the use of 3-patches, representing bare-soil, low-vegetation, high vegetation has been the approach taken by ECMWF since early 2000s already in ERA-40 (https://doi.org/10.21957/9aoaspz8)  more than 20 year ago. 

GLUX: The proposed cap of the conductivity at 5% below a snow fraction of 75% and then linear is a “pragmatic” approach (mentioned in the discussion), which is an acceptable justification, but more importantly it “supports the hypothesis of an overestimation of the ground heat flux, inducing basal melt in partially snow-covered surfaces”.  The harmonic average between snow/soil conductivities does not account for the air trapped between the snow base and soil top due to living/dead organic matter that will effectively reduce heat conductivity. This is difficult to represent due to the very high spatial variability of such conditions, but I suggest that this is also discussed as a possible explanation for the overestimation of the ground heat flux. 

Impact of initialization: I do not see any reason for the authors not to use the spin-up simulation ? Although it has a reduced impact for the actual period of validation, if the authors have a simulation with land initial conditions that are much better, why not use it ? 

Main result not much explored/discussed: The sensitivity results in Figure B1, in particular the activation of the 3-patches  shows that even at convection-permitting resolution of 2.5km representing the sub-grid scale variability of land surface heterogeneities of land cover are fundamental. I think that this is a very interesting result in particular within the current European destination earth program, showing that increasing horizontal resolution is not enough for a better representation of the land surface processes mostly due to the very high spatial scale of surface heterogeneities that impact the mean state over the grid-box. I would suggest the author discuss a bit on this in the conclusions and even mention it in the abstract, if they agree. 

Line 84: “e.g. HTESSEL (Balsamo et. al. 2009) : Suggest to change to ECLand (https://www.mdpi.com/2073-4433/12/6/723) as this is an updated reference for the ECMWF model that also has a multilayer snow scheme (https://doi.org/10.1029/2019MS001725) 

Lines 87-88: The sentence “The identify shortcomings may also explained some of the snow cover issues raised in …” is very unclear, please rephrase it. 

Eq. (5): Please provide in text (or table) and typical z0 values  in the presence of high vegetation used in the model to help understand actual behavior of the parameterization of snow cover fraction in these situations. THis also links with the WSN factor change from 5 to 1. Using a typical z0 value, showing a figure of the snow cover fraction as a function of snow depth with WSN=5 and WSN=1 would be illustrative. 

Table 1: Table presents in the last line “computational time relative to D95-3L” I must admit that I was very surprised to see such an increase when activating ES-DIF (+15%), and even more with -OPT just with 3 patches. In NWP/climate models the land surface component is normally a rather negligible part in the computational cost due to the “simplicity” of the calculations and to the 1D nature that typically fits very well MPI /OpenMP implementations. This is mostly a curiosity, but do the authors have some explanation for such a significant computational cost increase ? 

Line 424; “sigma” standard deviation is not used in the equations, no need to define it