Thank you for your nice work. Please find attached. 

The manuscript uses one new dataset over forests and peatlands situated in Northern and Southern Finland to evaluate different model configurations against surface energy fluxes, albedo, snow depth and soil temperatures. The authors emphasize the importance of testing these variables over organic layers and suggest which model configurations should or should not be used for some applications.

In its current form, not only is the manuscript too long, but it manages to be far too dense as well as too vague. There is not only too much information (e.g. there may “only” be 13 Figures in the main text and 3 in the appendices, but these 16 figures in fact include 86 windows altogether!), but also not enough detail where detail is needed. For example, many results seem to be relying on poorly understand parameters that are adjusted one way or another with little justification. I would not support the publication of this manuscript as it is, but I would strongly support it being split between a data paper and a modelling paper. This would not be “salami slicing”; the authors’ express their intention for the data to be used in future modelling exercises and by other modelling groups. As such, the data can stand on their own. So can the modelling study which, despite needing more work and clarifications regarding the adjustments mentioned above, has the potential to convey important messages to the growing community of SURFEX (and other LSM) users on how not to misuse models.

The major comments are separated into data and model:

Data:

• L311-312. How similar/different are the contiguous sites? And the meteorological stations? How much of the radiation data were missing? By R_g and R_A? How many timesteps were filled by ERA5? Ideally, either the timeseries of all meteorological observations or scatterplots showing how much these different sources differ when we do have overlapping data should be included. As this manuscript promotes a brand new dataset used for model evaluation, the gap filling in the dataset cannot be brushed off in two sentences.
• What area does the footprint of the eddy covariance towers cover? Does the vegetation cover or topography vary within the footprint? May this have consequences on the measurements? These questions may have been answered in the two cited papers, but such information is needed here.
• Much of the appendices could be transferred to a paper describing the data. Same for Section 4.1.
• L665-676. This could be added to a data paper.
• L735-737. The dataset presented here will not be widely re-used unless it is published in a data repository and a separate manuscript details the data. In a research landscape where open access to datasets and models is required by many funders and is often a pre-requisite for publication (I am surprised it is not compulsory in TC), a sentence like “Data are available upon request from the authors” raises red flags. You have done all the work on the dataset; with FAIR guiding principles becoming the norm rather than the exception, not publishing it suggests that the data are perhaps not as solid as should be.

Model

• As acknowledged by the authors, not only is w_sw important, but it is very poorly defined and, arguably, poorly understood. Table D1 clearly shows that w_sw is what I would call a “tweaking parameter”; it can range from 0.1 to 5 for no particular reason it seems. In addition, there does not seem to be any explanation given by the authors as to why different w_sw make so much difference in N-FOR but not in S-FOR. This needs to be explained, and preferably not in an Appendix (in fact, I am unclear as to what criteria were used to choose appendices over core text).
• One of the premises of this manuscript, with which I agree, is that peat and SOC are generally overlooked (l78-89) in snow modelling studies. L90: “The goal of this study is to evaluate the ability of SURFEX LSM (Surface Externalisée, Masson et al., 2013) to describe the surface energy balance and its drivers in boreal and subarctic peatlands and forests”. Yet, l290, we learn that the authors will not reach their goal by using a dataset fit for purpose, but instead “assumed the top 1 m of the peatlands to consist 100 % of SOC”. How can your whole study rely on an assumption?
• Table 2 suggests a fixed vegetation for ISBA, but nothing for MEB. How does MEB know how much of the grid box is vegetation and how much is vegetation air space? Does it change over time? What are the implications? Please clarify.
• In Section 4.4., the authors acknowledge that the lack of internal water vapour causes errors in heat exchanges between the snow and the soil and therefore potentially affects modelled soil temperatures. Knowing this, can the authors demonstrate that accounting for (an assumed) SOC is not a way to compensate for errors in the soil thermal regime that are caused by other processes that are badly or not represented?
• There are far too many plots that are not even referenced in the text. Then there are performance metrics in the plots. The more is not the better. Please reconsider whether you need 86 windows/plots in 16 figures and consider presenting your results in line with what you are highlighting in the text. For example, the different parts of the energy budget are important in different seasons, so why not have seasonal plots? You are asking the reader to compare seasonal plots (l516-517), it is your role to facilitate this if you believe this is important.

Minor comments:

The list of symbols and acronyms is huge and it makes it very hard to follow the manuscript, having to go back to previous pages to remember what is what. I would strongly advise the authors to make it easier for readers by having one or multiple tables describing the abbreviations, acronyms etc. It may also help the authors catch some that are not described (e.g. rho_sng).

Abstract: I disagree that the model included a “realistic” soil texture; the SOC values were assumed, not “real”. Please re-phrase.

L85: Incorrect reference. Krinner et al. (2018) do not provide any information at all about soil texture at ESM-SnowMIP sites; Menard et al. (2019) do.

Section 3.3.4: The models neglect LSA increases due to intercepted snow, but does intercepted snow sublimates? I could not find the answer in the manuscript even though a large percentage of snowfall is known to sublimate in coniferous forests (see Essery and Pomeroy, 2001 http://www.merrittnet.org/Papers/Essery_Pomeroy_2001.pdf for references).

Eq 9: Given how important snow density (rho) is to the calculation of the snow effective thermal conductivity, it would be helpful to know how rho is calculated.

Figure 1: What is the point of the top plot? We can hardly see where the sites are located in Finland, which would be more interesting than knowing where the boreal land biome is in the whole world. Also, could you please indicate the scale of the aerial images? Do they cover the EC tower footprint? If the scale is larger than the footprint, then, again, what is the point? The images should be proportional to how the sites are used. The manuscript presents site simulations, therefore we should have an idea of what the sites looks like. Otherwise, scrap Fig 1 altogether and simply present the sites as the parameters that represent them in the model.

Fig. 3: Add “Observed” at the start of the caption. Also, G is hardly visible

Fig 7 and others: Do we really need scatterplots, qqplots and timeseries? They are not all referenced in the text. If you want to use them all, please explain why, but I would advocate choosing.

Fig 7: Why did you choose that specific year for the timeseries?

L442: Same as in the abstract. I disagree that the model included a “realistic” soil profile; the SOC values were assumed, not “real”.

L493: “the summer energy fluxes were majorly improved by simply assigning the vegetation fraction to unity”. Is there a legitimate reason to assign the vegetation fraction to 1, or is it a tweak to “improve” the energy fluxes albeit for the wrong reasons?

L566: Do you mean Menard et al (2021)?

L584-592: This is a very important paragraph on how not to misuse or repurpose models. Splitting this manuscript into one data description and one model simulations paper would prevent such an important message from being buried deep under too much information. I would also like to see this message somewhere in the abstract.

Sections 4.3.1 + 4.3.2.: What I called “tweaking”, the authors call “compromise”. These sections are very honest about how some of the results were “improved” and are, in my opinion, the best in the manuscript. Would the authors consider be this transparent earlier in their manuscript?

Figure 12: This is very confusing. The legend convention is the opposite of Fig 4. where solid colours are the model, dashed are observations... Please be consistent.

L682-683: Does this mean that you may have achieved “satisfactory model performance” for the wrong reasons i.e. because one badly represented process compensates for another not represented at all (e.g. overestimating snow density may cause snow depth to be as low as if the model had accounted for lateral snow transport).

L699-701. Information about the footprint of the EC tower should be given earlier in the manuscript.