Review of “Impacts of snow assimilation on seasonal snow and meteorological forecasts for the Tibetan Plateau” by Wei Li, Jie Chen, Lu Li, Yvan J. Orsolini, Yiheng Xiang, Retish Senan, and Patricia de Rosnay

General comments:

The manuscript describes the impact of assimilating IMS snow cover extent data over the Tibetan Plateau (i.e. over 1500 m in altitude) on the snowpack state as well as on near-surface variables (temperature, precipitation, wind) and upper air variable (600 hPa geopotential height). It is therefore well within the scope of The Cryosphere.

Although the tools, models and datasets used in this paper do not bring novelty to the scientific community, the approach and goal of the study aim to provide substantial progress and answers to the community. Indeed, this study is very promising and the results bring very important information for scientists around the world.

However, the scientific method is not clearly outlined; this manuscript lacks precision leading to multiple gaps with respect to crucial details. For example, no information is given within the manuscript as to how the assimilation of IMS data is performed and very little information is given on the simulation setup leading to important doubts on the nature of the models used in this study. Furthermore, not enough details are given to describe and explain what is in the figures.

Because this study is very interesting but lacks precision, I suggest major revisions on the manuscript.

Specific comments:

Lines 55 to 57: Is SEAS5 a 3D model then? The wording “based on” confuses the reader as to what is the link between SEAS5 and IFS. Please clarify with great precision, as it is crucial for the readers to have a clear idea of the tools and models you use for this study.

Lines 94-96: Could you add more information about the upscaling method you used? Since IMS is a binary product (0 or 1), how do you obtain a snow cover fraction? What assumptions have you made?

Section 3: this section is insufficient, more information is necessary. It should include the method used to assimilate IMS data (assimilation method, analysed variable, snow model used for background, frequency of assimilation, etc.). Please clarify the role of IFS in your set-up.

Line 128: Did you evaluate/compare the two initial states?

Line 158: The reader is not yet aware of an “inherent” model precipitation excess. Please explain and clarify what you mean. Furthermore, IMS only gives information on the presence of snow, not on the state of the snowpack, so this sentence is misleading.

Line 173: Could you explain how snow density is affected by IMS data assimilation?

Line 179: Could you explain this statement? How is albedo affected by the assimilation of IMS data?

Line 188: Too vague! Which variable are you talking about?

Lines 190-191: Reduced in Depth or in Cover Fraction? Please clarify what you mean by that.

Lines 207-208: How would you explain the decrease in correlation when using DA?

Line 210: More information should be given as to what we see in these figures: what are the + signs, explain why the CC decrease with DA, explain the high median in CC in WTP (vs. ETP), etc.

Figures S1-5 cannot be in supplementary material if you are referring to them in the text. Multiple paragraphs refer to and explain these figures and the manuscript cannot be fully understood without direct access to them. Please insert them in the manuscript.

Line 241: Explain why.

Line 255: lacks precision. Is this cumulative and total (ie. solid + liquid) precipitation? In this case, how do you convert solid precipitation to mm? Is it averaged over the domain? Lots of information missing in the text and in the corresponding figures.

Line 335+: The discussion about the changes in snowpack states should be discussed before their impact on the atmosphere. Please consider reorganising the Discussion.

Line 345 and throughout the manuscript: please clarify whether you are talking about snow albedo or total land albedo (ie. snow-free and snow-covered land as well as vegetation)

269-271: Could you explain why?

Lines 390-391: This is already mentioned just above in point 2.

Conclusions: Point 2 should be snow specific. Point 3 should be specific to the impact of snow DA on the atmosphere.

The English language needs to be improved throughout the manuscript, but more specific corrections are detailed below.

Technical corrections:

• Hectopascal units are to be written hPa (and not hpa), please correct throughout the manuscript and figures.
• Line 20: replace underestimate by underestimating.
• Lines 64-66: please improve wording.
• Lines 138-139: please improve wording.
• Lines 195 and 257: replace ‘the’ 5-day by ‘a’ 5-day
• Line 204: is CC defined before?
• Line 285: please improve wording.
• Line 332: “perform not well”, please rephrase.
• Line 362: “leads to the more precipitation”
• Line 388: the use of smaller and larger in this sentence is incorrect, please rephrase.

The omission of snow cover assimilation over the Tibetan Plateau in the ECMWF forecast system gives the authors a compelling experiment (had ECMWF really not already done that experiment in making the decision not to assimilate snow cover above 1500 m elevation?). There are few surprises in the results once we have learnt that the assimilation decreases spring snow cover, but it is still worth seeing the results.

Corrections:

line 15

It would be more informative to say that IMS snow data are assimilated, but only below 1500 m elevation.

20

“while underestimating”

57

A statement of why IMS is not assimilated in IFS and hence SEAS5 is required. I think that this is not clear either here or in de Rosnay et al. (2012).

70

State the resolution in terms that will be comprehensible to general readers.

95

Why is “zonal statistics” stated here? It is not just zonal averaging that is required to go from 4 km to 0.5 degree resolution.

Section 4.1

Because only the TP is shown in Figure 2, “of the TP” does not need to be stated so many times.

169

“(the DA reforecasts minus the control reforecasts)” has already appeared in this sentence.

179

I don’t think that snow albedo depends directly on density in IFS. The differences in albedo are clearly dominated by differences in SCF.

204

CCs used here but not explained until line 266.

275

The supplementary figures should either be in the main paper (which will result in there being a lot of figures) or the discussion of them should be in the supplement.

299

“for the two reforecasts”

A lot of the Discussion section simply restates results that will be stated again in the Conclusions

440

de Rosnay et al. (2012) reference appears twice. de Rosnay et al. (2014) is missing.

Figure 1 shows elevation; DEM is just how elevation is specified. And even common acronyms should be explained.

The Figure 2 caption should state that the difference between reforecasts is with – without.

The description of columns in Figures 3, 8 and 13 is actually rows.

This work describes the impact of the usage of snow cover data assimilation above 1500m in the ECMWF seasonal prediction system over the Tibetan Plateau, by means of analysing two set of reforecast initialised in different ways. One set of reforecasts is initialised with an analysis using IMS snow cover observations for z>1500m in the snow data assimilation system. The other set was initialised without using these observations above the z=1500m orography threshold.

The experimental setup gives the opportunity to study the impact of snow initial conditions on seasonal time-scales over the Tibetan plateau. Results are interesting and worth publishing. However, I have some comments that I would like the authors to address before the work is published, that are reported below. For these reasons, I suggest major revisions on the manuscript.

General comments

• I found the description of the model setup not clear enough, missing important details or reported in a confusing way. I suggest the authors to reorganise the “Data” and “Methods” sections. All information regarding the model (horizontal resolution, number of vertical levels in atmospheric and ocean model etc.) should be reported in one section and meaning of acronyms clearly explained (e.g. Ln. 70, “ORCA1_Z75”). For instance, Ln. 90 says that “The reforecasts have a spatial resolution of 0.5°”; however, Ln. 70 and Ln. 120 says that a TCo grid is used.
• The Results section is in many places descriptive and can be shortened, improving conciseness and clarity. For instance, the discussion of scores (CC, MAE) in 4.2.1 and 4.3 could be simplified. Another example is the discussion about snow density, which is not linked with the other variables; the underlying physical mechanism for which density is lower in the DA experiment is not clear from the text. Also, there is large usage of “supplementary” figures, in particular in Sect. 4.3, two paragraph of discussion of “Supplementary” material. If those figures are important for the discussion maybe the authors can think of moving those in the main text? Otherwise, I would suggest the authors to rearrange the text, moving for instance details that are unnecessary to support the main conclusions to an appendix?
• I acknowledge that the proposed methodology was developed with the climatology differences between West and East Tibetan plateau in mind. However, it looks to me that the main differences in precipitation (Fig. 13), or snow depth (Fig.3), are in a south-located region on the edge of the (arbitrary?) 95° line chosen by the authors. How results are sensitive to the choice of this longitude value?

Specific comments

1. Abstract: I would make it clearer that reforecasts are initialised with analysis produced with/without snow assimilation above z=1500m.
2. I found the last sentence of the abstract rather vague. Can you be more specific, e.g. which component of the surface energy balance? A plot showing which surface flux is mostly affected would be important to support this last statement.
3. Ln 70: I think more details on the model setup should be provided for people not familiar with the specific model (see main comments).
4. From the “Methods” section, is not clear if the dedicated analysis experiments are land-surface analysis only or include the analysis of the entire atmosphere+land. Please clarify in the text.
5. Was the orography threshold for using IMS observations in the snow assimilation system only removed for the Tibetan plateau region? Or was removed globally, and then the analysis focussed on the Tibetan plateau region?
6. For how long the dedicated analysis were run? Are there possible model spin-ups in the land or atmospheric fields that should be taken into account?
7. Ln 159: What does “inherent” means? Also snow model biases can contribute to snow depth errors.
8. Sect 4.1: The mechanism linking the change in snowfall, snow density and albedo is not clear from this section. An increase of snowfall in the forecast would be associated with more (new) low-density snow depositing on the ground. A fresher snowpack would be associated with a higher albedo. However, the authors found that the albedo decreases in the DA simulation. Why?
9. 1: it would be useful to have an indication on where the data assimilation is acting, that is, highlighting the grid points with orography > 1500m (from the current figure it is hard to see).
10. Ln 204-213: Does the fact that CC is lower in the reforecast with snow DA mean that the temperature variability is worsened, but the temperature biases compared to CN05.1 are improved (as shown by the reduced MAE)?
11. Ln 233: How would you explain that the correlation against in situ observation gives a different result than the correlation against CN05.1 product?
12. Ln 239: It is not clear what “obvious” mean here. Please rephrase. See also at ln 358.
13. Ln 243: what is the (mean) height above the Tibet plateau of the 600hPa surface?
14. Ln 266: The Spearman’s correlation coefficients (CCs) should be defined the first time it is used in the text.
15. Ln 324: it improves in mean error but decrease correlation.
16. I found the argument of “horizontal heat transport” a bit speculative. The authors should also show horizontal temperature maps to clearly see if warmer air is advected with the wind. For instance, could the “convergence zone” cause colder temperature from surrounding snow area (or higher mountains) to be advected over the region?
17. It would be useful to have a time series of snow depth, similarly to what it is provided for air temperature and precipitation. It would enable understanding if the increased snowfall in the snow DA reforecast compensates to some extent differences due to the initialisation. It would also clarify differences in summertime snow melt in the two reforecasts.

Technical comments

1. Ln 121: typo, “OCEAN5”.
2. Ln 204: “CC” is not defined in the text.
3. Ln 269: suggestion: I would say “lower”, not “weak”.
4. Fig. 13 legend looks wrong to me. Is it not the “spatial differences in daily precipitation (mm) between the ensemble reforecasts and GPM data” the top and middle row (not column)?