Authors have developed a region-wise multifactor model to map the snow depth at 500m spatial resolution in the western Himalayas (particularly three lower, middle and upper zones). This study is essential regarding hydrology, climatology and other natural hazard perspectives and is recommended for consideration. Some of the observations are as follows:

1. Under the abstract, the authors have written about the very limited studies that were conducted on snow depth estimation using the passive microwave dataset. I suggest adding more information for more clarity. I also advised them to shorten the abstract by removing the information about the LHZ, MHZ, and UHZ (this is obvious information). Are you sure about the different regression approaches developed in this study? Recheck this statement. Mention the full form of AMSR2. Results need to be mentioned in a precise manner.  

2. Under the introduction part, the authors have covered all the aspects overall. But proofreading is required in some sentences such as their approach (L105), their models are developed (L110) etc. Revision is required for the second objective (L135) due to some grammatical issue.  Under the third objective, variables could be mentioned.

3. Under the study area, many terms are repetitively defined such as LHZ, MHZ, and UHZ (165) as you have already explained under L40. Underline 165 Upper Himalayan Zone is incorrectly abbreviated as MHZ. Under the methodology part, the flowchart is well-defined but the five steps need to be mentioned in the flowchart as explained in the subsections. The flowchart must be the stepwise reflection of the subsection (3.1 to 3.5). 

4. Results and discussion are well explained. The challenges are also defined in the discussion part. However, could you please also highlight any scope of the advanced machine learning or deep learning approach in the snow depth estimation? (Some of the previous studies also involved the neural network/deep learning approach in snow depth estimation). I think this point may increase the interest of the readers.

5. The conclusion part needs to be specific. However, it has been observed that many contents of the conclusion don't make any significant impact like the WH region is divided into three zones, i.e., LHZ, MHZ, and UHZ. So such types of lines could be removed.

Overall, many aspects have been disclosed in this article and recommended for further consideration.

Best Wishes

The authors present a novel technique for modelling snow depths using passive microwave observations in the Western Himalayan region. Their multiparameter approach is compelling and appropriate for the study area and remote sensing datasets utilized. The created multifactor model provides spatially distributed estimates of snow depth at a 500 m resolution, which are necessary for hydrologic modelling and understanding natural hazard risk in the region. While the author’s model shows promising results for the Western Himalayan region, the following points may be addressed to improve the clarity and strength of the analysis.

Notes

I recommend additional proofreading for the manuscript. Articles are used incorrectly or missing in places, and at times grammatical mistakes interfere with the meaning of the text.

Acronym use is high in the manuscript and can lead to confusion while reading. Removing some of the lesser-used acronyms would improve.

Repetition is an issue in the manuscript. Some sections of the Results and Discussion start by repeating methods or goals (e.g., lines 338, 354-356, 374-376, etc.). Please reduce repetition to improve the clarity of the text.

The Discussion section would be strengthened if the created multifactor model was discussed in more detail. This could include why the authors’ model outperformed existing models, transferability to other mountain ranges, and model error. Summaries of the methods and results are overemphasized in this section.

Specific Comments

Line 155: The area of the study region should read “360,866 km².”

Line 193: How is the AMSR2 snow depth product created? A citation would be helpful.

Line 194: The link did not work for me.

Line 210: “SCD” is not defined in the main body of the manuscript.

Line 214: More connection is needed to the studies from Sharma et al. (2014) and Singh et al. (2018). Were methods from these publications followed to estimate snow cover days?

Line 222: Was land cover different in the study period (i.e., 2012-2018) from the 2019 composite? How might this impact results?

Line 235: What was the impact of resampling the AMSR2 brightness temperature observations? How do results change if the spatial resolution is increased or decreased?

Line 236: What software was used to resample images, reproject images, and calculate brightness temperature difference? Does any other processing need to be done to AMSR2 brightness temperature data or is it done already?

Line 380: Was bias in the modelled estimates considered? I am curious if the models consistently over or underestimated snow depths.

Line 402: It is unclear to me why results from the MHZ refer to Figure 6 when the caption for Figure 6 states the reported regions are in the LHZ.

Line 410: It is unclear to me why results from the UHZ refer to Figure 6 when the caption for Figure 6 states the reported regions are in the LHZ.

Line 492: Figure 10(d) does not exist.

Line 532: Correlations of less than 0.5 are generally considered weak or moderate.

Line 534: How did local incidence angles impact the accuracy of snow depth estimates?

Figure 1, Figure 4: The acronyms “J&K” and “HP” are not defined in the figure caption.

Figure 5: Units for standard deviation are missing. Standardizing the standard deviation may improve the interpretability of the figure.  

Figure 6: Where are the Pir-Panjal, Greater Himalayan, and Karakoram regions? Perhaps these regions could be marked in Figure 1.

Figure 7: The scale for subplots b, c, and d should be the same for comparison, especially at depths of 0 cm and for regions with missing data.

Figure 8: RMSE may be better shown as scatter plots so variations in error between elevation, slope, land cover types, and snow cover days are more apparent. As depicted, these patterns are hard to assess visually considering the large study area, scale, and overlap between the stations. Further, this figure is discussed in terms of the lower, middle, and upper Himalayan zones (lines 465-470); however, only state borders are drawn on the map (rather than zone boundaries).

Table 1: The first two links in the table did not work for me. Full citations and access dates would also be helpful.

Table 6: Was the number of observations consistent between snow depth classes? On line 455 it is stated that only four of the 43 in-situ stations have mean snow depths of less than 25 cm. If there are fewer observations within this class it may influence the resulting error.

Table 7: The font for ‘Lower Himalayan Zone,’ ‘Middle Himalayan Zone,’ etc. should face the same direction as model names and RMSE values.

Review of “Passive Microwave Remote Sensing based High Resolution Snow Depth Mapping for Western Himalayan Zones using Multifactor Modelling Approach” by Dhiraj Kumar Singh et al.

 

This is a worthwhile and interesting study that explores new snow depth retrieval methodologies using passive microwave observations and associated datasets such as terrain and MODIS snow cover. A particular strength is the focus on the Western Himalaya and the use of a substantial dataset of in-situ snow depth measurements for training and validating the snow depth retrievals. An interesting aspect of the study is the way the high-resolution datasets (e.g. MODIS and terrain) are used to localise the 10 km passive microwave dataset down to 500 m. While the study could have used a more sophisticated machine learning approach, it instead uses multiple-parameter regression models including linear, power, logarithmic and so on. This is actually a very positive aspect to the study as the resulting simple models are clear and can easily be re-used by other investigators, both for retrievals and as a simple way of assessing sensitivity of snow depth to the various different parameters.

 

The work is clearly written and mostly well-presented, although some copy editing would be required in places. The main issue to be addressed is the use of the AMSR2 official snow depth product as a reference against which to assess the performance of the new algorithm. As is shown in figure 7, the official algorithm provides zero snow depth over most of the Western Himalaya at a moment when MODIS suggests the area is almost completely snow covered. The areas where AMSR2 does produce non-zero snow depths appear to be topographically linked - for example lower or flatter terrain areas. This suggests the AMSR2 algorithm is not working adequately, so it is a poor reference against which to compare. I wonder if it is possible that some bad quality flags have been set for this product, even if the data is not officially flagged as missing. The most problematic aspect of the use of the AMSR2 snow depth is that thanks to the zero values, it has apparently better performance in low snow depth conditions, in RMS error terms, than the new multi factor retrieval. This is almost certainly a spurious result (it is surely better to produce snow where it lies, even if too deep, than to “game” the results by setting almost all depths to zero). The comparison to AMSR2 needs to be de-emphasised within the results, and ideally a more adequate reference should be chosen, such as one of the legacy algorithms investigated elsewhere.

 

Given this one main issue and a number of smaller but important points, mainly relating to clarifying the work so that it can be re-used other scientists, I recommend major revisions.

 

Other main issues

 

1) The AMSR2 snow depth product needs to be described in more detail in the relevant methodology section, e.g. an overview of how the AMSR2 product is derived, whether there are any quality flags or areas where the product is known to perform poorly.

 

2) AMSR2 instrument should have a citation and there should be a DOI for the data (as for other datasets used).

 

3) The multi factor model targets dry snow conditions and hence the study is performed only in the winter period. It would be helpful to mention this in the abstract and conclusions, since this appears to be the main limitation on the validity of the model, other than the geographical specificity.

 

4) Of the 40 brightness temperature difference (BTD) pairs tested, only 8 are retained. The paper describes these 8 BTDs, but not the other 32. It is important to describe also which BTD pairs were rejected, to indicate what does not predict snow depth.

 

5) The paper needs to describe how it deals with missing data in any of the many input datasets, if there is any, and if there is not, to clearly state that.

 

6) The paper needs to explain the meaning of x_1 to x_15 in table 5 (see also line 371-372), without which the equations cannot be re-used by others.

 

7) Given the poor quality of the official AMSR2 snow product in the case study in this paper, the detailed analysis and comparison in table 7 is of very little interest to the community. It has already been established in Figure 4 and Table 6 that the AMSR2 product is not adequate in this example. Table 7 and associated text should ideally be removed.

 

8) Line 441 the snow depth classes are supposed to be grouped according to in-situ measurements, but since this analysis is done over the whole Western Himalaya, there cannot be in-situ measurement everywhere, so it is not clear what is going on.

 

9) Line 321 “a total of 72 parameters” - as with the 32 BTDs that were rejected, it is as important to know which parameters did not correlate well with snow depth, as much as those that did. Hence, in case there are any other rejected parameters that have not been described in the text, it also needs to be clear what these were, at least in summary form.

 

10) The MODIS snow cover days predictor is not clearly enough defined. I still am not sure if it is the days per year (making this a fixed map) or if it is the presence of snow cover on that specific day the snow depth retrieval is made. The text probably needs to be clearer here (see also point 6)

 

Minor issues

 

1) Linę 264: “Grody’s decision tree” should more fairly be “Grody and Basist’s decision tree”.

 

2) In defining Eq. 1- 4, "i" needs defining 

 

3) It seems incorrect to have two constant offset coefficients in equations 1, 2 and 4, since it is only possible to estimate one. One of alpha_0 and c_i should probably be removed. In any case, only one constant offset coefficient is seen in Table 5, so the equations do not appear to be consistent.

 

4) Line 362 could mention the likely reason for why descending and ascending passes have different results, namely as stated on line 527 that one local time is during daylight and hence more prone to melting snow.

 

5) Line 384-385 - consider using a table rather than loading the text with so many numerical results.