Contrasting patterns of change in snowline altitude across five Himalayan catchments

The authors established the automatic detection method for snowline altitude in the Himalayan region by using many images obtained from various satellites. Overall, the manuscript is written well. However, further manuscript improvement is necessary for the publication on The Cryosphere. I suggest the revision of the structure in the manuscript, the reconsideration of the analysis method for seasonal and interannual changes in SLA, and the enhancement of the study's significance. Please see the following comments. I hope my comments help to improve the manuscript.

Major comments:

1. I suggest you come up with a structure, figure, and table to explain the methodology. This is because you used various data (satellite images, DEM data, and reanalysis data) through many processes. A little confusing for me. Refer to the specific comments below.

2. I have concerns you use both Landsat-8 and Sentinel-2 for the trend analysis. This is because the accuracy of SLA using Landsat-8 is obviously lower than that using Sentinel-2 as you show Table 2 and Figure S3. I understand the advantage of detecting more SLA by using Landsat-8. However, you should explain the disadvantages too in the manuscript. In addition, I suggest you unify the satellite for the analysis images to discuss the SLA trends. The trends range from -15.6 m yr-1 to +14.4 m yr-1. Compared to the seasonal variations, the trends are sensitive to the bias resulting from the use of different satellites. If you choose a single satellite for the trend analysis, you might be able to see clear trends.

3. The significance of this study seemed weak. This is because it was unclear how much of the change in the detected SLA was related to the snow melt amount. Could you calculate the variation of the snow area with SLA changing in each catchment? Since it is a given that SLA has a sensitivity to weather conditions such as air temperature, I would like to see more value in this study by adding new findings. At least, you should quantitatively mention how this study contributes to the understanding of the hydrological cycle in the Himalayan region.

Specific comments:

L23: Remove “as Heading 1”.

L12: Remove or replace “Globally” in this sentence because you didn’t evaluate this algorithm globally.

2 Study site and data: I suggest the order of this section be replaced with the method section. The explanations of the data in this study appear suddenly, it is hard to read for me. 

3 Method: This is the evaluation paper for the SLA detection algorithm, you should move Table S2 to the main manuscript to show the detection method for SLA even if the flow is similar to that proposed by Girona-Mata et al. (2019). I suggest you add more detailed information, you used satellite products, atmospheric datasets, new categories (ice and water surfaces), and the point that you newly used Google Earth Engine, to the table. In addition, change and enhance the line color for the snowline. It is hard to see the black line (it is the same color as the catchment area!). It might be kind to add the explanation, you updated the previous study, to the caption.

Figure 1: Add the category of Snow to Overview in the legend. Does the white area mean the snow area, right? Also, I could not find the category of Water from the histograms. Also, it would be good if you could show the SLA in the figures (if possible).

L 73: Modify the table number. Do you have any references for the values? 

L77: What do you mean by “level”? Please explain it briefly.

L94-95: I don’t understand the expression that the SLA automatically detected using manual delineation. Is it manual or automatic?

Figure S3: What are the vertical lines in the panels (for example, the lines around the x-axis = 5500 m)? Probably, those mean the median values, but you should add an explanation in the caption.

4 Results: The Results section should be nominally limited to new results from the current observation or calculation and not include a literature review (L197, 277…). In addition, I saw the words “consider” and "suggesting" in the results section (L186, 261). The author’s speculation should be described in the discussion section. Please move the speculations to the discussion section.

L164-177: You should explain the differences in accuracy between Landsat-8 and Sentinel-2. The results from Sentinel-2 are better.

Figure 3: I suggest you show the SLA anomalies. Or, please add the value of the trend to the figure. It is hard to see the trends.

Figure 5: Why don’t you show seasonal anomalies of the SLA? Hard to see the differences…

Figure 6: I suggest you add the correlation coefficients of the interannual changes in the atmosphere variables with those in the SLA to the panels. Before the result of multiple regression analysis, I would like to see relationships between SLA and a single variable.

L296-298: I think an increase in cloud cover causes an increase in downward longwave radiation. Could you not consider that variations in downward longwave radiation contribute to SLA variations?

L376-379: The algorithm you proposed might be able to detect SLA globally, but the evaluation has not been done globally. You should add an explanation that further evaluation is necessary to apply the algorithm to glaciers worldwide.

L385-387: Related to major comment 3, please add more discussion and/or future challenges regarding effects of SLA variations on hydrological cycle (water resource management, surface mass balance, etc.)

The article, Contrasting patterns of change in snowline altitude across five Himalayan catchments, represents a tremendous effort, deriving snowline line altitudes from the Landsat and Sentinel-2  satellite missions for the period of 1999–2019. The findings are important and will likely set the stage for future High Mountain Asia studies. I found the article to be informative and well-written, but there were several areas throughout the text that I believe should be addressed before publication. I have provided thoughts, questions, and comments below.

Major Comments

1. The Introduction seems to lack important details about previous findings of the SLA or snow covered area that should frame the findings of this study. For example, the discussion of MODIS in lines 31–40 focuses on the limitations of the technique and thus largely seems to dismiss the findings of previous studies. I agree that coarser resolutions make snow covered area and snowline altitudes more challenging to interpret, but there is much that can be learned from these studies. I suggest adding in a paragraph to inform readers of the current knowledge of HMA snow cover and changing the language around the MODIS studies to better reflect their importance. Consider adding in references to Hammond et al. (2018; https://doi.org/10.1002/joc.5674), Lund et al. (2020; https://doi.org/10.3389/feart.2019.00318), and other related studies.
   
   
   
   
   
   
2. The Abstract and Conclusion of the manuscript seemingly present an overtly positive perspective of the presented methodology and its potential for global applications. While I take no issue with the positive tone and rhetoric, I think a thorough discussion of the uncertainties is lacking, but is well-warranted. In particular, there are three areas that I see as problematic: (1) the inter-satellite disagreement, (2) cloud and shadow impacts on SLA accuracy, and perhaps most critically, (3) the impact of image availability upon the calculated time series trends.
   1. Section 4.1 presents the inter-satellite comparison (Lines 178-187). I found this paragraph difficult to follow, particularly because there weren’t any figures/tables to aid in the presentation of the numbers. I think one or two supplemental figures would aid the reader here. I suggest including several nearest temporal neighbor pairs for the Landsat-5 and Landsat-7 era (I believe the orbits are offset by ~8 days?) and several pairs for the Landsat-7/8 and Sentinel-2 era.
   2. The impact of shadows and clouds are discussed in Section 5.3, but an analysis of clouds/shadows is not presented in the results. Additionally, I felt that the methods were unclear regarding how clouds in images were handled. I suggest including a few figures in the supplement of cloudy and shadowy imagery and presenting a formal analysis in the results that shows readers how these image artifacts can affect SLAs.
   3. Critically, the number of available images per basin changed after the Sentinel-2 constellation was launched. The 1999-2009 interval only saw two satellites (Landsat-5/7) each with repeat orbits of 16-days, whereas the 2009-2019 interval saw five satellites (Landsat-5/7/8 and Sentinel-2A/2B). The trends presented in the results are important and will likely be highly cited in the future. But these trends should be presented in the context of the image availability limitations, particularly given how cloudy these regions can be. I suggest that the authors include a supplemental table or figure that shows how many images were used in this analysis across time for each of the regions. I also think a simple analysis on how the image availability may affect the calculated trends is warranted.
      
      
      
      
3. Throughout the manuscript, I found the language that described the relative location of the SLAs to be inconsistent and confusing. Terms describing the SLAs included increasing vs. decreasing, maximum vs. minimum, peak vs. low peak, etc. Inherently, SLAs are more complicated to discuss than snow covered area, but I think the manuscript could still be improved. I suggest that the authors stick to one or two pairs of descriptors and apply them consistently throughout the text.

Minor Comments

Line 10 – As stated in the methods, the algorithm closely follows the Girona-Mata et al. (2019) algorithm. The primary difference appears to be a few parameters that were changed and the implementation within Google Earth Engine. To me, “propose an algorithm” implies originality. Consider using a different word or phrase.

Lines 11-12 – The study reveals significant uncertainties with this method that are not fully discussed in terms of the global implementation of this method. I think the uncertainties are worth mentioning in the abstract.

 Line 13 – Perhaps mention that the “meteorological factors” are defined by climate models.

Lines 17-19 – The phrasing here is a bit convoluted. Perhaps move the regions earlier in the phrases. For example, “We suggest that the increase in SLA in Nepal was caused by…”

Line 33 –The reference to Lievens et al. (2019) is perhaps a bit misleading. Lievens et al. (2019) used high resolution Sentinel-1 datasets to derive a coarser resolution snow depth product. Since the study did not emphasize snow covered area or snowline altitudes, I suggest removing. Additionally, the study was presented as more of a proof-of-concept and its accuracy is still highly debated in the literature.

Line 43 – For cloudy images, SLA may be less biased than snow covered area, but there can still be significant biases. I suggest revising.

Lines 46-48 – Is there anything that can be learned from these previous SLA studies that can better inform the interpretation of the results?

Line 53 – Given that this paper was submitted in 2024, I think it is misleading to use “past 20 years” when the period of study was 1999–2019.

Figure 1 – While I think the flow and presentation of the figure is excellent, I find it difficult to see the subplot labels a–e and the labels for the hypsometry in each of the subplots.

Line 71 – There is no mention of the hypsometry presented in the figure. Also, please state the source for the elevation data.

Line 73 – Please state the sources for precipitation and air temperature.

Line 78 – What is the significance of choosing watershed boundaries that have comparable size to Girona-Mata et al. (2019)? Perhaps a short clause or sentence is warranted to explain this significance.

Line 79 – Please provide an explanation for why TOA was used instead of the surface reflectance. Most publicly available snow covered or fractional snow covered area products use surface reflectance. I think the choice to use TOA is defensible, but it would be helpful for readers to understand the reasoning.

Lines 80-81 – Was resolution rescaling performed for either the Landsat or Sentinel-2 datasets to make the datasets more comparable? If so, please state how this was done. If not, please state why this choice was made.

Lines 82-83 – Given my major comment, I think a supplemental table or figure that shows how many images for each study site were obtained per year or season would be helpful for readers to better understand how image availability could bias the results.

Line 84 – Perhaps state what the DEM is used for. From the methods, it seems that the DEM was used for SLA extraction and calculation of the aspect.

Line 109 – Were clouds masked or removed from the images?

Line 112 – Perhaps a citation for NDSI is warranted.

Line 113 – Was 0.45 used as the threshold for both Landsat and Sentinel-2 NDSI products? As mentioned in line 145, the Landsat satellites and Sentinel-2 satellites have different radiometric resolutions. How might this choice have influenced the results presented in Section 4.1?

Line 116 – Perhaps a citation for improvements in Landsat-8 and Sentinel-2 is warranted.

Line 120 – please describe a bit more about this multistage process. What are the key stages in the process?

Line 166 – Why were scenes from Landsat 5 and 7 not evaluated? I would expect Landsat 7 to be particularly important to evaluate, given the scan line artifacts.

Line 185 – Seasonal snow, by definition, has large variability by season. Thus, this statement does not seem particularly meaningful to me and seems to minimize the importance of the inter-satellite variability in snowline detection. Perhaps it is sufficient to state the inter-satellite variability. This is a common problem for most studies that use multiple satellite platforms.

Lines 197-199 – The use of the word “peak” is a bit confusing. I understand that the “peak” refers to the period when snow covered area is at its greatest extent/SLA is at its lowest elevation, but this is not intuitive when looking at the plots on Figure 2, where peak conventionally refers to the time period when SLA is at its highest elevation. Please see major comments for more details.

Figure 2 – I am struck by the variability of the derived snow line altitude within the months. For example, the range in January for Figure 2d is nearly 2000 m. I think that Hammond et al. (2018) may be worth reviewing and discussing – they suggest an average snowline of ~3000 m for these catchments. Their methods and definition of snowline also differs from yours, but may be worth a brief mention/discussion. In these study regions, Hammond et al. (2018) seems to suggest relatively low snow persistence, so I am wondering if much of the variability in these derived snowlines can be attributed to snow storms that deposit snow at lower elevations but then melt out shortly after.

Line 217 – Please see major comments. Could this trend be influenced by image availability?

Line 224 – I did not see these trends presented in any of the tables or Table S1.

Figure 4f – I presume the rose diagram intervals are by 25%?

Line 242 – Per major comments, it would be helpful to share some metric for readers to better understand how image availability may be at play here.

Line 249 – the use of “lowering” and “decrease” to describe SLA in the same sentence is a bit confusing. Please see major comments.

Lines 249-262 – Much of this paragraph reads as discussion rather than results to me, mostly due to the connection to the snow/rain transition zone. In particular, the snow/rain transition zone was not defined in this study and its discussion here seems a bit unsupported to me. Further, the monsoon period is when the least number of images were available (figure 2), thus any firm claims about patterns during the monsoon season seem difficult to make without including a brief discussion of the uncertainties.

Line 274 – Since air temperature is not statistically significant, does this mean that the air temperatures tend to be below 0°C or are they poorly modeled?

Figure 6 – I suggest keeping consistent y-axes for each of the variables. For example, it is difficult to compare solar radiation across the different sites.

Lines 294-295 – Given the relatively low number of images obtained during the monsoon period, this statement seems difficult to support.

Line 302 – what does it mean for SLA to have a low peak – the rest of the sentence seems to indicate that this was when the elevation line was lowest? But this is a confusing way to describe it.

Line 304 – I’m not sure that the solar radiation from Figure 5a or Figure 6a supports this statement. Even Figure 4f seems to suggest approximately equal distribution between N and S aspects and NE and SE aspects.

Line 305-306 – Please provide a citation for the westerly winds and the snow storms. Many readers are likely not as familiar with this area as the authors and could benefit from the citations.

Line 313 – “Therefore” makes it seem as though this is a continuation of the previous sentence discussing the Satopanth Catchment, but I am not sure that another paragraph is warranted.

Line 333 – What does declining SLA mean? Does it mean that the SLA is moving up or down in elevation?

Lines 347-348 – I think this statement deserves a bit more analysis due to the cloudy nature of HMA. Please see major comments.

Line 351 – But the study does not cover the globe, only a portion of HMA. Surely this would be an issue in almost every mountainous region. Thus, would each region need to have a unique elevation masking parameter?

Lines 354-355 – The usefulness of this technique is also dependent on image availability, which has changed markedly with time and is affected by cloud cover.

Line 360 – The description of MODIS SLAs vs. derived SLAs should first be presented in the results. Please include a description of the MODIS SLA calculation and comparison to derived SLAs in the Methods.

Line 375 – This is only the second mention of the term, “arid”. Please describe for readers why arid regions are important in your study. i.e., Are the study areas considered arid?

Lines 388-392 – I’d like to see a few conclusions drawn on the limitations of the technique. As presented, the conclusions are overtly positive, but the authors identified several key uncertainties in the methods that should be considered before applying this algorithm to other regions.

Figure S3 – What do the vertical lines represent? Are those the basin-wide average? If so, please include in legend and discuss in caption.

Technical Corrections

Line 17 – Replace “while” with “whereas”.

Line 18 – Per the previous phrase in the sentence, “the decrease in SLA” is singular. Please replace “were” with “was”.

Line 23 – delete “(as Heading 1)”.

Line 73 – I believe this should be Table 1, not “Table 2”.

Line 150 – “t-tests” instead of “T-tests”?

Line 200-201 – Having a colon and a semi-colon in the same sentence seems unnecessarily complex. Consider revising to two sentences.

Line 209 – “snow line” is used here. Previously, “snowline”. Please check for consistency.

Line 228 – “aspect dependence” instead of “SLA dependence”?

Line 266 – Replace with “first half (1999-2009 in blue) and the second half (2010-2019 in red) of the study period.”?

Line 305 – SLA has already been defined.

Line 307 – “…with more snow cover on the west-facing…” As it currently reads, “snow” is arbitrary and could imply snow mass, which was not studied here.

Line 320 – Replace “nearby regions” with “study catchments”? Nearby regions is fairly arbitrary.

Line 384 – “HMA”, not “HAM”? Also, HMA is previously defined on line 24.

Lines 400-401 – This is a link to the Landsat portion of the GEE catalogue. Perhaps update to the more inclusive reference, or provide links for each of the HydroSHEDS data, Landsat data, and AW3D30.

Figure S4 – “Monsoon” instead of “Mnsoon” for figure labels? Also why are no data plotted for Sentinel-2 in Satopanth and Langtang in pre-monsoon?