the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Sep 2025
Sub-seasonal snowline dynamics of glaciers in Central Asia from multi-sensor satellite observations, 2000–2023
Abstract. Sub-seasonal glacier dynamics strongly influence the timing and magnitude of meltwater supply, a vital component of summer runoff in dry Central Asia region. Understanding of snowline evolution during the melt season is therefore essential for predicting seasonal water availability and glacier response to climate change. We present a novel method to infer 24-years of sub-seasonal snowline dynamic for four glaciers distributed throughout Pamir and Tien Shan mountain ranges using multi-sensor spaceborne observations. Our approach combines medium-resolution optical MODIS with high-resolution Sentinel-2 optical and Sentinel-1 radar imagery to produce close-to-daily estimates of the glacier Snow-Covered Area Fraction (SCAF – the ratio between snow covered area above the snowline and the total glacier area) throughout the melt season from 2000 to present. The method was validated against manually delineated Landsat snowlines, achieving RMSE values below 20 % for most sites. The resulting time series reveal substantial interannual and regional snowline variability with e.g. June SCAF ranging between 60–100 %. Recent warm years, show earlier exposure of bare ice and shifts in the melt season's end by as much as a month later in September. Accelerating snow depletion rates were found for all four glaciers, starting in 2000 and 2009 and reaching up to −1.25 %/day. Linking these dynamics to the annually measured and daily modelled mass balance data highlights that similar annual mass balance values can have large differences in sub-seasonal snow depletion and thus meltwater contribution, with implications for water availability during the critical dry-season months. Our findings demonstrate the potential of long-term, high-temporal-resolution snowline monitoring to improve understanding of glacier-climate interactions and to better constrain seasonal runoff forecasts in Central Asia's water-scarce river basins.
- Preprint
(2576 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 27 Dec 2025)
- RC1: 'Comment on egusphere-2025-3978', Anonymous Referee #1, 03 Oct 2025 reply
-
RC2: 'Comment on egusphere-2025-3978', Anonymous Referee #2, 21 Nov 2025
reply
The accurate monitoring of sub-seasonal snowline migration is of great necessity since it could provide more detailed SMB dynamics as well as meltwater distribution over glaciers. This manuscript has made nice attempts in this aspect by obtaining close-to-daily SCAF information which could potentially indicate SMB. However, to be honest, I find it a little hard to understand this paper and the language must be further polished. Although the connection between SCAF from Sentinel-1/2 and MODIS NIR exhibits some novelty, the current description about methods seems not so convincing to me, and I have serious doubts about its availability. Unfortunately, I think the current manuscript could not meet the publication standards of this journal.
Major improvements needed:
- There are too many errors in citations and punctuation marks which cannot meet the standard of a high-quality paper, please carefully check. If you use LaTex, please carefully check.
- Since the topic is snowline dynamics, it is a bit strange that you did not mention anything about SLA in the Abstract, it is better to add some direct explanations. Another possible way is to change the title to snow-covered area as you demonstrate in Line 47.
- The Introduction section can be streamlined, the current version contains too many unnecessary details, such as the in-situ observations during Soviet times.
- Line 46-47, I find it is still hard to understand due to the intrinsic influences of glacier locations and geometry on its SMB dynamics.
- As for the No.354 Glacier, why did it have the recorded average precipitation lower than the first two glaciers since it experiences frequent summer snowfalls?
- Why didn’t the authors employ Sentinel-2 data to manually snowline extraction after 2016? Provide more detailed information about the manual extracted snowline.
- When the authors conducted glacier surface classification, how did you treat other possible landcover types (e.g. rocks) as well as shadowed snow? In addition, only the NIR and backscattering information were utilized, I truly doubt if these properties could comprehensively demonstrate the differences between snow and other types of landcover and the availability of the current SCAF S2 and SCAF S1 methods. How did you determine the Otsu thresholds? Have you confirmed the availability of these thresholds in regional applications? The current description is confusing and needs modification. Some published studies which rely on SAR information to discriminate glacier zones (e.g. dry snow, wet snow, percolation and bare ice) have failed to generate a continuous snowline in the Antarctic Peninsula. Moreover, there were significant discrepancies even for same area in same year among different studies due to their various thresholds (please see Zhou et al., 2017, Idalino et al., 2024, Arigony-Neto et al., 2020).
- How did the authors treat the different spatial resolution when you merge SCAF S1 and S2 results?
- I don’t think there is a nice consistency between MODIS-based SCAF and manual results before the joint employment of S1 and S2 in Figure 4. If possible, please provide some quantitative analysis.
- Line 338, why is this large difference? As for Glacier No.354, the SCAF S1 is close to 0 while that from MODIS-based SCAF is higher than 0.75, which one could represent real conditions? Since there exhibits such significant discrepancies even in 2018, how confident the authors are when fetching SCAF by utilizing the regression functions?
- Line 374, I don’t see any trend to a later end of the melt season has become visible since 2009.
- Line 379, what do you mean that snowline rate changed by -1.25%/day? Could you illustrate this by utilizing meters for SLA dynamics? I don’t think it is appropriate to directly treat snowline as SCAF.
- It is better to provide a figure with SLA information overlapping with remote sensing images to directly exhibit the identification accuracy.
- Figure 3 to Figure 6, all these figures can be improved.
Specific comments:
Line 11, Give some quantitative descriptions about the earlier exposure of bare ice.
Line 13, Considering your research period began from 2000, and some glaciers have already exhibited accelerated snow depletion rates. I wonder how you selected these four glaciers. Is it a bit deliberate?
Line 19, Modify to by the World Meteorological Organization (WMO) in 2022.
Line 22, Add references.
Line 35-40, please add some newly published literature.
Line 57, As for reflectance, give some specific ranges.
Line 60, missing references.
Line 66, which albedo product? Be more specific.
Line 70, how long?
Line 75, give specific number of the coarse resolution.
Line 80, add a brief explanation on the low backscattering of melting snow.
Line 83, missing ‘.’.
Line 84, which new satellites?
Line 91, correct the misrepresentation. Moreover, ‘improve sub-seasonal mass balances and glacier melt water contribution’?
Line 105, missing ‘.’. There are too many similar problems in the whole manuscript.
Table 1, it should be ‘in-situ’, and as for locations, missing the specific °N and °E.
Figure 1. missing ‘.’ in the last sentence. Too many similar errors in the manuscript and I won’t list each one of them.
Line 107, (?). What do the authors mean?
Line 111, how could it be possible that a 1996 paper includes the data during same period as temperature data collected until 1998?
Line 161, remove ‘great’.
Line 166 and 171, add corresponding references.
2.2.2 Radar Data
Line 176, add corresponding references.
Line 233, why used the Italic style?
Line 243, ‘land images’?
Line 250, it should be ‘similar as in Rastner et al., (2019).’.
Line 256, ‘(3)’ ?
Line 276, how did you find the -6 dB is suitable? Need more explanation.
Line 290, why ‘20%’?
Line 295, Sentinel-1 data failed to detect dry snow? See Zhou et al., 2017, Idalino et al., 2024, Arigony-Neto et al., 2020.
Line 306, why 1.5 SD?
Figure 3, add the corresponding function in each sub-figure.
Line 316, why choose 2000-2023 and 2009-2023? Needs demonstration.
Line 330, where is Fig. 4.1.2??
Line 345, is it a satisfactory accuracy for an RMSE lower than 20%?
Line 391, more negative? Give quantitative description.
Line 439-445, this paragraph has to be re-organized.
Line 461-462, what do you mean?
5.2.2 Uncertainties and limitations, this part also needs to be re-organized.
Line 545-547, this sentence seems unreasonable to me, could the optical based method hamper the utilization of SAR data?
Line 593, could merely four glaciers represent the glacier conditions in the wide Pamir and Tien Shan regions?
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,833 | 123 | 20 | 1,976 | 36 | 34 |
- HTML: 1,833
- PDF: 123
- XML: 20
- Total: 1,976
- BibTeX: 36
- EndNote: 34
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of “Sub-seasonal snowline dynamics of glaciers in Central Asia from multi-sensor satellite observations, 2000-2023” by Kim et al.
I thoroughly enjoyed reading this paper. The paper proposes a robust and important new method. It successfully bridges a critical data gap by generating a 24-year, close-to-daily time series of the Snow-Covered Area Fraction (SCAF) for four reference glaciers in the Pamir and Tien Shan. The integration of MODIS, Sentinel-1, and Sentinel-2 data to derive SCAF is technically sound and innovative, offering a powerful tool for improving seasonal runoff forecasts and understanding glacier-climate interactions in water-scarce Central Asia.
The manuscript presents significant new data and a novel methodology that addresses a major observational deficit in glaciology. The core finding—that similar annual mass balance values can result from vastly different sub-seasonal melt dynamics—is highly relevant to water resource management. However, several methodological details, particularly regarding the data combination (SCAF S1 and SCAF S2) and validation metrics, require clarification and stronger justification to ensure the robustness and reproducibility of the proposed SCAF
My recommendation is to accept subject to a few major revisions.
Major:
General:
Title: Suggest to change to “Sub-seasonal snowline dynamics of four glaciers in Central Asia from multi-sensor satellite observations, 2000-2023”. The current title sounds like a regional assessment of all glaciers.
L19: Is this the appropriate way to format the WMO reference? Usually we see use: (WMO, 2022)
L20: Consider replacing “drastic” for a more common and scientifically precise term (e.g. Significant, Profound, Far-reaching, Severe)
L25: Unclear on the years of the “Soviet times”. Please be more specific (e.g. prior to 1985).
L26: Specify the year of the dissolution of the Soviet State?
L33: Please explain the AAR concept.
L36: Elaborate on the linear relationship, is this for all glaciers? Also suggest replacing usually with commonly.
L37: replace “on with “in” … remote sensing images
L42: replace “and concluded on the potential” with “and concluded that there is a potential”
L46: TSL not defined.
L47: Explain the difference between the SCAF and the AAR.
L56: Suggest “Snow in the accumulation…”
L66-68: Not sure this explanation does justice to the massive potential of ML workflows in recent papers assessing TSL etc. over large areas and for many glaciers, and for multiple classes, like snow, firn, ice, supraglacial ponds, off glacier areas, clouds, shadows, etc.
L76: Space in front of Aberle
L91/L107/L110/116/296/371/428/514/559: Many instances of “??” in manuscript for refs and figure cross-references. Please correct.
L97 / Fig.1: The overview map needs to better convey where we are in the world, and what the geopolitical boundaries are. Also, it would be great to see imagery of the glaciers themselves.
L97-98: Not clear on why these glaciers are chosen. There is mass balance data, but there are other glaciers with SMB data in the supplement as well. Please explain the choice of the glaciers.
L98: Can you expand on the idea of “contrasting glacier mass balance responses (Fig. 1)”. I’m looking at Figure 1 and not sure what is contrasting… Is it regional, aspect, elevation?
L100: Remove “.” In front of “(Pohl et al.[…]”
L102: add “e.g.” to the “Glacier No. 354”
Table 1: replace “in situ” with “in-situ” in caption, and add “°N and °E” to coordinates
L107: “of the four study glaciers”
Figure 1: I don’t think these are pie charts (like the caption suggests). There are colored bubbles/dots. I assume this is a regular grid of sorts. Can the grid resolution be specified in the caption (10x10km). Can the caption describe the SMB data a bit more in the caption (modelled, observed, years etc.)
Figure 1: Purple x and labels are hard to see.
Figure 1: Perhaps change legend label lower limit from -1.87 to -1.9.
L116-117: Can the annual summer, winter, and annual mass balance of the 4 glaciers be shown somewhere? Perhaps in a supplement. Fig. 9B is great, but I would like to see the rest of the glaciers as well.
L117: Without more information, I don’t think the mass balance data should be reported to three decimal places (for all glacier descriptions and Fig. 9).
L120: Chu River
L153: “. .” typo
L153: Can you quantify how many less images? What are the implications? Do you have a minimum number of images required to run the analysis?
L159: Version 6.1, not 61
L159: Yes, MODIS has 250m NIR, but the product you mention is 500m. https://ladsweb.modaps.eosdis.nasa.gov/missions-and-measurements/products/MOD09GA
L176: Description of S1A/B/C similar to the S2 description? Also, is it ascending/descending data? What are some of the challenged with SAR in mountains?
L178: replace “imagery” with “data”
L179: A stronger description of what GRD data is and what is it used for would be helpful.
L185-189: Are these polygons or lines? i.e. is it just the snowline being digitized or is it the SCA.
L187-188: Can you discuss more about the manual datasets that you created? Perhaps in a supplement? Which dates, what did you do for clouds, etc.
L190-191: This is a big issue. Having the outlines constant and using the meanNIR of the glacier presents a potentially large error that increases over time. As the glacier retreats, more land is exposed, lowering the meanNIR value of the polygon substantially. I think this requires more explanation and sensitivity analysis. This is likely somewhere ML methods would outperform Otsu.
L221: Not sure I would characterize S1 and S2 as “high resolution”. I’m assuming that’s what is being referred to here. Perhaps use “higher resolution” or simple “S1 and S2”… This comes up a few times in the manuscript. Please find more appropriate terminology and correct throughout, as “high resolution” has a definition in remote sensing. (e.g. L298)
L234-235: This statement depends on the size of the glacier. MODIS is sufficient resolution for a large enough glacier.
L236: Not sure what is meant here. Is there a fixed distance buffer (e.g. 500m)? It seems like a lot of non-glacier pixels are going to be included (pixels must contain 65% of the 1.4x larger polygon). This may exacerbate the issue listed in the first major comment.
L240: In cases when there are 3 or less cloudy pixels, are they masked out, or kept? The wording is unclear.
L243: What is meant by “land images”
L244: What is it about mountainous terrain that needs to be accounted for? Please specify.
L245: What does this algorithm do? Please explain briefly. Seems the referenced paper developed it for forested terrain?
Section 3.1.1: Please explain what happens to adjacent rock in this workflow. I assume they are groups in with “ice”. This limitation should be very clearly articulated.
L251-252: Unclear how thresholds were selected. Seems arbitrary. If they were selected manually, what dates were used? What are the implications of using different thresholds on each glacier, between seasons, and over time?
L255-256: Again, which dates, how was this done, seems arbitrary.
L256: Information missing in “(3)”.
L261: How is the total number of glacier pixels determined? (static over time from RGI?)
L266-267: What about ice? How does radar interact with ice? It is not clear how the S1 data is being used and if it is accurately identifying the snowline, as L278 suggests.
L276: How is the -6db threshold decided, please elaborate (examples, supplement, compare to other work, etc.)
L277: What method was used to resample the data?
L297: change “the” to “a linear function”
L307-308: Unclear why this is necessary? What situations would cause this, please give an example. Also, how many images are removed due to this?
L313: Please explain how the rate of change is calculated for each year.
Figure 3: Can the Landsat validation data be added here as well?
Figure 3: Not convinced that S1 is contributing much here. Also, why so few data points from S1?
Figure 3: Is this only a single year of data? If so, which year? Do you have different curves for every year?
L328: “in 4” should be “in Fig. 4” ?
L330-331: I’m not sure the wording of MODIS being more accurate is correct here. Landsat is the "validation" so how can you say MODIS is more accurate if it underestimates your validation dataset? Do you mean the higher temporal resolution allows a more precise determination of the end-of-season SCAF?
Figure 4: Hard to read x axis.
Figure 4: I thought S1 after August were removed?
Figure 5: Can this bias be explained because the glacier outlines are static? So over time more rock is exposed and the meanNIR is darker than if only snow and ice are incorporated?
Figure 6: Nice figure! A bit small and hard to dig into, but very nice way to communicate this. Perhaps remove the early season, that way the main summer months can be seen more clearly? Also, please change the black boxes. It is misleading as the color black is in your color ramp as well.
L362-L365: This may be easier to visualize in a line graph of the data in Fig 6.
Figure 7: Is it possible to add a metric of uncertainty in this graph? Perhaps the number of images in that year? Also, perhaps the min SCAF. So that we may see if the longer seasons are related to the lower SCAF.
L379: annual snowline rate of change or annual SCAF rate of change?
L391: The “annual” surface mass… ?
L439: SCAF or snowline? I think you mean SCAF?
L455: Not clear how figure B1 relates to this comment of decreased meltwater in autumn.
L505: Not the "same", but similar? What is different about these curves? can they be generalized or can the coefficients be predicted?
Figure 10: Nice figure, but not sure what is learned here?
L542-547: Unclear how useful S1 was in this study? Would the results be much different if S1 was not used?
L548: More discussion about the reliability of this method. The validation shows RMSE between 14-25%.
L550-552: This is an important point that should have more attention, as I’ve mentioned a few times now.
L555: space missing before Letréguilly.
L561: I’m not sure “interrupted” is the correct word.
L565-570: Would lag times be useful to look into here? I.e. could there be a delay in the response?
L573-575: It would be nice to include these years in Fig C1. i.e. Years where SCAF = 0, what is the SMB doing?
Figure A1: Can S1 and S2 be separated?
Figure B1: x axis is crowded and formulas are truncated.
Figure C1: change -1 to superscript in x axis label. Figure not cited in document? Could be one of the “??”