the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Sep 2023
Mapping and characteristics of avalanches on mountain glaciers with Sentinel-1
Abstract. Avalanches are important contributors to the mass balance of glaciers located in mountain ranges with steep topographies. They result in localised over-accumulation that is seldom accounted for in glacier models, due to the difficulty to quantify this contribution, let alone the occurrence of avalanches in these remote regions. Here, we developed an approach to semi-automatically map avalanche deposits over long time periods and at scales of multiple glaciers, utilising imagery from Sentinel-1 Synthetic Aperture Radar (SAR). This approach performs particularly well for scenes acquired in winter and in the morning, but can also be used to identify avalanche events throughout the year. We applied this method to map 16,302 avalanche deposits over a period of five years at a 6 to 12 days interval over the Mt Blanc massif (European Alps), the Everest (Central Himalaya) and Hispar (Karakoram) regions. These three survey areas are all characterised by steep mountain slopes, but also present contrasting climatic characteristics. Our results enable the identification of avalanche hotspots at the surface of these glaciers and allow us to quantify the avalanche activity and its spatio-temporal variability across the three regions. The avalanche deposits are preferentially located at lower elevations relative to the hypsometry of the glacierized catchments, and are also constrained to a smaller elevation range at the Asian sites, where they have a limited influence on their extensive debris-covered tongues. Avalanche events coincide with solid precipitation events, which explains the high avalanche activity in winter in the Mt Blanc massif and during the monsoon in the Everest region. However, there is also a time lag of 1–2 months, visible especially in the Everest region, between the precipitation and avalanche events, indicative of some snow retention on the mountain headwalls. Ultimately, this study provides critical insights into these mass redistribution processes as well as tools to account for their influence on glacier mass balance.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(3361 KB)
-
Supplement
(3002 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3361 KB) - Metadata XML
-
Supplement
(3002 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2007', Anonymous Referee #1, 25 Oct 2023
Dear authors,
Thank you for the interesting work combining avalanche detection with glacier mass balance. I agree with you that remote sensing has great potential for avalanche identification that goes beyond what has been done so far. However, I have identified several points that need to be addressed before publication. Below you find a collection of the most relevant and general comments followed by specific comments referring to lines in your manuscript:
- In your processing chain for the Sentinel-1 data you state that you have averaged the polarization modes VV and VH. Various studies however have observed a difference in backscatter based on the polarization of around 5 dB (e.g., Liu et al., 2022). Why did you still choose to treat the two polarizations the same and did you check if treating polarizations differently would affect your results? (see also comment to 147ff)
- The (description of the) process of manual avalanche mapping for parameter estimation followed by automatic detection but then again extensive manual corrections is confusing. How good was your automatic detection, why did you need to manually correct in the first place? How much time is gained by including the automatic mapping compared to going fully manual? Additionally, it is not always clear from which processing step the results presented in the result section came from.
- There is a lack of clarity in your discussion regarding the detectability of dry snow avalanches (line 440ff, 456ff). You state that “the performance of such approaches is generally very good under dry snow conditions […] most difficulties for periods with wet snow conditions”, this contradicts (other) previous research (e.g., Eckerstorfer et al., 2022, Abermann et al., 2019). But you also contradict yourself by writing later that “cold, low density snow […] is likely to be missed by this method, which likely also explains the upper elevation limits to avalanche detections, especially during the cold season.” Please carefully reexamine the passages and literature concerning this topic and adapt your discussion.
- Concerning the structure of your manuscript: There are 21 sections, subsections and subsubsections which make it hard for the reader to orient. Please try to simplify and keep an eye on descriptive headings for your sections.
- You have included many figures, please keep only those illustrating your main points/results (see also specific comments).
- There is a lot of spoken instead of written language used, for example “this is even more true”, “indeed”, “ultimately”. Please stick to scientific written language, especially for starting and connecting your sentences.
- Lastly, you have a mix of methodology, discussion and results in those three sections. Could you please go over this again and try to disentangle. If you want to keep discussion with results you could also do a section “Results and Discussion”.
Detailed comments with reference to line numbers:
39: You make it sound like rock(fall) contributes to mass balance. Debris cover may indirectly contribute by preventing surface melt, but since you want to help calibrate glacier mass models it is weird to mention rockfall with snow/ ice avalanches as if it’s the same process. I assume they are not calibrated the same way in the models. Not being able to differentiate between those mass movements in my opinion is a limitation of the chosen method (which should be mentioned in the discussion/limitations).
70ff: The citations on previous avalanche work in the introduction do not account for most recent work while you mention most of the recent work in the discussion/436ff. Could you please try to paint a more complete picture of recent work in the introduction and not bring “new” work only in the discussion without mentioning it before.
84: Could you please quickly explain how validity was proven and what challenges were found?
101: To me it is unclear how large your three study areas are. Please specify.
Figure 1: Connected to 101: are the numbers in the upper right corner your added up area of interest? Or is it the percentage of the added up?
Please add an overview map so that every reader can get an idea where your areas of interest are located. Additionally, you need to explain RGI 6.0 at first mention and as you cannot assume that every reader knows what it is.
(d) and (e): these two graphs are quite hard to read, and I am not sure the content is essential for understanding your work. Maybe replace them with the overview map.128: It would be interesting to know if all Pleiades imagery was taken on a day Sentinel-1 was acquired also, or (and if how long) the time gap was?
147/172: Here you mention a 500m AND a 450m high pass filter? Did you filter twice? Or if these sentences describe the same operation why is the kernel size different? Additionally, a 450/500m kernel is quite large- did you perform an investigation of the effects of different filter kernel sizes on visibility of avalanches and results?
147ff: You have averaged the polarization modes VV and VH. Various studies however have observed a difference in backscatter based on the polarization of around 5 dB (e.g., Liu et al., 2022). Furthermore, the different polarization modes do not provide information about the same objects (e.g., HV results in a higher noise floor, see also Howell et al., 2019) and a combination of this dual-polarization data should hence be treated as an index rather than a ‘normal’ dB scale image as suggested in Nagler et al. (2021). Why did you treat the two polarizations the same and did you check if treating polarizations differently affects your (manual and automatic) avalanche detections.
154ff: How much of your area of interest was masked out when excluding shadow and layover? Could you please specify how much of your area was at the end classified glacierized and considered for analyses (and consequently is the area you are referring to in the rest of your manuscript).
156: You cannot assume the reader to know what RGI 6.0 is, please explain this abbreviation.
Figure 3(e): Given recent work comparing manually mapped outlines (Hafner et al., 2023), I am wondering if the blue outline is “more correct” and how dependent it is on the operator (see also comment to 224/186).
Figure 4: What exactly do you want to show with this?
224/186: Could you please give a number for the agreement between the experts (e.g., Intersection over Union, IoU) to understand how much uncertainty is introduced if one operator manually corrects predictions (see also comment to Figure 3).
234: How did you determine a match between Pleiades and Sentinel-1? How much overlap was needed?
4.1.1: This chapter would fit better into the discussion, except for some specific results.
264: As mentioned in the general comments, discussion is mixed into the results, for example here as you try to give reasons for results.
Figure 5: Even though it is known that avalanche deposits, especially large ones remain visible for a long time, it seems weird to me to compare avalanches from 1.11.2019 Sentinel-1 imagery to 9.8.2020 Pleiades. I would be very careful with this comparison as I believe a comparison of what was visible around the same time is a lot more plausible. If you compare “everything with everything” it becomes hard for the reader to follow. Hence, could you go over your analyses again and leave those not very relevant out.
276: As mentioned in the general comments, methodology is mixed into the results, like here where you mention (again) how you compared operators. The overall IoU (comment to 224) would be very nice here in addition to the consensus numbers. Additionally, how do you define an avalanche event and how do you separate it? Hafner at al. (2021) found that avalanches in Sentinel-1 might be detected in more than one blob, so just taking connected pixels might be problematic/ lead to one avalanche being counted twice or more times.
298: The references to S9b &Co are of different style than in other places and seem to be a mixture between page number and Figure caption. There is quite a few of those throughout your manuscript, please check them all and correct them (e.g., 388, 391, 474).
312: What do you mean by referencing to S11-13? Is this supposed to be page numbers? Maybe it should be refereeing to a chapter (then it should be the chapter numbering)? There is quite a few of those throughout your manuscript, please check them all and correct them (e.g., 377, 388, 391).
302: Removing 36% and adding 41% is quite a lot. I already mentioned this in the general comments, could you please add a section to the discussion where you discuss the benefits of your approach despite it not being transferable and needing quite a bit of manual work.
312/Figure 6: It is unclear for which variables correlation was calculated exactly. Please make this clear.
4.1.3: Did you use the automated mapping prior to manual corrections for this analysis? Or after the step described in 302?
Figure 7: What is N-A and M-O for the Ascending and Descending?
347: What is R2?
Figure 8: It is not clear at first sight that the legend in (a) is true for all panels. Please move the legend outside the panels as it applies to all.
359ff: Here you are mixing methodology and results again, please disentangle.
362: What is meant by activity is not clear, I assume it is repeated occurrence of an avalanche in the same place. Since it is not possible to detect whole outlines in Sentinel-1, how did you determine deposits to be “the same”, in other words how much overlap did you require?
365: You are using Change detection with a D and D-i image, with change appearing green in your data. The green hue vanishes when moving forward in time even if the deposit remains well visible/ the backscatter of a single image high. Did you analyze backscatter separately or how did you come to that conclusion?
Figure 9: I do not see a benefit in (a), the graph is hard to read. I believe a simple table could be a better choice for bringing your point across. Furthermore, the choice of color makes the differentiation between ASC and DESC hard (it is also not color blind safe). Additionally, areas detected in both ASC and DESC cannot be identified. To get the full picture, it would also be nice to be able to see the area that was masked as outside the glacierized or in radar shadow/layover.
Figure 10: This figure is hard to read. Overlapping circles cannot be distinguished and the absolute values of circles of the same size in (a), (b) und (c) differ, conveying that values are the same, though they are not. I would remove that Figure.
4.3: Please carefully reexamine for mixture of results with discussion.
Figure 11/12/13: I would suggest displaying only one of those in the manuscript and moving the other two to the Appendix. Furthermore, you should not use the same color range for number of avalanches and for avalanche area. Additionally, the number of avalanches is discrete (I assume you are displaying per acquisition day), while the way you display it implies a continuous scale. Lastly, the x-axis is the same for all panels, I think they would be better readable/comparable if the time would only be displayed once at the bottom of all three panels. In (c) you could improve readability by adding a thin grey line at 0° Celsius. Additionally, you should indicate data gaps in Fig 12 the same for all panels and not once white and once black.
440ff: “the performance is generally very good in dry snow conditions”- this contradicts findings by Eckerstorfer et al. (2022) who found a low Probability of detection for solely dry snow avalanches and you also contradict yourself in 456ff where you state that “cold, low density snow avalanches are likely to be missed”. I suppose cold low density snow avalanches make up a good proportion of avalanches occurring under dry snow conditions. You also state that a rough surface is detected by its backscatter, generally wet snow avalanches tend to have a rougher surface. Could it be that the changes in overall snow wetness, wet to dry or dry to wet from D-I to D are one of the main drivers affecting detectability. For example, in Figure 11 it seems that the avalanche activity was very low for rain on snow (after a period of low temperatures) events which are generally known for critical avalanche situations and remained low until temperature conditions were stable again over a period.
447: “such approaches”, please be specific and precise.
469: “we therefore recommend”- therefore refers to previous reasoning and discussion which is absent here. Please elaborate why you believe morning scenes are better suitable before giving a recommendation. Following your argumentation- wouldn’t it be possible to mitigate the effects of snow wetness changes (caused by a rise of temperatures during the day) by comparing morning to morning and evening to evening scenes?
472: I am a bit skeptical of you being so sure about a good manual check/ correction. Could you please elaborate a bit on why you are so sure to be able to detect (true) false positives/ false negatives without additional information.
480: Didn’t you exclude all deposits smaller than 4000 m2?
491: How did you determine an overlap? See also comment to 362.
509-522: Not all discussion is related and relevant to your work, I suggest to significantly shorten this section. Additionally, I wonder if avalanches being more concentrated at lower elevations is mostly related to wetter snow conditions and better detectability (see also Eckerstorfer et al., 2022, Abermann et al., 2019). See also comments to 440ff.
538: You contradict yourself here regarding to 440ff.
5.3: Generally, since one of your research questions is to “evaluate implications for the glacier mass balance” I expect you to be a bit more specific and elaborate on this a bit more.
556: Did they use whole outlines for parametrization or just deposits/part of the avalanche as can be detected from Sentinel-1?
562: This sections content does not really fit the chapter headline.
5.3/6: I am missing a section with a throughout discussion of the limitations of your method. That would for example include that even though your methodology can detect the approximate area and frequency of avalanches, it does not give you any information on the mass of snow involved.
573/578/589: “we successfully established a semi-automated framework”- with the manual identification of thresholds, the automatic detection and the extensive manual correction (requiring extensive domain knowledge) it remains unclear to the reader how large this benefit is. What is the gain (e.g., time) compared to full manual mapping and how much of your methodology may be reused and saves whoever uses it time (this is linked to limitations, see comment to 5.3/6). Furthermore, are there ways to translate the avalanche area into mass without additional measurements (e.g., Hynek et al., 2023) that in your case are not available.
Could you please give an outlook on what is still needed for including avalanches into glacier models large scale with your methodology.References:
Abermann, J., Eckerstorfer, M., Malnes, E., and Hansen, B. U.: A large wet snow avalanche cycle in West Greenland quantified using remote sensing and in situ observations, Nat. Hazards, 97, 517–534, https://doi.org/10.1007/s11069-019-03655-8, 2019.
Liu, C., Li, Z., Zhang, P., Huang, L., Li, Z., and Gao, S.: Wet snow detection using dual-polarized Sentinel-1 SAR time series data considering different land categories, Geocarto International, 37, 10 907–10 924, https://doi.org/10.1080/10106049.2022.2043450, 2022.
Eckerstorfer, M., Oterhals, H. D., Müller, K., Malnes, E., Grahn, J., Langeland, S., and Velsand, P.: Performance of manual and automatic detection of dry snow avalanches in Sentinel-1 SAR images, Cold Regions Science and Technology, 198, 103 549, https://doi.org/https://doi.org/10.1016/j.coldregions.2022.103549, 2022.
Hafner, E. D., Techel, F., Leinss, S., and Bühler, Y.: Mapping avalanches with satellites – evaluation of performance and completeness, The Cryosphere, 15, 983–1004, https://doi.org/10.5194/tc-15-983-2021, 2021.
Hafner, E. D., Techel, F., Daudt, R. C., Wegner, J. D., Schindler, K., and Bühler, Y.: Avalanche size estimation and avalanche outline determination by experts: reliability and implications for practice, Natural Hazards and Earth System Sciences, 23, 2895–2914, https://doi.org/10.5194/nhess-23-2895-2023, 2023.
Hynek, B., Binder, D., Citterio, M., Larsen, S. H., Abermann, J., Verhoeven, G., Ludewig, E., and Schöner, W.: Accumulation by avalanches as significant contributor to the mass balance of a High Arctic mountain glacier, The Cryosphere Discuss. [preprint], https://doi.org/10.5194/tc-2023-157, in review, 2023.
Howell, S. E., Small, D., Rohner, C., Mahmud, M. S., Yackel, J. J., and Brady, M.: Estimating melt onset over Arctic sea ice from time series multi-sensor Sentinel-1 and RADARSAT-2 backscatter, Remote Sensing of Environment, 229, 48–59, https://doi.org/https://doi.org/10.1016/j.rse.2019.04.031, 2019.
Nagler, T., Schwaizer, G., Keuris, L., Rott, H., Luojus, K., Moisander, M., Small, D., Metsämäki, S., Malnes, E. and Eckertorfer, M.: SEOM S1-4Sci Snow: Development of Pan-European Multi-Sensor Snow Mapping Methods Exploiting Sentinel-1, Final Report, Deliverable 4.2, https://eo4society.esa.int/wp-content/uploads/2021/06/S14SciSnow.D4.2_v1_2_FR.pdf, 2021.
Citation: https://doi.org/10.5194/egusphere-2023-2007-RC1 - AC1: 'Reply on RC1', Marin Kneib, 16 Feb 2024
-
RC2: 'Comment on egusphere-2023-2007', Anonymous Referee #2, 26 Jan 2024
Please find all comments in the attached PDFs.
- AC2: 'Reply on RC2', Marin Kneib, 16 Feb 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2007', Anonymous Referee #1, 25 Oct 2023
Dear authors,
Thank you for the interesting work combining avalanche detection with glacier mass balance. I agree with you that remote sensing has great potential for avalanche identification that goes beyond what has been done so far. However, I have identified several points that need to be addressed before publication. Below you find a collection of the most relevant and general comments followed by specific comments referring to lines in your manuscript:
- In your processing chain for the Sentinel-1 data you state that you have averaged the polarization modes VV and VH. Various studies however have observed a difference in backscatter based on the polarization of around 5 dB (e.g., Liu et al., 2022). Why did you still choose to treat the two polarizations the same and did you check if treating polarizations differently would affect your results? (see also comment to 147ff)
- The (description of the) process of manual avalanche mapping for parameter estimation followed by automatic detection but then again extensive manual corrections is confusing. How good was your automatic detection, why did you need to manually correct in the first place? How much time is gained by including the automatic mapping compared to going fully manual? Additionally, it is not always clear from which processing step the results presented in the result section came from.
- There is a lack of clarity in your discussion regarding the detectability of dry snow avalanches (line 440ff, 456ff). You state that “the performance of such approaches is generally very good under dry snow conditions […] most difficulties for periods with wet snow conditions”, this contradicts (other) previous research (e.g., Eckerstorfer et al., 2022, Abermann et al., 2019). But you also contradict yourself by writing later that “cold, low density snow […] is likely to be missed by this method, which likely also explains the upper elevation limits to avalanche detections, especially during the cold season.” Please carefully reexamine the passages and literature concerning this topic and adapt your discussion.
- Concerning the structure of your manuscript: There are 21 sections, subsections and subsubsections which make it hard for the reader to orient. Please try to simplify and keep an eye on descriptive headings for your sections.
- You have included many figures, please keep only those illustrating your main points/results (see also specific comments).
- There is a lot of spoken instead of written language used, for example “this is even more true”, “indeed”, “ultimately”. Please stick to scientific written language, especially for starting and connecting your sentences.
- Lastly, you have a mix of methodology, discussion and results in those three sections. Could you please go over this again and try to disentangle. If you want to keep discussion with results you could also do a section “Results and Discussion”.
Detailed comments with reference to line numbers:
39: You make it sound like rock(fall) contributes to mass balance. Debris cover may indirectly contribute by preventing surface melt, but since you want to help calibrate glacier mass models it is weird to mention rockfall with snow/ ice avalanches as if it’s the same process. I assume they are not calibrated the same way in the models. Not being able to differentiate between those mass movements in my opinion is a limitation of the chosen method (which should be mentioned in the discussion/limitations).
70ff: The citations on previous avalanche work in the introduction do not account for most recent work while you mention most of the recent work in the discussion/436ff. Could you please try to paint a more complete picture of recent work in the introduction and not bring “new” work only in the discussion without mentioning it before.
84: Could you please quickly explain how validity was proven and what challenges were found?
101: To me it is unclear how large your three study areas are. Please specify.
Figure 1: Connected to 101: are the numbers in the upper right corner your added up area of interest? Or is it the percentage of the added up?
Please add an overview map so that every reader can get an idea where your areas of interest are located. Additionally, you need to explain RGI 6.0 at first mention and as you cannot assume that every reader knows what it is.
(d) and (e): these two graphs are quite hard to read, and I am not sure the content is essential for understanding your work. Maybe replace them with the overview map.128: It would be interesting to know if all Pleiades imagery was taken on a day Sentinel-1 was acquired also, or (and if how long) the time gap was?
147/172: Here you mention a 500m AND a 450m high pass filter? Did you filter twice? Or if these sentences describe the same operation why is the kernel size different? Additionally, a 450/500m kernel is quite large- did you perform an investigation of the effects of different filter kernel sizes on visibility of avalanches and results?
147ff: You have averaged the polarization modes VV and VH. Various studies however have observed a difference in backscatter based on the polarization of around 5 dB (e.g., Liu et al., 2022). Furthermore, the different polarization modes do not provide information about the same objects (e.g., HV results in a higher noise floor, see also Howell et al., 2019) and a combination of this dual-polarization data should hence be treated as an index rather than a ‘normal’ dB scale image as suggested in Nagler et al. (2021). Why did you treat the two polarizations the same and did you check if treating polarizations differently affects your (manual and automatic) avalanche detections.
154ff: How much of your area of interest was masked out when excluding shadow and layover? Could you please specify how much of your area was at the end classified glacierized and considered for analyses (and consequently is the area you are referring to in the rest of your manuscript).
156: You cannot assume the reader to know what RGI 6.0 is, please explain this abbreviation.
Figure 3(e): Given recent work comparing manually mapped outlines (Hafner et al., 2023), I am wondering if the blue outline is “more correct” and how dependent it is on the operator (see also comment to 224/186).
Figure 4: What exactly do you want to show with this?
224/186: Could you please give a number for the agreement between the experts (e.g., Intersection over Union, IoU) to understand how much uncertainty is introduced if one operator manually corrects predictions (see also comment to Figure 3).
234: How did you determine a match between Pleiades and Sentinel-1? How much overlap was needed?
4.1.1: This chapter would fit better into the discussion, except for some specific results.
264: As mentioned in the general comments, discussion is mixed into the results, for example here as you try to give reasons for results.
Figure 5: Even though it is known that avalanche deposits, especially large ones remain visible for a long time, it seems weird to me to compare avalanches from 1.11.2019 Sentinel-1 imagery to 9.8.2020 Pleiades. I would be very careful with this comparison as I believe a comparison of what was visible around the same time is a lot more plausible. If you compare “everything with everything” it becomes hard for the reader to follow. Hence, could you go over your analyses again and leave those not very relevant out.
276: As mentioned in the general comments, methodology is mixed into the results, like here where you mention (again) how you compared operators. The overall IoU (comment to 224) would be very nice here in addition to the consensus numbers. Additionally, how do you define an avalanche event and how do you separate it? Hafner at al. (2021) found that avalanches in Sentinel-1 might be detected in more than one blob, so just taking connected pixels might be problematic/ lead to one avalanche being counted twice or more times.
298: The references to S9b &Co are of different style than in other places and seem to be a mixture between page number and Figure caption. There is quite a few of those throughout your manuscript, please check them all and correct them (e.g., 388, 391, 474).
312: What do you mean by referencing to S11-13? Is this supposed to be page numbers? Maybe it should be refereeing to a chapter (then it should be the chapter numbering)? There is quite a few of those throughout your manuscript, please check them all and correct them (e.g., 377, 388, 391).
302: Removing 36% and adding 41% is quite a lot. I already mentioned this in the general comments, could you please add a section to the discussion where you discuss the benefits of your approach despite it not being transferable and needing quite a bit of manual work.
312/Figure 6: It is unclear for which variables correlation was calculated exactly. Please make this clear.
4.1.3: Did you use the automated mapping prior to manual corrections for this analysis? Or after the step described in 302?
Figure 7: What is N-A and M-O for the Ascending and Descending?
347: What is R2?
Figure 8: It is not clear at first sight that the legend in (a) is true for all panels. Please move the legend outside the panels as it applies to all.
359ff: Here you are mixing methodology and results again, please disentangle.
362: What is meant by activity is not clear, I assume it is repeated occurrence of an avalanche in the same place. Since it is not possible to detect whole outlines in Sentinel-1, how did you determine deposits to be “the same”, in other words how much overlap did you require?
365: You are using Change detection with a D and D-i image, with change appearing green in your data. The green hue vanishes when moving forward in time even if the deposit remains well visible/ the backscatter of a single image high. Did you analyze backscatter separately or how did you come to that conclusion?
Figure 9: I do not see a benefit in (a), the graph is hard to read. I believe a simple table could be a better choice for bringing your point across. Furthermore, the choice of color makes the differentiation between ASC and DESC hard (it is also not color blind safe). Additionally, areas detected in both ASC and DESC cannot be identified. To get the full picture, it would also be nice to be able to see the area that was masked as outside the glacierized or in radar shadow/layover.
Figure 10: This figure is hard to read. Overlapping circles cannot be distinguished and the absolute values of circles of the same size in (a), (b) und (c) differ, conveying that values are the same, though they are not. I would remove that Figure.
4.3: Please carefully reexamine for mixture of results with discussion.
Figure 11/12/13: I would suggest displaying only one of those in the manuscript and moving the other two to the Appendix. Furthermore, you should not use the same color range for number of avalanches and for avalanche area. Additionally, the number of avalanches is discrete (I assume you are displaying per acquisition day), while the way you display it implies a continuous scale. Lastly, the x-axis is the same for all panels, I think they would be better readable/comparable if the time would only be displayed once at the bottom of all three panels. In (c) you could improve readability by adding a thin grey line at 0° Celsius. Additionally, you should indicate data gaps in Fig 12 the same for all panels and not once white and once black.
440ff: “the performance is generally very good in dry snow conditions”- this contradicts findings by Eckerstorfer et al. (2022) who found a low Probability of detection for solely dry snow avalanches and you also contradict yourself in 456ff where you state that “cold, low density snow avalanches are likely to be missed”. I suppose cold low density snow avalanches make up a good proportion of avalanches occurring under dry snow conditions. You also state that a rough surface is detected by its backscatter, generally wet snow avalanches tend to have a rougher surface. Could it be that the changes in overall snow wetness, wet to dry or dry to wet from D-I to D are one of the main drivers affecting detectability. For example, in Figure 11 it seems that the avalanche activity was very low for rain on snow (after a period of low temperatures) events which are generally known for critical avalanche situations and remained low until temperature conditions were stable again over a period.
447: “such approaches”, please be specific and precise.
469: “we therefore recommend”- therefore refers to previous reasoning and discussion which is absent here. Please elaborate why you believe morning scenes are better suitable before giving a recommendation. Following your argumentation- wouldn’t it be possible to mitigate the effects of snow wetness changes (caused by a rise of temperatures during the day) by comparing morning to morning and evening to evening scenes?
472: I am a bit skeptical of you being so sure about a good manual check/ correction. Could you please elaborate a bit on why you are so sure to be able to detect (true) false positives/ false negatives without additional information.
480: Didn’t you exclude all deposits smaller than 4000 m2?
491: How did you determine an overlap? See also comment to 362.
509-522: Not all discussion is related and relevant to your work, I suggest to significantly shorten this section. Additionally, I wonder if avalanches being more concentrated at lower elevations is mostly related to wetter snow conditions and better detectability (see also Eckerstorfer et al., 2022, Abermann et al., 2019). See also comments to 440ff.
538: You contradict yourself here regarding to 440ff.
5.3: Generally, since one of your research questions is to “evaluate implications for the glacier mass balance” I expect you to be a bit more specific and elaborate on this a bit more.
556: Did they use whole outlines for parametrization or just deposits/part of the avalanche as can be detected from Sentinel-1?
562: This sections content does not really fit the chapter headline.
5.3/6: I am missing a section with a throughout discussion of the limitations of your method. That would for example include that even though your methodology can detect the approximate area and frequency of avalanches, it does not give you any information on the mass of snow involved.
573/578/589: “we successfully established a semi-automated framework”- with the manual identification of thresholds, the automatic detection and the extensive manual correction (requiring extensive domain knowledge) it remains unclear to the reader how large this benefit is. What is the gain (e.g., time) compared to full manual mapping and how much of your methodology may be reused and saves whoever uses it time (this is linked to limitations, see comment to 5.3/6). Furthermore, are there ways to translate the avalanche area into mass without additional measurements (e.g., Hynek et al., 2023) that in your case are not available.
Could you please give an outlook on what is still needed for including avalanches into glacier models large scale with your methodology.References:
Abermann, J., Eckerstorfer, M., Malnes, E., and Hansen, B. U.: A large wet snow avalanche cycle in West Greenland quantified using remote sensing and in situ observations, Nat. Hazards, 97, 517–534, https://doi.org/10.1007/s11069-019-03655-8, 2019.
Liu, C., Li, Z., Zhang, P., Huang, L., Li, Z., and Gao, S.: Wet snow detection using dual-polarized Sentinel-1 SAR time series data considering different land categories, Geocarto International, 37, 10 907–10 924, https://doi.org/10.1080/10106049.2022.2043450, 2022.
Eckerstorfer, M., Oterhals, H. D., Müller, K., Malnes, E., Grahn, J., Langeland, S., and Velsand, P.: Performance of manual and automatic detection of dry snow avalanches in Sentinel-1 SAR images, Cold Regions Science and Technology, 198, 103 549, https://doi.org/https://doi.org/10.1016/j.coldregions.2022.103549, 2022.
Hafner, E. D., Techel, F., Leinss, S., and Bühler, Y.: Mapping avalanches with satellites – evaluation of performance and completeness, The Cryosphere, 15, 983–1004, https://doi.org/10.5194/tc-15-983-2021, 2021.
Hafner, E. D., Techel, F., Daudt, R. C., Wegner, J. D., Schindler, K., and Bühler, Y.: Avalanche size estimation and avalanche outline determination by experts: reliability and implications for practice, Natural Hazards and Earth System Sciences, 23, 2895–2914, https://doi.org/10.5194/nhess-23-2895-2023, 2023.
Hynek, B., Binder, D., Citterio, M., Larsen, S. H., Abermann, J., Verhoeven, G., Ludewig, E., and Schöner, W.: Accumulation by avalanches as significant contributor to the mass balance of a High Arctic mountain glacier, The Cryosphere Discuss. [preprint], https://doi.org/10.5194/tc-2023-157, in review, 2023.
Howell, S. E., Small, D., Rohner, C., Mahmud, M. S., Yackel, J. J., and Brady, M.: Estimating melt onset over Arctic sea ice from time series multi-sensor Sentinel-1 and RADARSAT-2 backscatter, Remote Sensing of Environment, 229, 48–59, https://doi.org/https://doi.org/10.1016/j.rse.2019.04.031, 2019.
Nagler, T., Schwaizer, G., Keuris, L., Rott, H., Luojus, K., Moisander, M., Small, D., Metsämäki, S., Malnes, E. and Eckertorfer, M.: SEOM S1-4Sci Snow: Development of Pan-European Multi-Sensor Snow Mapping Methods Exploiting Sentinel-1, Final Report, Deliverable 4.2, https://eo4society.esa.int/wp-content/uploads/2021/06/S14SciSnow.D4.2_v1_2_FR.pdf, 2021.
Citation: https://doi.org/10.5194/egusphere-2023-2007-RC1 - AC1: 'Reply on RC1', Marin Kneib, 16 Feb 2024
-
RC2: 'Comment on egusphere-2023-2007', Anonymous Referee #2, 26 Jan 2024
Please find all comments in the attached PDFs.
- AC2: 'Reply on RC2', Marin Kneib, 16 Feb 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 449 | 220 | 32 | 701 | 52 | 21 | 25 |
- HTML: 449
- PDF: 220
- XML: 32
- Total: 701
- Supplement: 52
- BibTeX: 21
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cited
1 citations as recorded by crossref.
Marin Kneib
Amaury Dehecq
Fanny Brun
Fatima Karbou
Laurane Charrier
Silvan Leinss
Patrick Wagnon
Fabien Maussion
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3361 KB) - Metadata XML
-
Supplement
(3002 KB) - BibTeX
- EndNote
- Final revised paper