the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Detection and reconstruction of rock glaciers kinematic over 24 years (2000–2024) from Landsat imagery
Abstract. The importance of monitoring rock glacier dynamics is now widely acknowledged within the scientific community following the designation of rock glacier velocity as a parameter of the Essential Climatic Variable permafrost. However, the representation of long-term spatio-temporal patterns of rock glaciers velocity at regional scale remains challenging due to the unavailability of high-resolution remote sensing datasets. This study presents a robust methodological approach based on the redundancy of information, joint with the inversion of surface displacement time series and the automatic detection of persistent moving areas (PMA) applied to rock glacier monitoring, using annual open-access, medium-resolution Landsat 7/8 optical imagery. This methodology enables the detection, quantification and analysis of surface kinematics of 382 gravitational slope movements over a 24-years, of which 153 corresponds to rock glaciers. This is the first time that Landsat images were used to quantify rock glacier displacements and derived velocities. The results demonstrate an average velocity of 0.37 ± 0.07 m y-1 overall 24-year for all rock glaciers, with some exceptions where large rock glaciers and debris frozen landform exhibit surface velocities exceeding 2 m y-1. The results of this study shows a good agreement with high-resolution imagery and recent GNSS measurements. L7/8 imagery tends to underestimate surface velocity by approximately 10–20 %. The intrinsic limitations of Landsat imagery make it challenging to interpret annual velocity variations. Notwithstanding, decadal velocity changes can be depicted for the fastest and largest rock glaciers, revealing 10 % of the accelerations in one decade. Our study suggests a correlation between surface velocity and local topographic parameters (orientation, slope, elevation) as possible controlling factors. In conclusion, this study demonstrates the feasibility of using medium-resolution optical imagery, providing an alternative to InSAR, for monitoring rock glacier kinematics anywhere over the World.
- Preprint
(20728 KB) - Metadata XML
-
Supplement
(3118 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on Cusicanqui et al. egusphere-2024-2393', Jan Henrik Blöthe, 10 Oct 2024
In their manuscript, Cusicanqui et al. use freely available Landsat 7 and 8 imagery to derive velocities of rock glaciers (and other detectable movements) in the Andes of Chile and Argentina over the course of more than two decades. Checking their derived velocities against a complementary data set using InSAR approaches as well as high-resolution data sets and DGNSS measurements for two selected field sites in Chile, the authors demonstrate the feasibility of using medium resolution imagery for the robust quantification of rock glacier surface movement.
The manuscript is well written and contains ten figures, one table, and a range of supplementary tables and figures that illustrate the findings in a comprehensive and reproducible way. The data as well as the analysis are sound, and the findings and conclusions of the study are backed by the data presented. The study is of interest for the wider community focusing on alpine permafrost research and clearly deserves publication in The Cryosphere. However, before ready for publication, I recommend addressing a number of general and specific comments that I am outlining below, hopefully helping to improve the quality of the manuscript.
General comments
- In the introduction and the formulation of the aim of the study, the authors focus entirely on the rock glaciers and their surface kinematics. The justification of the study is well written and the work is a valuable contribution to the high-Andean permafrost research. That being said, the inclusion of landslides and other “gravitational slope movements” into the analysis introduces confusion and shifts the attention of the reader away from the subject of the manuscript. In my view, focusing the content and interpretation to the rock glaciers and summarizing different movements detected by the approach applied here as “other”, or “non-rock glacier” would not only increase the clarity of the work, but also make the work more in line with the aim formulated at the beginning as well as the community addressed here. The alternative would be to include a larger paragraph on landslide processes in the region in the introduction, addressing different types of landslides found and discussing their mechanisms of movement.
- I am very glad to see that the authors go beyond the idea of calculating mean velocities for rock glaciers. Despite the wide use of mean values for entire landforms, the very heterogeneous velocity distribution found in rock glaciers makes this very difficult. In my opinion, the authors propose a suitable metric in their analysis. However, in order to make their approach comprehensible and reproducible, a consistency of terminology is very important:
- In L335-336, the authors propose to use the “Top 50% fastest pixels within each PMA” as this metric (“top 50%” and “fastest” are redundant here). I would recommend to either elaborate this more clearly in the text, or, preferably, include a small figure of one of the rock glaciers, illustrating what the idea behind this approach is. If I am not mistaken, the authors suggest calculating the mean of all velocity values between median and maximum, i.e. the upper 50% of the data distribution. If I am not mistaken, this metric is a “spatial mean”, as opposed to the “average velocities” computed over different temporal intervals. Especially for the text in the sections 5.3 and 6.3 it is imperative to have an unambiguous terminology throughout.
- To allow the interested reader to follow the processing steps, and ultimately to allow the community to make use of the workflow, the processing chain outlined mainly in sections 4.1 and 4.2 should be explained in a bit more detail (also see specific comments on that sections below).
Specific comments:
- L14-15: This statement is a bit vague, as neither “long-term” nor “high-resolution” are clearly defined. Can you be more specific here?
- L18-19: From a geomorphic perspective, gravitational slope movement would not entail rock glaciers that are bound to the presence of permafrost.
- L74: It could be stated a bit more clearly here, what “the current warming context” refers to. Also for the specific study area, rising air temperatures have been documented, as the authors describe later.
- L81-82: And for short observation periods, or high temporal resolution of SAR imagery.
- L83-84: As the Landsat 7/8 data used in the present study spans the same time interval, I don’t think this statement adds to the justification here. As you are making this argument later on, why not focus this statement on freely available SAR imagery?
- L95: As the authors are citing our work (Blöthe et al. 2021) later anyway, it deserves to be mentioned here as well, given its spatial focus.
- L102-104: If it is really the aim of the study to investigate rock glacier velocities “for the late 20th century”, imagery from 2000 to 2024 maybe wasn’t the best choice :-)
- L122: Are these weather stations located in the study area? If so, please include their locations in Figure 1.
- L139-141: I am being a bit picky here, but our study investigates velocities between 2010 and 2017/18.
- Figure 1 / L148-152: As the map shows the Gruber 2012 model of permafrost distribution, I suggest the authors also use the terminology of Gruber in the legend presented here or explain how this was adopted.
- L159-162: It is somewhat confusing at this point… maybe first mention that for all analysis you were relying on the panchromatic band, before describing the ETM+ data gap.
- L191-192: The “landslide” and “other” classes as not necessarily periglacial processes.
- L222-223: Please elaborate this processing step in a bit more detail. At this point, it remains unclear why and how exactly the DEM was used to identify the PMAs.
- L237-240: Can you explain in a sentence how the “per-alignment strategy” applied here works? Also, it is not clear to me what “for the small ones” refers to.
- L245-246: I am not sure if I understand this processing step. What exactly does “median surface displacement” refer to? Are the authors subtracting the residual mismatches from the total displacement here, or does this refer to the results of the pre-alignment strategy?
- L247-248: In a nutshell, what processing step was done here?
- L255-256: I would suggest filling the bracket with the relevant processes here that might operate in your study area. A shifting river reach would also pass the “constant direction” test applied here but is not driven by gravity.
- L264: Please explain in a bit more detail how the definition of “moving areas” (I assume this refers to PMA here?) is done. I reckon this is done for the entire time span and based on the cumulative surface displacement? How was a lower threshold applied in this to discern moving and non-moving areas?
- L273: Maybe quickly explain what processes are subsumed under “landslides” and “other”. Based on high resolution imagery presented in GoogleEarth, it would be interesting for the reader to know which processes were identified.
- Table 1 / L295-296: It would be very good to extend this table and add columns that list the number of features that remain after the additional processing steps (i.e. the surface threshold). In the current form, the data presented in L298-320 is a bit confusing, as the numbers mentioned there are not appearing in the Table 1. Furthermore, column one does not extend to the last row, suggesting that “other” is not representing valid data?
- L304-306: These numbers for example are not contained in Table 1. Also, the class “deglaciated” is not mentioned there.
- L307-308: I recommend rephrasing this sentence to increase the clarity of what the “automatic surface threshold” is and how this translates into the PMA statistics? It refers to the 10 pixel threshold, right? But how is this automatic? And what does “the most smaller” refer to?
- Figure 4 / L322-325: Maybe show removed PMAs in light colours? In the caption, mention the threshold specifically, instead of “at left of”.
- L333-334: This is not only related to the technical approach, but also has a morphological cause (i.e. increased friction and low/no ice content in lateral margins) that deserves to be mentioned here.
- L335: This relates to one of my general comments above. I suggest you define this more clearly and use a clear and recognisable terminology throughout the manuscript. E.g. in the next sentences, (L342) does “average velocity” refer to the “Top 50% average velocity”?
- Figure 5 / L357-365: If I understand the figure correctly, the viridis colour bar is also valid for the small rock glaciers, but the label “Top 50% mean velocity” is not. In the plots showing the cumulative displacement, error bars would help placing the data into the broader context.
- Figure 6 / L404-407: In the panels a) and b), a label indicating which rock glacier is the Tapado and which is the Las Tolas would help the reader following the specific results and discussion sections. Alternatively, this could also be indicated in Figure 5. Furthermore, please elaborate what a “pseudo-GCP” is and how it was constructed – it is not mentioned anywhere else in the manuscript.
- Figure 7 / L419-420: Please elaborate more precisely what is shown in the figure, especially regarding the time span of the averaged data that is plotted here. What exactly does the point to pixel comparison contain? Are these velocities averaged over the temporal range of the entire analysis? In section 3.4 it is mentioned that DGNSS records started in 2009, but what time frame is the comparison with L7/8 and VHR data based on here?
- L433-435: Again, I don’t think this statement is valid here, as the Landsat 7/8 data exploited here also roughly covers the same time span.
- L449-451: What is the detection threshold mentioned here and how was this determined?
- L512-515: While I agree that landslides in general might show a higher motion variability, I am not sure that comparing the different processes here helps the line of argument of the manuscript. The term landslides encompasses a range of different gravitational processes and with that a range of movement mechanisms that operate on very different timescales, often rather instantaneous. Also see my general comment on this.
- Figure 9 / L557-559: It should be made clear what “mean average velocity” means exactly and how this compares with the mean velocity (panel a) and average velocity (panel b). Further, use same unit m yr-1 also here.
- L568-569: I might have missed this before, but are these different types of rock glaciers? If so, quickly explain in which respect these are different.
- L600-605: If I understand this correctly, it is not only the number of images that is different between both epochs, but also the length of the epochs differ – 14 years (with six images) versus 11 years (with 12 images).
- L639-641: Frankly, this seems a bit speculative, and I am not sure which data presented here supports this statement.
- L660-664: The topographic parameters are not only not covering the entire rock glacier but leave out the feeder basin that delivers material and water to the rock glacier.
- L676: Please explain how this topographical context was extracted from which data set.
Technical corrections:
- L22: “debris frozen landforms” ?
- L26: word missing or unclear: “10% of the accelerations in one decade”?
- L32: Please rephrase! Do you mean “snow cover variability”?
- L70: delete “and”
- L71: a lot of “thus” here
- L86: “set tracking techniques”?
- L94: If I remember correctly, Eriksen et al. 2018 investigate a rock glacier in Norway…
- L105: at not “ate”
- L106: explain “VHR” here
- L124: have shown a warming trend?
- Figure 2: the font of the panel labels as well as the text is not consistent throughout the figure.
- L203 & 216: L7/8
- L285: Please provide a more appropriate heading here
- L296: “is presented in” instead of “could be found”
- L311: This should be 58%, right?
- L332: “taking into account”
- L366: Although
- L387: rock glaciers and landslides instead of ‘rock glaciers’ and ‘landslides’, or alter in the remaining manuscript
- L390: delete “further”, as this is the results section
- L417: I am not sure if “were drawn” paints the correct picture here
- L435: Elaborate which datasets you are referring to here.
- L437: “with a global reach”?
- L446: Halla et al. 2021
- L458, 460, 475: I think the text here refers to Fig. 6, not 7?
- L523: Explain the abbreviation RoGI, also use consistently (L525, Rogi)
- L568: underestimation
- L630,635: I think the text here refers to Fig. 10, not 11?
- L646: a lot of permafrost here
- L671: Conclusions
- L715-716: repetition
Citation: https://doi.org/10.5194/egusphere-2024-2393-RC1 -
RC2: 'Comment on egusphere-2024-2393', Anonymous Referee #2, 06 Nov 2024
Cusicanqui et al. estimate rock glacier kinematics on annual to decadal time scales from medium-resolution Landsat imagery. Assessing the applicability of Landsat imagery for this purpose is important because Landsat images are more widely available than higher-resolution images, while the lower spatial resolution of 15 m (panchromatic) raises questions about the suitability for measuring rock glacier kinematics on subdecadal time scales. To appraise the applicability, the authors compare the Landsat-derived motion estimates to independent observations derived from GNSS and high-resolution images, as well as to an InSAR inventory.
The study raises and addresses a question of substantial interest to readers of The Cryosphere. Its novelty lies in its being the first to use Landsat imagery for estimating rock glacier kinematics. The methods and data interpretation are, for the most part, sound, and the results are valuable to the community. While the manuscript clearly advances the field, I have concerns about the extent to which the presented data support the authors' conclusions about the observational uncertainty. Furthermore, I have found the manuscript difficult to follow because of inconsistencies in content and terminology and, at times, a writing style I consider to be verbose and vague. As I agree with Jan Henrik Blöthe's three general comments, I will not comment further on these aspects, focusing instead on my concerns about the uncertainty analysis and the presentation.
Uncertainty:
My main content-related concern pertains to the accuracy assessment. There is currently no figure or table that shows aggregated accuracy metrics, making it difficult for the reader to appraise the evidence for claims found in the abstract and conclusion. I believe it would be helpful to reorganize the accuracy assessment, introducing relevant equations in the methods and presenting the estimates in the results (including a figure with metrics such as the root mean square deviation with respect to independent estimates, the NMAD over stable areas, or the estimated deviation between temporal changes). In the abstract and conclusions, these specific metrics can be reported, while clearly distinguishing observations from subjective interpretation.
Currently, I am concerned that there is insufficient evidence for the following conclusion:
"Despite underestimations due to pixel size and temporal gaps of images, decadal velocity changes were observable under certain conditions, notably when average velocities are greater than 1 m yr-1. Below this velocity threshold, changes in velocity using L7/8 data are not statistically significant and could not be safely assessed." Similar statements can also be found in the abstract.
Was the accuracy of decadal velocity changes evaluated directly based on quantitative data? Section 5.2 contains a back-of-the envelope appraisal of the expected uncertainty in the relative change (which would fit better in the methods), but a dedicated assessment is missing. Furthermore, equation (2) seems suspect, as the numerator can be negative (I suppose it should be replaced by the geometric mean). The assumptions should be made explicit: If two quantities are assumed to be equal, it is not sufficient to say they are assumed to be similar. Section 5.3 contains an evaluation of velocity estimates, not of changes in velocity. The statement (and Section 5.2) mentions statistical significance, but it is not clear to me how and under what assumptions statistical significance was determined.
In the statement from the conclusion, the observed underestimation is attributed to discrepant spatial scales and temporal gaps. It is not apparent to me what dedicated quantitative analyses were conducted to establish this conclusion, or whether it is a subjective interpretation.Presentation:
I have found the manuscript difficult to follow, primarily for two reasons:
1) Cohesion
Discrepancies in content and terminology between the different sections presented challenges to my understanding of the manuscript. I was repeatedly taken by surprise by sudden changes in direction: The results introduce new methods and analyses that were not covered in the methods, while the discussion introduces new results and analyses not mentioned previously. The introduction contains a long literature review, but I have found it difficult to relate it to the remainder of the manuscript. In particular, the discussion section comprises six subsections, of which at least three have no easily discernible (for me) connection to the introduction or conclusion:
- Section 6.2: I am not sure how this discussion relates to the objectives of the manuscript, as its connection to the presented results is tenuous. A new figure is introduced, but it is not described in detail. Is the primary purpose of the InSAR to classify the PMAs (together with Google Earth imagery) or does it also contribute to the quantitative appraisal of the Landsat results?
- Section 6.3: Consider moving the new results shown here to the results section. In addition, the more tightly this analysis is integrated with the remainder of the manuscript, the easier it will be for the reader to appreciate it.
- Section 6.6: Introduces new results (Fig 10) that are not referred to elsewhere in the main body of the manuscript. Consider cutting it or motivating it.
Inconsistency in terminology also presents challenges to the reader. For instance, the comparison to GNSS is referred to as "ground truth" in the results only, while the word GNSS is only used in the subsection header in the results. Furthermore, the expression "false PMA" is only used in the discussion.
2) Style
I find that the verbose style detracts from the content of the manuscript. I believe that reducing the word count by 25% is a realistic target. Removal of filler phrases such as "we can also state that", "we proceeded to compare surface velocity fields in more detail", "as mentioned in", "on the other hand", or "briefly", would help the reader focus on the content. So would strong topic sentences that succinctly summarize the content of the paragraph, thus guiding the reader through the manuscript. I provide more specifics in the minor comments. In addition, extensive language edits are advised, as illustrated by the following phrases from the abstract: "The results of this study shows [...]" -> show; "over a 24-years" -> over a 24-year period or over 24 years; "of which 153 corresponds" -> correspond; "providing an alternative to InSAR, for monitoring": remove comma.Minor comments:
Title: Kinematics?
Abstract: Mention the study area?
l23: Isn't it the method that underestimates the velocity?
l115: rapid: rugged or steep?
l129-130: Incomplete sentence? Consider cutting the entire paragraph.
l160: "data gap existed": Spatial data gaps due to SLC failure or data gap because you did not include these images?
l196: I do not know what you mean by "interferograms averaged in 2-looks", as subsequently a 2 by 8 boxcar filter is mentioned.
l236: "The MGM algorithm reduces the amount of high-frequency artefacts": Compared to what method? High-frequency: what (presumably spatial) frequencies are being referred to?
l249: "redundant system" of equations? It would help to be explicit on the assumptions and methods here. Any regularization?
l260: Is the slope direction oriented down or up the slope?
l285: Fix subsection header
Table 1: Consider changing the class names "non valid" (or invalid?) and "valid" to something like not corroborated by / corroborated by InSAR to better convey the substantial epistemic uncertainty
l305: of all PMAs?
l330: How was the NMAD computed (equation)? What normalization was applied?
l334: "due to window sizes of feature-tracking algorithms": Is this the only conceivable reason? If the attribution is speculative, consider removing it from the Results.
l377: "It can be observed that": where?
l392: A concise topic sentence that summarizes the entire paragraph would make this paragraph easier to read.
l395: What is the time period of the various estimates? Where changes in displacement rates evaluated?
l401: important != big
l409: Is this the root mean square deviation?
l449: I am not sure what the hypothesis is, why it is rough, and its precise relation to the rest of the sentence.
l464: what do you mean by "enhance/difficult"
l568: sub-estimation -> underestimation?
l578: "NMAD over stable areas corresponds to 60% of the [area]" What do you mean?
l671: -> ConclusionCitation: https://doi.org/10.5194/egusphere-2024-2393-RC2
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
252 | 78 | 53 | 383 | 18 | 3 | 3 |
- HTML: 252
- PDF: 78
- XML: 53
- Total: 383
- Supplement: 18
- BibTeX: 3
- EndNote: 3
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1