Automatic delineation of cracks with Sentinel-1 interferometry for monitoring ice shelf damage and calving

Libert, Ludivine; Wuite, Jan; Nagler, Thomas

doi:https://doi.org/10.5194/tc-16-1523-2022

Articles | Volume 16, issue 4

https://doi.org/10.5194/tc-16-1523-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/tc-16-1523-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 16, issue 4

Research article

|

27 Apr 2022

Research article |

| 27 Apr 2022

Automatic delineation of cracks with Sentinel-1 interferometry for monitoring ice shelf damage and calving

Ludivine Libert, Jan Wuite, and Thomas Nagler

Download

Final revised paper (published on 27 Apr 2022)
Preprint (discussion started on 12 Oct 2021)

Interactive discussion

Status: closed

RC1:
'Comment on tc-2021-296', Anonymous Referee #1, 10 Nov 2021

Please see supplement pdf for comments

Citation: https://doi.org/10.5194/tc-2021-296-RC1
- AC1: 'Reply on RC1', Ludivine Libert, 05 Dec 2021
  
  We thank the referee for its comments and corrections, as well as for raising some fruitful discussions on the analysis. We try to respond to all the points in a relevant and concise way in the PDF attached.
  
  Citation: https://doi.org/10.5194/tc-2021-296-AC1
RC2:
'Comment on tc-2021-296', Anonymous Referee #2, 16 Nov 2021

Libert et al. map the growth of a rift on the Brunt Ice Shelf, making use of the high frequency of Sentinel-1 imagery to provide a time series of crack evolution. Their interferometric method using edge detection of phase gradient magnitudes has an advantage over interferometrically-derived strain fields in that it does not rely on phase unwrapping, however, it has a disadvantage over other rift detection techniques (for example backscatter contrast or edge detection in optical imagery) in that it requires multiple images, with good coherence between image pairs. The method accurately tracks the location of a rift on the Brunt Ice Shelf. It also approximates the timing of rift growth. The timing is not validated by other observations, so it is currently difficult to assess the accuracy in detection of the rift tip itself.

The delineation of the rift is dependent on multiple (presumably tuned) threshold parameters, various stages of filtering and line cleaning to reduce unwanted noise. This appears to be a careful balance between keeping real cracks and removing artefacts. It would be valuable if the authors explained how the parameters are determined.

The paper is well written and provides a useful dataset for this particular ice shelf, but whether it would be valuable to apply to automatic crack detection on other ice shelves, given the added complexity and data processing requirements associated with interferometry and the lack of evidence that there is any improvement in positional accuracy, or detection success, is not clear.

Specific comments:

Line 23: I don’t think you should link fracturing and damage development to climate warming here. The large independent rifts you observe on the Brunt Ice Shelf are not related to climate warming and I have some doubts that your method would work well in areas where this is the case, for example where these is dense damage / cross-cutting cracks or complex shear margins.

Line 43: sp ‘…the majority of ice shelves are routinely monitored…’

Line 54-55: There is published work that suggests that SAR backscatter imagery can be used to detect narrow cracks under the right viewing geometry (e.g. Thompson et al., 2020, 10.1016/j.coldregions.2020.103128)

Line 107: It is not necessary to repeat what you are about to say in the next section

Line 223: I expect that in the event of a crack opening in pure mode I extension parallel to an ice shelf front, while the velocity may be different on either side of the crack, the phase gradient may not show a significant difference. In this case, edge detection on your offset tracking output may work better. Have you compared this to the interferometry?

Line 249 – 258: The method appears to suffer both from false positives and false negatives. The Halloween Crack was active during this period (continuing to widen by almost 0.5 m / day in the center).

Line 288: Does this 9 x 9 refer to the value for ‘w’ in the previous section? It would be useful to restate here (i.e. w = 9).

Paragraph 288: As the method is described as automated it is important to explain how these values were determined. What is the sensitivity of the results to these values? Do these parameters need to be changed if the velocity is different, or if the coherence is worse, or under a different viewing geometry, or on a different ice shelf?

Figure 10: There are only 7 points on this graph, but you say that 32 interferograms were generated (including 9 not used due to coherence issues). It would be useful to add the other 16 points, even if the detected crack length does not change between interferograms. The crack moves in discrete jumps and any periods of no movement are also of interest. This would also help readers to assess the uncertainty in the method.

Line 363: While it may be difficult to delineate the rift in Sentinel-1 backscatter, it was visible in Landsat-8 at a similar location to the interferometry on 19^th January and almost fully visible to the Stancomb-Brunt Chasm by the 6^th February, suggesting there is not necessarily a significant information gain using this InSAR method, relative to optical imagery.

Line 366: This sentence is a bit odd, consider rephrasing.

Fig 12 (a-c): As the fringes are fairly close, it is not immediately obvious from the figures that the color order of fringes is reversed on one side of the crack with respect to the other. This is an important point to highlight that the change in velocity is opposite on either side of the crack (particularly given that the fringe frequency is similar). This could be mentioned in the text to draw the readers’ attention.

Line 410: You should probably refer to strain rate (as they are being measured) rather than stress.

Line 443: What do you mean by ‘could be misinterpreted as ice flow acceleration by offset-tracking’?

Line 449: The derived rift growth pathway agreeing with the final calving pathway does not fully demonstrate the suitability of the approach for understanding the timing of rift growth. Would it be possible to validate the derived rift tip via other means (e.g. optical imagery)?

Line 456: Again I am not sure what you are referring to here with respect to ice flow speed up vs ice drift.

Citation: https://doi.org/10.5194/tc-2021-296-RC2
- AC2: 'Reply on RC2', Ludivine Libert, 05 Dec 2021
  
  Thank you for the careful reading and valuable comments. We believe that they helped to improve the paper content. We attempt to respond all your comments in a concise way in the PDF attached.
  
  Citation: https://doi.org/10.5194/tc-2021-296-AC2

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Publish subject to revisions (further review by editor and referees) (13 Dec 2021) by Reinhard Drews

AR by Ludivine Libert on behalf of the Authors (12 Jan 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (20 Jan 2022) by Reinhard Drews

RR by Anonymous Referee #1 (31 Jan 2022)

ED: Publish subject to minor revisions (review by editor) (16 Feb 2022) by Reinhard Drews

Dear Authors,

Thank you for submitting your revisions which was positively evaluated in the re-review. The manuscript is now almost ready for publication. Congratulations!

Unfortunately, I do not concur with the reviewer’s judgement of the high presentation quality. Below you will find many comments regarding language, definitions and layout of the manuscript.

In particular, I believe we do not have the same concepts/terminology to describe the different ice dynamic components. Although I understand your desire to differentiate between different mechanisms (e.g. “deformation” and “flow” in equation (1)), I don’t think this is correctly done here (cf comment to L457). Also I believe that InSAR cannot differentiate between those components without external knowledge. This must be worked out more clearly. Another example is that “stress”, “strain” and “strain rates” appear to be used interchangeably but those are surely not the same things.

None of my comments challenge the concepts or findings of the paper, but I still urge them to take them seriously so that the paper can be made more accessible to a wider readership.

Kind regards, Reinhard Drews

L 56 remove „for the first time“. This also doesn’t fit with the reference from 2016.

L 68 “about” -> “around” (?)

L 74 InSAR does not detect “stress field changes”, only if those lead to deformation. Consider rephrasing.

L 95 I don’t think the abbreviation SWIT is needed.

L 110 Again I don’t think InSAR captures “stress”. It captures strain rates related to stress gradients.

Figure 1: Add text “Fig 1c,d” to the white dashed box in Fig 1a.

L 140 I don’t see icebergs being treated here. Remove this example.

L 140 Somehow, I feel this is described more complicated than needed. How can you differentiate in an interferogram between d^{ij}_{LOS} and v^{ij}_{LOS}\Delta^{ij} ? I don’t think you can.

Fig 2a does not show “tidal bending” (nothing is bend here). Replace with “vertical tidal displacement”. Bending only occurs in the grounding zone.

L 166: add “..SAR interferometry captures [the cumulative effect] of spatially…”. Rephrase the stress field component.

L 168 In order for tides to be visible (rather than a constant phase offset) you require a spatially variable tilting of the ice shelf, correct? I don’t understand how you infer that “…the change of vertical position for a given pair of acquisitions shows smooth spatial variations of only a few centimeters”. The interferometric phase lumps all effects (ice-shelf tilting through tides, horizontal ice flow motion,..) together so how do you infer without external evidence that one of the components is small? I agree that horizontal ice-flow is more important than a residual tidal signal in the vertical but the chain of arguments for this is unclear.

L 177 “…crack detection is performed on interferograms containing the information of all three phase components”. This sentence and much of the information in the paragraphs above suggests that you can in theory differentiate between the first three phase components in eq. (1) in an interferogram. I don’t think that this can easily be done (e.g., you would need a model for ice-shelf tilting as a response to spatially variable tides).

L 200 Which DEM is used for the topographic phase correction? What is the effect if the DEM does not capture small-scale variability such as cracks?

L 204 “The phase signal in the interferogram is a sum of the ice motion component, the tidal component and the random phase noise…” . Eq (1) lists four terms, here only three are mentioned. Does this mean one component (supposedly \phi^{ij}_{flow} ) has been removed? I don’t think this is the case. The more I advance through the paper, the less convinced I am that the distinction of \phi^{ij}_{flow} and \phi^{ij}_{defo} is needed.

L 240 All tuning parameters should be explicitly mentioned so that the study is reproducible.

L 247 What is the reason that the 50 m contour line of the TanDEM X DEM is used rather than any published grounding line (e.g., the one displayed Fig 1a)? The clipping will be unnecessarily inaccurate.

L 255 Provide numbers for “small” to keep everything reproducable.

L 319 Here the parameters are provided that I asked for previously. I think this can be presented more succinctly by including the “processing” steps directly in the methods.

L 363 “As already stated above, …” no need to start the sentence like this. Either its worth repeating or it is not.

L 428 remove “))” and “see”.

Figure 10 What does the AA stand for in the title?

L457 “..changes in flow velocity or creep deformation..“. I believe we have different ice dynamic concepts. For me ice-flow has two components one due to internal ice-deformation and one to basal sliding. Both leads to a displacement at the surface. If “creep deformation changes” so does the “ice flow”. I believe what you are trying to distinguish here is “plug flow” (which is the most relevant flow regime for ice shelves with essentially 100% basal sliding and 0% deformation) and other components. Anyway, the way this is written here and also elsewhere repeatedly confuses me.

Later in that sentence only “ice flow” cancels during the doubled differencing. What happens to the “creep deformation” ? Further on in this sentence only the “natural variations…and the deformation” are left in the double differenced interferograms. What do you mean by “natural variations” ? Tides go “naturally” up and down, but this is not what is left in a double differenced interferogram. What is left is the differential tidal signal.

How about something like: “The differential phase of double differenced interferograms is composed out of the differential vertical displacement by tides and time-variable components in the horizontal velocities between the two interferograms.:

Section heading 6.5: Rephrase, InSAR does not see stress field variations. It sees related strain rate patterns.

This entire section starts be reiterating basic concepts of differential interferometry. This should be moved to the methods.

L430 “bulk contribution” seems like an unfortunate term. Define it as “the time-invariant contribution” or something like that.

L 470 Differential interferometry does not “isolate” the signal caused by rifting activity. As stated in the paragraphs before this differential signal still contains other processes (e.g. differential tides). Maybe “amplify”?

L 473 The first interferogram sees the “strain” the second “the stress field” and the third the “fringe pattern”. This is another example where I urge you to use consistent terminology. Please unify the discrepancy of “stress”, “strain”, and “strain rates” throughout the manuscript. Those terms cannot be used interchangeably. I don't think InSAR can pick up any stress patterns.

L 480 “..the observed phase ramp could either correspond to a vertical deformation or to a horizontal displacement..”. I agree, but what about the deformational component that you introduced in eq (1,12)?

L486f Use “the differential phase” when you refer to double differenced interferograms.

L502 “dominates..” -> “..that is larger than the differential tidal displacement in the vertical.”

L526 “iceberg drift”? Icebergs are fully disconnected from the ice sheet. I don’t think this is the case here. Rephrase.

Acknowledgements: Consider thanking the reviewers.

Hide

AR by Ludivine Libert on behalf of the Authors (24 Mar 2022) Author's response Author's tracked changes Manuscript

ED: Publish as is (03 Apr 2022) by Reinhard Drews

AR by Ludivine Libert on behalf of the Authors (05 Apr 2022)

Short summary

Open fractures are important to monitor because they weaken the ice shelf structure. We propose a novel approach using synthetic aperture radar (SAR) interferometry for automatic delineation of ice shelf cracks. The method is applied to Sentinel-1 images of Brunt Ice Shelf, Antarctica, and the propagation of the North Rift, which led to iceberg calving in February 2021, is traced. It is also shown that SAR interferometry is more sensitive to rifting than SAR backscatter and optical imagery.