the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Jun 2025
Characterizing Nearshore Icebergs in front of the Dalk Glacier, East Antarctica by UAV Observation
Abstract. Icebergs are the products of glacier calving, and they float on the ocean, influencing ocean circulation and maritime activities. Extensive research has been carried out on the distribution and shapes of icebergs. However, current research mainly focuses on medium and large icebergs. Smaller icebergs, though numerous, are less studied due to the difficulty of detecting them in satellite imagery. In our research, high-resolution images obtained by an Unmanned Aerial Vehicle (UAV) during the Chinese 36th Antarctic expedition were used for the study of smaller icebergs. We extracted nearshore icebergs in front of Dalk Glacier using a method combining superpixel segmentation and Random Forest classification. We directly calculated the area and freeboard of these icebergs by combining Digital Surface Model (DSM) and further calculated their volume. Our research found that DSM generated without control points exhibit a dome effect, and fitting with a surface effectively mitigates this error. Our research identified 187 icebergs, whose area follows a power law distribution with a slope of -1.13. The area/volume relationships obtained from the UAV survey align surprisingly well with existing iceberg parameterization in large-scale ocean models, which was firstly proved to also be valid in growler and berg bit scale by observation evidence. Our research provides new data and insights into the distribution and geometry of small icebergs. Our research reveals the capability of UAVs in extracting iceberg geometric features, highlighting their advantages and discussing the potential and challenges of using UAVs in polar iceberg research.
- Preprint
(14348 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2025-1884', Anonymous Referee #1, 24 Jul 2025
Thank you for inviting me to review this work, which I, unfortunately, do not recommend for publication or revisions. M. Nong and colleagues apply a segmentation algorithm and machine learning to UAV-derived DOM and DSM data acquired at the front of Dalk Glacier to estimate statistical distributions of iceberg areas, and the relationship to iceberg volumes. The study seems overall technically sound. It contains some valuable figures that support understanding. The study generally addresses the valid need for high-resolution reference data to validate satellite retrievals of iceberg properties, and to calibrate/validate ice/ocean models. While the study has potential, the results lack a critical and quantiative evaluation, in particular, with respect to the low number of icebergs analyzed, and concerning the reliability of the calculated power law slope. The manuscript also comes short in critically discussing the results in the context of the broader literature. Moreover, while some connection to an ocean model parameterization is established, a direct comparison to satellite data is neither done, nor discussed, leaving the scientific impact of the study overall vague. Besides, the manuscript lacks clarity and conciseness, particularly in the Abstract, Introduction, Discussion, and Conclusions. While the analyzed data are novel and valuable, the manuscript lacks scientific depth and a critical discussion, and has shortcomings in writing and presentation. Please find some detailed comments below.
Abstract: The abstract requires more quantitative results (size range of icebergs, reliability of power law fitting, e.g.), and a hint to the critical reflection of the presented results in the context of existing research. In its current form, the abstract does not explain the impact of the presented work.
Introduction: The introduction contains broad information of low relevance to the presented study. It should be restructured and streamlined, and summarize the state of the research to highlight the gap addressed by the presented study. Very old work can be mentioned if of high relevance, but with the evolution of sensors and methods, such older work does not seem to be crucial here.
Language: Use consistent past tense for work that was done in this study or in previous studies, e.g. L36 should read "Barbat (...) achieved", L167 should read "was used". Many other inconsistent usages appear throughout the manuscript. It is sensible to use present tense when describing algorithmic mechanisms. Please ensure overall consistency. The language is mostly grammatically correct, but writing style varies throughout the manuscript. The introduction contains many seemingly inflated sentences that contain phrases of low information content, such as, "notably" (L32), "critical insights" (L42), "an important asepct of iceberg research" (L49) - more could be found. Conciseness and consistency of language should be improved throughout the manuscript. In many instances, unspecific verbs like "explore" are used. Please ensure to precisely write what was done (e.g. "we quantified the (...) by (...)").
L9: The abstract should reflect, to a degree, the critical discussion of the reliability of the estimated power law slope - see further comments below.
L9: In the abstract, four sentences start with "Our research". Please rephrase the abstract for better readability.
L10: Please see the comment on this statement concerning L402.
L12: The novelty and impact of this study remain unclear, as "new data and insights" are rather unspecific. Please explain precisely how the study progresses the field in the context of the wider challenges in Antarctic ocean and ice research.
L12: While capabilities of UAVs seem generally undoubted, in my view, the manuscript does not prove and explain how the presented way of using the data specifically progresses "polar iceberg research". How is a set apart from very high resolution satellite data, datasets like ArcticDEM (see Shiggins et al. (2023), currently not cited in the manuscript), or the new SWOT data? How can the very high resolution of the measurement be leveraged to validate satellite remote sensing retrievals or ocean/glacier models? These aspects should be discussed in discussion and briefly reflected in the abstract.
L45: Related studies are mentioned, but Shiggins et al. (2023), TC, is missing, although it is strongly related and highly recent. It should be referred to in several instances, at least in the Introduction, in Table 4, and in the Discussion.
Figure 1: As a higher-resolution alternative to Landsat, was Sentinel-2 imagery available within a reasonable time before/after the UAV flight?
Figure 1: "Detailed" should read "detailed".Section 3.3: Although based on different data, the segmentation/classification methodology of this study is similar to Barbat et al. (2019), as super-pixel segmentation with subsequent classification based on features describing the super-pixel texture was employed in both studies. The authors should also cite Barbat et al. (2019) in section 3.3 (not only in the introduction), including any other relevant source and inspiration, and clearly highlight novel ideas and deviations from existing methodology.
Section 3.3: Why were no geometrical features used in the random forest classification, although done by Barbat et al. (2019)?
L198: This does not seem grammatically correct to me: "(...) and quantity statistical distribution across different size scales". Do the authors mean "quantify the statistical (...)"? Please correct/clarify.
L200: This does not seem grammatically correct to me: "(...) in varies areas". Apart from that, this sentence needs clarification.
Table 1: The size classification is widely known and citing it is sufficient. It does not need to be presented in the paper.
Figure 6: Consider mentioning why the x-marked iceberg was excluded. Consider writing "analysis" instead of "exploration", also in L245.
L246: Sub-section 4.3 requires a heading that specifies what type of iceberg statistics are analyzed (e.g. size statistics).
L248: The statistical summary of the iceberg population should be included in the abstract and in incorporated into the conclusions.
Figure 7: Subplots (a) and (b) should have labels on the x-axis, as they don't align with (c).
Figure 10: How was the 95% confidence interval calculated and what does it imply? What are confidence intervals only shown in log-log space (Figure 10b), not in linear space (Figure 10a)?
Table 4: "Slope" should be spelled "slope" and "power" should be capitalized. Consider writing "Power law slope". Please ensure consistent terminology; in the current manuscript, "slope" is used interchangeably with "exponent".
L287: Please discuss how the power law slope may be, potentially, biased by the data/methods used, also see the below comment regarding L292.
L288: The possessive marker in "Stern et al. (2016)’s" may disrupt reading flow. Consider rephrasing the sentence to avoid a marker behind (2016).
L289: Please discuss the power law slope in more depth, in particular, cite the "existing studies" (L290) here, discuss any aspects that could explain discrepancies between the obtained power law slopes. Specifically, figure 8 seems to show that a large share of small icebergs, indeed, seem to be missed by the segmentation. This would be an issue, since the value of such local rarely acquired high-resolution UAV data is its spatial resolution. However, if the applied segmentation method does not exploit the dataset's full potential, the use, e.g. for validating lower-resolution satellite data or ocean models is rather limited. This appears to be a limitation in the applied methodology and/or in the data themselves, which should be clearly discussed when interpreting the results.
L290: The open-ocean iceberg size distribution cannot be compared to the distribution found in the presented study.
L292: Here, a problem related to the previous comment is discussed: "This may lead to small icebergs being misclassified as sea ice". This is likely to affect the power law slope calculation. However, the problem is only discussed on a general basis. The implications for this study should be clarified in the text. As a reader, I would like to understand how biased the calculated power slopes may be due to confusion with sea ice, and possibly other issues.
L298: The issue of temporally varying iceberg populations is discussed, but this is currently not clearly discussed as a limitation of the presented study. The limitation of the single-time acquisition of UAV data, though, in favor of high spatial resolution, should be discussed.
L304-309: How is the ocean current relevant in the context of the presented study? Additionally, the statement on estimating freshwater fluxes in Greenland seems out of context. It can be reasonably expected that many researchers would disagree with the statement that estimating freshwater fluxes is straight forward in Greenland's fjords. Additionally, ocean forcing is relevant in those fjords too. Please consider erasing or strongly rephrasing this sub-section, and, more importantly, clearly explain how the mentioned aspects affect the methods/results/implications of the presented study.
L311-326: Power laws do not require basic explanation in my view, in particular not in the discussion of the paper. In addition, the sentence in L318 is not a complete sentence and thus needs rephrasing. The sub-section seems inflated, includes empty statements such as "it is reasonable to explore whether similar power law relationships can be applied to icebergs", and the message, implication, relevance in the context of the presented study is unclear to me. Consider erasing or strongly re-writing this sub-section.
L342: Here the authors mention that the study focuses on icebergs smaller than 30 m2 in contrast to previous studies. However, earlier in the manuscript the authors explained why icebergs smaller than 30 m2 were excluded in the study. Overall, the problem might rather be the low number of icebergs. This is somewhat discussed in the next lines, but it remains unclear whether "small iceberg population" refers to small icebergs or a low number of icebergs. The low number of icebergs is an issue for reliably fitting a power law. We also recognize the especially low number of samples at the top of the size range in figure 9, impeding the ability to reliably estimate a power law slope. Please explain your power law fitting method in the methods section and report quantitative metrics expressing the reliability of the fit. Consider excluding the largest sparse samples in the power law slope estimation.
L399-400: The authors mention "significant variation (...) in (...) length and freeboard". However, it was partly discussed and shown by figure 9 that a narrow size range and a low total number of icebergs was analyzed. This should be reflected in the conclusions too.
L402: The authors mention that the alignment with existing parameterizations was surprisingly good. At all instances where this is mentioned in the manuscript, including the abstract, more explanation is needed why this finding is surprising. Additionally, I would advise rephrasing the statement without calling the finding "surprising", while still explaining why the finding was, possibly, unexpected.
L405: While the value of very high resolution imagery is undoubted, it remains unclear to me how this study actually proves advantages over other datasets for the given application. If writing about advantages, it should be explained what quantiative metric they are backed by, and over what the used data have advantages (over high-resolution satellite data?).
L406: When mentioning "small icebergs" the size range of icebergs analyzed in this study should be specified, not only in the conclusions, but also in the abstract.
L406: The DOM and DSMs are no direct measurements of freeboard and volume in my understanding. First, UAV data are remote sensing measurements, although operated closer to the ground than airplanes or satellites. Second, deriving the iceberg area, freeboard, volume requires a significant amount of additional processing and assumptions, as presented in the manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-1884-RC1 -
RC2: 'Comment on egusphere-2025-1884', Anonymous Referee #2, 15 Aug 2025
General comments:
The authors of this study present work to generate a digital orthophoto map (DOM) and digital surface model (DSM) pair through structure from motion processing of UAV-acquired photos of icebergs and sea ice in Dalkoybukta Bay at the terminus of Dalk Glacier (east Antarctica). While more information is required regarding the photo processing, the study is strong in its data collection and correction for surface distortions (the “dome effect”) in the SfM product caused by a lack of ground control points. The authors go on to extract dimensional information for small icebergs to fit a power law model to the iceberg size-frequency distribution following a segmentation and random forest classification workflow.
I suggest that the paper be reconsidered after major revisions. While the applied methods and techniques seem generally suitable, the study is greatly limited by its small aerial-photo dataset and single DOM/DSM pair. The statements made about the applicability or significance of the results are poorly substantiated as a result.
The authors may consider re-focusing the paper on novel aspects of the data collection and processing without the use of ground-control points. This seems like an important conversation that the study can contribute to, given the data on hand and what is needed for scaling-up the use of UAVs for data collection in remote and environmentally challenging locations. Examples of potential investigations using this UAV data can then be incorporated but not focused on.
In addition, the work can be improved by connecting to more relevant and recent literature and fully discussing errors and limitations. A list of potentially useful literature is included at the end of this review, though some of these entries will not be relevant if the study refocuses on a particular topic.
General, minor comments:
Please remember to introduce acronyms where they are first used.
It seems that Figures 3 and 4 can be moved to an appropriate location in the Results. I understand how they were placed in their current locations, but it seems this move would allow the workflow and results to be presented in sequence.
Data sources should be cited in table captions.
Specific comments:
L7: I think a step is missing in the workflow that is referred to with “combining Digital Surface Model (DSM)”.
L8: Please add a little more detail regarding what was done to fit a surface to the DSMs.
L11: It would help to rephrase this last portion of the sentence. It is difficult to understand when this was “firstly provided” and with what observational evidence.
Introduction: It seems that this section jumps between topics, and it is difficult for the reader to follow the links between paragraphs.
L18: Based on the following sentence, it seems appropriate to divide this sentence so that the multiple impacts on the surrounding environment is not linked to the iceberg size differences. It doesn’t seem like the authors are looking to make the link between size and environmental impact here.
L22: I suggest removing “the drifting trajectories of icebergs obstruct navigation” and elaborate more on what maritime risks are increased with the presence of icebergs.
L26-64: It seems that there are many jumps in topic through this section, and it is difficult for the reader to follow the links between paragraphs. It would be valuable for the authors to revisit this section to consider a) which pieces of information are needed to identify knowledge/capability gaps and demonstrate the importance of filling these, and b) how this text can be best structured to build these arguments and ensure the reader follows the progression of ideas and information.
It also seems that some less-relevant information (e.g., some details related to megabergs) can be replaced by integrating reference to recent literature relevant to iceberg distributions, iceberg – ocean modelling, and iceberg surveying. A non-exhaustive list of some suggested literature is included at the bottom of the review.
L40: What size thresholds are being used to define small, medium and large icebergs?
L42: It would be helpful to rephrase “iceberg area decrease pathways”. What exactly is meant here?
L54-64: This section in particularly jumps between many different topics. It would be helpful to consider the flow of information here and likely split this text into multiple paragraphs.
L96: “The front of the glacier grows or calves periodically” can be rephrased to reflect that the glacier terminus advances and retreats, with calving contributing to that retreat.
L101-103: This seems a helpful statement to place somewhere in the Introduction.
L107: “over Dalk Glacier from a base at nearby Zhongshan Station”?
L116: Why were these 119 images from the first sortie selected? Would it be possible to split these into further subsets to more robustly calibrate and validate your segmentation and classification later on?
Figure 1: This is a well-presented figure. The final sentence of the caption can be removed as the point is made in Section 3.1, where it is more appropriately placed.
L121: Please mention, in general, what post-processing steps are conducted.
Section 3.1: This section can be expanded to include more details, including in relation to the use of positional data and the settings applied during photo stitching with SfM. Some information may also be captured in tabular form.
Figure 2: It would be helpful to have acronyms defined in the caption. Any relevant steps or settings related to the post-processing leading to the UAV DOM should also be included in Section 3.1.
Figure 3: This is very minor, but the DEM acronym in panel (a) does not match with what is used in the text/caption (DSM).
L148-150: Can this assumption be supported with other available data?
L159: I feel that the reader needs a little more information on how high-resolution UAV imagery may generate redundancy and interfere with the classification.
L160: The text “advantages of high resolution of the image” can be modified to “high resolution imagery”.
L160-163: This can be broken into multiple sentences for clarity.
L186: It would be valuable to cite the original IIP documentation.
L210: Fracture will be a dominate way that icebergs change shape, though this can occur because of melt.
L228-229: This statement can be moved to the discussion, supported quantitatively, and incorporate discussion of other studies.
L230: This statement can be elaborated on with select results from Table 3.
Table 3: The column headings seem to be offset? Is a column missing as well?
Section 4.2. The performance of the segmentation and classification workflow can be further evaluated. The use of a single DSM in the study likely impacts what performance analysis can be undertaken. Some of the sentences over L230-L234 may be better placed in a Discussion section, and the last paragraph of Section 4.2 would seem to fit better as the opener to Section 4.3.
Section 4.3: Please consider the precision of the iceberg dimension values that you are reporting. Are the values reported appropriate given DSM resolution and other potential error sources?
L255-256: It seems this last sentence can be removed from this location. Some focused consideration of shape variations could be included in the Discussion.
L258: It would be helpful to refer to the “left side of the image”. Alternatively, a compass could be included with each relevant figure, allowing for reference to cardinal/ordinal directions.
L258-268: Please consider what information is appropriate for the Results and what can be moved to and elaborated on in the Discussion.
L259-260: The middle sentence can be broken into two and can include further quantified details. It would be helpful to refer to the iceberg-size categories in a consistent manner.
L262-265: Comparison to other work can be saved to the Discussion, where there is already a dedicated section to the comparison of iceberg size distributions.
Figure 8: The caption can include a citation of the IIP classification scheme.
Section 4.4: The fit of the power law needs to be quantified, as does comparison with power-law distributions used elsewhere. Please re-state what distributions / parameterisations you are comparing to.
Figure 10: What ocean model is being worked with here? Please refer to where the size classes are derived from again. Has the 95 % confidence interval for the UAV survey data been discussed elsewhere? It could with help with visibility to thicken the yellow and grey shading lines.
L278-279: It seems that a more robust analysis or consideration of geography and previous observations would be needed to state that these icebergs’ capsizing could pose thread to Zhongshan Station?
Table 4: Units should be consistent within a given column. It looks like a population size and not iceberg size range is given for the Lu et al. (2013) row. Please check the details within the Åström et al. (2014) row.
L284-291: Please describe further how the number of categories impacts the power-law exponent. It would be helpful to have a more full and detailed comparison to the power-law exponents found or used in other studies. Which iceberg populations would you expect to have similar power law exponents, and is that found with this study’s results?
L292-295: The impact of sea-ice on iceberg classification needs to be quantified. Error introduced throughout the workflow documented in Figure 2 need to be reported. How does this error impact the reported power-law exponent?
L298-304: How will these different scenarios and deterioration processes impact on the power-law exponent? What does the possibility of fluctuating size distributions mean for this study’s methodology and results? What are the limitations of this study, and what do the authors suggest to make future studies more robust so that the methods or reported power law can be used more widely?
L304-309: I suggest removing the consideration of Greenland fiords and focus on how the ocean currents in the vicinity of Dalk Glacier may impact iceberg populations there, as well as any potentially impact on this study’s presented results.
L311-326: This background on the use of power laws throughout glaciology is unnecessary. Focusing on the applicability of a power law to iceberg distributions and iceberg research would be more helpful.
L331-334: It would be helpful to have the iceberg size ranges of these different iceberg populations continually reported. I would expect the Greenland icebergs to be considered small in the Antarctic context of this study. If that’s the case, then the contribution of this study needs to be re-worded and further contextualised.
L338: It is hard to understand what is being stated in the paragraph’s last sentence.
L340-342: The locational difference can be further explained. Please be explicit about which iceberg population is being referred to throughout. Please reconsider the statement within the sentence starting, “Small icebergs are more susceptible to…”, including small icebergs being more susceptible to ocean currents and this leading to enhanced fragmentation. See Kirkham et al. (2017, Scientific Reports) and Crawford et al. (2018, Journal of Geophysical Research: Oceans) for discussion of how dominant deterioration processes for different iceberg sizes will impact size distributions.
L344-345: For what size range is this study’s power law appropriate?
Section 5.3: The writing in this section is relatively strong, particularly in the discussion of environmental constraints on the methodology, impacts of limited control points, and correcting for the dome effect.
L400: I suggest removing, “with significant variation in length and freeboard”.
L403-405: While it may be appropriate to say that the geometrical data led to the extension of the iceberg size parameterization scheme to smaller iceberg sizes, it doesn’t seem that this work can validate an ocean model parameterization based on larger icebergs given the gaps in size classes observed in this study.
Suggested literature:
Åström, J., Cook, S., Enderlin, E.M., Sutherland, D.A., Mazur, A., Glasser, N., 2021. Fragmentation theory reveals processes controlling iceberg size distributions. J. Glaciol. 67, 603–612. https://doi.org/10.1017/jog.2021.14
Barbat, M.M., Rackow, T., Wesche, C., Hellmer, H.H., Mata, M.M., 2021. Automated iceberg tracking with a machine learning approach applied to SAR imagery: A Weddell sea case study. ISPRS Journal of Photogrammetry and Remote Sensing 172, 189–206. https://doi.org/10.1016/j.isprsjprs.2020.12.006
Barbat, M.M., Wesche, C., Werhli, A.V., Mata, M.M., 2019. An adaptive machine learning approach to improve automatic iceberg detection from SAR images. ISPRS Journal of Photogrammetry and Remote Sensing 156, 247–259. https://doi.org/10.1016/j.isprsjprs.2019.08.015
Braakmann-Folgmann, A., Shepherd, A., Hogg, D., Redmond, E., 2023. Mapping the extent of giant Antarctic icebergs with deep learning. The Cryosphere 17, 4675–4690. https://doi.org/10.5194/tc-17-4675-2023
Crawford, A.J., Mueller, D., Desjardins, L., Myers, P.G., 2018. The aftermath of Petermann Glacier calving events (2008-2012): Ice island size distributions and meltwater dispersal. Journal of Geophysical Research: Oceans 8812–8827. https://doi.org/10.1029/2018JC014388
Crawford, A., Mueller, D., Joyal, G., 2018. Surveying Drifting Icebergs and Ice Islands: Deterioration Detection and Mass Estimation with Aerial Photogrammetry and Laser Scanning. Remote Sensing 10, 575. https://doi.org/10.3390/rs10040575
Evans, B., Faul, A., Fleming, A., Vaughan, D.G., Hosking, J.S., 2023. Unsupervised machine learning detection of iceberg populations within sea ice from dual-polarisation SAR imagery. Remote Sensing of Environment 297, 113780. https://doi.org/10.1016/j.rse.2023.113780
Kirkham, J.D., Rosser, N.J., Wainwright, J., Vann Jones, E.C., Dunning, S.A., Lane, V.S., Hawthorn, D.E., Strzelecki, M.C., Szczuciński, W., 2017. Drift-dependent changes in iceberg size-frequency distributions. Scientific Reports 7. https://doi.org/10.1038/s41598-017-14863-2
Koo, Y., Xie, H., Mahmoud, H., Iqrah, J.M., Ackley, S.F., 2023. Automated detection and tracking of medium-large icebergs from Sentinel-1 imagery using Google Earth Engine. Remote Sensing of Environment 296, 113731. https://doi.org/10.1016/j.rse.2023.113731
Mazur, A.K., Wåhlin, A.K., Krężel, A., 2017. An object-based SAR image iceberg detection algorithm applied to the Amundsen Sea. Remote Sensing of Environment 189, 67–83. https://doi.org/10.1016/j.rse.2016.11.013
Shiggins, C.J., Lea, J.M., Brough, S., 2023. Automated ArcticDEM iceberg detection tool: insights into area and volume distributions, and their potential application to satellite imagery and modelling of glacier–iceberg–ocean systems. The Cryosphere 17, 15–32. https://doi.org/10.5194/tc-17-15-2023
Wagner, T.J.W., Eisenman, I., 2017. How climate model biases skew the distribution of iceberg meltwater. Geophysical Research Letters 44, 3691–3699. https://doi.org/10.1002/2016GL071645
Wesche, C., Dierking, W., 2015. Near-coastal circum-Antarctic iceberg size distributions determined from Synthetic Aperture Radar images. Remote Sensing of Environment 156, 561–569. https://doi.org/10.1016/j.rse.2014.10.025
Citation: https://doi.org/10.5194/egusphere-2025-1884-RC2
Status: closed
-
RC1: 'Comment on egusphere-2025-1884', Anonymous Referee #1, 24 Jul 2025
Thank you for inviting me to review this work, which I, unfortunately, do not recommend for publication or revisions. M. Nong and colleagues apply a segmentation algorithm and machine learning to UAV-derived DOM and DSM data acquired at the front of Dalk Glacier to estimate statistical distributions of iceberg areas, and the relationship to iceberg volumes. The study seems overall technically sound. It contains some valuable figures that support understanding. The study generally addresses the valid need for high-resolution reference data to validate satellite retrievals of iceberg properties, and to calibrate/validate ice/ocean models. While the study has potential, the results lack a critical and quantiative evaluation, in particular, with respect to the low number of icebergs analyzed, and concerning the reliability of the calculated power law slope. The manuscript also comes short in critically discussing the results in the context of the broader literature. Moreover, while some connection to an ocean model parameterization is established, a direct comparison to satellite data is neither done, nor discussed, leaving the scientific impact of the study overall vague. Besides, the manuscript lacks clarity and conciseness, particularly in the Abstract, Introduction, Discussion, and Conclusions. While the analyzed data are novel and valuable, the manuscript lacks scientific depth and a critical discussion, and has shortcomings in writing and presentation. Please find some detailed comments below.
Abstract: The abstract requires more quantitative results (size range of icebergs, reliability of power law fitting, e.g.), and a hint to the critical reflection of the presented results in the context of existing research. In its current form, the abstract does not explain the impact of the presented work.
Introduction: The introduction contains broad information of low relevance to the presented study. It should be restructured and streamlined, and summarize the state of the research to highlight the gap addressed by the presented study. Very old work can be mentioned if of high relevance, but with the evolution of sensors and methods, such older work does not seem to be crucial here.
Language: Use consistent past tense for work that was done in this study or in previous studies, e.g. L36 should read "Barbat (...) achieved", L167 should read "was used". Many other inconsistent usages appear throughout the manuscript. It is sensible to use present tense when describing algorithmic mechanisms. Please ensure overall consistency. The language is mostly grammatically correct, but writing style varies throughout the manuscript. The introduction contains many seemingly inflated sentences that contain phrases of low information content, such as, "notably" (L32), "critical insights" (L42), "an important asepct of iceberg research" (L49) - more could be found. Conciseness and consistency of language should be improved throughout the manuscript. In many instances, unspecific verbs like "explore" are used. Please ensure to precisely write what was done (e.g. "we quantified the (...) by (...)").
L9: The abstract should reflect, to a degree, the critical discussion of the reliability of the estimated power law slope - see further comments below.
L9: In the abstract, four sentences start with "Our research". Please rephrase the abstract for better readability.
L10: Please see the comment on this statement concerning L402.
L12: The novelty and impact of this study remain unclear, as "new data and insights" are rather unspecific. Please explain precisely how the study progresses the field in the context of the wider challenges in Antarctic ocean and ice research.
L12: While capabilities of UAVs seem generally undoubted, in my view, the manuscript does not prove and explain how the presented way of using the data specifically progresses "polar iceberg research". How is a set apart from very high resolution satellite data, datasets like ArcticDEM (see Shiggins et al. (2023), currently not cited in the manuscript), or the new SWOT data? How can the very high resolution of the measurement be leveraged to validate satellite remote sensing retrievals or ocean/glacier models? These aspects should be discussed in discussion and briefly reflected in the abstract.
L45: Related studies are mentioned, but Shiggins et al. (2023), TC, is missing, although it is strongly related and highly recent. It should be referred to in several instances, at least in the Introduction, in Table 4, and in the Discussion.
Figure 1: As a higher-resolution alternative to Landsat, was Sentinel-2 imagery available within a reasonable time before/after the UAV flight?
Figure 1: "Detailed" should read "detailed".Section 3.3: Although based on different data, the segmentation/classification methodology of this study is similar to Barbat et al. (2019), as super-pixel segmentation with subsequent classification based on features describing the super-pixel texture was employed in both studies. The authors should also cite Barbat et al. (2019) in section 3.3 (not only in the introduction), including any other relevant source and inspiration, and clearly highlight novel ideas and deviations from existing methodology.
Section 3.3: Why were no geometrical features used in the random forest classification, although done by Barbat et al. (2019)?
L198: This does not seem grammatically correct to me: "(...) and quantity statistical distribution across different size scales". Do the authors mean "quantify the statistical (...)"? Please correct/clarify.
L200: This does not seem grammatically correct to me: "(...) in varies areas". Apart from that, this sentence needs clarification.
Table 1: The size classification is widely known and citing it is sufficient. It does not need to be presented in the paper.
Figure 6: Consider mentioning why the x-marked iceberg was excluded. Consider writing "analysis" instead of "exploration", also in L245.
L246: Sub-section 4.3 requires a heading that specifies what type of iceberg statistics are analyzed (e.g. size statistics).
L248: The statistical summary of the iceberg population should be included in the abstract and in incorporated into the conclusions.
Figure 7: Subplots (a) and (b) should have labels on the x-axis, as they don't align with (c).
Figure 10: How was the 95% confidence interval calculated and what does it imply? What are confidence intervals only shown in log-log space (Figure 10b), not in linear space (Figure 10a)?
Table 4: "Slope" should be spelled "slope" and "power" should be capitalized. Consider writing "Power law slope". Please ensure consistent terminology; in the current manuscript, "slope" is used interchangeably with "exponent".
L287: Please discuss how the power law slope may be, potentially, biased by the data/methods used, also see the below comment regarding L292.
L288: The possessive marker in "Stern et al. (2016)’s" may disrupt reading flow. Consider rephrasing the sentence to avoid a marker behind (2016).
L289: Please discuss the power law slope in more depth, in particular, cite the "existing studies" (L290) here, discuss any aspects that could explain discrepancies between the obtained power law slopes. Specifically, figure 8 seems to show that a large share of small icebergs, indeed, seem to be missed by the segmentation. This would be an issue, since the value of such local rarely acquired high-resolution UAV data is its spatial resolution. However, if the applied segmentation method does not exploit the dataset's full potential, the use, e.g. for validating lower-resolution satellite data or ocean models is rather limited. This appears to be a limitation in the applied methodology and/or in the data themselves, which should be clearly discussed when interpreting the results.
L290: The open-ocean iceberg size distribution cannot be compared to the distribution found in the presented study.
L292: Here, a problem related to the previous comment is discussed: "This may lead to small icebergs being misclassified as sea ice". This is likely to affect the power law slope calculation. However, the problem is only discussed on a general basis. The implications for this study should be clarified in the text. As a reader, I would like to understand how biased the calculated power slopes may be due to confusion with sea ice, and possibly other issues.
L298: The issue of temporally varying iceberg populations is discussed, but this is currently not clearly discussed as a limitation of the presented study. The limitation of the single-time acquisition of UAV data, though, in favor of high spatial resolution, should be discussed.
L304-309: How is the ocean current relevant in the context of the presented study? Additionally, the statement on estimating freshwater fluxes in Greenland seems out of context. It can be reasonably expected that many researchers would disagree with the statement that estimating freshwater fluxes is straight forward in Greenland's fjords. Additionally, ocean forcing is relevant in those fjords too. Please consider erasing or strongly rephrasing this sub-section, and, more importantly, clearly explain how the mentioned aspects affect the methods/results/implications of the presented study.
L311-326: Power laws do not require basic explanation in my view, in particular not in the discussion of the paper. In addition, the sentence in L318 is not a complete sentence and thus needs rephrasing. The sub-section seems inflated, includes empty statements such as "it is reasonable to explore whether similar power law relationships can be applied to icebergs", and the message, implication, relevance in the context of the presented study is unclear to me. Consider erasing or strongly re-writing this sub-section.
L342: Here the authors mention that the study focuses on icebergs smaller than 30 m2 in contrast to previous studies. However, earlier in the manuscript the authors explained why icebergs smaller than 30 m2 were excluded in the study. Overall, the problem might rather be the low number of icebergs. This is somewhat discussed in the next lines, but it remains unclear whether "small iceberg population" refers to small icebergs or a low number of icebergs. The low number of icebergs is an issue for reliably fitting a power law. We also recognize the especially low number of samples at the top of the size range in figure 9, impeding the ability to reliably estimate a power law slope. Please explain your power law fitting method in the methods section and report quantitative metrics expressing the reliability of the fit. Consider excluding the largest sparse samples in the power law slope estimation.
L399-400: The authors mention "significant variation (...) in (...) length and freeboard". However, it was partly discussed and shown by figure 9 that a narrow size range and a low total number of icebergs was analyzed. This should be reflected in the conclusions too.
L402: The authors mention that the alignment with existing parameterizations was surprisingly good. At all instances where this is mentioned in the manuscript, including the abstract, more explanation is needed why this finding is surprising. Additionally, I would advise rephrasing the statement without calling the finding "surprising", while still explaining why the finding was, possibly, unexpected.
L405: While the value of very high resolution imagery is undoubted, it remains unclear to me how this study actually proves advantages over other datasets for the given application. If writing about advantages, it should be explained what quantiative metric they are backed by, and over what the used data have advantages (over high-resolution satellite data?).
L406: When mentioning "small icebergs" the size range of icebergs analyzed in this study should be specified, not only in the conclusions, but also in the abstract.
L406: The DOM and DSMs are no direct measurements of freeboard and volume in my understanding. First, UAV data are remote sensing measurements, although operated closer to the ground than airplanes or satellites. Second, deriving the iceberg area, freeboard, volume requires a significant amount of additional processing and assumptions, as presented in the manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-1884-RC1 -
RC2: 'Comment on egusphere-2025-1884', Anonymous Referee #2, 15 Aug 2025
General comments:
The authors of this study present work to generate a digital orthophoto map (DOM) and digital surface model (DSM) pair through structure from motion processing of UAV-acquired photos of icebergs and sea ice in Dalkoybukta Bay at the terminus of Dalk Glacier (east Antarctica). While more information is required regarding the photo processing, the study is strong in its data collection and correction for surface distortions (the “dome effect”) in the SfM product caused by a lack of ground control points. The authors go on to extract dimensional information for small icebergs to fit a power law model to the iceberg size-frequency distribution following a segmentation and random forest classification workflow.
I suggest that the paper be reconsidered after major revisions. While the applied methods and techniques seem generally suitable, the study is greatly limited by its small aerial-photo dataset and single DOM/DSM pair. The statements made about the applicability or significance of the results are poorly substantiated as a result.
The authors may consider re-focusing the paper on novel aspects of the data collection and processing without the use of ground-control points. This seems like an important conversation that the study can contribute to, given the data on hand and what is needed for scaling-up the use of UAVs for data collection in remote and environmentally challenging locations. Examples of potential investigations using this UAV data can then be incorporated but not focused on.
In addition, the work can be improved by connecting to more relevant and recent literature and fully discussing errors and limitations. A list of potentially useful literature is included at the end of this review, though some of these entries will not be relevant if the study refocuses on a particular topic.
General, minor comments:
Please remember to introduce acronyms where they are first used.
It seems that Figures 3 and 4 can be moved to an appropriate location in the Results. I understand how they were placed in their current locations, but it seems this move would allow the workflow and results to be presented in sequence.
Data sources should be cited in table captions.
Specific comments:
L7: I think a step is missing in the workflow that is referred to with “combining Digital Surface Model (DSM)”.
L8: Please add a little more detail regarding what was done to fit a surface to the DSMs.
L11: It would help to rephrase this last portion of the sentence. It is difficult to understand when this was “firstly provided” and with what observational evidence.
Introduction: It seems that this section jumps between topics, and it is difficult for the reader to follow the links between paragraphs.
L18: Based on the following sentence, it seems appropriate to divide this sentence so that the multiple impacts on the surrounding environment is not linked to the iceberg size differences. It doesn’t seem like the authors are looking to make the link between size and environmental impact here.
L22: I suggest removing “the drifting trajectories of icebergs obstruct navigation” and elaborate more on what maritime risks are increased with the presence of icebergs.
L26-64: It seems that there are many jumps in topic through this section, and it is difficult for the reader to follow the links between paragraphs. It would be valuable for the authors to revisit this section to consider a) which pieces of information are needed to identify knowledge/capability gaps and demonstrate the importance of filling these, and b) how this text can be best structured to build these arguments and ensure the reader follows the progression of ideas and information.
It also seems that some less-relevant information (e.g., some details related to megabergs) can be replaced by integrating reference to recent literature relevant to iceberg distributions, iceberg – ocean modelling, and iceberg surveying. A non-exhaustive list of some suggested literature is included at the bottom of the review.
L40: What size thresholds are being used to define small, medium and large icebergs?
L42: It would be helpful to rephrase “iceberg area decrease pathways”. What exactly is meant here?
L54-64: This section in particularly jumps between many different topics. It would be helpful to consider the flow of information here and likely split this text into multiple paragraphs.
L96: “The front of the glacier grows or calves periodically” can be rephrased to reflect that the glacier terminus advances and retreats, with calving contributing to that retreat.
L101-103: This seems a helpful statement to place somewhere in the Introduction.
L107: “over Dalk Glacier from a base at nearby Zhongshan Station”?
L116: Why were these 119 images from the first sortie selected? Would it be possible to split these into further subsets to more robustly calibrate and validate your segmentation and classification later on?
Figure 1: This is a well-presented figure. The final sentence of the caption can be removed as the point is made in Section 3.1, where it is more appropriately placed.
L121: Please mention, in general, what post-processing steps are conducted.
Section 3.1: This section can be expanded to include more details, including in relation to the use of positional data and the settings applied during photo stitching with SfM. Some information may also be captured in tabular form.
Figure 2: It would be helpful to have acronyms defined in the caption. Any relevant steps or settings related to the post-processing leading to the UAV DOM should also be included in Section 3.1.
Figure 3: This is very minor, but the DEM acronym in panel (a) does not match with what is used in the text/caption (DSM).
L148-150: Can this assumption be supported with other available data?
L159: I feel that the reader needs a little more information on how high-resolution UAV imagery may generate redundancy and interfere with the classification.
L160: The text “advantages of high resolution of the image” can be modified to “high resolution imagery”.
L160-163: This can be broken into multiple sentences for clarity.
L186: It would be valuable to cite the original IIP documentation.
L210: Fracture will be a dominate way that icebergs change shape, though this can occur because of melt.
L228-229: This statement can be moved to the discussion, supported quantitatively, and incorporate discussion of other studies.
L230: This statement can be elaborated on with select results from Table 3.
Table 3: The column headings seem to be offset? Is a column missing as well?
Section 4.2. The performance of the segmentation and classification workflow can be further evaluated. The use of a single DSM in the study likely impacts what performance analysis can be undertaken. Some of the sentences over L230-L234 may be better placed in a Discussion section, and the last paragraph of Section 4.2 would seem to fit better as the opener to Section 4.3.
Section 4.3: Please consider the precision of the iceberg dimension values that you are reporting. Are the values reported appropriate given DSM resolution and other potential error sources?
L255-256: It seems this last sentence can be removed from this location. Some focused consideration of shape variations could be included in the Discussion.
L258: It would be helpful to refer to the “left side of the image”. Alternatively, a compass could be included with each relevant figure, allowing for reference to cardinal/ordinal directions.
L258-268: Please consider what information is appropriate for the Results and what can be moved to and elaborated on in the Discussion.
L259-260: The middle sentence can be broken into two and can include further quantified details. It would be helpful to refer to the iceberg-size categories in a consistent manner.
L262-265: Comparison to other work can be saved to the Discussion, where there is already a dedicated section to the comparison of iceberg size distributions.
Figure 8: The caption can include a citation of the IIP classification scheme.
Section 4.4: The fit of the power law needs to be quantified, as does comparison with power-law distributions used elsewhere. Please re-state what distributions / parameterisations you are comparing to.
Figure 10: What ocean model is being worked with here? Please refer to where the size classes are derived from again. Has the 95 % confidence interval for the UAV survey data been discussed elsewhere? It could with help with visibility to thicken the yellow and grey shading lines.
L278-279: It seems that a more robust analysis or consideration of geography and previous observations would be needed to state that these icebergs’ capsizing could pose thread to Zhongshan Station?
Table 4: Units should be consistent within a given column. It looks like a population size and not iceberg size range is given for the Lu et al. (2013) row. Please check the details within the Åström et al. (2014) row.
L284-291: Please describe further how the number of categories impacts the power-law exponent. It would be helpful to have a more full and detailed comparison to the power-law exponents found or used in other studies. Which iceberg populations would you expect to have similar power law exponents, and is that found with this study’s results?
L292-295: The impact of sea-ice on iceberg classification needs to be quantified. Error introduced throughout the workflow documented in Figure 2 need to be reported. How does this error impact the reported power-law exponent?
L298-304: How will these different scenarios and deterioration processes impact on the power-law exponent? What does the possibility of fluctuating size distributions mean for this study’s methodology and results? What are the limitations of this study, and what do the authors suggest to make future studies more robust so that the methods or reported power law can be used more widely?
L304-309: I suggest removing the consideration of Greenland fiords and focus on how the ocean currents in the vicinity of Dalk Glacier may impact iceberg populations there, as well as any potentially impact on this study’s presented results.
L311-326: This background on the use of power laws throughout glaciology is unnecessary. Focusing on the applicability of a power law to iceberg distributions and iceberg research would be more helpful.
L331-334: It would be helpful to have the iceberg size ranges of these different iceberg populations continually reported. I would expect the Greenland icebergs to be considered small in the Antarctic context of this study. If that’s the case, then the contribution of this study needs to be re-worded and further contextualised.
L338: It is hard to understand what is being stated in the paragraph’s last sentence.
L340-342: The locational difference can be further explained. Please be explicit about which iceberg population is being referred to throughout. Please reconsider the statement within the sentence starting, “Small icebergs are more susceptible to…”, including small icebergs being more susceptible to ocean currents and this leading to enhanced fragmentation. See Kirkham et al. (2017, Scientific Reports) and Crawford et al. (2018, Journal of Geophysical Research: Oceans) for discussion of how dominant deterioration processes for different iceberg sizes will impact size distributions.
L344-345: For what size range is this study’s power law appropriate?
Section 5.3: The writing in this section is relatively strong, particularly in the discussion of environmental constraints on the methodology, impacts of limited control points, and correcting for the dome effect.
L400: I suggest removing, “with significant variation in length and freeboard”.
L403-405: While it may be appropriate to say that the geometrical data led to the extension of the iceberg size parameterization scheme to smaller iceberg sizes, it doesn’t seem that this work can validate an ocean model parameterization based on larger icebergs given the gaps in size classes observed in this study.
Suggested literature:
Åström, J., Cook, S., Enderlin, E.M., Sutherland, D.A., Mazur, A., Glasser, N., 2021. Fragmentation theory reveals processes controlling iceberg size distributions. J. Glaciol. 67, 603–612. https://doi.org/10.1017/jog.2021.14
Barbat, M.M., Rackow, T., Wesche, C., Hellmer, H.H., Mata, M.M., 2021. Automated iceberg tracking with a machine learning approach applied to SAR imagery: A Weddell sea case study. ISPRS Journal of Photogrammetry and Remote Sensing 172, 189–206. https://doi.org/10.1016/j.isprsjprs.2020.12.006
Barbat, M.M., Wesche, C., Werhli, A.V., Mata, M.M., 2019. An adaptive machine learning approach to improve automatic iceberg detection from SAR images. ISPRS Journal of Photogrammetry and Remote Sensing 156, 247–259. https://doi.org/10.1016/j.isprsjprs.2019.08.015
Braakmann-Folgmann, A., Shepherd, A., Hogg, D., Redmond, E., 2023. Mapping the extent of giant Antarctic icebergs with deep learning. The Cryosphere 17, 4675–4690. https://doi.org/10.5194/tc-17-4675-2023
Crawford, A.J., Mueller, D., Desjardins, L., Myers, P.G., 2018. The aftermath of Petermann Glacier calving events (2008-2012): Ice island size distributions and meltwater dispersal. Journal of Geophysical Research: Oceans 8812–8827. https://doi.org/10.1029/2018JC014388
Crawford, A., Mueller, D., Joyal, G., 2018. Surveying Drifting Icebergs and Ice Islands: Deterioration Detection and Mass Estimation with Aerial Photogrammetry and Laser Scanning. Remote Sensing 10, 575. https://doi.org/10.3390/rs10040575
Evans, B., Faul, A., Fleming, A., Vaughan, D.G., Hosking, J.S., 2023. Unsupervised machine learning detection of iceberg populations within sea ice from dual-polarisation SAR imagery. Remote Sensing of Environment 297, 113780. https://doi.org/10.1016/j.rse.2023.113780
Kirkham, J.D., Rosser, N.J., Wainwright, J., Vann Jones, E.C., Dunning, S.A., Lane, V.S., Hawthorn, D.E., Strzelecki, M.C., Szczuciński, W., 2017. Drift-dependent changes in iceberg size-frequency distributions. Scientific Reports 7. https://doi.org/10.1038/s41598-017-14863-2
Koo, Y., Xie, H., Mahmoud, H., Iqrah, J.M., Ackley, S.F., 2023. Automated detection and tracking of medium-large icebergs from Sentinel-1 imagery using Google Earth Engine. Remote Sensing of Environment 296, 113731. https://doi.org/10.1016/j.rse.2023.113731
Mazur, A.K., Wåhlin, A.K., Krężel, A., 2017. An object-based SAR image iceberg detection algorithm applied to the Amundsen Sea. Remote Sensing of Environment 189, 67–83. https://doi.org/10.1016/j.rse.2016.11.013
Shiggins, C.J., Lea, J.M., Brough, S., 2023. Automated ArcticDEM iceberg detection tool: insights into area and volume distributions, and their potential application to satellite imagery and modelling of glacier–iceberg–ocean systems. The Cryosphere 17, 15–32. https://doi.org/10.5194/tc-17-15-2023
Wagner, T.J.W., Eisenman, I., 2017. How climate model biases skew the distribution of iceberg meltwater. Geophysical Research Letters 44, 3691–3699. https://doi.org/10.1002/2016GL071645
Wesche, C., Dierking, W., 2015. Near-coastal circum-Antarctic iceberg size distributions determined from Synthetic Aperture Radar images. Remote Sensing of Environment 156, 561–569. https://doi.org/10.1016/j.rse.2014.10.025
Citation: https://doi.org/10.5194/egusphere-2025-1884-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 318 | 52 | 11 | 381 | 7 | 23 |
- HTML: 318
- PDF: 52
- XML: 11
- Total: 381
- BibTeX: 7
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1