Review Reports

Round 1

Reviewer 1 Report

This manuscript submitted by Zoklos et al. provides a valuable dataset and important results and approaches to the detection and interpretation of gas emission craters (GECs) in the north of West Siberia. I very much enjoyed reviewing the manuscript. In my view, both, the unique and valuable dataset as well as the state-of-the-art earth observation based change detection technique applied are very useful contributions with applied aspects of geoscientific relevance that warrant rapid publication, according to the journal criteria. While reading I had to withdraw many ideas of content-related suggestions because most of them were addressed later in the text. Overall, I highly recommend the paper for publication after some clarifications have been implemented.

General comments:

You correctly point to the aspect of limited GEC training data due to the sparse occurrence of this phenomenon. The combined use of reflectance and elevation data time series is a very promising approach to detect previously unrecognized GECs in limiting potential change areas to those where NDVI values suddenly drop and elevation changes occurred. However, I would have found it more convincing if the study would not have detected all previously reported GECs. The extensive description of the developed CDA suggests that areas previously covered with water bodies were excluded from further analyses: “To limit change in surface water to within our study period, surface water that was present both in 2018 and before the study period was omitted from the analysis” (lines 172-173) and “CDF-2 is not shown because its land surface change values were omitted by the water mask” (Caption figure 4). Given these constraints, how you have been able for example to automatically detect CDF 2 and SeYkh GEC in particular, which emerged within a river? Moreover, the predecessor mound of the latter was relatively small (< 2m high) and thus within the general elevation uncertainty when comparing ArcticDEM strip data. Does this imply that you still manually checked each change detection product separately, although figure 1 suggests a combined use of all inputs? As SeYkh seems to be difficult to detect, the authors may want to consider to include a showcase figure how the different inputs worked together in decision making for this particular feature.

Specific comments:

Lines 76-77: “GECs may be associated with thermokarst such as retrogressive thaw slump (RTS) features in the vicinity.” The authors several times refer to RTS development and I acknowledge that the CDA proved successful in detecting these features. However, I cannot see a direct link how GEC and RTS development is coupled. If you have other evidence, please expand on this topic.

Lines 263-264: “We validated 28 fishnet cells (4.6% of total), accounting for 21,000 km2 within the study area (Figure A3).” The authors report that validation cells were randomly selected. However, a cluster of six validation cells in the area of the Bovanenkovo gas field, where the first GECs were discovered, suggests a biased and non-random selection of validation cells. In case I did not understand the validation process correctly, please consider to include a schematic figure of the validation process (similar to Fig. 1), because subsection 2.3 is hard to follow for a reader that has not been involved in this process. Line 198: “manually interpreted high-resolution (sub-meter) satellite imagery”, again suggests that validation has been done where high-resolution imagery were available. What is the coverage of this data?

Lines 252-253: “All GECs were located within upland/non-wetland tundra except for SeYkhGEC, which was located within a river in a wetland complex”. Given this evidence, why you selected validation areas “Based on the majority landcover type within each cell, we used random stratified sampling to select 28 cells for validation” (lines 200-201).

Generally, I enjoyed subsection 4.2, because it takes a critical view on the newly identified GECs. However, lines 341-342 “In the absence of a talik, CDF-2 may have originated as a hydraulic (open-system) pingo, provided a sufficient groundwater source [46]” are hard to understand. How an open-system pingo should develop in the absence of a talik? Please clarify.

Line 354: “We surmise it was a degraded pingo” or “CDF‐3 could have originated as a pingo, whose ice rich core thaw and collapsed” (lines 356-357). Please consider the study of Wetterich et al. (2018, https://doi.org/10.1002/ppp.1979), where even a pingo that is being eroded by an adjacent lake cannot quickly transform into a depression like feature.

Lines 397-401: in the context of an outlook, also TerraSAR-X time series can be used to track e.g. RTS development (Zwieback et a l., https://doi.org/10.5194/tc-12-549-2018), and its higher spatial resolution might be more suitable for GEC change detection.

Author Response

Please see the attachment. Thank you!

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript under review is very interesting and can be published in Geosciences. I can only evaluate this work as a geologist, since I am not a remote sensing specialist. The ability to diagnose methane discharge craters using satellite images is very important and promising. The authors identified three new gas emission structures, which is an important result of their research. I believe that the article needs some very minor corrections.

In the caption to table 1, the full name must be added to the abbreviation CDF.

Why didn't you plot the position of the CDF points in Figure 2? This should be done.

In Figures 3, 5-7, you used letters (a, b, c, etc.) to name individual figures. It is advisable to describe in the figure caption each individual figure marked with the corresponding letter. Otherwise, it is not clear why these letters were used for each individual fragment of the figure, if they were not explained in the caption to the figure.

In the references in works 18 and 19, the list of authors contains their affiliation.

The reference to article 28 does not indicate the title of the journal.

Author Response

Please see the attachment. Thank you!

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper is an interesting application of batch processing of the Landsat / ArcticDEM archive for detecting changes in a very dynamic environment. Analyzing the data presented, I found convincing that the algorithm can definitely be used for detecting changes related to permafrost thaw, which can be published in a Geoscience journal. The pixel change values (> 0.5) related to known GECs are statistically different from the sample of other changes which is a valuable result and can be used for finding GEC occurrence further in the future. However, none of three newly found features (CDF-1,2,3) can definitely be classified as GEC, in my opinion. At least two of them (CDF-2,3) can likely be degraded pingos, which is also ‘result’ since pingos are rather unique cryogenic process for the climate of the North of Western Siberia. There were no signs of ejected material found at the area of these features. Moreover, according to pixel change values (Fig. 4), these features can rather be classified as changes not related to GEC, which of course can be associated with their smaller size. Anyway, I found it not convincing, that these features are certainly GECs, and Authors also do not claim that, as I understand. CDF-1 can be whatever.

 

The methodology part of the manuscript can significantly be improved by clarifying types of data used, algorithms of assigning change values for pixels etc (see further). Have Authors tried to analyze elevation changes only with higher (2 m) resolution? I think that could help in finding truly hazardous mound predecessors. Resampling these datasets to 30 m would not help in locating potentially growing objects. Further, extensive validation was performed for the area of ~ 21000 km2. But what are the results of this validation? What are the percentage of correctly classified changes related to permafrost thaw (RTS, lake drainage, thermal erosion etc)? Were there artifacts or changes linked to problems of co-location of scenes, inter-seasonal variability of tundra surface? Would be also positive to see pixel change statistics for other changes related to permafrost thaw.

To my knowledge, a new crater or GEC-like object appeared in Yamal in 2020. Authors could also test the algorithm for 2019-2020 in order to find this new feature. But it’s just a suggestion for potential enrichment of the manuscript with data, and can be ignored.

 

Further comments:

 

P1 L18 *emission craters

 

P1 L36 I think one can just use the term ‘permafrost’

 

P2 L53 Do you imply ‘Kamchatka’?

 

P2 L67 *hundreds of meters

 

P2 L76-77 I would re-phrase this sentence: “Since the important control of GEC occurrence is the presence of tabular ground ice in a geological section, GECs are often associated with abrupt permafrost thaw processes over the TGI bodies such as RTS – specific features found in the North of Western Siberia” or similar.

 

Table 1. CDA data are included before the introduction to the methods used to assess it. I think it would be better to place that part into the section ‘Results’. CFD-1,2,3 have never been introduced before. As I understand, these are new objects found within this research: this should definitely be placed in the Results.

 

P3 L116-118 The annual / summer mean temperatures vary within the study area. Therefore, it isn’t so informative to give the numbers for Salekhard only. There are more meteorological stations such as Se-Yakha, Marre-Sale for Yamal etc. I would suggest presenting more data.

 

P4 L133 I would indicate here the level of processing applied to Landsat data. Did you use a GEE standard available surface reflectance scenes (corrected with LEDAPS / LaSRC algorithms)? Or different correction algorithm has been applied to raw scenes?

 

The description of the algorithm is vague here (I’m not so familiar with LandTrendr). What do you imply by ‘greatest loss in NDVI’? Is this change in the index value by 0.5, 0.6, 0.7? In which case the pixel is classified as 1 in your SR classification? It is important to explain clearly since the NDVI values can vary significantly within the season in Yamal/Gydan at each ‘green pixel’: from zero values at the beginning of the summer season after abrupt snow melt (wet tundra) to 0.5-0.6 in August.

 

Table A1 LandTrendr parameters are not informative within this table. Please specify them in additional column.

 

P5 L171 I suggest using “thawed ground ice” instead of ground water

 

P5 L 179-182 Again, I think the methodology description of 0-1 classification can further be improved by providing additional information. As I understand from the text, Authors use ArcticDEM strips with initial spatial resolution of 2m. To assign the 0-1 values to the corresponding 30-m Landsat pixel, elevation changes must be resampled to the scale 30 within GEE. I do not know how GEE deals with such resampling, suppose it’s averaging the values of ca 225 pixels (15 pix for 2m resolution) ^ 2? And the 30-m pixel receives 1 if the average change value of 225 corresponding ADEM pixels is maximal and doesn’t exceed 20 m? Or at least 50% of them (110) show maximal change values for the entire study area? All these things are not clear for the reader in the text and could be improved.

 

P6 L196 RTS abbreviation has already been introduced

 

P6 L220-222 See my comment on the Temperature data above

 

P7 L258-259 Visually, RTS are quite similar to cryogenic landslides. What are the morphological / visual features that were chosen to associate these changes with RTS?

 

P8 L289 One can name the lake here, it is a famous Khalevto lake

 

P9 Fig. 3a I would describe these processes as ‘thermo-denudation’

 

Sub-section 3.3 There was no description of CDF-1 feature. I think it’s worth to describe it as well.

 

P10 Fig 4 There is no CDF-2 presented on the Figure. Please add.

 

P11 L354 There is no evidence that it was a degraded pingo. This small lake could have been resulted from the cryogenic landslide occurrence (as an example) and subsequent acceleration of thermokarst process (without water drainage).

 

P11 L357 “…or a GEC that released its methane non‐explosively”. That sounds unrealistic. The continuous permafrost here with tabular ground ice in the section would certainly prevent the methane release without an explosion. This might be an option in case of a lake (methane seep). Moreover, concerning the ‘pingo’ version: if there was a low change pixel values and no breakpoints, one could just analyze the changes in elevation at higher 2-m resolution. Perhaps, it could appear evident, that there was some/remarkable growth. ??? Otherwise, one can only guess about the potential reasons of the appearance of this new lake.

 

P11 L362-363 In my opinion, none of the three reported CDF are certainly represent GEC. The reason is that there is no evidence of the ejected material around any of them.

 

P12 Fig. 6j What is the dimension of NDVI here? Is it NDVI multiplied by 1000?

 

P14 L392 I would suggest using ‘paleo permafrost forming conditions’ instead of cryostructures. Cryostructures result from the type of sediments / type of freezing and GECs are not directly associated with cryostructures.

 

Please check references 18 and 19 for correct spelling.

Author Response

Dear Editor of Geosciences,

Thank you to the reviewers for their positive and constructive feedback. As detailed in our replies, we have incorporated most of their suggestions. We believe these revisions have improved the quality and clarity of our manuscript. Please find our replies in blue text below. Additionally, we directly replied to Reviewer’s detailed comments within the manuscript PDF they provided. Please find our replies included within the PDF in this re-submission package.

Sincerely,

Scott Zolkos, on behalf of the co-authors

 

Author Response File: Author Response.pdf

Reviewer 4 Report

This paper for the first time develops an approach to automatically find the gas emission craters recently found in Yamal and Gydan peninsulas based on publicly available remote sensing data, which is a very important step to understand the potentially hazardous areas. I highly support it to be published, however there are some shortcomings.

 

The main reason why I assign it for the major revision is because the readers non-specialists in remote sensing will be confused with the non-detailed and sophisticated description of the methods of detection of the surface changes. Why the scaling was needed and not the changes in hydrology vegetation and topography were treated in physical variables (meters, greenness, water table)? What was the reason you used the fishnet for validation? Etc. The questions are more detailed in comments all over the text of the paper in attached file.

 

The layout is uncomfortable for the reader. It is hard to address the figures put several pages after the text where these are referenced. I suggest there might be the pages fully consisting of the Figures, but just much closer to the first reference in the text.

 

There are some terminological issues with hazard and risk mitigation which need to be cleared.

 

The paper is missing the analysis of some non-GEC features, although it looks like there was an attempt to include the ones, like thermoerosional cirques formed on the shores by the retrogressive thaw slumps. I suggest authors to include the description of these features for the validation of their surface change study method with the real dataset from Arctic Coastal Dynamics GIS product.

Directly linked to the GEC formation – the pingo or predecessor mound problem. I believe this study could have accumulated enough information on pingo morphology in Yamal and could put forward the idea of what is the difference between pingos and pingos which might explode, which would be the real diagnostic indicator for geological risk reduction.

 

The attempt to link the GEC with the changes in heat exchange might be improved with the data from GTN-P database on changes in permafrost temperature to see whether there are links between GECs occurrence, Arctic warming and permafrost degradation. Using a single point to illustrate it does not look reliable.

 

 

One of the disappointing shortcoming is the lack of the indication of the period of study both in the text and on the figures. Figure panels should all be entitled in the figure caption according to MDPI author guide.

 

I would like to read this paper once again to make sure the methods are clearly described and other improvements to the text are made. I will not discuss anything in the form of response to my comments, because all my points and comments are aimed to improvement of the quality of the study.

Comments for author File: Comments.pdf

Author Response

Dear Editor of Geosciences,
Thank you to the reviewers for their positive and constructive feedback. As detailed in our replies, we have incorporated most of their suggestions. We believe these revisions have improved the quality and clarity of our manuscript. Please find our replies in blue text below. Additionally, we directly replied to Reviewer’s detailed comments within the manuscript PDF they provided. Please find our replies included within the PDF in this re-submission package.
Sincerely,
Scott Zolkos, on behalf of the co-authors

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

I have re read the new version of the paper. I found it much better, and, again, I'm fully support it to be published as it among all contains a lot of useful information on the new methods of study. However, I still unable to understand the validation section of the Methods. Authors are not consistent with terminology related to this part of the study along the manuscript. I am lost within different types of cells used for validation and do not understand if the method of validation chosen by authors is an optimum. 

Also, I believe authors may greatly increase the reader's experience if they contribute more efforts in rearranging the Figures within the text and improve some of the Figures.

There are also some statements which contradict the logic of the narrative.

Below you may find my comments for your further consideration. I would be happy to read the paper published, but now it needs further improvements:

Line 98 – Could you please specify the full names of the bioclimatic subzones?

Line 99 – 9-35 degrees C? where does this data come from? What kind of temperature is it – maximum and minimum? At which station?

Line 100 – I am not sure that it is correct to describe vegetation with the terms used for numbers: ranging from… to… Please check up and correct

Line 116-118 – “Infrastructure and activities are primary human uses” I think two different categories are combined here like rabbits and sawmills. Infrastructure could not be considered the use. Please correct

Line 120 – temperatures again. Why not to put them together with the line 99?

Lines 140-141 – line 73 states that calderas are ca. 35 m (from 10 to 90), quite comparable to Landsat pixel or even smaller. Changes might not be found for small GECs. Please provide more details on this issue.

Line 162 – again there is an ambiguity with the range from … to. Did you measure change by overlapping the layer of 2017 with the layer from 2008? Or did you measure change for every year in the range 2008-2017?

Lines 172-173 – Please write if you introduced any corrections to the rest of the strips which did not have any metadata?

Lines 181-182 – I do not understand why omission of water bodies before 1984 and after 2018 result in omission of changes within drained lake beds. Please write in more details.

Lines 186-189 – As far as I understand you assigned 1 unit for every type of change out of 3 within a pixel if it occurred. If not, you assigned 0. Please clarify because it seems at first that there is some continuum between 0 and 1, and you somehow scaled based on the level of change.

Lines 189-191 – Why did you took the mean and not the maximum change in altitude? It seems that was the way to find the small GECs.

Line 202 – Please discuss if you may have omitted the most dangerous GECs within this 100 m buffer from the infrastructure. It looks like this measure was excessive.

Line 210 – GEC crater hunters is better to write as GEC hunters or crater hunters because gas emission crater crater hunters is wordy

LINE 216 – Landcover data was derived from CAVM. This is a repeat.

Table 1 – you said that you filtered out mountains. Why did you include them in the Table?

Lines 226-228 – had all of the points co-occur to make feature defined as GEC, or any single one?

Section 2.3 – remains unclear why you made so many validation cells. Larger cell divided into smaller cells, and even to a smaller grid of 5x 5. What for? Why this number of cells 31 was chosen? Was this analysis made for every year in the range 1984-2017? This section has to be seriously reworked for clarity.

Line 238 – I believe it would be better to separate the weather trends analysis from significance statistics starting a new paragraph.

Line 266 – Pixel change values could be less than one. We definitely missing the description of scaling in methods.

Line 282 – fishnet cells – You did not use this term for any type of cells when described methodology. On the other hand you used Smaller cells, validation cells. Please make sure you use consistent terminology, otherwise readers cannot understand what exactly is meant.

Lines 335-336 – don’t you think that erosion changes are also associated with thermokarst lake drainage? Why?

I suggest putting Figure 1 between the lines 134 and 135

Lines 341-342 – largely unknown implications … - I doubt it is correct to say that, except for the craters which are really relatively new feature, all the other processes related to permafrost degradation are well described, and you made eference to them – Leibman wrote about retrogressive thaw slumps, coastal erosion is well represented by the ACD database. We cannot say about the unknown hazards sinde there was a study of Hjort et al 2018 in Nature about the risk assessment from permafrost degradation. Impications for biogeochemistry are by far studied best ofall the other changes. Note the studies of thermokarst lakes by Walter et al., 2007, Zimov et al. 1997, all the works of Germans in Lena delta and Hershel Island. This is not from this regiona but when we speak about the processes they are quite the same all over the arctic. So it includes both the unknown and well described change. Please consider to introduce changes to the text accordingly.

Line 383 – extra point in the middle of sentence. A typo. Please remove.

Figure 3 – I think it would be more valuable to show readers the previous state of the image with unchanged objects, and leave the newer images with colored pixels marking changes. Please do so if there was no objections from other reviewers. To reduce size of the figure, I suggest to put 4 panels in a row. Pairs of panels could be outlined to show they are related to one object. This way it allows you to put it closer to the text – before the line 282 forming a page with Figures 2 and 3.

Figure 4 – it should be resized. You might either use the logarithmic scale for axis Y or just compress it more, because some of the data around 1.0 is not seen anyway. There is also no need in values above 1.1 on X-axis, so you will have more room for writing the text in normal size (Geoscience does not recommend size of text less than 8).

Figures 4,5 should be placed on the page after the page with Figures 2,3 This way all the figures referred within the section 3.1 will stay in this section and readers won’t need to go through pages looking for the figure, and then returning back looking for the text they have read. This will improve the quality of representation and pleasure of reading.

Would be good to see Figure 6 after the line 305 and figure 7 after line 321.