Supervised Geomorphic Mapping of Himalayan Rivers Based on Sentinel-2 Data
Round 1
Reviewer 1 Report
Overview:
This work have a great interests because it is know that proglacial regions (refer to areas that have been impacted by glaciers, such as when a glacier melts and the water flows forming rivers and eroding the terrain) and their associated downstream river channels are rapidly changing due to the current swift pace of global warming. Morphological features of the rivers plains, such as channel width, channel pattern, and the prevalence and spatial distribution of riparian vegetation, are known to significantly affect river ecosystems and potential hazards. Due to this, a frequent morphological monitoring of proglacial streams is of the utmost importance to detect ongoing evolutionary trajectories and to anticipate future changes and hazards. However, a frequent morphological monitoring in these rivers is challenging due their generally low accessibility and terrain complexity, which are extreme in the Himalaya regions. In this case, satellite-based RS may be a suitable approach to monitor morphological changes in rivers of sufficient width.
The study has interest to geomorphologist community because the main goal is to accurately classify the areas within proglacial Himalayan fluvial corridors into three geomorphic macro-units defined in the Geomorphic Unit System (GUS), base-flow channels (submerged), emergent sediment units (unvegetated), and in-channel/riparian. The model was developed and trained in 3 rivers and then applied to a different river to properly evaluate its performance. The model uses Random Forest as classification method of a Sentinel-2 images ( Green and Near-infrared bands were used as predictors variables) and the results have an overall average accuracy of 96%. This means that the model was able to correctly predict or classify proglacial geomorphic features in 96% of cases. The authors used Kappa Index to evaluate the model accuracy, which reached a 0.94.
The authors think that the utility of this model lies in its ability to detect past and current morphological changes occurring in the Himalayan proglacial rivers with significant impact on the ecology and human communities in the region.
Main concerns
1.The authors discuss the importance of remote sensing in monitoring past and current fluvial dynamics in the proglacial zones of the Himalayas, due to their remote location and the logistical difficulties associated with traditional field-based and UAV-based morphological monitoring activities. On this point we agree, the use of images is the only way to monitor in a reasonable time the changes that are occurring in rivers in remote areas.
Key geomorphological concepts such as - "Proglacial zones" refer to areas that have been impacted by glaciers, such as when a glacier melts and the water flows forming rivers and eroding the terrain. - "Fluvial dynamics" refer to the patterns and processes of rivers, including how they flow, how they change over time, and how they affect the surrounding landscape.- "Morphological monitoring" refers to the study and tracking of how the shape of a physical feature changes over time. Change is implicit in all these concepts and this can be accelerated or forced by global warming.
Nevertheless, the authors focusing the classification in three "macro-units" within a river corridor. The three macro-units described likely refer to different landscape features of the river.
*Surrounding vegetation*: Vegetation can have a significant impact on the amount of sediment that gets transported in a river, as plant roots help hold soil in place, reducing erosion and sediment transport.
*Low-flow channels*: This term refers to areas of the river that carry water during periods of low or normal flow. These channels can change over time due to sediment transport.
*Unvegetated bars*: These are accumulations of sediment, usually sand or gravel, that form in a river's channel. These bars can change in position and size over time, due to erosion and deposition processes.
The model classifies well the pixels of the images fitting them in one of these three classes. However, it is not expected that significant changes will occur in the distribution of these units unless a catastrophic event occurs. The changes will occur precisely at the edge of the macro units as these are the areas susceptible to change. However, the authors assume that the edge pixels accumulate the model's error, so they have not given enough importance to the information contained in these mixed pixels. I believe that the authors should study the information from these pixels more deeply. It is necessary to see how these edge pixels evolve. To do this, these pixels should be classified according to the units involved and their fragmentation. Then, it should be observed which type of unit each of the defined edge pixel types change to. This is the way to find out what changes are occurring in the rivers studied. I think authors should include an analysis of edge pixel trends in their work.
2.The bibliography must be ordered by numbers even if it does not appear in alphabetical order.
Author rules that is provided on the instruction for authors page: ” References: A comprehensive list of references must be included, and numbered in order of appearance in the text (including table captions and figure legends) and listed individually at the end of the manuscript…”
Minor concerns
-Line 121: Check the text format of the table footers, they have different font size .
-Line 132-135: The text and the table 2 say the same thing, choose table or text
-Line 138:“Table 2: Macro-unit classes based on Google Satellite Imageries”.
If the authors prefer use the table 2, the foot of the table must be: Macro-units mapped based on Geomorphic Unit Systems (GUS) recently proposed by Belletti et al. (2017).
-Line 196: 29/09/20 (Saltoro, Figure 5), lack the year.
Author Response
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
Mukhtar et al have submitted their manuscript titled "Supervised geomorphic mapping of Himalayan rivers based on Sentinel-2 data" for review and potential publication in the Remote Sensing journal.
The authors present a machine learning approach to classify river development in the Himalayas, an area that exhibits considerable vulnerability regarding climate change effects.
The text is generally well-written and makes an overall interesting read. The machine-learning aspect is straightforward and nothing too new except for the specific use case it has been applied to. That is valid, but then I would expect better exploitation of the science background and implications. Either an innovative method or a well-discussed science case are needed and so I recommend **major revisions**, hoping the authors can come up with a more in depth-treatment on an otherwise interesting topic.
**Specific comments**
Figure 1: The authors may want to include the presented data source of the r/s images in the caption, it is unclear which data we are seeing being presented here without context. Supposedly S2 m/s but that is not clear without reading the text.
Figure 1: The map overview is -- due to its size -- not really helpful I must say. It would be great to have a larger-scale topographic/physiographic map that allows us to get a better idea about where the rivers are and from which areas they are derived. The arrows and lines do appear a bit cluttered, a larger overview map would allow to place text labels perhaps.
Figure 1: The scale bar lengths for AOI-1 and 4 seem a bit arbitrary. The authors might want to fix that.
L122. "The mapping of the fluvial corridors has been performed manually
within ArcGIS Pro software" --- could the authors provide more insights into the mapping process? They link to a publication which is fine, but I feel the baseline needs to be clearly communicated within this manuscript as well. How does this delineation work?
Also, how is a fluvial corridor defined? -- this journal has a broad audience.
L128: The reference is poorly cited and the title is cropped. Researchgate is not a citation source anyway. The authors may want to fox that issue.
L130: "As already mentioned in the introduction" -- either then do not mention again, or do not mention that it has been mentioned. It implies it is not relevant (just a suggestion).
L132: The reader would certainly appreciate if the macro units could be explained in a bit more detail. What do they represent and why is that considered to be important/representative? Are there others that have not been chosen and why. Even a sketch to show these features might be quite instructive. This journal has a broad audience. I understand that some more aspects follow in L171ff, but that paragraph does not present all units.
L140: I think this requires some explanations. The authors have chosen a certain m/s dataset (S2) but that only covers a few years back -- likely not significant for demonstrating effects due to climate change. Why has no other data been chosen, such as Landsat to study the longer-term effects? A standard reference period for climate mapping is 30 years. As a demonstrator, the presented approach still works of course, but the implications of climate change effects can not be communicated convincingly which exhibits a weak point of this manuscript. As I suppose it is about image resolution, a discussion would be helpful to make that transparent.
L150: The authors do not present image numbers and references to their data to allow review and potential reproduction.
L171ff: Could this be extended to cover all units and their relevance in more detail (see above)?
L246 "As mentioned above, ..."; L236 "As described in Section 3.2.1 ... " --- see above comment
Fig. 4: Have the author considered presenting three images of one type per row? The first row is for the optical images, second row is for classification. Inner grid labels could be removed so all images become a bit larger and easier to study. It is hard to see anything right now. Also, the map scale length seems arbitrary again.
Fig. 5: Could be stacked vertically to present more detail. The authors might want to think about it. Furthermore, the red arrows in X.3 are not explained in the legend/caption. Check grid labels, illegible and likely too granular anyway.
L346: A discussion on applicability would be highly needed. Also considering the limitation regarding time frame (it is only marginally mentioned in L370 . Monitoring river dynamics under consideration of climate change has some reasoning behind it -- what are these? What about physical and social vulnerabilities and so on? How could these mapped units help to better understand the impact of climate change? I believe this approach could be better sold in this manuscript by considering who is perhaps interested to use it.
Also, the authors might want to extend the scientific discussion (see entry statement), not only the technical one as the RF method is nothing new. Applying a conventional method to a new field without discussing the scientific implications and value feels a bit too thin.
I gave a number of hints and the authors might want to pick these up or find new ones. They could also link to a number of SDG topics in a broader discussion. There is lots of potential to expand.
L392: I very much appreciate authors provided the code on a data platform.
The problem is, it is inaccessible as of 12 Aug 2023:
https://github.com/ZarkaMukhtar/GEE-Code/blob/main/Sentinel-2%20based%20supervised%20classification%20modeling%20for%20proglacial%20river%20geomorphological%20macro-units%20extraction
The authors may want to check spelling and punctuation once more (space before units (km, m, nm), some prepositions are off) but for the rest it is absolutely fine and well written.
Very minor spelling and punctuation issues.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
I thank the authors for addressing my comments and for incorporating some new material/changes to improve the overall presentation. I do understand that the authors do not want to address the geoscientific/social topics and focus on the algorithms and methods, and that is certainly fine by me. What confuses me, however, is that the conclusions do not provide any deeper insights into the methods but rather summarize the overall findings quite unspecifically -- lacking all quantitative information. I believe a conclusion should provide more than just an abstract version of the main findings.
It is not about satisfying a reviewer's personal feelings, but I believe there is a chance to sell the method and approach much better to the research community who will essentially take nothing out of the conclusions as they are written now. Given that for citation researchers focus on the abstract and conclusions for screening, there is not much to extract here. The authors might want to consider that.
I would recommend a minor revision and I do not need to see the revision again. I would strongly suggest two things: (1) improving the conclusions by focusing on the real findings and not some generic statements that would be applicable to any applied model, and (2) moving supplementary material to a decent research data platform and not github. Zenodo and alike would certainly be better options.
These are, however, just suggestions, for the authors' consideration.
Kind regards.
Just minor issues that could potentially be addressed during copy-editing.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
