Identifying Alpine Lakes with Shoreline Features

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper is important and useful but the paper need to be rewritten for the better understanding of readers:

The paper used two step classification approach for the lake mapping however when reading the paper I was fully confused till I reached the discussion section.

Rf part is clear in the paper then the Deep learning based segmentation and classification both methods were used but the way paper is written no one will be able to understand these details.

Thus I have following suggestions:

1. The authors should divide the methodology into different segments like a, b, and c for the different methods used and clearly describe the method in the relevant section.

2. section 3.2.2. is Identification of Alpine Lakes where authors have mostly described the methodology but in section 3.2.3. Deep Learning-Based Alpine Lake Identification only some accuracy assessment method was presented totally confusing the readers.

3. The results starts with 4.1. Accuracy Assessment I think it should be brought to the end of methodology section and in results only the results should come.

4. Figures 7 and 8 What is My? I think the authors here want to say it is by their model, but it is not the way of doing it. I think authors can write a, b and c and in the caption they should describe what is a b and c.

 

Te authors have not cited other relevant work like Mitkari, K. V., Arora, M. K., & Tiwari, R. K. (2017). Extraction of glacial lakes in Gangotri glacier using object-based image analysis. IEEE Journal of selected topics in applied earth observations and remote sensing, 10(12), 5275-5283. whaich is importnt work in segmentation of the glacial lakes.

 

I think after these corrections the paper can be accepted  

 

 

Author Response

Comments 1: The paper used two step classification approach for the lake mapping however when reading the paper I was fully confused till I reached the discussion section. Rf part is clear in the paper then the Deep learning based segmentation and classification both methods were used but the way paper is written no one will be able to understand these details.

Response 1: Thank you for your feedback. We have revised the sections to clarify the RF and deep learning methods, ensuring that the methodology is presented in a more coherent manner. Your insights will help improve the overall clarity of our work.

Comments 2: The authors should divide the methodology into different segments like a, b, and c for the different methods used and clearly describe the method in the relevant section.

Response 2: Thank you for your suggestion. We have rewritten the methodology section according to your advice, dividing it into different segments to provide a clearer description of the various methods used.

Comments 3: section 3.2.2. is Identification of Alpine Lakes where authors have mostly described the methodology but in section 3.2.3. Deep Learning-Based Alpine Lake Identification only some accuracy assessment method was presented totally confusing the readers.

Response 3: We have made modifications to the sections to enhance clarity. In section 3.2.2, "Identification of Alpine Lakes," we have organized the key methodologies and processes. Additionally, we have revised section 3.2.3 to be titled "Accuracy Assessment" and included the assessment results.

Comments 4: The results starts with 4.1. Accuracy Assessment I think it should be brought to the end of methodology section and in results only the results should come.

Response 4: We have moved the "4.1. Accuracy Assessment" section to the end of the methodology section as per your recommendation and ensured that only the results are presented in the results section. Your feedback has helped improve the clarity of the article's structure.

Comments 5: Figures 7 and 8 What is My? I think the authors here want to say it is by their model, but it is not the way of doing it. I think authors can write a, b and c and in the caption they should describe what is a b and c.

Response 5: Yes, ‘My’ represents the results of this study. We greatly appreciate your suggestion and have revised Figures 7 and 8 accordingly. We have changed the labels to “a,” “b,” and “c,” and added corresponding descriptions in the figure captions to clarify each part. We hope this modification will improve reader comprehension.

Comments 6: The authors have not cited other relevant work like Mitkari, K. V., Arora, M. K., & Tiwari, R. K. (2017). Extraction of glacial lakes in Gangotri glacier using object-based image analysis. IEEE Journal of selected topics in applied earth observations and remote sensing, 10(12), 5275-5283. whaich is importnt work in segmentation of the glacial lakes.

Response 6: Thank you for highlighting this important reference. We have now included the citation of Mitkari et al. (2017) in the revised manuscript, as it provides valuable insights into glacial lake segmentation. We appreciate your suggestion and believe this addition enhances the context and relevance of our study.

L181-182: “The resulting terrain slope data can also provide useful information for distinguishing between water and non-water bodies [48,49].”

 

Reviewer 2 Report

Comments and Suggestions for Authors

Please check the attached file.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Comment 1: This paper titled “Alpine Lakes with Shoreline Features” presents a valuable contribution to remote sensing and environmental monitoring, particularly in the area of high-altitude lake detection. The study would be strengthened by a more comprehensive examination of methodological limitations, data-specific challenges, and generalizability. Authors are requested to check the following comments.

Response 1: Thank you for your review and feedback on our manuscript. We have included a more comprehensive discussion of the methodological limitations, data-specific challenges, and generalizability of our results in the revision. We hope these additions will help enhance the understanding of our research contributions.

Comment 2: Introduction Section. The introduction lacks a critical comparison of existing methods for lake identification, especially with respect to the advantages of the proposed hybrid method over standalone machine learning or deep learning techniques.

Response 2: As suggested, we expanded the Introduction to summarize the existing methods for lake identification and the advantages of our method compared to using machine learning or deep learning alone.

L109-117: “Currently, machine learning and deep learning techniques for identifying alpine lakes are widely applied. Compared to using machine learning or deep learning alone, combining the two methods leverages the strengths of each. Machine learning excels in the sensitivity of identifying individual pixels, making it effective for detecting small lakes, while deep learning more effectively captures the surrounding environmental context of targets, achieving higher classification accuracy. Therefore, integrating machine learning with deep learning enables a more efficient and accurate identification of a greater number of alpine lakes.”

Comment 3: The study is situated within the eastern Himalayas, more details about regional climatic challenges and data limitations should be in the text of the manuscript.

Response 3: As suggested, we expanded the discussion to provide a more detailed analysis of the regional climate challenges and data limitations specific to our study area.

L138-145: “The region experiences cloud cover for most of the year, with clear skies approximately half of the time. Snow cover persists for extended periods, and lakes remain icebound for long durations, with perennial ice and snow coverage. Compared to the western Himalayas, the eastern Himalayan region hosts larger and more numerous alpine lakes and exhibits more severe glacial loss. Additionally, the warming rate in this region surpasses the global average, making it one of the areas most susceptible to the impacts of climate change.”

Comment 4: Study Area Section. Page 03, Line no 131. Revise the figure 01. Location map of the study area. Figure 1 doesn’t show the actual location of the study area on the world map.

Response 4: As suggested, we revised the study area map to display its location on the world map.

Comment 5: Methodology Section. Despite the robustness, the methodology would benefit from a clearer rationale for selecting ConvNeXt over other convolutional neural network architectures.

Response 5: Thank you for your valuable feedback. As suggested, we added more reasons for choosing ConvNeXt for alpine lake extraction.

L242-250: “This method selects the ConvNeXt classification model as the core algorithm for the lake classification task, primarily due to its outstanding performance in complex environments, especially its enhanced adaptability in handling diverse images and transformation uncertainties. Compared to other convolutional neural network architectures, ConvNeXt strikes a good balance between performance and computational efficiency, effectively capturing detailed information without sacrificing speed. Other models, such as DeepLabV3, enhance the ability to capture contextual information through the use of zero convolution, but they struggle with precision when dealing with complex boundaries and small targets[50]. Additionally, the extensive use of zero convolutions can lead to higher computational and storage demands[51,52].”

Comment 6: Page 04, line no 155-159. The resampling of the ALOS PALSAR DEM data may introduce interpolation artifacts, Authors are suggested to discuss more about it.

Response 6: We adopted the bilinear interpolation for image resampling. Compared to other interpolation methods, bilinear interpolation produces relatively lower spatial errors and typically does not introduce significant artifacts. This makes it more reliable during the resampling process, especially for high-resolution images. Our evaluation and validation indicate that these artifacts have a minimal impact on the study results, thereby ensuring the reliability and validity of the data. We discussed this further in the manuscript.

L177-180: “Although interpolation artifacts may occur during the resampling process, our evaluation and validation indicate that these artifacts have a minimal impact on the study results, thereby ensuring the reliability and validity of the data.”

Comment 7: Page 07, Line 258. Figure 03, Flow diagram of the methods in the study. The figure 03 does not address computational efficiency, which is crucial for large-scale studies like this.

Response 7: We have added the average training time required for each epoch in Table 3 to demonstrate the computational efficiency of the model.

Comment 8: The paper does not provide sufficient statistical rigor, such as confidence intervals, to contextualize the performance metrics.

Response 8: As suggested, we will compare the results of this study with manually interpreted results through contour analysis to validate the credibility of the findings.

L374-376: “For alpine lakes larger than 0.1 km², a comparison of boundaries with manually interpreted results revealed an average boundary difference of 2 pixels. For lakes smaller than 0.1 km², the average boundary difference was 1 pixel.”

Comment 9: Authors are requested to explore the potential impact of seasonal variations, which might affect the generalizability of the model.

Response 9: The reviewer brought up an important point! Lakes in this mountainous region were greatly affected by seasons, especially snow and frozenness, which can also block the ability on observing surface water by satellites. Therefore, we only selected images acquired in late summer and early autumn in this study to reduce the impacts from seasonal variations. However, we discussed this limitation in the discussion.

L379-382: “Due to the significant seasonal influence on alpine lakes in mountainous regions, particularly from snow cover and ice formation, satellite observations of surface water can be hindered. Therefore, to minimize the effects of seasonal variability, this study exclusively selected images collected in late summer and early autumn.”

Comment 10: The authors could have tested their model’s robustness against different seasonal datasets to reinforce the model's validity across varying environmental conditions.

Response 10: As responded to the previous comment, we selected images from late summer and early autumn to avoid the impact of seasonal variations in this study. However, we did include a discussion regarding the limitation for seasonal monitoring.

L379-382: “Due to the significant seasonal influence on alpine lakes in mountainous regions, particularly from snow cover and ice formation, satellite observations of surface water can be hindered. Therefore, to minimize the effects of seasonal variability, this study exclusively selected images collected in late summer and early autumn.”

Comment 11: The discussion could be expanded to consider broader implications of the study, such as its applicability to other glacial regions outside the Himalayas.

Response 11: We appreciate the suggestion, and we are working on applying the method to the entire Himalayas as well as other mountainous regions beyond the Himalayas. We highlight this further work in the limitations and future requirement.

L437-439: “Future research should further explore the classification of alpine lake types, such as glacial lakes, thermokarst lakes, and non-glacial lakes, while also developing a global dataset of alpine lakes with intra-annual temporal resolution.”

Comment 12: The authors briefly mention that the RF model has limitations in distinguishing between lakes and snow or shadows, the discussion lacks an in-depth examination of these errors’ potential impacts on downstream applications like hydrological modeling.

Response 12: Following the suggestion, we include a citation in this revision to substantiate the limitation on the error on shadow identification by traditional machine learning algorithms, such as RF, and it impact to downstream application.

L80-86: “The potential impacts of these errors on downstream applications are significant. For example, in hydrological modeling, shadows affect the spatial heterogeneity of solar radiation in mountainous areas, significantly influencing the spatial distribution of ground temperature, evapotranspiration, snowmelt, and glacier mass balance. Shadows may reduce the quantity and quality of information obtained from these various properties, while also increasing errors and uncertainty in subsequent analyses.”

Comment 13: Conclusion section underscores the approach's success, it lacks practical recommendations for further improving or validating the model.

Response 13: We have added a discussion of future improvements and enhancements needed for the model in the discussion section.

L432-440: “In this study, we utilized the ConvNeXt classification model for water body identification, achieving robust performance in distinguishing aquatic features from various noise artifacts. Nevertheless, machine learning and deep learning approaches often necessitate extensive annotated datasets for training, demanding considerable time and human resources. Moreover, optimizing model parameters to effectively navigate complex geographic environments remains a significant challenge. Future research should further explore the classification of alpine lake types, such as glacial lakes, thermokarst lakes, and non-glacial lakes, while also developing a global dataset of alpine lakes with intra-annual temporal resolution.”

Comment 14: Figure 7 does not clearly illustrate the noise reduction difference between the RF and proposed methods.

Response 14: Thank you for your valuable feedback. We have revised Figure 7 to use different color schemes for the noise and lakes in the RF model results, highlighting the noise present in the RF outcomes.

Comment 15: Every study has its limitations. It is recommended that the authors include a separate section outlining the specific limitations of this study.

Response 15: As suggested, we have added a section titled "Limitations and Future Requirements" in the discussion part to provide a comprehensive discussion of the limitations of this study. This section aims to offer a clearer understanding of the constraints we encountered during our research and their impact on the interpretation of the results, as well as to outline prospects for future research directions.

Comment 16: In the revised version, the authors are also encouraged to clearly highlight the novelty of this work and explain why it merits publication.

Response 16: In the revised version, we have further highlighted the novelty of this work in the conclusion section and provided a detailed explanation of its significance and why it merits publication. We hope these improvements will help clarify the contributions of our research.

L445-451: “This study presents an innovative method that combines machine learning and deep learning for the accurate extraction of alpine lakes. The approach first employs a RF model to delineate water bodies, and then utilizes the environmental features surrounding these water bodies to leverage the powerful classification capabilities of deep learning, effectively eliminating misidentified areas. This method successfully addresses the impact of noise, such as mountainous shadows, on lake extraction, allowing for the effective delineation of numerous small alpine lakes in mountainous regions.”

 

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Authors, please check Figure 1; it does not show the world map. The authors have addressed all the comments well. Now, the paper may be accepted for publication.