Detecting Coseismic Landslides in GEE Using Machine Learning Algorithms on Combined Optical and Radar Imagery
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsGeneral comments
In the manuscript Detecting Coseismic Landsdlies in GEE using Machine-Learning Algorithms on combined Optical and Radar Imagery, the authors used five classifiers (CART, RF, GTB, SVM and NB) to detect four groups of coseismic landslides and concluded that their method of using both optical and microwave remote sensing data outperformed other methods. The technical part of this manuscript is solid, but the writing structure and presentations (figure and table) are poor.
Abstract
It might be better to write out the names of the classifiers in the abstract.
Introduction
The introduction section is poorly structured. I suggest combining 1.1 and 1.2 for the following reasons. The paragraph from Line 53 to 60 seems unnecessary. For the paragraph from Line 63 to 72, please add more context to the coseismic landslides. For example, what proportion of global landslides are related to earthquakes? What is the role of rainfall in coseismic landslides, is there a clear distinction between coseismic landslides and rainfall-induced landslides?
Line 95 to 99: To make the paper more accessible to general readers, please explain the mechanism of using remote sensing to detect landslides. For example, the rationale behind using a spectral index threshold is that the optical spectral signature can differentiate vegetation and bare ground caused by landslides. Similarly, please explain the mechanism of using microwave remote sensing to detect landslides. For example, do landslides alters the radar reflectance, hillslope geometry, roughness, …?
Line 104 to 107: The sentence is not clear, please consider rephrasing it.
Line 128: It is surprising to see some studies used TOA products, because I imagine it is less reliable to inform what is happening on the ground than the surface reflectance. What are the advantages of TOA products? Please make sure the introduction can help readers understand the context of this topic and know why the authors conducted the study, instead of piling up references of previous studies.
Line 133 to 139: Please classify these accuracy metrics to highlight the priority of each one. For example, TNE and FPR emphasize different things. It would be more organized to see different groups of metrics instead of a laundry list.
Line 198 to 254: Please considering consolidate these bullet points because they are taking up a lot of spaces.
Materials and Methods
Line 381 to 383: Please spell out the name of each country in Figure 1.
Line 462: Are there highly correlated conditioning factors? If so, the factor importance ranking will be unreliable.
Line 552: Consider using square kilometer as the unit of area for better readability in Table 4.
Results
Line 639: Please spell out the full name of each classifier and each accuracy metric in the caption of Figure 6. Each figure should be able to stand alone so readers can get the information without going back and forth to the main text.
Line 661: Please try to use a figure to replace Table 6 to improve readability.
Line 668: Please spell out the synonyms.
Line 685: It is not appropriate to use the lines to visualize feather importance, considering commonly used bar plots. Same for Figure 9, please consider a more appropriate visualization method.
Discussion
In the introduction, the authors talked about the importance of landslide detection to rapid disaster response. Please consider adding a paragraph to elaborate on the implications of results in this study to disaster management.
Comments on the Quality of English LanguageMinor editing of English language required
Author Response
Thank you very much for taking the time to review our manuscript. We truly appreciate your valuable feedback, constructive critique, and insightful comments, which have undoubtedly enhanced the quality of our work.
We have carefully considered each of your suggestions and critiques, and we are pleased to provide our detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe study is methodologically robust, employing innovative techniques for integrating data from multiple satellite sensors to improve the accuracy of landslide detection significantly. This multidisciplinary approach, combining remote sensing, machine learning, and geospatial analysis, showcases a novel application of these technologies in environmental disaster management. Below are detailed comments:
1. The manuscript mentions the use of imagery data from Sentinel-1, Palsar-2, and Sentinel-2 but does not detail the temporal range of the data acquisition, spatial resolutions, or any gaps in the temporal or spatial data coverage. Additionally, the discussion on how geological background, vegetation cover, and other regional factors might affect the model's generalizability is lacking, which could impact the method's widespread applicability.
2. The article discusses using Google Earth Engine (GEE) for data processing and analysis but does not specifically address the computational resources required for this study, including processing times and memory requirements. This information is crucial for other researchers who might want to replicate or expand upon this study.
3. Although the article shows high accuracy of the model, there is insufficient discussion on handling overfitting, a common issue when applying machine learning classifiers for land cover classification. The article should detail how overfitting risks were mitigated through techniques like cross-validation and regularization more thoroughly.
4. While certain terrain and geographic factors such as slope and elevation are mentioned, the manuscript does not thoroughly discuss other critical factors influencing landslide occurrence, such as soil moisture and geological structure. In this context, I recommend incorporating the following reference to enrich the discussion on terrain's impact:
Chang, S., Deng, Y., Zhang, Y., Wang, R., Qiu, J., Wang, W., ... & Liu, D. (2022). An advanced echo separation scheme for space-time waveform-encoding SAR based on digital beamforming and blind source separation. Remote Sensing, 14(15), 3585.
5. The manuscript's proposal to use transfer learning for improving model generalizability is promising yet the discussion regarding the selection of source and target domains and model adjustment is somewhat limited. For a deeper exploration of these strategies and their specific impacts on model performance, adding the following reference would be beneficial:
Ma, Y., Chen, S., Ermon, S., & Lobell, D. B. (2024). Transfer learning in environmental remote sensing. Remote Sensing of Environment, 301, 113924.
Comments on the Quality of English Language
Abstract (Lines 10-28): Consider breaking the abstract into shorter sentences to enhance readability. For instance, "This study proposes a novel approach, the ML-LaDeCORsat (Machine Learning-based coseismic Landslide Detection using Combined Optical and Radar Satellite Imagery), that integrates freely available Sentinel-1, Palsar-2, and Sentinel-2 imagery data, along with relevant spectral indices and suitable bands using machine-learning-based classification for coseismic landslide detection implemented in Google Earth Engine (GEE)." This sentence could be split into two or three shorter sentences for clarity.
Introduction (Lines 34-41): The use of "we here propose" could be replaced with "this study proposes" for a more formal tone. Additionally, "landslide triggered by earthquakes as case studies" could be rephrased to "earthquake-triggered landslides as case studies" for better flow.
Methodology (Lines 394-401): There's a slight inconsistency in the use of past and present tense, which can be streamlined for consistency. For example, "To prepare multispectral S2 data for landslide detection for each case study, two cloud-free sets of imagery covering the entire Area of interest (AOI) are necessary, representing the situations before and after the landslide events." It would be clearer if consistently maintained in the past tense, given the work has been completed.
Author Response
Thank you very much for taking the time to review our manuscript. We truly appreciate your valuable feedback, constructive critique, and insightful comments, which have undoubtedly enhanced the quality of our work.
We have carefully considered each of your suggestions and critiques, and we are pleased to provide our detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsOverall comment: the authors have made significant revisions to improve this manuscript. I really appreciate their effort. However, the quality of the figures and tables can be further improved. Figures and tables are the most efficient way to communicate a large amount of information. Readers should get a reasonably good understanding of these information from figures and tables alone. It would be better to write more detailed captions, for example, in Figures 9 and 10.
Author Response
Dear Reviewer,
Thank you very much for taking the time again to review our revised manuscript.
We have carefully considered your suggestions and improved all table and figure captions.
We also discovered a minor issue on Figure 9 c): Japan was still written as JPN instead of in full form. We corrected and updated Figure 9. For Figure 9 we have also removed redundant/duplicated text parts which is now part of the caption (see page 25, section 3.2, 1st paragraph, line 721: removed text: “sorted by their sum over all case studies”).
Kind regards
The authors:
Stefan Peters, Jixue Liu, Gunnar Keppel, Anna Wendleder and Peiliang Xu
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsCongratulations on the substantial improvements made to your manuscript. The concerns I previously raised have been largely addressed. I believe this version of the manuscript now meets the publication standards.
Author Response
Response to Reviewer 2:
Thanks for your congratulations and again for taking the time to review our manuscript. Your previous comments and suggestions had helped a lot to enhance the quality of our work.
It appears to us that you may have mistakenly ticked the wrong "Open Review" box. It is shown at our end that you have NOT yet signed your reivew report to officially accept our paper for publication from your end.
Could you please look into that. Many thanks
Warm regards,
The authors:
Stefan Peters, Jixue Liu, Gunnar Keppel, Anna Wendleder and Peiliang Xu
