Detection of Large-Scale Floods Using Google Earth Engine and Google Colab

Response to Reviewer 1 Comments

1. Summary

We greatly appreciate your time and effort in reviewing this manuscript. Below, you will find detailed responses to your comments and corresponding revisions and corrections, which have been highlighted and tracked in the resubmitted files. Thank you for your valuable feedback.

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: There are two duplicate titles "2. Materials and Methods" and "2. Equipment and Methods", please delete them according to the content.

Response 1: Second title was deleted

Comments 2: The clarity of the figures in the paper is too low, especially Figures 1 and 2, which affects the viewing, and it is suggested to improve the resolution.

Response 2: We updated both images with higher resolution versions.

Comments 3: In the paper, large-scale detection of floods is carried out by Google Earth Engine and Google Colab, but the detection accuracy is not compared with the latest research methods in section 3.1, which is an important evaluation index for detection methods.

Response 3: Comparisons are added in the discussion. Lines: 470-483

Comments 4. As mentioned in 2.3.6, the article generated 100 random points for both flood and non-flood areas, but the sum of each type of Reference mentioned in Table 2 was not 100.  Could it be interpreted that the positions of "Reference" and "Predicted" were wrong?  As shown in the table, the overall accuracy of the results is 85%, but in the detection results, the detection accuracy of the non-flood area is significantly higher than that of the flood area, and the method proposed in the paper should pay more attention to the detection accuracy of the flood area, please give corresponding explanations.

Response 4: We updated the text and the corresponding confusion matrix table to give more information about this validation data (user’s and producer’s accuracy). Lines: 395-401

Comments 5. In 3.3, the analysis and conclusion given in the comparison with other flood mapping products are too simple, please give an in-depth analysis.

Response 5: We updated the corresponding text to give quantitative data about the impacted surfaces, and strengthen analyses and comparisons between products. Lines: 434-437

Response to Reviewer 2 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: P.3 Lines 99-110:  The main contribution of this research should be listed in bullets in Introduction. Please revise.

Response 1: Main contributions are given from line 100 to line 103

Comments 2: The paper organization is not mentioned at the end of Introduction

Response 2: Paper organization is presented at the end of the introduction part. It was completed and enhanced to be more distinguishable (L108-112).

Comments 3: In P.8: The methodology of processing chain overview in Figure 4 is not clearly defined. Please add extra flowchart or diagram to show the details of your approach

Response 3: The processing chain overview is improved to show all important step of the processing chain

Comments 4. The Discussion Section needs some comparison with similar recent works.

Response 4: Comparisons with some similar recent works are added to the discussion. Lines xx - xx

Comments 5: A list of abbreviations is recommended at the end of paper

Response 5: According to reviewer recommendations a list of abbreviations is added in the end of the paper. (L470 -484)

Comments 6: Although the overall organization, language use, style, methodology, discussion, tables, and figures of the work are good; the readers may find this work very beneficial if the conclusions explicitly highlight future work and viewpoints.

Response 6: thank you for this suggestion, we have added it to a current project on which we are continuing our work. (L580-587)

Comments 7: In Discussion Section: What are the methods that can raise the flood detection system's Confusion Matrix's accuracy above 85%? Please mention these methods to enhance the outcomes.

Response 7: we added a paragraph in the discussion part on that point (L470-475)

Response to Reviewer 3 Comments

1. Summary

Thank you for dedicating your time to review this manuscript. You can locate the comprehensive responses and the corresponding revisions/corrections highlighted in red within the resubmitted files. Your input is greatly appreciated.

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: This paper is a study on flood extent analysis in large areas using Google Earth Engine and Google Colab, representing a typical application in the field. However, there are some peculiar aspects to this study, such as the utilization of the existing 'Global 30m HAND' data and the focus on a specific region as a case study. Nevertheless, for this paper to be considered for publication in the journal, there are many sections that lack overall organization.

Response 1: Thank you for this summary. We improved our work thanks to all the reviewers’ comments. We now believe it has been improved enough to be considered for publication in the Remote Sensing Journal.

Comments 2: In the basic structure of the paper, there are significant gaps in each section. For instance, in the Abstract, it is essential to provide a concise description of the main methods used and the results obtained, supported by statistical numbers or indices to demonstrate their accuracy and scientific significance. However, the current abstract lacks these details

Response 2: We appreciate the reviewer's feedback and worked on enhancing the presentation of results and their associated discussions. We have incorporated more details into the abstract to give a concise yet informative overview of our research. We provided a more in-depth and detailed analysis of the scientific achievements and findings in Sections 3 and 4 to better convey the contributions and significance of our research. We believe these revisions improve the quality and depth of the paper.

Comments 3: Apart from the Introduction, there is a particular need for detailed explanations of data and methods in each sub-section of Section 2. Currently, Section 2 lacks proper organization and detailed descriptions

Response 3: We value the reviewer's feedback, and we have taken the suggestions into account for improving the organisation and level of detail in Section 2. We have restructured the section to provide a more logical flow of the methodology, and have added more comprehensive descriptions of the data and methods within each relevant sub-section. We believe this adjustment will help readers better understand the specifics of the data and methods used in our study and the rationale behind our choices.

Comments 4. Furthermore, it is crucial to describe the scientific methodology employed in this study with equations and detailed theories.

Response 4: We appreciate the reviewer's suggestion to provide a more detailed description of the scientific methodology, including equations and theories. In our revision, we have addressed this concern by incorporating additional theoretical explanations to provide a more comprehensive understanding of the scientific methods employed in our study.

Comments 5: In Sections 3 and 4, the presentation of results and the associated discussion are too generic. I encourage you to emphasize the scientific achievements accomplished through this research and provide a more detailed presentation of the results and relevant discussions

Response 5: We appreciate the reviewer's feedback, and we recognize the importance of emphasizing the scientific contributions of our work. As per the reviewer's suggestion, we have placed a stronger focus on highlighting the scientific achievements of our research. This involved providing a more detailed description of our results, and engaging in more comprehensive discussions in relation to existing research within the same thematic domain. By doing so, we aimed to better convey the significance of our findings and their unique contributions to the field. We valued this input and were committed to addressing this concern in our revised manuscript. Thank you for your guidance.

Comments 6: Once these revisions are made, I will be happy to conduct a more detailed review. At this point, the paper lacks the necessary organization to be easily comprehensible as a research paper

Response 6: We now believe this article has been improved enough to be considered for publication in the Remote Sensing Journal. We’ll be happy if you could conduct a more detailed review, and answer your comments and questions.

Response to Reviewer 4 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: Line 112: Delete “2. Equipment and Methods”

Response 1: Line was deleted

Comments 2: Study zone – please change to “study area”

Response 2: Section title is changed as recommended  (L121)

Comments 3: Unclear here. Did you have the flood detected for the entire Madagascar? This is also the case when you are getting the data and processing it with GEE. However, your calibration and results analyses are focused on much smaller regions or areas (similar to Fig. 5 )

In your results though, it was showing entire Madagascar, but the analyses were based only on certain areas, not for the entire country. And so as the calibration. Please include a map identifying these and marking these areas, to see the geographic context and extent (similar to Figure 5). Also mark areas for calibration and testing

Response 3: In response to the reviewer's query, we conducted flood detection for the entire Madagascar during the cyclonic event. However, results analyses were focused on specific, smaller regions or areas within Madagascar. This approach allows for a more detailed evaluation of our processing chain's performance in regions where reference flood maps were available for comparison. The Madagascar-wide flood detection serves as an overarching assessment of our method's effectiveness. Some detected floods may not be related to the mentioned cyclone event due to the extended period detection. We emphasized these distinctions more explicitly in the revised manuscript to ensure clarity. (L129-136)

Figure 1 has been modified to help readers to locate calibration and analyses sites.

Comments 4. Lines 183 to 202. These are more for the validation. I suggest lift this entire part under 2.3.6 where you have the calibration and validation parts, so that it will be clearer how you directly use these data in these methods

Response 4: For better understanding, we moved lines on the recommended section.(L334-351)

Comments 5: Lines 203 to 211. You mentioned that “In order to provide practical and rapid information to risk managers, data from European Space Agency (ESA) World Cover [40], OpenStreetMap (OSM) [41] and Google 204 Open Buildings [42] were employed…:” But what is missing in this part is the relevance of these data in your study. Did you use them? If so, in what way? If you did not use them, then, omit this part

Response 5: We've removed this section as it wasn't complete enough, and added the possibility of integrating additional data to refine the analysis in the discussion section

Comments 6: Lines 213-214 (Processing chain): You wrote: “To make everything less compute-intensive, processing will be applied only to flood prone zones pre-determined according to topography.” How was this exactly done using the topographic information. It is also unclear how it exactly connects to the next sentence.

Response 6: Thank you for this note, we reformulate the paragraph

Comments 7: Lines 217-218 (Processing chain): “The choice of imagery will have a significant impact on results.” Which imagery do you exactly refer to here when you mention this part? Which images did you select then and why? Clarity on this step

Response 7: A phrase stating that selection procedure will be detailed later in the section is added

Comments 8: (Delineating flood prone areas): Start with the description of the DEM data (source and resolution from lines 246-247) in this part as the methods you are applying will all depend on the DEM.

Response 8: Corresponding sentences are moved earlier in the subsection

Comments 9: (Radar image filtering). In your results, you have shown the differences of using the different filters. However, in your methods, you haven’t mentioned clearly about each of these filters and what they do or their advantages for this purpose. From what is seen from the results section (Table 1) the differences were not so significant

Response 9: a description of each filter is added and table 1 is modified (comment 16) to demonstrate more the effect of image filtering (L277-289)

Comments 10: (Radar image filtering) In the results, you also mentioned multidate filters. What are these or what do you mean by these. Describe and clarify so that they can be matched as to what you have written in the Results section.

Response 10: A definition Multidate filter is added with a further explanation of radar filters purpose. (L277-289)

Comments 11: Lines 251-268 (Choice of reference image) – where did you use this reference image? Was it for the validation part (i.e. the validation data)? If it was, then put it under the same section. Otherwise, if you are referring to this image as the ones you used for taking out the results, you do not call them as reference images, to avoid confusion with validation data (i.e. for comparing results). You can write them simply as the sentinel images.

Response 11: Thank you for your comments. These images are called reference images because they are compared to post flood images. To avoid invoked confusion, we changed it to reference Sentinel-1 images. We also add additional definition in the beginning of the parent section

Comments 12: Sec 2.3.5. Processing chain overview – make sure that you briefly describe the figure that is shown.

Response 12: A brief description is added before the corresponding figure (L305-313)

Comments 13: Lines 298-300 (Calibration and validation of results) - which exact data was this?

Response 13: Explanations regarding the data to be compared have been included to provide a better understanding of the context and purpose of comparisons.

Comments 14: Lines 301-304 For a more unbiased validation, 100 random points were generated for each class 301 (flooded and non-flooded). Optical and radar images, along with water indices, were visually interpreted to establish the ground truth state of each point. This was carried out 303 while considering the available information on flooding in the study area.

Response 14: lines are deleted due to a duplicated information

Comments 15: (Calibration phase) – Figure 3 and 4. I will suggest to hatch the symbology for using both data, to see where they exactly overlap or not in the results. To see clearly whether they overlap in those areas being represented or not. The layers on top will take over the visualisation.

Response 15: Thank you for the suggestion. However, we had also considered hatching the symbols before submitting the paper. Unfortunately, the rendering of the small scattered flood patches was not visually effective with this approach. Legend is modified to improve reader understanding

Comments16: (Table 1) for these results, the percentage, is it percentage in area? Please write the total and also the actual size so that it can be judged based on the percentage if there are really significant differences in these results.

Response 16: The difference is not notably reflected in Table 1, primarily due to the dominance of large non-flooded areas in the accuracy assessment. However, these areas are predominantly derived from HAND data, which is not influenced by radar image processing parameters. To give greater consideration to misclassifications, even if they are relatively less represented, we computed the Kappa statistic instead of overall accuracy. As a result, the differences are now more evident. We believe this approach offers better readability than adding columns for actual size and total size for each filtering framework.

Comments 17: Comparison with other flood Copernicus (Figure 6). The maps only give a visual comparison. Include the total area flooded based on the method provided, and the validation data used to be able to quantify clearly the differences.

Response 17: Thank you for your suggestion. In the revised paper, we included the area of flood detected by our method only, by The EMSR, and by both (L430-439)

Comments 18: Same as previous comment for Figure 7

Response 18: Areas of detected flood are also added in addition to the kappa.(L458-459)

Comments 19: (Example applications) – how relevant are these land cover types for emergency responders? What will be interesting here are the built-up areas, roads, etc that can be affected directly, in addition to cultivated lands (for economic impact)

Response 19: We've removed this section as it wasn't complete enough, and added the possibility of integrating additional data to refine the analysis in the discussion section

Comments20: (Discussion) How will you put this study in context to earlier studies that applied the same method? Comparison:

Response 20: Comparisons with some similar recent works are added to the discussion (L170-484)

Comments 21: Sentence structure - one-sentence paragraphs. Avoid them

Response 21: More sentences are added to one-sentence paragraphs by adding new informations or splitting paragraph to several sentences