Mapping Urban Floods via Spectral Indices and Machine Learning Algorithms

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This research aims to address the issue of urban flood disasters by analyzing and studying remote sensing images, combining various machine learning methods with spectral index, and proposing a new spectral index that has several innovative and practical applications. However, the following issues still exist:

1. An addition to the study area scope map is suggested in section 2.1.

2. The figure number and name in Figure 1-7 are missing; it is advised to read the entire text.

3. This work offers a new spectral index, CNDWI, details how to concatenate the indices, and provides an additional explanation of the procedure in Section 2.3.

4. "Marching learning algorithms," on line 157, is spelled incorrectly.

5. The rationale for the two standard deviations and thresholds is provided in Table 5-8. It is recommended that you include an additional explanation.

6. Since the equipment utilized and the machine learning algorithm's time are strongly related, it is advised to add to the equipment configuration used in this paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors use four different machine learning algorithms to assess the influence and accuracy of two indexes, that are used to evaluate the impact of floods. The main finding is the validation of existing measures to identify floods in an urban environment.

The topic, methodology, and used data are up to date. The selection of the study area is relevant, as Brisbane is one of the main metropolitan areas in Australia which has historically experienced several large-scale floods.

To my knowledge, there was no comparable study published about this topic using the same methodologies. Another study (Levin and Phinn, 2022) which has already studied the 2022 flood in Brisbane using remote sensing data is cited in the manuscript. The used methods are adequate, and the illustrative figures are useful. Overall, the article is written well and represents high scientific standards. I really enjoyed reading it.

Overall, I recommend accepting the manuscript, after some minor revisions are done. Here are some comments that might help to improve the quality of the manuscript:

1) The titles of the figures on pages 9-11 are missing. From the text, I assume these are figures 4-7.

2) In tables 7 and 8, the train/test times are mentioned. What is the measurement unit? Seconds? It will be helpful to mention it somewhere below the table or in the text.

3) Typo: ANVOA (line 302)

4) What are the direct practical implications of the study? The introduction and conclusion mention keywords such as sustainable development, but it did not become clear how the presented results can achieve this. Maybe the authors can add a paragraph about this in the introduction and/or conclusion. Can the results help to make more precise estimations about the economic costs of flood disasters? Can the results help to improve the city planning? How? Are there other practical real-life implications of the results? These were some open questions from my side.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Overall: the aim of this paper – to learn how different spectral indices and ML algorithms perform – is worthwhile and needed.  The Introduction and Methods need minor improvement to appropriately convey what was already known before this study.  The results appear valid and relevant to achieving the aims; but, more clarity is needed to explain them.  The Discussion and Conclusions are weak: they largely repeat what has been written earlier in the paper without providing greater perspective on their results, identifying remaining problems, or proposing future advances and applications that could build on their work.

Specific issues:

Line 43: “notable inaccuracies” in application to urban environment.  What these are and their causes need to be made explicit, not just referenced.  Since this study aims to do better, problems to be addressed from past research should be brought out.

Lines 58-60: PlanetScope satellites capture “the Earth” “on average every day.”  But how representative is this for all land areas/cities?  A resolution of “three meters per pixel” seems to be problematic in accurately capturing flooded areas in urbs with tall buildings along narrow streets.  It should be noted that urbs which consist mainly of “suburbs” and “houses” as is the case in Brisbane (lines 123-124) are not typical of the situation in many other places.

Lines 104-107: It is said that the “case study” or the “floods” “contains an untested ground truth.”  It’s not clear in what sense a case study or a flood can contain a ground truth, or in what sense a ground truth is untested?  This needs clarification.

Lines 138-143 & Table 4: It should be stated what data were used to construct the Brisbane flood map and what its spatial resolution is.  Table 4 only gives pixel-based info; how large is a pixel in meters? What were the raw data on which the flood map was based, and how does the resolution of the image relate to the actual density of information on the ground?

Section 2.5.2: The accuracy ratio explained in 2.5.1 is straightforward.  The F1 score in 2.5.2 is much less intuitive, and readers should benefit from an explanation in plain English of what aspects of the difference between image and ground truth this score captures.

Figures 4-7:  These lack captions.  More importantly, they are not suited to the study, which emphasizes accuracy relative to ground truth.  These figures allow the reader to compare predictions to one another, so we can see which predictions are similar.  But we can't see which predictions are accurate and which are not.

Section 2.6.1: The formula is not quite correct; a sample SD is being calculated, not a population SD.  The presentation in this section needs to explain the notes below Tables 5-8 which refer to “first standard deviation” and “second standard deviation,” and “upper thresholds” and “lower thresholds.”  Procedures are unclear!  Likewise, Figures 8-11 showing SDs need to be explained better.  I suggest pairing the tables and figures; i.e., show Table 5 with Figure 8, Table 6 with Figure 9, etc.  Axis labels on these figures need to be larger!

Discussion: Mostly summarizes results.  Lacks substantial discussion of the results to show improvement with respect to prior work, issues they discovered with their research that they plan to improve, or the potential for themselves or others to build on the work shown in this paper (this point is addressed slightly in the Conclusions).  For example, this work emphasizes global measures of accuracy, which might not be the most useful for city planners or first responders.  What can be gleaned from studying the types of landcover/land use that are especially well or especially poorly predicted?

Comments on the Quality of English Language

English issues are very minor and require no special editing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

According to earlier comments, it seems like the authors have improved and corrected the work in addition to including new material and convincing arguments. The manuscript still has a serious problem with wording duplication, nevertheless, with a current percentage of 34%. It is crucial to reduce this duplication to improve the overall quality and originality of the document.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf