Inundation Map Prediction with Rainfall Return Period and Machine Learning

Round 1

Reviewer 1 Report

General comments to the Authors:

Generally, the subject of flood prediction in urban area is an important issue, and this manuscript tackles this issue, predicting the inundation maps in the Gangnam area of Seoul, Korea. However, in my opinion, this contribution has some deficiencies with respect to the (1) structure of the manuscript, (2) terminology used across the whole text, and (3) the applied methods. The approach is original, but the text in many parts is not clear. Unfortunately, the proof-reading was not done before the submission. The whole manuscript should be shortened, giving only the main points of the applied methods, and simultaneously referring to the original papers. As the reader is not always familiar with hydrometeorological features of the analyzed region, thus it would be useful to give some basic hydrometeorological features in the description of the study area, before You discuss the Maximum Probable Precipitation in the following sections.

(1) Comments with respect to the structure of the manuscript:

• The procedure of the conducted research should be moved from the Section 1 (Introduction) to the Section 2 (2. Study Area and Methodology), and clearly explained. In the first section You should only introduce the problem by giving  the general goal of the research as the development of  specific models for the purpose of predicting both the Maximum Probable Precipitation and Flood Patterns.
• Unfortunately, the description of the goal is presented in Lines 227-232, in section 3.Application, 3.1 Target Rainfall: “The purpose of this study is to estimate the frequency with the consideration of rainfall duration and to predict the flood map about observed rainfall in September 21, 2010 (Fig. 7). The rainfall data that was obtained from the AWS (Automatic Weather System) in Gangnam, and the rainfall between 12:00 to 18:00 were selected. In this study, the maximum rainfall for each time duration was used as target data. The duration of rainfall for the study was selected as 1, 2, 3hr. The total rainfall was 71mm for a duration of 1hr, 128.5mm for 2hr, and 204 mm for 3hr.” Actually, the information of data used in the analysis should be given before the Section in which results are presented.
• The title of the Section (2.1. Study Area and Verfication) is not clear. What do You verify here?? The whole section 2.1 with respect to the term “Verification” is not understandable.

  After the Introduction, You should first describe the study area, giving more details on hydrometeorological features of the analyzed area (including information on precipitation for this area, usual flood events, duration of historic rainfall causing flash floods, how quick do they form giving the reason to consider further the rainfall of 1-2-3hr duration, what is the geometry/channel system, imperviousness, if You have any hydrometric recording, etc.), then flow chart of the procedure, full description of the applied methods, and finally results of your analysis -application.

•  

  Line 242: “For this reason, an SOM was applied to produce 16(4×4), 25(5×5),and 36(6×6) expected flood maps(Fig. 8).” This should be moved to the methods section.

 

(2) Comments with respect to the applied terminology:

• The following terms are used which need verification:

  “Rainfall frequency”

  “frequency estimation for observed rainfall”.

  “Frequency estimation was performed with PNN(Probabilistic Neural Network )”

  “frequency of occurrence for observed rainfall” etc,

  “frequency of inundation map”

  “The design rainfall for the 2, 3, 5, 10, 20, 30, 50, 70, 80, and 100yr frequency was calculated” (meaning probably 100yr return period…)

  “To estimate the frequency of rainfall events that caused actual flooding in the study area, observed rainfall in September 21, 2010were entered into the trained PNN.” (not frequency but rather probability…)

  “5yr frequency”, “100yr frequency” ---- rather 100yr return period…

  but in the Figure 1, the term “probability” is used (Probability Rainfall scenario).

  In my opinion, the term “frequency” is wrongly used; in my opinion, the Probabilistic Neural Network is used to estimate probability of maximum rainfall (exceedance probability ?) rather than the frequency.

  One may estimate the probability using the frequency distribution.

  Therefore, the formulations like: “Frequency estimation was performed with PNN (Probabilistic Neural Network )…” is not clearly formulated. But of course, one may estimate “the frequency and extent of inundation map with consideration of observed rainfall” (but here again probably also probability is meant by the Authors…)

  Conclusion: The Authors should differentiate between the term “frequency” and “probability”, and make a correction across the whole text. Otherwise the context of such formulations is hardly understandable.

(3) Comments with respect to the applied methods:

In many parts of the manuscript, the applied methods are not clearly described, eg:

• Line 123: “To make design rainfall…” – not clear what does it mean “to make”
• Line 128: “Total 120 design rainfall was considered with different durations(1, 2, and 3 hr).” – not clear what is meant here by design rainfall.
• Table 4: Result of Frequency Estimation using a Probabilistic Neural Network(PNN): Please explain what is the Database (1) and Database (2)
• Line 126: The time distribution was performed with the Huff’s four-quantile method" Does the 4th q. give the highest flood volume and the highest peak flow?

(4) Other comments:

• “The Seocho and Sapyeung flood pumping stations form a part of the study area and discharge water to the Banpo river.” If this is important, please indicate the stations and the river on your map.
• Lines 101-103: “Fig. 2 and Fig. 3 show the location of the study area and the result of two-dimensional flood simulation with September 21, 2010 rainfall.” If You refer to former studies, the citation should be made. It is not clear if this concerns your study or former one.
• “the frequency of heavy rainfall is increasing annually” if this is really the case, then probability of rainfall occurrence is changing year by year or ?
• Fig 9 Result of Support Vector Regression(Gaussian Kernel Function): Please be more exact when describing (a), (b), (c)
• Fig 10 Inundation Map Simulated with the Two-dimensional Flood Analysis: Please be more specific-which rainfall data is used here
• Fig 11 Inundation Map Prediction with Machine Learning(Conjunction model 279 of Support Vector: Please be more specific-which rainfall data is used here

Author Response

Thank you very much for reviewing the paper.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The Authors propose in this manuscript a methodology for estimating the unknown frequencies of specific observed rainfall event and predicting the urban flood map. Their approach to achieve this goal by using PNN and SOM trained using results of SWMM simulations is original. The Authors also try to estimate the flood volume with SVR. Here I have some general comments.

a/ The term ‘frequency’ is not appropriate. By definition frequency is the number of occurrences a particular period of time. So it cannot be expressed in years. What is referred by the Authors should be called ‘recurrence interval’ or ‘return period’.

b/ The term ‘flood volume’ is also misused in the manuscript. Definition says that flood volume is the amount of water which flows through the cross-section during the duration of the flood current. Please provide your own definition if it means something else. However it cannot be measured in m³/s.

c/ The section 2 should be restructured. The first subsection cannot present verification of the model. First the (SWMM?) model should be described in details: its extend, boundary conditions, input/output parameters, calibration and then verification. Are the numerical models presented in section 2.2 part of SWMM? Please explain how they interact with each other.

I advise to revise the manuscript by native English speaker. Some sentences are difficult to understand. The word ‘which’ is not used correctly.

Additional specific comments which could improve the manuscript are as follow:

1. L44 The citation is missing on reference list.
2. L108 On Figure 3a replace word DEM with Elevation. Is there really 205 m difference in the elevation? On Figure 3b draw the urban catchment boundary which is modelled.
3. L140 Abbreviation SWMM was already expanded in L64. No need to repeat it.
4. L191 Typing error in the name of Vapnic.
5. L214-215 The sentence is unclear. Please rewrite.
6. L216 Please explain what is ‘maximum flooded area’?
7. L217 Is 4x4 a dimension or number of clusters?
8. L249 Please discuss what is presented on Figure 8 and what it is used for?
9. L254 what it means ‘The results area’?
10. L257 Please consider to move 3^rd column on 2^nd
11. L278-279 Please separate the images and draw the catchment area. The legend is required.
12. L290-300 Please rewrite as it is rather summary than conclusion.
13. AWS was already expanded in L229.
14. L318-319 The same sentence is in L313-314.

Author Response

Thank you very much for reviewing the paper.

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear Authors,

thank You for introducing the changes. Only minor remarks concern the following:

1. Line 90: "Validation" term is used, but in Line 111: "Verification" term is used. Please consider unification.
2. Line 81 : "the occurrence frequency of the predicted inundation map" Please check: Frequency or rather probability?
3. Line 253: In title of the table: Instead of "Max. flood area..." please write: "Maximum flood area...", abbreviation might be inside the table, but title should have the full word "maximum"

Author Response

Thanks for your review, please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

It’s good to see how the manuscript has improved after first round of comments. Unfortunately I am not fully convinced. Here are my concerns:

1/ I cannot agree with the statement that ‘flood volume’ is measured in m³/s. The Autors replied that “The flood volume indicated in this study means the sum of 10 minutes unit overflows (m³/s).” According to SWMM User's Manual Version 5.1 the result of Node Flooding could be:

“Total flood volume (million gallons or million liters)” which is the volume

or

“Maximum flooding rate (flow units)” which is volume per unit of time.

So please double check what you are referring to.

2/ Please replace DEM with Elevation in the Legend of Fig. 2a.

3/ If the content of 2.1 is about validation then please replace verification with validation in the title of Fig. 2.

4/ No need to write 120 in the title of Table 1.

5/ The English has improved but there are still grammatical errors. Replacing 'which' with 'that' does not change anything.  Please pass this manuscript for English correction.

Author Response

We appreciate for your review, please see the attachment.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

I strongly recommend the use of SWMM-CAT for this project, in predict and hydraulical analysis portion of the project. the study should be revised and resubmit.

 

Author Response

Thank you for your review.

 

Answer is uploaded.

Author Response File: Author Response.docx

Reviewer 2 Report

Abstract

The reported results of 21-y/2-h and 91-y/3-h come a bit out of the blue. Could you provide more general results or explain why you focus on those in the abstract?

 

Introduction

Due to the extensive use of passive formulation, at some places it is not clear to me, what has been done by previous studies or is presented in this paper.

For some references it is not clear to me, how they relate to the presented study, besides of applying a machine learning method on precipitation or flooding data. It may help to elaborate a bit on the conclusions of the cited references or to report on the context to the study at hand. Also, the literature review could be a bit more extensive. 

 

Methodology

The description of the applied methods needs to be improved. I consider myself to be quite experienced with machine learning approaches. Yet, it is difficult to understand, what exactly you have done. E.g. what are the input nodes in the PNN. Rainfall data is not a sufficient description. Which variables/units are used as input and output? What are the categories? Maybe a visualization or example may help. In terms of reproducibility, please provide the size of the individual layers (or how the size is chosen).

Similarly, for the other methods it is hard to get, what you have done.

 

Application

Please, provide more information about your data. What is the source for the "real storm events" (line 206)? What kind of rainfall data do you consider? What kind of data is reported by the NDMS?

How does your goodness of fit score deal with false positives (= an overestimation of the flooded area)? By consistently overestimating the flooded area, more NDMS will be included in the calculated flood area.

 

Conclusion

The conclusion is rather a summary than a conclusion. Please, elaborate the relevance of your study.

 

General Comments

Use more informative captions for figures and tables.

The size of some figures is way to small, e.g. Fig. 5, 6, 7.

Author Response

Thank you for your review.

 

Answer is uploaded.

Author Response File: Author Response.docx

Reviewer 3 Report

Please see the attached file

Comments for author File: Comments.pdf

Author Response

Thank you for your review.

 

Answer is uploaded.

 

Please check again.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

 I recommended EPA SWMM-CAT to be used for this project to justify the achieved result is correct and sightable. However, the author explained they are new for this software and suggested to write another journal using this software. Therefore, I deemed that the study presented is incomplete and the claimed results have not been justified using industry available for FREE software. For that reason at the ground of the research NOT be conducted correctly I recommend rejecting the paper. 

Reviewer 2 Report

Abstract

It’s hard to get from the abstract, what precisely the paper is about. E.g. ‘Flood maps for maximum historical rainfall events and 10-, 30-y frequencies have been presented to estimate the risk factors; however, it is difficult to suggest an intensive analysis due to climate change and the appearance of various rainfall events.’ Rather refer to what you show in the paper, not what has been done in the past. Further, you do not deal with climate change (and potential changes in rainfall frequency) and you do not consider various rainfall events, but rather study a specific rainfall event.

What do you mean by “observed rainfall”? Observed by whom, where, …? Several rainfall events with various intensities have been observed.

The sentence about the results (line 18-20) is not very informative, since - so far - the reader has no idea how ‘area fitness’ is defined. You may present your results in more general words.

 

Introduction

Again, what do you mean by “observed rainfall”?

Still, it is difficult to get, what has been done in previous studies and what is part of this paper. E.g. “The current state, cause of past damage, and risk factors are analyzed for selecting candidate sites to accurately represent flood hazards in urban areas. Flood simulation results for the design rainfall (10-, 30-y frequency) and the historical flood event are used to analyze the current risk factors.” seems to relate to previous studies. PROVIDE REFERENCES!

Also provide a reference for KMA (line 30).

 

Methodology

What is your data base? A data section should be added. It’s quite difficult to understand the methods without seeing the data or at least getting an understanding of the data. Most data are explained first time in the application section. Provide references for all your data sources.

Which data base do you use to train the PNN? How do you get the frequency of design rainfall / Where from? What kind of ‘design rainfall’?

The information about the available data is important for me to understand the proceeding and I don’t want to search for data information in the application section (especially when I read the paper the first time and do not know that the data are provided later on). At least refer to the corresponding tables.

The description of the applied methods still needs a thorough revision. Just adding some sentences is not sufficient.

Please elaborate on why use the applied methods. What is their advantage compared to state-of-the-art approaches. Why not using simple regression to estimate rainfall frequency? What do you gain by using PNNs? (also explain for the other methods) You may also adjust the literature review to give a better overview on what are state-of-the-art approaches and what is the benefit of using machine learning techniques.

Make sure your results are reproducible. Provide the required information.

 

Application

Still it is not clear how your validation deals with an overestimation of flooded area. I would receive a very high fitness, if I simply assume that a very large area is flooded (since most NDMS points would be in that area).

Section 5.3: Check your data. The intensity for 3h is higher (68mm/h) than for 2h (64mm/h). Did you select the highest intensity intervals?

 

General Comments

Use more informative figure captions!

Size of the figures is still to small.

Language needs to be revised.

Reviewer 3 Report

The manuscript was revised by the authors, still, this new version needs further work

The introduction should give a context with the review of previous studies; however, this manuscript is not clear why the cited literature is relevant. As mentioned in the authors instructions, the introduction should define the purpose of the work and its significance, including specific hypotheses being tested. For instance, the discussion in lines 460-478 should be placed in the introduction to give an idea of the context of the analysed problem. In addition, the purpose of the study should be clearly established and justified: for example, why it is necessary to estimate the precipitation return period in real time?How this information will provide better inundation maps? The methodology does no describe the data used in this study. The methods are described in a general way. For example, the study does not provide information of the choose of the number of hidden layers and nodes in the neural network, neither the performance obtained. Please make sure that your methodology can be really understood by the scientific community in order to assure that the results are reproducible. The abstract mentioned that “it is difficult to suggest an intensive analysis due to climate change and the appearance of various rainfall events”; However, this s confusing as the manuscript is not related to climate change. Figures still have problems, there is not enough information in the captions and they are very small.