An Effective Rainfall–Ponding Multi-Step Prediction Model Based on LSTM for Urban Waterlogging Points

Round 1

Reviewer 1 Report

General Comments

This is an interesting article comparing the performance of MSE, MAE and MSLE for rainfall-ponding multi-step prediction model. However, the numbering of the section and sub-section of the manuscript can be improved. The abstract should be improved by adding numerical result, instead of general statement. The introduction is too specific at part, and the general issue should be elaborated more and written clearly. The arrangement of the sub-section can be improved. For example, I don’t understand the need for many sub-section in the section of “Algorithm verification”. The title of the section also should be improved to give a clear picture what the results and discussion entails. This can improve the flow and continuity of the research process from the beginning until the end, particularly on the results section. At this point, the authors should reconsider the manuscript for major revision before it can properly publish. The specific comments are provided below. Please provide line numbering for quick reviewing process.

Specific comments

Abstract

·       Please specify the accuracy predictors used for validation and evaluation, as well as numerical results.

·       What do you mean by good applicability of LSTM model? Please give numerical result.

·       Please specify the numerical result of trained MLSE, and MAE. Otherwise, the abstract is too general.

Introduction

·       Introduction section should be “1. Introduction”.

·       Please revised “gathered more and more people and wealth”

·       Please revised “more and more extreme weather”

·       Extraordinary Storm > extraordinary storm

·       “In the face of a serious disaster, the problem that needs to be solved is the accurate and rapid prediction of ponding to provide a basis for disaster prevention and mitigation”. There is no continuity for this sentence. Please elaborate more.

·       What do you mean by “gradient disappearance or explosion”? Please elaborate more on this, to ensure a general reader can grasp the concept.

·       At the end of the paragraph, please briefly discuss where the study is conducted, the data used and experimental step to be taken. Indicate why do you select this study area.

Modeling principle

·       h^(t-1)denotes > h^(t-1) denotes. Please make similar correction for the rest of the manuscript.

Data source

·       This section should be “Study Area and Data Sources”

Data pre-processing

·       the raw data need to be data pre-processed > the raw data need to be pre-processed

·       Please use a proper sub-section numbering. For example, “1) Continuous series segmentation” > “2.2.1 Continuous series segmentation”. Please make correction for the rest of the manuscript.

·       Figure 3. Please label the x- and y- axis. Please make similar correction for other figure.

Analysis of prediction results

·       Fig3 > Fig.3. Please make similar correction throughout the manuscript.

·       Table 1 can be improve.

·       Tab 1 > Table 1

Conclusion

·       Please revised “has received more and more attention”

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

This study uses long short-term memory (LSTM) neural network to model rainfall-ponding. This research claims to utilize multi-step prediction as a novelty, however, the reader cannot understand the subject since it is poorly explained and not supported by relevant evidence. Some of the arguments need to be changed or improved before it can be considered as a scientific paper. Therefore, as a reviewer, I believe the paper should be rejected. The following comments and ideas for enhancement are provided:

(1)         Innovation should be elaborated more clearly in the manuscript. You need to use more creative approaches to solve the real-world problems. Using LSTM to model rainfall-ponding and demonstrating its ability cannot be a novelty from my point of view.

(2)         You need to provide supporting material such as data, code, and trained model for your scientific work in the way that scholars can implement your proposed method easily.

(3)         All the figures in the manuscript need to have legends with units, more clear titles, and better descriptions.  

(4)         You must discuss more comprehensively about the achieved results and expand the conclusion part. For example, it was mentioned that the MAE can perform better as the loss function; you can discuss about the probable reasons and argue your hypotheses.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

When performing very well-established approaches it is almost mandatory to bring some new findings and/or an extensive testing. Honestly, I can't see any significant added-value in the methodology. Moreover, there are many issues with theoretical definitions and mathematical equations. There is an obvious lack of references (this topic is very well documented in the scientific literature) and some choices are not strongly motivated. The analyses and interpretation of the results are very poor and sometimes not relevant. The paper is quite well structured and the topic is of interest, but I unfortunately recommend the majors recommendation of the paper and I highly encourage a re-submission after rewriting the article in a simple way. However, i have the following comments that I hope the authors could address in the revision:

1.     In introduction the authors mentioned that LSTM has not been used much for multivariate. This is not true. Try to add the use of LSTM in multivariate in hydrology. 

2.     2)   Resampling : In this part the author have chosen to use the fill method to process the precipitation data. I suggest to give a little more details about this method and its validation

3.      1.3. Model framework : In the whole article the authors mentioned only the data splitting method and not the number of data. I suggest to add for each case the number of training data and the number of data of the validation with percentage.

4.     1.4. Hyperparameter setting : set dropout = 0.4 and  dropout = 0.3 Based on which method you have set these values. input layer units = 256, nodes units = 128 and the output layer units = 32 Based on which method you have set these values.

5.     2.1. Data source : According to the prediction objectives, this paper ……………… 1 hour, and the time interval of the original ponding data varies, and the accuracy is at  the "minute" or "second" level. I ask the authors to rewrite this paragraph. As it is, it remains unclear.

6.     Inconsistency in the exponent of equation (11). Please check them

7.     Why you fixed batch size at 256.

8.     The discussion of the results in Figure 3 remains incomplete. It seems to me that the results presented are not those of the validation. I think that the training parameters are not fixed during the validation.   Please indicate the results of the learning and validation parameters in the responses to the reviewers.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

General Comments

This is an interesting article comparing the performance of MSE, MAE and MSLE for rainfall-ponding multi-step prediction model. The manuscript has been substantially improved compare to the previous version. I’ll accept the paper with only few minor revision as shown in the specific comments. The specific comments are provided below.

Specific comments

Abstract

·       Line 41: MSE(mean square error), > MSE (mean square error),

·       Line 41-42: MSLE(mean squared logarithmic error)were > MSLE (mean squared logarithmic error) were

·       Line 46: “mae (mean absolute error),”. This is repetitive abbreviation

·        

1.     Introduction

·       Line: 78-79: “long short-term memory neural network (LSTM)”. This is a repetitive abbreviation.

2.3. Model framework

·       Line 174-175: Revised the sentence > The original data was randomly divided into training and test sets according to the ratio of 80% and 20%.

2.4. Optimization algorithm in model compilation

·       Line 193: Please give full description of ADAM.

4.1. Prediction accuracy index

·       Line 303-304: Repetitive abbreviation of RMSE, MAE, MAPE, and NSE

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

It seems that the paper was improved since last time, however, the authors don't want to provide supporting materials. At least they have to mention this sentence in their paper. 

Readers who are interested in our work and wiling to use our model can contact corresponding author.

Author Response

We thank the reviewers for the time and effort that they have put into reviewing the previous manuscript. We carefully read the journal's data availability policy, and according to the comments of the reviewer, a statement of the data availability is added to the revised manuscript. Please see the details (Line380-382).

Data Availability Statement: Readers who are interested in our work and willing to use our model can contact corresponding author. Data and codes used for this work are available from the authors upon reasonable request.

Reviewer 3 Report

The authors have provided answers to all my comments 

Author Response

We are very glad that you are satisfied with our modifications and replies. We thank you again for your time and effort on our manuscript.