Predicting Inflow Rate of the Soyang River Dam Using Deep Learning Techniques

Round 1

Reviewer 1 Report

This study applied LSTM technique to predict the daily inflow rate in the Soyang River Dam. This study

This study only applied 15 years of daily meteorological data and predict the inflow rate over the next two days. I think this study will be useful to readers after revision.

Major Comments

1. This manuscript is very similar to Hong et al. (2020). It is necessary to emphasize the novelty of this study and revise the title. Particularly, it needs to show the improvement of this study comparing with Hong et al. (2020).
2. I do not find the LSTM model training results. So, I think it should add the model training test results.

Minor comments

1. in line 54, add abbreviation (MLP) of multilayer perceptron.
2. in lines 227 and 313, Lee et al. [9] are not matched with the relevant reference. Check and revise.

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with your comments and answered about your questions as follows:

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper entitled "Predicting Inflow Rate of Soyang River Dam using 2
Deep Learning Technique" is within the scope of water journal. The paper was well-organized and written. However, before the paper is undertaken any decision, some sections need improvements:

(1) Quantitative results of the best models can be merged to the abstract section.

(2) Applications of machine learning models and deep learning techniques in the various fields of science need for improving introduction section:

-Decoding Clinical Biomarker Space of COVID-19: Exploring Matrix Factorization-based Feature Selection Methods

-Reliability assessment of water quality index based on guidelines of national sanitation foundation in natural streams: integration of remote sensing and data-driven models

-Pipe break rate assessment while considering physical and operational factors: a methodology based on global positioning system and data-driven techniques

-A Novel Multiple-Kernel Support Vector Regression Algorithm for Estimation of Water Quality Parameters 

(3) Results need improvement for better illustration of robust artificial intelligence models by using F-test, reliability analysis, and uncertainty analysis, as seen in the following papers:

-Comparative study of different wavelet-based neural network models to predict sewage sludge quantity in wastewater treatment plant

- Receiving More Accurate Predictions for Longitudinal Dispersion Coefficients in Water Pipelines: Training Group Method of Data Handling Using Extreme Learning Machine Conceptions

 

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with your comments and answered about your questions as we attached:

Author Response File: Author Response.pdf

Reviewer 3 Report

Manuscript is devoted to predicting Inflow Rate of River Dam proposing a new Long Short-Term Memory model. Manuscript present the method using the data of a case study which is from the largest multi-purpose dam in Korea. It is indeed uncertainty due to the global warming to accurately predict the Inflow. However the avalability of an extensive time series data and deep learning would be essential. Manuscript proposes using the meteorological, dam, and weather warning data to predict the inflow rate over the next two days. The study design, data and method shows novelty and relevance. Following minor comments would be essential to improve the quality and clarity of the manuscript.   Research gap and the sate of the art in respect of the former DL and machine learning methods for inflow prediction are not included in the introduction. Thus novelty is yet to be described and elaborated.   The LSTM method had not been described adequately. several references to deep learning methods can be useful to describe the LSTM, e.g., "list of deep learning methods".   The validation is not elaborated.   Manuscript wont present an effective comparative analysis with other DL or ML methods. Thus it is essential to point this out as a limitation and future of this study.   Why LSTM had been used not for instance Random Forest, or other DL methods? describe the reason behind proposing LSTM.   Insert an acronyms and abbreviation table.

Although the paper has appropriate length and informative content, several parts must be improved and written in better grammar and syntax. It would be essential if authors would consider revising the organization and composition of the manuscript, in terms of the definition/justification of the objectives, description of the method, the accomplishment of the objective, and results. The paper is generally difficult to follow. Paragraphs and sentences are not well connected. Furthermore, I advise considering using standard keywords to better present the research.

The research question, method, and the results must be briefly communicated. The abstract must be more informative. I suggest having four paragraphs in the introduction for; describing the concept, research gap, contribution, and the organization of the paper. The motivation has the potential to be more elaborated. You may add materials on why doing this research is essential, and what this article would add to the current knowledge, etc. The originality of the paper is not discussed well. The research question must be clearly given in the introduction, in addition to some words on the testable hypothesis. Please elaborate on the importance of this work. Please discuss if the paper suitable for broad international interest and applications or better suited for the local application? Elaborate and discuss this in the introduction.

State of the art needs improvement. A detailed description of the cited references is essential. Several recently published papers are not included in the review section. In fact, the acknowledgment of the past related work by others, in the reference list, is not sufficient, e.g., "Prediction of hydropower generation using grey wolf optimization adaptive neuro-fuzzy inference system". Consequently, the contribution of the paper is not clear. Furthermore, consider elaborating on the suitability of the paper and relevance to the journal.

Limitations and validation are not discussed adequately. The research question and hypothesis must be answered and discussed clearly in the discussion and conclusions. Please communicate the future research. The lessons learned must be further elaborated in the conclusion by discussing the results to the community and the future impacts. What is your perspective on future research?   

   

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with your comments and answered about your questions as we attached:

Author Response File: Author Response.pdf

Reviewer 4 Report

Kindly  revise according attached comments.

Comments for author File: Comments.pdf

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with your comments and answered about your questions as we attached:

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have revised based on the reviewer's comments very well. The manuscript is now qualified to be published. 

Congratulations!

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with other reviewer's comments. 

Followings are the summary of the modifications:

• Section 2.4.7 algorithm are presented in algorithmic format. In addition, we include a flow chart for training Seq2Seq LSTM.
• We changed all Figures. Especially, the font in figures is changed to be more legible.
• We added a unit for RMSE and MAE in Table 8, 11,15, and 16.
• We added more graphs of baseline models. These graphs are used to compare the proposed model and the baselines.

Reviewer 2 Report

Accept as is

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with other reviewer's comments. 

Followings are the summary of the modifications:

• Section 2.4.7 algorithm are presented in algorithmic format. In addition, we include a flow chart for training Seq2Seq LSTM.
• We changed all Figures. Especially, the font in figures is changed to be more legible.
• We added a unit for RMSE and MAE in Table 8, 11,15, and 16.
• We added more graphs of baseline models. These graphs are used to compare the proposed model and the baselines.

Reviewer 4 Report

Authors revised major comments properly.however still some issues need clarification and justification to recommend this paper for publication.

section 2.4.7 algorithm must be presented in algorithmic format i.e. input, output then logic or code flow.In addition present flow chart about inflow prediction through seq2seq lstm.

Figures still not in best presentation form.Authors should use clearly readable fonts figures.In figures,different fonts used,some places bold fonts used.Not acceptable.Should use uniform font style in all figures and keep figure aspect ratio properly.do not paste or stretch figures.

In tables,should mention the units of RMSE and MAE.

With proposed methods, you used many base models for comparison but where is graphical comparison of these models with proposed models.Table 15 shows your proposed model dominancy, on these results, you should draw time variation graph,violin plots  and scatter plots .Therefore it could be justify and can see proposed model performance with other models.For your clarity, in time variation graph,scatter and violin plots,observed data should include,therefore could compare which models give more closely results to the observed data and in addition for scatter plots,trend line equation and r2 value should mention on plots to see models performance.Kindly do these necessary changes.

Authors did not describe the literature about MARS model to justify it.Use the following studies in your Introduction section, about the bench marked model MARS's acceptable performance in  modeling studies:

https://doi.org/10.3390/w12102927

https://doi.org/10.3390/w11102060

https://doi.org/10.1007/s00521-020-05164-3

https://doi.org/10.1016/j.asoc.2020.107008

https://doi.org/10.1016/j.jhydrol.2019.123981

https://doi.org/10.1016/j.jhydrol.2019.124371

Author Response

Thank you for your precious comments. We revised the manuscript in accordance with your comments and answered about your questions as we attached:

Followings are the summary of the modification and answers to comments:

• Section 2.4.7 algorithm are presented in algorithmic format. In addition, we include a flow chart for training Seq2Seq LSTM.
• We changed all Figures. Especially, the font in figures is changed to be more legible.
• We added a unit for RMSE and MAE in Table 8, 11,15, and 16.
• We added more graphs of baseline models. These graphs are used to compare the proposed model and the baselines.

Author Response File: Author Response.pdf