Combining Multiple Machine Learning Methods Based on CARS Algorithm to Implement Runoff Simulation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript discusses a novel approach for machine learning based flood forecasting. I can see big efforts on analyzing and writing this research paper and suggest this paper to be improved within the reviewer’s comment.

 

My comments are following,

1. In the table 1 – LSTM neural network, the optimal value of epochs is 0.01 while the range of value is 50-300. I think the value of epochs cannot be less than 1. I think it is simple typo and hope the authors can revise it.

 

2. About normalization, only Min-max scaler was used in the normalization process, inconsistent with other items (machine learning methodology, evaluation metrics, etc.) that are compared and analyzed by applying many techniques. I hope the authors think about using other methods such as z-score so that it can be matched with evaluation techniques.

 

3. There are so many variables expressed in the charts (figures 4 and 7) that it is difficult to see the difference for each variable. It would be nice to improve the scale of the chart and the optimization of the line expression method if it simply shows the result that the derived results converge well without significant difference from the actual observed results.

 

4. Authors mentioned about evaluation metrics with 7 different types and 6 are selected and calculated as table 2. I think they are too many and the paper becomes complicated. That’s why it was too hard to understand the physical (or logical) meaning provided from each metrics. It seems that the table 3 shows the final results from your research and the table 3 shows only RMSE. Please make the number of metrics reduced or provide their clear description and physical meaning of them.

 

Thanks for your huge effort to develop the study.

 

 

 

Comments on the Quality of English Language

I am not a native speaker but I think the quality of English writing in the paper is fluent. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper deals with the topic of runoff forecasting. The study utilizes the Competitive Adaptive Re-weighted Sampling (CARS) algorithm, combining various machine learning models to construct a data-driven rainfall-runoff simulation model. A comparative analysis of forecast results from different types of machine learning models is performed to improve the accuracy of flood forecast.

The subject matter is quite novel and it addresses the issues of Water.

The manuscript is an original contribution, and the analysis on a real case study (i.e., the Bahe river basin, Shaanxi Province in Northwest China) strengthens the research.

English language is clear, the presentation is satisfactory; anyway, I have detected some criticisms in the text that should be properly addressed. Since the figures are missing in the pdf document, it was not possible to check the correspondence between the text and the results depicted in the figures. Anyhow, the analysis seems complete and exhaustive even if some aspects can be further explored/deepened.

Authors can benefit from the comments below to improve their paper. These have to be accomplished before manuscript acceptance.

 

 

Title

Title is appropriate.

 

 

Abstract

The abstract reflects the content of the article. It summarizes the main outcomes of the study. It could be a little shortened

 

 

Keywords

Five relevant keywords are provided.

 

 

Introduction

Aims of the study are properly clarified in the Introduction. Relevant references are included.

Lines 44-47-51: Concerning process-driven models in rainfall runoff modelling, and the difficulty in parameter tuning, Authors are recommended to mention available enhanced calibration techniques. Integrated event calibration approach, distinct from traditional

event-based methods, calibrates parameters across multiple events for a comprehensive system-wide analysis. In this regards, Authors are recommended to include, among others, the following relevant reference in order to deepen the introductory discussion on process-driven and data-driven models:

-        Assaf et al. (2024) New optimization strategies for SWMM modeling of stormwater quality applications in urban area. Journal of Environmental Management 361:121244, https://doi.org/10.1016/j.jenvman.2024.121244.

 

 

Methodology

This section is clear and adequately detailed.

 

 

Study area and data description

This section is clear and the proposed figures seem all necessary. Unfortunately, only the figure caption of Figure 1 is available, while figure 1 is missing.

It should be clarified whether the Bahe River and the Ba River are the same river.

Lines 186-190: Authors should specify the criteria adopted for flood events selection. Also the choice of the training set and validation set should be motivated since it affects the efficacy of the forecasting.

 

 

Data-driven runoff forecasting model based on CARS

The text of the section is clear and it follows a logical sequence. Unfortunately, only the figure captions of Figures 2 and 3 are available, while Figures 2 and 3 are missing.

The adopted performance evaluation metrics are appropriate and comprehensive.

 

 

Results and analysis

The text of the section is clear and it follows a logical sequence. Unfortunately, only the figure captions of Figures 4, 5,6 and 7 are available, while Figures 4, 5, 6 and 7 are missing.

Therefore, it is not possible to check the correspondence between the text and the results presented in the figures.

Authors should consider including further discussion on the examined forecast horizons, i.e. T = 1h, 3h, 6h.

The authors investigate the results in terms of predicted flow rate and delay. It would be of interest an analysis/comment on predicted volume.

Concerning the forecasting accuracy and the results for low flow, normal flow and flood events with the tested models, Authors should also evaluate/discuss how the choice of the training set and validation set could have affected the results.

Lines 395-397: it is not evident from table 9 that NRMSE values of the seven models on the testing dataset are typically higher than those on the validation dataset. Authors are recommended to revise this comment.

 

 

Conclusions

Conclusions seem reasonable and consistent with the previous sections of the manuscript.

Point (4): “The NRMSE values of the seven models on the testing dataset are generally higher  than those on the validation dataset but lower than those on the training dataset, indicating good generalization ability”. According to my previous comment, this sentence should be revised. Sometimes (not generally) NRMSE values of the seven models on the testing dataset are higher  than those on the validation dataset.

 

 

References

One reference is recommended in order deepen the introductory discussion on process-driven and data-driven models. Apart from this references, based on my knowledge, no important reference is missing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Editor-in-Chief

water-3138247

Title: Combining multiple machine learning methods based on CARS algorithm to implement runoff simulation

 

After carefully evaluating the manuscript, here is some feedback for consideration during revision, which may strengthen the paper.

The manuscript is well organized, so I recommend a major revision. Detailed comments are as follows:

 

Abstract and Title:

-It is recommended to incorporate key results and numerical findings into the abstract.

For example, the following sentences would be better expressed with numerical findings:

“The model achieves high forecasting ac-curacy for rainfall-runoff predictions”

“Compared to other machine learning models, the SVR model demonstrates better simulation performance and generalization ability.”

Introduction:

- Elaborate further on the different data-driven methods and approaches used to simulate river flow runoff.

- It is necessary to explain the difference between continuous modeling approaches aimed at assessing water yield (on a daily timescale) and event-based modeling approaches aimed at flood modeling (on an hourly timescale) in the introduction.

- Before stating the objective, provide a summary of previous research and highlight the significant aspects of the research discussion for better emphasis.

- Before stating the objective, it would be beneficial to mention the specific characteristics of the study area for the current research.

- Include a closure statement highlighting the exact focus and significance of your research.

- Clearly show the gaps or problems identified in the literature that your research addresses. Indicate the novelty and impact of your study in the closure paragraph.

Study area and Methodology:

- In section "2.2. Data description," has a criterion for selecting flood events been considered? Please explain whether the base flow in the observational data used for simulating the flow hydrograph has been separated.

- Overall, the methodology section, especially "3.2. Settings of model parameters," should be supported with relevant references and citations.

- It is unclear whether rainfall has been considered as an input for the model. Please provide clarification on this.

- When using statistical indices for model evaluation in "3.3. Performance evaluation metrics," it is advisable to reference authoritative statistical sources.

Results:

- Unfortunately, the figures were not included in the submitted version for review, so some questions may arise in the next stage.

Conclusions:

- In the discussion section, try to compare the results with similar approaches. Some aspects require interpretation (regarding possible reasoning for the results), such as the following sentence:

"As the lead time increases, however, the forecasting performance of the seven models decreases, with instances of underestimating peak flow."

- Explain the theoretical and practical implications of the current research.

- It would be more appropriate to address the research limitations and sources of error in the conclusion section.

- Based on the results, provide practical recommendations for users regarding model selection and flood modeling processes.

- Suggest considerations for applying the model to other regions, including necessary data and any precautions.

- Additionally, offer suggestions for future research directions.

 

Comments on the Quality of English Language

The manuscript's English level is satisfactory, but some minor revisions are needed. These can be addressed by revising the manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for responding faithfully and very quickly and revising the manuscript despite the many opinions suggested from this reviewer. As a reviewer of this manuscript, I would like to inform you that there are no additional opinions.

It is believed that this manuscript may be completed after the intensive review of typos and the use of English language.

Thanks for your huge effort to develop this study. 

Comments on the Quality of English Language

English language written in the manuscript is fine and I don't give any comment on it. 

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been significantly improved following the recommendations of the Reviewers; all my concerns have been addressed and convincingly justified by the Authors.

Reviewer 3 Report

Comments and Suggestions for Authors

Editor-in-Chief

The authors have responded to the comments and majority of suggestion/comments have been addressed in the text. Therefore, the paper has been greatly improved and is acceptable for publication.

Final Reviewer