Understanding the Effect of Hydro-Climatological Parameters on Dam Seepage Using Shapley Additive Explanation (SHAP): A Case Study of Earth-Fill Tarbela Dam, Pakistan
, Saad Salman 3,*, Khan Zaib Jadoon 4, Abid Ali Khan Danish 3, Kifayat Ullah Bangash 5 and Dai Qianwei 1,2,*
Round 1
Reviewer 1 Report
Summary and general comments:
I am pleased to review the manuscript “Machine Learning and Feature Importance for dam seepage modelling based on Hydro-climatological parameters: A case study of Earth-fill Tarbela Dam, Pakistan” by Ishfaque et al. for potential publication in the journal Water. This study uses different machine learning methods to predict dam seepage, which is a very important aspect of water dam safety and the methodologies used are of great novelty. Among all methods, the newer CatBoost method outperforms the others. Sensitivity analysis has been conducted to investigate the uncertainties introduced by each parameter. Based on their training, calibrating, and testing procedures, and the equipped SHAP method, the authors have figured out the most significant parameter as the reservoir level. I believe these findings are of great interest to the research communities of both hydrology, engineering, and machine learning. With some minor revisions, this paper can be considered publishable.
Specific comments:
- Line 34 in the Abstract, “effecting” may be modified to “affecting”.
- Line 161: nice summary.
- Table 1: font format issues such as ft3/s and m3/hr. Should use superscripts.
- Section 2.4: Though the definition of SHAP method has been highly compacted in the literature, I suggest the authors to provide some mathematical basis for better understanding.
Author Response
Thank you very much for your kind letter.
The manuscript has undergone substantial revision in response to the suggestions and insightful comments of reviewers 1. We found all the comments very helpful and constructive, which we feel has contributed considerably to the improvement of our manuscript. In particular, during the present submission, we revised the text as suggested by the reviewer. We present right below a point-by-point response to all points that the reviewers have raised in the Attached file.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Review Report
The authors of manuscript no. water-1847204 has estimated seepage from Tarbela Dam of Pakistan using machine learning models. The topic of the paper is interesting and within the scope of the journal. But it needs some modifications for better improvement of the paper. Therefore, I have proposed a “major revision” with the following comments:
1. Write the full form of all the used abbreviations as they come in the manuscript.
2. Update the introduction by highlighting major difficulties and challenges, and your original achievements to overcome them by citing the latest state-of-arts.
3. Superscript “2” with R throughout the entire manuscript.
4. Correct as “sediment inflow”, not “sediments inflow”, in the entire manuscript.
5. Explain the reason, why did the authors select the Earth-fill Tarbela Dam of Pakistan as a case study?
6. Correct the unit of Water Inflow and Average Seepage in Table 1 by Superscripting “3” with “ft” and “m”. Also, check the whole manuscript and correct it.
7. What are the advantages of the applied models (i.e., ANN, RF, SVM, and CB) over others in simulating the complex hydrological processes?
8. Any pre-processing is done with the training data? Explain
9. Many works have already proved the potential of machine learning models in seepage flow prediction. What is the transferability of such results to other locations in terms of impact or usefulness? Is it really the novelty in a true and specific sense or just to test the applied models?
10. The authors have used only two model evaluation metrics i.e., root mean square error and coefficient of determination. Why only two? It is recommended to add “Nash-Sutcliffe Efficiency” and support all these by citing the suitable reference. Also, write their range.
11. R2, RRMSE, MAE, NSE, and GPI to evaluate the performance of applied models. Also, write their range.
12. Furnish more debate about the practical utility of work in the discussion section.
13. In conclusion, it includes the direction for future works.
14. The reviewers recommend some useful references, that need to be cited which help the authors for improvement of the paper.
Lake water level modeling using newly developed hybrid data intelligence model. Theoretical and Applied Climatology, 141(3-4): 1285-1300, https://doi.org/10.1007/s00704-020-03263-8.
Prediction of daily water level using new hybridized GS-GMDH and ANFIS-FCM models. Engineering Applications of Computational Fluid Mechanics, 15(1): 1343-1361, https://doi.org/10.1080/19942060.2021.1966837.
Simulation of seepage flow through embankment dam by using a novel extended Kalman filter based neural network paradigm: case study of Fontaine gazelles dam, Algeria. Measurement, 176: 1-16, https://doi.org/10.1016/j.measurement.2021.109219.
Author Response
Thank you very much for your kind letter.
The manuscript has undergone substantial revision in response to the suggestions and insightful comments of reviewer 2. We found all the comments very helpful and constructive, which we feel has contributed considerably to the improvement of our manuscript. In particular, during the present submission, we revised the text as suggested by the reviewer. We present right below a point-by-point response to all points that the reviewer has raised in the attached file.
Author Response File: Author Response.pdf
Reviewer 3 Report
This article is a case study on dam seepage modeling using machine learning techniques. The article has an interesting topic. There are some issues that need to be addressed.
1- The title can be improved to avoid any confusion for the readers. The keyword "Feature Importance" is not clear.
2- In Figure 1, the scale is needed for the map.
3- Table 1, please correct the formatting for the Average Seepage unit.
4- Table 3, parameters need units. The number of decimal places goes to 4 for inflow, and water level, and up to 8 decimal places for precipitation. Do we have such precision in measuring them? I would suggest limiting decimal places based on the precision level in measurement.
5- Since the data is split into training, testing, and validation it is necessary to show the data statistics separately for these three datasets.
6- RMSE formula is missing the square root function.
7- Table 4, RMSE has a unit. Please add it.
8- Peaks estimation in time series simulation is very important. It is suggested to add a specific error measure (relative peak error) for those specific peaks in Testing Dataset. That will help to look into models comparison from a new angle.
Author Response
Thank you very much for your kind letter.
The manuscript has undergone substantial revision in response to the suggestions and insightful comments of reviewer 3. We found all the comments very helpful and constructive, which we feel has contributed considerably to the improvement of our manuscript. In particular, during the present submission, we revised the text as suggested by the reviewer. We present right below a point-by-point response to all points that the reviewer has raised in the attached file.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
It is okay from my side and recommended for publication in Water MDPI Journal.
