A Machine Learning-Based Surrogate Model for the Identification of Risk Zones Due to Off-Stream Reservoir Failure
Round 1
Reviewer 1 Report
A Machine Learning based surrogate model for the identification of risk zones due to off-stream reservoirs failure.
Interesting manuscript. I have checked the similarity scores for the manuscript using Turnitin and found out that it is 11% and well below the limit. Therefore, this manuscript can be further processed to reviewing process.
The presentation is not the best when it comes to language. This may not be the case of grammar but the way and the style. For example, the 1st sentence in the abstract.
The research gap is not clear in the abstract.
Introduction is good with a good number of citations. However, the authors have not introduced the 3 hydraulic models there. They used the abbreviations.
What is Iber?
Some issues in text. “2.1 Data generation the synthetic cases were defined on the basis of ten parameters shown in Error! Reference source not found. The length of the domain is 4000 m (LA)”
These reference issues can be seen in number of places. Please correct them.
You define Iber in section 2.2 But you have used this in your abstract. Mismatching in presentation.
Why Random Forest? Why not any others? Regular Genetic algorithms? Any advantage of Trees compared to traditional ways?
Concern on units and their superscript powers!
Can the authors connect and join the results and discussion together?
Conclusions – rather than a summary can the authors short the direct findings?
Good number of references.
Overall this paper is well written, however, I found the comparison of the work to previous work in the sense of finding is lacking. Can the authors think of that?
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper needs major revisions:
1-It seems that all sections of paper have been fully developed. "2.4.4. Measures of accuracy" needs reference "Flood risk mapping by remote sensing data and random forest technique".
2-How does RF work? A comprehensive description is required.
3-Describe the data sets for training an testing stages. Is data allocated to checking stage? Author can use material of "Riprap incipient motion for overtopping flows with machine learning models".
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Review for “A Machine Learning based surrogate model for the identification of risk zones due to off-stream reservoirs failure”
The manuscript is generally well-structured and the topic seems to present some interesting results for readers. I was nearly to suggest a minor revision. Overall, I suggest a "major" revision before possible consideration of the application in Water. My comments are listed as:
1. Although the manuscript is generally well-written, a language check by a professional native speaker or an editing agency is needed to fix some syntax, style, and phrasing problems.
2. Please use different keywords that not mentioned in the manuscript title.
3. All paragraphs in introduction needs more references.
4. Citation style is not correct in the manuscript. Please check it.
5. Figure 1 text should be increased.
6. Many figures and tables citations were wrongly written.
7. What is GiD in line 199?
8. Why the authors used random forest as a ML technique?
9. L291 to L293: Style mistake.
10. Figure 8: what is the axis title?
11. Figure 11: please explain in details.
12. Check the degree sign in line 460
13. L467: style mistake
14. What is the y-axis of figure 17(a)
15. Please check the references as there are some references without doi or wrongly inserted and some the location of publication is not written. Please check them carefully.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
The manuscript discusses alternative ML models to evaluate safety impact caused by off-stream reservoir failure. However, the benefit of running ML model instead of hydraulic model is not very convincing. If it is a lack of resource from property owners, how exactly does a sophisticated ML training and prediction workflow reduce the needed resource?
It’s nowadays trivial to perform prediction with multiple ML methods and choose the method most appropriate and with top performances. Within sklearn there are many options too, see https://scikit-learn.org/stable/tutorial/machine_learning_map/index.html, why did the author directly go for RF? Were other methods attempted? Also, since the authors discussed probability map, perhaps there are better probabilistic approaches than what was described? What about Bayesian networks among others? These need to be included in the discussion.
Is there a way to design a classification “success criteria”? The appendix A provides 600 classification results. Using a criteria, we can statistically tell how many of those 600 results succeeded and how many failed.
Cosmetic:
- Python variables in the author’s code are used without explaining their meaning. For experienced sklearn user this is not an issue but for others it can be.
- For researchers familiar with ML, some of section 2.4 is redundant, those formulas are available in sklearn.
- text display issues: L139, L151,L292,L483
- typos: L181 where, figure 11 numbering wrong
- figure 14 is duplicated
- L475 not sure what are the “15 para.”
- Figure 1(a) symbols cropped out, can’t see. Also wrong symbols.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Accept as is
Reviewer 3 Report
-
