Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting: A Case Study in the Wei River Basin, Shaanxi, China

Round 1

Reviewer 1 Report

A BRIEF SUMMARY

The paper titled The Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting Model: A Case Study in the Wei River Basin, 4 Shaanxi, China” presents a good topic for readers of this Journal. The topic represents a line of research as interesting as studied. The paper is well structured.

The results are carefully described and analysed in the paper. Hovewer, some open questions remain after reading the paper. Below is the list of some questions that need to be addressed.

 

• Why have you choose this study area? In my opinion, you have to add more details in “study area description”.
• In conclusion, at line 458-462, the authors have reported: “From the aspect of robustness, the rainfall data's inclusion can enhance the model's performance when the hyperparameters vary”.

I agree this sentence, because this is a very detailed study. But, in my opinion, you have to rewrite the sentence in order to better show the difficulty to have sufficient rain data for forecasting. In addition, you could suggest use of continuous hydrologic method for flood forecasting (see suggested references).

• Why the references is so few? I understand the very particular topic developed in the paper, but probably an more accurate references research could help to add value for this topic. I strongly suggest that the authors try to add some more references especially in the "part 1 (introduction)" of the paper. I have indicated some suggestions for specific section of paper, but more can be added to make the foundation for the arguments stronger.

 

 

SPECIFIC COMMENTS

In fact, in international literature there are recent studies on “continuous hydrologic modelling”.

• Petroselli, A., Grimaldi, S. 2018. Design hydrograph estimation in small and fully ungauged basins: a preliminary assessment of the EBA4SUB framework. Journal of Flood Risk Management, 11, S197-S210.
• Grimaldi, S., Nardi, F., Piscopia, R., Petroselli, A., Apollonio, C.. Continuous hydrologic modelling for design simulation in small and ungauged basins: A step forward and some tests for its practical use. Journal of Hydrology, 2020, 125664, ISSN 0022-1694. doi: 10.1016/j.jhydrol.2020.125664
• Piscopia, R., Petroselli, A., & Grimaldi, S. (2015). A software package for the prediction of design flood hydrograph in small and ungauged basins. Journal of Agricultural Engineering 2015 XLVI, 432, 74–84.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Please, have a look at the enclosed file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The Research Paper presents an interesting issue on the importance of ANN technique in runoff forecasting. 

The paper addresses a topic that is abundant in literature what could compromise the novelty. However, there are some aspects that should be revised in order to improve paper quality.

The introduction is a little bit large I suggest simplifying from line 60 to 81 because it is strongly focused on ANN description ( you are writing on a hydrological journal!). I suggest to revise the last part of the introduction because it contains a little methodology,  study area description. Moreover, a working hypothesis and what did you aspect from the results are missing...

The 2.1 part of the methodology explains detailedly topics well known. Please, simplify and focus your attention on your methodology.

In the model evaluation part, indexes range and optimal/bad values are missing. Please, provide them.

In the 3.3 part, the PCA denoises data could result. Please, assess if can be moved in the next section.

Results are well arranged and discussion seems targeted (in line 366 maybe NSE?).

The results, after focusing the ANN performance evaluation, should be addressed to the implication of your research on the runoff prediction strategy avoiding to stress the methodology (it is still a study case!).

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

In the study titled “The Impact of Input Filtering and Architecture Selection Strategies on GRU Runoff Forecasting Model: A Case Study in the Wei River Basin, Shaanxi, China”, The goal of this study is to show the case study about runoff analysis by introducing the Neural network system for estimating the runoff. The outline of this thesis is acceptable. But I would like to confirm several points before accepting.

I hope you will check the comments below:

Major Comments

1. Consideration based on the hydrological background
I was impressed that there looked less discussion about the result of this case study from the point of view of hydrology or geography, only from the technical matter. I think it would be better to discuss it from the point of view of those. For example, according to the result, inclusion of rainfall data or runoff data make worsen the performance of the estimation of T+1 or T+2. The result is judged based on the statistical indicator like RMSE... But these results could be estimated regarding the geographical background, for example, by considering the distance between target station (Huaxian) and observation station. (The values from the stations further from the target station might be related to the estimation results of the later date, because of the time lag. Isn’t it possible for this system to consider this kind of time lag?) I think this paper would be more interesting, if these kind of matter would be discussed in this paper.

2.bi-GRU network
Two methods (GRU and bi-GRU) are mainly introduced in this paper. But bi-GRU is used for case 6 only. Should I understand that bi-GRU model.is not important, or the outcome of bi-GRU is almost same as the result of GRU? The reason why bi-GRU is not used for the calculation would be desirable. (I am asking this question because you explain the outline of bi-GRU by using the many lines (section 2.3))

3.Snowfall
Did this system consider snowfall? I am not sure about the seasonal change in this area. If snowfall is contained in the learning process, I think that the result would be affected by it.

Minor comments

1.You use the words “rainfall” and “precipitation” in your thesis. It looks that these two words are used as same meaning. Is it necessary to use both words?

2.Line 366: NES -> NSE

3.IQR
You mentioned the result of RMSE and MAE by referring to IQR. Is it possible to define the maximum value as 4 quarter for RMSE, and the min-max range for MAE?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

see attachment

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx