Flood Discharge Prediction Based on Remote-Sensed Spatiotemporal Features Fusion and Graph Attention

Round 1

Reviewer 1 Report

Review of the paper “Flood Discharge Prediction based on Remote-Sensed Spatiotemporal Features Fusion and Graph Attention”.

The paper aims to propose a methodology for flood discharge prediction based on remote-sensed spatiotemporal features fusion and graph attention. The paper suffers from different perspectives as detailed below:
1. The paper suffers from the lack of an appropriate introduction addressing the following issues.
a. The introduction needs to be revised substantially.
b. What are the current problems in flood prediction that this paper seeks to solve?
c. What are the study gaps?
d. What are the contributions of this work?
e. There are multiple uses of aims and objectives at different sections of this paper which create unnecessary confusions. Authors should put the aims and objectives at the same place, not in segregated way.
All these questions need to be addressed in the introduction briefly and then elaborated at the end of literature review.

2. Abstract to be revised to incorporate actual outcomes.
   3. There is no proper literature review section. Most of the references are outdated. More references from 2020 and 2021 to be added to present the current state of the art. You should propose other predictions methodologies like Random Forest, Artificial based prediction models etc. You should compare these models and discuss their advantages and limitations. Based on that you should define the scientific gap.
3. Why do we need this kind of model for flood prediction?
4. How we can judge about quality of the training set? What is the minimum data set for properly adjusting the NN? This should be clarified and explained.
5. You should add validation section after section 5. In the validation section you have to compare the results with other existing prediction algorithms. This will help to the readers to see advantages of your methodology over existing in the literature. One would expect to find the previous empirical work enriching the discussions of the results, but unfortunately, that has not been done.
6. Add limitations of proposed methodology. This should be discussed in the conclusion section.
7. Conclusion - The part of the recommendations is rather short, maybe you can strengthen that part in a way which really show the implication of the findings more clearly.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Re: “Flood Discharge Prediction based on Remote-Sensed Spatiotemporal Features Fusion and Graph Attention”

by Chen et al., 2021

General comment

This work focuses on predicting floods using machine learning methods, with the consideration in extracting spatial distribution information of rainfall data to improve flood prediction accuracy. Overall, the main findings seem promising and can potentially be attractive to readers of Remote Sensing.  After careful examination and evaluation of the manuscript, I find that some contents (especially in Section 3) are quite unclear and sometimes really difficult to follow, and more importantly, many basic details of the work and in-depth discussion are not presented. Therefore, I can only recommend major revision of this manuscript at this time.

Some major comments

This manuscript lacks of detailed description in several sections, which is fundamental for readers to understand this work. (1) How many rainfall data points were used for model training and forecasting, respectively? At least some figures should be presented for the data used in the models. (2) Are there any model training results to show the performance of discharge matching? (3) Will the density of observation data (i.e., rainfall and discharge) affect the training and prediction performances of the model used in this manuscript? (4) How the models described in Section 4 were implemented for the study area?

Meanwhile, this manuscript lacks in-depth discussion about the results in the hydrology fashion. Say, in Figure 15, what are the causes of the significant differences for using different pred_step? In Figure 16, what are the model training details for the results of LSTM. In Figure 17, what are the potential reasons for the over-prediction of floods #2 and #3, and also reasons for the under-prediction of floods #1 and #6.

Other comments

In Abstract 

• Line 9.What is the meaning of “to extract the investigate the”?
• Line 14.  The first letter of “xixianhe” should be capitalized.

In Introduction

• Line 1: why “also one of the most”? Have any other disasters already being introduced?
• In the fourth paragraph, what does “ANNs” stand for? Same for “RNN”, “CA-LSTM” in other sentences.
• “In order to fully excavate the spatial distribution information of rainfall and improve the overall prediction ability of the model, this paper uses remote sensing images to extract digital elevation information and obtain regional spatial information.” This sentence is quite unclear (In the fourth paragraph).
• what does “GAT” stand for?

In Data profile and preprocessing

• The names of hydrological stations and basins are confusing.  Is it “Xi country hydrological station” OR “Xixian hydrological station”? In some places, Xixian County. Mixed use of county name exists throughout the manuscript.
• In Figure 1, are there any other stations? If yes, please specify. If not, there will be no need to use “stations”
• Where are the flow data and digital elevation information in Figure 2?, as the authors stated, “As shown in Figure 2, the historical flow data of Xixian hydrological station from 2012 to 2018 and the historical rainfall data of 50 stations, including Xixian hydrological station, are provided. It also provides the longitude and latitude distribution of 50 stations and the digital elevation information of the area.”

Section 2

• In thethird paragraph, some statements by the authors are somewhat misleading. For example, “the progress of spatial feature mining of rainfall data is not obvious in the field of flood forecasting.” Scientists have been working decades trying to obtain spatial distribution of rainfall data as accurate as possible for flood predictions. It will be good if the authors can provide a thorough review for this part on this point.

Section 4.1

• What is the meaning of “European data” / “non-European data”?

Section 4.2

• Most part of the fourth paragraph is repeating contents in the third paragraph.

Section 5.1

• Why different forecast periodsettings in the forecast models led to the significant differences in Figure 15? Are there any theoretical explanations?

Section 5.3

• In Figure 16, which model is used for flow prediction, GRU or LSTM?

Section 5.4

• What does “D model”in the sub heading stands for? Meanwhile, how the evaluation rule in Table 8 is determined? Is there any government or scientific guidelines for this rule?

 Section 6

• The author concluded that the proposed model predicted the historical floods with high accuracy. It will be good if they can also list some potential limitations and disadvantages of the proposed model to provide readers with a full understanding of their work.

 

Author Response

"Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear authors,

Please, read the attached file.

Best regards,

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed the point of my concern. I am happy with their corrections. Hence, I would like to recommend this manuscript to be published.