Data-Driven Forecasting and Modeling of Runoff Flow to Reduce Flood Risk Using a Novel Hybrid Wavelet-Neural Network Based on Feature Extraction
Round 1
Reviewer 1 Report
This study investigates the efficiencies of wavelet transformation coupled with Neural Networks for forecasting and modeling Runoff Flow. This research claims to use Feature Extraction as a novelty, however, this method is not well-described in paper and the reader cannot understand it. The manuscript includes many points that should be revised or improved to be considered as a scientific paper. Therefore, as a reviewer, I suggest that the manuscript should be rejected. Comments and suggestions for improvement are presented below:
- Innovation should be elaborated more clearly in the section of methods and materials.
- The manuscript should have more coherency and avoid using different terms without explaining them. As an example, the authors stated “hybrid method has been used in previous researches” at line 113. On the other hand, the authors reviewed the papers about Groundwater (lines 62-75). It is recommended to review literature more related to runoff modeling.
- English writing should be improved and omit the redundancy in the manuscript. For instance, the first sentences of sections 2.1 and 2.2 stated the same point.
- Also, the information about wavelet transformation is insufficient. It is expected to speak about the selected wavelet family and the reason as well as the level of decompositions.
- Coefficient of determination is the right term in line 233, moreover, it is suggested to mention the formula when a term is used in the study.
- Figures 5 and 6 would be more understandable with legends.
- RMSE and MSE are from the family of least square errors. It is proposed to use one of them and replace another with one of the absolute errors such as Mean Absolute Error (MAE) or Relative Error (RE)
Author Response
Response to Reviewer 1 Comments
We would like to express our most sincere gratitude for your effort and patience in reviewing our manuscript. We deeply appreciate your constructive comments which greatly help us to improve the technical quality and the presentation of this manuscript. This paper has been revised according to the relevant comments. The following is provided to outline each change made (point by point) as raised in the reviewer comments. The corrected sentences in the manuscript are highlighted in red fonts.
Point 1: Innovation should be elaborated more clearly in the section of methods and materials.
Response 1: Thank you very much for your time and valuable comments. The innovation has been mentioned in abstract, intriduction and conclusion section.
Point 2: The manuscript should have more coherency and avoid using different terms without explaining them. As an example, the authors stated “hybrid method has been used in previous researches” at line 113. On the other hand, the authors reviewed the papers about Groundwater (lines 62-75). It is recommended to review literature more related to runoff modelling.
Response 2: The authors are grateful for the comments of Reviewer. Some more related reference has been add to the literature review part.
Point 3: English writing should be improved and omit the redundancy in the manuscript. For instance, the first sentences of sections 2.1 and 2.2 stated the same point.
Response 3: Thank you very much for the valued suggestion. The language editors have modified the paper in order to improve the readability of this sections. Sorry for our negligence.
Point 4: The information about wavelet transformation is insufficient. It is expected to speak about the selected wavelet family and the reason as well as the level of decompositions.
Response 4: The authors appreciate the comment of the kind reviewer.
We were used Discrete Wavelet Transform DWT:
- a) Provides sufficient information both for analysis and synthesis
- b) Reduce the computation time sufficiently
- c) Easier to implement
- d) Analyze the signal at different frequency bands with
different resolutions
- e) Decompose the signal into a coarse approximation
and detail information
In DWT, the original signal, passes through two complementary filters and emerges as one approximation and one detail components. Approximation (A) is the high-scale, low frequency and details (D) are the low-scale, high frequency components of the signal. Normally approximation is the most important part of the signal that represents the background information of data. In DWT, scale and translation parameters are usually based on powers of two (dyadic) instead of every possible scale and translation.
whereW(α, β) are the wavelet coefficients, N is the length of discrete time series and the asterisk corresponds with the complex conjugate function of ψ which is the mother wavelet.
The wavelet transform is a three-dimensional space, including scale, time and wavelet spectrum.
In this study rbio 6.8 (Reverse biorthogonal) wavelet is taken for the analysis.
The rbio wavelet is a family of biorthogonal wavelet defining a discrete wavelet transform
rbio 6.8’ belong to the family of biorthogonal wavelets and to its reverse, respectively, and is more advanced. The best suggestion is trial and error procedure by applying different types of mother wavelets. Based on the nature of the signal and the purpose of analysis, the best mother wavelet can be applied. This wavelet have been chosen because it has shown best performance in analyzing disturbance signals and also have chosen between all type of wavelet with test and trial.
Rbio wavelets are dual spline wavelets which have compact support, biorthogonality and symmetric Finite Impulse Response filters. The rbio wavelet family comprises a total of 15 wavelets listed as: rbio 1.1, rbio 1.3, rbio 1.5, rbio 2.2, rbio 2.4, rbio 2.6, rbio 2.8, rbio 3.1, rbio 3.3, rbio 3.5, rbio 3.7, rbio 3.9, rbio 4.4, rbio 5.5 and rbio 6.8.
Point 5: Coefficient of determination is the right term in line 233, moreover, it is suggested to mention the formula when a term is used in the study.
Response 5: Thank you for the suggestion of the kind reviewer. Correlation coefficient or Pearson correlation measures the strength of the linear relationship between forecasted and observed time series. R varies from -1 to 1. Zero value for R indicates that there is no linear relationship between modelled and observed time series. Coefficient of determination is the squared value of the Pearson correlation. Consequently, the range of coefficient of determination lies between 0 and 1. The efficiency of the model enhances as the value of ?2 increases and the optimal modelling occurs when ?2 reaches 1. In general a model with the ?2 greater than 0.5 is considered as an acceptable match to the real system. In the case of linear regression this coefficient is equivalent to NSE coefficient.
Point 6: Figures 5 and 6 would be more understandable with legends.
Response 6: The suggestion of the kind reviewer is appreciated and the Figures has been corrected with legend.
Point 7: RMSE and MSE are from the family of least square errors. It is proposed to use one of them and replace another with one of the absolute errors such as Mean Absolute Error (MAE) or Relative Error (RE)
Response 7: The authors agree with the comment of the kind reviewer. For replacing another with one of the absolute errors we need to take long time to simulate it. Hence, we will do for the future work.
Thank you
Authors
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The manuscript presents Data-driven Forecasting and Modeling of Runoff Flow to Reduce Flood Risk Using a Novel Hybrid Wavelet-Neural Network Based on Feature Extraction. Data-driven Forecasting and Modeling of Runoff Flow to Reduce Flood Risk is important to the society.
The article still needs to be modified and explain in a more detailed way.
Recommendations for addition:
- In the study is not any connection to the field of sustainability or sustainability studies except only one phrase in the conclusion (P.12, 297). Therefore, the scope of the article and the way it is presented do not meet the scope of Sustainability.
- There is a lack of a deeper scientific discussion about the achieved results linked on indexed scientific work in this field. Recommendation, discuss your results with suitable references.
- Do not authors consider to add supplementary file of presenting project via github repository for better reader understand the interaction feature extraction method? It would significantly increase the interest of readers.
Author Response
Response to Reviewer 2 Comments
We would like to express our most sincere gratitude for your effort and patience in reviewing our manuscript. We deeply appreciate your constructive comments which greatly help us to improve the technical quality and the presentation of this manuscript. This paper has been revised according to the relevant comments. The following is provided to outline each change made (point by point) as raised in the reviewer comments. The corrected sentences in the manuscript are highlighted in red fonts.
Point 1: In the study is not any connection to the field of sustainability or sustainability studies except only one phrase in the conclusion (P.12, 297). Therefore, the scope of the article and the way it is presented do not meet the scope of Sustainability.
Response 1: Thank you very much for your time and valuable comments.
Sustainable water resources management is a critically important priority across the globe. While water scarcity limits the uses of water in many ways, floods may also result in property damages and the loss of life. To more efficiently use the limited amount of water under the changing world or to resourcefully provide adequate time for flood warning, the issues have led us to seek advanced techniques for improv- ing streamflow forecasting accurate stream flow forecasting is a vital component of sustainable water [planning and management.
In this paper, a new combined neural network and wavelet transform method for forecasting river flow is proposed. While this hybrid strategy has been employed in prior studies, an intermediary step is examined in this paper to minimise complexity and improve accuracy. This is the middle step in obtaining certain unique characteristics from the wavelet-processed signals and data. Furthermore, feature extraction that reduces the volume of neural network input data improves predicting accuracy and even reduces processing and training time when utilising this method. Furthermore, unlike references that consider only a few preceding days, it allows you to use the properties of all data in a given range. By providing the novel WNN model and applying it to three daily, weekly, and monthly time scales, this proposed method focuses on enhancing accuracy and lowering risk of river flow forecasting (Ellen Brook River, Western Australia). The data are applied daily, weekly, and monthly in this method, and an attempt has been made to enhance the forecasting of short, medium, and long-term river flows with various structures by taking these concerns into account. Given the seasonality of the Ellen Brook River, this research can improve the suggested method to provide more accurate tools to aid decision-makers in planning sustainable water resources and flood prevention.
The below paper has been published in Sustainability Journal and is so close to our study:
[] Novel Ensemble Forecasting of Streamflow Using Locally Weighted Learning Algorithm
Point 2: There is a lack of a deeper scientific discussion about the achieved results linked on indexed scientific work in this field. Recommendation, discuss your results with suitable references.
Response 2: Thank you for the suggestion of the kind reviewer. As reviewer comment, more discuss has been add to the Results and Discussion section.
Point 3: Do not authors consider to add supplementary file of presenting project via github repository for better reader understand the interaction feature extraction method? It would significantly increase the interest of readers.
Response 3: Kindly inform you that, We want to extend our project and after publishing it, we will add the supplementary file of presenting the project via the GitHub repository in future.
Thank you
Authors
Author Response File: Author Response.pdf
Reviewer 3 Report
This is a very interesting study. I enjoyed reading the manuscript. Nevertheless, it needs some further improvements. In general, there are still some occasional grammar errors throughout the manuscript, especially the article "the," "a," and "an" is missing in many places; please make a spellchecking in addition to these minor issues. The reviewer has listed some specific comments that might help the authors further enhance the manuscript's quality.
- Specific Comments
A list of acronyms is needed
- Introduction
- The objectives should be more explicitly stated.
- Please elaborate a bit on the introduction section. In this regard, the following literature may be helpful to Flood susceptibility modelling using advanced ensemble machine learning models>>, Characterization of the 2014 Indus River Flood Using Hydraulic Simulations and Satellite Images>>, you may consider additional references as well.
- What is the novelty of this work?
- Methods
- The methodology limitation should be mentioned.
- All variables should be explained.
- Results
- This section is well written.
- Please improve the text size for all figures.
- Discussion
- The discussion should summarize the manuscript's main finding(s) in the context of the broader scientific literature and address any limitations of the study or results that conflict with other published work.
Author Response
Response to Reviewer 3 Comments
We would like to express our most sincere gratitude for your effort and patience in reviewing our manuscript. We deeply appreciate your constructive comments which greatly help us to improve the technical quality and the presentation of this manuscript. This paper has been revised according to the relevant comments. The following is provided to outline each change made (point by point) as raised in the reviewer comments. The corrected sentences in the manuscript are highlighted in red fonts.
Point 1: A list of acronyms is needed
Response 1: Thank you for the suggestion of the kind reviewer. The nomenclature
has been added to the manuscript as below:
WNN Wavelet Neural Network
ANN Artificial Neural Network
CI Computational Intelligence
GWL Ground Water Level
Qsim Simulated Stream flow
Qobs Observed Stream flow
STD Standard deviation
Max Maximum
Min Minimum
ANFIS Adaptive Neural Fuzzy Inference System
FIS Fuzzy Inference System
WNF Wavelet Neural Fuzzy
R2 Coefficient of determination
WT Wavelet Transform
RBF Radial Basis Function
MSE Mean Square Error
RMSE Root Mean Square Error
NSE Nash-Sutcliffe Efficiency
f(x) Function
R Coefficient of correlation
φ Scaling Fucntion
ψ Mother wavelet
α scale parameter
β translation parameter
Point 2: The objectives should be more explicitly stated.
Please elaborate a bit on the introduction section. In this regard, the following literature may be helpful to Flood susceptibility modelling using advanced ensemble machine learning models>>,
Characterization of the 2014 Indus River Flood Using Hydraulic Simulations and Satellite Images>>, you may consider additional references as well.
What is the novelty of this work?
Response 2: The suggestion of the kind reviewer is appreciated. As reviewer sugesstion the references has been hadded to the introduction. The novelty of this work is, a new combined method for forecasting river flow is presented based on neural network and wavelet transform. While this type of hybrid method has been used in previous researches, but in this paper, an intermediate step is considered to reduce the complexity and increase the accuracy. This is the middle step of extracting some special features from the signals and data processed in the wavelet section. Also, by using this method, feature extraction that reduce the volume of neural network input data increase the accuracy of forecasting and even reduce the processing and training time. Additionally, it makes it possible to use the properties of all data in a certain range, unlike references that consider only several previous days. This proposed method focuses on improving accuracy and reducing risk of river flow forecasting (Ellen Brook River, Western Australia) by presenting the new WNN model and applying it to three daily, weekly and monthly time scales. In this method, the data are applied daily, weekly and monthly, and by taking into account these considerations, an attempt has been made to improve the forecasting of short, medium, and long-term river flows with different structures. Given the seasonality of the Ellen Brook River, this paper can provide more accurate tools to assist decision-makers in planning sustainable water resources and flood prevention by improving the proposed method.
A summary of the contributions of this study is as follows:
- Presenting a hybrid WNN method;
- Using wavelet to increase forecasting accuracy;
- Using feature extraction (energy, standard deviation and maximum values, etc.);
- Reducing the computation time using feature extraction
- Reducing computational complexity by using feature extraction;
- Using all daily, weekly, and monthly data;
- Comparison with other previous methods.
Point 3: The methodology limitation should be mentioned.
All variables should be explained.
Response 3: Thank you for your valuable suggestion. The methodology limitation has been mentioned and all variable has been explained.
Point 4: This section is well written. (Results)
Please improve the text size for all figures.
Response 4: Thank you for the suggestion of the kind reviewer. The text size for all figures ahs been improved.
Point 5: The discussion should summarize the manuscript's main finding(s) in the context of the broader scientific literature and address any limitations of the study or results that conflict with other published work.
Response 5: The authors appreciate the comment of the kind reviewer. The literature review and discution sections have been improved as reviewer comment.
The authors are grateful to the kind reviewer for the dedicated time to evaluate this work. Please let us know if something is missing in the document, and we can improve it.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The manuscript has improved a lot since the last time but there are still many points that should be changed.
It was suggested to add legends to the figures. The authors have done but the legends are incorrect. Also, the authors refused to add another error such as MAE which was proposed. These points imply that either they don't have access to the results or they are not able to interpret the results.
Therefore, I highly recommend to reject the manuscript.
Author Response
The authors would like to thank the reviewer for giving helpful comments and the following productive suggestions. We have revised the manuscript according to your comments. The modified contents have been marked in red color in the revised manuscript.
Reviewer 2 Report
Thanks to the authors for revising and supplementing study based on my comments from last time. All comments have been incorporated. The article has been reworked in very large scale, which significantly increased the quality of scientific processing of the article.
Author Response
We would like to thank you and the referee for the time and efforts on our manuscript.
Reviewer 3 Report
I don't see any major contribution from this paper.
Author Response
We would like to thank you and the referee for the time and efforts on our manuscript. This paper proposes a novel hybrid wavelet-neural network (WNN) method with feature extraction to forecast river flow. To do this, initially, the collected data are analyzed by the wavelet method. Then, the number of inputs to the ANN is determined using feature extraction, which is based on energy, standard deviation, and maximum values of the analyzed data. The proposed method has been analyzed by different input and various structures for daily, weekly, and monthly flow forecasting at Ellen Brook river station, Western Australia. Furthermore, the mean squares error (MSE), root mean square error (RMSE) and the Nash-Sutcliffe efficiency (NSE) is used to evaluate the performance of the suggested method. Furthermore, the obtained findings were compared to those of other models and methods in order to examine the performance and efficiency of the feature extraction process. It was discovered that the proposed feature extraction model outperformed their counterparts, especially when it came to long-term forecasting.
