Round 1

Reviewer 1 Report

Please See my report

Comments for author File: Comments.pdf

Author Response

Please refer to the attached document

Author Response File: Author Response.docx

Reviewer 2 Report

The present paper is devoted to signal processing and seismic activity prediction using historical data and deep learning methodology. It is necessary to appreciate that real data are used for the proposed method verification.  I have the following specific comments to the present submission:

 

Major comments:

 

11.     ABSTRACT: The text should include more details to the proposed methodology, and specification of numerical results achieved (the prediction length,…).

 

22.      Section 1. INTRODUCTION: I suggest to include here notes to the prediction horizon, sensors used to collect data sets, specification of data collected, the sampling frequency and appropriate mathematical methods related to data processing. Specifically I suggest to add notes to methods of data analysis and digital filtering both in the time, frequency and scale domains. Additional relevant references should be added, including those that use similar sensors for accelerometric data acquisition and motion analysis in different applications using similar mathematical methods, for instance:  

 

[1] Dostál O. et al.: Recognition of Motion Patterns Using Accelerometers for Ataxic Gait Assessment, SPRINGER: Neural Computing and Applications, 33:2207-2215, 2021

 

[2] Procházka A. Et al: Integrating the Role of Computational Intelligence and Digital Signal Processing in Education, IEEE Signal Processing Magazine, 38(3): 154-162, 2021

[3] Bnou K., et al.:  A wavelet denoising approach based on unsupervised learning model, EURASIP Journal on Advances in Signal Processing, 36, 2020

 

 

33.   Section 2.1  DATA COLLECTION: This section should describe into more details data sets, specification of observed signals, and the sampling frequency.  Were multichannel signals observed? How their time synchronization was achieved?

 

44.     Section 2.2 PRE-PROCESSING: How preliminary signal analysis was done? Which mathematical methods were used? Was both time-frequency and time-scale analysis done? Which spectral components were included in observed signals? Which component were dominant? Was digital filtering applied? And how its coefficients were selected?

 

55.     Section 2.3 FEATURE ENGINEERING: Gutenberg-Richter law should be better explained, and especially its relation to seismic activity prediction; all variables used should be described in the text

 

66.     Section 2.4 THE PROPOSED DEEP NEURAL NETWORK ARCHITECTURE: How feature vectors are specified? Are sigmoidal layers used only? And how their number is selected? How long prediction horizon was selected? Which software tools were applied?

 

77.     Section 3 RESULTS AND DISCUSSION: I suggest to specify into more details processing goals and specification of signal prediction tasks and classification applications.

 

88.   Section 4 CONCLUSION: Further research should be specified. Is the real time processing possible?  

 

 

Minor comments:

 

11.     The whole text should be carefully examined and corrected (page 5 line 136: Where >> where, …)   

 

22.    The formal level of the whole text should be increased and mistakes in English corrected.

 

33.     Refences should be examined, publishers added in some cases ([1], [15], [20], …)

Author Response

Please refer to the attached file.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors  applsci-2156228

A Generalizable Deep Learning Approach for Seismic Activity Prediction

 by

 Dost Muhammad , Iftikhar Ahmad, Muhammad Imran Khalil , Wajeeha Khalil, and Muhammad Ovais Ahmad.

 The manuscript deals with a novel methodology for prediction of earthquake using feature engineering and deep learning-based technique. New features are calculated based on the various  seismic laws. The features include seismic rate of changes, fore-shock frequencies, release seismic energy, total time of recurrence, maximum minimum  relevance and redundancy etc. These features are extracted and used as input for our deep learning model. The results show that the proposed deep neural network is more accurate  than the other ML approaches.

 Reviewer's comments:

·      To improve the quality of Figure 1, please describe in more detail in the text.

·       ·Explain in detail the variations of your proposal in Figure 3. Argue the cause of the variations

·    ·  Explain in the text how the engineering features used were selected, if there are others, what is the reason they were discarded?

·    ·   In section 2.5, why compare the results with algorithms of the same nature? Why not compare with other prediction methods? I think it is not valid to compare the method with others of the same nature.

·   ·Explain in detail in the text whether the data used were suitable after post-processing for use in all algorithms.

·        ·Include a section on scope and limitations of your proposal.

·       ·  L26 -27 Machine Learning (ML) approaches play a pivotal role in prediction and forecasting in various fields including different disasters such as floods, earthquakes, and landslides etc [2– 8].

·       ·  L28 -30 These studies have utilized a variety of machine learning approaches such as artificial neural network [5], support vector machine [5], random forest [5], and convolutional neural network [6].

·   ·L34-35 Artificial neural network is implemented by Oktarina1 et. al. [5] for earthquake prediction in Indonesian region and calculated the mean square error.

·         ·Standardize in the text whether et. al. or only et al. will be used.

·Conclusions must be rewritten, they must be supported by the results obtained. Especially the following:

 Our proposed model can help the policy makers on the safety level to formulate evidence-based policies to avoid catastrophe and human loses from the earthquake. Moreover, the proposed model enriches the knowledge-based decision making of the Government and their disaster preparedness departments. This model is more is more potent for Earthquake engineers, Geologist, Ecologist, Disaster managers to predict earthquake and their magnitude in the concern region.

Author Response

Please refer to the attached file.

Author Response File: Author Response.docx

Reviewer 4 Report

The manuscript "A Generalizable Deep Learning Approach for Seismic Activity Prediction" presents a machine learning approach that aims to predict dangerous seismic activities. In particular, the authors considered deep learning algorithms based on Deep Neural Network architectures to evaluate data recorded in Southern California, Chile, and Hindu Kush seismic zones. The topic interests data scientists, seismologists, and professionals who must make political decisions about seismic risk issues. The introduction presents the research topic in a broad context of learning applications using the mean square error as the primary metric, highlighting various applications through references. However, in the introduction, the authors ignore the Gutenberg–Richter law and its applications in seismic risk analysis. Although the manuscript deals with a subject worthy of investigation, it still needs improvement. The presentation is sometimes confusing, with missing or poorly motivated definitions. Authors should pay more attention to the methodology section, which is not presented accurately. Also, the English language and style must be improved. Please see some comments below.

 

(1) The methodology section is poor and must be improved. Figure 1 is not didactic. This figure was included in the manuscript without any clarification. There is no need to include the United States Geological Survey (USGS) logo in the workflow. Nor any from other institutions. After all, the authors claim that their proposal is a generalized approach; the methodology should work in analyzing several databases. 

 

(2) A new section must be created to present the data set and its pre-processing. Data and their collection shall only form part of the methodology section if the work is about, for instance, seismic surveys. Authors must merge subsections 2.1, 2.2, and 2.3 into this new section. Also, such subsections should be better clarified.

 

(3) Authors should make it clear what their data is. Magnitude information? Waveforms? Earthquakes spatial distribution? Please clarify these points and others that the authors think are appropriate.

 

(4) The data summary is in Table 1. What are the geographic coordinates of each studied region? What is the range of amplitudes? How many earthquakes have been recorded? Authors must detail the main characteristics of each seismic region.

 

(5) In subsection 2.2, the authors state, "After obtaining the data from US Geological Survey, it was evaluated initially for cut-off magnitude. As no seismic event was found missing, cut-off magnitude corresponds to the earthquake magnitude.". The authors should explain how they determined the magnitude of completeness. It is unclear which criteria were adopted by the authors to state that "no seismic event was found missing". How did the authors determine the cutoff magnitude?

 

(6) In line 34 (among others), the authors mention "accuracy". Authors should better explain what this means in the present context, as a mathematical explanation is only shown in Table 3. Accuracy in what sense? In predicting significant earthquakes? Earthquake occurrence? Please clarify, as such a term is vital in understanding this work.

 

(7) Bullet points presented in lines 79-87 are results and should be in the results or conclusion sections, not the introduction section.

 

(8) Authors should dedicate a paragraph in the introduction to the Gutenberg-Richter law and its generalized approaches. The importance of this law in the analysis of the seismicity of a region has been treated in several works in its original formulation (see, for instance, https://doi.org/10.1038/s41561-020-0544-y and https://doi.org/10.3390/app112412166) and in its generalized formulations (see, for instance, https://doi.org/10.1103/PhysRevLett.92.048501 and  https://doi.org/10.1016/j.chaos.2020.110622).

 

(9) Authors must explain all parameters presented in the equations. For example, the definition of M_mean in Eq. 4 only appears in Eq. 10. Arrange the order of the equations and explain them.

 

(10) The value 1.5 in Eq. (11) is well-known in seismology applications and has been used in several analyses. However, is the value 11.8 valid for any seismic catalog? Please explain it.

 

(11) What does Max mean in Eq. (13)? And $h\teta(x)$ in Eq.(16)?

 

(12) Please double-check the References. Standardize them. Some references display the journal's name in abbreviated form, while others do not. The first reference must be included in full. Please see the author's guide (https://www.mdpi.com/journal/entropy/instructions).

 

(13) The English language and style really needs to be improved. Some parts are hard-to-read sentences or wordy. Below, I show a few points where authors can improve the text. Nonetheless, I suggest authors perform a spell check in all work to make the article more readable. For example: 

 

(13.1) Lines 18-19: “Earthquakes are is considered the one of the most dangerous among natural disasters as it they comes without any warning.” leading to “Earthquakes are considered one of the most dangerous natural disasters as they come without warning.”

 

(13.2) Lines 20-21: “The ratio of deaths caused by earthquakes is more than half of the that of other natural disasters [1].” leading to “The ratio of deaths caused by earthquakes is more than half of that of other natural disasters [1].”.

 

(13.3) Lines 22-23: Please consider rephrasing  “It means that either they were they were either injured, or they lost their houses and valuable properties.” leading to “It means they were either injured or lost their houses and valuable properties.”

 

(13.4) Lines 24-25: “Seismic activity prediction can be the optimal technique for avoiding economic and human tragedies.” leading to “Seismic activity prediction can be optimal for avoiding economic and human tragedies.”

 

(13.5) Lines 34-36: “An Aartificial neural network was is implemented by Oktarina1 et. al. [5] for earthquake prediction in the Indonesian region and calculated the mean square error.” leading to “An artificial neural network was implemented by Oktarina et al. [5] for earthquake prediction in the Indonesian region and calculated the mean square error”.

 

(13.6) Lines 36-37: Please consider rephrasing and correct the spelling: “Jena et al. [6] conducted a study studied of the Palu region in Indonesia to identify earthquake-prone areas using cluster anlaysis analysis techniques.” or ” “Jena et al. [6] conducted a study studied of the Palu region in Indonesia to identify earthquake-prone areas using the cluster anlaysis analysis technique.” leading to “Jena et al. [6] studied the Palu region in Indonesia to identify earthquake-prone areas using cluster analysis techniques.” or “Jena et al. [6] studied the Palu region in Indonesia to identify earthquake-prone areas using the cluster analysis technique.”

 

(13.7) Lines 40-42: “Majhi et al. [11] used a moth-flame optimized functional link with an artificial neural network for prediction of to predict seismic magnitude on earthquake catalog data and the mean square error was calculated by considering the mean square error as a metric.” leading to “Majhi et al. [11] used a moth-flame optimized functional link with an artificial neural network to predict seismic magnitude on earthquake catalog data by considering the mean square error as a metric.”

 

Overall, the authors do not present the methodology accurately. The present manuscript must be better presented, especially in the dissociation of methodology and data set. The present manuscript should address the above points for the work to be interesting for several readers, in addition to a more detailed and accurate explanation of the methodology, especially around equations.

Author Response

Please refer to the attached file.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Most comments were answered but I still suggest a careful reading of the whole text.

 

1. The line with authors should be divided into two lines

2. Page 5, line 146: Where >> where

Author Response

Thank you very much for the prompt review. Our response to comments are;

 

Reviewer 01: 
1. The line with authors should be divided into two lines

Reply : The comment is addressed

2. Page 5, line 146: Where >> where

Reply : The comment is addressed

Reviewer 4 Report

The authors addressed almost all questions. Two more points still need to be addressed.

 

(1) Sub-section 2.1. Data Collection really should be outside 2. Methodology. A new section (not a sub-section) should be created for the data presentation and processing (for instance, 2. Data Collection), while the section Methodology presents the approaches used to analyze the processed data set (for instance, 3. Methodology). In other words, all processes described in the first line of Figure 1 must be in a particular section, while all processes described in the second line of Figure 1 must be in another section.

 

(2) In the new version of the manuscript, the authors included the following sentence in lines 120-123: "As discussed in Asim et al. [9], the cut-off magnitude for California region is 2.6, for Chile 3.4 and for Hindukush 4.0. Weimer and Wyess [21] discussed various methodologies for determining the cut-off magnitude, however, in line with the existing literature, we used Gutenberg-Richter law [9]." Weimer and Wyess [21] discussed various methodologies for determining the cut-off magnitude, and for this reason, the authors state that they use the Gutenberg-Richter law. Indeed, there are other motivations for using the Gutenberg-Richte law. It is worth remembering that determining the magnitude of completeness must be carried out independently of the Gutenberg-Richter Law. After all, recording earthquakes smaller than the completeness magnitude is due, in many cases, to the technological inability to record low magnitude earthquakes or the masking of low magnitude earthquakes due to more significant earthquakes. Please clarify these points.

Author Response

Thank you for the prompt review, our reply to learned reviewer are as following;

 

Comment 1. Sub-section 2.1. Data Collection really should be outside 2. ..... while all processes described in the second line of Figure 1 must be in another section.

Reply
We carefully evaluated the argument presented by the learned reviewer. Although, the suggestion seems appropriate, we feel that it may lead to deviation from the standard norm of MDPI Applied Sciences journal as well as other journals. As a standard matter (there can be exceptions), data collection is part of the methodology and as such is verified from a number of articles published in MDPI Applied Sciences. We are are referring to some articles which are recently published in MDPI Applied Sciences (where data collection is part of the methodology section).

https://www.mdpi.com/2076-3417/12/21/11129 (Please refer to Figure 3)
2. https://www.mdpi.com/2076-3417/13/3/1443 (Please refer to Section 2.1 Study Population)
3. https://www.mdpi.com/2076-3417/13/3/1442 (Please refer to Section 2.1 Participants)

If the learned reviewer suggests otherwise, we will comply and place it as a separate section.

Comment 2. .... It is worth remembering that determining the magnitude of completeness must be carried out independently of the Gutenberg-Richter Law. After all, recording earthquakes smaller than the completeness magnitude is due, in many cases, to the technological inability to record low magnitude earthquakes or the masking of low magnitude earthquakes due to more significant earthquakes. Please clarify these points.

Reply:
The comment is addressed and a line is added in subSection "Data Cleaning".