Novel Deep Learning Approaches for Mapping Variation of Ground Level from Spirit Level Measurements

Zarzoura, Fawzi; Kaloop, Mosbeh; Samui, Pijush; Hu, Jong Wan; Sabri, Md Shayan; ElGharbawi, Tamer

doi:https://doi.org/10.5194/gmd-2023-62

Preprints

https://doi.org/10.5194/gmd-2023-62

Preprints

Submitted as: development and technical paper

12 Apr 2023

Submitted as: development and technical paper |

| 12 Apr 2023

Status: this preprint has been withdrawn by the authors.

Novel Deep Learning Approaches for Mapping Variation of Ground Level from Spirit Level Measurements

Fawzi Zarzoura, Mosbeh Kaloop, Pijush Samui, Jong Wan Hu, Md Shayan Sabri, and Tamer ElGharbawi

Abstract. This study investigates the use of new machine learning techniques in mapping variation in ground levels based on ordinary spirit levelling (SL) measurements. Convolution Neural Network (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and bi-directional LSTM (BI-LSTM) were developed and compared in the current study to estimate the leveling through SL measurements. SL measurements of the Manzalla region, Egypt, were used in the current study. 3253 datasets of SL observation points, including 229 benchmarks of precise levelling (PL), were used to design and verify the proposed model’s results. The results show the developed LSTM model outperforms CNN, RNN, and BI-LSTM in modeling ground leveling in the training and testing stages. The root mean square error and correlation determination of the LSTM model are 7.4 cm and 0.99, respectively, in the testing stage. The accuracy of mapping ground levelling through the developed LSTM model is close to 99 % in terms of model error.

This preprint has been withdrawn.

Received: 15 Mar 2023 – Discussion started: 12 Apr 2023

Download & links

Preprint (PDF, 1761 KB)

Withdrawal notice
This preprint has been withdrawn.
Preprint (1761 KB)

Supplement (114 KB)

Download & links

This preprint has been withdrawn.

Fawzi Zarzoura et al.

Interactive discussion

Status: closed

CEC1:
'Comment on gmd-2023-62', Juan Antonio Añel, 06 May 2023

Dear authors,
Unfortunately, after checking your manuscript, it has come to our attention that it severely infringes the "Code and Data Policy" of our journal.

https://www.geoscientific-model-development.net/policies/code_and_data_policy.html
You have not published the code that you use in your manuscript and which is necessary to replicate your work. What is more, your manuscript does not even includes the mandatory "Code Availability" section. In this way, we will have to reject your manuscript for publication unless you reply to this comment with the information about the repository (DOI and link) for the code (one acceptable according to our policy - please, check it carefully).
I should note that, actually, your manuscript should not have been considered in our journal, given this lack of compliance with our policy. Therefore, the current situation with your manuscript is irregular.
Please, address and solve this issue in a prompt manner, replying with the necessary information.
Also, in case you reply to this comment with the necessary information, and we consider this problem solved, remember that you must include in any potentially reviewed version of your manuscript the modified 'Code and Data Availability' section, with the DOI of the code.
Moreover, I should note that the data added as supplementary material is in .xlsx format, which is a format that depends on proprietary software to access the data. It would be better if you save the data in an open ISO standard format, such as .ods, or even .dat or .csv. Also, the supplementary material only contains the data, without any explication about its structure, what is each column or row, etc. Please, you should add some metadata, maybe in an additional Readme file.
Juan A. Añel
Geosci. Model Dev. Executive Editor

Citation: https://doi.org/10.5194/gmd-2023-62-CEC1
- AC1:
  'Reply on CEC1', Jong Hu, 14 May 2023
  
  Dear Editor
  Thank you for your comment. Please check the attachment. We have submitted the codes and datasets.
  Regards
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC1
  - CEC2: 'Reply on AC1', Juan Antonio Añel, 16 May 2023
    
    Dear authors,
    Many thanks for sharing the code. Doing it as supplementary material is enough. However, I would recommend you change the name "file" in the Python code and use the actual name of the file.
    Please, if your manuscript is accepted for publication or goes through additional review stages, do not forget to update the supplementary material with the files attached to this comment and add the Code Availability section to the manuscript.
    Regards,
    Juan A. Añel
    Geosci. Model Dev. Executive Editor
    
    Citation: https://doi.org/10.5194/gmd-2023-62-CEC2
    
    AC2: 'Reply on CEC2', Jong Hu, 16 May 2023
    
    Dear Editor
    Thank you for your comment; we will add the "code availability" section in the paper with the revised version form.
    My best regards
    
    Citation: https://doi.org/10.5194/gmd-2023-62-AC2
RC1:
'Comment on gmd-2023-62', Junye Wang, 04 Jun 2023
The authors nvestigated mapping variation in ground levels using machine learning techniques. They compared the ordinary spirit levelling (SL) measurements using four MLs, including Convolution Neural Network (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and bi-directional LSTM (BI-LSTM), in the Manzalla region, Egypt. Their results showed the developed LSTM model outperforms CNN, RNN, and BI-LSTM in modeling ground leveling in the training and testing stages. While the subject is relevant and interesting, some details have been missing. Furthermore, it is unclear what are novelties of this paper. The authors should highlight the novelties of this paper. The reviewer lists some comments as the below:
This paper presents a comparison of estimating SL using several MLs. However, there are many ML algorithms. It is unclear why the authors selected these methods. What advantages of every method did you select them?

L77-82, these should be merged with Method section since they are description of methods.

L86-124, the following paragraphs are a brief description of several methods rather than a theory summary. I would suggest you add more details of every algorithm used in this study.

It is unclear how you implement the simulations, such as training, testing, input and output variables. You should present what software of MLs have been used for this study if your codes were not in-house. The authors should add a section to present your implementation or describe them after each algorithm.

L229-232, what means E, N and Z in Table 1? Why did you not use input and output variables that have physical meaning?

In Table 2, how did you calculate “Total Score”?

I would suggest you add a figure to show results of CNN and RNN.
Citation: https://doi.org/10.5194/gmd-2023-62-RC1
- AC3: 'Reply on RC1', Jong Hu, 24 Jun 2023
  
  Dear Reviewer
  Thank you for your comments. We have revised and improved the manuscript based on your comments. Please check the attachment.
  thank you
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC3
RC2:
'Comment on gmd-2023-62', Deepak Subramani, 05 Sep 2023

The authors claim to develop a deep learning method for mapping variation of ground level from spirit level measurements. There are several major issues with this manuscript as follows:
1. What is the objective of the work? What are the input and output? How is the model trained? Fig 1,2,3 and Sec 2 are describing known Deep Learning models, the authors have not explained what is the input and output here.
2. Since the objective of the model is not clear, the results and discussion section cannot be followed. What result are the authors discussing?
3. Section 3 of data collection needs to have more details of the collected data and how to access it. A generic statement of it can be found in the supplementary information is not enough.
A clearer explanation is necessary including the novelty of the work. Thus, the manuscript cannot be published in its current form

Citation: https://doi.org/10.5194/gmd-2023-62-RC2
- AC4: 'Reply on RC2', Jong Hu, 06 Sep 2023
  
  Dear Editor
  The authors are grateful for the Editor's constructive comments, which helped the authors to improve the quality and organization of the manuscript. The manuscript has been thoroughly revised and rechecked incorporating the reviewer's comments. Please find attached the revised manuscript and our responses to your comments.
  Thank you
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC4
EC1:
'Comment on gmd-2023-62', Deepak Subramani, 07 Sep 2023

Dear Authors,
Thank you for trying to address the comments from referees. The new information sheds more light on what exactly you have done and further reveals major issues with the work. Why is a sequential model (RNN, LSTM,1D CNN etc) needed for a task with only two input variables? Please show that all the features of CNN and LSTM described are being used in the two variable input models. The final results also suggest that there is overfitting with an overparametrized neural network. I suggest the authors rethink the utility of using a DNN, compare it with linear regression or a simple MLP with a couple of dense layers, and clearly show why other models are needed. A gradient-boosted tree model may also be tried. However, even if all that is done, the novelty of the work is minimal, except perhaps in the data gathering. What new scientific insight is the model giving? Is it generally usable? What novel model development and/or training strategy was used? Is the complexity of the model justified? All these are important considerations for GMD. Just a motivation that soft computing methods are not available for this problem is not enough.
I would also encourage the authors to use the best practices for training ML models in the future. Please consider cross-validation and careful hyperparameter tuning instead of "trial-and-error."
Considering all the above, I would encourage the authors not to submit a revised manuscript but instead work on a substantially revised model development in case they consider GMD in the future for this work. The authors may also consider reporting just the data with the model analyses, which may be suitable for other venues.
Thanks,
Deepak

Citation: https://doi.org/10.5194/gmd-2023-62-EC1
- AC5: 'Reply on EC1', Jong Hu, 07 Sep 2023
  
  Dear Editor
  Thank you for your cooperation
  We will attempt to revise and improve the manuscript; we would like to thank you for your effort and time.
  Regards
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC5

Interactive discussion

Status: closed

CEC1:
'Comment on gmd-2023-62', Juan Antonio Añel, 06 May 2023

Dear authors,
Unfortunately, after checking your manuscript, it has come to our attention that it severely infringes the "Code and Data Policy" of our journal.

https://www.geoscientific-model-development.net/policies/code_and_data_policy.html
You have not published the code that you use in your manuscript and which is necessary to replicate your work. What is more, your manuscript does not even includes the mandatory "Code Availability" section. In this way, we will have to reject your manuscript for publication unless you reply to this comment with the information about the repository (DOI and link) for the code (one acceptable according to our policy - please, check it carefully).
I should note that, actually, your manuscript should not have been considered in our journal, given this lack of compliance with our policy. Therefore, the current situation with your manuscript is irregular.
Please, address and solve this issue in a prompt manner, replying with the necessary information.
Also, in case you reply to this comment with the necessary information, and we consider this problem solved, remember that you must include in any potentially reviewed version of your manuscript the modified 'Code and Data Availability' section, with the DOI of the code.
Moreover, I should note that the data added as supplementary material is in .xlsx format, which is a format that depends on proprietary software to access the data. It would be better if you save the data in an open ISO standard format, such as .ods, or even .dat or .csv. Also, the supplementary material only contains the data, without any explication about its structure, what is each column or row, etc. Please, you should add some metadata, maybe in an additional Readme file.
Juan A. Añel
Geosci. Model Dev. Executive Editor

Citation: https://doi.org/10.5194/gmd-2023-62-CEC1
- AC1:
  'Reply on CEC1', Jong Hu, 14 May 2023
  
  Dear Editor
  Thank you for your comment. Please check the attachment. We have submitted the codes and datasets.
  Regards
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC1
  - CEC2: 'Reply on AC1', Juan Antonio Añel, 16 May 2023
    
    Dear authors,
    Many thanks for sharing the code. Doing it as supplementary material is enough. However, I would recommend you change the name "file" in the Python code and use the actual name of the file.
    Please, if your manuscript is accepted for publication or goes through additional review stages, do not forget to update the supplementary material with the files attached to this comment and add the Code Availability section to the manuscript.
    Regards,
    Juan A. Añel
    Geosci. Model Dev. Executive Editor
    
    Citation: https://doi.org/10.5194/gmd-2023-62-CEC2
    
    AC2: 'Reply on CEC2', Jong Hu, 16 May 2023
    
    Dear Editor
    Thank you for your comment; we will add the "code availability" section in the paper with the revised version form.
    My best regards
    
    Citation: https://doi.org/10.5194/gmd-2023-62-AC2
RC1:
'Comment on gmd-2023-62', Junye Wang, 04 Jun 2023
The authors nvestigated mapping variation in ground levels using machine learning techniques. They compared the ordinary spirit levelling (SL) measurements using four MLs, including Convolution Neural Network (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and bi-directional LSTM (BI-LSTM), in the Manzalla region, Egypt. Their results showed the developed LSTM model outperforms CNN, RNN, and BI-LSTM in modeling ground leveling in the training and testing stages. While the subject is relevant and interesting, some details have been missing. Furthermore, it is unclear what are novelties of this paper. The authors should highlight the novelties of this paper. The reviewer lists some comments as the below:
This paper presents a comparison of estimating SL using several MLs. However, there are many ML algorithms. It is unclear why the authors selected these methods. What advantages of every method did you select them?

L77-82, these should be merged with Method section since they are description of methods.

L86-124, the following paragraphs are a brief description of several methods rather than a theory summary. I would suggest you add more details of every algorithm used in this study.

It is unclear how you implement the simulations, such as training, testing, input and output variables. You should present what software of MLs have been used for this study if your codes were not in-house. The authors should add a section to present your implementation or describe them after each algorithm.

L229-232, what means E, N and Z in Table 1? Why did you not use input and output variables that have physical meaning?

In Table 2, how did you calculate “Total Score”?

I would suggest you add a figure to show results of CNN and RNN.
Citation: https://doi.org/10.5194/gmd-2023-62-RC1
- AC3: 'Reply on RC1', Jong Hu, 24 Jun 2023
  
  Dear Reviewer
  Thank you for your comments. We have revised and improved the manuscript based on your comments. Please check the attachment.
  thank you
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC3
RC2:
'Comment on gmd-2023-62', Deepak Subramani, 05 Sep 2023

The authors claim to develop a deep learning method for mapping variation of ground level from spirit level measurements. There are several major issues with this manuscript as follows:
1. What is the objective of the work? What are the input and output? How is the model trained? Fig 1,2,3 and Sec 2 are describing known Deep Learning models, the authors have not explained what is the input and output here.
2. Since the objective of the model is not clear, the results and discussion section cannot be followed. What result are the authors discussing?
3. Section 3 of data collection needs to have more details of the collected data and how to access it. A generic statement of it can be found in the supplementary information is not enough.
A clearer explanation is necessary including the novelty of the work. Thus, the manuscript cannot be published in its current form

Citation: https://doi.org/10.5194/gmd-2023-62-RC2
- AC4: 'Reply on RC2', Jong Hu, 06 Sep 2023
  
  Dear Editor
  The authors are grateful for the Editor's constructive comments, which helped the authors to improve the quality and organization of the manuscript. The manuscript has been thoroughly revised and rechecked incorporating the reviewer's comments. Please find attached the revised manuscript and our responses to your comments.
  Thank you
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC4
EC1:
'Comment on gmd-2023-62', Deepak Subramani, 07 Sep 2023

Dear Authors,
Thank you for trying to address the comments from referees. The new information sheds more light on what exactly you have done and further reveals major issues with the work. Why is a sequential model (RNN, LSTM,1D CNN etc) needed for a task with only two input variables? Please show that all the features of CNN and LSTM described are being used in the two variable input models. The final results also suggest that there is overfitting with an overparametrized neural network. I suggest the authors rethink the utility of using a DNN, compare it with linear regression or a simple MLP with a couple of dense layers, and clearly show why other models are needed. A gradient-boosted tree model may also be tried. However, even if all that is done, the novelty of the work is minimal, except perhaps in the data gathering. What new scientific insight is the model giving? Is it generally usable? What novel model development and/or training strategy was used? Is the complexity of the model justified? All these are important considerations for GMD. Just a motivation that soft computing methods are not available for this problem is not enough.
I would also encourage the authors to use the best practices for training ML models in the future. Please consider cross-validation and careful hyperparameter tuning instead of "trial-and-error."
Considering all the above, I would encourage the authors not to submit a revised manuscript but instead work on a substantially revised model development in case they consider GMD in the future for this work. The authors may also consider reporting just the data with the model analyses, which may be suitable for other venues.
Thanks,
Deepak

Citation: https://doi.org/10.5194/gmd-2023-62-EC1
- AC5: 'Reply on EC1', Jong Hu, 07 Sep 2023
  
  Dear Editor
  Thank you for your cooperation
  We will attempt to revise and improve the manuscript; we would like to thank you for your effort and time.
  Regards
  
  Citation: https://doi.org/10.5194/gmd-2023-62-AC5

Fawzi Zarzoura et al.

Supplement

https://doi.org/10.5194/gmd-2023-62-supplement

Fawzi Zarzoura et al.

Viewed

Total article views: 599 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
481	91	27	599	28	5	6

HTML: 481
PDF: 91
XML: 27
Total: 599
Supplement: 28
BibTeX: 5
EndNote: 6

Views and downloads (calculated since 12 Apr 2023)

Month	HTML	PDF	XML	Total
Apr 2023	148	23	5	176
May 2023	113	22	10	145
Jun 2023	63	11	4	78
Jul 2023	55	8	0	63
Aug 2023	44	5	0	49
Sep 2023	58	22	8	88

Cumulative views and downloads (calculated since 12 Apr 2023)

Month	HTML	PDF	XML	Total
Apr 2023	148	23	5	176
May 2023	113	22	10	145
Jun 2023	63	11	4	78
Jul 2023	55	8	0	63
Aug 2023	44	5	0	49
Sep 2023	58	22	8	88

Viewed (geographical distribution)

Total article views: 572 (including HTML, PDF, and XML) Thereof 572 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 14 Sep 2023

Download

This preprint has been withdrawn.

Preprint (1761 KB)
Metadata XML

Short summary

The study aims to map variation in ground levels based on ordinary spirit levelling (SL) measurements. New machine learning techniques were developed and compared in the current study to estimate the leveling through SL measurements. The results show the developed LSTM model outperforms CNN, RNN, and BI-LSTM in modeling ground leveling in the training and testing stages. The accuracy of mapping ground levelling through the developed LSTM model is close to 99 % in terms of model error.


Total:	0
HTML:	0
PDF:	0
XML:	0