Tropical Cyclone Intensity Prediction Using Deep Convolutional Neural Network
Round 1
Reviewer 1 Report
The manuscript titled " Tropical Cyclone Intensity Prediction Using Deep Convolutional Neural Network" by Xiao-Yan Xu, Min Shao, Pu-Long Chen, and Qin-Geng Wang.
The manuscript presents the simulation of tropical cyclone intensity (TCI), minimum central pressure (MCP), and maximum 2 minutes mean wind speed at near centre (MWS) based on ocean, atmospheric reanalysis data, using deep convolutional neural network (CNN) method.
In the manuscript the authors explore the interpretability of model structure, sensitivity experiments with various combinations of predictors.
The authors claimed that model results show simplified VGG-16 (VGG-16s) outperforms the other two general models (LeNet-5 and AlexNet).
Overall paper is well written and study are interesting, and can be valuable for scientific community. However, the study need some more analysis and evidence to come to claimed conclusion.
Major Comments
- I would suggest authors to include the comparison/validation (other than R2,RMSE,SMAPE & MAE) plots and some more analysis for TCI, MCP and MWS.
Minor Comments
- Use some appropriate word instead of ‘mine’ from the sentence ‘which was designed to process images exclusively, has been widely applied to mine information from satellite imagery data [16,17]’
The given questions and suggestions recommended to be answered and included in text. Finally, I recommend to take into account these suggestions and to accept the Article.
Author Response
Dear Reviewers:
Thanks for your constructive suggestion, which is highly appreciated. We have carefully scrutinized the manuscript, and made corresponding revisions. Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In the paper, machine learning is used to obtain tropical cyclone parameters from data of ERA5 reanalysis and Best Tracks of tropical hurricane. The VGG-16 model was adopted for the studied task, and the results were estimated using statistical characteristics. In the work, advantages of the applied VGG-16s model over LeNet5 and AlexNet are shown. The text is predominantly comprehensible, the methods and results are presented in the required form and conclusions are clearly related to the main part of the work. The main input parameters are identified for different tropical cyclone characteristics (intensity, minimum central pressure, mean wind speed near center). Moreover, future prospects are presented in detail. Probably, the discussion could be expanded but a deeper analysis of the model possibilities should be done in the next publications.
Proposals are collected in the attached file. They are mainly technical or involve little text errors. Accordingly, I assume that the paper should be minimally corrected and can be published in Atmosphere after minor revision.
Comments for author File: Comments.pdf
Author Response
Dear reviewer: Thank you very much for your kindly comments on our manuscript. There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. In what follows, we would like to answer the questions you mentioned and give detailed account of the changes made to the original manuscript. Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
- VGG 16 model is used for prediction and the results are compared with LeNet-5 and AlexNet.
- Though, VGG 16 performs better than other two models, how Vanishing gradient problem in is solved in VGG16.
- Justify how the learning rate values are chosen for the models.
- How to perform joint modelling of weather variables and the predictors used in the study.
- Mention the mathematical notations for the various parameters used in the study and how they are interrelated.
- Specify the reason why ResNet architecture is not considered for comparison which incorporates the residual learning and overcomes the complexity of VGG 16 layers.
- Remove the discussion section from the conclusion and can be given separately. Also, improve the discussion section with detailed analysis.
- The performance metrics like accuracy, precision, recall, F1 score etc. are mentioned in the paper. But the computed values of the metrics are missing.
- Conclusion and future work can only be highlighted at the end.
Author Response
Dear reviewer:
Thank you for your comments concerning our manuscript. These comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made corrections which we hope meet with approval. Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Thanks for the considering and including the point which point out in 1st round of review. But there is still some point which you can just improve.
My comment append below:
As you added conclusion section and then after discussion section. I think after conclusion don't keep discussion section.
After considering my comments I would like to recommend, accept the paper for publication.
Author Response
Dear reviewer:
Thank you very much for your time involved in reviewing the manuscript. These comments are very constructive and valuable. We have modified our manuscript according to the suggestions.
Response to Reviewer 1 Comments
Point 1: As you added conclusion section and then after discussion section. I think after conclusion don't keep discussion section.
Response 1: Thank you for pointing out this problem. We are grateful for the suggestion. We have changed the order of ‘Discussion’ and ‘Conclusion’ section.
Reviewer 2 Report
During the first review I did not find any important errors but the new version is supplemented and improved. Minor errors noticed in the first version are corrected. Now there is only one incomprehensible item with the section ‘Discussion’ before ‘Conclusions’. I assume that their order needs to change. A few more notes are related to possible typos and can be easily corrected. Accordingly, the manuscript can be published in ‘Atmosphere’ almost unchanged.
Minor remarks:
Abstract, L. 4 from below. ‘850hPa and 500hPa’ – spaces should be before ‘hPa’. This also applies to the main text.
P. 5, L. 1 from below. ‘F1-socre’ – should be ‘F1-score’.
P. 6, L. 7. ‘the comparsion observed values and prediction results’ – probably, ‘the comparison between observed values and prediction results’.
P. 6, Par. 2, L. 8. ‘are shown in figure 4’ – should be ‘Figure 4’ (capital letter).
P. 9, Section 4, L. 5. ‘VGG-16s’ model structure’ – it seems that the apostrophe after ‘VGG-16s’ is unnecessary.
P. 9. The place of the section ‘Discussion’ after ‘Conclusion’ is too unusual. When such a structure is used, it seems that ‘Conclusions’ become local and do not include arguments from ‘Discussion’. In my opinion, this is incorrect and the order of the sections should be changed to typical: ‘4. Discussion’ and ‘5. Conclusions’.
P. 9, Section 5, L. 4. ‘efficient method’ – probably, ‘efficient methods’.
P. 9, L. 3-4 from below. ‘In the study, we chose VGG-16s as sensitivity experiments model and studied the interpretability of the model structure’ – it is worth editing the phrase that is grammatically unclear.
P. 10, [12]. ‘atlantic basin’ – should be ‘Atlantic basin’.
P. 11, [16]. ‘Chen, B.; Chen, B.-F.; Lin, H.-T.; Acm.’ – who is ‘Acm’? The paper was written by only three co-authors.
P. 11, [16]. ‘ENGLAND’ – capital letters are unnecessary; ‘United Kingdom’ is more appropriate than ‘England’.
Author Response
Dear reviewer:
We appreciate your clear and detailed feedback and hope that the explanation has fully addressed all of your concerns.
Response to Reviewer 2 Comments
Point 1: Abstract, L. 4 from below. ‘850hPa and 500hPa’ – spaces should be before ‘hPa’. This also applies to the main text.
Response 1: We have corrected it throughout the manuscript.
Point 2: P. 5, L. 1 from below. ‘F1-socre’ – should be ‘F1-score’.
Response 2: We have corrected this mistake.
Point 3: P. 6, L. 7. ‘the comparsion observed values and prediction results’ – probably, ‘the comparison between observed values and prediction results’.
Response 3: We apologize for this confusion. We have modified this sentence.
Point 4: P. 6, Par. 2, L. 8. ‘are shown in figure 4’ – should be ‘Figure 4’ (capital letter).
Response 4: We have corrected it.
Point 5: P. 9, Section 4, L. 5. ‘VGG-16s’ model structure’ – it seems that the apostrophe after ‘VGG-16s’ is unnecessary.
Response 5: We have corrected it.
Point 6: P. 9. The place of the section ‘Discussion’ after ‘Conclusion’ is too unusual. When such a structure is used, it seems that ‘Conclusions’ become local and do not include arguments from ‘Discussion’. In my opinion, this is incorrect and the order of the sections should be changed to typical: ‘4. Discussion’ and ‘5. Conclusions’.
Response 6: Thank you for raising this important issue. We have changed the order of these two sections.
Point 7: P. 9, Section 5, L. 4. ‘efficient method’ – probably, ‘efficient methods’.
Response 7: We have corrected it.
Point 8: P. 9, L. 3-4 from below. ‘In the study, we chose VGG-16s as sensitivity experiments model and studied the interpretability of the model structure’ – it is worth editing the phrase that is grammatically unclear.
Response 8: We apologize for this confusion. We have modified this expression.
Point 9: P. 10, [12]. ‘atlantic basin’ – should be ‘Atlantic basin’.
Response 9: We have corrected it.
Point 10: P. 11, [16]. ‘Chen, B.; Chen, B.-F.; Lin, H.-T.; Acm.’ – who is ‘Acm’? The paper was written by only three co-authors.
Response 10: We have corrected this mistake.
Point 11: P. 11, [16]. ‘ENGLAND’ – capital letters are unnecessary; ‘United Kingdom’ is more appropriate than ‘England’.
Response 11: We have corrected it.
