A Novel Tropical Cyclone Track Forecast Model Based on Attention Mechanism
Round 1
Reviewer 1 Report
1. In the fourth paragraph of "Introduction", the authors should clearly point out the limitations of integrating 2d features with 3d features in recent literature, and then propose your novelty of the 2d and 3d combination module.
2. In the last paragraph of “Introduction”, the contributions are not discussed precisely. Authors only wrote the new technique/model name, but did not explain the novelty of these models, and why do you use them.
3. On page 4, line 168, the number of this title is incorrect.
4. On page 6, line 206, why 31°×31°?
5. For Data Preprocessing, does that mean the tropical cyclone center is always at the center of the 2D structure and 3D structure? What if the overall tropical cyclone path is beyond the figure size?
6. The "MLP" should give the full name when it appears the first time in the article (page 8, line 274).
7. Figure 7 and Figure 8 are not depicted explicitly, what are dense feature, sparse feature, and embedding vectors in this tropical cyclone scenario, and what is X0? Does the same color in figure 7 and figure 8 represent the same things? Do these figures reflect the time series characteristics?
8. The proposed model should be compared with some traditional meteorological methods and other up-to-date deep learning models in this research area.
9. Figure 10 and Figure 11 are exactly the same, and what does the baseline tropical cyclone path mean?
10. Does the model have the capability to predict the future 36, 48 hours path? The limitations of this model should be discussed more.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
please see the attached file.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper deals with the task of forecasting Tropical Cyclone Track using deep learning and the attention mechanism. The authors deeply understand the deep learning concepts.
- The paper is well-written, has well coherence, and is well-structured.
- The introduction section explains the concept in a nice and understandable way. Also, it mentions the motivation and the contribution of this study, something that is really important.
- The dataset is sufficient and appropriate for conducting various experiments.
- The evaluation results are analytically explained and fruitful discussion paragraphs are presented.
Suggestions and observations:
- It is recommended that the authors upload their Code on GitHub.
- It is suggested the authors mention the used technologies for conducting the experiments and training their model (e.g. did you use PyTorch or any other deep learning model?)
- In the last paragraph of the Introduction section, I suggest you mention the contributions of this paper in a bulleted list.
- How do you assure the tropical cyclone forecasting model produces a valid combination of latitude and longitude?
- Could you please run a statistical test to show that the results have statistical significance in Tables 5, 6, 7, and 8?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
1. The grammar of this manuscript needs a thorough proof reading.
2. The description of Figure 7 said "We take 2D features of tropical cyclones as dense features (light blue circles in Figure 7) and 3D features as sparse features (dark blue circles in Figure 7). After embedding, they are transformed into low-dimensional dense features." But from the figure, it seems like the sparse feature is transformed into dense feature and embedding vectors. What does the arrow from sparse feature to embedding vector mean?
3. Page 15, line 462, is CMO a deep learning method? How does the Central Meteorological Observatory of China forecast tropical cyclone path? Where are the CMO results from?
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Line 384: "The CNN was compared with several state-of-art machine learning methods". i still believe the use of 'state-of-art' is not inappropriate, u may be considering 'existing models'. After all, i suggest to accept the manuscript since the authors have made a major revision.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Dear Authors,
Thank you for your changes. There are a few things you need to improve further.
* Response 1: Could you please translate the comments of the code into English?
* Response 2: You mention you use "the Python deep learning framework to build the model and conduct our experiments". However, Python is not a deep learning framework but a programming language. Could you please specify which is the exact deep learning framework?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
