Next Article in Journal
Detecting the Archaeological Traces of Tar Production Kilns in the Northern Boreal Forests Based on Airborne Laser Scanning and Deep Learning
Previous Article in Journal
An Improved YOLOv5 Method to Detect Tailings Ponds from High-Resolution Remote Sensing Images
 
 
Communication
Peer-Review Record

Skillful Seasonal Prediction of Typhoon Track Density Using Deep Learning

Remote Sens. 2023, 15(7), 1797; https://doi.org/10.3390/rs15071797
by Zhihao Feng 1,†, Shuo Lv 1,†, Yuan Sun 2,*, Xiangbo Feng 3, Panmao Zhai 4, Yanluan Lin 5, Yixuan Shen 6 and Wei Zhong 2
Reviewer 1:
Reviewer 2:
Reviewer 3: Anonymous
Remote Sens. 2023, 15(7), 1797; https://doi.org/10.3390/rs15071797
Submission received: 24 February 2023 / Revised: 22 March 2023 / Accepted: 25 March 2023 / Published: 28 March 2023
(This article belongs to the Topic Big Data and Artificial Intelligence)

Round 1

Reviewer 1 Report

Comments on the manuscript entitled "Skillful seasonal prediction of typhoon track density using deep learning" by Feng et al. submitted to Remote Sensing.

 

Recommendation: Minor revision

Typhoon track density prediction is a challenge for numerical models, and no agency has the capability to predict typhoon track density up to now. Different from the previous studies, this study tries to investigate the seasonal predictions of tropical cyclones over the western North Pacific using newly developed methodology based on deep learning. This manuscript presents a verification of new method for predicted period, which is effective and useful. Comparing with the previous studies on deep learning, this paper skillfully uses transfer-learning method to greatly expand the size of training data. Their results are better than the main numerical models in most cases. This method of transfer-learning shed lights on the other weak points of numerical models (e.g., precipitation and ocean wave which are relatively lack of long-term observations), especially under complex conditions.

Overall, the topic is important as a new method for the users of seasonal forecasts of tropical cyclones over the western North Pacific, the results can support their main conclusions, the paper is well written. I am happy to recommend “Minor revision” before the acceptance of this paper. The detailed comments are listed below.

 

Minor comments:

1.     Line 16: “despite of decades of efforts” -> “despite decades of efforts” 

2.     Line 17-18: “skillful seasonal prediction of TC distribution” is not consistent with title “Skillful seasonal prediction of typhoon track density”. It would be better to make the same description. 

3.     Lines 63-65: The authors should explain the reasons why select H500, subsurface ocean temperature and zonal vertical shear to identify predictors.

4.     Line 68: “the TC track density in WNP”. It may be better to use “over” instead of “in”.

5.     Line 81-82: “includes Historical, and RCP4.5 and RCP8.5 experiments”. -> “includes Historical, RCP4.5 and RCP8.5 experiments” 

6.     Figure 3a and b: Title and top border overlap.

7.     I would like suggest read these published papers, and the authors may benefit from these papers. These papers may help the authors better understand the current key points of seasonal tropical cyclone predictions using deep learning.

 

References:

1.     Pang et al., 2021: A New Detection Framework of Tropical Cyclones from Meteorological Satellite Images with Deep Transfer Learning. Remote Sensing. 2021, 13, 1860.

2.     Wimmers A., Velden C., Cossuth J. H., 2019: Using deep learning to estimate tropical cyclone intensity from satellite passive microwave imagery. Monthly Weather Review, 2019, 147, 2261-2282.

3.     Klotzbach et al., 2019: Seasonal tropical cyclone forecasting. Tropical Cyclone Research and Review, 8, 134-148.

Author Response

We would like to thank you for your thoughtful and constructive comments. Please see the attachment for the detailed response.

Author Response File: Author Response.docx

Reviewer 2 Report

I think that this manuscript is well written. On the other hand, some discussion and/or explanation will be helpful for readers’ understanding. Considering them, I would like to accept this paper after moderate revision.

 

Comments:

- Related to Fig.4, I want to know which factor (Tsub, H500, Ushear) is important for the “heat map”.. Is it possible to write these anomalies in 2010 and 2018 and discuss about it?

 

- Fig.S4 is hard to see.. Could you please rewrite Fig.S4 with color scale?

 

Author Response

We would like to thank you for your thoughtful and constructive comments. Please see the attachment for the detailed response.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents a new method to predict seasonal tropical cyclone distribution in the Western Pacific Basin based on a convolutional neural network. The article is of importance for the field of study of tropical cyclones and provides a novel approach to determining the seasonal distribution of TC. The manuscript should be accepted for publication if the editor thinks this journal is suited as the paper does not deal with remote sensing. Also, the article should only be accepted after major revisions are done. Besides the comments below, I have included a few comments in the attached document.

There is a lot of material in the supplemental material, which is very good, but I think the Data and Methods section will benefit from more details about the methodology. It is hard to understand unless it is read several times and together with the supplementary material.

It is unclear why there is CMIP data under the RCP 4.5 and 8.5 scenarios, as the manuscript does not deal with climate change. Is it to have more events for the training? How would that influence the results?

I understand the pre-training is done with CMIP data and training with reanalysis, but it does not clearly explain why and how that contributes to the development of SeaUnet.

The authors claim that SeaUnet provides better results than the other prediction models, but they do not offer any discussion on why it is that. Clearly, this is the expected result as SeaUnet is based on the historical data it is expected to reproduce, and the others are predictive models based on physics. It is not very surprising that SeaUnet provides better results. That does not demerit the new method but should be clearly discussed and included in the conclusions.

 

Comments for author File: Comments.pdf

Author Response

We would like to thank you for your thoughtful and constructive comments. Please see the attachment for the detailed response.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The authors have addressed all the comments, and I consider the article suited for publication with minor English editing. 

Back to TopTop