Typhoon Tracks Prediction with ConvLSTM Fused Reanalysis Data
Round 1
Reviewer 1 Report
This paper proposed to use more convolution layer in ConvLSTM units to improve the model performance for image features extraction, and mark position for obtaining each physical variables. The idea introduced by the authors is quite interesting. However, there are still some issues that need to be improved.
- The manuscript needs English proofreading. Moreover, it is strongly recommended to take the grammar check-up and language polishing. For example, in line 9, 'have bought' seems to be corrected as 'have brought.'
- To improve the model performance, a convolutional layer in ConvLSTM units was used. However, it is difficult to admit that it has sufficiently considered the influence of physical factors on typhoon movement by using convolutional layer only.
- Canonical correlation analysis (CCA) is a very powerful multivariate tool to jointly investigate relationships among multiple data sets. However, the samples must be pairwise for CCA realization. Moreover, the class information of the samples should be fully exploited in CCA. For the readers’ better understanding, the authors should mention more clearly and deeply regarding this problem.
- In line 53-54, researchers already proposed to predict typhoon tracks with ConvLSTM [11]. This paper should mention how this article is different from the previous researches in terms of contributions and novelty.
- In Figure 7, to identify the prediction error and performance of proposed method conspicuously, it is necessary to add some extended scope images for readers.
- In the Section 3-4 , 'comparison with other methods', the comparison results are supposed to contain additional contents that explain which event was compared, and whether "such a feature" described in lines 364-366 was added to the training process of the comparison target in ref. [10, 11], [30-35] or not.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Typhoon Tracks Prediction with ConvLSTM Fused Reanalysis Data describes an innovative method of predicting dangerous events such as typhoon traces. The metotology used is the use of artificial neural networks in the ConvLSTM structure as well as Canonical Correlation Analysis and Gray Relation Analysis. The article is very interesting and correctly written from a scientific perspective. The literature review is sufficient, there is also a reference and comparison to other studies in this field, which is a great value of the work. The only deficiency that is visible here is the lack of analysis with statistical tools and probability calculus. Perhaps the authors could introduce the consideration of their specific case to the analysis of rare events. I think this could enrich the article significantly.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
All of my comments are well answered and this paper has been improved.