The manuscript presents a study on the submesoscale eddy identification. The high-resolution chlorophyII from GOCI I is used. The technology and methodology employed in this paper is well presented and the results are notably impressive. This work is poised to contribute to the future research in understanding the marine dynamics and environmental process.

Comments: 

1. Coda availability and code example. It will be beneficial if the authors could provide code for this research. Making the code available would not only enhance transparency but also facilitate further research by enabling other researchers to replicate and build upon the findings. Besides, to code examples for how to use the dataset would be highly beneficial.

2. A discussion on the selection and optimization of these parameters would be particularly valuable, like learning rate, the number of weights.

3. How the author solves potential biases introduced by the predominance of cyclonic over anticyclonic eddies in the training dataset is not mentioned in the manuscript. In general, an imbalance in the number of data samples can significantly influence the behavior of a deep learning model.

Technical Corrections: 

Paragraph 15: confidence minimum of 0.2 is unclear. “minimum confidence threshold of 0.2” may be more clear.

 Ten years of eddies in the NE pacific ocean are presented in this paper.  The fine spatial and temporal patterns derived from  the GOCI-I chlorophyll-a data is very interesting although the widely-used deep leaning model for eddy extraction is not so novel. 

This paper can be published after amending some information:

In Fig.10, please amend error bar to illustrate the standard deviations in hourly ,  monthly and yearly eddy numbers; amend the inter-annual changes.  And, accordingly, the impacts of cloud cover on the above patterns should be discussed.  It will be much better if the cloud dataset is presented.