A Reconstructing Model Based on Time–Space–Depth Partitioning for Global Ocean Dissolved Oxygen Concentration

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

With the intensification of global climate change, the marine environment has also been constantly changing. Many international institutions have developed a series of grid data products around changes in the marine environment, such as ocean temperature, salinity,pCO₂ etc. Dissolved oxygen plays a crucial role on global marine ecosystem health. However, due to the lack of dissolved oxygen observation data, there have been no good products available at present.Not like the other marine environmental elements, such as SST, Chla, SLA etc, which can be observed from remote sensing. How to use the Argo profiles to develop the spatial gridded-dataset is a hot issue. To deal with this challenge, the authors designed a novel method based machine learning, named TSD-ML. The core idea of the TSD-ML adopted the hierarchical partitioning in space and depth to establish the relationship between dissolved oxygen and temperature and salinity equipped with BGC Argo profiles, and then develop the dissolved oxygen from the core Argo profiles with temperature and salinity. The manuscript is very interesting and important, which can contributes to develop global ocean dissolved oxygen concentration dataset. Thus, this manuscript can be accepted after finishing the following minors.

- Line 92 – The Batur et al. paper doesn’t seem like a good support here – it looks like a different study area (lake) and not applicable to the global ocean. 

 - Line 140-142 – Why would the partitioning dataset only use DO and not temperature and salinity? 

 - Line 185-188 – If you are only selecting ± 5(or ±3) meters from every depth layer, does that mean you are throwing out all data between, say 205 and 245 m? 

 - Line 345 – What is the spatial distribution of the GLODAPv2 samples you selected? You didn't describe it.

 - Line 349 – Where do you get the numbers in the paragraph starting at line 349 from? It's better to give precise values.

 - Line 461– Is there any reference for the definition of OMZ range here? If so, please provide relevant literature.

 Why do you not use any shipboard data, as in Sharp et al.?  Seems like an omission of a large and valuable dataset.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

In order to solve the problem of unevenly distributed Argo DO data, the author developed a novel machine learning model(TSD-ML) to enhance reconstruction accuracy in data-sparse regions. TSD-ML partitions Argo data into segments based on time, depth, and spatial dimensions, and conducts model training for each segment. The results show that TSD-ML significantly improves the reconstruction accuracy in areas with uneven distribution of DO data. The conclusion is consistent with the evidence and arguments presented. this study fits the scope of this journal and will be of interest to its readers. I recommend this work for publication after some revisions:

1. On line 16 of the manuscript, the authors mention TSD-ML for the first time, which is an abbreviation, and the authors should give the full name of "TSD-ML" as it appears for the first time. Although the author explains this abbreviation on line 163.

2. In Section 2.2.1 of the manuscript, the author mentioned Time-Space-Depth partition and introduced its construction method in detail. This is a good idea. Can the author show the results of Time-Space-Depth partition in more detail in the results chapter?

3. In Section 3.1 of the manuscript, the author mentioned "500 independent BGC Argo observations". Are these points evenly distributed on the global scale? Is it representative?

4. In lines 204-206 of the manuscript, the author mentions "Spline with Barriers (SWP) interpolation". Does the author have any other interpolation methods? Compared with other methods, what are the advantages of this interpolation method?

5. In lines 334-336 of the manuscript, the author mentions the difference between the two data and the reasons for the difference. Can you explain it in more detail?

6. In lines 211-212 of the manuscript, the author uses K-means and SSE as the basis for spatial partitioning. However, the results of K-means clustering are easily affected by the initial value and the number of iterations. How did the author eliminate this effect？

Author Response

Please see the attachment.

Author Response File: Author Response.docx