Statistical Approach for the Imputation of Long-Term Seawater Data Around the Korean Peninsula from 1966 to 2021

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Ocean characteristics around the Korean Peninsula were analyzed using four statistical methods. The results showed that the long-term change in 20 water temperature aligns with previous studies. The title is not informative. The manuscript also suffers from a lack of novelty and poor organization. The main methodology was given in the Introduction, and some results were demonstrated in the methods section. First, the manuscript needs a substantial revision from an organization and originality perspective. some other minor points are given below:

- Add a scale bar and north sign to Figure 2.

- Why is the data divided into three layers? not less or more. Justify.

-Legend in Figure 6 cannot be read.

- To analyze trends, why don't use more recent and innovative techniques?

- an in-depth discussion of the results is needed.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Comments 1: Add a scale bar and north sign to Figure 2.

Response 1: We added a scale bar and north sign to Figure 2.

Comments 2: Why is the data divided into three layers? not less or more. Justify.

Response 2: The density of seawater exhibits distinct distribution patterns in the surface, middle, and bottom layers. The surface layer experiences mixed by wind stress and cooling in winter and stratified by heating in summer, while the middle layer contains a thermocline. The bottom layer has a higher density and remains relatively stable with minimal influence from surface conditions. For these reasons, this study categorizes the ocean into three distinct layers.

Comments 3: Legend in Figure 6 cannot be read.

Response 3: We revised scale bar and contour line in Figure 6.

Comments 4: To analyze trends, why don't use more recent and innovative techniques?

Response 4: Machine learning and deep learning offer strong predictive capabilities, their black-box nature limits interpretability. Since our study aims to analyze ocean temperature trends rather than solely predict the future, we chose Piecewise Regression (PR) for its transparency and effectiveness in detecting trend changes at key inflection points (2000, 2009).

PR provides robust estimates with limited data, reducing the risk of overfitting common in deep learning models. It is more flexible than simple linear regression while remaining interpretable, making it well suited for assessing climate change impacts over time. We have revised the manuscript to clarify this methodological choice and appreciate the reviewer’s valuable feedback.

We added some paragraph in methods (page 7) and discussion (page 13).

"While modern machine learning and deep learning techniques provide strong predictive power, they have a black-box nature that makes it difficult to interpret results. Therefore, we want to use a traditional linear-based model for trend changes and interpretability."

"The aim of this study is not to simply predict the future, but to analyze the evolution of ocean temperatures based on historical data and compare the impacts of climate change over time. PR can quantitatively analyze trend changes over time based on a specific point in time, making it well suited to effectively assess the impacts of climate change around inflection points (2000, 2009).

Modern machine learning and deep learning techniques require large amounts of data and can suffer from overfitting. PR, on the other hand, provides reliable estimates with relatively little data and allows for analyses that reflect the nature of the data. In particular, when analyzing trends around inflection points suggested by existing research, PR is more flexible than simple linear regression and easier to interpret than complex machine learning models."

Comments 5: an in-depth discussion of the results is needed.

Response 5: We analyzed PDO index and added the result of PDO and below sentences in discussion (page 13).

"We analyzed the Pacific Decadal Oscillation (PDO) index, which is closely related to climate change, in conjunction with sea surface temperatures (SST) around the Korean Peninsula. A cross-correlation analysis was conducted to determine the time lag between the PDO index and the SST. To eliminate seasonal variations, the monthly mean values were subtracted from both the PDO and SST data. The results showed no time lag in the Yellow Sea and the East Sea, whereas the South Sea exhibited an 8-month lead. However, as all correlation coefficients (R values) were below 0.05, the relationship between the PDO index and SST around the Korean Peninsula was found to be insignificant.

The aim of this study is not to simply predict the future, but to analyze the evolution of ocean temperatures based on historical data and compare the impacts of climate change over time. PR can quantitatively analyze trend changes over time based on a specific point in time, making it well suited to effectively assess the impacts of climate change around inflection points (2000, 2009).

Modern machine learning and deep learning techniques require large amounts of data and can suffer from overfitting. PR, on the other hand, provides reliable estimates with relatively little data and allows for analyses that reflect the nature of the data. In particular, when analyzing trends around inflection points suggested by existing research, PR is more flexible than simple linear regression and easier to interpret than complex machine learning models."

Reviewer 2 Report

Comments and Suggestions for Authors

Review of “Statistical approach for the imputation of long-term seawater data around the Korean Peninsula from 1966 to 2021” by Kwak et al.

 In this study, the author used subsurface temperature and salinity data from the 25 KODC  sites comprising 204 stations from 1966 to 2021. These data were collected around the coastal water of South Korea every 2 and 3 months. The author tried to generate the gap-free data with different algorithms and tried to show the trend in the temperature observation. Although it is a good topic, the author failed to give a clear take-home message to the reader from this manuscript. In this form, the paper may not be suitable to publish. My comments are given below

1. One of the basic drawbacks of this study is the asymmetric distribution of data on time. The authors themselves pointed out this in the conclusion that the data may not be from the same period as the survey period. This makes the analysis a bit erroneous, and the conclusions made from this may not be correct. The author should make the concurrent data to make this observations to be scientifically used for any conclusion.
2. Nowhere is it mentioned how much the actual data was and how much was covered by the imputation method to make the complete gap-free time series. The author should show the time series of these two from a few representative areas already chosen. Also, they should show the vertical profile plots for both these data sets raw and after filling the gap.
3. Figure 5 looks very odd. A does not show any trend, and b shows slight, but the distribution looks like some systematic pattern. The author should explain why this pattern. Appeared in figure
4. The author should show a similar to Fig 5 for all the depths defined in this study.
5. Conclusions need clearer statements in terms of data preparation, uniqueness, and scientific outcomes using this data.  

 

 

Comments on the Quality of English Language

English looks good to me. 

Author Response

Comments 1: One of the basic drawbacks of this study is the asymmetric distribution of data on time. The authors themselves pointed out this in the conclusion that the data may not be from the same period as the survey period. This makes the analysis a bit erroneous, and the conclusions made from this may not be correct. The author should make the concurrent data to make this observations to be scientifically used for any conclusion.

Response 1: We recognize these limitations in our study and have used piecewise regression to compensate for them.

Piecewise regression is a method that enables trend-robust analysis of data with time-varying distributions, as it automatically detects structural breaks in the data and applies a separate regression model for each bin to account for differences in the distribution of the data over time. This approach minimizes the impact of data bias in a particular period on the overall analysis and helps to reliably estimate trends even in the presence of temporal imbalances.

In addition, in this study, we aimed to mitigate the errors that may occur due to temporal inconsistency based on these advantages of piecewise regression, and to ensure the reliability of the analysis results. We will explain this more clearly in the paper.

Furthermore, we added below sentences in conclusion (page 14).

"Traditional regression models struggle to estimate trends in time series data with nonlinearity, heteroscedasticity, and structural breaks because they are sensitive to outliers and abrupt changes. Piecewise regression solves this problem by detecting points of change and applying a localized regression model to each segment, allowing for adaptive trend estimation. This method minimizes bias and improves robustness by adapting to different distributions, unlike a single regression model that assumes a uniform trend. The work of ref. [44-45] demonstrated the effectiveness of this method in detecting structural changes in economic and time series data. Applications in climate analysis, finance, and physiological signal processing have also confirmed its superiority in producing trend-robust estimates compared to traditional models."

Comments 2: Nowhere is it mentioned how much the actual data was and how much was covered by the imputation method to make the complete gap-free time series. The author should show the time series of these two from a few representative areas already chosen. Also, they should show the vertical profile plots for both these data sets raw and after filling the gap.

Response 2: We added time series and vertical profile of temperature and salinity in Supplementary Figure file (Figure s4~s5). There were too many figures, including over 1,800 time series plots and more than 57,000 vertical profiles, making it impossible to include all of them. Therefore, only a selected few were presented.

Comments 3: Figure 5 looks very odd. A does not show any trend, and b shows slight, but the distribution looks like some systematic pattern. The author should explain why this pattern. Appeared in figure

Response 3: The periodic temperature variations observed in the bottom layer of the East Sea are not as prominent in the mid and upper layers due to strong stratification, limited deep water exchange, and stable circulation patterns. The presence of a well-developed pycnocline and thermocline restricts vertical mixing, preventing temperature fluctuations from propagating upward (Kim et al., 2020). In contrast, the bottom layer experiences deep water renewal, where periodic inflows of deep, colder water contribute to significant temperature changes, while the mid and upper layers remain relatively stable due to weaker turbulent mixing (Talley et al., 2011). Additionally, the mid-layer is influenced more by ocean currents such as the Tsushima Warm Current and Liman Cold Current, which moderate temperature changes compared to the more isolated bottom layer (Isobe et al., 2002). These factors collectively contribute to the observed stability in the mid and upper layers, highlighting the need for further research on vertical mixing processes and their role in long-term ocean temperature trends.

References

• Kim, K.; Min, C.; Nam, S. Vertical mixing processes and water mass transformation in the East Sea. J. Oceanogr. 2020, 76, 123–138. DOI: 10.1007/s10872-020-00563-7.
• Talley, L. D.; Pickard, G. L.; Emery, W. J.; Swift, J. H. Descriptive Physical Oceanography: An Introduction, 6th ed.; Academic Press: Cambridge, MA, USA, 2011.
• Isobe, A.; Kuroda, K.; Saito, T. Seasonal variations in the Tsushima Warm Current paths in the Japan Sea. J. Phys. Oceanogr. 2002, 32, 1147–1156. DOI: 10.1175/1520-0485(2002)032<1147:SVITTW>2.0.CO;2.

Comments 4: The author should show a similar to Fig 5 for all the depths defined in this study.

Response 4: We added temperature trend for all the depths in Supplementary figure file (Figure s1~s3).

Comments 5: Conclusions need clearer statements in terms of data preparation, uniqueness, and scientific outcomes using this data.  

Response 5: For an in-depth discussion of this study, we first quantitatively compared the rate of sea temperature change by region and water layer and analyzed differences from previous studies. Second, we further examined the statistical significance of inflection points to explain their relationship with subsequent changes in the marine environment. Third, we assessed the impact of missing data imputation using the multiple imputation method on the analysis results and evaluated the appropriateness of the imputation technique.

Future research will focus on a more in-depth analysis of the relationship between marine environmental changes and climate change, utilizing advanced methodologies for more precise predictions. For example, we plan to explore the feasibility of comparative studies between the PR method used in this study and state-of-the-art techniques such as time-series deep learning and Bayesian change point detection, providing directions for future research.

We revised first paragraph in conclusion (page 14).

(before) "In this study, we analyzed NIFS Oceanographic observation data by dividing it into three layers, after which the imputation data were verified through statistical methods. Finally, the long-term imputation data were used to analyze the water temperature changes for each layer around Korean Peninsula."

(after) "In this study, we proposed a systematic approach to analyzing NIFS ocean observation data by disaggregating it into three layers. To ensure the reliability of the reconstructed data, we implemented and rigorously validated an estimation strategy using statistical methods. Using these estimated long-term datasets, we comprehensively analyzed water temperature changes in different layers surrounding the Korean Peninsula, providing new insights into long-term oceanographic trends."

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors improved their manuscript relatively regarding my minor comments. However, no reaction was given to my major comments. Unfortunately, the manuscript is still too poor in novelty and organization. In addition, the authors' replies to my fourth and fifth minor comments are not reasonable. Innovative trend analysis does not mean the use of machine learning.  I am sorry that I cannot suggest this manuscript for publication.  

Author Response

To the editors and reviewers of Water Journal,

We sincerely appreciate the time and effort of the reviewers in evaluating our manuscript. Their insightful feedback and constructive suggestions have contributed greatly to the quality and clarity of our work. In this letter, we carefully address the reviewers' concerns and outline the revisions to our revised manuscript.

Summary of key revisions

In response to the reviewers' valuable comments, we have made significant improvements to the manuscript. The key revisions include

Improved trend analysis model: We improved our trend analysis by incorporating a deep learning-based piecewise regression model that allows us to more flexibly and accurately identify points of change in water temperature trends.

Seasonality modeling: We integrated a seasonality model that uses Fourier transforms to better capture cyclical fluctuations in water temperature, allowing for a more comprehensive understanding of climate-induced variations.

Gaussian distribution for confidence estimation: We modeled water temperature using a Gaussian distribution, as opposed to traditional methods that rely solely on point estimates. This approach allows us to estimate confidence intervals, which improves the interpretability and confidence of our results.

Improved manuscript organization: We have reorganized key sections of the manuscript to improve logical flow and readability. We have also added supplementary figures and explanations to provide greater clarity on our methodology.

Responding to reviewer comments

Reviewer's concern: Lack of novelty

Comment: “The manuscript still lacks too much novelty. Innovative trend analysis does not imply the use of machine learning.”

Response: We sincerely appreciate this point and recognize the importance of demonstrating the novel contribution of our work. To clarify, our approach is not simply the application of machine learning, but rather the integration of several advanced methodologies to increase interpretability and robustness. To be more specific

Our work introduces deep learning-based piecewise regression models to effectively detect structural changes in long-term ocean temperature trends that are difficult to capture with traditional statistical methods.

We incorporated Gaussian distribution-based uncertainty estimation to provide confidence intervals and improve the reliability of climate change impact assessments.

The inclusion of Fourier transform-based seasonality modeling allows for a more holistic analysis of temperature trends by accurately representing cyclical variations.

To further emphasize these points, we have explicitly detailed these aspects in the revised manuscript (Sections 2 and 3) to highlight how our methodology advances existing research.

Reviewer's concern: organization of the manuscript

Comment: “The manuscript is not well organized.”

Response: We recognize the importance of clear and organized presentation. In the revised manuscript, we made several improvements to improve the organization of the manuscript:

Revised Introduction: the introduction has been improved to clearly describe research gaps and the unique contribution of our study.

Enhanced Methodology section: We now have a more organized and detailed description of our methodology, with figures added for clarity.

Improved Discussion section: We have reorganized the Discussion section to systematically compare our findings to previous studies and provide a clearer interpretation of our results.

We expect these modifications to make the manuscript more coherent, engaging, and easy to understand. We also sincerely hope that these modifications will clarify our responses and address reviewers' concerns appropriately.

We sincerely appreciate the reviewers' constructive feedback, which has enabled us to significantly improve the quality of the manuscript. We are confident that the revised manuscript presents a more robust and innovative approach to long-term ocean temperature analysis that meets the high standards of Water Journal. We sincerely hope that the revised manuscript satisfactorily addresses all concerns, and we look forward to its publication.

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

The authors improved the manuscript.

Author Response

We would like to thank you and the reviewers for your thoughtful and constructive feedback.