Deep Learning-Based Rolling Forecasting of Dissolved Oxygen in Shandong Peninsula Coastal Waters

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Editor

Water

To enhance the predictive capability of DO in marine ranching areas, this study evaluates multiple forecasting approaches, including AutoARIMA, XGBoost, BlockRNN-LSTM, BlockRNN-GRU, TCN, Transformer, and an ensemble model that integrates these methods. Using hourly DO observations from coastal buoys, we performed multi-step rolling forecasts and system-19 atically assessed model performance across multiple evaluation metrics (MAPE, RMSE, and R²), complemented by residual and error distribution analyses. The results show that the ensemble model, based on deep learning techniques, consistently outperforms individual models, achieving higher forecast robustness and more effective variance control, with MAPE values maintained below 4% across all three buoys. The subject addressed falls within the scope of water. However, some revisions have been found:

• In the introduction, please also review the pioneering studies in the field of this study.
• Please add flowchart to show steps of performing this study.
• Please list the parameters values that presented in Table 2. You can present them in one Table.
• Which machine learning or deep learning algorithm was used for SHAP analysis? Using different algorithms may yield different results. Please state it for readers.
• In the section 4, the results is for which training or testing period? Please determine it.
• The results section is short. Please provide the results of training and testing periods separately.
• It is better, to time series plots or scattering plot of other algorithms are demonstrated in the section 4.
• The conclusions is too short. Please add more details about your methodology and quantities results.
• If possible, please also provide forecast results for future time steps.
• In the figures, use subcaptions for each subfigure.
• I recommend adding a graphical abstract to this study.

Good luck,

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In the study, the authors developed time series models to predict Dissolved Oxygen levels. The study holds significant value due to its potential to function as an early warning system. Overall, the work is well-designed; however, there are a considerable number of points that require clarification. My comments and questions are listed below:

1. What is the innovation that distinguishes this study in the literature? The claim in the introduction that previous studies focused on short-term predictions appears rather weak. I believe the study lacks a strong novelty that sets it apart from general machine learning applications. The authors need to clearly state the aspects of the work that present an original contribution.
2. The introduction should mention the existence of similar early warning systems in the literature.
3. Abbreviations such as ARIMA (line 46) and SHAP (line 180) should be written out in full upon their first mention.
4. The main objective section should not include any findings or results. This section is meant to present the study's aims and provide a very brief description of the models used. The main objective section should be rewritten accordingly.
5. A “Unit” column should be added to Table 1.
6. What does “Reinhardt” refer to in line 121?
7. Equation 3 is unnecessary. Stating the unit used for DO within the text is sufficient. Including this equation unnecessarily extends the manuscript.
8. The equation number mentioned in line 157 appears to be incorrect.
9. To which model do the SHAP results pertain (XGBoost or DL)?
10. It is unclear whether Pearson or Spearman correlation coefficients were used for the correlation matrices.
11. Up to which lag value were delays tested and used?
12. Figure numbers on page 9 are incorrect.
13. It is recommended that the key hyperparameters listed in Table 2 be explained in the supplementary file.
14. Detailed descriptions of the machine learning models mentioned in the manuscript should be provided in the supplementary material. Models such as BlockRNN-LSTM, BlockRNN-GRU, and TCN are not widely known and require clarification.
15. The input combinations used in the study are not clearly presented. A table should be provided listing which lagged values of independent variables were included in the models.
16. How were the machine learning model hyperparameters optimized?
17. Were past DO values used as input? If so, which lag times were considered and how were they determined?
18. Are the predicted values or errors of the models statistically different? This should be tested and reported.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

this is an extremely well written article, there is a clear message, and an interesting set of findings.  some small areas for improvement I would suggest include, a little more detail about the computation, language used and the specific software adopted.  I think also some of the choices made in the specific modelling approaches would be beneficial. it is necessary to be more specific in detail about the ensemble model, and the approaches taken. Ensemble models are often used to allow the evaluation of uncertainty, and there is no discussion of uncertainty.  re the interesting work on the early warning system, I would expect some additional evaluation of the performance of the early warning system , eg false positives and negatives.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Editor

Water

This manuscript is previously carefully evaluated. The current version of the manuscript is being suggested to be published in this admir journal.

With kind regards,

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have meticulously made all the requested corrections.