Exploring the Effect of Meteorological Factors on Predicting Hourly Water Levels Based on CEEMDAN and LSTM

Round 1

Reviewer 1 Report

The amount of tidal energy is contingent upon fluctuations in ocean water levels. By properly forecasting these changes, tidal power plants may strategically plan and optimize the time of power output, hence maximizing energy harvesting efficiency. The temporal variability of water level fluctuations gives rise to the categorization of water level data as time-series data, which holds significant importance in the context of both short-term and long-term forecasting. The acquisition of real-time water level data is of paramount importance in the investigation of tidal power. The National Oceanic and Atmospheric Administration (NOAA) possesses a repository of real-time water level information, rendering the utilization of NOAA data highly advantageous for conducting research in this field. This study employs long short-term memory (LSTM) and its variations, namely stack long short-term memory (StackLSTM) and bi-directional long short-term memory (BiLSTM), to forecast water levels at five specific locations. These LSTM models are then compared to classical machine learning algorithms such as support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM). This study aims to examine the impact of wind speed (WS), wind direction (WD), wind gust (WG), air temperature (AT), and barometric pressure (Baro) on the hourly water level (WL). The findings indicate that, except for the La Jolla site, all climatic factors yielded the highest Correlation coefficient when employed as inputs. Burlington station has a coefficient of determination value of 0.721, whereas the Kahului station has a coefficient of determination value of 0.852. The concluding section of this article presents the integration of the complete ensemble empirical mode decomposition adaptive noise (CEEMDAN) method into different models, demonstrating notable enhancements in the accuracy of water level predictions at each respective site. Among the algorithms considered, the CEEMDAN-BiLSTM method demonstrated superior performance, achieving an average root mean square error (RMSE) of 0.0759 m h^(-1) for the prediction of three specific sites. The findings suggest that the utilization of the CEEMDAN algorithm in the context of deep learning yields a more consistent and reliable prediction performance when projecting water levels across various geographical areas.

 

The current study represents an excellent work in the field of water studies; the paper is well-written and organized. However, it needs some minor suggestions to be suitable for publication in this journal:

I.                 Authors should motivation using the support vector machine, the random forest, XGBoost and LightGBM models.

II.                I do think that the paper still needs to be motivated well. You may add all motivations at the end of the introduction.

III.              L222: “Hochreiter et al. (1997) [12]” should be “Hochreiter et al. [12]”.

IV.              L231: “follows[36]” should be “follows [36]”.

V.               L235: “(8” should be “(8)”.

VI.              Make Tables 3, 4 and 6 on one page.

VII.            Remove the space before Table 4.

VIII.           It’s required to add some future points for potential works, readers may need to know how to expand this work.

IX.              The main codes of the paper should be added into the appendix.

X.            The level of writing in English is generally good, but the paper needs to be improved through scientific writing in general and language specifically in order to be in the best possible way.

The level of writing in English is generally good, but the paper needs to be improved through scientific writing in general and language specifically in order to be in the best possible way.

Author Response

Hello, I have uploaded your comment. Please check it out!

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors used long short-term memory (LSTM) and its variants to predict water levels at five sites and compared with classical machine learning algorithms and finally introduced the complete ensemble empirical mode decomposition adaptive noise (CEEMDAN) algorithm into the above models to verify the compatibility and forecasting efficiency of the hybrid combinations. The CEEMDAN-BiLSTM algorithm performed the best among all the combinations chosen in this study. They infer that applying the CEEMDAN algorithm to deep learning has a more stable predictive performance for water level forecasting in different regions. Although the results are quite impressive, the manuscript is loaded with too many known matters and the English is below the level of publication. Many sentences are of length beyond the patience of the readers. I suggest a revision of the manuscript to be suitable for publication in the present journal.

1.       Lines 20-22 Rewrite the sentence which is wrongly formatted “his paper at-20 tempts to investigate the…  of the hourly.”

2.       Line 22-24 Rewrite the sentence “The results 22 show that the highest Correlation…

Kahului station (?2) = 24 0.852).”

3.       Line 22 I do not understand what the authors wish to express, what input is taken, correlation between which parameters and referring to which results, and what is the cause for relatively low correlations at two specific sites, the sentences in the abstract are confusing.

4.       Lines 34-37 The first sentence of the introduction section is not clearly understandable “In order to alleviate the consumption…”

5.       Line 48 What is SAR, define. Moreover a space is expected before the left brace “autoregressive model(SAR)”

6.       Line 49 The authors wrote a sentence mentioning many researchers but cited a single article “Many researchers use ANN in predicting water levels [5]”

7.       Lines 53-54 The term  water level prediction is repeatedly used in a single sentence “In water level prediction many researchers have tried to use deep learning 53 to accomplish the water level prediction task [7-9].”

8.       Line 55-56 Incomplete sentence “Deep learning includes many kinds of algorithms, among which the classical models 55 are Recurrent Neural Network (RNN) [10], Glavnoe Razvedivatelnoe Upravlenie (GRU) 56 [11], LSTM [12],”

9.       Line 61 Unexpected author etal “Assem et al.”

10.   Line 67-71 Rephrase the sentence “However, the relationship between the water…”

11.   Lines 79-81 Rephrase the sentence.

12.   Lines 81-85 Sentence having huge length, it's beyond the patience of the reader to read such sentences.

13.   Line 90 “are explored by exploring” Same word consecutively used.

14.   Section 2.3 and 2.4 Major parts of the paper are covered with known things, a recitation of existing forecasting models.

15.   Lines 227-229 The LSTM model description has already been given by many articles. For example, Reddybattula et al., 2022 (https://doi.org/10.3390/universe8110562) discusses in brief about the main architecture and function of LSTM. The authors are advised to refer to such articles for further discussion on the models rather than describing all details of the models.

1.       Lines 442-443 Rewrite “It can be concluded from Table 5-7…”

2.       Discussion: A significant portion of this section is literature review which must be transferred to the introduction. The authors should focus on discussing their results with respect to the contemporary results at same or different sites with similar algorithms, which is not reflected in the present manuscript.

3.       Conclusion: The computational complexity and time are not discussed for all the models as well as the proposed model in the last.

4.       An acknowledgement section is missing in this manuscript though the authors used a great resources database.

Although the results are quite impressive, the manuscript is loaded with too many known matters and the English is below the level of publication. Many sentences are of length beyond the patience of the readers.  A thorough English correction is expected before the next submission.

Author Response

Hello, I have uploaded your comment. Please check it out!

Author Response File: Author Response.pdf

Reviewer 3 Report

This study examined the “Exploring the effect of meteorological factors on predicting hourly water levels based on CEEMDAN and LSTM”. Given that fluctuations in water level are increasing due to climate change, this paper is timely and could offer new insights in the mentioned research area. The manuscript is generally well written and easy to understand. I suggest that the authors revise the manuscript incorporating the following comments and suggestions into an updated version.

·         What is the criterion for selecting the best model, "if the selection is mare a visual interpretation, then I recommend using some programming technique like Compromise programming or Global performance indicators or tailor skill score and rating matrix to rank the models”.

·         On what basis the authors selected the combination of input parameters.

·         Add legends in the study area map?

Needs minor changes

Author Response

Hello, I have uploaded your comment. Please check it out!

Author Response File: Author Response.pdf