Short-term Prediction of the Significant Wave Height and Average Wave Period based on VMD-TCN-LSTM Algorithm

Ji, Qiyan; Han, Lei; Jiang, Lifang; Zhang, Yuting; Xie, Minghong; Liu, Yu

doi:https://doi.org/10.5194/egusphere-2023-960

Preprints

https://doi.org/10.5194/egusphere-2023-960

Preprints

11 May 2023

| 11 May 2023

Short-term Prediction of the Significant Wave Height and Average Wave Period based on VMD-TCN-LSTM Algorithm

Qiyan Ji, Lei Han, Lifang Jiang, Yuting Zhang, Minghong Xie, and Yu Liu

Abstract. The present work proposes a prediction model of significant wave height (SWH) and average wave period (APD) based on variational mode decomposition (VMD), temporal convolutional networks (TCN), and long short-term memory (LSTM) networks. The wave sequence features were obtained using VMD technology based on the wave data from the National Data Buoy Center. Then the SWH and APD prediction models were established using TCN, LSTM, and Bayesian hyperparameter optimization. The VMD-TCN-LSTM model was compared with the VMD-LSTM (without TCN cells) and LSTM (without VMD and TCN cells) models. The VMD-TCN-LSTM model has significant superiority and shows robustness and generality in different buoy prediction experiments. In the 3-hour wave forecasts, VMD primarily improved the model performance, while the TCN had less influence. In the 12-, 24-, and 48-hour wave forecasts, both VMD and TCN improved the model performance. The contribution of the TCN to the improvement of the prediction result determination coefficient gradually increased as the forecasting length increased. In the 48-hour SWH forecasts, the VMD and TCN improved the determination coefficient by 132.5 % and 36.8 %, respectively. In the 48-hour APD forecasts, the VMD and TCN improved the determination coefficient by 119.7 % and 40.9 %, respectively.

Received: 10 May 2023 – Discussion started: 11 May 2023

Download & links

Preprint (PDF, 8779 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (8779 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

09 Nov 2023

Short-term prediction of the significant wave height and average wave period based on the variational mode decomposition–temporal convolutional network–long short-term memory (VMD–TCN–LSTM) algorithm

Qiyan Ji, Lei Han, Lifang Jiang, Yuting Zhang, Minghong Xie, and Yu Liu

Ocean Sci., 19, 1561–1578, https://doi.org/10.5194/os-19-1561-2023,https://doi.org/10.5194/os-19-1561-2023, 2023

Short summary

Qiyan Ji et al.

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2023-960', Brandon Bethel, 20 Jun 2023

Considering the need to enhance predictions of ocean wave parameters, Ji et al. considered the adoption of a joint VMD-TCN-LSTM algorithm to forecast significant wave height and wave period with minor computational expense and using direct buoy observations. The paper is of use to the community and while I have no technical objections, there are, however, a few issues the authors should consider. They are as follows:
1. The introduction of the model and its settings is very detailed but far too long as it consumes the first 13 pages and 6 pages of the article. These can be either reduced significantly or placed within a supplement to join the manuscript. This will allow for readers to focus on the results section which should be the manuscript's centrepiece.
2. On L32, the reference (P. et al., 2020) does not follow the format of the other references. Please revise.
3. There is a space missing on L47 before Zhao et al., 2019.
4. I don't understand why the SST or water temperature would direectly affect wave activity. Indeed WTMP and ATMP are negatively correlated with wave parameters in Figure 3. Please justify the usage of these variables and check if the forecast skill improves with their addition/subtraction in a new experiment. If forecast skill does not change with their removal, you'll have your answer on if its necessary to include it them in an already extensive list of predictands.
5. The range of APD in Table 2 and in Figure 2 seem to indicate the occurence of both wind waves and swell. Were wave forecasts done on both systems together, or individually? As swell is generally insensitive to wind information, using wind to predict swell may be ineffective.
6. There should be a colon (:) instead of a period (.) at the end of the sentence on L130. Same for L206.
7. There is a duplication of a comma on L372 after Table 7.

8. It might be useful in the conclusion to discuss the implications of the research on, for example, ocean wave energy projects that would be heavily dependent on wave height and period forecasts.

Citation: https://doi.org/10.5194/egusphere-2023-960-RC1
- AC1: 'Reply on RC1', Lei Han, 17 Jul 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-960/egusphere-2023-960-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-960-AC1
RC2:
'Comment on egusphere-2023-960', Anonymous Referee #2, 27 Jul 2023

Wave prediction is very important for fisheries, wave power generation and marine transportation. Numerical modelling (e.g., SWAN or WAVEWATCHIII model) is a common method for operational wave forecasting. Data-driven methods, such as neural network methods, are also very popular. This paper proposed a hybrid VMD-TCN-LSTM model to forecast significant wave height and wave period. The results show that the method is effective in predicting ocean waves. However, some issues need to be clarified.
1.The descriptions of the VMD, TCN and LSTM methods are very detailed. As these methods are widely used in other fields, the corresponding description can focus more on the improvement of these methods in this study.
2.In situ measurements from four buoys were used in this study. Does the hybrid VMD-TCN-LSTM wave prediction model use the same parameters measured at these buoy stations?
3.Line 104~105: The GST has a positive relation with SWH. Why not use this physical parameter to drive the model?
4.Line 211~212: The BO has two critical components. First, establish an agency model of the objective function through a regression model (e.g., Gaussian process regression) and subsequently use the acquisition function to decide where to sample next (Frazier, 2018). The word “build” and “use” should be revised as “establishing" and “using".
5. Line 628 :“To quantify the prediction model performance” should be revised as “To quantify the performance of the prediction model”.
6.Line 294 "in 3-hour SWH and APD forecasts" and Line 414 "In the 3-hour SWH and APD forecasts". The word “in” should be revised as “for”.
7. Line 375: Please add "at" before “Buoy 51004”.
8. Line 391: "the TCN cells is." Here, "is” should be “as”.

Citation: https://doi.org/10.5194/egusphere-2023-960-RC2
- AC2: 'Reply on RC2', Lei Han, 11 Sep 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-960/egusphere-2023-960-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-960-AC2
RC3:
'Comment on egusphere-2023-960', Anonymous Referee #3, 08 Aug 2023
The manuscript proposes a VMD-TCN-LSTM hybrid model to predict significant wave height and average wave period. The theoretical innovation of this article is not remarkable. However, before considering publishing in this top journal, this study lacks an in-depth comparative analysis of the data. The issues listed below should be addressed by the authors.
In the introduction, much more references related with wave period prediction are expected to cite for overall literature review.

It is recommended to set a threshold to distinguish whether the center frequency has changed significantly.

Have other wave parameters such as MWD or WSPD been decomposed by VMD for prediction? If they are decomposed, please add their K values, otherwise explain the parameter composition of input.

Are the hyper-parameter optimization results in Table 4 obtained from these search intervals? Are they obtained from search spaces containing several specific values? Much more explanation are suggested to provide.

What’s the maximum epochs set for each model during training?

Please check all bold metrics values. It seems that the MAE, RMSE, MAPE and R² of VMD-TCN-LSTM in SWH prediction at 51101 given in Table 6 are not the best.

What are the lags of each input variable chosen for prediction?

Compared with previous methods, the properties of the proposed method should be summarized to describe clear findings of this study.
Citation: https://doi.org/10.5194/egusphere-2023-960-RC3
- AC3: 'Reply on RC3', Lei Han, 11 Sep 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-960/egusphere-2023-960-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-960-AC3

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2023-960', Brandon Bethel, 20 Jun 2023

Considering the need to enhance predictions of ocean wave parameters, Ji et al. considered the adoption of a joint VMD-TCN-LSTM algorithm to forecast significant wave height and wave period with minor computational expense and using direct buoy observations. The paper is of use to the community and while I have no technical objections, there are, however, a few issues the authors should consider. They are as follows:
1. The introduction of the model and its settings is very detailed but far too long as it consumes the first 13 pages and 6 pages of the article. These can be either reduced significantly or placed within a supplement to join the manuscript. This will allow for readers to focus on the results section which should be the manuscript's centrepiece.
2. On L32, the reference (P. et al., 2020) does not follow the format of the other references. Please revise.
3. There is a space missing on L47 before Zhao et al., 2019.
4. I don't understand why the SST or water temperature would direectly affect wave activity. Indeed WTMP and ATMP are negatively correlated with wave parameters in Figure 3. Please justify the usage of these variables and check if the forecast skill improves with their addition/subtraction in a new experiment. If forecast skill does not change with their removal, you'll have your answer on if its necessary to include it them in an already extensive list of predictands.
5. The range of APD in Table 2 and in Figure 2 seem to indicate the occurence of both wind waves and swell. Were wave forecasts done on both systems together, or individually? As swell is generally insensitive to wind information, using wind to predict swell may be ineffective.
6. There should be a colon (:) instead of a period (.) at the end of the sentence on L130. Same for L206.
7. There is a duplication of a comma on L372 after Table 7.

8. It might be useful in the conclusion to discuss the implications of the research on, for example, ocean wave energy projects that would be heavily dependent on wave height and period forecasts.

Citation: https://doi.org/10.5194/egusphere-2023-960-RC1
- AC1: 'Reply on RC1', Lei Han, 17 Jul 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-960/egusphere-2023-960-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-960-AC1
RC2:
'Comment on egusphere-2023-960', Anonymous Referee #2, 27 Jul 2023

Wave prediction is very important for fisheries, wave power generation and marine transportation. Numerical modelling (e.g., SWAN or WAVEWATCHIII model) is a common method for operational wave forecasting. Data-driven methods, such as neural network methods, are also very popular. This paper proposed a hybrid VMD-TCN-LSTM model to forecast significant wave height and wave period. The results show that the method is effective in predicting ocean waves. However, some issues need to be clarified.
1.The descriptions of the VMD, TCN and LSTM methods are very detailed. As these methods are widely used in other fields, the corresponding description can focus more on the improvement of these methods in this study.
2.In situ measurements from four buoys were used in this study. Does the hybrid VMD-TCN-LSTM wave prediction model use the same parameters measured at these buoy stations?
3.Line 104~105: The GST has a positive relation with SWH. Why not use this physical parameter to drive the model?
4.Line 211~212: The BO has two critical components. First, establish an agency model of the objective function through a regression model (e.g., Gaussian process regression) and subsequently use the acquisition function to decide where to sample next (Frazier, 2018). The word “build” and “use” should be revised as “establishing" and “using".
5. Line 628 :“To quantify the prediction model performance” should be revised as “To quantify the performance of the prediction model”.
6.Line 294 "in 3-hour SWH and APD forecasts" and Line 414 "In the 3-hour SWH and APD forecasts". The word “in” should be revised as “for”.
7. Line 375: Please add "at" before “Buoy 51004”.
8. Line 391: "the TCN cells is." Here, "is” should be “as”.

Citation: https://doi.org/10.5194/egusphere-2023-960-RC2
- AC2: 'Reply on RC2', Lei Han, 11 Sep 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-960/egusphere-2023-960-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-960-AC2
RC3:
'Comment on egusphere-2023-960', Anonymous Referee #3, 08 Aug 2023
The manuscript proposes a VMD-TCN-LSTM hybrid model to predict significant wave height and average wave period. The theoretical innovation of this article is not remarkable. However, before considering publishing in this top journal, this study lacks an in-depth comparative analysis of the data. The issues listed below should be addressed by the authors.
In the introduction, much more references related with wave period prediction are expected to cite for overall literature review.

It is recommended to set a threshold to distinguish whether the center frequency has changed significantly.

Have other wave parameters such as MWD or WSPD been decomposed by VMD for prediction? If they are decomposed, please add their K values, otherwise explain the parameter composition of input.

Are the hyper-parameter optimization results in Table 4 obtained from these search intervals? Are they obtained from search spaces containing several specific values? Much more explanation are suggested to provide.

What’s the maximum epochs set for each model during training?

Please check all bold metrics values. It seems that the MAE, RMSE, MAPE and R² of VMD-TCN-LSTM in SWH prediction at 51101 given in Table 6 are not the best.

What are the lags of each input variable chosen for prediction?

Compared with previous methods, the properties of the proposed method should be summarized to describe clear findings of this study.
Citation: https://doi.org/10.5194/egusphere-2023-960-RC3
- AC3: 'Reply on RC3', Lei Han, 11 Sep 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-960/egusphere-2023-960-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-960-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Lei Han on behalf of the Authors (11 Sep 2023) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (13 Sep 2023) by Meric Srokosz

RR by Anonymous Referee #2 (13 Sep 2023)

RR by Anonymous Referee #3 (20 Sep 2023)

ED: Publish as is (20 Sep 2023) by Meric Srokosz

AR by Lei Han on behalf of the Authors (24 Sep 2023) Manuscript

Journal article(s) based on this preprint

09 Nov 2023

Qiyan Ji, Lei Han, Lifang Jiang, Yuting Zhang, Minghong Xie, and Yu Liu

Ocean Sci., 19, 1561–1578, https://doi.org/10.5194/os-19-1561-2023,https://doi.org/10.5194/os-19-1561-2023, 2023

Short summary

Qiyan Ji et al.

Viewed

Total article views: 410 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
262	120	28	410	9	7

HTML: 262
PDF: 120
XML: 28
Total: 410
BibTeX: 9
EndNote: 7

Views and downloads (calculated since 11 May 2023)

Month	HTML	PDF	XML	Total
May 2023	59	15	4	78
Jun 2023	41	13	2	56
Jul 2023	50	17	6	73
Aug 2023	43	14	3	60
Sep 2023	42	33	6	81
Oct 2023	23	25	6	54
Nov 2023	4	3	1	8

Cumulative views and downloads (calculated since 11 May 2023)

Month	HTML	PDF	XML	Total
May 2023	59	15	4	78
Jun 2023	41	13	2	56
Jul 2023	50	17	6	73
Aug 2023	43	14	3	60
Sep 2023	42	33	6	81
Oct 2023	23	25	6	54
Nov 2023	4	3	1	8

Viewed (geographical distribution)

Total article views: 404 (including HTML, PDF, and XML) Thereof 404 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 09 Nov 2023

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (8779 KB)
Metadata XML

Short summary

Accurate wave forecasts are essential to marine engineering safety. The research designs a model with combined signal decomposition and multiple neural network algorithms to predict wave parameters. The hybrid wave prediction model has good robustness and generalization ability. The contribution of the various algorithms to the model prediction skill was analyzed by the ablation experiments. This work provides a neoteric view of marine element forecasting based on artificial intelligence.


Total:	0
HTML:	0
PDF:	0
XML:	0