Improving Significant Wave Height Forecasts Using a Joint Empirical Mode Decomposition–Long Short-Term Memory Network

Round 1

Reviewer 1 Report

The paper presents an approach for significant wave height forecasting that couples Empirical Mode Decomposition (EMD) and Long Short-Term Memory Networks (LSTM). The topic is highly interesting for the sector and I believe the paper has the potential to impact in the offshore engineering community. However, the added value of the present study to the already extensive literature is unclear. To that end the literature review must be significantly improved, the contribution of the present study clearly stated and results selected in order to support that contribution.

Therefore, the present paper cannot be accepted for its publication in its current form and should address the comments listed in the attached document before its publication.

Comments for author File: Comments.pdf

Author Response

Thank the reviewer a lot for your time spent on our manuscript. Your encouragements are very appreciated. Your comments have been carefully addressed one by one below. The details to see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This paper assesses coupled EMD-LSTM method for achieving cheaper and faster significant wave height forecasts (HSIG) at various time horizons. The paper use of EMD-LSTM method for breaking the HSIG signal of 2018 observations into a set of intrinsic mode functions for training the model and later using it to forecast 2019 HSIG is impressive. Additionally, the paper is written very concisely, which is great as well. However, I have some major concern that needs to be included in the paper for proving this coupled method’s novelty in making accurate HSIG forecasts. First, the HSIG are a functions of local wind forcings, mean sea level pressure and offshore wind generated waves (Swells) therefore it is difficult for coastal modeling community to see the application of this method for flood forecasting without taking into account the driving forcing of HSIG (local winds, pressure and swells). Secondly, the paper only talks the analysis over a very small time period which also is questionable to confidently accept the accuracy of the coupled EMD-LSTM method, therefore a much extensive time period and a large number of wave buoys should be used for analysis. Overall, considering the interesting methods used in this study for HSIG forecasting, I recommend major revisions to the manuscript.

Author Response

Thank the reviewer a lot for your time spent on our manuscript. Your encouragements are very appreciated. Your comments have been carefully addressed one by one below. The details to see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The article is well written and interesting, at this point in time, given the more focused univariate prediction the complexity of this work is not has striking as in other similar contributions at JMSE and similar journals. However, it does have interest given the practical case application and quite well presented work in a simpler but logic manner that helps other researchers to look for practical applications in field. So the work is more valuable for its potential application to different engineering cases than for its novel/high quality findings. Nevertheless the novelty and main contribution of this article should be made a bit clearer in the introduction and eventually in the abstract, if there is space for it. Some corrections are given below for further improvement. Final section should be named conclusions and perhaps a better framing of these findings into the literature could be given at this section as well.

L21 demonstrates

L31 for the benefit of a large portion of the readers of JMSE I strongly suggest you to include in these examples a mention to the importance of wave heights for offshore engineering design, e.g. in scour protections, offshore wind foundations, breakwaters and many others, please consider to include these two works as well, which provide wave heights modelling and discussion on their importance for marine structures design, DOI: 10.1177/0309524X18777323 and these example and others you may find suitable 10.1016/j.coastaleng.2020.103671

Up to line 44 - There many other numerous models to discuss here, for example, the rather interesting use of copulas approaches, conditional models etc... many of these are mandatory in several design norms. Please elaborate 1 or 2 sentences on this topic and address to the following review DOI: 10.1680/jmaen.2019.20

L64 - which other wind forecast models? Please be more specific 

L80 - It seems that the study is contributing to the literature by covering periods of forecasting of 48 to 72 hours. Is this the only novelty of the work? Perhaps the novelty and contribution of the paper should be made more clear at this stage. Note that many other models allow for 48 and 72 hours predictions with a fair degree of accuracy. Please consider to explain a bit better what is the novelty of this work.

L84 - were

Figure 1 - Can you give the buoys coordinates in text? Can this image's resolution be improved?

L106 - either explain what the Hadamard product is or give at least a reference for it.

COmment - somewhere in section 2 a figure with the wave's timeseries and a table with the common descriptive statistics would be interesting for a better visualization of the data at hands. Although there is indeed a time-series representation in Figure 4.

L167 - The authors say "LSTM’s forecast errors accumulate precipitously." Please explain how this has been concluded.

L167 - 169: A zoom in Figure 4 for the places where the larger wave heights are less well predicted would be of value for a better graphical perception of the method's accuracy.

L222-224 - was the lower RMSE only valid in your case or is it ever like that? Perhaps you should limit the sentence saying that EMD-LSTM should always be applied instead of LSTM alone. As this may not always be the case, or is it?

Figure 6 - it would be better to include units in the yy axis.

Comment - Section 4 should be named conclusions. Additionally, if there are any other works dealing with a similar model for the same application it would be nice to have a comparison/framing of such conclusions into the literature.

Author Response

Thank the reviewer a lot for your time spent on our manuscript. Your encouragements are very appreciated. Your comments have been carefully addressed one by one below. The details to see the attachment.

Round 2

Reviewer 3 Report

The manuscript has improved and my concerns were attended.

It is now good to be published.

Congratulations.