Siamese-Derived Attention Dense Network for Seismic Impedance Inversion

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Review report:  mathematics-3173144

Title: Siamese-derived attention dense network for seismic impedance inversion

The author has proposed a ML based seismic inverse method to estimate the seismic wave impedance from seismic data. The method is novel and the research work has a great interest and application to the researchers and industry people.  A few of my comments are given below:

The introduction and method is well written and explained.  However, the method part is mixed up with the Experiments and Results section as a part methodology and equation are written in the result section. It is better to keep both sections separate. For example, section 3.4. Evaluation Metrics should be part of method.

From the manuscript, it looks that no other author is involved than the first author but in the it is being referred (e.g., we apply, our results) that the work is carried out by many authors.

Although the results of the DenseNet looks promising compared to other methods, but I have few concerns about the applications of the presented work. For example, the author mention - section 3.1. Datasets: We apply the proposed method to real seismic data. But all the presented examples are 1D well logs data. Without 2D/3D real seismic or even synthetic dataset application, I do not think this work should be accepted for publication. Therefore I like to recommend to include the more examples.

Conclusion is fine.

Comments on the Quality of English Language

Overall the quality is fine. Few points are:

From the manuscript, it looks that no other author is involved than the first author but in the it is being referred (e.g., we apply, our results) that the work is carried out by many authors.

 

The author has repeated this sentence a lot, e.g., There are many ups-and-downs... It is better to define the disagreements between true and predicted models in an other way.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors, I have carefully revised the manuscript. My main concerns are:

-The described condition may lead to potential overfitting, making it less effective when applied to new data, as it might have learned specific patterns rather than generalized from them. Please discuss further the potential for overfitting;

-The high dimensionality resulting from the large number of traces and samples significantly impacts computational efficiency. It is important to further discuss and compare the time required for the training process in each case;

-Incorporating features into the training of a CNN can provide implicit regularization, especially in cases involving lower impedances. Given this, it is important to further discuss the role of regularization. Since the initial step with only the DenseNet already introduces implicit regularization, the question arises: which step has the greater impact on regularization?

This is an interesting paper and the results obtained are of interest. After making the necessary changes, this work can be recommended for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author has incorporated all the comments and I do not have any further comment.