A Robust Algorithm for Photon Denoising and Bathymetric Estimation Based on ICESat-2 Data

Round 1

Reviewer 1 Report

This manuscript performed study on photon denoising and water depth estimation in underwater topography mapping based on the ICESat-2 data. The manuscript provides valuable insights into the challenges of underwater topography mapping using Lidar altimeters. Thus, I would like to recommend the publication of this manuscript. However, due to the comments listed below, the manuscript should be carefully revised before publishing.

1. The written English should be checked by a native speaker through the whole draft to avoid or exclude any unacademic expressions.

2. The abstract needs to be rewritten, with research content and conclusions described separately.

3. More related works should be introduced, and a summation about the related work should be added.

4. In the method section, the DBSCAN in section 3.1 and the Two-dimensional window filter in section 3.2 should be integrated into one section named two-step method.

5. The innovation of this manuscript is further sorted out and highlighted in the abstract, introduction, methods and conclusions.

6. In the accuracy assessment section, it necessitates knowing the elevation datum utilized for both the ICESat-2 ATL03 product and the in-situ data obtained from the CZMIL system. Additionally, it would be beneficial to know whether the author performed an elevation datum conversion and, if so, how this conversion was conducted. These details would help to provide a more comprehensive understanding of the study's methods and results.

7. As the ICESat-2 ATL03 data and CZMIL system in-situ data were collected on different dates, it is pertinent to determine whether the authors applied tidal correction to the bathymetric data, particularly given the significant time gap between the two sets of measurements. The article would benefit from an explanation of the methods used for tidal correction, providing readers with a more thorough understanding of the study's approach and findings.

8. The reason why the two-step method denoising is better than the single method denoising should be further discussed and analyzed.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

1. I only saw the research progress of DBSCAN algorithm in the introduction of this article. Compared with other methods (such as the denoising algorithm based on raster image processing, and OPTICS (https://ieeexplore.ieee.org/document/9126850 ），What are the advantages of this algorithm in the paper?

 

2. Is the application scope of this method limited to the near shore. Can it be applied to larger areas and other complex terrain? 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

ICESat-2 is a great resource for those working on shallow water bathymetry but, as the authors discuss, the noise level of the data very high and this limits the ability to process large amounts of this data. I suspect most people making use of ICESat-2 data are currently doing a moderate level of filtering and then manually cleaning the data. What is really required is a fully automated method. The method described in this paper seems to require at least two sets of parameters to be set by visual interpretation (?). So it is not clear whether it offers much advantage over manual cleaning, because adjusting those parameters to optimise the result could be a frustrating process. Manually cleaning with an efficient system, i.e. literally erasing bad points, could be almost as fast and give a better result. That the method is not fully automated must be made clear in the abstract as this is the main limitation. Ideally there would be a comparison to manual cleaning, because that is the test of interest. There is some mention of "manual cleaning" in the results but I don't understand what that refers to as it isn't mentioned in the methods. What I am referring to is literally taking Fig. 2b (for e.g.) and erasing the noise points in a drawing package (like photoshop) by visual interpretation. With the right set-up this can be very fast and I believe it gives the best results, until it's demonstrated otherwise. This is the question of practical interest.

The paper also needs substantial improvement in the writing and important information is missing. The introduction is good but later some of the writing later is incomprehensible. The accuracy metrics (F1, etc) aren't defined so can't be interpreted (I'm not even sure if a big number or a small number is better). There is no description of how the sea surface line is determined, but this must be as important as the bottom surface.

Below are some specific comments, but there are other typos and parts that are unclear. Because I couldn't understand the accuracy metrics or what was done I didn't review past section 4.5.

Specific comments:

Line 54 - can you add a few words just to give a basic idea of what the "after-pulse effects of lidar" are?

Line 120 - typo, sentence doesn't make sense.

Line 125 - improve the bathymetric accuracy relative to what?

Line 155 - "based"

Line 156 "radius eps" what does it mean ? if eps is a variable name make that clear.

Line 157 This description of DBSCAN is incomprehensible. "Search area is charged" typo? "circle's radius eps as the center", a radius can't be a center? falls in the neighborhood" means what? "any point of the boundary point"? Suggest to refer to the figure before talking about points A, B, N etc. What happens to the core points, boundary points and noise points? The description does not actually say how the denoising is done. What is rejected, what is the output?

Line 174 "the tape to scan"?

Section 4.1 - how is this classification done? How is the "dividing line" determined? It actually looks a little bit too low? The establishment of the sea surface line is as important as the sea bottom points so must be described. Is it really always completely flat, I'm not sure that it would be.

Figs. 6 and 7 and others - suggest not to vary the stretch of the horizontal or vertical scales - present all figures in the same aspect ratio otherwise they are hard to compare, and it's not even clear when the same data is presented.

Line 272 "It can is clear"? typo.

Line 273 - so if I understand correctly the method requires human interaction and visual interpretation for every data set? It should be made clear in the abstract that this is not a fully automated method.

Line 307 - the setting of the second filter parameters is a second human intervention and visual interpretation step? The issue is now the method has two points of parameter setting, so it becomes cumbersome in practice. You may be going back to the first step to change parameters to try again the second step with other parameters. In the context of this paper good results are shown with parameters quoted but what is not clear, is the effort required to determine these optimal parameters. It's not easy to tell if the method offers any less effort than manual cleaning of data, especially as I think manual cleaning will preserve more good points in sparse areas.

Table 2 - hard to interpret due to poor layout. F1, accuracy, precision and recall are not defined so I do not know what these numbers represent. The text says manual denoising is compared but I don't see that in the table?

Line 344 - I don't understand what the manual denoising is. It is not described in the methods. From Table 2 I have difficulty to understand what the different treatments are. This should be laid out clearly in a bullet-point list in the methods.

Author Response

Please see the attachment.

Author Response File: Author Response.docx