Multi-Scale Representation of Ocean Flow Fields Based on Feature Analysis

Round 1

Reviewer 1 Report

The paper presents an interesting idea for analyzing vector field data sets by sub-sampling fewer data points to accelerate the analysis and visualization. The authors have used several statistical measures such as auto-correlation and classification techniques to find patterns in the data and then use rough set and evidence theory to assign weights to data points. The result is compared with the equidistant thinning (regular sampling) technique to show the effectiveness of the proposed method. Overall, the paper presents an interesting solution but needs improvement in several aspects:

1. Comparing the method only with regular sampling is not enough. It is well known that regular sampling is not good at capturing features in data sets. It would be interesting to see how this technique compares to a random sampling approach, specifically, stratified random sampling.
2. The description and adaptation of the rough set theory are somewhat confusing and hard to follow. Section 3 needs more explanations and possibly descriptions of simple concepts about rough sets in detail so that the technique is easy to understand.
3. The results are promising, however, one key aspect is missing from the discussion which is the size of data sets and the performance study. The data set used seems quite small, whereas, the technique is developed for large vector fields. The authors need to provide results of the technique by applying to a large vector field and also report the computational performance.
4. There are some typos, which need to be fixed for future revision.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The topic of the manuscript is of interest and the presented methods seem to perform well on the one dataset.

Firstly, the language in places needs to be significantly improved. The structure of the sentences can be heavy leading to poorly expressed thoughts.

Secondly, the authors have not created a flow in the story - a few methods have been stringed together to achieve the final result. The questions are: Why these methods? What about comparative analysis with other existing approaches? References are tightly focused on the described two approaches, but the base of comparison should be broader.

The Results are also not presented in a cohesive flow to represent the motivation of the approach and emphasize its novelty. The conclusions are leaving the reader deflated as there is no solid basis for comparison and relating to different data sets.As a matter of fact, the method is tested on one region only. Do the results and performance apply for other regions as controlled by different conditions and weather patterns?

Some statements should be supported by references, e.g. in 2.1 section.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The problem being addressed in this paper is clearly outlined in the introduction and addressed in a systematic fashion throughout. The paper is well written, with only one clear correction that came to my attention

————(line 58)

The method in this paper, first, this performs an auto-correlation analysis and classification

Strike the comma and “this” following “first” in that line. 

————

The only suggestion for improvement would be to provide a bit more of an introduction and interpretation for non-specialists in the numerical methods used in the modeling and analysis. As written, and given my lack of specific expertise in the methods used, I did not find myself in a strong position to judge the scientific soundness of the paper (I need to leave that to the other reviewers). Given the brevity of the paper (contributing to its clarity) there is likely some additional room for additional backgrounding and interpretation for non-specialists - potentially increasing the accessibility of the content and likely increasing the impact of the paper. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

The authors present a method for feature retention of ocean flow fields in gridded data using spatial auto-correlation and different weighting schemes. I would say the study made an overall positive impression on me. I particularly like how a relatively simple, but effective method can rise after AI starts to dominate a niche. The writing quality is good, there are a few grammar errors, but nothing too severe. The illustrations are representative, they help to understand the problem better, although the could be more illustrating the phenomenon itself.

Please take extra care when revising the formulas and mathematical notations in Section 3. While I could grasp the essence of rough set theory applied to the results, I could not follow some of the equations and descriptions.

I find the lack of literature study one of the two major problems in this paper. The introduction contains some references to past methods for the same problem, but I could not find a proper introduction, nor a proper literature review. This way only professionals in the same niche can place the study properly judge its novelty or contribution, etc.. 

I think the introduction should contain more background information on the problem itself and the nature of the phenomenon maybe with some schematic illustrations. Furthermore, there should be a literature review outlining past methods with their efficacy in feature detection. Were there any similar approaches? The study goes from a '91 paper to AI so quickly that I can hardly believe there weren't any efforts in the meantime. If so, please elaborate on it.

My other main concern is the sample data used in the study. There is no information on the data used for the study. The problem with this is datasets not only differ in spatial resolution, but often in other, domain-specific aspects. For example, Landsat and SPOT instruments have different spectral resolutions and  number of sensors. Processed datasets are even more diverse, as the processing methods used have a cumulative effect on the accuracy, dimensionality, precision, etc. of the represented phenomena. Such diversity can significantly affect any scientific result. I suggest to describe the dataset and disclose its source, or even better, share the availability if it is open data. The same applies to the verification process. We have no knowledge about how the verification datasets were prepared nor about their origin.

l78 How? If it depends on locality, please say so. If there is a common rule of thumb weight matrix, please elaborate.

Figures 1,3,6,9 Tones are hardly distinguishable. I had to tilt my laptop screen to be able to differentiate between not significant and low-high areas.

Equation 2 is badly formatted.

l113 Sentences starting with a mathematical notation usually look bad.

l132 What is "defining interval" method?

Section 3 is very dense. Please use a math font consistently for annotations.

l173 What is ind?

l174 Notations are hard to interpret. Please use proper indexing. How can be Xi=U while Xi∩Xj=null? What is j if not x+-something? Where is k used?

--- MINOR ---
l42/l43 Superfluous adjectives, something like "[...]reduce the amount of data points by filtering out uninteresting data" would sound better.

l60 [...]and extract specified area. Please rephrase.

l65 Analysis of attributes, feature of flow fields, and weight allocation

l79 patterns -> pattern

l85 negative spatial negative

l135 Capitalize. Please check other titles as well.

Figure 4 If data are not interpolated between integer weights, please remove decimal values from the legend.

l221 Generalized?

l257 Capitalize figure and figures.

l258 Figure 7

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed most of my comments, added new comparison results and relevant literature works to the paper. The description of rough set theory has been improved. These modifications improved the quality of the paper and so this paper can now be accepted.

Author Response

Thank you for reviewing this paper, the latest revised paper has been uploaded

Reviewer 4 Report

I am really happy to see the results put into context. I think now less qualified readers will also know why the method presented in this paper is significant. With the new Introduction section I could better understand the novelty of this study. The mathematical formulas were also made more clear.

My only recommendation left is to make the legend of the figures consistent. I gladly observed that the coloring was changed in Figure 2 in order to provide more contrast, but in the other similar figures, the old legend pertained.

Author Response

Thank you for reviewing this paper. The legend has been modified，the latest revised paper has been uploaded.