Prediction of Unpaved Road Conditions Using High-Resolution Optical Satellite Imagery and Machine Learning

Round 1

Reviewer 1 Report

Interesting paper and addressing the need of industry.  

Author Response

Thank you for the feedback. In light of other reviews and the need to miprove the English and readability we have substantially revised the paper:

The method and results for each of the three analysis methods are now combined, which should make the paper easier to follow. For example:

• The problem statement is integrated as part of the introduction: line 42 to 48.
• Related work has been condensed and clarified: line 54 to 83
• Figure 1 and associated text has been moved to 2.1: line 111 to 133
• Variation in width, method and results: line 167 to 246
• Variation in pixel intensity, method and results: line 247 to 362. Three tables have been removed from this section to make reporting of the results clearer
• Machine learning, method and results: line 363 to 465

The section on lower resolution imagery has been removed. This detracts from the main message of this article and can be revisited in a separate publication.

The writing style has been revised throughout the article to be more suited to a scientific journal.

The conclusions in section 5 have been adjusted, the first paragraph has been repositioned as the second paragraph and edited, lines 568 to 575.

Reviewer 2 Report

The manuscript titled “Prediction of unpaved road condition using high-resolution optical satellite imagery and machine learning” covers very interesting topic but the research could be done much better. The writing style is more suited for some professional journals, not scientific ones. It does not follow standard rules for writing research papers, making some parts hard to follow. There are some parts of the research that are very strange and make no sense, for example:

• researchers said that they used high-resolution satellite imagery from the Pleiades satellite, but they used only RGB bands for analysis, why not use other bands?

• line 283, there is no sense in adding the value of bands into one number, I see no connection between the sum of the band with the condition of the road. That's more guessing than scientific proof

• I see no connection between road width and the condition of the road, especially if you look at Figure 6 – 8. Does that mean that roads that are more wide are generally in bad condition than the narrower ones? The geometry of the road has nothing to do with its condition, even if you take into the account explanation from section 2.3.1

 

Some of the Tables and Figures (e.g. Table 1 and Figure 5) are not something that is common in scientific articles.

 

There is not even one satellite image in the manuscript as a Figure (except a screenshot of Google Earth). I understand that satellite imagery has its license and restrictions, but every satellite image provider will allow you to put a part of the image in your manuscript if you provide a source of the image.

 

Authors often refer to GIS for everything. GIS is too general to explain what you have done. You can not say “implemented within the GIS software..” or “manual measurements were conducted using GIS tools” or “on the satellite imagery, using GIS tools” or “identify and create samples, using GIS tools” ... etc. GIS and GIS tools are general terms, that are just like saying “We used mathematics to calculate something”. What algorithms, and tools did you use?

 

The authors said that they used a large number of ML methods, but none of them explained how they did it. Some of the ML methods are quite advanced and most of them work in a different way. It looks more like the Authors used ArcGIS or some similar GIS software, imported the data, used default settings and exported the results and then concluded that “classical” ML methods are better than CNN, which I am very sceptical.

 

Overall, the manuscript is very interesting but I think the research is done poorly.

 

 

Author Response

Thank you for the detailed and useful response. We have made substantial changes to the draft, which I hope will satisfy the issues that were identified. The detailed response is attached.

Author Response File: Author Response.pdf

Reviewer 3 Report

The abstract lacks a clear and concise statement of the problem being addressed. Begin by clearly defining the challenges faced by Local Roads Authorities (LRAs) in Africa regarding data collection on the condition of rural roads. Emphasize the limitations and difficulties LRAs encounter in maintaining these roads effectively.

The abstract should provide more context on the significance of the problem. Discuss the importance of maintaining rural roads for transportation, economic development, and social welfare in Africa. Include relevant statistics or references to highlight the impact of inadequate road maintenance on local communities.

Expand on the proposed novel framework by briefly explaining the key steps and methods employed. Specify the earth observation techniques used and their relevance to predicting road conditions. Mention the specific machine learning techniques applied for automated prediction and highlight their suitability for this task.

 Instead of merely stating the prediction accuracy values, provide more details on the experimental setup and the specific metrics used for evaluation. Describe the ground truth information used for comparison and explain how it was obtained. Include a discussion of the strengths and limitations of the experimental results.

Compare the proposed framework's performance with existing methods or alternative approaches used by LRAs to assess road conditions. Highlight the advantages and disadvantages of the proposed framework, particularly in terms of accuracy, cost-effectiveness, and scalability.

Provide a more detailed analysis of the trade-off between spatial resolution and prediction accuracy. Explain the reasoning behind selecting a 0.5m resolution for the trial area in Tanzania and discuss the potential benefits and drawbacks of using lower-resolution imagery. Consider the impact

 Moderate editing of English language required

Author Response

Thank you for the detailed and useful response. We have made substantial changes to the draft, which I hope will satisfy the issues that were identified. The detailed response is attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Upon careful examination of the second revision and comparing it to the original submission, I regret to inform you that the fundamental issue we raised regarding the methodology has not been adequately addressed. The methodology serves as the backbone of any research, and the lack of substantial changes in this crucial aspect raises concerns about the validity and reliability of the study's findings. The authors based their findings on the intensity of the pixels on the road, the sum of RGB bands and the width of the road as a factor to determent the quality of the road surface. As this is a Q1 journal, I am very reluctant to accept this paper for publication using this methodology. I could be wrong, maybe the Authors found something new, but I am not convinced this is possible.

While I acknowledge the improvements made to the manuscript, the unchanged methodology is a substantial barrier to its acceptance.

I want to emphasize that the decision to reject your paper is not a reflection of its potential contribution to the field and the manuscript is well-written. Still, I am not convinced that the methodology of the research is adequate. 

Author Response

Thank you again for the constructuive feedback. We recongise the shortcomings in the methodoology and have added to it to try and explain the approach taken, with added examples of related research that informed the methodology, especially for pixel variation. Our approach has been based on previous research we did with roads authorities in Africa, and our learning that a system where they can understand the process and visulaise the results is important for acceptance, hence the relatively simpe approach. 

Author Response File: Author Response.docx

Reviewer 3 Report

The authors have revised the paper as suggested by the reviewers, in which case the paper can be accepted.

 

Moderate editing of the English language required

Author Response

Thank you for your review and support, we appreciate your acceptance of the revised paper