Business Intelligence through Machine Learning from Satellite Remote Sensing Data

Round 1

Reviewer 1 Report

The paper provides a comprehensive overview of the available literature and solutions related to Remote Sensing (RS). It highlights various successful applications in fields such as agriculture, tourism, and energy production. The proposed solutions aim to generate actionable information using freely available satellite imagery. Additionally, the results can be utilized to create information products, especially for SMEs that may lack the knowledge and resources required to access relevant information.

In the study, several issues have been well identified. In addition to the mentioned problems, such as the challenge of accurately assessing model performance (due to the difficulty in gathering relevant data and constructing a high-quality large dataset) and the influence of hyperparameters and the size of the fishnet, the incorporation of energy-efficient lighting, particularly those designed to reduce light pollution, could also have an impact on the final results. Please provide comments on this matter.

The results are presented in a very clear and detailed manner. The analysis process, along with the Machine Learning Pipeline Diagram, is well-documented. While the experimentation and evaluation are extensively covered, it would be beneficial to present them in tabular form in Section 5.

In the final section, there are a few missing links.

-    Need to spellcheck. For example, '... for satellite data acquisition and 662 preprocessing techniques.'

-         Also, please try to clarify and divide some sentences to make them more comprehensible.  For example, '...  While remote sensing satellite data have found application in a plethora of different 233 Business Intelligence fields we here focus on the selected ones highly correlated with the 234 Greek economy such as the tourism and agriculture industries paying special attention to 235 applications related to Nighttime lights and urbanization which are good proxies of the 236 economy will also be presented.'

Author Response

The paper provides a comprehensive overview of the available literature and solutions related to Remote Sensing (RS). It highlights various successful applications in fields such as agriculture, tourism, and energy production. The proposed solutions aim to generate actionable information using freely available satellite imagery. Additionally, the results can be utilized to create information products, especially for SMEs that may lack the knowledge and resources required to access relevant information.

In the study, several issues have been well identified. In addition to the mentioned problems, such as the challenge of accurately assessing model performance (due to the difficulty in gathering relevant data and constructing a high-quality large dataset) and the influence of hyperparameters and the size of the fishnet, the incorporation of energy-efficient lighting, particularly those designed to reduce light pollution, could also have an impact on the final results. Please provide comments on this matter.

The reviewer is right here. We totally missed these light pollution related issues. For this, we have added paragraphs 7 and 8 in the section "Synopsis and Future Work".

The results are presented in a very clear and detailed manner. The analysis process, along with the Machine Learning Pipeline Diagram, is well-documented. While the experimentation and evaluation are extensively covered, it would be beneficial to present them in tabular form in Section 5.

In the final section, there are a few missing links.

We have fixed the missing links.

Need to spellcheck. For example, '... for satellite data acquisition and 662 preprocessing techniques.'

We carefully revised the whole text and made several corrections.

-         Also, please try to clarify and divide some sentences to make them more comprehensible.  For example, '...  While remote sensing satellite data have found application in a plethora of different 233 Business Intelligence fields we here focus on the selected ones highly correlated with the 234 Greek economy such as the tourism and agriculture industries paying special attention to 235 applications related to Nighttime lights and urbanization which are good proxies of the 236 economy will also be presented.'

The suggested sentences, and several others, have been clarified.

Reviewer 2 Report

The paper is very good: publication is recommended.

Yet, the authors are encouraged to review the English language and add a discussion subsection where they clearly highlight the innovation in the proposed approach.

A thorough revision of the English language is recommended

Author Response

The paper is very good: publication is recommended.

Yet, the authors are encouraged to review the English language and add a discussion subsection where they clearly highlight the innovation in the proposed approach.

A thorough revision of the English language is recommended

The paper has been extensively revised to improve the use of the English language. A commercial grammar checker has been used and a colleague whose mother tongue is  English further proofread it.

Two paragraphs (9th and 10th) concerning the innovation of the approach have been added to the section "Introduction"

Reviewer 3 Report

This is an interesting study in which the authors apply ML algorithms and satellite data to applications in urban planning. In order to analyze and interpret satellite data, the authors have created an open-source software system that makes use of Google Earth Engine (GEE) and a number of spectral indices. By offering useful spatial data, this program attempts to support business intelligence and policymaking. One of the software's unique capabilities is its capacity to identify vacant buildings in a certain location, providing urban planners and decision-makers with a useful tool. I have some suggestions:

1. I am a bit confused about the two ways suggested by the authors in tackling the lack of ground truth data: The study lacks ground truth data for their case study area, which is crucial for training and evaluating machine learning models. Although they have suggested two ways to overcome this, I believe they should elaborate on how their suggested methods help to overcome this , and whether they are really effective. 

2. Thus, the paper could benefit from a more detailed explanation of the methodologies used, especially how they combat the lack of ground truth data.

3. Instead of the classical machine learning methods, did the authors try other ML methods or DL, which might affect the robustness of the results.

4. There i a lack of comparative analysis. The study does not compare its methods and results with existing solutions in a detailed manner, making it hard to gauge its relative effectiveness.

5. Finally, when is the opensource software archived? i cannot seem to find the URL in the manuscrupt.

6. There are some sentences like: "It is worth to note that all software and data considered in this study are publicly 1136 available at ???." which should be corrected.

7. in abstract: change "offer" to "offered". there are many other grammatical errors. do check through carefully.

8. I personally think that the first part of the manuscript is too heavy on the literature review. some parts can be summarized.

--

Author Response

This is an interesting study in which the authors apply ML algorithms and satellite data to applications in urban planning. In order to analyze and interpret satellite data, the authors have created an open-source software system that makes use of Google Earth Engine (GEE) and a number of spectral indices. By offering useful spatial data, this program attempts to support business intelligence and policymaking. One of the software's unique capabilities is its capacity to identify vacant buildings in a certain location, providing urban planners and decision-makers with a useful tool. I have some suggestions:

1.I am a bit confused about the two ways suggested by the authors in tackling the lack of ground truth data: The study lacks ground truth data for their case study area, which is crucial for training and evaluating machine learning models. Although they have suggested two ways to overcome this, I believe they should elaborate on how their suggested methods help to overcome this , and whether they are really effective.

Unfortunately, we were unable to locate any ground truth data. Please note our comments in the paper and in particular the first two paragraphs in the section "Evaluation of our Methodology".

2. Thus, the paper could benefit from a more detailed explanation of the methodologies used, especially how they combat the lack of ground truth data.

Perhaps the reviewer missed our discussion on the lack of ground truth data, and the way we deal with it in paragraphs 5-7 in the section "Data Collection".

3. Instead of the classical machine learning methods, did the authors try other ML methods or DL, which might affect the robustness of the results.

No, we have not considered other ML methods. We plan to work in this direction in the near future.

4. There i a lack of comparative analysis. The study does not compare its methods and results with existing solutions in a detailed manner, making it hard to gauge its relative effectiveness.

We do not provide any quantitative comparison for two reasons. (1) Our data are closely related to the location of the study and to the best of our knowledge, no related results exist. (2) Programming/simulation tools mentioned in other similar studies are not available to us.

5. Finally, when is the opensource software archived? i cannot seem to find the URL in the manuscrupt.

Software, data, and bibliography are now available at https://github.com/ckyriakos/thesis_front_end 

6. There are some sentences like: "It is worth to note that all software and data considered in this study are publicly 1136 available at ???." which should be corrected.

We now provide the URL (see above) to the data in the sentence mentioned. SImilar typos have been corrected also.

7. in abstract: change "offer" to "offered". there are many other grammatical errors. do check through carefully.

Extensive language revision fixed our grammatical errors for which we apologize.

8. I personally think that the first part of the manuscript is too heavy on the literature review. some parts can be summarized.

We are more than willing to summarise our literature review, at least to some extent, if the reviewer insists. This might move the "center of gravity" of the paper though, and we are not sure that this will improve its scientific merit.

Round 2

Reviewer 3 Report

I am afraid that the revised manuscript is not ready for publication yet.

1. "Perhaps the reviewer missed our discussion on the lack of ground truth data, and the way we deal with it in paragraphs 5-7 in the section "Data Collection"--->I have said it very clearly that I have read that part and I am not convinced. Better ways should have been considered.

2. No, we have not considered other ML methods. We plan to work in this direction in the near future---> Then this manuscript is really not ready for publication yet. 

3. There is a lack of comparative analysis and my previous opinion about this remains. It is unacceptable for the authors to just brush away my previous comments like this.

4. "We are more than willing to summarise our literature review, at least to some extent, if the reviewer insists. This might move the "center of gravity" of the paper though, and we are not sure that this will improve its scientific merit."---> have you even tried? does it worsen the situation? have you considered asking another neutral party to read/comment? I am not talking about improving its scientific merits per se. I am saying how it can further enhance the readability.

5. The research value/impact of this work is limited and the work in its current state is not ready for publication yet. The methodology needs to be further improved.

-

Author Response

We believe we understand the concerns of the reviewer and we have tried to accommodate them, as follows

1) "Perhaps the reviewer missed our discussion on the lack of ground truth data, and the way we deal with it in paragraphs 5-7 in the section "Data Collection"--->I have said it very clearly that I have read that part and I am not convinced. Better ways should have been considered.

A ground truth analysis for two algorithms on the only available data to us has been added to subsection 5.3. "Ground Truth Comparisons and Neural Network Implementation and Testing."

2. No, we have not considered other ML methods. We plan to work in this direction in the near future---> Then this manuscript is really not ready for publication yet. 

Other ML methods are considered in subsection 5.3. 

3. There is a lack of comparative analysis and my previous opinion about this remains. It is unacceptable for the authors to just brush away my previous comments like this.

We are sorry but our intentions were not to brush away but to improve our paper. We hope that the content of our answer in the above two items will satisfy the reviewer for this.

4. "We are more than willing to summarise our literature review, at least to some extent, if the reviewer insists. This might move the "center of gravity" of the paper though, and we are not sure that this will improve its scientific merit."---> have you even tried? does it worsen the situation? have you considered asking another neutral party to read/comment? I am not talking about improving its scientific merits per se. I am saying how it can further enhance the readability.

We made an effort to make the section a bit shorter and more readable. Two of my colleagues also went over it and made suggestions. They state that it will be worth going over for several readers of FI.

Several other changes (and corrections) have been made. We could provide the PDF that presents all the changes we have made if needed.

Round 3

Reviewer 3 Report

the revised manuscript can now be accepted.

-