Uncertainty-Aware Interpretable Deep Learning for Slum Mapping and Monitoring
, Harry Gibson 1, Yunzhe Liu 1, Moloud Abdar 2, Marius Posa 3, Gholamreza Salimi-Khorshidi 1, Abdelaali Hassaine 1, Yutong Cai 1, Kazem Rahimi 1 and Mohammad Mamouei 1,*
Round 1
Reviewer 1 Report
Dear authors,
this paper presents mapping slums using satellite imagery. There was not single mention which date and why you used this imagery.
Line 214: "We used modified Copernicus Sentinels data [2015-2020] processed by Descartes Labs." - why and how it is modified?
Line 219 - Section ??
Figure 1 should be described before in text - figure title should be brief. Apply on all figures (2,3,4, 5, 6...)
Line 237 - AUPRC scores reported - what is the accuracy of this reports?
Figure 2 - legend is not readable at all.
Lines 249 - 252 This statement is yours or you took it from somewhere? I would like to know how you decided for this parameters, and how you are so sure that improves generalization?
From line 260 - 261 - what is MCD?
Procedure shown in Figure 3 should be in detail explained before figure, this is key part of manuscript.
What information can be obtained from SWIR1 and SWIR2 - seems like to strong results..
Whole paper seems like you are not clear enough what and why are you trying to do. I suggest you review your manuscript according to all reviewers comments.
Author Response
We would like to thank the reviewer for the valuable comments and suggestions. Your comments and suggestions in this round helped us to improve both the scientific quality and the readability of our manuscript.
The new version of the manuscript has been revised according to the given comments, and the major updates include: a) improvement of the presentation style of our main research contribution, b) enhancement of other related issues indicated by the reviewers, and finally c) general corrections of the English language applied throughout the manuscript.
Please find below our point-by-point responses.
With kindest regards,
The authors
Comment: This paper presents mapping slums using satellite imagery. There was not single mention which date and why you used this imagery. Line 214: "We used modified Copernicus Sentinels data [2015-2020] processed by Descartes Labs." - why and how it is modified?
Our response: We have added additional details as to exactly what imagery was used and why. We have included signposting to the supplementary material for details of how and why the images were preprocessed.
Comment: Line 219 - Section ??
Our response: Correct signposting to the supplementary material has now been included.
Comment: Figure 1 should be described before in text - figure title should be brief. Apply on all figures (2,3,4, 5, 6...)
Our response: We have ensured that all figures are described in the text before they are shown and that they all have succinct titles and captions.
Comment: Line 237 - AUPRC scores reported - what is the accuracy of this reports?
Our response: We did not use accuracy as a metric in this paper for reasons outlined in Section 3.2.7.
Comment: Figure 2 - legend is not readable at all.
Our response: We have removed the legend in Figure 2. The information previously contained in the legend is outlined in the caption of the figure.
Comment: Lines 249 - 252 This statement is yours or you took it from somewhere? I would like to know how you decided for this parameters, and how you are so sure that improves generalization?
Our response: We have included additional references and explanations regarding the choice of the Random Forest baseline model and the hyperparameters used.
Comment: From line 260 - 261 - what is MCD?
Our response: MCD is a Monte Carlo Dropout. We have made sure to ensure it is defined more clearly.
Comment: Procedure shown in Figure 3 should be in detail explained before figure, this is key part of manuscript.
Our response: We have ensured that the Figures all appear after being described in the text and that the architecture shown in Figure 3 is explained thoroughly within the text.
Comment: What information can be obtained from SWIR1 and SWIR2 - seems like to strong results.
Our response: We have included comments in the discussion about the interpretation of why the models use short-wave infra-red as powerful features for prediction.
Comment: Whole paper seems like you are not clear enough what and why are you trying to do. I suggest you review your manuscript according to all reviewers comments.
Our response: We have made all of the changes suggested by other reviewers and improved the clarity of the motivation, context and results in the paper.
Reviewer 2 Report
Figure 2 needs to improve. Current format is not readable.
Also I recommend to include a study area map with basic Geographic details including population.
Author Response
We would like to thank the reviewer for the valuable comments and suggestions. Your comments and suggestions in this round helped us to improve both the scientific quality and the readability of our manuscript.
The new version of the manuscript has been revised according to the given comments, and the major updates include: a) improvement of the presentation style of our main research contribution, b) enhancement of other related issues indicated by the reviewers, and finally c) general corrections of the English language applied throughout the manuscript.
Please find below our point-by-point responses.
With kindest regards,
The authors
Comment: Figure 2 needs to improve. Current format is not readable.
Our response: We have made changes to Figure 2 to remove the legend and make sure all information previously contained in the legend is outlined in the caption of the figure.
Comment: Also I recommend to include a study area map with basic Geographic details including population.
Our response: Details about the overall Mumbai population and its slum-dwelling population are included in Section 3.2 and a map of the study area is shown in Figure 2
Reviewer 3 Report
The paper focuses on a topic that can make a positive contribution to slum management and improving the quality of life of slum dwellers in large cities.
The paper is well designed and organized. It contains a relevant state-of-the-art presentation and easy-to-follow scientific content.
However, there are some aspects that should be changed to increase the value of the article.
In the Section 1, the authors discuss about different „gaps”. In line 70 we find out about „the first of the three main gaps in the literature”. In lines 78-79 we found out that „four main research gaps ... are investigated in this paper.” In line 168 is written „There are several gaps that we recognize in the literature which we will address in this paper”
We suggest the authors to redesign a bit this Section in order to achieve a better consistency of their statements.
At the end of the first Section (lines 107 – 120), when the contributions are specified, if you do not specify to each contribution its label (the first contribution is, the second,...) do not specify for any. Just enumerate and list them.
In lines 198 – 200, it is stated: „we emphasize that in the literature there are no existing machine learning models for slum 1 mapping which have uncertainty quantification and so this motivates this aspect of our work.”
This can be true, and the bibliography of the article proves that the authors documented well. Anyway, it would be better, to relax a little this statement, adding a „as far as we know”, to the statement from the lines 198-200: „We emphasize that, as far as we know, in the literature there are no existing machine learning models for slum... ”
In line 344, explain the acronym SHAP (i.e. SHapley Additive exPlanations). It is the its first apparition of this acronym in the paper (excepting the Abstract).
The basic contributions of the article are presented in the lines 107-129. The same contributions are presented, using a slightly different expression in Section 6 Conclusions (lines 430-454).
We suggest to modify the Section Conclusions, such that the reader not have the feeling of a deja-vu content.
Author Response
We would like to thank the reviewer for the valuable comments and suggestions. Your comments and suggestions in this round helped us to improve both the scientific quality and the readability of our manuscript.
The new version of the manuscript has been revised according to the given comments, and the major updates include: a) improvement of the presentation style of our main research contribution, b) enhancement of other related issues indicated by the reviewers, and finally c) general corrections of the English language applied throughout the manuscript.
Please find below our point-by-point responses.
With kindest regards,
The authors
Comment: In the Section 1, the authors discuss about different „gaps”. In line 70 we find out about „the first of the three main gaps in the literature”. In lines 78-79 we found out that „four main research gaps ... are investigated in this paper.” In line 168 is written „There are several gaps that we recognize in the literature which we will address in this paper”. We suggest the authors to redesign a bit this Section in order to achieve a better consistency of their statements. At the end of the first Section (lines 107 – 120), when the contributions are specified, if you do not specify to each contribution its label (the first contribution is, the second,...) do not specify for any. Just enumerate and list them.
Our response: We have made sure that the references to gaps in the existing literature are consistent and that they are enumerated in the introduction.
Comment: In lines 198 – 200, it is stated: „we emphasize that in the literature there are no existing machine learning models for slum 1 mapping which have uncertainty quantification and so this motivates this aspect of our work.”.This can be true, and the bibliography of the article proves that the authors documented well. Anyway, it would be better, to relax a little this statement, adding a „as far as we know”, to the statement from the lines 198-200: „We emphasize that, as far as we know, in the literature there are no existing machine learning models for slum... ”
Our response: This recommended change has been made.
Comment: In line 344, explain the acronym SHAP (i.e. SHapley Additive exPlanations). It is the its first apparition of this acronym in the paper (excepting the Abstract).
Our response: We have now included an explanation of the SHAP initialism.
Comment: The basic contributions of the article are presented in the lines 107-129. The same contributions are presented, using a slightly different expression in Section 6 Conclusions (lines 430-454). We suggest to modify the Section Conclusions, such that the reader not have the feeling of a deja-vu content.Our response: We have edited the conclusion section to be less of a repetition of previous sections.
Round 2
Reviewer 1 Report
Dear authors,
the authors did make some changes. But I have highlighted captions of
figures which should not be in figure caption.
I have mentioned that in my first report. Please have authors correct
that.
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer,
We have ensured that all figures are described in the text before they are shown and that they all have succinct titles and captions. We have edited the captions of the figures to ensure that they are all short and succinct a and no longer than one sentence. We have moved explanations about the figures to when the figures are first mentioned in the text.
