Spatiotemporal Modeling of Urban Growth Using Machine Learning
Round 1
Reviewer 1 Report
The paper proposes a framework for modeling the growth of three important variables for cities: population distribution, binary urban footprint, and urban footprint in color. The development of the system is covered in detail in the whole manuscript. The approach, in general, is clear and well written. However, it presents some concerns mainly in the description of the methods and analysis of results. Please find the comments in the following lines:
The authors introduced some related works in the introduction. The idea, in general, presents no novelty as other approaches have already addressed similar applications. The visual representation shown in Figure 19, 26, 29 does not show any relevant differences. I recommend the authors to find a better way to present the variables. Maybe pixel-level differences can represent better the changes rather than using the same images. At the current status, it will result difficult for the readers to measure the efficiency of the approach by visual observation of the images. There are some parts regarding the related works in the manuscript that are described in detail, however, they just occupied space and can be referenced to their respective approach. The spaces should be optimized. Figures should include precise and keywords that make easier for the readers to follow it. For example, Figures 3 and 6. Information can be supported in the text. I recommend the authors to find an appropriate way to present the figures regarding the proposed method. The proposed methods are not adequately presented. There are too many figures and sections that somehow confuse the contributions of the approach. An evaluation of the generated results with the reality could provide the application better applicability. Minor errors in grammar should be revised. Conclusions should provide a better image of the results and the proposed method by focusing on the contributions. Some ablation studies are required to compare the present method with other alternative applications.Author Response
"Please see the attachment."
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This well written manuscript depicts, in a very extensive (perhaps too extensive) and sustained manner, a Machine Learning application for assessing and predicting urban growth dynamics.
The theme is not new, but it certainly is current and interesting.
The problem is quite well defined, yet the research gap could have been better explained, as there are other tools for the same goals and the machine learning employed technique is not the innovative part of the research. Without a clearer analysis to current analytical and AI-based models, this approach novelty is not strongly sustained.
Methods are adequate and thoroughly explained, except for the ML techniques, whose algorithms, selection and development are not addressed in-depth. It should be noted that a good technical level is found both in data processing and ML techniques handling. Authors expertize is evident.
The research beneath the manuscript is very substantial, but hardly replicable. ML technique and implementation was not described in a way that it can be replicated by someone who has good expertize in such matters. For example, the autotuning algorithm is nearly impossible to replicate with complete fidelity to what has been done in this programme by other researchers who are not very experimented ML scientists.
Datasets and frequently codes are provided in ML applications deemed to be replicated by the community. Data repositories can be used for such an end, and cited in the article. I recommend you to do so.
The abstract, figures and flowcharts are very good.
While the aforementioned outlook is clearly positive, even taking into account some easily solvable issues, my two major concerns are related with data quality and with Results and Conclusions sections of this manuscript.
Used data is clearly of low quality, which also led to employing low accuracy interpolations for dealing with population metrics. I will leave a stronger opinion on this for other reviewers with background in spatial analyses of urban data. Someone with a background on exact sciences, as me, would feel highly inclined for not using this kind of data with this ML tools.
My concerns about using machine learning techniques to problems whose data is strongly affected by multiple causes, which are closely interrelated and very susceptible to extreme events and quasi-random political decisions, are partly recognized by the Authors, who endeavoured efforts to filter or accurately account such factors. However, ML only works in its full potential when such corrections are not necessary. In fact, these later causes have long been pointed out as very difficult to be dealt with ML techniques. ML and, broadly speaking, AI applications, are tolerant to uncertainty and error to a certain degree, but its replication for such kind of data remains a problem when dealing with extreme events.
Discussion and conclusions sections, which should always be the most important part of a research paper are, here, surprisingly modest. Basic quantitative assessment of the developed solutions application is necessary. How good was the effectiveness of your options? Have other algorithms and techniques been reporting as more effective in bibliography?
Likewise, framing these findings within the existing body of knowledge would be paramount. Why using this model for Rionegro and Valledupar, considering the current data quality, was a better solution compared to using other models? What is the real gain of employing your proposal? This is the main question to assess whether your research was just a good exercise or is an important contribution for your research filed.
All things considered, Discussion and Conclusions weaknesses are the main reason why I recommend a more extensive upgrading of your manuscript.
As a last comment, please note that “With the sustainable development goals set by the United Nations, the world is finally recognizing that our sustainability as species will be highly influenced by the sustainability of our cities. For this reason, we must aim to develop more capable, interpretable, and accurate prediction models and planning tools for making better decisions” is something that I truly agree. However, it is far from being yielded from your research. Therefore, it should not be included in the article’s Conclusions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors followed the recommendations and addressed the comments accordingly. The text has been improved to be understandable to readers. In addition, the figures and descriptions have been revised.
Reviewer 2 Report
I thank the Authors for their improvements to the manuscript.
Most of my comments were either satisfactorily answered or lead to sensible modifications.
In these conditions I recommend the article acceptance.
