Coordination and Adaptation: An Analysis of the Spatial Compatibility Between Primary Schools and Adjacent Facilities in China’s Central Cities

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper's research topic, analyzing the spatial compatibility of primary schools and surrounding facilities using multi-model machine learning, has practical significance. However, to meet publication requirements, the following revisions are necessary:
1. The current abstract is too long and lacks quantitative presentation of the core research findings. The authors are encouraged to remove unnecessary terminology and provide specific numerical data.
2. The authors could include quantitative data at the beginning of the introduction, such as trends in high-quality education, primary education, and child-friendly policies over the past decade, to enhance the persuasiveness and urgency of the research context.
3. Section 2.2 only describes the data source but does not specify the timeframe for data collection or the criteria used to filter the POI data. The authors should provide a specific timeframe (e.g., the entire year of 2020), the percentage of invalid points removed, and explain how closed or duplicate facilities were handled to enhance data transparency.
4. The authors' description of the ensemble framework in the methodology section is overly conceptual. The authors should clearly define the key hyperparameter settings (e.g., tree depth, learning rate, sampling rate), and cross-validation folds for CART, RF, and XGBoost, preferably in a table format. 5. Lines 344-349 of the article only mention Borderline-SMOTE, without specifying the oversampling ratio or the extent to which it improves model performance. The authors should provide more information on the changes in the ratio of positive and negative samples, recall, and F1 score before and after oversampling.
6. The Results section presents only precision, recall, and F1 score, lacking significance tests.
7. The authors' discussion of NIMBY is too brief, merely mentioning it as "something to consider." At least some quantifiable metrics (such as noise decibels, road volume, and distance to waste disposal facilities) should be presented, and how these could be incorporated into future prediction models should be explained.
8. Furthermore, the authors note the limitations of the power grid and service radius but fail to validate their sensitivity.
9. The conclusions are too broad and lack practical policy recommendations.

Author Response

Dear Reviewer,

We are most grateful for your valuable comments and thorough review of this paper. We have carefully revised and refined the manuscript in accordance with your suggestions. Your comments are listed below in bold type, with specific issues numbered. Our responses are provided in normal type, while changes/additions to the manuscript are indicated in yellow text.

1.The current abstract is too long and lacks quantitative presentation of the core research findings. The authors are encouraged to remove unnecessary terminology and provide specific numerical data.

Response：Thank you for your valuable suggestions. We have streamlined the abstract section by removing redundant terminology and incorporating quantitative metrics such as model accuracy and feature importance to highlight the core research findings.

2.The authors could include quantitative data at the beginning of the introduction, such as trends in high-quality education, primary education, and child-friendly policies over the past decade, to enhance the persuasiveness and urgency of the research context.

Response：We are most grateful for your guidance. We have supplemented the opening of the introduction with data on China's investment in basic education and achievements in child-friendly city development over the past decade, alongside relevant references. This has strengthened the policy context and practical relevance of the research.

3.Section 2.2 only describes the data source but does not specify the timeframe for data collection or the criteria used to filter the POI data. The authors should provide a specific timeframe (e.g., the entire year of 2020), the percentage of invalid points removed, and explain how closed or duplicate facilities were handled to enhance data transparency.

Response：Thank you for your interest in data transparency. We have explicitly stated in Section 2.2 that data collection spanned the entirety of 2020. We have also detailed the proportion of invalid points removed during POI data cleansing (approximately 8.2%), alongside the spatial deduplication and manual verification applied to duplicate or closed facilities.

4.The authors' description of the ensemble framework in the methodology section is overly conceptual. The authors should clearly define the key hyperparameter settings (e.g., tree depth, learning rate, sampling rate), and cross-validation folds for CART, RF, and XGBoost, preferably in a table format.

Response：Thank you for your correction. We have now added a table to the “Model Training” section, clearly outlining the key hyperparameter settings and cross-validation fold counts for CART, RF, and XGBoost models.

5.Lines 344-349 of the article only mention Borderline-SMOTE, without specifying the oversampling ratio or the extent to which it improves model performance. The authors should provide more information on the changes in the ratio of positive and negative samples, recall, and F1 score before and after oversampling.

Response：Thank you very much for your correction. Upon verification, this study actually employed the SMOTE algorithm to mitigate sample imbalance. We have amended the original text and added the following clarification: following oversampling, the ratio of positive to negative samples across all cities was adjusted to 1:1. This resulted in an approximate 12% improvement in model average accuracy and a roughly 10% increase in the F1 score.

6.The Results section presents only precision, recall, and F1 score, lacking significance tests.

Response：Thank you for your suggestion. We have supplemented the section on model performance evaluation with a significance test for the prediction results between models (using a paired t-test), and have annotated the significance levels in the figures.

7.The authors' discussion of NIMBY is too brief, merely mentioning it as "something to consider." At least some quantifiable metrics (such as noise decibels, road volume, and distance to waste disposal facilities) should be presented, and how these could be incorporated into future prediction models should be explained.

Response：Thank you for raising this important point. We have supplemented the discussion section with specific NIMBY indicators to be introduced in future, such as noise levels in decibels, road traffic volumes, and distances from pollution sources. We have also outlined how these will be incorporated into predictive models through multi-level buffer zone analysis.

8.Furthermore, the authors note the limitations of the power grid and service radius but fail to validate their sensitivity.

Response：Thank you for bringing this to our attention. We shall systematically conduct multi-level buffer zone analysis in subsequent research to validate the sensitivity of service radii across different urban contexts, and we apologise for the lack of thorough verification in this instance.

9.The conclusions are too broad and lack practical policy recommendations.

Response：Thank you for your guidance. We have comprehensively rewritten the conclusions section, formulating feasible policy recommendations tailored to specific cities and scenarios based on the model outputs. These include optimising supporting facilities, standardising the layout of training institutions, and prioritising site selection in areas with high compatibility.

Additionally, to enhance the transparency of the research, this revision includes the uploading of the raw experimental data, complete code, images, and tables for review by the peer reviewers.Once again, we extend our sincere gratitude for your assistance in enhancing the rigour and completeness of our paper.

Sincerely yours,

The Author Team

Reviewer 2 Report

Comments and Suggestions for Authors

I see two major problems with studies of this kind:

Models based on historical data. These models are trained on past information, but that doesn’t necessarily mean the current situation is fine. If we are planning for the future, it makes more sense to look at the best and most up-to-date methods/case studies, not only the past.

Infrastructure should follow people’s needs. School and social infrastructure are closely tied to demographics: number of pupils, level of economic development, and social processes. In many EU countries, young families are moving to more developed cities with stronger social infrastructure. This means the demand for schools rises in those regions but falls in poorer ones. The real challenge is that in well-developed cities it’s not easy to find parcels of land to build schools—especially if we aim for a “15-minute city.” So, the key practical question becomes: where do we actually find land in expensive cities to build a school in the right place?

Major notes:

Overall, I think the work is worth publishing. But I would suggest making the aim and tasks of the study clearer in the introduction (lines 111–124), and then answering these tasks one by one in the conclusion (lines 668–678).

It looks like Task 3 wasn’t addressed at all.

Minor notes:

Line 96: Please spell out SHAP the first time you mention it — SHapley Additive exPlanations.

Lines 155–164: You use data from 2000 and 2020. This needs some explanation, since 2020 is pre-COVID. Why is this timeframe still relevant and selected? Also, I think birth rate numbers are very important for primary school planning and should be included.

Figure 2 is hard to read.

Figure 8: the map is also hard to read.

Author Response

Dear Reviewer,

We are most grateful for your valuable comments and thorough review of this paper. We have carefully revised and refined the manuscript in accordance with your suggestions. Your comments are listed below in bold type. Our responses are provided in normal type, while changes/additions to the manuscript are indicated in yellow text.

I see two major problems with studies of this kind:

Models based on historical data. These models are trained on past information, but that doesn’t necessarily mean the current situation is fine. If we are planning for the future, it makes more sense to look at the best and most up-to-date methods/case studies, not only the past.

Response：We are most grateful for your profound insight. We fully concur that ideal urban planning should be forward-looking and draw upon exemplary precedents. This study employs historical built environment data to identify spatial matching patterns between primary/secondary schools and facilities that have emerged during the development of various cities, thereby providing empirical evidence for understanding urban spatial structures. We have further emphasised in the Discussion and Conclusions section that the model results should serve as a ‘decision-support reference’ rather than a ‘planning blueprint’. It is explicitly stated that subsequent research will integrate population projections with policy scenario analyses to enhance its guidance for future development.

Infrastructure should follow people’s needs. School and social infrastructure are closely tied to demographics: number of pupils, level of economic development, and social processes. In many EU countries, young families are moving to more developed cities with stronger social infrastructure. This means the demand for schools rises in those regions but falls in poorer ones. The real challenge is that in well-developed cities it’s not easy to find parcels of land to build schools—especially if we aim for a “15-minute city.” So, the key practical question becomes: where do we actually find land in expensive cities to build a school in the right place?

Response：We are most grateful for your identification of this core challenge in planning practice. This study identifies areas theoretically suitable for educational facilities from a spatial compatibility perspective, providing quantitative grounds for preliminary screening. As explicitly stated in the text, actual site selection must comprehensively consider practical factors such as land ownership, regeneration costs, and community preferences. Our findings serve as a crucial component within multi-criteria decision-making, particularly in advancing the development of the “15-minute city” concept by aiding the identification of potential areas.

Major notes:

Overall, I think the work is worth publishing. But I would suggest making the aim and tasks of the study clearer in the introduction (lines 111–124), and then answering these tasks one by one in the conclusion (lines 668–678).

Response：Thank you for your valuable suggestions. We have rewritten lines 111–124 of the introduction to clarify the research objectives and specific tasks, and have addressed each point in the conclusion to ensure logical consistency and coherence throughout the text.

It looks like Task 3 wasn’t addressed at all.

Response：We are most grateful for your observation regarding this omission. We have now supplemented the conclusions section with a specific implementation pathway for Task 3, emphasising how the model outputs can be integrated with planning scenarios such as urban renewal and new district development, thereby enhancing the research's practical applicability.

Minor notes:

Line 96: Please spell out SHAP the first time you mention it — SHapley Additive exPlanations.

Response：Thank you for bringing this to our attention. We have now supplemented the full name: ‘SHapley Additive exPlanations’.

Lines 155–164: You use data from 2000 and 2020. This needs some explanation, since 2020 is pre-COVID. Why is this timeframe still relevant and selected? Also, I think birth rate numbers are very important for primary school planning and should be included.

Response：We appreciate your rigorous standards regarding data selection. We have supplemented the relevant paragraph to clarify that the 2020 data represents the most recent pre-pandemic population and POI figures currently available, while the 2000 data was utilised to identify early built-up areas for constructing negative samples. Despite the overall decline in China's birth rate, the nine central cities under study continue to face pressures from population influx and spatial restructuring, rendering primary school planning an enduring priority. We shall incorporate demographic indicators such as birth rates in subsequent research.

Figure 2 is hard to read.

Figure 8: the map is also hard to read.

Response：Thank you for highlighting the issue. We have redrawn Figures 2 and 8, optimising the colour scheme, annotations and layout to enhance clarity and readability.

Furthermore, we have uploaded all original data, process outputs, source code, and raw figures pertaining to this paper to the supplementary materials to enhance the transparency and reproducibility of our research.

Once again, we extend our heartfelt gratitude for your insightful comments and for your recognition and encouragement of this research direction. Each of your suggestions has been immensely valuable to us.

Sincerely yours,

The Author Team

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

after reading your work 3 times, I honestly feel that it has potential but NOT as currently written as it does not meet the minimum methodological standrads. 

Generally, it needs stronger validation, better labeling strategy, clearer police discussion and relevance. The conclusions are also weak and need to be better structured and reinforced.

What I would do

Methodology needs to get stronger

1. Labels must be redefined carefully (assumptions like “built up in 2000 but no schools = incompatible.” look a bit akward to me.
2. Try to use a sensitivity analyses. It is highly recommended in such models
3. Maybe adopt spatial block cross-validation and leave one city for testing
4. Report SD for metrics and robustness purposes
5. Discuss how service catchements vary by density and transport
6. I do not understand the "conflict". It is on the Title - maybe remove it
7. I would improve SHAP interpretation by comparing ranking across models with rank correlation metrics
8. It is really important to sho how modeling outputs can be integrated into planning
9. I believe the POI classification codebook needs to be published.
10. Shorten the abstract, and remove exaggerated claims

I would restructure the paper make more clear and transparent to the reader and resubmit.

Comments on the Quality of English Language

many typos, units need to be consistent km2, % thousands

Author Response

Dear Reviewer,

We are most grateful for your valuable comments and thorough review of this paper. We have carefully revised and refined the manuscript in accordance with your suggestions. Your comments are listed below in bold type. Our responses are provided in normal type, while changes/additions to the manuscript are indicated in yellow text.

Dear Authors,

after reading your work 3 times, I honestly feel that it has potential but NOT as currently written as it does not meet the minimum methodological standrads.

Generally, it needs stronger validation, better labeling strategy, clearer police discussion and relevance. The conclusions are also weak and need to be better structured and reinforced.

What I would do

Methodology needs to get stronger

1.Labels must be redefined carefully (assumptions like “built up in 2000 but no schools = incompatible.” look a bit akward to me.

Response：Thank you for highlighting the importance of label definitions. We have reviewed and strengthened the rationale for negative sample selection, clarifying that designating ‘areas constructed before 2000 without primary schools’ as negative samples is based on the fact that such areas have reached functional maturity over the past two decades. They predominantly serve commercial, industrial, or office purposes with scarce residential land, and existing educational needs are already met by surrounding schools. Thus, they genuinely lack the feasibility and necessity for new primary schools, making their classification as ‘incompatible’ samples reasonable.

2.Try to use a sensitivity analyses. It is highly recommended in such models

Response：Thank you for your suggestion. We have supplemented the methodology section with the following clarification: by training the model using multiple random seeds (e.g., 42, 100, 2024), the performance metrics (accuracy, recall, F1 score) exhibited fluctuations within a 5% range, indicating acceptable model stability.

3.Maybe adopt spatial block cross-validation and leave one city for testing

Response：We are most grateful for this valuable suggestion. We previously attempted to construct a unified cross-city model, but due to significant differences in urban development stages and spatial structures, the model demonstrated poor generalisation capabilities. Consequently, this study adopts a “city-specific modelling” strategy, randomly partitioning data within each city into 80% training and 20% testing sets. We have outlined the rationale for this choice in the discussion section and noted that developing a cross-city generalised model represents a future research direction.

4.Report SD for metrics and robustness purposes

Response：Thank you for your reminder. We have supplemented the standard deviation for all model performance metrics and displayed the fluctuation range from multiple cross-validation runs in the charts to enhance the robustness of the results.

5.Discuss how service catchements vary by density and transport

Response：Thank you for raising this important perspective. We have expanded upon this point in the discussion section, noting that service radii may vary according to population density and public transport accessibility, and indicating that future work will incorporate multi-modal transport networks for more detailed analysis.

6.I do not understand the "conflict". It is on the Title - maybe remove it

Response：Thank you for your correction. We have removed the term ‘conflict’ from the original title, and the revised title is now:Coordination and Adaptation: An Analysis of the Spatial Compatibility between Primary Schools and Adjacent Facilities in Central Cities of China.

7.I would improve SHAP interpretation by comparing ranking across models with rank correlation metrics

Response：We are most grateful for this constructive suggestion. We have conducted a consistency analysis of SHAP feature importance across 27 models (9 cities × 3 algorithms). The complete ranking results are presented in the appendix, while the main text displays the standardised overall feature importance for each city to enhance interpretability and comparability.

8.It is really important to sho how modeling outputs can be integrated into planning

Response：Thank you for highlighting this point. We have further clarified in the introduction, discussion and conclusion how the model results serve three planning scenarios: improving the surroundings of existing primary schools, coordinating the layout of facilities in new districts, and identifying potential school sites within built-up areas. Diagrams illustrate the pathway through which these findings translate into planning recommendations.

9.I believe the POI classification codebook needs to be published.

Response：Thank you for your interest in research reproducibility. We have now made the POI classification coding table publicly available as supplementary material.

10.Shorten the abstract, and remove exaggerated claims

I would restructure the paper make more clear and transparent to the reader and resubmit.

Response：Thank you for your suggestions. We have streamlined the abstract, removing exaggerated claims and emphasising the quantitative findings and methodological contributions.

Furthermore, we have optimised the overall structure of the manuscript to enhance readability and logical coherence, and have uploaded all original data, process results, code, and figures to meet your requirements for research transparency.

Once again, we extend our sincere gratitude for your invaluable guidance, which has enabled us to significantly improve the methodological rigour and overall presentation quality of the paper.

Sincerely yours,

The Author Team

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you to the authors for carefully revising the paper according to my comments. Overall, all the revisions are necessary and effective. The revised manuscript shows significant improvements in structure, content, and expression. This is a very good work, and I congratulate the author.

Author Response

Comment 1: Thank you to the authors for carefully revising the paper according to my comments. Overall, all the revisions are necessary and effective. The revised manuscript shows significant improvements in structure, content, and expression. This is a very good work, and I congratulate the author.

Response: We are deeply grateful to the Reviewer for this generous and encouraging feedback. We sincerely appreciate the time and insightful comments provided during the previous round of review, which were instrumental in helping us strengthen the manuscript. It is highly rewarding to know that our revisions have successfully addressed the points raised. Thank you once again for your valuable guidance and for your kind congratulations.