Review Reports - Beyond Vegetation Quantity: The Role of Horticultural Configuration in Shaping Human Perception of Urban Parks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Review

Comments for author File: Comments.pdf

Author Response

Comment 1: Abstract

Materials and methods are not adequately described

Was the visual impression of individual parts of the park evaluated

How were the landscape segments in the park evaluated

What is the scientific and what is the economic contribution of the work

The aim of the work is missing

Response 1: Thank you for this important comment. We have revised the Abstract to clarify the study aim, sampling and evaluation procedures, and contributions. Specifically, we now state that 500 landscape segments were sampled from representative local areas of 10 urban parks using standardized photographs. Their visual impressions were evaluated by 150 participants using 17 attributes on a seven-point scale, with at least 10 valid ratings per image. We also clarify that the study aims to identify horticultural configuration gradients and determine whether they improve perception prediction beyond conventional greenness metrics. Finally, the scientific contribution is defined as advancing park assessment from vegetation quantity to spatial organization, while the potential economic contribution lies in supporting targeted park renovation and reducing inefficient design experimentation. Because no formal cost–benefit analysis was conducted, the latter is presented as practical economic relevance rather than a directly measured economic outcome.

Comment 2: Introduction

The problem is correctly analyzed, but certain parts of the text are repeated, so they need to be shortened

The methodology of data sampling and analysis is not completely clear

Are they working on assessments of the quality and aesthetics of space using AI and how reliable are these methods?

State the objective of the work more clearly and answer it clearly in the conclusion

State several working hypotheses

Response 2: Thank you for these constructive suggestions. We shortened the Introduction by removing repeated descriptions of greenness indicators, horticultural configuration and visual perception. We also clarified the sampling and analytical workflow, including the standardized collection of 500 landscape segments from 10 parks, the evaluation of 17 visual attributes by 150 participants, and the use of multivariate analysis, cross-validated XGBoost and SHAP. Visual quality and aesthetics were assessed directly by participants rather than AI; machine learning was used only to model and interpret relationships between objective configuration metrics and human evaluations. Model reliability was examined using leave-one-park-out cross-validation and park-level bootstrap refitting. Furthermore, we restated the objective as determining whether horticultural configuration explains multidimensional visual perception beyond conventional greenness indicators and revised the Conclusion to answer this objective directly. Four working hypotheses were added: parks with similar greenness may differ in configuration structure; representative configurations differ across perceptual domains; configuration metrics provide predictive information beyond NDVI and greenspace coverage; and different perceptual domains are associated with distinct, potentially nonlinear structural predictors.

Comment 3: 2.1 Study Area and Sampling Design

Is the sample of parks random?

State the design styles of the parks

Are the parks properly maintained and how old are they?

Why the samples are representative

Why the choice of parks is important for the international context

Have these parks been evaluated so far, aesthetically, dendrologically, etc.

Response 3: Thank you for this important comment. The parks were not randomly selected; purposive sampling was used to capture contrasting horticultural configurations while maintaining a shared climatic and urban context. We revised Section 2.1 and added a table reporting each park’s establishment or renovation period, size, dominant design style, vegetation characteristics and maintenance condition. The selected parks represent naturalistic woodland, open-lawn, waterfront, ornamental mixed-planting and built-dominant settings. All were publicly accessible and under routine municipal maintenance during sampling. Representativeness was strengthened by selecting 50 spatially distributed segments per park across typical local landscape settings and applying identical sampling criteria. We also clarified that Chengdu provides an internationally relevant case because its high-density subtropical “Park City” context combines rapid urban development with diverse green-space forms. Nevertheless, the parks are not claimed to represent all international contexts; rather, the identified configuration metrics provide a transferable analytical framework requiring validation elsewhere. Existing assessments mainly concern general park management or vegetation resources; no previous integrated evaluation combining structural configuration, multidimensional aesthetic perception and interpretable machine learning was identified.

Comment 4: Figure 1. The data in Figure 1 is illegible

Response 4: Thank you for pointing this out. Figure 1 has been redesigned and exported at higher resolution. The font size, labels, symbols, colour contrast and panel spacing were enlarged and adjusted to ensure readability in both digital and printed formats. A high-resolution version has also been uploaded as a separate file.

Comment 5: How were the respondents in the sample selected

Was it a random selection or were experts and members of the public chosen, as has already been done in previous similar studies

In what time period was the interview conducted?

Describe the conduct of the survey in more detail

Response 5: Thank you for this comment. We expanded the participant and survey procedures in the revised Methods. Respondents were members of the general public rather than landscape experts because the study aimed to capture ordinary users’ visual perceptions. They were recruited through Wechat using non-probability sampling; therefore, the sample is not presented as statistically representative of the wider population. After informed consent and standardized instructions, 150 participants evaluated randomly presented park images using 17 attributes on a seven-point scale. Image allocation and presentation order were randomized to reduce order and fatigue effects, and each of the 500 landscape segments received at least 10 valid evaluations. We also added the eligibility criteria, participant characteristics, survey setting, completion procedure and data-quality screening rules to improve reproducibility.

Comment 6: The results

Pictures 3-6 are illegible, display differently

To try to shorten this chapter and bring it closer to the readership in a simpler way, a lot of statistics are used, which must give rise to a simpler way of presenting data with an application solution for the profession. The results are loaded with values: R², SHAP, ω², ARI, which makes it difficult to follow

Response 6: Thank you for this suggestion. Figures 3–6 were redesigned using consistent dimensions, larger fonts, simplified legends, improved colour contrast and high-resolution export. The Results section was shortened to emphasize the principal findings, while detailed statistics, including ω², ARI, NMI and model diagnostics, were moved to the Supplementary Materials. Only statistics essential to interpretation, such as cross-validated ΔR² and the dominant SHAP predictors, remain in the main text and are defined at first use. We also added a practitioner-oriented summary linking perceptual objectives to configuration attributes: vegetation aggregation and connectivity for restorative immersion, patch size for spatial openness, and shape complexity for affective richness. These are presented as evidence-based diagnostic directions rather than causal design thresholds.

Comment 7: Discussion

Reduce the chapter to a maximum of 2 pages of text. It is unnecessary to repeat the results already presented. Add 10 more tak references for correlation.

Response 7: Thank you for this suggestion. The Discussion was reduced to approximately two pages by removing repeated numerical results and retaining only their interpretation, comparison with previous research, practical implications and limitations. We also added ten relevant references to contextualize the observed associations between vegetation aggregation, connectivity, patch size, geometric complexity and visual perception. The revised text clearly distinguishes the study’s empirical findings from supporting literature and avoids interpreting predictive correlations as causal effects.

Comment 8: Conclusion

Some general attitude is given about the design of greenery and the design styles, which define the shape of the park, are not taken into account

The professional contribution of the work is given on a smaller scale

How planners can apply these results

Response 8: Thank you for this important comment. We revised the Conclusion to acknowledge that park form is also shaped by design style, historical development and management practices. The three identified configurations are therefore interpreted as representative structural states occurring across design styles, rather than as substitutes for formal design-style classifications. We also expanded the professional contribution from individual planting decisions to park-scale planning. Planners can first diagnose existing spaces using configuration metrics, define the intended perceptual objective, and then prioritize vegetation aggregation and connectivity for restorative immersion, larger visual patches for openness, or greater geometric complexity for affective richness. These metrics can also be used to compare alternative renovation scenarios and monitor post-construction outcomes. Because the study is observational, these recommendations are presented as evidence-based planning directions requiring site-specific validation.

Comment 9: Literature

The references are numerous and adequate, but they are missing in the discussion chapter

Response 9: Thank you for this observation. We added relevant references throughout the Discussion to compare our findings with previous studies and support the interpretation of the identified structural–perceptual associations. We also clearly distinguished the present study’s empirical findings from explanations derived from existing literature.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript addresses a significant limitation in urban green space evaluation by shifting the focus from basic vegetation quantity (e.g., NDVI) to nuanced horticultural configurations and structural mechanisms. The integration of 500 visual samples from Chengdu, spatial metrics, multidimensional perceptual assessments, and interpretable machine learning is well-designed, offering practical insights for urban design and planning. The methodology is robust, and the paper is logically structured.

The introduction is well written.

MM section - Figure 2 - Add a citation for 'Some images in this table are directly adopted from a previous paper, with additional images created and provided by the authors of the current study for the purpose of enhancing the comprehensiveness of the data presentation'

Lines 508–512, and so on:

The framing in lines 508–512 makes it ambiguous whether the authors are describing the specific empirical findings of this study or summarizing existing literature. While the variables mentioned (restorative immersion, spatial openness, affective richness, structural gradients) mirror the results of this manuscript, the passage ends with a citation block [53–60]. Please rephrase this section to clearly distinguish between the authors' own empirical discoveries and the supporting background literature. If these are the key conclusions of the current study being contextualized with prior research, consider revising to explicitly state this (e.g., "In line with previous literature [53–60], our findings demonstrate that..." or "Consistent with prior studies [53–60], the results of this study show...").

The entire discussion section must be carefully rewritten.

However, the authors are to be complimented for an exceptionally thorough, transparent, and methodologically sound limitations section. Explicitly acknowledging the spatial nesting of data and highlighting the use of leave-one-park-out cross-validation demonstrates strong methodological awareness regarding potential data leakage, while framing the three configurations as continuous structural gradients rather than rigid categories adds valuable theoretical clarity. Crucially, the authors display excellent scientific restraint by noting that SHAP values reflect predictive associations rather than direct causal effects, explicitly warning against treating zero-crossing points as absolute design thresholds. This level of nuance is frequently absent in machine-learning applications within urban planning and landscape research. Overall, this section reflects high academic integrity, clearly delineating the empirical scope of the current findings while setting a well-grounded agenda for future multi-city and experimental validation.

Author Response

Comment 1: MM section - Figure 2 - Add a citation for 'Some images in this table are directly adopted from a previous paper, with additional images created and provided by the authors of the current study for the purpose of enhancing the comprehensiveness of the data presentation'. Lines 508–512, and so on:

Response 1: Thank you for this comment. We added the original reference to the Figure 2 caption and clarified which images were reproduced or adapted from the previous publication and which were created by the present authors, including permission information where required. We also revised Lines 508–512 and similar passages throughout the Discussion to distinguish our empirical findings from literature-based interpretations. Statements now begin with phrases such as “In this study, we found…” for our results, followed by “Consistent with previous studies…” when contextualizing them with references.

Comment 2: The framing in lines 508–512 makes it ambiguous whether the authors are describing the specific empirical findings of this study or summarizing existing literature. While the variables mentioned (restorative immersion, spatial openness, affective richness, structural gradients) mirror the results of this manuscript, the passage ends with a citation block [53–60]. Please rephrase this section to clearly distinguish between the authors' own empirical discoveries and the supporting background literature. If these are the key conclusions of the current study being contextualized with prior research, consider revising to explicitly state this (e.g., "In line with previous literature [53–60], our findings demonstrate that..." or "Consistent with prior studies [53–60], the results of this study show...").

Response 2: Thank you for identifying this ambiguity. We revised Lines 508–512 to distinguish our empirical findings from their literature-based interpretation. The revised text first states that the three configuration states differed in restorative immersion, spatial openness and affective richness in the present study, and then relates this multidimensional pattern to previous research [53–60]. Similar passages throughout the Discussion were also revised using explicit phrases such as “Our findings demonstrate…” and “Consistent with previous studies…” to prevent confusion between the current results and supporting literature.

Comment 3: The entire discussion section must be carefully rewritten.

Response 3: Thank you for this recommendation. We comprehensively rewrote the Discussion rather than making isolated revisions. The revised section now focuses on interpreting the principal findings, comparing them with previous research, explaining possible mechanisms, translating them into planning implications and acknowledging limitations. Repeated results and excessive statistics were removed, empirical findings were clearly separated from literature-based explanations, and causal language was replaced with association-based wording where appropriate.

Comment 4: However, the authors are to be complimented for an exceptionally thorough, transparent, and methodologically sound limitations section. Explicitly acknowledging the spatial nesting of data and highlighting the use of leave-one-park-out cross-validation demonstrates strong methodological awareness regarding potential data leakage, while framing the three configurations as continuous structural gradients rather than rigid categories adds valuable theoretical clarity. Crucially, the authors display excellent scientific restraint by noting that SHAP values reflect predictive associations rather than direct causal effects, explicitly warning against treating zero-crossing points as absolute design thresholds. This level of nuance is frequently absent in machine-learning applications within urban planning and landscape research. Overall, this section reflects high academic integrity, clearly delineating the empirical scope of the current findings while setting a well-grounded agenda for future multi-city and experimental validation.

Response 4: We sincerely thank the reviewer for this positive assessment. We have retained these methodological qualifications and further ensured that the revised manuscript consistently presents the configurations as representative states within continuous gradients, SHAP values as predictive associations, and zero-crossing points as diagnostic references rather than causal design thresholds. Future multi-city and experimental validation is also emphasized.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Manuscript ID: horticulturae-4475222

Title: Beyond Vegetation Quantity: The Role of Horticultural Configuration in Shaping Human Perception of Urban Parks

Authors: Yi Peng, Zongsheng Li, Yuzhou Liu, Qibing Chen, Huixing Song

General comments. The manuscript addresses timely and relevant research on the role of spatial vegetation configuration in shaping human perceptions of urban parks beyond conventional measures of vegetation quantity. The topic is relevant to landscape architecture, urban green spaces, environmental psychology, and evidence-based landscape design.

The manuscript has several strengths, including a clear research rationale, a substantial and relatively up-to-date bibliographic base, a multidimensional perceptual assessment, and the integration of landscape metrics with multivariate statistics and explainable machine learning approaches.

- Relevance and topicality of the theme. The theme is very well anchored in current concerns regarding: the quality of urban green spaces; landscape ecology; green infrastructure; landscape perception; psychological health and restoration; evidence-based design; the use of artificial intelligence and machine learning in landscape research.

- Clearly formulated objectives: The four objectives are coherent and logical:

identify structural gradients;
identify perceptual domains;
determine the contribution of the configuration beyond greenness metrics;
identify nonlinear relationships.

- The integration of several analytical methods that give the manuscript an interdisciplinary approach and allow the investigation of the problem from multiple perspectives.

- The implications for landscape design are interesting. The proposal of differentiated strategies for: restorative spaces; open spaces; spaces with experiential richness, are relevant for landscape practice.

The Introduction provides adequate context on the importance of urban parks for environmental quality, ecosystem services, psychological restoration, and human experience. The shift from quantitative assessment of vegetation, through indicators such as NDVI, green space cover, and GVI, to the analysis of spatial vegetation configuration is well presented. The bibliography is adequate and relatively up-to-date, supporting the theoretical underpinnings of the study.

However, the concept of “horticultural configuration” should be defined more precisely in relation to the variables actually measured.

The Research design is relevant and well-designed for the research objectives. The study is done on 500 samples organized within the 10 parks and combines standardized visual sampling, multidimensional perceptual assessment, landscape metrics, multivariate analysis, clustering and explainable machine learning.

Methods are described in detail and use appropriate analytical tools and are presented in an elaborate manner (data standardization; Spearman + FDR; VIF; PCA; PERMANOVA; PERMDISP; K-means; bootstrap; Kruskal-Wallis; Dunn; XGBoost; SHAP; LOWESS).

However, some important aspects require clarification:

- choice of clustering (K = 3) - the authors acknowledge that K = 2 is the statistically optimal solution, while K = 3 has poorer internal performance, but is chosen due to ecological interpretability. This decision should be justified more rigorously.

- characteristics of participants (the characteristics of the participant sample should be reported: age, sex, education, possibly familiarity with green spaces, etc.),

- interpretation and justification of the choice of PCA in relation to an exploratory factor analysis.

The results are logically organized and reported quantitatively. However, several technical editing errors, incorrect figure references and too extensive detailing of explanations for certain figures should be corrected.

The structure of the figures is logical and covers the main stages of the analysis. However, the readability, information density and consistency of numbering require revision.

- Some graphs are very dense and can be difficult to read. Figure 9 and Figure 5 contain a lot of information in a single figure, they could benefit from simplification.

- Lines 379-380 - "Figure 3.2b" instead of Figure 5b;

- Line 467 - "Figure (c)" instead of "Figure 9c".

Conclusions. The main conclusions are supported by the results obtained, in particular the conclusion that the spatial configuration of vegetation provides additional predictive information compared to conventional indicators of vegetation quantity. The analyses presented support the conclusion that certain characteristics, such as vegetation aggregation and connectivity, patch size, and geometric complexity, are associated with different dimensions of park quality perception.

However, some statements in the conclusions section seem to go beyond what can be directly demonstrated by the research design. Because the study is based on observational data and predictive modeling, the relationships identified should be interpreted as associations or predictive relationships and not as causal effects.

Conclusions should avoid causal formulations such as “enhances,” “determines,” “increases,” or “leads to,” except in cases where these are explicitly presented as implications for design or hypotheses to be tested in future experimental research.

I recommend revising the conclusions to make a clearer distinction between empirical results and implications for landscape design. Authors should emphasize that vegetation configuration metrics improve the prediction of perceptual responses, while recognizing that the causal mechanisms underlying these relationships need to be demonstrated through experimental studies.

The bibliography is extensive and relatively current, including work on: environmental/landscape restoration and perception; green space morphology; spatial openness; visual complexity; landscape connectivity; and the effects of green spaces on health.

Overall Assessment. In conclusion, I believe that some methodological clarifications and revisions are needed before the manuscript can be considered for publication. I believe that the manuscript is potentially valuable and publishable after revision. The central finding - that the spatial configuration of vegetation provides information about the perceived qualities of urban parks beyond the amount of vegetation - is potentially significant.

Before publication, I recommend the following changes:

˗ Clarification and redefinition of the concept of “horticultural configuration”.

˗ Presentation of the characteristics of the participants.

˗ Interpretation and justification of the choice of PCA for the 17 variables.

˗ Justification of the choice of K = 3, given that K = 2 has better statistical results.

˗ Clear separation of the “restorative perception” from the “restorative effect”.

˗ Correction of editing and numbering errors.

Author Response

Comment 1: The Introduction provides adequate context on the importance of urban parks for environmental quality, ecosystem services, psychological restoration, and human experience. The shift from quantitative assessment of vegetation, through indicators such as NDVI, green space cover, and GVI, to the analysis of spatial vegetation configuration is well presented. The bibliography is adequate and relatively up-to-date, supporting the theoretical underpinnings of the study.

However, the concept of “horticultural configuration” should be defined more precisely in relation to the variables actually measured.

The Research design is relevant and well-designed for the research objectives. The study is done on 500 samples organized within the 10 parks and combines standardized visual sampling, multidimensional perceptual assessment, landscape metrics, multivariate analysis, clustering and explainable machine learning.

Response 1: Thank you for this positive assessment. We have now defined “horticultural configuration” more precisely as the spatial organization and morphological characteristics of vegetation patches within each visual scene, operationalized through fragmentation, aggregation, connectedness, mean patch area, shape complexity and fractal dimension. NDVI and greenspace coverage represent vegetation quantity and were treated as comparison variables. We also clarify that the term does not refer to species composition, dendrological quality, maintenance intensity or formal park design style. This definition has been added to the Introduction and applied consistently throughout the manuscript.

Comment 2: Methods are described in detail and use appropriate analytical tools and are presented in an elaborate manner (data standardization; Spearman + FDR; VIF; PCA; PERMANOVA; PERMDISP; K-means; bootstrap; Kruskal-Wallis; Dunn; XGBoost; SHAP; LOWESS).

However, some important aspects require clarification:

- characteristics of participants (the characteristics of the participant sample should be reported: age, sex, education, possibly familiarity with green spaces, etc.),

- interpretation and justification of the choice of PCA in relation to an exploratory factor analysis.

The structure of the figures is logical and covers the main stages of the analysis. However, the readability, information density and consistency of numbering require revision.

- Some graphs are very dense and can be difficult to read. Figure 9 and Figure 5 contain a lot of information in a single figure, they could benefit from simplification.

- Lines 379-380 - "Figure 3.2b" instead of Figure 5b;

- Line 467 - "Figure (c)" instead of "Figure 9c".

Response 2: Thank you for this detailed assessment. We strengthened the justification for retaining K = 3. Although K = 2 achieved the best internal validation indices, K = 3 was not presented as the statistically optimal partition but as a representative-state solution that distinguishes three structurally interpretable profiles. Its adequacy was supported by bootstrap Jaccard values of 0.86–0.93, high resampling agreement (ARI = 0.853; NMI = 0.837) and clear centroid differentiation. This distinction and the exploratory status of the three states are now emphasized. We also added participants’ age, sex, education and familiarity with urban green spaces. PCA was selected to reduce correlated perceptual variables into orthogonal composite dimensions for subsequent comparison and prediction, rather than to confirm latent psychological constructs. Its distinction from exploratory factor analysis and the corresponding interpretive limitation are now stated explicitly.

We additionally corrected all figure citations and numbering, including “Figure 3.2b” to “Figure 5b” and “Figure (c)” to “Figure 9c.” Figures 5 and 9 were simplified by reducing redundant annotations and visual density, while detailed information was transferred to the Supplementary Materials. Font sizes, legends, panel labels and numbering were standardized throughout.

Comment 3: Conclusions. The main conclusions are supported by the results obtained, in particular the conclusion that the spatial configuration of vegetation provides additional predictive information compared to conventional indicators of vegetation quantity. The analyses presented support the conclusion that certain characteristics, such as vegetation aggregation and connectivity, patch size, and geometric complexity, are associated with different dimensions of park quality perception.

Response 3: Thank you for this important clarification. We revised the Conclusions to consistently describe the identified relationships as associations or predictive relationships rather than causal effects. Terms such as “enhances,” “determines,” “increases” and “leads to” were replaced with “is associated with,” “corresponds to” or “improves prediction of.” The revised section now separates the empirical conclusion—that configuration metrics provide predictive information beyond NDVI and greenspace coverage—from their potential design implications. We also explicitly state that the underlying causal mechanisms and proposed design applications require validation through controlled experiments and longitudinal or multi-city studies.

Comment 4: The bibliography is extensive and relatively current, including work on: environmental/landscape restoration and perception; green space morphology; spatial openness; visual complexity; landscape connectivity; and the effects of green spaces on health.

Response 4: Thank you for this positive assessment. We retained the broad and current literature base while reviewing all references for relevance, consistency and formatting accuracy. Additional citations were incorporated into the revised Discussion where needed to contextualize the observed structural–perceptual associations.

Comment 5: Overall Assessment. In conclusion, I believe that some methodological clarifications and revisions are needed before the manuscript can be considered for publication. I believe that the manuscript is potentially valuable and publishable after revision. The central finding - that the spatial configuration of vegetation provides information about the perceived qualities of urban parks beyond the amount of vegetation - is potentially significant.

Response 5: We sincerely thank the reviewer for this encouraging overall assessment and for recognizing the potential significance of our central finding. In response to the comments, we have thoroughly revised the manuscript and clarified the sampling strategy, indicator construction, perceptual-dimension reduction, cluster selection, predictive validation, and interpretation of nonlinear responses. We have also distinguished empirical associations from causal claims and strengthened the discussion of practical applicability and study limitations. These revisions improve the methodological transparency and reinforce the conclusion that vegetation configuration provides additional information about perceived urban park quality beyond vegetation quantity.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have taken into account all reviewer's comments. The work has now been significantly improved. I have no additional remarks.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for providing the necessary adjustments.