Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The framework of the paper is complete, but some key aspects, such as the results analysis and discussion of driving factors, still need improvement.

（1）The figure numbering in the paper appears to be inconsistent. It is recommended to label the figures in sequential order.

1. Introduction

（1）Line 49-93：The comparison between the study’s innovative aspects and traditional methods needs to be expanded. While the existing content notes the "limitations of traditional regression analysis," it fails to explicitly elaborate on "how the RF-SHAP framework proposed in this study resolves these limitations."

2. Materials and Methods

（1）Line 144-147. It is recommended to provide the screening criteria for the 62 participants in order to ensure the scientific rigor of the study.

（2）For Figures 2, 3, and 4, it is recommended to add legends to make the conclusions more intuitively accessible.

4. Discussion

(1) Line 460-466: It is recommended to incorporate more existing research findings and strengthen the comparison with classical theories to enhance the depth of the discussion section.

5. Result

(1) Line 531-556: It is recommended to broaden and refine the conclusion section to achieve a higher level of synthesis.

Author Response

Dear Editor and Reviewers,

We sincerely thank you for your positive assessment of our manuscript's framework and for the constructive and meticulous suggestions provided. We appreciate the reviewer's recognition of the study's completeness while also acknowledging the need for deeper elaboration in several key areas. Your comments have been invaluable in helping us significantly improve the rigor, clarity, and impact of our work. We have thoroughly addressed all the points raised, with revisions made throughout the manuscript to enhance the innovation narrative in the Introduction, the methodological transparency in Materials and Methods, the depth of the Discussion, and the synthesis of the Results and Conclusion. Below, we provide a point-by-point response to each comment, detailing the specific changes implemented.

Reviewer comment 1: [“The figure numbering in the paper appears to be inconsistent. It is recommended to label the figures in sequential order.”]

Response: [We sincerely thank the reviewer for pointing out this oversight. We have thoroughly checked and corrected the figure numbering throughout the entire manuscript to ensure it is now entirely consistent and sequential. The figures and their in-text citations have been updated accordingly.]

Reviewer comment 2: [The comparison between the study’s innovative aspects and traditional methods needs to be expanded. While the existing content notes the "limitations of traditional regression analysis," it fails to explicitly elaborate on "how the RF-SHAP framework proposed in this study resolves these limitations."]

Response: [We greatly appreciate this constructive suggestion, as it has allowed us to more explicitly articulate our framework's advantages. To address this, we have expanded the relevant paragraph in the introduction (lines 80-87 in the revised manuscript) by adding a new sentence that details how the RF-SHAP framework resolves the limitations: "The RF-SHAP framework overcomes the limitations of regression analysis by utilizing random forest (RF) to capture non-linear relationships and interactions among landscape elements. Additionally, the SHAP method enhances interpretability by providing insights into the individual contributions of each element, reducing reliance on expert judgment and offering a more objective and scalable approach for landscape analysis." This addition is supported by logical reasoning from our methodology, where RF handles complex, non-linear data patterns that linear regression cannot， and SHAP provides game-theoretic attributions for interpretability。]

Reviewer comment 3: [It is recommended to provide the screening criteria for the 62 participants in order to ensure the scientific rigor of the study. ]

Response: [To ensure the scientific rigor and accuracy of the data, participants must meet the following screening criteria(lines 175-181 in the revised manuscript) : all participants must have normal color vision, with no color blindness or color weakness; they must possess normal cognitive abilities, free from cognitive impairments or neurological disorders, and be capable of understanding and performing the experimental tasks. Additionally, participants with backgrounds in landscape design, urban planning, or related fields are excluded to prevent professional knowledge from influencing preference judgments.]

Reviewer comment 4: [For Figures 2, 3, and 4, it is recommended to add legends to make the conclusions more intuitively accessible.]

Response: [We agree with the reviewer that clear legends are crucial for the intuitive understanding of the figures. We appreciate this excellent suggestion. Action taken in the manuscript: As recommended, we have added comprehensive and descriptive legends to Figures 2, 3, and 4. Each legend now clearly explains the various elements, symbols, or data series presented in the figure, directly linking the visual information to our key findings and making the conclusions more immediately accessible to the reader.]

Reviewer comment 5: [It is recommended to incorporate more existing research findings and strengthen the comparison with classical theories to enhance the depth of the discussion section.]

Response: [We greatly appreciate this suggestion, as it has allowed us to more robustly situate our findings within the established literature. To address this, we have integrated additional references and explicit comparisons into the existing paragraphs of the discussion section without creating a standalone segment, ensuring a natural flow. Specifically, we added a sentence referencing Browning et al. (2014) in the second paragraph (lines 495-498 in the revised manuscript), highlighting how their "14 Patterns of Biophilic Design" corroborate our WVI prominence by emphasizing visual connections to nature as key to well-being, while refining the focus on unobstructed sightlines. We also incorporated a comparison to Prospect-Refuge Theory (Appleton, 1975) at the end of the second paragraph (lines 501-504), explaining how it aligns with our WVI and Freedom effects by positing preferences for landscapes with prospect (open views) and refuge (enclosure), extending the theory through our SHAP-quantified interactions. Finally, we added a reference to Peters and D'Penna in the third paragraph (lines 511-515), linking their work on "Biophilic Design for Restorative University Learning Environments" to our Freedom observations, paralleling prospect-refuge dynamics in built settings for evidence-based trail design implications. These additions are supported by logical reasoning from our SHAP analyses, which reveal non-linear patterns consistent with these theories (e.g., visibility enhancing restorative affordances, as per biophilic literature), and draw from empirical studies like Appleton's evolutionary framework and Browning et al.'s design patterns, which empirically validate preferences for balanced natural elements—directly reinforcing our threshold-and-interaction model without introducing new data or unrelated content.]

Reviewer comment 6: [It is recommended to broaden and refine the conclusion section to achieve a higher level of synthesis.]

Response: [We appreciate this valuable insight, as it has enabled us to transform the conclusion into a more integrated and forward-looking synthesis of our findings. To address this, we have expanded and restructured the conclusion section (lines 567-571 in the revised manuscript) by: (a) adding an introductory paragraph that synthesizes the broader implications for urban sustainability and human well-being, drawing connections across all key findings; (b) refining the summaries of the three main points for conciseness and logical flow, emphasizing interconnections; (c) incorporating a new subsection on implications and future directions (lines 612-619), discussing policy recommendations for urban green space design and potential extensions to other contexts. These changes are supported by logical reasoning that our RF-SHAP framework's nonlinear insights (e.g., thresholds in WVI and GVI) align with emerging evidence on perceptual dynamics in green spaces, as evidenced by recent studies showing that visual and psychological factors mediate well-being (e.g., Frontiers in Psychology, 2025, on AHP-TOPSIS-POE models for behavioral perception). This higher synthesis reinforces the manuscript's contributions to sustainable landscape design, enhancing its theoretical and practical value.]

Kind regards,
Jinlin Teng

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript titled "Landscape Preferences of Recreational Walkways in Urban  Green Spaces:Bada Shanren Meihu Scenic Area, China" delves into the urban populations and the natural  environment. It is a good work, the statistical methods valid and correctly are applied, the quality of the figures and tables is satisfactory, the reference list cover the relevant literature adequately and in an unbiased manner, and the methods are sufficiently documented to allow replication studies. So, I suggest that it is accept for publication in this journal.

Author Response

Dear Editor and Reviewers,

We deeply appreciate the reviewer's positive assessment and endorsement for publication. In response to this comment, we have carefully reviewed the manuscript to ensure that all aspects highlighted by the reviewer remain consistent and of high quality. Specifically, we verified that the statistical methods (as described in the Methods section, lines 150–180) are applied correctly and align with established practices in landscape preference studies, such as those cited in our references. The figures and tables have been cross-checked for clarity and accuracy, and the reference list has been updated to maintain comprehensiveness and neutrality, drawing from key literature in the field. We agree with the reviewer's evaluation and have made no substantive changes to the manuscript, as it already meets the standards highlighted. We are committed to upholding these qualities in the final version.
Best regards,
Jinlin Teng

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

The topic of the manuscript aligns well with the scope of the journal. It is both current and relevant, and likely to be of interest to a relatively broad readership. While the idea of identifying objective indicators of landscape quality is not entirely novel, the methodology proposed offers an interesting approach to spatial evaluation.

The paper is well formatted according to the journal's template. Although I am not a native English speaker, I did not encounter any significant language or stylistic issues—the manuscript is written clearly.

The introduction is supported by sufficient recent literature and provides a solid explanation of the research problem. The Materials and Methods section presents the case study location in a clear and sufficiently detailed manner. However, the rationale behind the choice of methods needs to be better explained. While the employed methods are described, it is not clear why they were selected?! A brief introduction to the type of data used and the overall analytical approach would make this section more comprehensible, especially for readers who are not specialists in this particular research field. Figure 4 is too small; the ranking of the four factors on the right side is completely illegible. I suggest increasing its size and improving readability.

Furthermore, the justification for the choice of methods should be elaborated. In several places, there is an unclear and technically inappropriate comparison between Depth Pro and LiDAR acquisition, which are fundamentally different technologies. Also, the reported processing times seem inaccurate. If a Time-of-Flight ToF camera is used, a precise depth map is obtained immediately, with no need for post-processing—this should be clarified.

The conclusion merely restates the results. I believe it should go beyond this by proposing practical applications of the findings. For example, is the purpose of the evaluation to compare different development alternatives before implementation, or is it intended to assess the existing state of the landscape for the purpose of intervention in areas with the lowest perceived quality? This section could also benefit from some international context and potential application examples, as mentioned earlier in the paper.

In conclusion, I find the manuscript to be of good quality and it has potential for publication, but only after revisions in line with the comments and suggestions above.

Author Response

Dear Editor and Reviewers,

We sincerely thank the reviewer for their positive assessment of the manuscript's topic, relevance, formatting, language clarity, and introduction, as well as for the constructive suggestions that have helped us refine the Materials and Methods section. Below, we provide a point-by-point response to the reviewer's comments, detailing the revisions made to address each concern.

Reviewer comment 1: [The Materials and Methods section presents the case study location in a clear and sufficiently detailed manner. However, the rationale behind the choice of methods needs to be better explained. While the employed methods are described, it is not clear why they were selected?!A brief introduction to the type of data used and the overall analytical approach would make this section more comprehensible, especially for readers who are not specialists in this particular research field.]

Response: [We are grateful for this feedback, which highlights opportunities to enhance the accessibility of the manuscript for diverse readerships. To address this, we have revised the introductory subsection (2.2 Methodological Workflow, lines 121–138 in the revised manuscript) to briefly describe the data types—such as RGB images from tracking sampling points, processed for depth, segmentation, and color features—and the overall workflow. This includes illustrating how subjective ELO scores are integrated with objective metrics via machine learning and SHAP for interpretable preference modeling, as depicted in Figure 2. Additionally, Table 2 now elaborates on the meaning and computational methods of each data metric, providing a clear representation of relevant data types.

Furthermore, within the Methods section, we have inserted concise rationale statements for each methodological subsection:  For the ELO rating (lines 151–154), we emphasize its selection to mitigate biases associated with traditional scales in subjective evaluations. For Depth Pro (lines 194–196), it was chosen for its capacity to perform rapid, zero-shot monocular depth estimation from single RGB images, making it suitable for in-situ photographs captured without specialized hardware. For Mask2Former (lines 226–228), its use is justified by its Transformer-based efficiency in semantic segmentation of complex landscapes, pre-trained on the ADE20K dataset. For color quantization (lines 245–246), the HSV system was adopted due to its alignment with human perceptual characteristics. For machine learning (lines 262–266), six algorithms were evaluated to identify the optimal one for capturing nonlinear relationships in visual data. For SHAP (lines 275–279), it was selected for its game-theoretic approach to interpreting black-box model outputs.]

Reviewer comment 2: [Figure 4 is too small; the ranking of the four factors on the right side is completely illegible. I suggest increasing its size and improving readability.]

Response: [Thank you for pointing out this issue, which improves the manuscript's visual clarity. We have enlarged Figure 4 by 50% in the revised manuscript (now spanning a full column width) and enhanced the font size and contrast for the ranking labels on the right side, making them fully legible. No other changes were made to the figure content, as supported by the need for clear visualization of color extraction results to aid reader comprehension of the HSV-based metrics, consistent with best practices in graphical representation for scientific communication.]

Reviewer comment 3: [ Furthermore, the justification for the choice of methods should be elaborated.]

Response: [We are grateful for this recommendation, as it strengthens the methodological rigor. Building on the additions in response to comment 1, we have elaborated justifications throughout section 2.3.1 by adding specific sentences linking each method to the study's objectives: e.g., ELO for reliable pairwise comparisons reducing evaluator bias (lines 151-154 ); Depth Pro for cost-effective depth from monocular images (lines 194-196); Mask2Former for accurate multi-object segmentation in natural scenes (lines 226-228 ); HSV color extraction for perceptual alignment (lines 245-246); multiple ML algorithms for comparative optimization (lines 262-266 ); and SHAP for feature contribution analysis (lines 275-279). This elaboration is grounded in literature, such as Lundberg and Lee's SHAP framework [37] for interpretability, and logical inference that these choices enable scalable analysis of 424 trail images, enhancing applicability without overcomplicating the manuscript.]

Reviewer comment 4: [In several places, there is an unclear and technically inappropriate comparison between Depth Pro and LiDAR acquisition, which are fundamentally different technologies.]

Response: [We appreciate this astute observation, which ensures technical precision. To resolve this, we have revised the relevant sentences in section 2.3.2 (lines 204-207) to clarify that Depth Pro, as a software-based monocular AI model using RGB images, serves as a complementary alternative to hardware-intensive methods like LiDAR, rather than a direct equivalent. Specifically, we rephrased: "Compared to hardware-based systems like LiDAR, which provide millimeter-precision but require specialized equipment and extensive fieldwork, Depth Pro offers sufficient accuracy for visual landscape analysis with advantages in speed, cost, and flexibility from single images." This adjustment is supported by the model's description in its original paper (arXiv:2410.02073), emphasizing zero-shot metric depth without intrinsics, and aligns with logical reasoning that our study prioritizes efficient processing of field photos over survey-grade precision, avoiding inappropriate equivalency.]

Reviewer comment 5: [Also, the reported processing times seem inaccurate. If a Time-of-Flight ToF camera is used, a precise depth map is obtained immediately, with no need for post-processing—this should be clarified.]

Response: [Thank you for raising this point, which allows us to enhance accuracy. We have clarified in section 2.3.2 (lines 207-215) that the 0.3-second processing time refers to Depth Pro's GPU-based inference on a single RGB image, as confirmed in the model's GitHub repository, and is not comparable to real-time hardware sensors like ToF cameras, which capture depth directly without AI computation. We added: "Unlike immediate depth capture from ToF cameras, Depth Pro's time reflects software processing, making it suitable for post-field analysis of monocular photos." ]

Reviewer Comment 6: [The conclusion merely restates the results. I believe it should go beyond this by proposing practical applications of the findings. For example, is the purpose of the evaluation to compare different development alternatives before implementation, or is it intended to assess the existing state of the landscape for the purpose of intervention in areas with the lowest perceived quality? This section could also benefit from some international context and potential application examples, as mentioned earlier in the paper.]

Response: [We sincerely thank you for this valuable suggestion. To enhance the practical relevance of the Conclusion section, we have incorporated specific application contexts and backgrounds to clarify the real-world value of our study in landscape design. We have emphasized how the findings can guide the improvement of low-quality areas in urban landscapes—particularly through interventions focused on water presence, building visibility, spatial openness, and Green View Index at specific locations to elevate landscape preference. Furthermore, we have introduced a discussion on scalability, proposing that the framework could be extended to different geographical regions and seasonal variations to further refine and improve prediction accuracy. We also note that future research could integrate methods such as AHP-TOPSIS-POE to better inform policy-making for urban green infrastructure. These revisions provide a more practical and internationally relevant perspective to the study.]

Kind regards, 

Jinlin Teng