The Impact of Artificial Intelligence on Urban Green Total Factor Productivity—Evidence from Chinese Cities

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

General Comment:

This manuscript addresses a timely and relevant topic and adopts a generally sound empirical framework. However, several important issues remain, including missing references, incomplete table presentation, unclear variable specification, and the reliance on patent-based AI indicators that may not adequately reflect actual application. In addition, the mechanism discussion is somewhat generic, and the overall presentation requires improvement. Therefore, major revision is recommended before the manuscript can be considered for publication.

Specific Comment:

1. The manuscript refers to multiple tables (Tables 1~8) throughout the text. However, these tables are completely missing in the current version. Please ensure that all referenced tables are properly included, correctly numbered, and clearly formatted in the manuscript. This is essential for readers to verify the data, methodology, and empirical results.
2. The manuscript does not clearly identify or analyze specific AI applications or tools at the city level. Instead, AI is proxied by patent-based indicators, which reflect innovation activity rather than actual deployment. The authors are encouraged to clarify this distinction and, if possible, provide more direct evidence of AI application in practice.
3. The manuscript provides a general explanation of how AI may enhance GTFP through green innovation and industrial upgrading. However, the mechanisms are described in a rather generic manner and rely mainly on proxy variables (e.g., patents), lacking direct evidence of how AI applications improve efficiency at the city level. The authors are encouraged to clarify the causal pathways and provide clearer and more context-specific explanations.

To improve the manuscript, the authors are suggested to:

• Section 2 (Theories and Hypotheses): Add a clear causal chain (e.g., AI adoption → energy efficiency/resource allocation → emissions reduction → GTFP) and briefly include sector-specific examples.
• Section 3.2 (Variables): Better justify the use of patent-based AI indicators and clarify that they capture innovation rather than actual application; if possible, include or discuss alternative indicators reflecting AI deployment.
• Section 4.5 (Mechanism Analysis): Strengthen empirical interpretation by linking results to more concrete channels, such as how green patents or industrial upgrading translate into efficiency gains, and avoid purely abstract explanations.

4. The manuscript contains a lot of “Error! Reference source not found.”, which appears to result from broken cross-references or formatting issues. The authors should carefully check and correct all references to ensure proper citation formatting.
5. The reference list is insufficient and lacks completeness. Many references do not include DOIs, which limits traceability and transparency. The authors should revise and update the references, ensure proper formatting, and provide DOIs for all applicable sources.
6. There is noticeable redundancy between the Introduction and the theoretical framework, with several arguments repeated in similar wording. The authors are encouraged to streamline the text and reduce repetition to improve clarity and conciseness.
7. Several parts of the manuscript, particularly the theoretical framework (Section 2) and mechanism analysis (Section 4.5), would benefit from visual presentation. The authors are encouraged to include a conceptual framework diagram/Figure for illustrating the relationships among AI, GTFP, mediating variables (green innovation and industrial upgrading), and moderating factors (e.g., city size, resource dependency). This would significantly improve clarity and readability for the readers.
8. Line 320, the word “Indframworkh”. Please verify whether it is correct or not.
9. Line 52 and 53, the sentence “while systematic and causal evidence from the city level remains limited.” is duplicated.

Comments on the Quality of English Language

The overall English is understandable. However, it can be improved for clarity and academic precision. Some sentences are repetitive or overly long, and minor grammatical and wording issues are present. For example, there are typographical and editing errors such as the unclear term “Indframworkh” (Line 320) and duplicated sentences (Lines 52–53). Careful language editing and thorough proofreading are recommended to enhance readability and coherence.

Author Response

Dear Reviewer,

Thank you very much for your detailed and constructive comments. We have carefully considered all the issues you raised—ranging from missing tables and incomplete references to the need for clearer mechanisms and better language presentation. In response, we have thoroughly revised the manuscript by inserting all tables, clarifying our AI measurement, providing city-level examples to support the mechanism analysis, correcting cross-references and DOIs, reducing redundancy, adding a conceptual diagram, and polishing the English throughout. We believe the manuscript has been substantially improved. We greatly appreciate your time and insightful suggestions.

The comments from Reviewer1

General Comment:

This manuscript addresses a timely and relevant topic and adopts a generally sound empirical framework. However, several important issues remain, including missing references, incomplete table presentation, unclear variable specification, and the reliance on patent-based AI indicators that may not adequately reflect actual application. In addition, the mechanism discussion is somewhat generic, and the overall presentation requires improvement. Therefore, major revision is recommended before the manuscript can be considered for publication.

 

Comment 1: The manuscript refers to multiple tables (Tables 1~8) throughout the text. However, these tables are completely missing in the current version. Please ensure that all referenced tables are properly included, correctly numbered, and clearly formatted in the manuscript. This is essential for readers to verify the data, methodology, and empirical results.

 

Response:

We apologize for this oversight. The omission occurred because the tables were uploaded as separate files during submission and were not embedded in the main manuscript. This has now been corrected. All tables (Tables 1–8) have been properly inserted, renumbered, and formatted within the revised manuscript to ensure consistency and readability.

 

Comment 2: The manuscript does not clearly identify or analyze specific AI accumulations or tools at the city level. Instead, AI is proxied by patent-based indicators, which reflect innovation activity rather than actual deployment. The authors are encouraged to clarify this distinction and, if possible, provide more direct evidence of AI accumulation in practice.

 

Response;

Thank you for this important comment. We agree that patent-based indicators mainly capture innovation output rather than actual AI deployment. To clarify this distinction, we have added a detailed explanation in Section 3.2. Specifically, we distinguish three dimensions of AI development: technology accumulation, adoption, and application intensity. In this study, our indicator reflects AI technology accumulation, measured by the stock of AI-related patent applications at the city level. This captures a city’s innovation capability rather than actual usage of AI in production or governance. We also discuss this limitation in Section 5.2 . As you rightly suggested regarding more direct evidence, we highlight that future research could incorporate more direct measures of AI deployment, such as firm-level AI adoption data, digital investment records, or cloud-based AI service usage, to further strengthen empirical analysis.

 

Comment 3; The manuscript provides a general explanation of how AI may enhance GTFP through green innovation and industrial upgrading. However, the mechanisms are described in a rather generic manner and rely mainly on proxy variables (e.g., patents), lacking direct evidence of how AI accumulations improve efficiency at the city level. The authors are encouraged to clarify the causal pathways and provide clearer and more context-specific explanations.

 

Response:

Thank you for this helpful suggestion. We agree that clarifying the concrete causal pathways through which AI technologies influence urban Green Total Factor Productivity (GTFP) is critical for strengthening the theoretical and empirical contributions of the study. We also appreciate the reviewer’s suggestion to provide city-level examples, as this allows the mechanisms to be illustrated in practice rather than solely relying on proxy variables. We have revised the manuscript accordingly. First, in Section 2, we clarify the causal chain linking AI to GTFP.: AI accumulation → real-time data monitoring and intelligent optimization → energy efficiency improvements and resource allocation optimization → reductions in pollutants and carbon emissions → enhanced urban GTFP. This delineation makes the mechanisms linking AI to GTFP more transparent and theoretically grounded. Second, to provide concrete empirical illustrations, we incorporated city-level examples throughout the relevant sections. In Section 2.1, we highlighted that in Shenzhen, the deployment of AI-enabled smart grids has reduced electricity transmission losses and improved urban power utilization efficiency, while in Hangzhou, the City Brain system has optimized traffic flows, decreasing energy consumption and emissions. In Section 2.2, we further demonstrated that Shenzhen has leveraged AI to optimize energy management in buildings and industrial parks, reducing energy intensity and accelerating green technology adoption, whereas Hangzhou has integrated AI into environmental monitoring platforms to enable real-time emissions tracking and adaptive regulatory interventions. In Section 2.3, we provided examples from Shenzhen’s smart manufacturing and logistics systems, which optimize resource allocation and improve operational efficiency, and Hangzhou’s AI-enabled e-commerce logistics and digital services, which enhance cross-sector coordination and promote industrial upgrading. Finally, in Section 4.5 Mechanism Analysis, we illustrated measurable efficiency gains at the city level: Shenzhen’s AI-enabled smart grids and Hangzhou’s City Brain optimize energy consumption and resource allocation to support green innovation, while Guangzhou’s smart logistics and Suzhou’s predictive maintenance systems enhance energy efficiency and industrial output. These revisions allow us to link the previously abstract mechanism pathways to tangible outcomes in specific urban contexts, thus demonstrating that AI adoption not only promotes green technological innovation and industrial upgrading but also generates measurable improvements in urban GTFP. The manuscript now provides a clear, theory-grounded, and empirically supported explanation of how AI technologies affect urban productivity, addressing the reviewer’s concerns regarding generalization and causal clarity.

 

Comment 4: The manuscript contains a lot of “Error! Reference source not found.”, which appears to result from broken cross-references or formatting issues. The authors should carefully check and correct all references to ensure proper citation formatting.

 

Response:

We are grateful the reviewer for identifying this formatting issue. It was caused by broken cross-references during manuscript editing. We have now carefully checked the entire manuscript and corrected all errors related to figures, tables, and section references. A full consistency check has also been conducted to ensure that no similar issues remain.

 

Comment 5; The reference list is insufficient and lacks completeness. Many references do not include DOIs, which limits traceability and transparency. The authors should revise and update the references, ensure proper formatting, and provide DOIs for all applicable sources.

 

Response：

We appreciate the reviewer’s feedback regarding the reference list. We acknowledge that the previous version contained incomplete reference information and missing DOIs. In the revised manuscript, we have thoroughly updated the reference list following the formatting requirements of Sustainability. All references have been checked for completeness, standardized formatting, and DOI inclusion where available. We have also conducted an additional verification to ensure accuracy and consistency throughout the reference section.

 

Comment 6; There is noticeable redundancy between the Introduction and the theoretical framework, with several arguments repeated in similar wording. The authors are encouraged to streamline the text and reduce repetition to improve clarity and conciseness.

Response：

We thank the reviewer for this valuable comment. In response, we carefully revised the Introduction and Section 2 to clearly delineate their respective functions and reduce redundancy. The Introduction now focuses on research motivation, questions, empirical context, and the main contributions of the study, providing a concise overview of AI, urban green growth, and GTFP without delving into detailed mechanisms. Section 2, in contrast, is dedicated to developing the theoretical framework and hypotheses, emphasizing mechanistic reasoning. In particular, we highlight the role of AI in urban GTFP through endogenous growth theory and digital transformation, and introduce dynamic capability theory to explain how AI enhances cities’ sensing, learning, and coordination capabilities, forming the theoretical basis for the mediating channels. Three hypotheses are clearly articulated: H1 (AI positively affects urban GTFP), H2 (AI promotes urban GTFP indirectly through green technological innovation), and H3 (AI promotes urban GTFP indirectly through industrial upgrading). In addition, we incorporated city-level examples, such as Shenzhen’s AI-enabled smart grids, Hangzhou’s City Brain system, and Guangzhou’s smart logistics, to illustrate these mechanisms in practice, adding empirical context without repeating the general background from the Introduction. Overall, these revisions ensure that the Introduction and Section 2 are logically complementary, with the Introduction setting up the motivation and Section 2 establishing the theoretical rationale and hypotheses, effectively eliminating the previous redundancies. All changes are highlighted in the revised manuscript.

 

Comment 7：Several parts of the manuscript, particularly the theoretical framework (Section 2) and mechanism analysis (Section 4.5), would benefit from visual presentation. The authors are encouraged to include a conceptual framework diagram/Figure for illustrating the relationships among AI, GTFP, mediating variables (green innovation and industrial upgrading), and moderating factors (e.g., city size, resource dependency). This would significantly improve clarity and readability for the readers.

Response：

We sincerely thank the reviewer for this valuable suggestion. In response, we have added a conceptual framework diagram at the end of Section 2 (Theories and Hypotheses) to visually illustrate the relationships among artificial intelligence (AI), urban Green Total Factor Productivity (GTFP), the mediating variables (green technological innovation and industrial upgrading), and moderating factors (such as city size, resource endowment, and transportation conditions). The figure explicitly illustrates the direct effect of AI on GTFP (H1), as well as the two indirect pathways through green innovation (H2) and industrial upgrading (H3), while also incorporating the moderating role of urban heterogeneity. This visual representation complements the theoretical discussion and mechanism analysis in Sections 2 and 4.5, enhancing the clarity, readability, and overall coherence of the manuscript. The figure has been included at the end of Section 2 and is referenced accordingly in the revised text.

 

Comment 8：Line 320, the word “Indframworkh”. Please verify whether it is correct or not.

Response：

We are grateful to the reviewer for carefully identifying this typographical error. The term “Indframworkh” was indeed incorrect and resulted from an oversight during manuscript preparation. We have corrected it and standardized the terminology as “Indupgrading” throughout the manuscript to ensure consistency in variable naming and clarity of presentation. A thorough search has been conducted in the relevant sections of the manuscript.

 

Comment 9：Line 52 and 53, the sentence “while systematic and causal evidence from the city level remains limited.” is duplicated.

 

Response：

We sincerely thank the reviewer for pointing out this issue. We apologize for the oversight. The duplicated sentence (“while systematic and causal evidence from the city level remains limited.”) has been removed accordingly in the revised manuscript to ensure the conciseness of the text

 

Comments on the Quality of English Language：The overall English is understandable. However, it can be improved for clarity and academic precision. Some sentences are repetitive or overly long, and minor grammatical and wording issues are present. For example, there are typographical and editing errors such as the unclear term “Indframworkh” (Line 320) and duplicated sentences (Lines 52–53). Careful language editing and thorough proofreading are recommended to enhance readability and coherence.

 

Response：

Thank you for this constructive comment. We recognize that the manuscript can be further improved in terms of clarity, conciseness, and academic expression. In response, we have undertaken a comprehensive language revision of the entire manuscript. This includes improving sentence clarity, restructuring overly long or complex sentences, eliminating redundant expressions, and correcting grammatical and lexical inaccuracies. We have also carefully addressed all typographical and editing issues identified by the reviewer. For instance, the unclear term “Indframworkh” (Line 320) has been corrected, and the duplicated sentences (Lines 52–53) have been removed. Moreover, the manuscript has undergone an additional round of thorough proofreading to ensure consistency, readability, and overall linguistic quality.

Reviewer 2 Report

Comments and Suggestions for Authors

1. In the abstract (Lines 16–25), the information appears somewhat overloaded, particularly with excessive details on mechanisms and heterogeneity, which weakens the presentation of the core findings. It is recommended to streamline this section. In addition, it is unclear whether “0.75 units” refers to a change in standard deviations or an absolute change.
2. There is an obvious repetition in Lines 48–50, where “remains limited” appears twice. It is advisable to carefully proofread the entire manuscript to eliminate similar repetitions and grammatical issues.
3. The description of the research contributions (Lines 70–78) is currently somewhat vague and lacks specificity.
4. In the construction of the AI indicator, the use of a “keyword + patent retrieval” approach may introduce substantial measurement error. It is recommended to include at least one robustness check (e.g., alternative measures such as a digital economy index or the number of AI-related firms), or to provide a simple manual validation to demonstrate the accuracy of the classification.
5. The measurement of GTFP needs to clearly specify the components of inputs, desirable outputs, and undesirable outputs to ensure transparency and replicability.
6. The instrumental variable used to address endogeneity—“terrain ruggedness × lagged AI”—is debatable. Terrain characteristics may directly affect economic structure and environmental efficiency, and thus GTFP, potentially violating the exogeneity assumption. It is recommended to provide a more convincing theoretical justification or consider using more commonly adopted historical instruments.
7. The implementation details of the Double Machine Learning approach are not sufficiently explained. More information on model specification and estimation procedures is needed.
8. The mechanism analysis does not formally test mediation effects. It is recommended to conduct mediation analysis using methods such as the Sobel test or bootstrap techniques, or alternatively adopt a structural equation modeling (SEM) framework.
9. In the descriptive statistics (Lines 361–362), the reported population density of 6.16 persons/km² is clearly inconsistent with typical Chinese urban data, suggesting a possible unit error or logarithmic transformation issue. The variable definition and units should be carefully re-examined, as this constitutes a critical data issue.
10. There is a logical inconsistency in the interpretation of the lagged model. The analysis uses lagged AI variables but interprets the results as evidence of path dependence in GTFP. Methodologically, path dependence should be examined using lagged GTFP; otherwise, the interpretation should be revised.
11. In the discussion of results, the role of AI appears somewhat overstated, with terms such as “transformative” and “central driver.” It is recommended to adopt more cautious academic language, such as “suggest” or “indicate,” in line with standard scholarly practice.
12. The conclusions and policy recommendations (e.g., “enhancing AI application levels” and “strengthening integration”) lack specificity. It would be more meaningful to link these recommendations to the heterogeneity findings and provide targeted suggestions for different types of cities (e.g., resource-based cities or smaller cities), such as investments in AI infrastructure or green finance support.
13. There is a duplication issue in the references section (the same reference appears more than once). A thorough check of the reference list is required.

Author Response

Dear Reviewer,

Thank you very much for your detailed and constructive comments. We have carefully addressed all the issues you raised, including the abstract clarity, measurement transparency, instrumental variable justification, mechanism testing, language tone, and policy specificity. We have streamlined the abstract, added robustness checks with alternative AI measures, clarified key variable definitions, revised the interpretation of the lagged model, adopted more cautious academic language, and linked policy recommendations to heterogeneity findings. All duplicated references have also been removed. We believe the manuscript has been greatly improved. We deeply appreciate your time and thoughtful suggestions.

The comments from Reviewer 2

Comment 1：In the abstract (Lines 16–25), the information appears somewhat overloaded, particularly with excessive details on mechanisms and heterogeneity, which weakens the presentation of the core findings. It is recommended to streamline this section. In addition, it is unclear whether “0.75 units” refers to a change in standard deviations or an absolute change.

 

Response：

We appreciate this helpful suggestion. In response, we have revised the abstract to improve clarity and focus by streamlining the presentation. Specifically, we reduced excessive details related to mechanisms and heterogeneity, and emphasized the core empirical findings more clearly. We also clarified the interpretation of the estimated coefficient. The revised manuscript explicitly states that a one-standard-deviation increase in AI development is associated with an approximately 0.75 increase in GTFP, where GTFP is measured in its original index scale rather than a standardized form.

Comment 2：There is an obvious repetition in Lines 48–50, where “remains limited” appears twice. It is advisable to carefully proofread the entire manuscript to eliminate similar repetitions and grammatical issues.

 

Response：

We thank the reviewer for carefully identifying this issue. The repeated phrase has been corrected in the revised manuscript. In addition, we have carefully proofread the entire manuscript to eliminate similar redundancies and improve overall linguistic accuracy and readability.

 

Comment 3：The description of the research contributions (Lines 70–78) is currently somewhat vague and lacks specificity.

 

Response：

We sincerely thank the reviewer for this helpful comment. We agree that the original description of the research contributions was somewhat vague and lacked sufficient specificity. In response, we have carefully revised this section (Lines 70–78) to enhance clarity and precision. Specifically, we have made the contributions more explicit by clearly articulating the city-level focus of the analysis, the role of AI in green growth, and its positioning within a general-purpose technology framework. We further clarify the key transmission channels through which AI affects urban GTFP, namely green innovation and industrial upgrading, and explicitly situate these mechanisms within a technology–innovation–structure–green growth transmission framework. In addition, we refine the heterogeneity analysis by clearly linking it to differences in resource endowments, urban size, and transportation accessibility, thereby emphasizing the boundary conditions and context-dependent nature of the effects. These revisions substantially improve the specificity and theoretical clarity of the contribution statements.

 

Comment 4：In the construction of the AI indicator, the use of a “keyword + patent retrieval” approach may introduce substantial measurement error. It is recommended to include at least one robustness check (e.g., alternative measures such as a digital economy index or the number of AI-related firms), or to provide a simple manual validation to demonstrate the accuracy of the classification.

 

Response：

We sincerely thank the reviewer for this insightful comment regarding the potential measurement error associated with the “keyword + patent retrieval” approach used in constructing the AI indicator. In response, we have conducted additional robustness checks using two alternative proxy measures to validate the reliability of our baseline indicator. First, we use the employment scalein the “Information Transmission, Computer Services, and Software Industry”at the city level. This indicator captures the human capital and infrastructure foundation necessary for AI development and diffusion. These industries constitute the core of the digital economy and provide essential technical support for AI applications, including software development, data processing, and algorithm-related services. Therefore, the employment scale in this sector reflects the underlying capability of a city to support AI-related technological activities. Second, as suggested by the reviewer, we incorporate the number of AI-related firms in each city to capture the actual industrial presence and commercialization intensity of AI technologies. Compared with patent-based indicators, which mainly reflect innovation outputs, firm-level measures provide a closer approximation to the real-world deployment and market adoption of AI, thereby reflecting the degree of AI application across economic activities. Together, these two alternative measures capture complementary dimensions of AI development—namely, the human capital foundation and the industrial application level. The empirical results based on these alternative indicators remain consistent with our baseline findings, confirming the robustness of our main conclusions. These additional analyses have been incorporated into the revised manuscript.

 

Comment 5：The measurement of GTFP needs to clearly specify the components of inputs, desirable outputs, and undesirable outputs to ensure transparency and replicability.

Response：

 

Response：

We thank the reviewer for this helpful suggestion. In response, we have revised the description of the GTFP measurement to improve clarity, transparency, and replicability. Specifically, we now explicitly detail the construction of the SBM Malmquist–Luenberger index by clearly specifying the input variables (capital stock and labor input), the desirable output (real GDP at constant prices), and the undesirable outputs (industrial wastewater emissions, sulfur dioxide emissions, and industrial smoke and dust emissions). Capital stock is estimated using the perpetual inventory method, and labor input is measured by urban employment levels.

These revisions ensure a more transparent and replicable measurement framework for GTFP, in line with the reviewer’s suggestion.

 

Comment 6：The instrumental variable used to address endogeneity—“terrain ruggedness × lagged AI”—is debatable. Terrain characteristics may directly affect economic structure and environmental efficiency, and thus GTFP, potentially violating the exogeneity assumption. It is recommended to provide a more convincing theoretical justification or consider using more commonly adopted historical instruments.

 

Response：

 We thank the reviewer for this insightful comment regarding the validity of the instrumental variable. We fully agree that terrain characteristics may, in principle, be correlated with long-run economic structure and environmental outcomes, and thus may raise concerns regarding the exclusion restriction. In our identification strategy, the instrumental variable is constructed as the interaction between time-invariant city terrain ruggedness and lagged AI intensity. The identifying variation therefore mainly comes from exogenous geographic constraints that shape the spatial diffusion and adoption of AI technologies, rather than from terrain characteristics directly affecting contemporaneous changes in GTFP. To further clarify this point, we have revised the manuscript to emphasize that while the exclusion restriction cannot be directly tested, the inclusion of city and year fixed effects, together with a rich set of control variables, helps isolate the exogenous supply-side shock potential confounding influences. We appreciate the reviewer’ s constructive suggestion, and we have revised the relevant sections to more clearly acknowledge this limitation and strengthen the discussion of identification validity.

 

Comment 7：The implementation details of the Double Machine Learning approach are not sufficiently explained. More information on model specification and estimation procedures is needed.

 

Response：

We thank the reviewer for this helpful comment. We agree that the initial version did not provide sufficient detail regarding the implementation of the Double Machine Learning (DML) framework. In response, we have revised the manuscript to provide a more detailed description of the model specification and estimation procedures. Specifically, we now clarify that the sample is randomly partitioned into five folds for cross-fitting estimation. Lasso and Ridge regression are employed as auxiliary machine learning models to estimate nuisance parameters, with tuning parameters selected via grid-search cross-validation. Based on the orthogonalization principle, we then estimate the partial effect of AI intensity on urban GTFP while mitigating overfitting and omitted-variable bias. These additional details improve the transparency and reproducibility of our empirical strategy. We appreciate the reviewer’s constructive suggestion, which has helped us significantly improve the clarity of the methodology section.

 

Comment 8：The mechanism analysis does not formally test mediation effects. It is recommended to conduct mediation analysis using methods such as the Sobel test or bootstrap techniques, or alternatively adopt a structural equation modeling (SEM) framework.

 

Response：

 We thank the reviewer for the valuable suggestion regarding mediation analysis. We would like to clarify that the mechanism analysis in this study follows the two-step framework proposed in the literature (引用江艇和m-y文章.), rather than a formal mediation analysis approach such as SEM or Sobel tests. Specifically, the two-step approach first strictly focus on identifying the causal effect whether the explanatory variable (AI accumulation) is exogenous and has been shown to significantly affect the outcome variable (GTFP) in the baseline model. Given this established causal relationship, the mechanism analysis then examines the effect of AI on the proposed transmission channels (green innovation and industrial upgrading).In the second step, the relationship between the transmission channels (M) and the outcome variable (GTFP) is supported by existing theoretical and empirical literature, rather than being re-estimated within a structural mediation framework. This approach ensures that the identification of the core explanatory variable remains consistent with our baseline causal inference strategy. We have revised Section 4.5 to explicitly clarify this reduced-form mechanism testing strategy and its distinction from formal mediation analysis. We appreciate the reviewer’ s constructive suggestion, which has helped improve the clarity of the empirical framework.

 

Comment 9；In the descriptive statistics (Lines 361–362), the reported population density of 6.16 persons/km² is clearly inconsistent with typical Chinese urban data, suggesting a possible unit error or logarithmic transformation issue. The variable definition and units should be carefully re-examined, as this constitutes a critical data issue.

 

Response：

We thank the reviewer for this careful observation. We agree that the initial description of the population density variable may have caused confusion regarding its scale and units. In the revised manuscript, we have clarified that the population density variable is measured in natural logarithmic form. Accordingly, the reported mean value of 6.16 reflects the logarithmic transformation of population density (persons per square kilometer), rather than the raw level. We have also revised the variable definition table to explicitly state this transformation to avoid any ambiguity. We appreciate the reviewer’ s attention to detail, which has helped us improve the clarity and transparency of the data description.

 

Comment 10；There is a logical inconsistency in the interpretation of the lagged model. The analysis uses lagged AI variables but interprets the results as evidence of path dependence in GTFP. Methodologically, path dependence should be examined using lagged GTFP; otherwise, the interpretation should be revised.

 

Response：

We thank the reviewer for this insightful comment regarding the interpretation of the lagged model. We agree that the original wording may have led to confusion by implying path dependence in GTFP, which would indeed require the inclusion of lagged dependent variables. In the revised manuscript, we have corrected this issue by clarifying that Models (3)–(5) are designed to capture the dynamic and delayed impacts of AI technology on current urban green total factor productivity (GTFP), rather than to test path dependence in GTFP itself. Specifically, we now interpret the results as reflecting the dynamic and cumulative influence of prior AI development on current green productivity. In addition, we have removed all references to “path dependence” and revised the relevant discussion to emphasize mechanisms such as knowledge accumulation, learning-by-doing, and technology spillovers, which explain the persistence of AIs impact over time. These revisions ensure consistency between the model specification and its interpretation. We appreciate the reviewers constructive suggestion, which has significantly improved the clarity and rigor of our analysis.

 

Comment 11；In the discussion of results, the role of AI appears somewhat overstated, with terms such as “transformative” and “central driver.” It is recommended to adopt more cautious academic language, such as “suggest” or “indicate,” in line with standard scholarly practice.

 

Response：

We thank the reviewer for this valuable suggestion regarding the tone of interpretation in the discussion of results. We agree that some expressions in theinitial manuscript  , such as “transformative” and “central driver,” may have been overly assertive. In the revised manuscript, we have carefully revised the relevant sections and adopted more cautious academic language throughout the discussion of results. Specifically, we have replaced strong causal or normative expressions with more appropriate terms such as “suggest,” “indicate,” in line with standard scholarly conventions. These revisions improve the tone and ensure that the interpretation of results remains appropriately measured and consistent with empirical evidence. We appreciate the reviewers constructive comment, which has helped improve the academic rigor and presentation of the manuscript.

 

Comment 12；The conclusions and policy recommendations (e.g., “enhancing AI accumulation levels” and “strengthening integration”) lack specificity. It would be more meaningful to link these recommendations to the heterogeneity findings and provide targeted suggestions for different types of cities (e.g., resource-based cities or smaller cities), such as investments in AI infrastructure or green finance support.

 

Response：

We thank the reviewer for this valuable suggestion regarding the specificity and practical relevance of the policy recommendations. We fully agree that the original version provided relatively general policy implications.In response to the reviewer’s comment, we have substantially revised the conclusions and policy recommendations section by explicitly incorporating the heterogeneity results of the empirical analysis. Specifically, we now provide differentiated and targeted policy suggestions based on city characteristics, including resource-dependent cities, small cities, and large and medium-sized cities.For resource-dependent cities, we emphasize strengthening AI infrastructure construction and digital skill training to reduce dependence on resource-intensive industries and promote industrial diversification. For small cities, we highlight the role of AI in improving public service efficiency and supporting smart agriculture and basic digital governance. For large and medium-sized cities, we stress the importance of deeper integration of AI into industrial chains, innovation systems, and interregional green governance networks to fully leverage agglomeration and spillover effects.These revisions directly align the policy implications with the heterogeneous empirical findings, thereby enhancing the practical relevance and policy applicability of the study. We appreciate the reviewer’s insightful suggestion, which has significantly improved the clarity and usefulness of the policy section.

 

Comment 13；There is a duplication issue in the references section (the same reference appears more than once). A thorough check of the reference list is required.

 

Response：

We thank the reviewer for pointing out this issue. In response, we carefully screened the entire reference list to identify duplicated entries.  All duplicated references have now been removed in the revised manuscript. In addition, we performed an comprehensive cross-check of the bibliography section to ensure that all citations are unique, correctly formatted, and consistently presented.

Reviewer 3 Report

Comments and Suggestions for Authors

I found this manuscript timely and relevant. The paper addresses an important question about whether artificial intelligence applications can improve urban green total factor productivity in Chinese cities. The focus on 120 prefecture-level cities from 2010 to 2023 gives the study a useful empirical scope, and the attention to green innovation and industrial upgrading as mechanisms makes the paper relevant to current debates on AI-enabled sustainable development. Overall, I think the manuscript has potential, and my comments below are offered as minor suggestions to further improve its conceptual clarity and presentation.

First, I encourage the authors to position AI applications more clearly within the broader digital transformation literature. At present, AI is mainly treated as a standalone technological input. However, AI adoption is often part of a wider process of digital transformation involving digital infrastructure, organisational capability building, resource orchestration, innovation processes, and multilevel change. This broader framing could help the authors explain why AI matters not only as a technology, but also as part of a wider transformation process that reshapes how cities, industries, and firms create green value. In this respect, the authors may find it useful to engage with recent work on digital transformation and innovation, including Salunke and Kowalkiewicz’s work on digital transformation in SMEs and the digital transformation–innovation nexus. The definition of digital transformation could also be framed in terms of its links to the firm's competitive advantages (Saeedikiya et al., 2025, p. 3), which could be particularly helpful in strengthening the paper's conceptual foundation.

Second, I suggest clarifying the distinction between “AI application,” “AI technology accumulation,” and “AI adoption.” The paper’s title and theoretical discussion refer to AI applications, but the empirical measure is mainly based on AI-related patent stock. Patent data are useful for capturing technological invention and accumulation, but they may not fully reflect actual AI use in production systems, urban governance, firms, or green transformation practices. The authors already acknowledge this limitation, which is appreciated, but it would be helpful to make this distinction more explicit throughout the manuscript. This would allow the authors to interpret the findings more carefully and avoid overstating what the AI index captures.

Third, the theoretical mechanism could be developed a little further. The argument that AI improves GTFP through green innovation and industrial upgrading is reasonable, but the paper could explain more clearly how this process unfolds. For instance, AI may help cities and firms sense environmental inefficiencies, optimise energy and resource use, process large-scale environmental data, support cleaner production decisions, and reconfigure industrial activities toward greener and higher-value sectors. Bringing in a dynamic capability or digital transformation perspective could make the mechanism more convincing, because it would show how AI-enabled sensing, learning, coordination, and reconfiguration contribute to green productivity improvement.

Fourth, the instrumental variable strategy would benefit from a clearer justification. The interaction between topographic relief and lagged AI intensity is interesting, but the exclusion restriction needs to be explained more carefully. Topographic conditions may also influence transport accessibility, industrial structure, urban expansion, environmental pressure, and regional development patterns, all of which may be related to GTFP. I suggest that the authors provide a stronger explanation of why the instrument affects current GTFP only through AI application, or add further robustness checks if data availability allows.

Finally, inconsistent hypothesis labels such as H2, H2a, and H2b, repeated subsection titles in the heterogeneity analysis, and variable names should be improved. These issues are not substantive, but addressing them would improve the readability and professional presentation of the paper.

Author Response

Dear Reviewer,

Thank you very much for your positive and constructive comments. We greatly appreciate your recognition of the manuscript's potential. We have carefully addressed all your suggestions, including positioning AI within the broader digital transformation literature, clarifying the distinction between AI accumulation and adoption, deepening the theoretical mechanism using a dynamic capability perspective, strengthening the instrumental variable justification, and correcting presentation inconsistencies. We believe these revisions have substantially improved the manuscript. Thank you again for your time and valuable feedback.

The comments from Reviewer 3

I found this manuscript timely and relevant. The paper addresses an important question about whether artificial intelligence applications can improve urban green total factor productivity in Chinese cities. The focus on 120 prefecture-level cities from 2010 to 2023 gives the study a useful empirical scope, and the attention to green innovation and industrial upgrading as mechanisms makes the paper relevant to current debates on AI-enabled sustainable development. Overall, I think the manuscript has potential, and my comments below are offered as minor suggestions to further improve its conceptual clarity and presentation.

 

Comment 1；First, I encourage the authors to position AI accumulations more clearly within the broader digital transformation literature. At present, AI is mainly treated as a standalone technological input. However, AI adoption is often part of a wider process of digital transformation involving digital infrastructure, organisational capability building, resource orchestration, innovation processes, and multilevel change. This broader framing could help the authors explain why AI matters not only as a technology, but also as part of a wider transformation process that reshapes how cities, industries, and firms create green value. In this respect, the authors may find it useful to engage with recent work on digital transformation and innovation, including Salunke and Kowalkiewicz’s work on digital transformation in SMEs and the digital transformation–innovation nexus. The definition of digital transformation could also be framed in terms of its links to the firm's competitive advantages (Saeedikiya et al., 2025, p. 3), which could be particularly helpful in strengthening the paper's conceptual foundation.

 

Response：

We appreciate this insightful suggestion. It helped us rethink the positioning of AI more broadly within the digital transformation literature. In the revised manuscript, AI is no longer treated as an isolated technological input, but is explicitly embedded in the broader process of digital transformation. We now emphasize that digital transformation involves coordinated changes in infrastructure, organizational capabilities, resource allocation, and innovation systems, rather than technological adoption alone. Within this framework, AI is framed as a core enabling technology that interacts with these dimensions and contributes to reshaping how cities and industries generate green value. We also draw on recent studies on digital transformation and innovation, including work on SMEs and the digital transformation–innovation nexus, as well as literature linking digital transformation to competitive advantage (e.g., Saeedikiya et al., 2025).Importantly, this conceptual refinement is aligned with our empirical design, where AI is measured through patent-based technological accumulation. The revision therefore strengthens theoretical positioning without altering the identification strategy.

 

Comment 2：Second, I suggest clarifying the distinction between “AI accumulation,” “AI technology accumulation,” and “AI adoption.” The paper’s title and theoretical discussion refer to AI accumulations, but the empirical measure is mainly based on AI-related patent stock. Patent data are useful for capturing technological invention and accumulation, but they may not fully reflect actual AI use in production systems, urban governance, firms, or green transformation practices. The authors already acknowledge this limitation, which is appreciated, but it would be helpful to make this distinction more explicit throughout the manuscript. This would allow the authors to interpret the findings more carefully and avoid overstating what the AI index captures.

 

Response：

This is an important clarification, and we appreciate the reviewer’s careful reading. We fully agree that clearly differentiating these concepts is essential for the accurate interpretation of our empirical results and for avoiding any potential overstatement of what our measure captures.In response, we have made substantial revisions to improve conceptual clarity throughout the manuscript. First, in Section 3.2 (Variables), we have explicitly introduced and distinguished three related but conceptually different dimensions of AI development: (i) AI technology accumulation, referring to the stock of AI-related knowledge creation and inventive activity; (ii) AI adoption, referring to the introduction of AI technologies into firms, industries, or governance systems; and (iii) AI accumulation intensity, reflecting the scale and depth of AI usage in real economic and administrative processes.We further clarify that the empirical proxy used in this study—city-level AI-related patent stock—corresponds specifically to AI technology accumulation, which captures cities’innovation capacity and technological trajectory in the AI domain, rather than the actual deployment or operational use of AI technologies. This clarification has been consistently incorporated into the variable definition, empirical interpretation, and discussion of results to ensure conceptual alignment throughout the manuscript.Moreover, we have strengthened Section 5.2 (Limitations and Prospects) to explicitly acknowledge that patent-based measures reflect innovation output rather than real-world AI deployment or usage intensity. We emphasize that this distinction is crucial for interpreting the estimated effects and for appropriately framing the policy implications of our findings. Accordingly, we also suggest that future research could incorporate more direct indicators of AI adoption and application, such as firm-level AI investment, software procurement data, or cloud-based AI service utilization.Finally, we have systematically revised the terminology throughout the manuscript to ensure consistency. In particular, references to “AI accumulation” have been carefully replaced or clarified where necessary to avoid conceptual ambiguity and to fully align the empirical analysis with the actual measurement framework.We greatly appreciate the reviewer’s constructive suggestion, which has significantly improved the conceptual rigor and interpretability of the paper.

Comment 3: Third, the theoretical mechanism could be developed a little further. The argument that AI improves GTFP through green innovation and industrial upgrading is reasonable, but the paper could explain more clearly how this process unfolds. For instance, AI may help cities and firms sense environmental inefficiencies, optimise energy and resource use, process large-scale environmental data, support cleaner production decisions, and reconfigure industrial activities toward greener and higher-value sectors. Bringing in a dynamic capability or digital transformation perspective could make the mechanism more convincing, because it would show how AI-enabled sensing, learning, coordination, and reconfiguration contribute to green productivity improvement.

 

Response：

We thank the reviewer for this helpful suggestion. It encouraged us to rethink the mechanism from a more process-oriented perspective. We fully agree that while our original arguments on green innovation and industrial upgrading are reasonable, the underlying transmission process can be further strengthened by explicitly incorporating a dynamic capability and digital transformation perspective.In response, we have substantially revised Sections 2.2 and 2.3 to deepen the theoretical mechanism and improve its process-level explanation. Specifically, we now explicitly embed the dynamic capability framework to better illustrate how artificial intelligence (AI) translates into improvements in urban green total factor productivity (GTFP). From this perspective, AI is conceptualized not merely as a technological input, but as an enabling force that systematically enhances a sequence of organizational and regional capabilities within cities and firms. Firstly, AI strengthens sensing capability by enabling real-time environmental monitoring, detection of energy inefficiencies, and identification of pollution patterns through large-scale data collection and intelligent perception systems. Second, it enhances learning capability by facilitating data-driven knowledge accumulation, pattern recognition, and continuous improvement in green innovation through advanced analytics and machine learning techniques. Third, AI improves coordination capability by reducing information frictions and enabling more efficient cross-sectoral and cross-organizational allocation of capital, labor, and environmental resources across production systems. Finally, AI fosters reconfiguration capability by supporting the optimization of production processes and accelerating the structural transformation of industries toward high-value-added, low-energy-consuming, and low-emission sectors. Building on this dynamic capability logic, we further clarify how these four processes jointly contribute to green productivity improvement. Specifically, enhanced sensing reduces information asymmetry in environmental management; improved learning accelerates green technological innovation; stronger coordination increases resource allocation efficiency; and effective reconfiguration enables industrial upgrading toward greener production systems. Together, these mechanisms provide a more complete and process-oriented explanation of how AI leads to improvements in urban GTFP. We believe that this revision significantly strengthens the theoretical foundation of the paper by moving from a static mechanism description to a dynamic and capability-based framework, thereby improving both the explanatory power and internal consistency of the study.

We greatly appreciate the reviewer’s insightful suggestion, which has substantially improved the theoretical depth and rigor of the manuscript.

 

Comment 4：Fourth, the instrumental variable strategy would benefit from a clearer justification. The interaction between topographic relief and lagged AI intensity is interesting, but the exclusion restriction needs to be explained more carefully. Topographic conditions may also influence transport accessibility, industrial structure, urban expansion, environmental pressure, and regional development patterns, all of which may be related to GTFP. I suggest that the authors provide a stronger explanation of why the instrument affects current GTFP only through AI accumulation, or add further robustness checks if data availability allows.

 

Response：

We appreciate this constructive comment regarding identification. We fully agree that the validity of the exclusion restriction requires careful justification, particularly given that geographic characteristics may be associated with multiple dimensions of urban development. In response, we would like to further clarify the identification logic from a different perspective, focusing on the source of identifying variation in our empirical strategy. The instrumental variable is constructed as the interaction between time-invariant terrain ruggedness and lagged AI intensity. Importantly, this design implies that the variation used for identification does not stem from terrain itself, but from how exogenous geographic constraints shape the marginal diffusion of AI technologies over time across cities. In this context, terrain ruggedness operates as a structural constraint that affects the incremental cost and spatial feasibility of digital infrastructure deployment and AI diffusion, rather than directly determining contemporaneous changes in economic output or environmental performance. Therefore, the identifying variation primarily comes from heterogeneous diffusion dynamics of AI under different geographic constraints, rather than from direct geographic effects on GTFP. Moreover, the inclusion of city fixed effects absorbs all time-invariant geographic characteristics, including baseline terrain conditions, while year fixed effects control for macro-level shocks. As a result, the remaining variation exploited by the instrument is effectively orthogonal to static geographic development conditions and instead reflects differential AI diffusion trajectories conditioned by terrain constraints. This interpretation helps to clarify that the instrument does not rely on the direct economic implications of terrain, but on its role in shaping the pace and unevenness of AI technology diffusion, which provides the necessary variation for identifying the causal effect of AI on GTFP .We appreciate the reviewers constructive suggestion, which has helped us further clarify the identification strategy from a diffusion-based perspective.

 

Comment 5：Finally, inconsistent hypothesis labels such as H2, H2a, and H2b, repeated subsection titles in the heterogeneity analysis, and variable names should be improved. These issues are not substantive, but addressing them would improve the readability and professional presentation of the paper.

 

Response：

We thank the reviewer for carefully pointing out these presentation issues. We fully agree that although these issues are not substantive, improving the clarity of hypothesis labeling, subsection structure, and variable naming is important for enhancing the overall readability and professional presentation of the paper. In response, we have carefully revised the manuscript to ensure full consistency across hypothesis labels (including H2, H2a, and H2b), corrected repeated subsection titles in the heterogeneity analysis, and standardized variable names throughout the empirical sections and tables. These revisions help improve the internal coherence of the paper and ensure that all theoretical propositions, empirical specifications, and results are clearly and consistently presented.We greatly appreciate the reviewer’s careful reading and constructive feedback, which has contributed to improving the clarity and overall quality of the manuscript.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

1. The issue of broken references has not been adequately addressed in the revised manuscript. Multiple instances of “Error! Reference source not found.” remain in the text (e.g., Line 46, Lines 304–307, and Line 495), indicating unresolved citation or cross-reference errors. This is a serious editorial and formatting issue that affects readability and undermines the professionalism of the manuscript. The authors should carefully review the entire document and ensure that all in-text references and cross-references are correctly linked and properly displayed before publication.
2. The authors have attached a supplementary document listing the references; however, this file is not fully consistent with the reference list presented in the manuscript and still contains formatting and citation errors. Since the manuscript already includes a complete reference section, this supplementary document appears unnecessary for publication and may instead create confusion for readers and editors. The authors are recommended to remove this attachment and ensure that the reference list in the main manuscript is complete, accurate, and fully consistent.
3. The redundancy issue has been partially improved; however, noticeable repetition remains in the revised manuscript. In particular, the term “AI technological accumulation” is used excessively, and similar explanations regarding the mechanisms through which AI influences GTFP are repeated across the Introduction, theoretical framework, and mechanism analysis sections. Further streamlining is recommended to improve conciseness, readability, and the overall quality of presentation.
4. Table 1 contains apparent inconsistencies and editing errors. For example, both the Resource Dependency and Spatial Location/Transportation Accessibility variables are labeled as “transport,” which suggests incorrect variable coding or careless editing. In addition, the current formatting of Table 1 makes it difficult to read and interpret efficiently. The authors should carefully verify all variable labels, column width, row height, and improve the table formatting for clarity and consistency. All Tables are suggested to decrease the row height to increase their readability.
5. Reference [11] contains author information that appears potentially ambiguous or inconsistent (e.g., repeated author initials with the same surname, such as Liu, Y). Please verify the accuracy of the author listing and ensure that all bibliographic details are correctly presented.

Comments on the Quality of English Language

The overall English is understandable and has improved in the revised version. However, some wording remains repetitive, and minor grammatical or editorial issues are still present. Careful proofreading and language polishing are recommended to further improve readability and presentation quality.

Author Response

Comment 1: The issue of broken references has not been adequately addressed in the revised manuscript. Multiple instances of “Error! Reference source not found.” remain in the text (e.g., Line 46, Lines 304–307, and Line 495), indicating unresolved citation or cross-reference errors. This is a serious editorial and formatting issue that affects readability and undermines the professionalism of the manuscript. The authors should carefully review the entire document and ensure that all in-text references and cross-references are correctly linked and properly displayed before publication.

Response:

We sincerely thank the reviewer for pointing out this important issue. In the revised manuscript, we have carefully checked and corrected all instances of broken references. Specifically, we have reinserted and updated all cross-references in the Word document to ensure that they are properly linked and can be correctly displayed within the editable file. We have verified that all references and cross-references function normally in the source Word file. However, if the issue of “Error! Reference source not found.” still appears in the compiled PDF version, it may be due to technical errors occurring during the file conversion process, which is beyond our control. We would greatly appreciate the editorial office’s assistance in ensuring that the final PDF generation process preserves all cross-references correctly. We apologize for any inconvenience caused and have made every effort to ensure the accuracy and professionalism of the manuscript.

Comment 2: The authors have attached a supplementary document listing the references; however, this file is not fully consistent with the reference list presented in the manuscript and still contains formatting and citation errors. Since the manuscript already includes a complete reference section, this supplementary document appears unnecessary for publication and may instead create confusion for readers and editors. The authors are recommended to remove this attachment and ensure that the reference list in the main manuscript is complete, accurate, and fully consistent.

Response:

Thank you for this constructive suggestion. We have removed the unnecessary supplementary reference document from the submission system. The reference list in the main manuscript has been carefully checked to ensure it is complete, accurate, and consistent.

Comment 3: The redundancy issue has been partially improved; however, noticeable repetition remains in the revised manuscript. In particular, the term “AI technological accumulation” is used excessively, and similar explanations regarding the mechanisms through which AI influences GTFP are repeated across the Introduction, theoretical framework, and mechanism analysis sections. Further streamlining is recommended to improve conciseness, readability, and the overall quality of presentation.

Response：

We sincerely thank the reviewer for this insightful and constructive comment. In response, we have carefully revised the manuscript to further reduce redundancy and improve overall clarity and readability. First, we have systematically addressed the excessive use of the term “AI technological accumulation” throughout the manuscript. In the revised version, this term has been replaced with a more consistent and concise expression, “AI capabilities,” and all related descriptions have been carefully standardized to avoid unnecessary repetition. Second, we have substantially revised the Introduction, theoretical framework, and mechanism analysis sections to eliminate repeated explanations of how AI influences urban Green Total Factor Productivity (GTFP). In particular, we have clearly differentiated the functional roles of these sections: the Introduction now focuses on research motivation and gaps, the theoretical framework develops the underlying mechanisms in a structured manner, and the mechanism analysis section is limited to empirical testing of transmission channels without reiterating theoretical explanations. Third, redundant and overlapping descriptions of mechanism pathways have been streamlined to improve logical separation across sections and enhance narrative efficiency. This revision reduces repetition while preserving the theoretical integrity and empirical rigor of the study. Overall, these revisions significantly improve the conciseness, coherence, and presentation quality of the manuscript, and we believe they adequately address the reviewer’ s concern regarding redundancy.

Comment 4：Table 1 contains apparent inconsistencies and editing errors. For example, both the Resource Dependency and Spatial Location/Transportation Accessibility variables are labeled as “transport,” which suggests incorrect variable coding or careless editing. In addition, the current formatting of Table 1 makes it difficult to read and interpret efficiently. The authors should carefully verify all variable labels, column width, row height, and improve the table formatting for clarity and consistency. All Tables are suggested to decrease the row height to increase their readability.

Response:

Thank you for your careful and constructive comments. We sincerely apologize for the labeling inconsistency in Table 1. The duplication of the label “transport” for both Resource Dependency and Spatial Location/Transportation Accessibility was caused by an editing oversight. We have now carefully reviewed and corrected all variable labels to ensure that each variable has a unique and accurate abbreviation. In addition, following your suggestion, we have improved the formatting of Table 1 and all other tables in the manuscript by adjusting column widths, reducing row height to enhance readability, standardizing the presentation of variable names and definitions, and conducting a thorough consistency check across all tables. These revisions have significantly improved the clarity and overall presentation quality of the tables. Thank you again for your valuable suggestions, which have helped us improve the manuscript.

Comment 5: Reference [11] contains author information that appears potentially ambiguous or inconsistent (e.g., repeated author initials with the same surname, such as Liu, Y). Please verify the accuracy of the author listing and ensure that all bibliographic details are correctly presented.

Response:

We appreciate this helpful observation regarding Reference [11]. After a careful re-examination of the citation, we confirm that the apparent inconsistency in author information (e.g., repeated surname and initials such as Liu, Y) arises from the presence of different authors sharing the same name, rather than an error in bibliographic entry. To ensure accuracy, we have also systematically reviewed the entire reference list and corrected any potential inconsistencies where necessary. All references have now been carefully verified to ensure consistency and correctness. We are grateful for this suggestion, which has contributed to improving the accuracy of the manuscript.