Urban-Scale Chikungunya Risk Mapping in the Western Guangdong-Hong Kong-Macao Greater Bay Area Using Remote Sensing

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript developed a spatially explicit risk identification framework based on Earth observation data and population-constrained statistical calibration. It tried to identify urban-scale risk during the 2025 Chikungunya fever outbreak in the western Guangdong-Hong Kong-Macao Greater Bay Area. There are some comments.

1)Introduction, the epidemiological background of the 2025 local chikungunya outbreak in the western Greater Bay Area is overly brief and vague. Only the cumulative case number is mentioned, but no detailed spatiotemporal distribution, transmission chain, or urban-specific epidemic differences in the study area are provided, weakening the practical motivation of the research.

2)Page 3, it should clarify the specific blank of existing studies: e.g., lack of 10 m high-resolution mosquito habitat assessment, population-constrained statistical calibration of MHSI, and residual diagnosis integrated with multi-type POIs in urban agglomerations.

3)Author should distinguish the research gaps between global/regional scales and refined urban agglomeration scales.

4)The novelty is repeatedly stated in Highlights, Abstract, and Introduction, and the core innovation is not refined.

5)Section 2.3, the ecological rationality of the MHSI multiplicative model (the four factors act independently) is not explained.

6)The assumption of no spatial autocorrelation in the random forest residual correction for 10 m LST downscaling is not verified.

7)Section 3.1, the MHSI only explains 21.3% of the incidence variation, the remaining 78.7% variation is not deeply analyzed.

8)The effect size calculation (10% increase in MHSI results in 36% increase in incidence) does not show the formula derivation. Clarify this calculation and add in manuscript.

9)Abstract "This index is driven by the product of water proximity (W), temperature suitability (T), and humidity suitability (H), temperature suitability (T), and humidity suitability (H)." Authors are required to carefully proofread their manuscripts prior to submission.

Author Response

Response to Reviewer #1

Urban-Scale Chikungunya Risk Mapping in the Western Guangdong-Hong Kong-Macao Greater Bay Area Using Remote Sensing

 

Reviewer #1

We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive and detailed comments. These suggestions have been extremely helpful in improving the clarity, methodological transparency, and overall presentation of the revised paper. Below we provide a point-by-point response.

Comment 1: Introduction, the epidemiological background of the 2025 local chikungunya outbreak in the western Greater Bay Area is overly brief and vague. Only the cumulative case number is mentioned, but no detailed spatiotemporal distribution, transmission chain, or urban-specific epidemic differences in the study area are provided, weakening the practical motivation of the research.

Response 1: Thank you for this important comment. We agree that the original Introduction did not provide sufficient outbreak context. In the revised manuscript, we substantially expanded the epidemiological background of the 2025 Guangdong outbreak and added a more explicit description of its spatiotemporal progression, likely transmission context, and inter-city heterogeneity within the western Greater Bay Area.

Specifically, the Introduction now describes that the outbreak did not emerge synchronously across Guangdong, but instead showed a clear progression from early intense transmission in Foshan to subsequent spillover toward neighbouring western Greater Bay Area cities. We added the concentration of early cases in Shunde District and in highly connected towns such as Lecong, Beijiao, and Chencun; the subsequent emergence of cases in Zhongshan, Jiangmen, and Zhuhai; and the later divergence between Foshan and Jiangmen in cumulative burden. We also clarified that the event was characterized as a locally transmitted outbreak initiated by imported infection, and used this context to motivate why urban continuity, inter-city commuting, and reporting heterogeneity are epidemiologically relevant in our study area.

These revisions were made to strengthen the practical motivation for urban-scale risk mapping and to make clear that the 2025 outbreak was spatially clustered and epidemiologically differentiated across the study area, rather than a uniform regional event.

Comment 2: Page 3, it should clarify the specific blank of existing studies: e.g., lack of 10 m high-resolution mosquito habitat assessment, population-constrained statistical calibration of MHSI, and residual diagnosis integrated with multi-type POIs in urban agglomerations.

Response 2: Thank you for this suggestion. We agree that the original manuscript did not state the specific research gap clearly enough. In the revised Introduction, we now explicitly identify the three main gaps addressed by this study.

First, we clarify that existing global and regional chikungunya risk models are generally too coarse to capture neighbourhood-scale heterogeneity within densely connected urban agglomerations. Second, we state that even within urban remote-sensing studies, few frameworks combine 10 m habitat-related indicators with empirically derived, outbreak-specific weighting. Third, we note that environmental suitability alone cannot fully explain urban transmission patterns, yet residual spatial mismatch is rarely interpreted together with human-process proxies such as transport connectivity, residential activity spaces, outdoor exposure environments, and healthcare accessibility.

We then directly link these gaps to the revised study design, namely the development of a reproducible 10 m MHSI workflow, empirical weight estimation from observed 2025 sub-city case locations relative to background samples, and residual interpretation using POI-based human-process proxies. This revision makes the contribution of the study more explicit and better aligned with the reviewer’s concern.

Comment 3: Author should distinguish the research gaps between global/regional scales and refined urban agglomeration scales.

Response 3: Thank you. We fully agree and have revised the Introduction to distinguish more clearly between the roles and limitations of broad-scale versus fine-scale frameworks.

In the revised text, global and regional risk models are now discussed as valuable for identifying broad climatic and environmental gradients, but limited in their ability to resolve fine-scale heterogeneity within continuous built-up corridors, river-network environments, and connected township-scale urban spaces. We then contrast this with the need for refined urban-agglomeration-scale mapping during a documented local outbreak, where sub-city heterogeneity becomes central.

This distinction is carried not only in the Introduction, but also in the Discussion, where we now compare the present framework with broader transferable models and explain the trade-off between geographic transferability and local spatial detail. This makes the scale-specific positioning of the study much clearer.

Comment 4: The novelty is repeatedly stated in Highlights, Abstract, and Introduction, and the core innovation is not refined.

Response 4: Thank you for pointing this out. We agree that the original version repeated novelty claims too often and did not distill the main contribution sufficiently. In response, we streamlined and refocused the Highlights, Abstract, and Introduction.

In the revised manuscript, the core contribution is now framed more consistently around three linked elements: (1) a reproducible 10 m urban environmental suitability workflow, (2) empirical calibration of the suitability index using observed sub-city outbreak locations, and (3) residual interpretation beyond environmental suitability through human-process proxy layers and exploratory spatial analysis. Repetitive or overly broad novelty statements were removed or softened, especially where the original wording overstated the level of inference.

This revision was intended to reduce redundancy, sharpen the central contribution, and present the innovation in a more focused and disciplined way.

Comment 5: Section 2.3, the ecological rationality of the MHSI multiplicative model (the four factors act independently) is not explained.

Response 5: Thank you for this important methodological comment. We agree that the ecological rationale for the original multiplicative formulation was not adequately explained, and that the previous wording could be read as implying independence among the four components.

After reconsidering this issue, we revised the framework itself. In the revised manuscript, the MHSI is no longer presented as a multiplicative model. Instead, we now use a weighted additive formulation and explicitly explain why this structure is more appropriate for the present purpose. The revised text states that the index is intended to characterize overall environmental suitability for mosquito habitat at the urban scale rather than to represent a strictly mechanistic transmission model. It also clarifies that the four variables were chosen to capture different, but partially overlapping, dimensions of environmental support, and are not assumed to be strictly independent.

We further clarified the ecological interpretation of each component, including the use of EVI as a proxy for moist vegetation, canopy closure, and shaded microhabitats rather than as a direct measure of humidity. In this way, the revised manuscript addresses the reviewer’s concern not only by adding explanation, but by replacing the earlier formulation with one that is more defensible for the stated application.

Comment 6: The assumption of no spatial autocorrelation in the random forest residual correction for 10 m LST downscaling is not verified.

Response 6: Thank you. We agree that this point required explicit verification. In the revised manuscript, we substantially expanded the description and validation of the LST downscaling procedure.

First, we now report hold-out validation statistics for the random forest downscaling model, including R2, MAE, RMSE, and MBE, to document its predictive performance against the Landsat-based thermal reference. Second, and more importantly for this comment, we explicitly test spatial dependence in the residual field using sample-based global Moran’s I. We report Moran’s I values for both the raw 30 m residual field and the smoothed residual field used in the residual-correction step, and show that both retain positive spatial structure rather than satisfying a no-autocorrelation assumption.

We therefore revised the wording accordingly. The manuscript now states that the residual field was treated as containing both local noise and spatially structured bias, and that Gaussian kernel smoothing was used to retain the interpretable low-frequency component rather than assuming spatial independence. We also added a limitations paragraph noting that the downscaled 10 m LST should be interpreted as a statistically refined environmental proxy rather than as a physically observed independent thermal product. These revisions directly address the reviewer’s concern.

Comment 7: Section 3.1, the MHSI only explains 21.3% of the incidence variation, the remaining 78.7% variation is not deeply analyzed.

Response 7: Thank you for this insightful comment. We agree that the unexplained component required deeper interpretation. In revising the manuscript, we also reconsidered the role of the original city-level Poisson pseudo-R2 framework.

In the revised version, the study no longer relies on the previous city-level Poisson model as the central inferential basis, and the earlier emphasis on “21.3% explained variation” has been removed. Instead, we now use a revised framework in which MHSI weights are empirically estimated using observed sub-city case locations versus background samples, and the remaining spatial mismatch is interpreted through a dedicated residual-difference analysis against an external comparative risk surface.

To address the reviewer’s substantive concern, we substantially expanded the Discussion. We now analyze the unexplained spatial mismatch using multiple POI groups, descriptive overlay, and exploratory OLS/GWR models. These analyses suggest that the residual structure may plausibly be related to crowd aggregation, exposure settings, mobility and importation pathways, and healthcare accessibility or reporting configuration. At the same time, we explicitly state that these findings are exploratory and hypothesis-generating rather than causal proof. This provides a much deeper treatment of the variation not captured by environmental suitability alone.

Comment 8: The effect size calculation (10% increase in MHSI results in 36% increase in incidence) does not show the formula derivation. Clarify this calculation and add in manuscript.

Response 8: Thank you for noting this. We agree that the previous effect-size statement was insufficiently documented and, in its original presentation, could be misunderstood as stronger evidence than the model structure justified.

Because the revised manuscript no longer uses the previous city-level Poisson elasticity formulation as the central analytical result, we removed the earlier “10% increase in MHSI corresponds to a 36% increase in incidence” statement rather than retaining a potentially over-interpreted claim. The revised manuscript now avoids presenting that unsupported effect-size expression and instead focuses on the more defensible results of the revised case-background consistency analysis, the external comparative consistency analysis, and the residual-difference interpretation.

We believe this revision addresses the reviewer’s concern more appropriately than simply inserting a derivation for a statement that is no longer central to the revised framework.

Comment 9: Abstract "This index is driven by the product of water proximity (W), temperature suitability (T), and humidity suitability (H), temperature suitability (T), and humidity suitability (H)." Authors are required to carefully proofread their manuscripts prior to submission.

Response 9: Thank you for catching this error. We apologize for the duplication and for the insufficient proofreading in the previous version.

In the revised manuscript, this sentence has been corrected. More broadly, we carefully proofread the full manuscript to remove duplicated wording, improve consistency of terminology, and reduce awkward or repetitive phrasing. We also revised multiple sections for clarity and style so that the manuscript presents a more consistent academic voice throughout.

We appreciate the reviewer’s reminder and have taken this issue seriously in the revision.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study proposes and implements a spatially explicit risk identification framework based on Earth observation data and population-constrained statistical calibration to derive a Mosquito Habitat Suitability Index (MHSI). While the topic is relevant and the proposed approach has potential, the manuscript presents several substantive issues that require attention.

First, the title is confusing and should be simplified to better reflect the core objectives and methods of the study. The abstract is excessively long and difficult to follow, containing many technical details that would be more appropriate for the Methods section. I recommend substantially streamlining the abstract; detailed suggestions are provided in the attached PDF.

Throughout the manuscript, the use of acronyms should be corrected, ensuring that all abbreviations are defined at first mention and used consistently. Figures must be explicitly cited in the text, and all maps (e.g., Figure 2) should include essential cartographic elements, such as a scale bar, north arrow, and a clear explanation of legend values. This issue applies to all map figures.

In addition, equations should be cited in the text using a consistent format (e.g., “Equation (2)”), and the variables in Equation (2) need to be clearly explained. Section 3.1 should not begin with a figure; instead, the results should first be described in the text. In Figure 6, the layout and positioning of the maps should be corrected, and the apparent difference in pixel size or resolution between panels (a) and (b) must be clarified.

The Conclusions section is overly long and contains material more appropriate for the Discussion. It should be substantially reduced to focus on the main findings and implications. Additionally, several important references are missing and should be included to better situate the study within the existing literature.

There are also more substantive methodological concerns. The description of LST downscaling using Random Forest is complex but reads like a technical manual rather than a clear methodological explanation authored by the researchers. Key assumptions and limitations of the statistical downscaling approach are not sufficiently discussed. Furthermore, the validation strategy requires clearer justification: the use of a global ENM-based risk map should be better framed and explained, particularly in terms of whether it is intended as a validation dataset or a comparative reference.

For these reasons, I recommend major revisions before the manuscript can be considered for publication.

 

Comments for author File: Comments.pdf

Comments on the Quality of English Language

English language should be revised. 

Author Response

Response to Reviewer #2

Manuscript: Urban-Scale Chikungunya Risk Mapping in the Western Guangdong-Hong Kong-Macao Greater Bay Area Using Remote Sensing

We sincerely thank Reviewer #2 for the careful evaluation of our manuscript and for the constructive comments. We have revised the manuscript substantially in response to these suggestions. Below we provide a point-by-point response and indicate how each issue has been addressed in the revised version.

Comment 1. The title is confusing and should be simplified to better reflect the core objectives and methods of the study.

Response: Thank you for this helpful suggestion. We agreed that the original title was too long and placed too much emphasis on procedural details. We therefore simplified the title from “Risk Identification of Chikungunya Fever Transmission in the Western Guangdong-Hong Kong-Macao Greater Bay Area Based on Remote Sensing Data and Population Models” to “Urban-Scale Chikungunya Risk Mapping in the Western Guangdong-Hong Kong-Macao Greater Bay Area Using Remote Sensing.” The revised title more directly reflects the scope, scale, and core methodological orientation of the study.

Comment 2. The abstract is excessively long and difficult to follow, and includes many technical details that would be more appropriate for the Methods section.

Response: We appreciate this comment and have substantially streamlined the abstract. In the revised version, we removed or condensed methodological details that are more appropriate for the Methods section, including overly specific optimization and modelling descriptions. The abstract now focuses on the study objective, core data sources, principal findings, and the main implication of the work. We also revised the wording so that the external Riskmap is framed as a comparative reference rather than a direct validation dataset. Overall, the abstract has been made shorter, clearer, and more readable.

Comment 3. Throughout the manuscript, acronyms should be defined at first mention and used consistently.

Response: Thank you. We carefully checked acronym usage throughout the manuscript and standardized the first mention and subsequent use of key abbreviations, including CHIKV, MHSI, LST, NDVI, NDWI, NDBI, EVI, OLS, and GWR. We also removed inconsistent or redundant forms and ensured that the abbreviations list is consistent with the main text.

Comment 4. Figures must be explicitly cited in the text, and all maps should include essential cartographic elements such as a scale bar, north arrow, and a clear explanation of legend values.

Response: We appreciate this important observation. In the revised manuscript, we checked figure citation consistency and ensured that all figures are explicitly referred to in the main text before or together with their discussion. We also revised the map figures and captions to strengthen cartographic completeness. Specifically, map panels were checked to ensure the inclusion of essential cartographic elements where appropriate, including a north arrow, scale information, and clearer legend/value explanations in the captions and figure layouts. We agree that these elements are necessary for interpretability and reproducibility, and they have been improved accordingly.

Comment 5. Equations should be cited in the text using a consistent format, and the variables in Equation (2) need to be clearly explained.

Response: Thank you. We revised the manuscript so that equations are referred to consistently as “Equation (1)”, “Equation (2)”, and so forth. We also expanded the explanatory text surrounding the equations. In particular, for Equation (2) and the related model expressions, we now explicitly define the response term, the offset term, the regression coefficients, and the meaning of the variables involved. This revision improves the interpretability of the modelling framework for readers who are less familiar with generalized linear models and spatial statistical notation.

Comment 6. Section 3.1 should not begin with a figure; instead, the results should first be described in the text.

Response: We agree. In the revised manuscript, the opening of the Results section was reorganized so that the key findings are first stated in text, followed by the figure callout and visual interpretation. This improves the narrative flow and makes the section more consistent with standard journal presentation conventions.

Comment 7. In Figure 6, the layout and positioning of the maps should be corrected, and the apparent difference in pixel size or resolution between panels (a) and (b) must be clarified.

Response: Thank you for pointing this out. We revised the Figure 6 layout and clarified the meaning of the apparent resolution difference between panels. In the revised manuscript, we explain that panel (a) represents the locally derived MHSI at a much finer spatial resolution, whereas panel (b) is derived from a coarser global CHIK Riskmap used only as an external comparative reference. The two surfaces were spatially aligned and normalized to a common analysis framework for comparison, but they do not originate from the same native resolution. We therefore added clarifying text to avoid the misleading impression that the two panels should visually display identical pixel sizes.

Comment 8. The Conclusions section is overly long and contains material more appropriate for the Discussion. It should be substantially reduced to focus on the main findings and implications.

Response: We appreciate this suggestion and fully agreed. The Conclusions section has been substantially shortened and rewritten. Material that was interpretive, comparative, or limitation-oriented was moved to the Discussion where it belongs. The revised Conclusions now focus on three core points only: the high-resolution suitability mapping framework, the main spatial and weighting findings, and the broader implication that environmental suitability alone is insufficient to explain urban outbreak heterogeneity.

Comment 9. Several important references are missing and should be included to better situate the study within the existing literature.

Response: Thank you. We reviewed the reference list and added relevant literature to better position the study within the broader fields of vector-borne disease mapping, remote sensing based risk assessment, and urban environmental suitability modelling. In particular, we strengthened the comparative discussion with prior work on ecological niche modelling, remote-sensing-derived vector indicators, and urban landscape factor mapping. These additions help clarify both the contribution and the limitations of the present framework relative to existing studies.

Comment 10. The description of LST downscaling using Random Forest is complex and reads like a technical manual rather than a clear methodological explanation. Key assumptions and limitations are not sufficiently discussed.

Response: We appreciate this methodological comment very much. In response, we rewrote the LST downscaling subsection to improve clarity and authorship voice. The revised text now explains the rationale for choosing random forest regression, the role of the Sentinel-2 predictors, the purpose of residual correction, and the interpretation of the final 10 m LST surface in a more concise and readable way. We also expanded the discussion of assumptions and limitations. Specifically, we now state more clearly that the downscaled product is a statistically refined environmental proxy rather than an independent thermal observation, that the residual field retains spatial structure, and that the final product may be sensitive to predictor choice, sampling, smoothing radius, and validation split. These additions directly address the reviewer’s concern that the original text was too procedural and did not sufficiently discuss uncertainty.

Comment 11. The validation strategy requires clearer justification, especially the use of a global ENM-based risk map and whether it is intended as a validation dataset or a comparative reference.

Response: Thank you for highlighting this important point. We substantially revised the framing of this part of the manuscript. In the original version, the wording could indeed suggest that the global ENM-based CHIK Riskmap functioned as a conventional validation dataset. In the revised manuscript, we now explicitly describe it as an external comparative reference surface rather than as ground-truth validation data. We clarify that its value lies in assessing macro-spatial consistency between our locally derived MHSI and an independently developed large-scale risk surface, not in establishing predictive accuracy against independent local surveillance observations. We also revised the corresponding Methods, Results, and Discussion text to make this distinction explicit and to avoid overstatement.

Comment 12. Detailed suggestions are provided in the attached PDF.

Response: We thank the reviewer for the additional annotated suggestions in the PDF. We checked these comments carefully and incorporated them into the revision. In particular, we used them to further simplify the title and abstract, improve terminology consistency, refine the presentation of equations and figures, and strengthen the distinction between methodological description, comparative reference, and interpretation. These detailed editorial and structural suggestions were very helpful in improving the readability and overall presentation of the manuscript.

We hope that these revisions have satisfactorily addressed the reviewer’s concerns and have improved the clarity, rigour, and presentation of the manuscript. We are grateful for the reviewer’s thoughtful comments, which helped us substantially strengthen the paper.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript proposes a spatially explicit risk identification framework for chikungunya fever (CHIKV) transmission in the western Guangdong-Hong Kong-Macao Greater Bay Area, integrating high-resolution remote sensing environmental indicators with population models. The study is timely, given the scale of the 2025 CHIKV outbreak in Guangdong, and addresses a relevant gap in operationalising remote sensing data for urban-scale vector-borne disease risk mapping. The construction of a Mosquito Habitat Suitability Index (MHSI) at 10 m resolution using Sentinel-2 and downscaled Landsat LST data, calibrated through a Poisson regression framework with SLSQP weight optimisation, is a methodologically interesting contribution. The residual diagnostic analysis overlaying POI data to interpret human behavioural mechanisms beyond environmental suitability adds an interpretive dimension that is conceptually valuable.

However, the manuscript presents several substantial methodological, analytical, and interpretive limitations that must be addressed before publication. These include insufficient external validation, a very low number of analytical units in the Poisson regression, limited diagnostic assessment of model assumptions, inconsistent terminology, and overstatement of conclusions relative to the evidence. The writing quality also requires significant improvement, with numerous instances of unclear phrasing and apparent machine-translation artefacts. 

Major comments:

1. Insufficient sample size for Poisson regression and parameter estimation

The Poisson GLM (Equation 2) is fitted at the city level, yet the study area encompasses only a small number of cities (Table 1 shows data for approximately 5–6 cities). Fitting a model with one predictor (log(MHSI)) plus an offset using fewer than 10 observations raises serious concerns about statistical power, overfitting, and the reliability of parameter estimates. The bootstrap analysis (Section 3.2.1) confirms this directly: the 95% CI for β₁ spans from –18.97 to 14.33, encompassing zero and reversing sign, which fundamentally undermines the claimed statistical significance (p = 2.28×10⁻²⁷). The authors must reconcile the extremely narrow frequentist confidence interval from the single model fit with the wide bootstrap interval. This discrepancy suggests the p-value from the single fit is unreliable due to the small sample size. The authors should consider whether a finer spatial unit (e.g., district or sub-district) could increase analytical power, or alternatively, reframe the regression as exploratory rather than confirmatory.

2. Model validation is weak and methodologically problematic

The external validation relies on comparison with a global ecological niche model (ENM) risk surface for chikungunya (referred to as the “CHIK Riskmap” from Lim et al. 2025, Ref. 39). However, this reference surface operates at approximately 5 km resolution and was constructed using global covariates and occurrence data; it is not an independent ground-truth validation dataset. The Pearson r = 0.331 between MHSI and the reference surface is modest and, given the spatial autocorrelation inherent in both gridded surfaces, likely inflated. The authors do not report whether this correlation was tested against a spatially structured null (e.g., using modified t-tests for spatial correlation or permutation-based inference). A more rigorous validation would involve comparison with sub-city case data (e.g., at township or street level), entomological survey data (Breteau or Container indices), or temporal out-of-sample prediction.

3. MHSI formulation and the additive vs. multiplicative inconsistency

There is a critical inconsistency between the MHSI formula as presented and the underlying ecological rationale. The abstract and introduction describe a “multiplicative” relationship (line 152: “nonlinear multiplicative MHSI model”), yet Equation 1 defines MHSI = W·a + V·b + T·c + H·d, which is an additive (linear weighted sum) model. This is not a trivial distinction: a multiplicative model (MHSI = Wᵃ × Vᵇ × Tᶜ × Hᵈ) embeds the ecological logic that suitability drops to zero if any single factor is absent, whereas the additive form allows compensation across factors. The authors must clarify the actual functional form used, correct any inconsistencies, and justify the chosen specification against the stated ecological rationale. 

4. Absence of diagnostic assessment for the Poisson GLM

The manuscript reports pseudo R² and AIC but does not present any diagnostic assessment of the Poisson regression model. Specifically, there is no evaluation of overdispersion (which is common in disease count data and would invalidate standard Poisson inference), no residual diagnostic plots for the city-level model, no assessment of influential observations (which is critical given the very small sample size), and no comparison with alternative distributional assumptions (e.g., negative binomial or quasi-Poisson). 

5. Temporal mismatch between exposure window and case data

The environmental data are extracted for the exposure window of March–May 2025 (line 42), while the case data shown in Table 1 correspond to epidemiological week 23 (June 2–8, 2025). The rationale for this temporal lag is not explicitly justified. While a lag between environmental conditions and case reporting is epidemiologically expected (accounting for vector development, incubation, and reporting delay), the authors should clearly specify and justify the assumed lag structure. Furthermore, the study appears to use cumulative weekly case counts from a single reporting week, which may not adequately represent the full epidemic curve. If additional temporal data were available, this should be noted; if not, the limitation of relying on a single-week snapshot should be discussed.

6. POI-residual analysis is descriptive and lacks formal statistical support

The residual overlay analysis with POI data (Section 4, Figure 7) is presented as evidence that human behavioural factors explain residual variation. However, this analysis is entirely qualitative—no formal spatial statistics (e.g., kernel density estimation, spatial regression of residuals on POI density, geographically weighted regression) are applied to test whether POI covariates are statistically associated with residual patterns. The current interpretation risks confirmation bias. The authors should either formalise this analysis with appropriate spatial statistics or reframe the POI discussion explicitly as hypothesis-generating rather than hypothesis-testing.

7. Overstatement of conclusions relative to evidence base

Several conclusions extend beyond what the data robustly support. For example, the claim that the framework provides “actionable spatial evidence and intervention directions for refined Chikungunya fever prevention and control” (line 71) is premature given the modest model fit (pseudo R² = 0.213), the weak external validation (r = 0.331), and the absence of prospective or operational evaluation. The effect size interpretation (10% MHSI increase → 36% incidence increase) is presented without appropriate caveats about the ecological fallacy inherent in city-level inference. The conclusion that “water-related conditions are key factors shaping the spatial gradient of mosquito vector suitability” (lines 542–543) may partly reflect the choice of proxy variables (EVI as “humidity”) rather than a genuine ecological signal. The authors should moderate claims throughout to align with the descriptive and exploratory nature of the analysis.

Minor comments:

1. Abstract

• The abstract is excessively long and reads as a compressed Methods section. It should be restructured to clearly state the objective, summarise key findings, and state the main conclusion in a more accessible format.
• The abstract contains a duplication error: “temperature suitability (T), and humidity suitability (H)” is repeated in lines 50–51.
• Reporting six-decimal-place precision for weights and coefficients in the abstract (e.g., a(W)=0.203430) is inappropriate; round to two or three significant figures.

2. Introduction

• The introduction is very long and covers substantial background on Aedes biology that, while relevant, could be condensed. The transition from global CHIKV epidemiology to the specific study rationale should be tightened.
• Reference numbering appears disordered (e.g., citations jump from [12–16] to [1–11] and back). This should be checked and corrected per journal style.
• The phrase “urban microhabitats are difficult to capture by traditional monitoring and macro-climate indicators” (lines 142–143) is presented in quotation marks but is not attributed. Remove quotes or cite the source.

3. Methods

• Section 2.2.2 (“Clinical Data”): The term “PCH cases” appears in line 274 without definition. Clarify this abbreviation.
• Table 1 shows duplicate rows for Zhongshan (same week, same date range, same population). Explain or correct this.
• The description of the global ENM validation dataset (Section 2.3.3) is extremely detailed relative to the local study methodology. Consider moving some of this to supplementary material and focusing on how the validation was operationalised in the current context.
• The constraint a + b + c + d = 1 is a common identifiability constraint, but the authors do not discuss sensitivity to the chosen optimisation bounds (0.01–0.99) or initialisation values for SLSQP.

4. Results

• Section 2.2.2 (“Clinical Data”): The term “PCH cases” appears in line 274 without definition. Clarify this abbreviation.
• Table 1 shows duplicate rows for Zhongshan (same week, same date range, same population). Explain or correct this.
• The description of the global ENM validation dataset (Section 2.3.3) is extremely detailed relative to the local study methodology. Consider moving some of this to supplementary material and focusing on how the validation was operationalised in the current context.
• The constraint a + b + c + d = 1 is a common identifiability constraint, but the authors do not discuss sensitivity to the chosen optimisation bounds (0.01–0.99) or initialisation values for SLSQP.

5. Discussion

• The discussion does not adequately compare the MHSI approach with existing vector suitability indices or remote sensing-based risk mapping frameworks for arboviral diseases (e.g., Marti et al. 2020 is cited but not discussed comparatively).
• The limitation section is absent. 

Comments on the Quality of English Language

• Multiple instances of incorrect disease names: “Kenyan fever” (lines 308, 392, 582) and “Kikunganya fever” (line 304) must be corrected to “chikungunya fever.”
• The writing quality is uneven throughout and suggests heavy reliance on machine translation. A thorough English-language edit by a proficient speaker or professional editing service is strongly recommended.
• Figures 6a and 6b use pixel-coordinate axes (0–250–2000) without geographic reference. Add latitude/longitude axes or scale bars for interpretability.
• The abbreviation list (line 588–589) uses “CHIKV” to denote both “Chikungunya virus” and “Chikungunya fever” interchangeably throughout the text. Standardise usage. 

Author Response

Response to Reviewer #3

Manuscript: Urban-Scale Chikungunya Risk Mapping in the Western Guangdong-Hong Kong-Macao Greater Bay Area Using Remote Sensing

We sincerely thank the reviewer for the careful and constructive evaluation of our manuscript. We appreciate the recognition of the timeliness of the topic and the value of combining high-resolution remote sensing with outbreak-related spatial analysis. The comments have helped us substantially improve the manuscript in terms of methodological framing, interpretation, terminology, and writing quality. Below we provide a point-by-point response to each concern.

Major Comment 1. Insufficient sample size for Poisson regression and parameter estimation

Response:

We agree that the original city-level Poisson GLM was not sufficiently robust for confirmatory inference because the number of analytical units was too small. We therefore removed the city-level Poisson regression framework, including the associated SLSQP optimization based on city counts, from the revised manuscript. In its place, we redefined the empirical calibration strategy using publicly reported sub-city case locations as positive samples and randomly sampled environmental background points. The revised MHSI weights are now estimated by maximizing case-background discrimination rather than by fitting a city-level count model.

This change directly addresses the reviewer’s concern about unstable parameter estimation and the inconsistency between the original single-fit p-value and the wide bootstrap interval. In the revised manuscript, we no longer present the previous Poisson coefficient, effect-size interpretation, or bootstrap interval as confirmatory evidence. Instead, the case-based comparison is explicitly framed as an internal consistency assessment, while the overall framework is described as exploratory and outbreak-oriented rather than statistically confirmatory. We also state clearly in the Discussion that stronger epidemiological testing would require finer and more complete district-, street-, or other sub-city surveillance data in future work.

Major Comment 2. Model validation is weak and methodologically problematic

Response:

We appreciate this point and agree that the global CHIK Riskmap is not a ground-truth validation dataset. In the revised manuscript, we no longer describe the comparison as strict external validation against independent truth. Instead, we consistently refer to the Lim et al. (2025) surface as an external comparative reference used to assess macro-spatial consistency. We explicitly note that the product is model-derived, coarser in resolution than the local MHSI, and designed for a different modelling context.

To address the reviewer’s concern about spatial autocorrelation, we added spatially corrected inference for the pixel-level comparison. Beyond the ordinary Pearson correlation, we now report a modified t-test using effective sample size correction, a torus-shift spatial null model, and a block-permutation test. These analyses show that the observed positive association remains robust after explicit correction for spatial dependence. At the same time, we temper the interpretation: the comparison is now presented only as evidence of externally supported macro-spatial consistency, not as proof of predictive accuracy. We also emphasize in the Discussion that stronger validation would require independent local surveillance or entomological data, which were not available for the present study.

Major Comment 3. MHSI formulation and the additive vs. multiplicative inconsistency

Response:

We fully agree that this inconsistency required correction. In the earlier version, the text referred to a multiplicative MHSI while Equation (1) was additive. In the revised manuscript, we corrected this throughout. The final model is explicitly defined as an additive weighted composite: MHSI = aW + bV + cT + dH. All references to a “multiplicative” or “nonlinear multiplicative” MHSI have been removed from the Abstract, Introduction, Methods, Results, and Conclusions.

We also strengthened the ecological justification for the additive specification. Specifically, we now explain that the purpose of the MHSI is to represent overall environmental suitability at the urban scale rather than to encode a strictly mechanistic transmission process in which any single factor would force suitability to zero. Because urban environmental support may arise from partially overlapping dimensions, an additive structure was judged more interpretable and more appropriate for a pragmatic, outbreak-oriented composite index. We further clarify that the selected variables are not assumed to be strictly independent, but instead capture different and partially overlapping dimensions of mosquito habitat support.

Major Comment 4. Absence of diagnostic assessment for the Poisson GLM

Response:

We agree. Because the city-level Poisson GLM has been removed from the revised manuscript, the concerns regarding overdispersion, influential observations, residual diagnostics, and alternative count distributions are no longer applicable to the revised analytical framework. Rather than trying to defend a weak count model with too few analytical units, we chose to redesign the calibration strategy so that the manuscript no longer depends on unsupported distributional assumptions at the city scale.

In the revised paper, the main quantitative evaluation now consists of: (i) case-background discrimination for empirical weight estimation, (ii) internal consistency assessment using observed sub-city case locations versus random background points, (iii) comparative analysis with the external reference surface under spatial-autocorrelation correction, and (iv) exploratory OLS and GWR analyses of the residual-difference field against major POI groups. We believe this revised structure is methodologically more defensible given the available data.

Major Comment 5. Temporal mismatch between exposure window and case data

Response:

We thank the reviewer for highlighting the need to justify the lag structure more clearly. In the revised manuscript, we added an explicit explanation for the March–May 2025 environmental exposure window. We clarify that this lagged window was selected to reflect the approximate time required for mosquito development, viral extrinsic incubation, human incubation, and subsequent case detection and reporting.

We also revised the clinical-data description to make clear that the updated empirical positive samples are no longer limited to a single weekly city-level snapshot. Instead, they are derived from a manually curated public dataset of geocoded sub-city outbreak records spanning a broader reporting period during the 2025 epidemic. We nevertheless acknowledge that the temporal alignment remains approximate rather than event-specific, because the available public records do not permit a perfect reconstruction of local exposure histories. This limitation is now stated explicitly in the Methods and revisited in the Discussion.

Major Comment 6. POI-residual analysis is descriptive and lacks formal statistical support

Response:

We agree that the original overlay analysis, on its own, was descriptive and potentially vulnerable to confirmation bias. In response, we substantially revised this section. First, we now distinguish clearly between descriptive overlay and formal exploratory modelling. The visual POI overlay is explicitly described as hypothesis-generating rather than hypothesis-testing.

Second, to move beyond qualitative interpretation, we added a global OLS model and a geographically weighted regression (GWR) analysis in which residual intensity (defined from the Riskmap–MHSI difference field) is related to four major POI groups: Medical, Crowd aggregation, Mobility & importation, and Exposure scenarios. We report both the modest overall explanatory power of OLS and the stronger spatially varying associations revealed by GWR, while carefully stating that these patterns constitute exploratory spatial statistical support rather than causal proof. This revision was made specifically to address the reviewer’s concern that the original POI analysis lacked formal statistical underpinning.

Major Comment 7. Overstatement of conclusions relative to evidence base

Response:

We appreciate this criticism and have moderated the claims throughout the manuscript. Expressions implying operational readiness or definitive intervention guidance have been softened. For example, we no longer claim that the framework directly provides actionable intervention directions in a strong operational sense. Instead, we describe it as providing preliminary, externally supported spatial evidence for outbreak-oriented urban risk assessment.

Similarly, we removed the original city-level effect-size interpretation linking a 10% increase in MHSI to a 36% rise in incidence, because that inference depended on the now-removed Poisson model and risked ecological overinterpretation. In the Conclusions and Discussion, we now emphasize that the framework is a proof of concept for high-resolution environmental suitability mapping and exploratory spatial interpretation, not a causal transmission model or an operational prediction system. We also note that the stronger contribution of humidity- and water-related components should be understood under the present variable specification and not as definitive evidence of universal ecological dominance.

Minor Comment 1. Abstract

Response:

We revised the abstract substantially. It is now shorter, more readable, and less method-heavy. Redundant technical detail was removed or condensed, and the structure was tightened to emphasize the objective, main findings, and appropriately qualified conclusion. We also corrected the duplication error involving repeated mention of temperature and humidity suitability and removed excessive decimal precision in the reporting of model weights and statistics.

Minor Comment 2. Introduction

Response:

We revised the Introduction to improve focus and readability. Background material on Aedes biology was condensed, and the transition from global chikungunya epidemiology to the specific outbreak context and study rationale was made more direct. The quotation marks around the statement about urban microhabitats were removed, because it was not a direct quotation from a source.

In addition, the citation order problem identified by the reviewer was corrected in the revised manuscript so that references now appear in sequential order according to journal style.

Minor Comment 3. Methods

Response:

Several method-related editorial and technical issues were corrected. The undefined term “PCH cases” in the previous version has been removed because the city-level count-model framework was removed entirely. The duplicate Zhongshan row in the original table is also no longer relevant because the old weekly city-level case table has been replaced by a revised description of the curated sub-city case-location dataset.

We also streamlined the section describing the external reference product. Rather than reproducing an overly long technical description of the global ENM workflow, the revised manuscript now focuses on how the external Riskmap was used operationally in the current study. Finally, we added sensitivity-oriented discussion clarifying that the results of the refined temperature field and the downstream suitability surface may depend on predictor choice, training samples, smoothing radius, and other modelling settings.

Minor Comment 4. Results

Response:

The Results section was reorganized and cleaned for consistency with the revised analytical framework. The section no longer begins with a figure alone; instead, the main findings are introduced textually before the corresponding figure is interpreted. Figure references were checked for explicit citation in the text. We also clarified that differences between the MHSI and the external Riskmap partly reflect differences in scale, covariates, and modelling objectives rather than direct model error.

Moreover, the revised manuscript now reports the spatially corrected comparison with the external reference surface, which strengthens the interpretation of the revised comparative-results section.

Minor Comment 5. Discussion

Response:

We strengthened the Discussion in two main ways. First, we added a more explicit comparative discussion of how the present framework relates to existing vector-suitability and remote-sensing-based mapping approaches, including the broader synthesis represented by Marti et al. (2020) and other vector or surveillance-oriented studies. We now clarify more clearly what this framework adds, what trade-offs it makes, and what it does not attempt to replace.

Second, we added a distinct limitation-oriented discussion. This now covers the non-independence of the internal case-based assessment, the model-derived nature of the external reference surface, uncertainty propagated from the LST downscaling procedure, sensitivity to spatial aggregation and GWR bandwidth, and the continued need for finer and more complete epidemiological or entomological data.

Comments on the Quality of English Language

Response:

We carefully reviewed the manuscript language throughout and corrected the terminology errors identified by the reviewer, including all incorrect occurrences of “Kenyan fever” and “Kikunganya fever,” which were corrected to “chikungunya fever.” We also standardised acronym usage so that CHIKV refers to chikungunya virus, while the disease is referred to as chikungunya fever where appropriate.

In addition, the revised manuscript underwent comprehensive sentence-level editing to improve clarity, grammar, flow, and academic tone. We also corrected figure-related presentation issues where possible in the revised version, including clearer figure citation and improved explanation of map content and comparative panels.

We thank the reviewer again for these thoughtful and rigorous comments. We believe that the revised manuscript is substantially improved in methodological transparency, interpretive caution, and overall clarity, and we hope that the revisions satisfactorily address the concerns raised.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript has been improved.

Author Response

Response to Reviewer 1

We sincerely thank Reviewer 1 for the constructive comments and suggestions. We have carefully revised the manuscript to improve language clarity, formatting consistency, and the presentation of figures and data.

Comment 1.
The English expression could be improved.

Response:
Thank you for this helpful comment. We have carefully revised the manuscript for English expression, grammar, sentence structure, and academic style. In particular, we improved the consistency of technical terminology and acronyms, revised several long or awkward sentences, and checked punctuation and formatting throughout the manuscript. We also corrected remaining typographical and formatting issues, including inconsistent capitalization, missing punctuation, and residual formatting artifacts. These revisions have improved the overall readability and professionalism of the manuscript.

Comment 2.
Figures, tables, and data presentation could be improved.

Response:
Thank you for this valuable suggestion. We have revised the presentation of figures and data to improve clarity and reproducibility. Specifically, we reformatted figure captions to make them more complete and consistent, corrected the Figure 6 caption so that it appears as a single continuous caption, and checked the figure numbering and panel descriptions throughout the manuscript. We also strengthened the reporting of key analytical data, including the optimized AUC value for the MHSI weighting procedure, the number of background points and resampling iterations used in weight optimization, and the final GWR model parameters, including adaptive bandwidth, effective number of parameters, and degrees of freedom. These additions make the data presentation more transparent and allow readers to better assess the robustness and reproducibility of the results.

Reviewer 2 Report

Comments and Suggestions for Authors

This study proposes and implements a spatially explicit risk identification framework that integrates Earth observation data with population‑constrained statistical calibration to derive a Mosquito Habitat Suitability Index (MHSI). While the topic is relevant and the proposed approach shows potential, the manuscript previously presented several substantive issues requiring attention.

The title and abstract have been improved. However, the use of acronyms still requires correction to ensure consistency throughout the manuscript. The citation of equations has been addressed, but the authors are encouraged to verify instances where equations are repeated (e.g., Equation 1). The remaining comments and suggestions have been considered, and the manuscript has been substantially improved overall.

In light of these revisions, I recommend minor revision.

 

 

Comments on the Quality of English Language

English language was revised.

Author Response

Response to Reviewer 2

We sincerely thank Reviewer 2 for the careful re-evaluation of our revised manuscript and for the positive assessment that the title, abstract, equation citation, and overall manuscript quality have improved substantially. We are grateful for the recommendation of minor revision.

Comment 1.
The title and abstract have been improved. However, the use of acronyms still requires correction to ensure consistency throughout the manuscript.

Response:
Thank you for this helpful comment. We carefully checked all acronyms throughout the manuscript and revised inconsistent or undefined acronym usage. In particular, “points-of-interest (POI)”, “ordinary least squares (OLS)”, “geographically weighted regression (GWR)”, “remote sensing (RS)”, “Google Earth Engine (GEE)”, and “area under the receiver operating characteristic curve (AUC)” are now defined consistently at first use and listed in the Abbreviations section. We also corrected the keyword phrase from “Mosquito-borne habitat suitability index (MHSI)” to “Mosquito Habitat Suitability Index (MHSI)” for consistency.

Comment 2.
The citation of equations has been addressed, but the authors are encouraged to verify instances where equations are repeated, e.g., Equation 1.

Response:
Thank you for pointing this out. We rechecked the equation formatting and removed the redundant repeated representation of Equation 1. The text now introduces the equation only once, followed by a single displayed equation and the corresponding explanation of parameters and constraints. This avoids duplication and improves readability.

Comment 3.
The remaining comments and suggestions have been considered, and the manuscript has been substantially improved overall.

Response:
We appreciate the reviewer’s positive assessment. In this minor revision, we further checked terminology, formatting, figure captions, abbreviation consistency, and equation presentation to improve the clarity and polish of the manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for submitting the revised version of this manuscript. The authors have made substantial and commendable revisions that address the majority of the major concerns raised in the first review. The manuscript has improved considerably in terms of methodological transparency, analytical rigour, and interpretive balance. The revised title, the restructured abstract, the comprehensive epidemiological context added to the introduction, the shift from a city-level Poisson regression to a sub-city case-location framework with AUC-based weight estimation, the addition of spatially corrected inference for the external comparison, the formalisation of the POI-residual analysis through OLS and GWR, and the substantially moderated conclusions together represent a thorough and responsive revision.

Most of my major concerns from the first review have been adequately or substantially addressed. However, a small number of points require further clarification before the manuscript can be considered for publication. These are detailed below. 

1. The environmental composites cover March–May 2025, but the case locations span June–November 2025. While the lag justification for the initial phase is reasonable, it is less convincing for cases occurring four to five months after the exposure window (e.g., cases from September–November). The authors should add one or two sentences in Section 2.2.1 or in the limitations acknowledging that static composites may have reduced relevance for later-phase cases, and briefly discuss whether seasonal environmental variation (e.g., monsoon effects on NDWI and LST) could affect the MHSI-case correspondence over the full outbreak period.

2. The weight optimisation is based on maximising AUC for case-background discrimination (Section 2.3.2), yet the final AUC value achieved by the optimised weights is not explicitly reported in the Results. The Mann-Whitney U test (Section 3.2) provides a related measure, but the AUC itself—which was the objective function—should be reported alongside the optimised weights to allow readers to assess discriminatory performance directly.

3. Section 2.3.2 states that background points were randomly generated within the study area and that the optimisation was repeated across multiple resamplings (lines 379–381). However, the number of background points per run, the number of resampling iterations, and whether background points were constrained (e.g., excluding water bodies or uninhabited areas) are not specified. These details affect reproducibility and the stability of the weight estimates. Brief additional specification would strengthen this section.

4. The GWR used an adaptive bi-square kernel with automatic bandwidth selection and a fallback to a larger neighbour set if initial convergence failed (lines 436–437). It would be helpful to report the final bandwidth (number of neighbours) actually used, whether the fallback was triggered, and the effective degrees of freedom, as these affect the interpretability of the R² = 0.698. High R² values in GWR can sometimes reflect overfitting when bandwidths are narrow, and reporting these parameters would allow readers to assess this.

Comments on the Quality of English Language

Minor language and formatting issues

• Line 363: The displayed Equation 1 notation (“MHSI = W·a + V·b + T·c + H·d”) is immediately followed by the same equation in display format with different notation (“MHSI = Wa + Vb + Tc + Hd”). One representation should be removed to avoid redundancy.
• Figure 6 caption (lines 494–496): The caption appears split across two paragraphs with a line break after “Figure 6”. This should be reformatted as a single continuous caption.
• Section 2.2.3 heading (line 308): “Verified case data” is not capitalised consistently with other section headings.
• Lines 578–579: The sentence beginning “Third, following the broader synthesis of Marti et al.” lacks a full stop before it, running into the preceding sentence (“...a documented epidemic episode.Third”).
• The manuscript still has occasional instances of underlined text visible in the PDF, which may be residual tracked changes or formatting artefacts. These should be cleaned before final submission.

Author Response

Response to Reviewer 3

We sincerely thank Reviewer 3 for the thoughtful and constructive second-round review. We are grateful for the positive evaluation that the manuscript has improved considerably in methodological transparency, analytical rigour, and interpretive balance. We have carefully addressed each remaining point, as detailed below.

Major/technical comments

Comment 1.
The environmental composites cover March–May 2025, but the case locations span June–November 2025. While the lag justification for the initial phase is reasonable, it is less convincing for cases occurring four to five months after the exposure window, e.g., cases from September–November. The authors should add one or two sentences in Section 2.2.1 or in the limitations acknowledging that static composites may have reduced relevance for later-phase cases, and briefly discuss whether seasonal environmental variation, e.g., monsoon effects on NDWI and LST, could affect the MHSI-case correspondence over the full outbreak period.

Response:
Thank you for this important clarification. We agree that the March–May environmental composites are more directly relevant to the early outbreak phase and may have reduced explanatory relevance for later cases reported in September–November. We have added explicit statements in Section 2.2.1 acknowledging that this temporal alignment is an epidemiologically informed approximation rather than a strict event-specific exposure reconstruction. We also added a limitation noting that monsoon-related seasonal changes in surface moisture, NDWI, vegetation moisture, and LST may have altered local habitat suitability during the later outbreak phase and weakened the correspondence between the static MHSI surface and later case locations.

Revision made:
Section 2.2.1 now includes the following clarification:

“Because the environmental composites were constructed for March–May 2025, their relevance may be stronger for the early phase of the outbreak than for cases reported later in September–November. Seasonal changes during the monsoon period, including variation in surface moisture, NDWI, and land surface temperature, may have altered local habitat suitability and weakened the correspondence between the static MHSI surface and later-phase case locations.”

The limitation section was also revised to state:

“The use of static March–May environmental composites may also reduce explanatory relevance for later-phase cases reported in September–November, when monsoon-related changes in water availability, vegetation moisture, and LST could have altered local habitat suitability.”

Comment 2.
The weight optimisation is based on maximising AUC for case-background discrimination, yet the final AUC value achieved by the optimised weights is not explicitly reported in the Results. The Mann-Whitney U test provides a related measure, but the AUC itself—which was the objective function—should be reported alongside the optimised weights to allow readers to assess discriminatory performance directly.

Response:
Thank you for this helpful suggestion. We agree that the AUC value should be reported directly because it was the objective function used for weight optimisation. We have now added the optimized AUC value in the Results section immediately after reporting the final weights.

Revision made:
Section 3 now reports:

“The optimized weighting scheme achieved an AUC of 0.762 for case-background discrimination, indicating acceptable discriminatory performance under the present balanced sampling design.”

This addition allows readers to assess the discriminatory performance of the optimized MHSI directly.

Comment 3.
Section 2.3.2 states that background points were randomly generated within the study area and that the optimisation was repeated across multiple resamplings. However, the number of background points per run, the number of resampling iterations, and whether background points were constrained, e.g., excluding water bodies or uninhabited areas, are not specified. These details affect reproducibility and the stability of the weight estimates. Brief additional specification would strengthen this section.

Response:
Thank you for identifying this reproducibility issue. We have revised Section 2.3.2 to specify the number of background points, the number of repeated resampling iterations, the random seed, and the constraints applied to valid background pixels.

Revision made:
Section 2.3.2 now states that each optimisation run used 96 background points, matching the 96 reported case-location records and producing a balanced 1:1 case-background comparison. The procedure was repeated 100 times using a fixed random seed of 42. Background points were sampled only from valid land pixels with finite values for all four MHSI components and were screened using the same mask applied during MHSI construction, excluding pixels with NDWI > 0, NDVI < 0, or population = 0. We also clarified that these background points represent available environmental conditions rather than confirmed absence locations.

Comment 4.
The GWR used an adaptive bi-square kernel with automatic bandwidth selection and a fallback to a larger neighbour set if initial convergence failed. It would be helpful to report the final bandwidth, number of neighbours, actually used, whether the fallback was triggered, and the effective degrees of freedom, as these affect the interpretability of the R² = 0.698. High R² values in GWR can sometimes reflect overfitting when bandwidths are narrow, and reporting these parameters would allow readers to assess this.

Response:
Thank you for this valuable methodological comment. We agree that reporting the final bandwidth and model flexibility statistics is important for interpreting the GWR results and assessing potential overfitting. We have revised both the Methods and Results sections to report the final adaptive bandwidth, whether fallback was triggered, the effective number of parameters, and degrees of freedom.

Revision made:
Section 2.3.4 now states:

“In the final model, the automatically selected adaptive bandwidth was 122 grid cells, and the fallback procedure was not triggered. The effective number of parameters (ENP = 143.49), model degrees of freedom (1242.51), and residual degrees of freedom (1194.93) were also reported to support assessment of local model flexibility and potential overfitting.”

The Results section now further reports:

“The GWR model achieved an R² of 0.6977 and an adjusted R² of 0.6627, with an automatically selected adaptive bandwidth of 122 grid cells. The model retained 1,386 grid cells after filtering and had an effective number of parameters (ENP) of 143.49, model degrees of freedom of 1242.51, and residual degrees of freedom of 1194.93. The fallback procedure for bandwidth enlargement was not triggered, indicating that the initial automatic bandwidth search converged successfully.”

We also added a cautionary interpretation that, although the selected bandwidth involved a relatively broad neighbour set, the GWR results remain exploratory and sensitive to aggregation scale, bandwidth selection, and predictor structure.

Comments on English language, formatting, and presentation

Comment 5.
Line 363: The displayed Equation 1 notation “MHSI = W·a + V·b + T·c + H·d” is immediately followed by the same equation in display format with different notation “MHSI = Wa + Vb + Tc + Hd”. One representation should be removed to avoid redundancy.

Response:
Thank you. We removed the redundant inline representation and retained only one displayed version of Equation 1. The surrounding text was revised to introduce the equation without repeating it.

Comment 6.
Figure 6 caption: The caption appears split across two paragraphs with a line break after “Figure 6”. This should be reformatted as a single continuous caption.

Response:
Thank you for noting this formatting issue. We reformatted the Figure 6 caption as a single continuous caption.

Revision made:
The caption now reads:

“Figure 6. Comparison between the MHSI and the external comparative Riskmap: (a) MHSI after normalization to 0–1; (b) Riskmap after normalization to 0–1; (c) residual-difference map between the two surfaces; and (d) pixel-level Pearson correlation scatter plot.”

Comment 7.
Section 2.2.3 heading: “Verified case data” is not capitalised consistently with other section headings.

Response:
Thank you. We corrected the heading to “Verified Case Data” to ensure capitalization consistency with other section headings.

Comment 8.
Lines 578–579: The sentence beginning “Third, following the broader synthesis of Marti et al.” lacks a full stop before it, running into the preceding sentence.

Response:
Thank you for identifying this typographical error. We inserted the missing full stop and separated the sentence properly. The revised text now reads:

“Its contribution therefore lies not in long-term operational forecasting, but in achieving closer alignment between a high-resolution environmental suitability surface and the fine-scale spatial pattern of a documented epidemic episode. Third, following the broader synthesis of Marti et al. [25], …”

Comment 9.
The manuscript still has occasional instances of underlined text visible in the PDF, which may be residual tracked changes or formatting artefacts. These should be cleaned before final submission.

Response:
Thank you for this careful observation. We checked the manuscript and removed residual underlined text, tracked-change artifacts, and other formatting inconsistencies before resubmission. We also reviewed figure captions, headings, abbreviations, punctuation, and equation formatting to ensure a clean final version.