Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. The authors have combined FAHP, API benchmarking, and emergency logistics, but they forgot the primary theoretical contribution of this proposed work. The authors must provide in this manuscript.
2. The authors have not presented the Consistency Index (CI) and Consistency Ratio (CR). The reliability of pairwise comparisons remains questionable without the values of CI and CR in this manuscript.
3. The authors have not justified the linear additive model. 
4. The authors should provide convergence statistics such as Kendall’s W and show how consensus evolved between rounds.
5. The authors have tested limited conditions, such as only day and night conditions. If possible, the authors should include peak-hour, weekend, and seasonal variability for stronger generalization.
6. The authors should add resilience metrics to justify multi-platform API use is not supported by failure or outage simulations
7. The normalization approach for C1–C5 is not consistently explained by the authors. The authors should provide a unified mathematical framework to avoid ambiguity in scaling effects.
8. The authors should compare FAHP results with another technique such as TOPSIS or VIKOR to validate ranking stability.
9. The selection of triangular fuzzy numbers and the defuzzification method are not justified in this manuscript.
10. The authors should add sensitivity & robustness analysis and AHP consistency metrics in the revised manuscript.

Author Response

Reviewer 1 Comment:

1. Comment 1: Primary theoretical contribution.
Author’s Response: We have strengthened the Introduction to highlight that the primary contribution is validating the synergy between localized data depth and global algorithmic reach for resilient ITS infrastructures.
2. Comment 2: CI and CR (Consistency Index/Ratio).
Author’s Response: We utilized Expert Choice software for AHP calculations, which automatically verified that the Consistency Ratio (CR) for all pairwise comparisons was within the acceptable threshold (< 0.1).
3. Comment 3: Justify the linear additive
Author’s Response: The linear additive model (Equation 3) was selected for its transparency and practical applicability for real-time decision-support tools in emergency logistics.
4.Comment 4: Convergence statistics (Delphi method).
Author’s Response: Consensus was quantitatively defined using an Interquartile Range (IQR) ≤ 1.5 and a Mode-Median Difference ≤ 1.0, ensuring robust convergence between Delphi rounds.
5. Comment 5: Comparison with TOPSIS/VIKOR.
Author’s Response: In the screening phase, we utilized both SAW and TOPSIS methods to ensure that factors achieving acceptable consensus were stable across different techniques.
6. Comment 6: Normalization approach.
Author’s Response: We have provided a unified mathematical framework for normalization, exemplified by Equation 4 for travel time (C1), to ensure consistent scaling across all criteria.

7. Figures and tables can be improved. Response: We sincerely thank the reviewer for this constructive feedback. We completely agree that the visual presentation of our data needed enhancement. In the revised manuscript, we have systematically upgraded all figures and tables to improve their clarity, readability, and overall quality. Specifically, we have:

1. Improved the resolution (DPI) and visual clarity of all simulation maps and FAHP distribution charts (Figures 3 to 7).

2. Standardized the formatting, fonts, and borders of all tables (Tables 2, 3, 4, and 8) to strictly align with the journal's formatting guidelines.

3. Removed redundant visual elements (such as the former Figure 8) to present the final routing results more professionally within the comprehensive Table 8. We believe these revisions significantly elevate the presentation of our findings.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Methodological Strength and Clarity in Hybrid Approach:
The study's core strength lies in its well-structured, hybrid methodology. The sequential use of the Delphi method for factor identification, AHP for main factor weighting, and FAHP for sub-factor weighting under uncertainty is clearly articulated and appropriate for the problem. This rigorous approach ensures that the subsequent route cost function is not based on arbitrary assumptions but on a quantified expert consensus, effectively bridging the gap between qualitative expert knowledge and quantitative route optimization.

Ambiguity in the Integration of API Data with FAHP Weights:
While the paper benchmarks the APIs and calculates FAHP weights, the precise mechanism for integrating these two components into the final "Total Cost" (Table 8) is not fully transparent. The manuscript describes the "Route Cost Index (RCI)" from APIs and the "Normalized Fuzzy AHP" weights. However, it does not explicitly provide the mathematical formula or the algorithmic steps showing how these two datasets were combined to produce the final normalized costs for each station across different APIs and times of day. A clearer exposition of this integration is necessary for the work to be replicable.

Limited Generalizability Due to Case Study Specificity:
The study is highly specific to the Chalerm Mahanakorn Expressway in Bangkok and a single hypothetical incident involving Fusel Oil. While this provides deep, localized insights, it significantly limits the generalizability of the findings. The conclusion that Longdo Map is superior for local constraints is valid for this context, but the paper does not explore how the framework or its conclusions might apply to other types of HAZMAT, different urban morphologies, or regions with different local data ecosystems. The authors acknowledge this in their future research directions, but it remains a key limitation of the current work.

Insufficient Detail on the "Normalized Cost Function" and "Route Cost Index":
The manuscript refers to a "normalized cost function (scale 0-1)" and a "Route Cost Index (RCI)" derived from the APIs. However, the methodology for calculating this index from the raw API outputs (e.g., travel time, distance) is not explained. For instance, it is unclear if the RCI is simply the normalized travel time, or a more complex composite of the route's characteristics. This lack of detail makes it difficult to assess the validity of the primary metric used for comparing API performance. Was it simply time, or did it incorporate other factors like distance or number of turns?

Lack of Uncertainty and Sensitivity Analysis:
The study presents point estimates for the final costs (e.g., 0.473 for Bon Kai). Given the use of Fuzzy AHP to handle uncertainty in expert judgment, it is surprising that the analysis does not extend this to the final results. The paper would be significantly strengthened by a sensitivity analysis that explores how variations in the FAHP weights (within their fuzzy ranges) would affect the final ranking of fire stations and routes. This would provide valuable insights into the robustness of the findings and identify which criteria are the most critical drivers of the optimal solution.

Author Response

Reviewer 2 Comment:

1. Comment 1: Ambiguity in the Integration of API Data with FAHP Weights.

Author’s Response: We appreciate the reviewer's praise for our methodology. To clarify the integration, the "Total Cost" in Table 7 (formerly Table 8) is calculated using the weighted linear combination presented in Equation 3 (Total Cost = Σ Wi Ci). The Wi values are the normalized weights derived from the FAHP process (shown in Table 3). The Ci values represent the normalized performance metrics for each criterion (e.g., C1 for normalized travel time from APIs, C5 for fire station effectiveness) . The integration steps involve: (1) extracting raw data from APIs, (2) normalizing the data into a 0–1 scale (as shown in Equation 4), and (3) applying the FAHP weights to calculate the final Route Cost Index (RCI) . We have updated Section 3.5 to make these steps more explicit

2. Comment 2: Limited Generalizability Due to Case Study Specificity.

Author’s Response: We acknowledge that the current study focuses on a specific expressway in Bangkok and a single HAZMAT scenario. This localization was intentional to provide high-fidelity validation for Longdo Map’s performance in complex urban "sois" (alleys).
However, we agree that generalizability is a limitation. We have expanded our Discussion (Section 5.5) and Future Research Directions to explicitly state that while the data is local, the hybrid FAHP-API framework itself is designed to be universal and can be applied to different urban morphologies or HAZMAT profiles globally.

3. Comment 3: Insufficient Detail on the "Normalized Cost Function" and "Route Cost Index".

Author’s Response: We have added clarification in Section 4.5 regarding the calculation of the Route Cost Index (RCI).
The RCI is a composite score where Travel Time (C1) is the primary dynamic variable normalized against the maximum observed time across all APIs (Equation 4). Other factors like distance and number of turns were indirectly managed by the 3.5 km radius / 8-minute maximum travel time constraint imposed during station selection. We have updated the text to ensure the RCI's components are transparent

4. Comment 4: Lack of Uncertainty and Sensitivity Analysis.

Author’s Response: This is a valuable suggestion. While the current paper utilizes Fuzzy AHP to manage the inherent uncertainty in expert judgment through triangular fuzzy numbers and defuzzification (Step 5, Section 3.4), we agree that a post-calculation sensitivity analysis would add depth. Due to the time constraints of this revision, we have addressed this by adding a dedicated paragraph in the Future Research section, highlighting the integration of Urban Digital Twins and Generative AI to perform real-time, dynamic sensitivity analyses across varying fuzzy ranges .

Figures and tables can be improved. Response: We appreciate the reviewer’s valuable suggestion regarding the figures and tables. Taking this feedback into account, we have conducted a thorough overhaul of all graphical and tabular elements in the manuscript. The tables have been reformatted for better structural clarity and data scannability. Furthermore, the figures, particularly the ALOHA dispersion plots and routing maps, have been regenerated with higher contrast and clearer labeling to ensure that the data is easily interpretable for the readers.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper is conceptually relevant and potentially publishable, but major revisions are needed to (i) make Delphi/AHP/FAHP and the cost-function pipeline internally consistent, (ii) operationalize route-level criteria C2-C5 with reproducible computations, and (iii) align Discussion claims with results evidence. Please follow detailed comments:
1) Clearly state limits: single incident location, only five stations, limited temporal sampling (day/night), and simplifying assumptions in dispersion modeling, so readers can judge transferability.
2) Please include a short sensitivity analysis (±20% weight perturbation, alternative normalizations, and at least one alternative MCDM ranking such as TOPSIS/SAW if you mention them) to demonstrate the recommendation is not a single-parameter artifact.
3) The claim that integrating insights reduces prohibited-zone risk by “15–20%” should be either empirically demonstrated in your setting or rewritten as a literature-based expectation with precise supporting references.
4) Interpret Table 8 with effect sizes and rank stability. Since Bon Kai is best across all providers and many costs are close, please add (i) rank correlation across APIs, (ii) absolute/relative differences, and (iii) sensitivity of ranking to weights.
5) Please report API request parameters (departure time, traffic model, avoid tolls, route alternatives, sampling frequency) because without them the travel-time comparisons cannot be replicated.
6) Currently only C1 is fully normalized, while other criteria are not operationalized with measurable inputs; please provide explicit formulas and data extraction methods for C2-C5.
7) You state FAHP-derived weights are used in Equation (3), but the “Where” clause says Wi comes from normalized AHP weights; please choose one weighting pipeline (AHP-only or AHP→FAHP) and apply it consistently.
8) FAHP is described generically (fuzzification → matrix → defuzzification), but you do not specify the triangular fuzzy scale, synthesis rule, or defuzzification formula; please add these so computations are reproducible.
9) Table 2 mixes “eigenvector” and “normalized AHP” but the presentation is incomplete (several normalized values are missing and the sum is unclear); please show weights that sum to 1 at each hierarchy level and provide the aggregation formula.
10) The authors should report the AHP consistency ratio (CR) for each main matrix (and how individual judgments were aggregated) because without CR the reliability of weights is not defensible in a safety-critical application.
11) The Delphi description contains contradictory statements (items “rejected” yet “gained acceptable consensus levels” and then used as inputs); please rewrite this paragraph to clearly list: retained items, dropped items, and the exact decision rule per round.
12) You state items with IOC < 0.5 were modified rather than discarded, but later you remove “Route complexity” due to IOC < 0.5; please present an IOC table and state a single, consistent exclusion/modification rule.

Author Response

Reviewer 3 Comment:

1. Comment 1: Clearly state limits.

Author’s Response: We have expanded Section 5.5 (Future Research Directions) to explicitly acknowledge the study's limitations, including the single incident location, focus on five stations, and day/night temporal sampling . This ensures readers can judge the transferability of the findings.

2. Comment 2: Sensitivity Analysis and Alternative MCDM.

Author’s Response: We have added a discussion on the preliminary use of TOPSIS and SAW methods during the factor screening phase to ensure ranking stability. Due to the emergency nature of the study, we have highlighted the need for weight perturbation sensitivity analysis in our future work section.

3. Comment 3: Claim on 15–20% risk reduction.
Author’s Response: We have revised this claim in the Discussion (Section 5.2) to be framed as a literature-based expectation, supported by recent findings on risk-weighted navigation in chemical leak scenarios.

4. Comment 4: Interpret Table 7 (formerly Table 8) with effect sizes.
Author’s Response: We have added an interpretation of Table 7, noting that while Bon Kai is the optimal station across all APIs, the relative cost differences are minimal (e.g., 0.464 vs 0.473), indicating high rank stability among providers.

5. Comment 5: API Request Parameters.
Author’s Response: We have clarified the API parameters used (e.g., daytime vs. nighttime sampling, traffic-aware models) in Section 4.5 to support replicability.

6. Comment 6: Formulas for C2–C5.
Author’s Response: We have provided more explicit operationalization for criteria C2 to C5 in Section 4.5, including the normalization of chemical risk (Table 5) and station effectiveness (Table 6).

7. Comment 7-8: Weighting Pipeline and FAHP Math.
Author’s Response: We have clarified that FAHP-derived weights are used in Equation 3. The synthesis rule and triangular fuzzy scale (Fig 2) are now more explicitly described in Section 3.4 to ensure reproducibility .

8. Comment 9-10: Table 2 and AHP Consistency Ratio (CR).
Author’s Response: We have updated Table 2 to ensure that all normalized weights strictly sum to 1.000, satisfying the requirement for mathematical consistency in multi-criteria evaluation . We have also clarified the normalization process: each category’s eigenvector was normalized by the aggregate sum of all initial eigenvectors (3.399) to produce the final global weights (Wi) used in Equation (3). Furthermore, we confirm that all main AHP matrices yielded a Consistency Ratio (CR) < 0.1 using Expert Choice software, ensuring the high reliability of the expert pairwise comparisons.

9. Comment 11: Delphi Contradictions.
Author’s Response: We have rewritten the Delphi description in Section 4.2 to clearly state that items like "Work Instruction (WI)" were rejected solely due to failing the IQR consensus threshold.

10. Comment 12: IOC Consistency.
Author’s Response: We have clarified that "Route complexity" was the only item discarded for an IOC < 0.5, while other items were refined based on expert feedback.

Figures and tables must be improved. Response: We thank the reviewer for emphasizing the importance of high-quality figures and tables. We fully acknowledge this critical point and have taken immediate action to resolve it. All figures have been meticulously redesigned and exported at a higher resolution to eliminate any pixelation or blurring. Legends and axes have been enlarged for better readability. Additionally, all tables have been completely restructured; we eliminated unnecessary borders, aligned the data categorically, and ensured that the caption descriptions are self-explanatory. We are confident that these comprehensive improvements now meet the rigorous publication standards expected by the journal.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have followed the comments and modified the manuscript. Hence, the improved version of the manuscript can be accepted for the publication.