Review Reports - Development and Validation of a Regionally Optimized Newmark Model for Coseismic Landslide Hazard Assessment in Southwest China

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the opportunity to review your manuscript titled “Development and Validation of a Regionally Optimized Newmark Model for Coseismic Landslide Displacement Prediction in Southwest China”. I have read your work with great interest and am pleased to inform you that I am recommending Acceptance with Minor Revisions. Your study addresses a well-defined and important gap in regional coseismic landslide hazard assessment by systematically recalibrating seven existing Newmark displacement models for the unique seismotectonic conditions of Southwest China. It also developed an improved model that incorporates peak ground velocity (PGV) as a key predictive parameter.

To facilitate an efficient revision process, I have attached a line-by-line annotated PDF containing detailed comments and suggested edits. The annotations identify specific locations requiring attention, provide recommended rephrasing where appropriate, and flag minor grammatical and formatting issues.

I believe the manuscript makes a meaningful contribution to the literature on coseismic landslide hazard assessment and aligns well with the scope of Sustainability.

Warm regards,

Reviewer

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This study focuses on development and validation of a regionally optimized Newmark model for coseismic landslide displacement prediction in Southwest China, and this work underscores the value of region-specific calibration and offers a refined tool for coseismic landslide hazard assessment in Southwest China. However, the article needs major revise before publish it.

The background sentence of the abstract needs to be re-structured to highlight the scientific issue. Furthermore, include some quantitative results in the abstract.
The paragraphs in the introduction section are too numerous. It is recommended to reorganize them to categorize and summarize the methods for coseismic landslide displacement prediction, and then discuss the advantages and disadvantages of each method separately, thereby highlighting the advantages of the method you have used.
The resolution of figures should be improved, for example, figure 1, 2, 10,
Line 34, you should pay attention to the symbols “.；”, and you should verify throughout the entire text for the similar issues.
In Figure 3, you should list the basis for your choice.
You need added the relevantly references in the table 2.
It is suggested that the discussion section include the existing research results for further argumentation.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

This study develops a region-specific Newmark displacement prediction model for Southwest China using 591 strong-motion records from nine earthquakes. Existing models are recalibrated, with the Yiğit (2020) model performing best. A new model incorporating a log-transformed peak ground velocity (PGV) term is proposed, achieving improved predictive performance (R² ≈ 0.92). Validation using the 2022 Luding earthquake demonstrates enhanced landslide discrimination (AUC = 0.687). The work highlights the importance of regional calibration in seismic landslide hazard assessment.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This paper evaluates seven existing empirical models for Newmark displacement using 591 strong-motion records from Southwest China. The authors identify the Yiğit2020 model as the best baseline and improve it by introducing a logarithmic PGV term, achieving a higher R² value. This topic is of significant engineering and geological importance. However, the proposed model has certain flaws, particularly regarding structural bias in the dataset and the absence of key methodological details.

According to Table 1, out of the 591 records, 278 are from the Wenchuan aftershock sequence and 216 are from the Wenchuan mainshock. This means that approximately 83.5% of the dataset originates from a single seismic event (and its aftershocks). The current 70%/30% random split for training and validation sets is highly prone to data leakage and overfitting to the Wenchuan fault mechanism. Consequently, the trained model resembles a "Wenchuan-optimized model" rather than a "Southwest China-optimized model." To avoid this misconception, alternative validation methods (such as leave-one-event-out cross-validation) should be employed to more effectively demonstrate the model's regional generalization capability.
The newly proposed model uses I_a, PGA (included in the a_c/PGA term), and PGV as predictor variables. It is well-established in earthquake engineering that I_a, PGA, and PGV are highly correlated ground motion parameters. Introducing PGV into a model that already includes Ia and PGA will trigger multicollinearity issues among the parameters, which artificially inflates the R² value and renders the regression coefficients highly unstable. The researchers must conduct diagnostic tests for multicollinearity (e.g., calculating the Variance Inflation Factor, VIF). If severe multicollinearity exists, appropriate feature selection or dimensionality reduction techniques must be discussed.
In Section 5.2 (Line 344), the authors state: "The physical and mechanical parameters of the rock and soil mass are directly taken from published values." However, these specific values are entirely missing from the manuscript. Because the calculation of the static factor of safety and critical acceleration is highly sensitive to cohesion and friction angle, a summary table detailing the specific physical and mechanical parameters assigned to each engineering geological unit must be provided to ensure the reproducibility of the calculated results.
The proposed model yields an AUC value of 0.687 for the Luding earthquake. Although this is an improvement compared to the recalibrated models (0.600–0.636), the absolute AUC value of 0.687 still indicates relatively poor discriminative ability in practical hazard risk assessment. The authors should explicitly point this out in the discussion section. The use of static, homogeneous geotechnical parameters (which ignores the spatial variability and dynamic strength degradation of the rock mass) might be the bottleneck preventing the model from achieving a higher AUC value (>0.75).
There are issues with equation numbering. Equation (8) is followed immediately by another Equation (9) (Line 249), then the merged equation is also labeled as (9) (Line 252), and the final regression equation is labeled as (10) (Line 256). Please revise and double-check all formula numbering throughout the text.
The content of Figure 14 needs modification to ensure all text is in English. Coordinate formats across different figures should be consistent. Some figures have directional annotations while others do not, which may reduce the article's readability.
The referencing is not standardized; for instance, the citation number at Line 551 is 1, which is incorrect.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 5 Report

Comments and Suggestions for Authors

This manuscript investigates the regional applicability of Newmark displacement prediction models for coseismic landslide assessment in Southwest China. Using strong-motion records from multiple earthquakes, the authors recalibrate existing models, develop a new model by incorporating PGV, and validate its performance using the Luding earthquake case. Overall, I recommend minor revision, as the manuscript is generally well structured and of potential value, but the following issues still require clarification and improvement before it can be accepted.

Although the authors compiled strong-motion records from multiple earthquakes for model development, the distribution characteristics of the key ground-motion parameters in the dataset are still not presented clearly enough. At present, it remains difficult to judge whether parameters such as PGA, PGV, Arias intensity, and epicentral distance are well distributed and sufficiently representative. Could the authors further clarify the statistical distribution of these key parameters, for example by providing corresponding plots or summary tables?
The numbering and in-text referencing of Equations (8)–(11) appear somewhat confusing, and the logical linkage among these equations is not always easy to follow. This affects the readability of the model-development process and may cause ambiguity in understanding how the final model was derived. Could the authors carefully recheck whether the equation numbering and cross-references in this section are fully consistent?
The authors used ROC curves and AUC values to validate the proposed model using the Luding earthquake case, and the reported AUC value of 0.687 is indeed higher than those of the comparison models. However, the overall discrimination performance still appears to remain at a moderate level. In this regard, could the authors further explain the possible reasons why the AUC value is still not particularly high?
In Figure 11, the color contrast among several lithological units appears insufficient, and some rock groups are displayed with very similar colors. This reduces the readability of the figure and makes it more difficult for readers to distinguish different geological units efficiently. Could the authors reconsider whether the current color scheme is sufficiently clear for publication-quality presentation?
Chinese labels can still be observed in Figure 14, which is not fully appropriate for an English-language manuscript. This affects the consistency of figure presentation and may reduce the overall polish of the paper. Could the authors check whether all figure annotations have been fully converted into English?
In the Discussion section, the authors note that pulse identification was not incorporated and that pulse-like strong ground motions may lead to significantly larger displacement responses. This is an important point, but it also raises a further question regarding the dataset used in the present study. Could the authors clarify whether pulse-like records are actually included in the dataset, and if so, whether this factor may have affected the model fitting and validation results?