Please refer to the enclosed PDF file.

The NHESS manuscript “Modeling Seismic Hazard and Landslide Potentials in Northwestern Yunnan, China: Exploring Complex Fault Systems with multi-segment rupturing in a Block Rotational Tectonic Zone” by Cheng et al. focuses on forecasting earthquake activity on the complex northwestern Yunnan fault system.  This paper is generally well written and logically organized.  The authors have broadened the scope of this study by also mentioning implications of their modeling results to ground-motion assessment, regional landslide hazard, and local tectonics.  These ancillary topics are treated superficially, but the core modeling methodology is well founded. However, characterization of potential ruptures needs to be broadened and better justified (see Comment 1). Major comments are included below, as well as some minor details that should be easily addressed by the authors.

Major comments:

1. The authors develop four models of multi-segment and multi-fault rupture combinations based on “the segmentation model and fault rupture behaviors”, informed largely by historical earthquake ruptures. Given the limited record of finite-rupture observations, this is prone to a great deal of bias [see Stein et al., 2012]. A more objective method is to evaluate all possible segment combinations for a given fault and establish “plausibility filters” (as suggested in Section 4.1) for multi-fault ruptures [Field et al., 2014]. Then, the results from SHERIFS can be evaluated against the historical record for verification. At minimum, more explanation is needed in Section 2.2 to firmly establish the authors’ preferred combinations and perhaps include more possibilities for multi-segment/multi-fault rupture.
2. Uncertainty analysis of the model results is not well described and perhaps incomplete. For example, it is unclear whether uncertainty in fault slip rates, which is detailed in Section 2.1, the regional MFD parameters and the M-A relations are all propagated through to the results.
3. In addition, evaluation of the model results is based on NMS ratios, rather than rigorously establishing quantitative prediction errors or goodness-of-fit metrics.
4. Description of the PGA calculation is cursory, and it is unclear whether source of uncertainty other than the GMPEs are included.
5. Similarly, uncertainty associated with the landslide hazard analysis is incomplete. See for example, Wang and Rathje [2015]. It is even unclear in this analysis what the parameters of the hazard calculation are (e.g., exposure time, probability model, etc.).

Minor comments:

(6) L20: Specify “ductile flow of the lower crust” to be clearer.

(7) L32 and throughout: “averagely” -> “on average”.

(8) L65: Is the “low velocity belt” delineated by the faults located in the lower (i.e., ductile region) or upper crust (i.e., the host rock of the faults)?

(9) L106: Unclear what the “pre-earthquake period” refers to.

(10) L108, 112: “errors”-> I think you mean “uncertainty”.

(11) L151-153: Indicate some brief description of GMPEs and site conditions used, as this is key to PGA estimates.

(12) L193: Reference Figure 2.

(13) L303 and throughout this section: “integrated”->”included” or similar.

(14) L313-316: This seems like conjecture.  Any evidence to support this inference?

(15) L336 and throughout this section: Need to distinguish the regional MFD (input to model) from the on-fault MFD (output).

(16) L357, 434: The Wells and Coppersmith (1995) relations are dated at this point. Better to use, for example Leonard [2010], or a similar recent study as an alternative to Cheng et al. (2020).  See summary by Stirling et al. [2013].

(17) L406: Shouldn’t some goodness-of-fit metric be used then?

(18) L448: There hasn’t been any explanation on how these prediction intervals are calculated.  Please include a detailed description, particularly which sources of uncertainty this pertains to.

(19) L463: “branches” of what?

(20) Figures: Font size is very small, to the point where the labels and numbers are unreadable.

References cited in review

Field, E. H., et al. (2014), Uniform California Earthquake Rupture Forecast, version 3 (UCERF3)--The time-independent model, Bull. Seismol. Soc. Am., 104, 1122-1180.

Leonard, M. (2010), Earthquake fault scaling: Self-consistent relating of rupture length, width, average displacement, and moment release, Bull. Seismol. Soc. Am., 100(5A), 1971-1988.

Stein, S., R. J. Geller, and M. Liu (2012), Why earthquake hazard maps often fail and what to do about it, Tectonophys., 562-563, 1-25, doi:https://doi.org/10.1016/j.tecto.2012.06.047.

Stirling, M., T. Goded, K. Berryman, and N. Litchfield (2013), Selection of Earthquake Scaling Relationships for Seismic‐Hazard Analysis, Bull. Seismol. Soc. Am., 103(6), 2993-3011, doi:10.1785/0120130052.

Wang, Y., and E. M. Rathje (2015), Probabilistic seismic landslide hazard maps including epistemic uncertainty, Eng. Geol., 196, 313-324, doi:https://doi.org/10.1016/j.enggeo.2015.08.001.

Review of manuscript nhess-2024-96

## General comments

In my opinion, the manuscript could be accepted after a major revision. It does not make clear what is the problem it tries to solve, and it lacks details about key elements of the methodology (i.e., the use of machine learning to calculate landslide hazard). Moreover, the discussion of the results –and essentially the manuscript itself– focuses on the seismic hazard model, while the title suggests that it is about landslide hazard too. Moreover, the documentation calculation of the landslide hazard should be improved. Moreover, given that landslide hazard modelling and the results with respect to landslide hazard are given so little coverage in the manuscript, please consider revising the title, removing from the manuscript whatever concerns landslide hazard, and focusing on seismic hazard.

Although I am neither an English native speaker nor an English language professional, I believe I have found more than a few instances, where the writing should be improved. Therefore, the manuscript does not meet editorial standards, in my opinion. Please consider having the manuscript edited by an English language professional.

As far as the figures are concerned, which have been published elsewhere and are included in the manuscript as they are or after some modification, please make sure that the reproduction rights have been secured, and inform the editor, or please consider removing them.

## Abstract

Please considering stating clearly what is the main topic of the paper. It is not clear what is the problem that this paper tries to solve. It states that it presents a new probabilistic seismic hazard assessment model that accounts for multi-segment faults, and that it uses this new model to do landslide hazard assessment. As suggested in line 28-29, the new seismic hazard model makes better predictions of some ground motion intensity measures, which may lead to a better assessment of landslide hazard. Moreover, please consider finishing the abstract with a statement about the implications of the findings.

- Line 1, “Potentials”: please consider replacing with “hazard”.

- Lines 21-23: Please clarify why the abstract mentions this historical earthquake.

- Line 24, “presented”: incorrect tense. Please replace with “presents”.

- Line 24, “a novel seismic hazard modeling study”: please replace with to “a new probabilistic seismic hazard model”

- Line 25, “integrating fault slip parameters and assessing multi-segment rupturing risks”: Please explain why is this being done by this paper. What is the necessity? A classical PSHA would not do?

- Line 28-29, “emerges as”: Please consider replacing with “is proposed as”.

- Line 28-29, “most suitable”: Please explain by which criteria and for which use.

- Line 28-29, “supported by the alignment”: Please consider replacing with: “as suggested by the agreement”.

- Line 30, “demonstrated”: incorrect tense. Please replace with “demonstrates”.

- Line 30, “peak ground-motion acceleration (PGA) values, calculated with a 475-year

31 return period from modeled seismicity rates, exhibited”: incorrect terminology. Please consider replacing with: “the peak ground acceleration for a mean return period of 475 years, which is calculated with the developed probabilistic seismic hazard model, has”

- Line 32, “fault distribution”: Please clarify if the manuscript refers to the spatial distribution of the faults. 

- “than the China Seismic Ground Motion Parameters Zonation Map”: please consider revising replacing with “than the PGA given by the Chinese seismic ground motion parameters zonation map.”

- Line 33: Please give a one-sentence description of the simulations.

- Line 34, “across”: Please consider replacing with “as a function of”.

- Line 37, “highlighted”: incorrect tense. Please replace with “highlights”.

1. Introduction

Please justify why this study only uses the peak ground acceleration. Please state what are the ground motion intensity measures used in the literature for landslide hazard and for vulnerability to landslides.

Line 156: The use of a machine learning model is suddenly mentioned here. Please consider mentioning it in the title, and in the abstract. Please justify the use of machine learning and consider adding comparisons of this calculation using machine learning with classical methods or cite a reference that validated this method.

Line 158-159: “disaster preparedness… in the area”. Indeed, this study may help in this direction, but please consider mentioning in the abstract and in the opening of the introduction that this is also part of the context of this study.

 3. Multi-segment rupture hazard Modeling

Line 335: Please consider describing what is the state of the art in probabilistic seismic hazard assessment, and then explain why accounting for the slip rate would be an improvement

It is not uncommon to take into account the characteristic earthquake in seismic hazard models. It is not clear why this paragraph mentions this in its opening.

Please explain why the manuscript focuses on the the estimation of the PGA with a 10% probability of exceedance in 50 years. Please consider discussing the PGA for other annual probabilities of exceedance, and other intensity measures. If the national hazard map is only in terms of PGA for 475 years, please consider comparing the other intensity measures with other hazard models.

Line 581: The reader may have questions about this method, but its description is missing. The machine learning model is trained using scenarios which include the PGA as an entry parameter. However the footprint of the PGA in a scenario is different from a map of the PGA for a specific return period. Moreover, please state if the landslide hazard calculation accounts for all (or a very wide range) of annual probabilities of exceedance of the PGA (or for a very wide range of return periods), and not just the PGA for 475 years. If it does not, please explain why.

## Conclusions

Line 671: Please add section title for the conclusions.

Line 671: The opening statement claims that the manuscript has given insights. This sentence seems out of place, because the manuscript first needs to briefly state the insights, then explain their importance, and then claim that it made valuable insights. Please consider dedicating the biggest part of the conclusions to the importance of the findings.

Lines 687-693: In my opinion, this is rather vague. Please consider making precise recommendations for future research.