The manuscript aims to select representative earthquake scenarios for the communication of seismic risk and for the planning of risk mitigation actions. The authors define these scenarios as the scenarios which are most likely to cause a certain regional loss value. For this purpose, they apply a methodology that was proposed by two of the authors in a separate manuscript. This methodology is claimed to require less simulation runs than a conventional loss disaggregation, which commonly relies on a large stochastic event catalog.

The topic is interesting and within the scope of NHESS. The manuscript is well written and shows the expertise of the authors. It is also clear and generally well-structured. Yet, some parts of the manuscript would benefit from additional explanations and discussions to improve its clarity and its contribution to the field. Please see some suggestions below.

Major comments

1. The y-year loss refers to the loss value that is (on average) exceeded every y years. Importantly, it does not refer to the loss value, which (on average) occurs every y years. The identified scenario, however, relates to the most likely scenario that causes this loss value (and not an exceedance of this loss value). Therefore, the link between the return period y and the identified scenario is not straightforward. I am sure that the authors are well aware of this important difference, but the readers (and more importantly, potential users of the identified scenarios) would certainly benefit from additional explanations on this aspect.
2. To resolve the above-mentioned issue, the scenario could be identified as the one that contributes most to the losses larger than the y-year loss, ly. In other words, one would aim to find the mode of p(theta | L > ly) rather than the mode of p(theta | L = ly). For hazard disaggregation, for example, Fox et al. (2016) note that such an exceedance-based approach is preferable if one aims to establish a direct link to a ground motion with a specified return period. The results, shown in Figure 7, suggest that the authors performed loss simulations for the entire stochastic event catalog. With these simulations, it should be straightforward to perform such an exceedance-based disaggregation and to compare the resulting scenarios with the already identified ones. Such a comparison would be very valuable and, given that it does require very little additional effort, I recommend that the authors include a discussion on this aspect in the main body of the paper. The detailed results of this comparison could be shown in an appendix.    
3. I agree that scenario-based analyses are valuable for many risk mitigation actions, as well as for risk communication purposes. Yet, the manuscript would benefit from some comments on potential pitfalls related to the use of a single representative scenario. For example, emergency managers may use the estimated spatial distribution of damage and losses from a representative scenario to optimize the placement of machinery and personnel before an event. What if another event (with a different distribution of damage and losses) is almost as likely to cause the considered return period loss. Is the proposed methodology capable of identifying such alternative scenarios?
4. A large part of the manuscript focuses on the description of the scenario selection algorithm, which is quite different to the conventional loss disaggregation approach. Yet, the conclusion section contains very little – if any – information on the advantages and limitations of the proposed method (in comparison with the conventional one). I recommend that the authors try to improve the clarity of their conclusions. The summary in the discussion section is much appreciated. 

Minor comments

1. Over the past years, the term Ground Motion Model (GMM) seems to be more commonly used than the term Ground Motion Prediction Equation (GMPE). Mainly, because most of the modern empirical GMMs do no longer consist of a single equation. See also the notes of David Boore for some terminological discussion (Boore, 2020), and feel free to adapt it.
2. Lines 94-97: Do the authors have any quantitative comparison (in terms of computation time) between the two loss disaggregation methodologies? This would be interesting and certainly help to highlight the advantages of the proposed method.   

References

Boore, D. M. (2020): Thoughts on the acronyms “GMPE”, “GMPM”, and “GMM”, available at https://daveboore.com/daves_notes.html

Fox, M. J., Stafford, P. J., and Sullivan, T. J. (2016): Seismic hazard disaggregation in performance-based earthquake engineering: occurrence or exceedance?, Earthquake Engineering & Structural Dynamics

The manuscript with title “Risk-informed representative earthquake scenarios for Valparíso and Viña del Mar, Chile” has as objective setting a criterion for selecting earthquake scenarios to carry out specific studies and activities around disaster risk management and reduction. Overall, the paper is well written and structured, although the manuscript could benefit from adding additional explanations and definitions of multiple concepts and input data which are required for its application. My recommendation is to carry out a major revision before it can be accepted.

Below, I provide general and specific comments that authors may find useful.

 General comments

• My main concern is that the manuscript does not include in any section a clear explanation of why this proposed approach is better than the “classic” loss disaggregation obtained from an event loss table and used, for over 10 years, in the catastrophe risk modelling field.
• I suggest authors to review previous works on different topics covered by this manuscript and include several references that in my opinion are missing. I will provide examples of this in the specific comments section.
• For the application of this methodology, an event-based earthquake risk assessment must be carried out. However, this is never mentioned or explained in detail and the equations that show how the loss computations are performed are not explicit enough for this.
• It is not clear why in the case study, two (very) different synthetic earthquake catalogs are used. What is the benefit of doing so? Are the results at any stage combined?
• There are different statements made by the authors that are not accompanied by evidence or references. I will provide examples of this in the specific comments section.
• Some of the conclusions of the paper are contradictory, between the two case studies.
• Authors in my understanding are referring interchangeably to synthetic earthquake catalogs and stochastic event sets, whereas these are two very different representations. I suggest that the difference between them is explicitly mentioned, and also which representation is the one used in the proposed methodology.
• Case studies in 4.1 and 4.2 show an example for a single building. However, the EQ loss assessment explained with the equations in the paper is not a good approach for this type of assessment and is usually preferred only for portfolio assessments.
• A discussion about how the methodology performs in a case with multiple buildings and how the treatment of the spatial correlation may introduce changes with respect to the results of the two case studies presented.
• Some decisions/assumptions made by the authors are not very clear. As for instance, why if the two case studies are located within the same area/country, different GMMs are used for each of them?

 Specific comments

• At the abstract, I suggest changing “risk management tasks” for “risk management activities”.
• At the abstract (and the introduction), it must be explained why the mentioned activities make use of scenarios of earthquake events.
• In the abstract it says that earthquake scenarios are defined in terms of the loss exceedance. Is this referring to rates? Probabilities? If the latter, in which timeframes?
• The introduction mentions that earthquake scenarios are the starting point for detailed risk assessments. However, this is not true nowadays and even more, today it is more common to carry out a fully probabilistic and event-based EQ risk assessment, and from the results (e.g., ELT), choose events to carry out scenario analyses.
• The classic PSHA formulation by Esteva (missing reference) and Cornell, did not aim to generate synthetic earthquake catalogs or stochastic event sets (note that these two are not the same). This statement at the introduction must be revised and adjusted.
• L21: it says that the classic hazard disaggregation does not consider the losses of the affected systems. This is evident and correct since as authors mention, it has to do only with the hazard component. I suggest removing that sentence.
• L25: what is the accumulated loss? Spatially accumulated? Temporal accumulation?
• L28: please clarify if the return period mentioned is that one for the loss, and if so, it is worth highlighting that it is usually very different than the one of the event.
• L47: I suggest adding “network” after power supply.
• Authors refer along the text to seismic catalogs. A definition and comprehensive explanation of what these are, what they include, etc. is needed. On L48 for instance, it is not clear if authors are referring to historical catalogs, synthetic catalogs, or both. Only in L106 it is mentioned a “stochastic seismic catalog” which in the cat-risk modelling jargon is not common.
• Section 2 required adding a better description of event-based PSHA (plus the corresponding appropriate references). Also, I think that in this section is where the explanation between stochastic event-sets and synthetic catalogs must be included.
• L67: PGA and Sa are not inputs to assess the vulnerabilities but the losses to the exposed systems.
• Section 2 also requires adding an explanation of event-based EQ risk assessment (including the appropriate references”
• The proposed methodology seems to work well in cases where only one source is controlling the EQ hazard and risk. Some discussion about its applicability in other (more common) contexts where multiple sources contribute to the overall EQ hazard and risk levels is required.
• The explanation of Eq. 4 starting in L93 is only one way of treating the aleatory uncertainty in probabilistic risk assessments. Others (perhaps more efficient and with similar results) exist and must be mentioned and referenced.
• A map with the epicenters could accompany Table 1 for a better understanding.
• L279: is that the original or the modified G-R relationship?
• To me, it is not clear what is the purpose of using two (very) different synthetic EQ catalogs and why, if one covers the small (buildings) and larger (power network) areas, the other one is needed. Also, a comparison of the two catalogs (e.g., rates by bins) for the “common area” would be useful if authors decide to keep the two.
• 5: the size of the dots as a function of Mw is not very visible in the maps.
• L352: authors state that loss estimations are “similar” from 10^8 onwards, but the results shown in Figure 7 show a very different thing. The EP curves even overlap. Again, in this point is not clear what is the purpose and benefit of using two synthetic EQ catalogs.
• L362: a better justification of why the spread is deemed as acceptable is missing.
• The light purple color for RT 100yrs does not contrast well with the grey background in Figures 8, 10 and 11. It could be changed to other tone.
• The discussion section includes some interesting conclusions and statements that can be better understood if more evidence or explanations are included. For instance, what useful validations can be made? (L394). Why the 500 and 1000yr scenarios are larger than the expected? (L397).