The manuscript by Danciu et al. provides a concise and short description of the European Seismic Hazard Model 2020. The overview paper lists, describes and references the vast amount of work that build the foundation the ESHM20 and will serve as primary reference for this model, pointing to more detailed descriptions of the ingredients of the model. This manuscript is a significant contribution to improve seismic hazard assessment efforts on the regional European scale.

I recommend publication of this manuscript after minor revision. I primarily request clarifications and clearer statements of the authors opinion about some of the results. Please guide the reader with your interpretation of the results rather than let the reader free to do so.

Below I have listed my requests pointing to the lines in the manuscript.

Lines 117-118: Please make the link to the additional data more prominent.  It took me a while to figure out that I have to go to section 8 to access supplementary data. Add a link here and reference the section, a table, something that is easy for the reader.

Line 143: An example would be handy and would give real numbers for the reader: What is the overall rate change for M≥6 (or some threshold) in the model (overall). What is the change in active vs stable regions? Please add an MFD rate comparison plot similar as in Woessner et al. (2015) for the source models and the weighted combination.

Lines 154-155: Isn’t this (at least partly) mitigated by the regional adjustments of the non-ergodic GMM?

Lines 150-162: In this section, a discussion on the regional impact would be good. PGA475 increases in Romania strongly, increases are also seen in Spain. Decreases are seen in Italy and Turkey. Can you point out spatially (by tectonics, by country) the main drivers of change: Is it the rate model and if so which part of the rate model?  Is it the weighting or the models?  Is it the ground motion model, which decision therein?  Is the ground motion model update more impactful than the update of the source model? Are there spatial differences?

Lines 212-214: Could you elaborate on these tests and the criteria used to decide on a declustering method? More precisely, was the catalogue (against which the model was evaluated) declustered as well with the same method? If no, how insightful is a comparison of a mainshock forecast with a full (foreshock, mainshock, aftershock) catalogue? Sentence is not clear on which method was used in the end. Please clarify.

Line 216: “this branch”  which branch?

Lines 264-265: “as a proxy of the spatial organization of the source models”  what does this mean?

Lines 274-276: How exactly were the three Mmax values selected? Do they correspond to specific quantiles of area?

Line 278: “lower value of the expected Mmax”  maybe remove “expected”

Lines 319-320: Are “similar predictions for well-represented scenarios” not desirable? Please clarify.

Lines 367-369: Some of the uncertainties explicitly explored in ESHM20 were “collapsed” In ESHM13 (e.g. b values). This sentence suggests that ESHM13 featured a single collapsed source logic tree branch, although in reality it had three alternative branches representing different methodologies. Please clarify.

Lines 383-385: What sensitivity studies were conducted, and what criteria were considered to decide on using the “correlated end-branches”?

Line 415-418: This statement is vague. Please outline what would be required to be able to trust seismic hazard values at APE of 1/5000 or lower if required for a certain location and give references to example studies?

Line 430-444: Selecting these six locations to illustrate the complexity of the hazard curves and its uncertainty is illustrative. Despite this being a combined effect, could you elaborate in addition to the current description whether the uncertainty of the ground motion model or the rate model is more influential and for which location? Towards the end, the hazard curve 'decay' is described. The description does not discuss the parameter that is responsible which I believe is the uncertainty of the GMPEs? Please clarify and describe more precisely.

Lines 458-460: Maybe also the fact that ESHM20 uses “correlated end-branches” which will overestimate the uncertainty, whereas ESHM13 only considers mean recurrence parameters (which underestimates uncertainty)?

Figure 5: Please add a table with values to make comparisons easier or provide in the appendix. Including a table with these values for each capital city would be handy for reference.

Line 477 - 485: For the Azores, Canaries and Iceland: The Canaries are tectonically different from the Azores and Iceland. Has there been any differentiation in approaching this in the model?

Line 485: Please add statements on whether differences for SA=0.2s and 1s are similar or different from what is seen for PGA. If not, please give examples for locations where differences occur and explain.

Lines 506 - 507: The statement that PGA hazard patterns 'differ slightly' between the area and the smoothed seismicity model is not correct. Firstly, how do you define slightly? Secondly, there are differences in pattern and levels of PGA in Norway, Sweden, the UK, Belgium and Belgium - France border, German-Czech border, Southern Germany. It is positive that there are differences, otherwise two branches would not be needed. Please note and state these differences. Clarify what is considered different with a quantifiable metric? What does this tell the reader about the level of knowledge in a certain area or is it a consequence of subjective judgement? Help the reader to better understand your interpretation of the model / model component results.

Line 594-600: The European seismic risk model is not based on exactly the same hazard model. Please highlight the major differences otherwise I think this conveys a wrong message.

Review of “The 2020 European Seismic Hazard Model: Overview and Results” by Danciu et al.

This manuscript gives an overview of an update to the European Seismic Hazard Model. This is a large collaborative effort and the authors succeeded in their goal of building a more comprehensive, updated model. I am especially impressed with their treatment and presentation of epistemic uncertainties throughout the manuscript. I understand that the model is more or less final, and that my comments are most useful as they relate to the presentation of information about the model. The following are comments and requests for clarifications at various points throughout the manuscript. One area that could use improvement is the description of the active crustal fault source model. But, overall, this is a well organized overview paper that understandably can’t cover every detail of such a complex community model.

Lines 77-79: “and it was optimized for large-scale computation of the ground shaking hazard depicted by Peak Ground Acceleration (PGA) and a pseudo-acceleration spectrum (SA) with 5% damping at fifteen spectral ordinates from 0.05s to 5s.”

I don’t understand what this means, please clarify; how was it optimized?

Figure 1: Although this information is available in the caption, it would be nice to include labels for each subpanel so the reader can quickly identify which map is for ESHM13 and ESHM20

Line 161: “methodological enchantments” -- although some may find seismic hazard models enchanting, I think this may be a typo (“enhancements”?).

Lines 208-224: I could be misunderstanding, but I find this section to be odd. It sounds as though you decided to use the Reasenberg (1985) model because it left the most events in (ie., it declustered the least). Is this because of the papers you mention, such as Marzocchi and Taroni (2014) that argue for not declustering (for purposes of determining earthquake rates; you still need to decluster for the spatial distribution)? That seems like a bad reason to use an antiquated declustering model. At least in the US NSHM23, Reaseberg got the lowest weight for spatial seismicity declustering because it did such a poor job of removing aftershocks (Field et al., 2024; doi: 10.1785/0120230120)--which, again, seems to be the reason that you seem to have chosen it? Because it did the worst job at removing aftershocks and got closest to using no declustering model at all? I know it’s too late to change, but I would advocate for using a modern declustering algorithm for the spatial component, and no declustering for the overall rate model. Regardless, you need to more explicitly state why Reasenberg was chosen.

Lines 213-214: “suggested that the cluster method (Reasenberg, 1985) and an alternative to the declustering method (Grünthal, 1985) used in ESHM20” Missing a word here? “Should be used”?

Figure 2: The green volcanic area sources in 2a are barely perceptible. I suggest a larger figure and more distinct color from the gray. Also suggest that you change “(black)” to “(black polygons)” to make it more clear.

Lines 273-283: I find this description to be lacking, it’s possible that some of the details are covered in other papers but I believe that they should at least be summarized.

1. Are all faults fully segmented in your model? Multifault ruptures are commonly observed in nature and are included in other leading regional seismic hazard models (UCERF3, NZ NSHM22, US NSHM23). If multifault ruptures are not included in your model, that should be stated and explained. If multisegment ruptures are not included in your model, that is a major issue that affects the usefulness of your model and needs to be explained. Is the 2023 Türkiye-Syria rupture in your model as a single event?
2. You mention that you use slip rates to determine your activity rate parameters; are the slip rates well fit in the final model? In other words, if you take your branch-averaged rate model and compute average slip per rupture and multiply it by the rate of each rupture, then sum those values across all fault patches, does it match the original slip rate?

Lines 364-398: I like this description of the computational challenges associated with uncorrelated logic trees, or even large correlated ones. Did you do any sensitivity tests to see how mean hazard computed with your simplified logic tree compares to the full correlated tree?

Line 409: Missing space in “84thand”

Figure 7: I recommend that you remind the reader somewhere around here of Figure 1 and the electronic supplement in order to see comparison maps with ESHM13. By the timeI got to this figure I had forgotten the earlier mention of the supplement.

Figure 9: You need to specify that these are Log10(ratio), not just ratios, both in the caption and above the color scale. I came across this figure before reading the description in the text and thought that I must be going crazy seeing ratios <1.

Lines 533-548: I think this section would be cleaner if you stuck to giving linear ratios (even if you plot the log ratios in the figure for more dynamic range), rather than switching back and forth between log and linear (and sometimes omitting the linear).

Line 620: Missing space in “ 2016)and”

Line 625: “These methods are likely to involve the use of physics-based simulations of both earthquake ruptures and/or ground shaking (refs)”

You references examples of physics-based earthquake rupture *or* ground motion simulation, but not an example of both (except for, arguably, Li et al., which is both but for a handful of ruptures and not a probabilistic calculation). I suggest referencing Milner et al. (2021; https://doi.org/10.1785/0120200216), which is the only study I’m aware of that does physics-based earthquake rupture *and* ground motion simulation in a probabilistic calculation.

Supplement:

Figures S1 and S2: which model is which in the left panel? Please label. You also need to make the map titles and axis labels larger, they are barely readable (especially for the left panel).