General comments

The manuscript presents design criteria for improving the communication of earthquake hazard maps by carefully considering colour and classification schemes. They apply these considerations to redesign a map of earthquake hazard in Germany. They evaluate the effectiveness of the redesigned map with a small survey, finding the new map improves users understanding of several important concepts related to earthquake hazards. They conclude with a discussion about applying these design criteria to other hazard maps.

This topic is relevant to NHESS, the manuscript is very well written with clearly defined objectives and conclusions that are supported by the results. One general comment is the scope is relatively narrow to a single map, and it would be interesting to see broader implications of how these design criteria could improve other hazard maps (e.g., low-probability high-consequence scenarios, regions with different hazard patterns). However, the authors clearly identify their scope, and their discussion section offers some insights into how these criteria could apply to other scenarios. Overall, this is a high-quality manuscript and I recommend it for publication with some minor revisions.

Specific comments

• Scope and other applications. One topic that would be interesting to further explore is the mapping of low-probability high-consequence scenarios, which are a common challenge across all natural hazards. The introduction begins by discussing this scenario and the typical right skewness of earthquake hazard, but the remainder of the paper narrows the focus to a single hazard variable (i.e., 475-year return period). It would be interesting to return to this topic in the discussion with some insights about what types of communication goals or colour and classification criteria could be helpful to communicate low-probability high-consequence events. Line 564 briefly mentions communicating skewness and I wonder if this could be expanded.
• Capping the classification at the maximum value. While I understand how capping the classification with the maximum value shown on a map will optimize the communication goals on a single map, but I think there are limitations to this approach that are worth acknowledging. First, this method could distort the understanding of extreme hazard if the highest values always reach the highest end of the scale, because in some datasets the maximum hazard may only result in light or moderate ground shaking. If a communication goal is understanding the severity of extreme hazard, then I think there is value in having some external constraint to avoid every map from appearing to have extreme hazard. Another concern is that fitting classification schemes to individual datasets could diminish communication for users who need to compare hazard in different scenarios, such as different regions or different return periods. In this case, it would perhaps be more meaningful to combine the data from all the maps in the analysis and create a standard classification to this data using similar methods done for the single map.
• Hazard level terminology. Are there standard definitions for extreme, high, moderate, and low earthquake hazard? Are these relative terms, or does it directly relate to levels of peak ground acceleration for a set return period? It seems these terms are loosely defined in the manuscript and could be clarified. If one of the communication goals is to understand the difference between extreme and high hazard, it would help to know what this precisely means. Clarifying these terms would also help readers who are less familiar with earthquake hazards (including myself), since the only value provided for reference is the building code threshold of 0.4 m/s2. If there are objective definitions of extreme hazard, then these may also be meaningful thresholds to include in the classification. One specific place this impacts understanding is in lines 318-326 where the distinction between high and extreme hazard is used to compare the Fisher, Head-Tails, and quantile classification methods.
• Keeping track of goals and criteria. When first reading the manuscript, I was slightly confused by the various lists of goals and criteria, since there are research questions, communication goals, and two sets of criteria. The criteria were easy to track with the C1-C5 and L1-L3 labels, but it was a little more difficult to reference back to the communication goals. Also, I found the current structure slightly confusing because the communication goals were presented before the research goals, which made the objective of the study unclear when reading. I wonder if some minor restructuring, short explanations, or labelling could make this easier to follow.
• I appreciate the high quality of the figures and tables. They are clear and simple to understand and support the main arguments of the manuscript.

Technical comments

• Line 276: Perhaps be more specific than “previous three sections” since there are only two previous sections that contain several levels of subsections.
• Sect 3.3: The justification of the Yl-Or-Rd-8 colour palette is very strong in this section! It would be nice to see this colour scheme applied more broadly for all natural hazard communications.
• Line 363-365: What colour property was modified to change the dark red to brown? The argument about preserving perceptual order, uniformity, and discriminability is clear, but I am curious whether it was the hue, saturation, or lightness that was modified to make the extreme end of the scale distinct from the dark reds.

The manuscript of Schneider et al. addresses the problem of the correct representation of seismic hazard models, to give the public the right information on the hazard levels at the various sites. The problem is deeply felt by those who produce seismic hazard models and I find it is important that it is tackled in rigorous scientific terms. I’m talking as a researcher involved in seismic hazard model elaborations and in their dissemination and the reading of the manuscript was very pleasant and flowing.

The work is very well organized and reaches the result after very serious analyses, also including tests with possible users.

I believe the manuscript can be published with sligthly revisions.

Regarding the detail of the structure of the article, I find the bibliographic research on the choice of the right colors very interesting, based on studies also carried out in different disciplinary fields.

I have no critical observations to make about the manuscript, except the constant reference to the criteria defined in sections 1 and 2, which require you to scroll back and forth through the pages.

What in my opinion needs an explanation to the reader, for the work to be usable in other countries as well, is the problem of the limited number of classes and the definition of class intervals. The German model proposes very low values, if compared with those of other European nations. In Italy the map with 10% of probability of excess reaches 3 m/s2, double the maximum value in Germany. In Greece and Turkey there are higher values. If we then take into consideration estimates of 2% in 50 years or estimates of spectral accelerations, the values are much higher, even up to 20 m/s2, an order of magnitude higher. Since the authors never mention this issue, I would like to know if the authors plan to use for any output the same color palette and modify the classes, or to add more classes; in the first case it would be impossible to compare maps for different return periods or for different spectral periods and the information that there are higher values in one map than in another would be lost. I think this information will be useful for many future users of the method, myself included.

Another question is whether the 3 classes used for values lower than 0.4 m/s2 are not too many, to the detriment of a higher resolution of the higher values, for which I personally would prefer a greater number of classes. Personally, I am convinced that up to 10-12 classes are better for covering wider range of values.

Figure 3b has a different legend than those reported in the supplement as Figure S2. Maybe I misunderstood the caption of figure 3?