Pls see attached review report. 

The manuscript studies for the first time the conditions of reference stations across 60 rock stations belonging to the broadband and accelerometric networks of the National Observatory of Athens. The analysis is based on selecting stations situated on rock and whose data availability is sufficient for a meaningful collection of recordings.

The study is relevant since no systematic previous site effects studies were conducted for the ensemble of stations under examination. Similar studies have been conducted in other contexts and at different scales by Lanzano et al. (2020, 2022) for Italy and by Pilz et al. (2020) for Europe.

I believe that this work is a very useful contribution to the scientific community and to seismological studies in Greece. I suggest accepting the publication with minor revisions.

Some comments below:

Station and data selection

1. Please explain further, on the basis of which criterion you consider the 60 stations to be installed on rock. Did you use geological and/or topographical proxies? Do any of the stations also have a geophysical survey?
2. Letters a and b are not present in Figure 1 and 2

Creation of a new strong motion dataset

The analysis of the signals for the creation of the dataset is a very important step. I realised that the authors did a very thorough job, taking advantage of already available codes. However, I suggest that this section be reorganised schematically by indicating the data processing work in steps.

For example:

- Identification of clipped records (which criteria?)

- Waveform picking

- Calculation of signal-to-noise ratio

- Identification of corner frequencies

- Etc.

I am not saying to repeat things that have been explained in other works, but to list the actions in a schematic and sequential manner. I think the work would benefit in terms of clarity. I think it would also be helpful to understand which signal analyses are done by the NGA-East code and which are not.

Do you check for double events recordings? How do you treat?

I also suggest improving figure 3 to make it more self-explaining. A legend is missing.

Transfer function

1. For the calculation of transfer functions with the horizontal-to-vertical spectral ratio method, I would pay attention to the recordings of co-located stations.

From what I understand, all recordings for stations equipped with accelerometer and velocimeter were considered for the estimation of HVSR. My experience with INGV's Italian seismic network, which has a large number of collocated stations, is that this should be done carefully. I suggest conducting a preliminary analysis by keeping the instruments separate (consider them as two different stations) to verify that the transfer function is equal. First of all, the sensors could suffer from fixed scaling (a comparison of intensity measurements of recordings of the same event is also recommended) caused by incorrect conversion constants in the station xml. In addition, Hollander et al. (2020) and Castellaro et al. showed that the behaviour could be different at specific frequency ranges (especially in high frequency) due to different station installations.

1. I think it is also useful to explain how the groupings in Figure 7 were made. Was a clustering analysis conducted? Or is it based on a visual analysis of the curves? I have the impression that the transition from one group to another may not be clearly defined and some stations may be in one group rather than another arbitrarily. Wouldn't it be useful to set a quantitative criterion to isolate stations that have a flat response?

Discussion and conclusions

I noticed that the authors never considered the high-frequency near-site attenuation parameter k0 among the parameters for identifying reference sites. Considering the great experience of the authors, do you think this parameter could have any weight in the future? For example, in the work of Morasca et al. (2023) we started from the 36 stations of the study by Lanzano et al. (2020) and restricted to 6, as reference sites for a GIT in central Italy. The selection was made on the basis of k0, identifying those with k0<0.015s.

Additional reference

Castellaro, S., Alessandrini, G., & Musinu, G. (2022). Seismic station installations and their impact on the recorded signals and derived quantities. Seismological Society of America, 93(6), 3348-3362.

Hollender, F., Roumelioti, Z., Maufroy, E., Traversa, P., & Mariscal, A. (2020). Can we trust high‐frequency content in strong‐motion database signals? Impact of housing, coupling, and installation depth of seismic sensors. Seismological Research Letters, 91(4), 2192-2205.

Morasca, P., D'Amico, M., Sgobba, S., Lanzano, G., Colavitti, L., Pacor, F., & Spallarossa, D. (2023). Empirical correlations between an FAS non-ergodic ground motion model and a GIT derived model for Central Italy. Geophysical Journal International, 233(1), 51-68.