The authors investigate the different expressions of isentropic PV that govern the occurrence and role of the mistral wind using a SOM clustering approach, which reveals new aspects of the upper-level circulation impact on low level features.

The paper is beautifully written and concise, with a thorough discussion of the findings and it should be accepted for publication in Weather and Climate Dynamics. Though there are some points that in my opinion need to be clarified, mostly regarding methodological choices. I hope the authors find the following comments useful.

specific comments

• In the title, why “synoptic” and not “large-scale” drivers is chosen?
• I don’t quite understand why for the identification of the mistral days the cyclone masks are firstly applied and you did not apply the wind criterion alone. Could you please elaborate on that? Additionally, did you test the sensitivity of the results on different wind speed thresholds? (lines 118-123)
• Commonly normalization refers to the use of min value and of max-min range to bring the values between 0 and 1. I would say that the choice here is more of a rescaling. Most importantly, is this step necessary for the SOM application? If not, why this modification was applied? (lines 52-56)
• The set-up and validation of SOM method are extensively discussed, though I would like to ask if the frequently used quantization and the topographical errors were examined as well.
• I would suggest to add the resulted PV maps for the 16 clusters along with the STD maps in the Appendix A2 in order to facilitate the reader to follow the discussion, otherwise one should go back and forth between Figure 3 and A2. In any case, I think that the resulted PV maps is necessary to be provided per se.

minor comments

• In line 23, I am not sure that “filament” is the appropriate term here
• In lines 42-43, please define “altitude-crossing mechanism”
• In Figure 3, the colorbar for the PV units is missing

This analysis provides a new perspective in explaining the observed variability of Mistral events in the Mediterranean recognizing different large-scale dynamical patterns in terms of PV. This work follows an approach already tested for heavy rainfall events but not yet explored for characteristics of strong winds outbreaks. In that respect is highly original. The methodology, through the use of self-organizing map clustering, is rather innovative in this field I think.

The paper is very well written, easy to follow, and right to the point. I have few remarks and I think, after having improved on the following points,  it will be ready for publication.

Specific comments:

- In line 69, " ...(1986) designed a numerical QG experiment..", the abbreviation QG has not been introduced before

- Between lines 188 and 121. The method to identify the cyclone Era-Interim is not very clear. Could you expand on this ? And, even more. how the choice of the cyclone criteria is impacting the population of the database. I guess you should mention some sensitivity here since this is directly controlling the number of the Mistral events

- Figure2: Any comments on the odd presence of very long duration events (14 days and 15 days). Maybe you could start drawing when the frequency is above 2 to avoid spurious counting probably due to a very low threshold on the average wind (2m/s over the GOL domain).

- Line 150 - 155: Could you clarify the normalization (which I do not fully understand in this way) ? Wouldn't it have been better to work with standardized anomalies (Pv- Pv_mean)/STD ?

- In Fig3 you can improve readability by plotting geopotential with a greater contour interval (every 40m ?) . In addition the colormap for PV is missing

- In Fig5 the coastlines can be confused with MSLP isolines (same color). Have you tried with thin black coastlines to see if the readability improves?

- By the way, I would change coastline color in all maps (and maybe thin out other fields as well) to improve readibility of Fig.6 and case studies

- I wonder plotting the number of occurrences inside the colored rectangles would help the interpretation of the transition matrix in fig.7. It is not easy to appreciate changes in likelihood with this color scale. And the same in fig 8