Overall comments:

The aim of this study is to investigate potential sources of European winter windstorm predictability. The overall methodology appears to make sense, but I found the article hard to read (it really needs a thorough proof-read as there are many poorly-worded or grammatically incorrect sentences; in some cases to the extent that it wasn’t really clear what you mean). In particular, I didn’t fully understand the “process-based view” method, so please could you review the description of this to make sure it is clear.

Some of the figures need a bit of work to make them clearer – in particular I found the schematics in Fig 1 and 2 confusing (and also some of the text on them was too small and hard to read – not sure if this was in the conversion to pdf or an issue with the actual figures).

Specific comments:

Abstract: The second paragraph of the abstract seems rather vague. Please could this be rewritten – perhaps if you look at the conclusions section and summarise the key results in a more solid way.

L38: what do you mean by “signal” here?

L 50: can you say what the equivalent-potential temperature actually is (either here or later) as I, and presumably other readers, am not familiar with this quantity.

L105: do you think your method will detect sting-jet storms? These storms have some of the strongest and most damaging surface winds (e.g. the Great storm of 1987) but are relatively short-lived and affect a relatively small area compared with other types of storms where the strong winds are due to e.g. the cold-conveyor belt winds. I wonder if your tracking method would capture these events and if so, whether they might have different dynamical drivers than other types of storms. This may be beyond the scope of your study but definitely worth considering/mentioning.

L122 check reference format, should be in brackets

Throughout: you seem to use “i.a.” instead of “i.e.” or “e.g.”

Figure 1: Caption should be “schematic” rather than “scheme”. Some of the text within the figure is very small.

Figure 1: In general, I didn’t really understand this schematic and I think more description is needed in the figure caption and/or text. In particular:

• What do the different coloured boxes mean? In the text it says “The coloured boxes indicate in which physical view (Quasi-geostrophic Omega- and PV-theory) these factors are included” but it doesn’t say anywhere what each individual colour corresponds to (this should be in the figure caption).
• What are your definitions of “cyclone” and “windstorm” that means these are considered as separate things? And for instance, why is tropical precipitation labelled as a “source of predictability” for windstorm but not for cyclone?
• What is the PV 350K bandpassfilter and how is it different from 350K?

 Figure 2: Text is small/fuzzy and the text on the diagonal lines is squashed and hard to read. There is some description of this figure in the text but it wasn’t clear to me how some of the text corresponded to what is in the schematic and there are many aspects of the schematic which are not explained. For example what are the small rain-cloud shapes which appear to be at the surface? I appreciate that there are lots of complex processes involved and it would be useful to summarise them in some sort of schematic, but this particular schematic does not seem to show them very clearly and requires some work.

L140: Do you do this for each ensemble member separately? So for each ensemble member you do the 10-3-10 classification, which means that when you then look at composites they are based on different sets of years for the different members and for the observations? Could you make this clearer? Also if this is the correct interpretation, could you comment on how much similarity/difference there is between the ensemble members and the obs: do the same years get picked out by most of the ensemble members as high/low storm count years? Are they consistent with the obs? This could have implications for the interpretation of the results in terms of the influencing factors.

L140: what do you mean by “at least a decade long duration”? A decade is a period of 10 years, but I think you mean 10 years of data?

L144: is this the total absolute difference of individual members or of the ensemble mean?

L150 Please can you give  a brief description of what the ranked Tb-Kendall correlation is

L151: “chapter” should be “section”

L165: This is strangely written: what do you mean by “represented in the model as derived from reanalysis”? Do you mean something like “Does the model represent the same physical connections between causal factors as the reanalysis”?

Figure 3: It’s not clear to me what has been plotted here. Is it the respective quantity in the strong storm seasons minus the weak storm seasons, at each grid point? Why is the magnitude in the GloSea5 mean so much smaller than in ERA5?

Figure 4: I found this figure very confusing! Please add more description in the figure caption. Where you say “connection” I think you mean “correlation” and where you say 1^st/2^nd column you mean row not column. The bar graphs don’t have a scale so I’m not sure how useful they really are. And I thought the colours corresponded to correlation values but then the numbers in the boxes don’t seem to correspond. I’m also a bit confused by the interpretation of the figure and the conclusion you draw from it that the connection between the factors and windstorms is well represented in ensemble members as well as ensemble mean (L204/5).

L213: do you mean upstream of the British Isles rather than downstream?

Figure 6:  I didn’t find it clear what this is showing, please can you explain more clearly in the caption or text. My understanding: you’re showing the Kendall correlation difference in windstorm forecast at each grid point, between seasons in which the various factors or processes are well- and poorly-forecast. And the boxes show the regions that you’re interested in for the factors/processes. But how do you combine the boxes to determine if the season is well or poorly forecast? What if it’s well forecast in one box region but not another?

Figure 6: the dots and triangles are too small to be able to differentiate between them. And  it’s also not clear to me what they mean. What does it mean for a well predicted year or a badly predicted year to be significant?

L255 (and similar description/interpretation of results): does your definition of well forecast or badly forecast necessarily mean that the factor is well forecast in all the boxes? Where you have several boxes could the quantity be very well forecast in just one or two of the boxes but not in the others?

L317: it might also be worth saying that the skill is increased over other parts of Europe – NW Europe, SW Europe?

Review of wcd-2023-12

The seasonal forecast skill of European windstorms is investigated in this study. In particular, the dynamical factors that potentially drive the known skill in seasonal forecasts of storms are assessed. It is shown that for four key dynamical drivers of cyclones: their representation in the seasonal forecast model is similar to ERA5, the seasonal forecast of storms is correlated to the dynamical factors in various upstream regions, and well forecast storm seasons are correlated with well forecast dynamical factors. This topic is definitely of interest to the community and fits well within the scope of the journal. The methods used seem appropriate (though they are not always well explained) and the results are interesting. However, I found the paper to be very hard to follow. There are many poorly-worded sentences and poorly-described figures. I would think the manuscript would be suitable for publication after a thorough proof read and strong edit for clarity and completeness. More in depth comments are included below.

Major comments:

1. Clarity of writing.

I found much of the text hard to follow. The article would benefit from a thorough rewrite to draw out the main aims, results and implications of the study, which are currently being lost within the somewhat unclear text/structure and confusingly worded/long sentences. For example, the introduction contains several paragraphs about seasonal forecasts and the dynamics of extratropical cyclones, but they are not well linked with each other or related to the aims of the study (which are not really mentioned until the end of the introduction). I have listed some sentences that were unclear to me below, but it is not an exhaustive list and I recommend the entire manuscript be checked for clarity. Also, several sections begin with a question, which is presumably the question that the section aims to address. The questions need to be properly introduced and answered if they are to be included, it reads as draft-like in its current format.

The methods need to be more clearly explained as well. In particular, it is not clear what exactly is being shown in the Figures or how it is calculated. Most of the results show correlations but there is no information on exactly what is being correlated. The results certainly could not be reproduced with the information that is currently included.

Unclear sentences:

L7: “Following Glosea5 factors’ validation contributing to windstorms”

L23: "Windstorms in this study are thus more related to the direct impacts of a cyclonic system”. More related than what?

L59: “Hence, it is connected to cyclonic systems and can be an indicator for their strength and location over 60 the North Atlantic”.

L190: “but the time coherent link between storms and factors is also of great interest, hence a correlation analysis between the factors’ time development and windstorm frequency is used for validation”

L204: “After knowing that relevant factors are well represented in their connection to windstorms not only from an ensemble mean perspective, but also within individual ensemble members and thus representing a consistent physical development, the next step tests if these factors themselves are well predicted.”

L207: “Thus, in those regions of important connections between factors and windstorms (section 4.1) they should be well predicted to make an influence for the windstorm forecast performance.”

L306: “With mostly agreeing physical connection between windstorms and individual factors within the observational and model data these connections may enhance model forecast performance when the individual factors are well forecast themselves”.

L323: “For all four factors the model provides positive forecast skill within relevant regions, means the model performance for the individual factor is positive and well predicted seasons in these regions, supporting skilful windstorm forecasts.”

L333: “A similar scattered result is resulting for all approach steps for the SST gradients.”

L344: “which give new knowledge where the windstorms forecast skill might originate and where additional efforts, beside the also for windstorms existing signal-to-noise paradox”

2. The dynamical factors.

Much of the analysis focuses on four of the dynamical factors that are deemed most influential for cyclone development, yet there are 20 (by my count) that are included in Table 1 and Figure 1. I wonder if it is necessary to include all the factors in Table 1 and Figure 1 as you do not really mention them in the text (the coloured boxes in Figure 1 are not defined either). The schematic in Figure 2 is also not properly described.  I would recommend removing the Figures and Table and simply listing the predictors you chose to analyse in the study. If you do keep all the predictors in the manuscript then there should be a much more thorough description of what each means and how they relate to cyclone development (though I’m not sure what the point of this would be as the majority of the predictors are not included in the main text).

There is also no clear explanation on how the four included predictors are chosen (you say they “highlight the postulated link to winter storms clearly and best”). What metric is used to determine this? This information would potentially be more beneficial to show than the schematics.

3. Selection of good and bad forecasts.

I am somewhat confused on how you separate good and bad forecasts for the results presented in Figure 6. In section 3.3 it says you separate forecast years into good and bad by comparing their storm counts to that in ERA. But then in section 4.3 it says you separate them into good and bad by considering the skill of the forecast factors (though it is not clear exactly what you mean by this). I have a number of concerns about the approach regardless:

-Are you considering at all the temporal aspect of forecast skill or if the skill is actually related to wind storms? If you are just comparing the mean values of the factors in the different regions across the entire forecast then I’m not sure you can relate this purely to windstorms. For example, you might have a low value of MSLP gradient in the different regions that is well predicted and which is associated with a good prediction of a reduced number of storms. Therefore the skill may increase over the UK but not in relation to storms. (I could be misunderstanding what is shown in the plot.)

-To me, a more intuitive approach would be to consider the factor skill in the regions when a storm is identified. Then you could show that when a storm is in the forecast and the factor regions are well predicted, the storm is well predicted over the British Isles, and vice versa. You have the tracks for the storms so this should be feasible.

-Do you require the skill to be good in all the factor regions? If so, have you tested if a particular region is most important. I.e. does the forecast skill over the UK increase more if the factor is well predicted in a particular region?

-Is the difference in the left and right columns of figure 6 just that the regions used to define the good and bad forecast skill are different? But the method is the same apart from that?

-Have you tested different metrics of forecast skill? The results presented may be sensitive to the metric you use to determine forecast skill (please state clearly what this is). It would be good to try other metrics and compare results.

Minor comments:

L45: the Eady Growth Rate parameter is not itself a source and intensifying factor for extratropical cyclones. Strong baroclinicity is (i.e. what high values of the EGR parameter represent). So this sentence needs rephrasing.

L50: “These variables were also used in other studies”. Other studies about what? How were the variables used? Some additional context is needed here.

L77: “This could lead”. This seems vague and weak. You could say something like “the aim of this study is to better understand…”. Or something similar.

L115: Do you mean local PV? Remote PV anomalies can influence cyclones via action at a distance.

L118: Did you test if your results are sensitive to the averaging length? 3 months seems quite a long time period to average for dynamical factors relating to cyclones.

L150: Please define what is meant by tau_b Kendall correlations. How are they calculated?

Throughout: use of chapter instead of section!

Figure 3: Are there compensating errors here? I.e. do the strong storm seasons look similar in GloSea and ERA, as well as the weak storm seasons, their differences might look similar for the wrong reasons.

L187: would a dipole suggest a shift in precipitation location rather than an overall increase?

Figure 4: unclear exactly what you are correlating here? You correlate the number of storms with what metric of the dynamical factor (mean in the boxes?).

Figure 4: can you include the correlation values for GloSea as well? This would allow for easier comparison than comparing redness/blueness. The histograms are very small here as well.

Figure 5: again, it is not clear what is being correlated here.

L231: what aspect of factor skill are you referring to here? The mean value of the factor in the region? Temporal evolution? Please state explicitly.

L276: the aim of the study here should be more clearly stated in the introduction

Technical corrections:

L2: the seasonal forecast of —> seasonal forecasts of

L5: I’m not sure if ERA5 and GloSea5 should be included in the abstract without defining them. Perhaps change to “a reanalysis product and a seasonal forecast system”.

L10: What three steps?

L21: use rare or extreme. Do not need both.

L26: remove “from” before “different regions and hazards”.

L44: “investigated” —> “have investigated”.

L46: I’m not sure if i.a. is right here?

L54: Need to define theta_e before you use it.

L73: “Further on”. Further on than what? The study you refer to is from 2015 which is earlier than those mentioned previously.

L90: GloSea5 is defined earlier, though not fully?

L109: in —> is

L127: “exemplary” means very good. I do not think that is what you mean here.

L143: bad —> badly

L155: up on —> upon

L174: less strong —> stronger?

L189: is —> are

L198: outside —> upstream?

Figure 4 caption: column —> row

L206: using “Thus” to start two sentences in a row, should be changed.

L212: upstream —> downstream?

L213: downstream —> upstream?

L235: Does —> do

L235: remove “would” after storm.

L264: might be theta_e —> which might be theta_e

L265: is SST an atmospheric state?

L328: which —> who