Review of “What determines the predictability of a Mediterranean cyclone?” by Benjamin Doiteau and coauthors

General:

In this paper, the authors examine the predictability of Mediterranean cyclones by using a large reference dataset of 2853 cyclone tracks from 21 years and ensemble reforecasts for this period. This topic is very important for both research and operational scientific communities. The work is ambitious in the way that both location and intensity errors of the cyclone tracks are systematically examined, with decomposition into sub categories of cyclones according to their season of occurrence, sub-region of the Mediterranean, intensity, propagation velocity, and deepening rate. The authors find significant predictability differences among the subcategories in terms of cyclone location and intensity at different lead times. Overall, the paper is very well structured and clearly written, with good visualizations of the data.

The manuscript will clearly provide a valuable contribution to WCD once the following concerns will be addressed. Of the major comments, the first concerns the underlying tracking methods, while the other two raise issues which are largely not addressed in the current paper, but are nonetheless key for making insightful conclusions (and are feasible given the existing datasets).

Major comments:

1. Differences in the cyclone tracking algorithm between the reference and ensemble forecast datasets:

Using non-identical tracking methods for the reference tracks and the forecasted tracks can introduce biases into the analysis. This issue should be either corrected or the differences in tracks carefully assessed for any systematic influence on the predictability metrics.

2. Prediction of cyclones per se:

In addition to the systematic investigation of the predictability of location and intensity, it is even more fundamental to understand if the mere existence of the cyclone (or track) is captured by the ensemble. By construction of the verification technique against a reference track, this aspect required a separate quantification, such as tracking the mean number of members having a cyclone at all (orange line in Fig. 5). In my view, if this aspect is developed it could greatly enhance the take-home messages from the paper. For example, one can readily decompose this metric (fraction of ensemble members with a cyclone) to the different cyclone categories.

3. Consideration of the cyclone lifetime relative to cyclogenesis/peak intensity:

Currently, the examination of the cyclone tracks predictability does not consider one of the most important aspects, which is the cyclone stage in the lifecycle, namely relative to cyclogenesis/peak time/lysis. Instead, the large variability in Fig. 5 envelops both the variability among ensemble members (for a given cyclone track point), and different cyclones, with all subcategories and at all track times (i.e., the time relative to genesis/peak etc.). It can be insightful to look into this variability, and see how Figs. 5 and 6 differ when decomposing the analysis to different track times. I understand that, by construction, the genesis point is more predictable when forced to start from a vicinity of the reference, but I hope the authors can think of a way to still address this aspect.

Minor comments:

Line 1 and 20: use of “beneficial” is unclear here

Line 4: “characterize cyclone predictability”: mention the time scale / range you consider here

Line 22: add “negative” before “impacts”

Paragraph ending in line 31: the paragraph currently misses a description of heat lows over land.

Line 47: mention that Baumgart et al considered hemispheric-wide simulations

Line 108: the usage of data on the 700 hPa level comes as a surprise and was not mentioned before. Please clarify. Also on line 111: is the smoothed relative vorticity field evaluated at both 850 and 700 hPa? Line 120: “…maximum remains the track point…” - on which level?

Lines 116-117: how is the selection made? Do you mean that closer and more intense is favoured?

Lines 130-131: it is unclear then why not to use a lower confidence interval to obtain a comparable number of tracks. Please also mention how many tracks are obtained here?

Line 164: “combination” = mean?

Fig. 3a: it is unclear how the subset density is normalized (% of the strongest 10% of all cyclones?), same for Fig. 4.

Lines 305-306: please clarify if the “number of members in which a cyclone is found” means that there is a cyclone track point within a 3.5-degree radius near the reference track?

Fig. 5: Bringing the orange line to the front can make it more readable

Line 377: good prediction of W. Med cyclones: it can be expected because they are mostly slow-moving so errors in location are not expected to arise compared to faster moving systems.

Line 389: add “36 h,” before 42.

Fig. 10c,d and accompanying text: this view is confusing, as it does necessarily capture the cyclone at the timing of the maximum deepening (12 h until max. intensity). It will be more interesting to isolate only this section of the track and examine its predictability metrics.

Line 428: Need to reconcile this statement with the best predictability of W. Med cyclones in terms of track location. It will be good to come back to this issue in the conclusions section.

Line 480: cyclogenesis: this stage is no directly shown here (see also major comment 3).

Technical corrections:

Line 54: robuster => more robust

Line 70: systematical => systematic

Line 107: in => into

Lines 192/384/392: delete “for its/their part” without loss of information

Line 209: spell out “CDFE”

Line 213: add “.” after “applied”

Line 247-248: “do not enter over the sea” - unclear wording

Line 306: on => in

Line 315: of => to

Fig. 5 caption: in function => as a function

Line 382: “the both” - delete “the”

Line 392: firsts => first

Line 399: add “visualized” after “previously”

Line 406: every => all

Line 436: statically => statistically

Line 473: a same area => the same area

Line 491: firsts => first