Overview

This manuscript constitutes a case study for cyclone/medicane “Daniel” which, at different stages in its lifecycle, delivered devastating weather and impacts in parts of both Greece and Libya, in early September 2023. Four aspects are examined, as listed at the start of the title.

The most novel and publication-worthy features of the study are the moisture source analyses, for both the Greek and Libyan floods, the sea wave analysis and the use of cyclone and precipitation objects.

The remainder of the study does not add much to previous published literature on this case (notably Hewson et al, 2024) which is cited and latterly Couto et al (2024), which focusses on broadscale aspects. Admittedly the Couto paper, available here: https://www.mdpi.com/2073-4433/15/10/1205, has only just appeared so was probably unavailable to the authors pre-submission. In some respects these two papers go much further than the one under review, particularly with regard broadscale patterns, local details of the extreme weather and considerations with regard to high impact warnings. Given that a standard requirement for publication, in any journal, is that one adds to previously established knowledge (rather than detracting from it) it is clear, in the opinion of this reviewer, that a very substantial reworking of the paper’s content would be required for acceptance.

There is a clear reluctance to include observations in this paper – notably rainfall measurements. If numerical model analyses were perfect, this might be acceptable, but given that they are not, particularly with regard to rainfall, which is the impact centrepiece of this study, this is a major omission.

Furthermore, reviewing the paper has been a frustrating process due to the many inconsistencies in different segments of the text, inconsistencies between what the figures show and what the text says, simple errors, poorly explained figures, and unsubstantiated conclusions. Rather than go through absolutely everything which is of concern, which would take a very long time and replicate the checks the authors themselves should have carried out before submission, I will instead go through the figures, which are the bedrock of the paper, and highlight the key issues via those.

Main points

Figure 1a: Some of the red spots are missing (assuming the time interval is 6 hours, which should anyway be stated); some of the labels show the wrong time, and the size-for-mslp legend is hard to interpret. At one stage in the text it is stated that the cyclone was intense early in its lifecyle – by normal measures 1004mb is not intense – and indeed elsewhere in the text this statement is contradicted. Somewhere else in the text it says that the cyclone intensified on 6^th and 7^th, as can be seen on this Figure; this was not the case and nor does the figure show it, even allowing for mislabelling errors. Somewhere else the text says the minimum pressure of 997mb was reached on 9^th September. This is not correct either, nor does the figure show this.

Figure 2a: The rainfall area is too small to see properly (even when zooming on the pdf), as are the wind barbs. Where does the data come from – ERA5 or ECMWF analyses/short range forecasts? The latter is much higher resolution (9km versus 31km) and so would likely show much more useful rainfall detail (if one could see it). I can quite imagine that the PV of 2 PVU, in the caption, is actually 2. In the text it is stated that “there is a high wind speed pattern aligned with the PV streamers’ orientation”. I do not know what aligned with means here. The text cites 750mm rainfall in 24 hours on 5^th – why not say where! This actually occurred east of Volos, at 3 close sites (see Table 2 in Dimitriou et al, 2024), and is where the model rainfall pattern, when zoomed in maximally, shows about 110mm. So the reference to a 50% shortfall in the model should be 85%. The authors actually refer directly to purple colours (which are seen elsewhere), which represent 200mm, so quite clearly even those don’t represent 50% of 750mm. This all then makes the statement that (ECMWF) models provided good guidance somewhat incorrect.

Figure 2b: The model rainfall, up to 00UTC on 11^th, is not much in the Derna catchment (location unfortunately not shown but included in Hewson et al, 2024) despite the fact that the dams broke an hour or two later. Again this is very concerning with regard to model validation and impact predictability. These aspects are not discussed at all. Text says the PV streamer was “much weaker” at this time. I do not know what this means. It gives the wrong impression too as this streamer is, on the contrary, probably a marker for a substantial lobe of upper level forcing that helped trigger the main intensification of Daniel.

Figure 3: It was nice to see the moisture sources, even if the propensity to uptake most moisture just upwind of the heaviest rain for Daniel, in strong wind areas, was not hugely surprising. The uptake in the composited cases is much harder to second guess, so this is a nice result. I am not sure why 10 day trajectories were used. That seems quite long? Also I am not sure what “30km grid” means, in the main text. Coastlines and sea areas are impossible to see on the figure in this form, so that aspect has to be improved. The main discussion of the moisture uptake elects to ignore any sources over land, yet clearly they are relevant – more so than the Atlantic Ocean which is mentioned. In the conclusions uptake over landmasses is mentioned for the first time.

Figure 4: Although this looks initially quite convincing on closer inspection one sees that there is virtually no signal in (b) of a particularly high discharge near to where the heaviest rainfall was in Greece (its all time 24h record), east of Volos, nor in Derna in Libya, or its catchment. These aspects should have been extensively discussed. Maybe this relates to the rainfall errors on Figures 2a and 2b that I reference above, which were also not discussed.

Figure 5a,b: The colour scheme used is poorly chosen as it does not allow for accurate values to be read off. However to me it looks like the value of a +2C anomaly quoted in the text should actually be +1C (save perhaps for the area N of Derna on (b) where it may be +2C). This is especially true if one references both 5a and 5b instead of just 5a, which would be justified as the lifecycle is then better covered. This would be a bit of a counter argument against the misleading statements regarding climate change influence made late in the manuscript. Furthermore the blue patch of negative SST anomalies on 5b, which may be a legacy of Daniel’s upward fluxes, is not discussed; indeed the manuscript contains no reference to 5b at all, so far as I can see.

Figure 6: This is a nice figure. However a related statement in the text that “it is impossible to evaluate the relative socio-economic impact of each threat (storm surge, waves, rain, river flood)” seems rather preposterous when we know that >5000 people lost their lives in Derna as a result of a dam burst (due to rain and river flooding causing overtopping).

Figure 7: Ok but the right-hand panels are not valid for 10 September at 12UTC. I also have doubts about the valid time of the left hand panels given that the cyclone centre seems to have a rather different position to that shown on Figure 2a. Or maybe Fig 2a is the one that’s wrong? The statement in the text that 7a shows a much larger area of high PV than Fig. 2 is not correct. It is the other way round (note we only have 2PVU on Fig 2). And for PV averaging it might anyway be better to take the log first, given PV structure/ranges? An analogy is that one cannot meaningfully average visibility (across several orders of magnitude). Whilst this figure and the next one highlight clear convergence in the EPS solutions, which is OK, the text fails to acknowledge that relative to what came beforehand, the forecasts from 12UTC 1^st (the first one included) actually represented a big positive step in skill – at least they had cyclones – due to much better handling of the mid-Atlantic Rossby wave train, due in turn to better handling of a tropical cyclone (as in Hewson et al, 2024). This is an example where one sees that the manuscript is not adding much to previous work, and indeed is contradicting it somewhat. These two results would need to be placed alongside each other in this paper to give the full context of cyclogenesis predictability for this case, and thereby advance the science as is required for a publishable standard. For Fig 7h the discrete 300hPa high PV blob west of Daniel is not mentioned. This very likely links to the upper level low moving in from the west from Hewson et al (2024), that is referenced, so a useful connection could be made here, pointing out also the increased specificity of this feature as lead times reduce, as shown by 7b,d,f,h.

Figure 8: The left hand panels do not appear to be valid for 5 Sep 12UTC. Judging from the cyclone spot cluster they may be for 5 Sep 18UTC. Similarly the spots on (g) do not seem to correspond with the mslp minimum on Fig 7g, suggesting these panels are not for the same time. This is all rather confusing. The valid time for b,d,f,h looks to be correct.

Figure 9: The reader is left to guess what the valid time range is for the precipitation objects. It may be that it is the 24h periods ending at the stated valid times, yet if that is the case why use 5 Sep 12UTC as an end time when the main 24h rainfall period was 00-24UTC on 5 Sep or a bit later (again reference Table 2 in Dimitriou et al, 2024)?

Figure 10: This figure is fine but I do not understand what it intends to show – the text “This shift is plausibly relevant..” I have not managed to decipher. Adding spots on the grey track lines, to show cyclone centres at a particular valid time, could help.

Figure 11: This figure looks potentially informative but the elements of it are not explained, and furthermore some elements are barely visible (grey tick marks overlapping the box and whiskers). First readers should be pointed to where the Pinios river outlet is, and what its catchment is. According to Wikipedia the spelling should be Pineios (though I concede that could be “wrong”). It would also help to see the Wadi Derna catchment – the relative size of this, versus the Pineios catchment, is very important for predictability and impact prediction and this is not discussed. Then where does the “perfect forecast” benchmark come from. Is it related to the rainfall in Figure 2b, which as stated above looks wrong (hardly perfect!) in the critical area? Then what do the box and whiskers relate to, and why do they have a strange shape? What are all the percentiles represented? It is fairly clear to me that the forecasts for Greece converge onto the “right” solution (if the red curve can be trusted), whilst the forecasts for Libya, though overall they get a bit better with lead time, basically do not converge. The forecasts from 9^th for Derna, which might be at the most critical for triggering preventative measures, step back from those of the previous day, and then even from 10^th we still have huge spread and a big shortfall in the box and whisker median (if that’s what the middle black line is). Yet all the text says about the Derna forecast is that it follows a “similar pattern” to the one for Greece. This is an incorrect and unhelpful sweeping statement. Furthermore, the following paragraph goes on to say that Fig 4b highlights the unprecedented nature of the event, when for Derna and its catchment the signal is rather weak. The much stronger signal is well to the west (also discussed above).

Figures 12 and 13: On many of the panel legends the numbers do not align with the colour bars. So the reader does not know what the colour bars mean. This is obviously important when one tries to cross-reference with the text – e.g. on Fig 12h it is stated that temperatures have gone up by 2C in the Ionian Sea when it looks like rather less than that. Then why are there contours as well as shading on panels d,h,l and p? The fact that the rainfall amounts for the 2023 case in this depiction under-represent reality by a large margin is not mentioned, when clearly this has relevance (panels i). The worst part about this part of the study is that the conclusions in the text do not reflect what the figures show. For example, the authors state “we conclude that Mediterranean depressions like Daniel hitting Greece and Libya show lower MSLP and higher precipitation in the present climate than in the past”. The evidence for this is supposed to be panels d which show basically no mslp change at all; and panels l which show drier over Greece and slightly wetter over the seas around Libya. And maybe +2mm or so per day over northern Libya itself, but when >400mm/24h was recorded at one site for Daniel this seems irrelevant. The text of Section 5.2 contains many other errors and inconsistencies, too numerous to go into here. In my opinion the vast majority of Section 5.2, for which these Figures are the “evidence” should be removed from the paper, as it shows very little of substance. One could much more usefully and honestly say, in brief, that “an in-depth study using standard methods indicates that in the ERA5 dataset there is no evidence of climate change influencing features like Daniel in the 1980-2020 period”. The only non-neutral “result” I can see on these figures is a signal for an increased frequency for cyclones, in the SOND period, in the SE Mediterranean near the N African coast (Fig 13x). So that could be referenced too. Furthermore, it seems to me that trying to link El Nino, the PDO and the AMO to Daniel-like cyclones over just a 40-year period is stretching physical credibility beyond its natural limit.