The manuscript “Changes in Mediterranean flood processes and seasonality” by Y. Tramblay et al. is an interesting study analysing changes in flood event characteristics, flood types and their seasonality in 98 catchments in Southern France. The results presented are coherent with other recent literature about flood changes in the Mediterranean and demonstrate and confirm that soil moisture is the primary driver of flood changes in this region. The manuscript is overall well written and logically organised. Please find my comments below.

Major comments:

My main concern is about the reliability/suitability of the reanalysis product used for the retrieval of precipitation and soil moisture information for the catchments (L135-136). What is the spatial resolution of this product? Is the spatial resolution fine enough for the relatively small catchments in the analysis? At L394-395 the authors state “[…] despite the large sample of basins considered, the patterns are consistent and homogeneous across different basin sizes and locations”. Could this be due to the (coarse) spatial resolution of the reanalysis data compared to the (small) size of the catchments?

Specific comments:

L46-47: “[…] the mean flood date being on average advanced by one month”. Please specify that the shift refers to two sub-periods.

L116-117: I suggest adding in the introduction (and discussion) one recent study by Tarasova et al. (2023) about changes in flood processes in Europe. Tarasova, L., Lun, D., Merz, R. et al. Shifts in flood generation processes exacerbate regional flood anomalies in Europe. Commun Earth Environ 4, 49 (2023).https://doi.org/10.1038/s43247-023-00714-8

L155: in other words, did you adopt a peak-over-threshold (POT) approach?

L163-167: it’s unclear how the maximum precipitation is calculated. Is it the maximum daily precipitation within the same time interval where total precipitation is calculated?

L213: does the first period start in 1959 or in 1950 (as stated in the abstract)? Please check.

L214-215: it is not clear how the pivot year is selected and used in the analysis. Is the extension of the two periods always the same in all catchments (as also shown in all figure legends) or does it vary? Please clarify.

L230: I suggest renaming this section “Results and discussions” as it also contains, alongside with the results, a considerable amount of interpretation of the findings in the context of the literature.

L234: please specify that the changes refer to the difference between the two periods. This also applies to subsequent occurrences in the manuscript, especially in the caption of the figures, where it is not always clear what exactly “changes” refers to.

L256: how is the runoff coefficient calculated for each event?

L258-261: Correlations between antecedent soil moisture and runoff coefficients are analysed and reported in the text. I suggest adding a table or adding these results as a panel of figure 3 (or modifying figure 3) to make it easier to follow. The same suggestion (i.e. adding a table/plot) for L395-399 and  L401-405.

Figure 2: I suggest inverting the colours of the colorscale and adding the units of the relative change to the axis.

L293: a mountain range is mentioned. To facilitate reader that are not familiar with this area I suggest adding a label to the map of Figure 1 to locate the mountain range.

Figure 4: the coloured dots look all a bit brownish and therefore the map is not so easy to read. I suggest making the colours brighter.

L330: “Association between flood occurrence and weather patterns”. How is the association done? Is the WT selected based on the date of occurrence the flood peaks or the preceding days? Please specify it here or in the method section.

L340: “Change in seasonality (of what? Of floods?) can be ascribed to changes in the seasonal occurrence of the weather types”

L332-347: please check coherence of WT numbers and names in the text and in figure 6. The WT numbers and names seem to be different in some occurrences in the text and in the figure. E.g., WT2 is “Atlantic circulation” in the text but “Steady Oceanic” in figure 6, where Atlantic is WT1 instead.

L340-347: the described changes are tiny in figure 7 and therefore do not seem very significant in the context of the description and interpretation of the results. There are other larger changes in figure 7 that are instead not described. Perhaps figure 7 could be further discussed.

L402-404: “For short rain and long rain, the maximum contributions observed are 36% and 32%, respectively, but these maximum values are only found in small basins.” Do these findings refer to the same 30 basins mentioned above?

Figure 7: please add a label to the vertical axis

L513: “[…] related to higher evapotranspiration rates” could you add a reference?

L470: please specify how the regional distributions in fig 12 are obtained.

L419 and L426: the word “flood drivers” is here introduced and I believe it refers to the flood types mentioned in the rest of the manuscript. Please use consistent terminology to avoid confusion.

L423: is the peak in January or February? (I think February is correct, fig 9)

L426: where are the long-term changes shown?

The manuscript titled “Changes in Mediterranean flood processes and seasonality” by Tramblay et al. investigates how flood types and seasonality evolve in Southern France during the past 50 years and attempts to link the change in flood seasonality to changes in mechanisms. Overall, the logic is clear and the results are well presented. However, deeper scrutiny reveals some issues that, in my opinion, might undermine the quality of the manuscript in its current form. My main concerns are the robustness of the results and the contribution of the study, which are relatively weak in the present manuscript.

Firstly, in the introduction part, the authors argue that “Most of these studies rely on flood classification schemes, with various complexity depending on the type of data available, allowing a data-based separation of floods into their distinct generation mechanisms”, which I agreed with. While I assumed that the study would propose an improved approach to partially overcome the current limitations of the flood classification scheme (e.g., the relatively subjective threshold selection, etc.), it seems that the study only used a simple decision tree with hard thresholds, without any justification or discussion of the threshold selection (Tarasova et al. 2020; Zhang et al, 2022). Currently, all the results are based on the seemingly arbitrary threshold and structure of the decision tree. For example, the authors use “50% saturation”, “95th percentile of rainfall”, etc. as the threshold to distinguish the “excess rainfall”, “long rainfall”, and “short rainfall”, while if some events are distributed around these critical points, the conclusion about their changes might be quite sensitive to these values. I wonder if there are some sensitivity tests to ensure the robustness of the results.

In addition, the manuscript argued that “Yet, beside the trend detection, no study has provided an in-depth analysis of the long-term evolution of flood processes in these regions”. This seems to be a very strong statement. However, a recent study by Jiang et al. (2022) has examined the change in flood mechanism in 1000 catchments in Europe, which also include catchments in southern France. The study showed similar trends in flood generation processes as in the present manuscript. I think it would make more sense to compare the trend results with other literature. Also, my concerns about the contribution arise because the conclusion is not very strong currently based on the significance test results. Although the authors argued that it is due to the short records of samples and interannual variability, it would impair the reliability of the conclusion somehow, particularly in the case of lack of sensitivity test for the method.

References:

Tarasova, L., et al. A process-based framework to characterize and classify runoff events: The event typology of Germany. Water Resources Research, 56, e2019WR026951 (2020). https://doi.org/10.1029/2019WR026951

Zhang, S., et al. Reconciling disagreement on global river flood changes in a warming climate. Nat. Clim. Chang. 12, 1160–1167 (2022). https://doi.org/10.1038/s41558-022-01539-7

Jiang, S., et al. River flooding mechanisms and their changes in Europe revealed by explainable machine learning, Hydrol. Earth Syst. Sci., 26, 6339–6359 (2022), https://doi.org/10.5194/hess-26-6339-2022.

Other comments follow:

Abstract: what do the “flood event characteristics” mean? Please specify.

L131: What is the spatial resolution of the SIM reanalysis data used? Please clarify if the size of the study catchments is comparable to the spatial resolution of the hydrometeorological data sets.

L133: Please clarify how the nival regime is identified from the river discharge hydrographs. It is also not clear why they should be removed, since even snow-covered catchments can also be affected by rainfall, e.g. rain-on-snow events.

Figure 1: It would be better to add an inset map to show the location of the study area in Europe.

L155: I understand the reason for using POT1 instead of AM, but please clarify if this will affect the subsequent trend analysis.

L164: Please specify whether rainfall is the precipitation that excludes snowfall.

L165: Is the precipitation on the same day as the flood peak considered, please clarify.

L174: How is the duration of the flood event calculated, i.e. how are the start and end points of the event determined?  Also for the runoff coefficient calculation process.

L183: How is September 1 of the hydrologic year determined?

L197-206: As I noted earlier, the classification reads rather arbitrary.

L207: What does "other" mean?

L220: What test was used to check statistical significance?

Figure 2: what is the unit of "relative change"? I also suggest showing the statistical significance of these changes in the map.

Table 1: one-tailed or two-tailed test? please specify.

L262-264: How is the conclusion related to the results?

L269-273: I would expect a figure to support the results.

Figure 3: please give the number (of events?) in each bin.

Figure 4: not quite sure if POT1 events instead of AM events will introduce a bias in the flood dates.

L298-302: it is difficult to follow without showing these basins on the map.

Figure 5: Please also show the results of the significance test in the maps.

L304: Which time period (1991-2021 vs. 1959-1990) was referred to?

L341: I would like to add significance tests on the difference between the proportions, which would be more supportive of the discussion.

L398: I think we should be cautious about this conclusion, given the relatively high p-value and low correlation coefficient.

L431-434: How statistically significant are these results?

Figure 10: Please also include the significant test results. I am also not clear how the change in frequency is calculated for each catchment. Do you compare the frequency in 1991-2021 with the frequency in 1959-1990?

L445: How was the conclusion reached? I don't understand the logic. I am not surprised that the trend level is not consistent. Even a small change in the driver magnitude can lead to a change in the flood type (because the threshold leads to a hard boundary in the classification, that's why I asked for a sensitivity analysis).

Figure 11: Please show the 25th and 75th percentiles of the regional frequency to show the spatial variance.

L477: "This is mainly due to a decrease in the specific discharge of short rain floods" I can't understand it, because if the short rain has been observed intensified, if we only consider short rain floods, it is more reasonable that the flood magnitude will also increase.

This paper shows that floods in southern France have generally been coming earlier in the wet season (water year in the U.S.) over the last 60 years, and while the associated precipitation extremes have increased, there are no corresponding trends in flood magnitudes.  The explanation (which I find reasonably convincing) is reduced antecedent soil moisture, due to the shift of events earlier in the year (and perhaps also a general drying of soils associated with climate warming).  The paper constitutes yet another bit of evidence that while precipitation extremes mostly are increasing, there is little evidence of concurrent increases in flooding.

My main misgiving about the paper (which may well not affect the overall results of shifting of floods earlier in the year, without magnitude trends) is their approach to segregating events by flood generating mechanisms.  It seems a bit peculiar that the classification of the majority category is based on antecedent soil moisture, but the other two on precipitation intensity only (short and long duration, respectively).  I would be inclined to lump all the events together, then examine the role of antecedent soil moisture separately (and as well, whether the flood-associated precipitation is, or is not, sufficient to satisfy the soil moisture deficit by the end of the event.  That said, they may have stumbled on something in their finding that the fraction of events in their excess soil moisture category has decreased slightly with time, whereas the short (one-day) events have increased.  The question that it would be nice to answer is whether there’s been a shift in the distribution of extreme precipitation to shorter events?  And if so, could such a shift (doesn’t necessarily have to be just one day, could for instance be 1-2 day) becoming sufficient to exceed the initial soil moisture deficit, so that at some point (although clearly not now) this might lead to an increase in extremes.  Stated otherwise, while the recent balance between increased extreme precipitation and reduced antecedent soil moisture seems to have shifted so that antecedent dryness is cancelling increased precipitation intensity, is that balance likely to shift in the future?

One final comment:  I don’t see much value in the weather type discussion.  What really matters to floods are a) antecedent soil moisture, b) precipitation intensity and c) precipitation duration (there are in addition factors such as storm extent and movement relative to catchment size and orientation, but these are more difficult to analyze given that the catchments are fixed by prior decisions as to where to locate gauges.  So I would stick to the precipitation characteristics that matter to floods, and how they might or might not have changed.