the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Jul 2026
From climatology to hazard: environmental regimes and tornado wind-speed exceedance in peninsular Spain and the Balearic Islands
Abstract. Tornadoes and waterspouts are infrequent but locally damaging hazards in the Iberian–Mediterranean region, where uneven reporting and intensity uncertainty hamper their assessment. We compiled an updated 1992–2025 catalogue for peninsular Spain and the Balearic Islands by merging the national meteorological agency's tornado records with a pan-European severe-weather database, harmonising event times to UTC and removing duplicates, which yielded 1185 events: 567 tornadoes and 618 waterspouts. The record is temporally heterogeneous, with a late-summer to autumn maximum and a marked diurnal split between morning waterspouts and late-afternoon tornadoes; its rising counts appear to track reporting practices rather than any climatic trend. Using 54 pre-event atmospheric-reanalysis (ERA5) parameters, we tested a six-region geographical framework that proved only partially separable, so a simpler structure emerged: Ward clustering of 50 km hexagonal cells resolves three broad regimes — shear-dominated in the west and southwest, thermodynamically charged in the central-southeast, and weakly forced in the north and east. Adapting a footprint-based occurrence-to-exceedance model, we derived the first regime-dependent wind-speed exceedance estimates for the region. Notably, the least active regime holds four of the five Enhanced Fujita Scale 3 (EF3) tornadoes and dominates exceedance at 180 and 220 km h⁻¹, so where tornadoes are most frequent is not where the strongest winds are most likely. These results provide a first catalogue-based baseline for regime-dependent tornado and waterspout hazard assessment in peninsular Spain and the Balearic Islands.
- Preprint
(2159 KB) - Metadata XML
-
Supplement
(997 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'REVIEW of egusphere-2026-4053', Anonymous Referee #1, 10 Sep 2026
-
RC2: 'Comment on egusphere-2026-4053', Anonymous Referee #2, 11 Sep 2026
The manuscript is a very interesting contribution to the study of tornadoes and waterspouts in Spain and also in Europe. The results are relevant, but in the current form the manuscript is not ready for publication.
Major comments
1. The Methodology section is very dense and with no sufficient details. In the current version of the manuscript it reads as a collection of statistical analyses without a clear structure. I recommend to the authors to reorganise the Methodology section so that all the relevant details, the purpose of each part of the analysis and the link to the manuscript objectives are explicit.
2. The authors mention that the rising number of reports over the study period is probably due to better reporting and increase public awareness and not due to more tornadoes. I agree with this observation, because this is the situation in other European countries as well regarding severe weather reports in general (not just tornadoes). Then the authors have used the data to calculate tornado rates assuming that the reporting is homogeneous both in space and time. This is an important point, given that regime 2 covers the sparsely populated central-southeastern Spain, where reporting is likely low (and only big damaging events are reported). The authors can conduct a sensitivity analysis using recent years (e.g., after 2005), and normalise by a reporting proxy (e.g., population, road density). This can show whether the regime ranking is robust or partly a reporting artefact.
3. The labels for the regimes are obtained from medians at the hexagon level of the pre-event predictors derived from ERA5. The classifier is then used to recover the labels from event values of the same predictors. But the hexagon is already a summary of the events. This seams like a circular argument, or is just the methdology that is not described clearly. One suggestion here is to repeat the Ward clustering keeping out each hexagon, relabel the hexagon and use the classifier on those independently obtained labels.
4. One of the main points of the manuscript is that the least active regions are also the regions where the stronger winds are more likely. This is based on 4 (out of 5) EF3 tornadoes from regime 2. This number of events is very small. The authors should provide a table with the number of tornadoes by EF class and regime. Also, the authors should test whether the intensity distribution is different between different regimes.
5. Waterspouts represent approximately 50% of the events in the dataset and are concentrated in regime 0. Thus, the regime 0 "weakly forced, low-shear" signature may be a marine boundary-layer signature rather than a tornadic environment. This is then used as the spatial frame for a tornado only hazard calculation. The Ward regionalization should be done also on tornadoes alone to show that the partition is stable. Also, some of the hexagons had no reports and were labelled by neighbour inference (Figure 5b). These hexagons contribute to the regime areas A_r used in equation 1 affecting the occurrence rate. The authors should calculate the rates based on observed hexagons only and include this in manuscript as well.
Minor comments
1. The ESWD data are not described in details, for example what are QC1, QC2?
2. For the analysis based on ERA5 reanalysis, how many vertical levels were used for the pseudo-soundings? Are this model levels?
3. The tornado rating in the SINOBAS and ESWD databases were both EF scale or the F scale was used for early reports? If so, how the tornado ratings were homogenised between the two databases?
Citation: https://doi.org/10.5194/egusphere-2026-4053-RC2
Model code and software
iberian-tornado-hazard: code repository for climatology, regionalization, and wind-speed exceedance analysis Gonzalo Agurto Barragán https://doi.org/10.5281/zenodo.21225731
Interactive computing environment
Jupyter notebooks implementing the tornado climatology, regionalization, and exceedance workflow Gonzalo Agurto Barragán https://doi.org/10.5281/zenodo.21225731
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 177 | 67 | 21 | 265 | 58 | 20 | 15 |
- HTML: 177
- PDF: 67
- XML: 21
- Total: 265
- Supplement: 58
- BibTeX: 20
- EndNote: 15
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General assessment
The manuscript presents an interesting and potentially valuable contribution to the understanding of tornado and waterspout climatology and wind-speed hazard over peninsular Spain and the Balearic Islands. The updated event database, the environmental analysis based on ERA5, the objective spatial regionalization and the subsequent exceedance analysis provide a broad and ambitious framework. I also find the manuscript well contextualized, with a large number of relevant references, and the general presentation of the results is quite easy to follow.
In particular, I find the results section substantially clearer than the methodology section. The regional patterns and the distinction between tornado occurrence and high-threshold wind-speed exceedance are interesting and potentially useful. The paper therefore has good potential.
My main concern is not the general scientific interest of the work, but rather the complexity and presentation of the methodology, and, consequently, the difficulty for the reader to follow exactly how the final results are obtained. Several methodological decisions are introduced in a rather compressed way, while their practical meaning only becomes clear much later in the Results section. In my opinion, the manuscript would benefit considerably from a clearer and more pedagogical presentation of the methodological workflow, together with some additional discussion connecting the statistical results with the underlying meteorology.
I would therefore recommend major revision. Most of the issues below appear addressable without changing the general framework of the study.
Major comments
1. The methodological framework should be introduced and explained more clearly
The last paragraph of the Introduction introduces the footprint-based framework of Balbi and Barbieri (2018), but the reader is only given a relatively general description of it. Since this framework is an important part of the transition from climatology to hazard in the present study, I think it would be useful to explain already in the Introduction, in somewhat more concrete terms, which elements of the original framework are adopted and which are adapted for the Spanish case.
For example, the authors could briefly explain how occurrence rate, tornado intensity, path geometry and the distribution of wind speeds are combined to obtain the exceedance probability. This would help the reader understand from the beginning what the final objective of the study is and how the different methodological parts are connected.
2. The definition of an independent event deserves more attention
The treatment of duplicate events should be clarified. The manuscript states that reports within 50 km and 2 h are considered duplicate candidates, but it is not sufficiently clear how the final decision is made in such cases.
This point is particularly important for tornado families, outbreaks, or sequences of tornadoes produced by the same convective system. Two tornadoes occurring within 50 km and 2 h are not necessarily duplicate observations of the same tornado. They may represent two physically different tornadoes associated with the same storm or convective system.
Related to this issue, Section 2.6 describes the occurrence process as involving "rare and approximately independent events". The authors should discuss whether this assumption is fully compatible with possible tornado families or outbreaks containing several tornadoes. If some degree of dependence between events is expected, the implications for the occurrence and exceedance calculations should at least be discussed.
3. The terminology "regime", "region", and "class" should be defined early and used consistently
The manuscript uses the terms region, subregion, regime, and class, but their precise meaning is not always immediately clear to the reader. In particular, the use of regime sometimes feels somewhat abstract and the reader has to discover its meaning progressively.
I suggest defining these terms explicitly before they are used extensively. For example, the manuscript could clearly distinguish between:
I also think it would be useful to explain why regime is preferred to region for the final spatial units, since the latter may sometimes be more intuitive.
4. The PCA and clustering methodology needs considerably more methodological detail
This is probably the main issue I have with the manuscript.
Section 2.4 introduces a relatively complex sequence involving standardization, PCA, cumulative explained variance, DBSCAN, grid-based density analysis, Ward clustering, supervised classification, several performance metrics, normalization of MCC, a composite score and a dummy-relative skill score. Section 2.5 then introduces another sequence involving hexagonal aggregation, medians, imputation, standardization, PCA, Ward clustering with spatial connectivity, different grid sizes, different numbers of clusters, spatial cross-validation and several criteria for the final selection.
Although all these elements may be reasonable individually, their presentation is very compressed. It is difficult to reconstruct the complete workflow and understand which methodological choices are actually used to obtain the final three-regime solution.
The PCA procedure in particular should be described in more detail. There are several possible ways of constructing and applying PCA in a dataset of this type, and the manuscript should specify more clearly the exact procedure used, including the variables entering the PCA, their preprocessing, the number of retained components, and how the resulting components are subsequently used in the clustering.
The paragraph beginning around line 156 also needs more explanation. It would help to give a simple example showing what is meant by noise fraction, dominant-cluster fraction and the screening of candidate solutions.
More generally, I suggest that the authors consider reorganizing Sections 2.4 and 2.5 so that the methodological steps are presented in a more intuitive sequence and are directly connected to the corresponding results. At present, the reader is sometimes asked to remember a complex methodological procedure and only much later discovers why a particular step was important.
5. The spatial-connectivity constraint in the Ward clustering should be explained
The manuscript states that a spatial-connectivity constraint was used to prevent geographically fragmented regions. However, the exact way in which this constraint is imposed is not sufficiently clear.
This is an important methodological point because spatial coherence is one of the main arguments for selecting the final regionalization. The authors should explain more explicitly how neighbouring hexagons are defined, how the connectivity matrix/constraint is constructed, and how this constraint modifies the Ward clustering algorithm.
A simple schematic example would be very helpful.
6. The relationship between the methodological complexity and the physical interpretation should be strengthened
The final regionalization is interesting, but the physical interpretation of the three regimes is currently based mainly on standardized anomalies of ERA5-derived predictors. I think the paper could be significantly strengthened by connecting these anomalies more directly with the actual meteorological environments.
In particular, I would encourage the authors to show, either in the main paper or in the Supplement, composites or representative values in physical units for the main thermodynamic and kinematic ingredients characterizing each regime. For example, CAPE, low-level and deep-layer shear, SRH, LCL and other key variables could be shown using their actual values rather than only standardized anomalies.
This would make the result much more physically tangible.
At present, Figure 6c gives the impression that regime 2 has a particularly strong signal in many of the physical predictors, whereas regimes 0 and 1 appear much less distinctive in this respect. This raises an interesting meteorological question: are the environments associated with regimes 0 and 1 genuinely much less characterized by these ingredients, or is part of this appearance simply a consequence of calculating anomalies relative to the same overall event population?
The paper would benefit from discussing this point more explicitly. It is particularly relevant because one of the potentially interesting contributions of the study is its possible connection with environmental prediction.
7. The transition from occurrence frequency to high-threshold hazard requires stronger physical interpretation
One of the most interesting results is that regimes 0 and 1 dominate the overall tornado occurrence at moderate thresholds, while regime 2 becomes disproportionately important for the highest wind-speed exceedance.
This is an important and potentially powerful result. For that reason, I think it deserves a clearer meteorological interpretation. The reader would benefit from a discussion of why the environmental characteristics of regime 2 may favour the occurrence of the most intense reported tornadoes, and why the more frequent environments in regimes 0 and 1 do not necessarily produce the same high-end hazard.
The interpretation should be made carefully, given the limited number of high-intensity observations, but I think this is one of the places where the manuscript could make a stronger scientific contribution.
8. The high-wind-speed results need careful treatment because of the very small EF3 sample
The authors correctly acknowledge that the 220 km h⁻¹ results are based on only five EF3 reports and on the Brooks (2004) intensity–path scaling. However, because these results receive substantial attention in the paper, I think the limitations should be integrated even more strongly into their interpretation.
In particular, the statement around line 461 concerning the "cautious interpretation" is difficult to understand in its present form and should be rewritten more clearly.
The authors should also explain the shape of the curves in Figure 7a. Some of the apparent "bulges" within the different EF intervals look potentially artificial or at least non-intuitive. I would appreciate an explanation of whether these features arise from the mathematical formulation of the footprint model, the fitted distributions, the scaling between intensity classes, or another methodological aspect.
This would increase confidence in the exceedance curves.
Other comments and specific points
9. Definitions of QC1 and QC2
QC1 and QC2 should be defined at their first occurrence in Section 2.1. Readers who are not familiar with the ESWD quality-control terminology should not have to search for the definitions elsewhere.
10. Clarify the term "filtering"
The manuscript repeatedly refers to "filtering" of the database, but this is not sufficiently specific. Please state exactly which filters are applied and in what order.
11. Time convention: UTC versus local time
Section 2.1 states that event times were harmonized to UTC, while Section 2.2 discusses climatological distributions in local time. Although the difference may not be very large in this geographical domain, the distinction should be made completely clear and the terminology should be consistent. I would personally recommend using UTC as the standard time convention throughout the methodological description, with an explicit conversion to LT only where the diurnal climatology is discussed.
12. Figure 1
It would be useful to include latitude and longitude axes in Figure 1. This would make the geographical limits of the proposed regions easier to interpret.
The sentence in the caption describing how the map was generated with GeoPandas–Matplotlib and the geographical base layers may also be reconsidered. If retained, it should be clear why this technical information is useful to the reader.
13. Clarify the "hourly series" and selected times
Around line 137, the manuscript refers to hourly series but then summarizes them at t = −6, 0 and +6 h. Please clarify whether the complete hourly series are used in the analysis and these three times are only selected for illustration, or whether the analysis is actually based on these selected times and the other descriptors.
14. Clarify the PCA dimensionality in Section 3.2
There appears to be some inconsistency, or at least potential confusion, between the statement that up to 10 PCs are retained to explain approximately 90 % of the variance and the fact that the discussion in Section 3.2 focuses on the first two PCs.
Please explain explicitly how many PCs are used in the actual clustering/classification analyses and why Figures 3a and the associated discussion focus only on PC1 and PC2.
15. Figure 3: explained variance
Please indicate in Figure 3, or in its caption/text, what percentage of the total variance is explained by PC1 and PC2. This is important for interpreting how representative the two-dimensional projection is.
16. Rejection of DBSCAN and other candidate methods
The final part of Section 3.2, where DBSCAN and the other approaches are rejected in favour of Ward clustering, would be easier to understand if the authors showed some details or a simple example of the candidate solutions.
For example, one additional panel or a concise table showing the main candidate solutions and why they were rejected would make the methodological decision much more transparent.
17. Clarify the last paragraph of Section 2.6
The final paragraph of Section 2.6, particularly its last two sentences, is difficult to read and assimilate. I recommend rewriting this paragraph using shorter sentences and making the sequence of the occurrence, footprint and exceedance calculations more explicit.
18. Length and width notation
At line 231, the notation "L × W" is introduced. Please define explicitly that L and W correspond to path length and mean/path width, respectively, and check that the terminology is used consistently throughout the manuscript.
19. Figure 8 and the presentation of hazard
For hazard communication, return periods are often more tangible and easier to interpret than very small exceedance probabilities. Since the manuscript already calculates return periods, I suggest considering whether Figure 8 could also be presented, or complemented, in terms of return periods.
This could make the spatial results more useful for readers interested in hazard assessment and applications.
20. Potential exposure/socioeconomic perspective
The final part of Section 3.4 could potentially be strengthened by comparing the hazard climatology with population density or another simple socioeconomic/exposure indicator, using the same hexagonal grid adopted in the study.
I consider this an optional suggestion rather than a requirement, but such a comparison could considerably increase the practical relevance of the work by showing where the identified hazard overlaps with areas of greater exposure.
21. Conclusions
The Conclusions are concise and acceptable, but they largely repeat points already presented in the Discussion. The authors could consider making this section slightly more focused on the main new findings and implications rather than repeating the preceding discussion.
22. Supplementary Figure S7
I recommend carefully checking and comparing the anomalies shown in Figure S7 with those presented in Figure 6c. At first sight, there appear to be regime-dependent differences, including some apparently discrepant signs.
This should be checked carefully because it may affect the desired meteorological interpretation of the three regimes. If the two figures use different variables, rankings, normalizations or samples, this should be stated very clearly. If they are intended to represent the same quantities, the apparent differences should be resolved.