the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 23 Jun 2026
Hurricanes that haven’t happened, yet: Identifying unprecedented tropical cyclone scenarios
Abstract. Tropical cyclones (TCs) can be unprecedented in many dimensions and can result in disasters when they are unforeseen. We conduct an intercomparison of four TC databases to identify plausible synthetic events that would exceed observational records, and argue that these provide robust and evidence-based scenarios for disaster management use. We compare datasets produced by two statistical TC track models and a newly published TC track hindcast archive from numerical weather predictions: STORM (n = 712,800), IRIS (n= 472,162), and WATTCH (n= 36,793). For all six TC basins, we explore how each dataset characterises unprecedented extreme events in terms of lifetime maximum intensity, 24 h changes in wind speed, monthly frequency of Category 4 and 5 storms, and latitude at first landfall. We assess how each dataset represents the basin-level observational record from IBTrACS by conducting a series of fidelity tests (mean, standard deviation, kurtosis, and skewness) to assess whether their most extreme events could be considered plausible in the current climate. Between 50 % and 89 % of dataset-basin combinations pass at least 2/4 fidelity tests if we include where models indicate underestimation. From this, we identify several hundreds of plausible simulated TCs that exceed historical records in different ways. Where datasets show good fidelity, we illustrate the potential use of these datasets by extracting unprecedented scenarios such as a Category 5 TC hitting southern Madagascar or a TC making landfall on the city of Xai-Xai in Mozambique south of the country's most southerly landfall. Based on this work, we underscore an opportunity for disaster management practitioners to access unprecedented TC scenarios relevant to their work and that would be both robust and imaginative, going beyond current practice.
- Preprint
(2347 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-2758', Anonymous Referee #1, 01 Jul 2026
-
RC2: 'Comment on egusphere-2026-2758', Anonymous Referee #2, 28 Jul 2026
Review of Hurricanes that haven’t happened, yet: Identifying unprecedented tropical cyclone scenarios by Heinrich et al
General comments:
This paper provides a comparison of databases from available tropical cyclone (TC) generation systems with an observed database. The analysis is conducted basin by basin and employs a range of different indicators of tropical cyclone activity. The evaluation of the tropical cyclone generation systems uses a statistical comparison with the statistical moments of the observed distributions. The paper seeks to identify which generation databases are most appropriate in each basin for each indicator in identifying unprecedented tropical cyclone behaviour. The analysis is thorough, the research is novel (perhaps the first systematic comparison of this kind), and the topic is important and appropriate to the journal. My comments are therefore in the 'minor' category. This paper makes an important contribution.
Specific comments:
1. Figures. Many of the figures are too small to be able to discern the details in the figures. This is particularly true of printed copies, but the problem persists for electronic copies -- when increasing the magnification on the electronic copy the fonts and lines tend to just blur in my pdf-reader. This comment could go in the 'technical comments' section below, but it is actually an issue in evaluating the paper, so is included in this section.
2. Observed cyclone dataset uncertainty. The observations here are IBTrACS (this term is used in the abstract but isn't defined there). Since IBTrACS extends from 1851 in some historically very sparsely observed basins (e.g. SP, SI), I would imagine that there are a number of missed systems and missing or poor observations pertaining to some systems. It would be helpful here to provide some indication to the reader how this might affect the results, particularly since IBTrACS is used as a form of truth against which to evaluate the TC generated datasets. For example, divide the IBTrACS dataset in half (first half period, second half period) and compare the statistics of the two samples against one another and against the full period sample. Notable differences could be due to long period variation and sampling, but they could also be due to missing and poor quality data in the first half of the dataset. The reader would like to know if this is potentially important or not, and how it might be allowed for in the comparison with the generated datasets. For example, redo all the comparisons with the generated datasets with just the second half period of the IBTrACS data and report on the differences for the results with the whole IBTrACS period.
3. Moments-based comparisons. The first four statistical moments are used to evaluate and compare the distributions of the sampled generated TCs with the statistics of the observed TCs. It would be helpful here to discuss some of the shortcomings of a moments-based approach for the reader. For example, matching the first few moments of two given distributions does not guarantee that they are the same or have the same shape. The focus on moments makes sense when you are concerned about the tail, but the shape is another important indicator of the distribution which can change the space of plausible outcomes. The statistical comparison would be stronger if you added in another test, such as the KS test, that relates more specifically to the shape of the distribution. The KS test has its limitations too in that it can be insensitive to the tails, but the combination of moments tests and a KS test (e.g. Irving et al. https://doi.org/10.1002/met.2217) provides a more robust overall evaluation with diagnostic information on tail and shape.
4. Section 3.2.5. This section is not like the others. It says that it is a presentation of "work done" elsewhere and seemingly already published "by Archer et al (2024)"? While I understand that rainfall is an important part of the impacts of TCs and wasn't in any of the datasets you analysed til now, I don't think this section adds enough to warrant the disjunction and the implied questions about whose work this is. For example, I was looking at the captions in figures 9 and 10 to see if the figures were from Archer et al., and if not, what was done in this work? Furthermore, figure 10b is a mystery to me; it is mostly black! The caption simply says that it is a "flood footprint map". I have no clue what this is showing me and no idea even if the axis coordinates are latitude and longitude or something else? Since the paper is already long and I have suggested some extensions to the analysis above (points 2 and 3), I suggest deleting this section. You can still note that rainfall and flooding are important impacts from TCs, but are not covered here.
5. Title. The title is in some ways misleading and inconsistent in the first half where it says "Hurricanes that haven't happened, yet". All the TC datasets in the paper are based on TCs that did actually happen. They are then extrapolated into larger samples by statistical or dynamical means. But the point is that they are all framed around events that actually happened. This isn't a problem as such, but this is an important caveat, and the point is somewhat obscured by the first half of the title. The first part of the title is also partly inconsistent in that the body of the paper seems to mostly prefer the more general term 'tropical cyclone' to the more provincial term 'hurricane'. The first half of the title is a nice alliteration, but that nicety should not come at the cost of accuracy in describing the content.
Technical corrections:
6. Figure 2. This figure is needlessly clumsy in that it splits the single SP region box into two boxes (as noted in the caption) due to the longitudes running -180--180. There is no need to introduce this ambiguity around the box regions, since there is no box that spans the 0 longitude line. i.e. the same plot with the longitudes running 0--360 would not split any of the boxes and would be clearer to the reader.
7. In most journals the acronyms need to be defined at first use. I'm not sure if this applies here or to the abstract, but the dataset acronyms are not defined in the abstract.
8. Line 35. The citation to Frank 2003 is recursive/repeated?
Citation: https://doi.org/10.5194/egusphere-2026-2758-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 56 | 33 | 5 | 94 | 4 | 6 |
- HTML: 56
- PDF: 33
- XML: 5
- Total: 94
- BibTeX: 4
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Summary
The author team uses several datasets to explore different characteristics of unprecedented TC events. Thereby they tested the fidelity of the datasets in realistically representing their most extreme TC events. Further they showed in some examples how the data set can be used, underscoring the opportunity for disaster management.
General
The author team present an interesting study which certainly fits NHESS. Still, the structure lacks clarity and the quality of the figures does not meet the standard for publication. Also, the results are in a stage of the zero-order draft so I recommend major revision.
Major
Minor
Title: The title is a bit repetitive, so I suggest to use the second part: “Identifying unprecedented tropical cyclone scenarios”
L15: TCs can also result in disaster when they are foreseen, so remove “when they are unforeseen”.
Introduction: It is rather uncommon the use subsections in the introduction (except in review articles), please make one flow text.
L58: “Hodges et al. (2017) used the “
L85: Please cite studies which assess TCs in past present and future, one example is DOI: 10.3402/tellusa.v64i0.15672 but there is certainly more work which could be cited.
L96: “such as Philip et al. (2022) who derived “
L99: “who modelled”
L106: For convection permitting models you can cite DOI: 10.1127/0941-2948/2013/0472
L146 Missing dot
L151: “who argued that scientists “
L204: It remains unclear how the matching is done. The authors need to be specific about this.
L204: “. Tropical cyclones”
L209-213: This is a kind of repetition of the paragraph before and again no proper explanation of how the matching is done is presented.
Fig. 1: too small
256: Please change methodology to method.
L261: “then the models produce “
L270: Two sentences for a sub-subsection are not enough, maybe avoid sub-subsections.
Fig.2: Resolution is not sufficient and labels too small.
L323-326: This is a repetition.
Fig. 3: too small, low resolution
L360-366: This bullet list contains a lot of speculations and draws conclusions, I suggest avoiding this a move the conclusions to section 5
Fig.4,5,6,8 too small, axes labels missing, resolution bad.
L394-399: This bullet list contains a lot of speculations and draws conclusions, I suggest avoiding this a move the conclusions to section 5
L427-437: This bullet list contains a lot of speculations and draws conclusions, I suggest avoiding this a move the conclusions to section 5
Fig.7 bad quality
L486-497: This bullet list contains a lot of speculations and draws conclusions, I suggest avoiding this a move the conclusions to section 5
Fig.9: Is this really needed as a figure, I have the feeling that a description is sufficient. Also is it mm per day ?
L525: “Archer et al. (2024) identified”
Fig.10: axes labels missing in a, (a), (b) labels missing and (b) is unclear what we see and what the shading means.
L591: “Kelder et al. (2025) who advocated “
L639: You may cite DOI: 10.1127/0941-2948/2013/0472 here
L646: The publication of Jeffries seems to be quiet important but it is still not accepted, so I suggest to remove submitted publications and find published works which supports your statements.
L648: The third aim of this study”
Conclusion: This section is more a summary/abstract than a conclusion, so please draw conclusions.
L659-662: Avoid asking rhetorical questions, so I suggest removing this paragraph.