Loading today’s stories
Science1 distinct publisher3 min readPublished
A Birmingham-led team grew the sample from 55 observed tropical-origin cyclones to several thousand simulated ones, and the return period that falls out for a Lili-scale wind event is shorter than the observed record could ever have revealed.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
The method is only as good as the number it replaces, so start with the sample size. The historical atmospheric record used for comparison held 55 tropical-origin cyclones that reached the study area [5]. The two simulated archives held roughly 45 and 69 times as many, which works out to about 2,500 and 3,800 storms [6][13]. The UNSEEN approach buys many more draws from a modelled version of the same century, rather than new observations [4].
The observed record shows why anyone would bother. The researchers count two named tropical-origin cyclones that caused major disasters in western Europe since 1990, Lili in 1996 and Leslie in 2018 [9]. Two events give you one interval, 22 years long [15]. You can quote that number, but an uncertainty band is a different matter, and almost everything a risk model needs lives in that band.
Converted into the units people budget in, once every 21 to 33 years is a 3.0% to 4.8% chance in any given year [2][14]. Lili itself did more than $300 million of damage in western Europe [3], in 1996 building stock and 1996 money.
The country percentages describe something narrower than annual risk, and the authors say so themselves: Ireland's 13% is a share of the tropical-origin cyclones that affect Europe at all, not a 13% chance per year of a hurricane arriving [7][11]. The study also measures hazard rather than loss. It counts areas of unusually strong wind and heavy rain, which is how the team separated a storm passing over European water from one able to do damage on land [17], and it stops there. For a sense of the impact side you have to go back to an observed case: Ophelia in 2017 killed several people in Ireland, brought down trees, damaged buildings and left around 300,000 households without electricity [10].
I take the frequency estimate more seriously than the league table of countries, and I would put a condition on it. These storms only matter to Europe if they undergo extratropical transition and still find favourable atmospheric conditions as they meet midlatitude weather systems [12]; whether the two ensembles reproduce that handoff faithfully is what the 21-to-33-year figure rests on. The simulations also cover the 20th century [4], so this is a 20th-century climate estimate, not a projection of a warmer one.
Even with those caveats, the direction of the correction is hard to argue with. A hazard with 55 observed cases in a century and two damaging landfalls in 36 years is exactly the shape of problem where a catalogue fitted to observations alone will place too little weight on the worst outcome, because the worst outcome has had almost no chance to show up yet. Ireland and the UK topped both the wind and the rainfall rankings [8], and neither is a place where a once-in-a-generation windstorm is a hypothetical.</body_markdown> </invoke>
Ranked by verification strength, evidence, and original report placement.
Dr Kelvin Ng of the University of Birmingham and colleagues used thousands of simulated hurricane seasons to study tropical cyclones reaching Europe, finding the UK and Ireland among the parts of Europe most likely to be affected; results published in Geophysical Research Letters.
The study suggests storms comparable in severity to Hurricane Lili could occur roughly once every 21 to 33 years.
Hurricane Lili caused more than $300 million in damage in western Europe in 1996.
The researchers used the UNSEEN approach (UNprecedented Simulation of Extremes using Ensembles), drawing on two sets of computer simulations covering the 20th century, which collectively gave thousands of possible storm tracks.
The historical atmospheric data set used for comparison contained just 55 tropical-origin cyclones that reached the researchers' European study area.
The two simulated data sets contained roughly 45 and 69 times as many tropical-origin cyclone events as the historical data set.
Distinct publishers with included, body-backed reporting in this cluster.
phys.org
1 article · August 31, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
science
Eastern US extreme rain is pooling into fewer, wider storms, and station records hide it1 distinct publisher
security
Windows 11's secure kernel trusts a RAM chip that never checks who is writing to it1 distinct publisher
science
The service robot mirrors the customer, and the customer mirrors back1 distinct publisher
science
Antarctica's own seismometers hear hundreds of icequakes at Thwaites and Pine Island1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Peer-reviewed underneath, one retelling on top
The specifics are unusually clean for a single-source science story: 55 observed storms, 45- and 69-fold expansions, return windows of 23.4–32.9 and 21.2–29.3 years, per-country percentages that sum sensibly, and arithmetic that survives being redone. What is missing is anyone outside the author team. The ensembles go unnamed, no validation against those 55 real storms is reported, and every number reaching the reader has passed through exactly one write-up of a Geophysical Research Letters paper.
No one on record acting on the figure
A return period earns its keep when somebody prices, plans or regulates against it. Nothing in this reporting shows a met service, a grid operator, a reinsurer or a building-standards body taking the 21-to-33-year estimate on board, and a paper's publication is not uptake. We would rather say nothing than convert a citation into an adoption signal.
A model output wearing a fact's clothes
The 21-to-33-year figure describes storms that have never happened, produced by two unnamed 20th-century ensembles whose fidelity to real European landfalls is not shown — and it is stated with the flatness of an observed frequency. That is the overshoot. It stays small because the coverage does the two things that keep such numbers honest: it reports the estimate as ranges from two separate simulation sets, and it hands the reader the researchers' own refusal to let 13% be read as an annual chance of a hurricane hitting the UK.
The authors' own account, unopposed
Every figure travels one hop: research team, then science-news write-up. 'The benchmark disaster was less rare than the record suggests' is the kind of result that gets picked up, and no outside voice is present to push back on it. Against that, there is no commercial party anywhere in the chain, the venue is named, and the team volunteered the caveat that most cuts against overreading its own headline — which is not how a pitch behaves.
One thread, well spun
We are reading a single publisher on a single day about a peer-reviewed paper, with no replication, no second retelling and no downstream user to steady the picture. The internal coherence is good enough that we are not worried about the numbers being garbled in transit; we simply have no independent purchase on whether the simulations deserve the weight the return period puts on them.