One 2008 cyclone killed more than twice as many people as 675 other storms combined

Work the cyclone number backwards. If Cyclone Nargis killed almost 140,000 people and that is more than double the other 675 deadly tropical cyclones of the period put together, then those 675 killed fewer than about 70,000 between them ^[2][12]. The average is under roughly 104 deaths each. Nargis is about 1,350 times that average ^[13], and it accounts for more than two-thirds of everyone who died in a tropical cyclone across the 25-year record ^[14].

For fitting a distribution, that leaves one point. No other single storm in the record can have killed more than about 70,000 people ^[15], so the part of the curve that decides the answer has a sample size of one. A return period cannot be estimated from one observation. The mean of 676 events tells you nothing useful about the next Nargis.

Heat has the same shape on a shorter clock. Around 70,000 people died in the 2003 European heat wave, and almost half of those deaths fell in the first two weeks of August ^[3]. That is roughly 35,000 in a fortnight, about 2,500 a day ^[16].

Tom Matthews, senior lecturer in environmental geography at King's College London and lead author of the paper, said: "History is almost guaranteed to make us underestimate future risk. The events that have shaped our sense of what extreme weather can do are not the worst that can happen. They are the worst that have happened so far, in a short record, in a climate that no longer exists." ^[6]

This is a perspective paper ^[17], and its argument is about method, not probability. The authors say it is hard to rule out very rare future events with death tolls in the millions, and they state plainly that possible is not the same as likely ^[4]. The analysis stops short of a frequency for that case.

They describe a narrow, specific blind spot. The authors say current risk assessment handles a familiar hazard at unfamiliar intensity reasonably well, a hurricane with higher wind speeds for instance, and handles the unprecedented sequence badly: a devastating heat wave arriving just before a strong hurricane, with many millions evacuating and no air conditioning to fall back on ^[9]. Their broader charge is that the what-if scenarios now widely used are not designed to identify worst possible outcomes, and so cannot prevent them ^[10].

The fix they propose runs through process. Matthews said the work "means working backward from an outcome nobody wants and asking what could produce it, and it means bringing in people who are not usually in the room, such as those who can harness the power of human imagination, including risk scholars and writers of climate fiction" ^[8]. The paper's three priorities are horizon scans with diverse participants including existential risk scholars, frontline communities, insurers, climate fiction writers and climate modelers; better understanding of the thresholds beyond which impacts grow very steeply; and mapping the physical limits of weather in climates far hotter and far colder than today's ^[5].

If the thresholds where impacts steepen are not yet well understood ^[5], then the relationship between hazard severity and death toll is unsettled, and a scenario can get the wind speed right while getting the consequence badly wrong. The warning arrives after a summer in which thousands of excess deaths were recorded across Europe during the June heat waves alone ^[11].