How Scientists Put a Death Toll on a Single Heatwave
A heatwave rarely leaves an obvious body count. There is no single moment, no wreckage to photograph. Most of the people it kills die at home or in hospital wards over a stretch of days, their deaths recorded as heart failure or stroke, with the heat never named on a certificate. So when a spell of dangerous weather ends, one question hangs in the air with no easy answer: how many did it kill?
For France's late-June 2026 heatwave, two climate scientists have offered a fast reply. Writing in a guest analysis for Carbon Brief, Christopher Callahan of Indiana University and Andrew Dessler of Texas A&M estimate that the heat caused roughly 2,700 more deaths across the country between 12 and 29 June than a comparable stretch of normal weather would have. The daily toll, by their reckoning, peaked at about 300 deaths on 24 and 25 June, the hottest days.
The number is worth sitting with. But so is how it was produced, because that is the real story. This is not a coroner's tally. It is a modelled estimate, and understanding the machinery behind it is what separates a meaningful figure from a scary one.
From a temperature curve to a death toll
The method rests on a simple, grim regularity: people die at predictable rates depending on how hot or cold it is. Plot daily mortality against daily temperature for a given place and you get a U-shaped curve. Deaths bottom out around a mild optimum (roughly 20C for a daily maximum) and climb as the weather turns either colder or hotter. The steepness of that climb, and exactly where the curve bends, differs from region to region depending on what people are acclimatised to.
Callahan and Dessler did not invent that curve for this heatwave. They borrowed it from a study Callahan co-authored in 2025 in the Proceedings of the National Academy of Sciences, which fit the temperature-mortality relationship to four decades of French data. That PNAS paper, led by Callahan with colleagues then at Stanford, used daily death records from all 94 departments of continental France between 1980 and 2019 to model how mortality responds to heat and cold. Crucially, it did more than link a single day's temperature to that day's deaths. It carried five days of temperature history forward to capture heat's delayed toll, and it let consecutive hot days compound, so that a run of scorching days does more damage than the same days scattered apart.
That refinement matters. Tested against the catastrophic European heatwave of August 2003, a standard model underestimated France's excess deaths by more than half, predicting about 7,200 against roughly 15,900 that actually occurred. Once the compounding effect of back-to-back hot days was included, the model's estimate rose to nearly 15,000, close to the real figure. A heatwave, in other words, is more than the sum of its hot days.
To turn that relationship into a 2026 death toll, the two researchers fed in the observed temperatures for June and compared the mortality the curve predicts against what a normal late-June would produce. The gap is the estimate: about 2,700 deaths attributable to the heat.
How hot it actually got
The weather that drove the number was genuinely extreme. France's average June daily maximum reached 36.9C, the authors report. That is well past the previous June record of 34.5C set in 2022, and about 2.4C above it. Climate models had projected temperatures of that order for France around the 2070s, not the 2020s. The heat that arrived was, on this measure, decades ahead of schedule.
The official count, and why the two numbers differ
France's public health agency, Sante publique France, runs its own tally through excess mortality, counting how many more people died in a period than statistical models expected, without attributing each death to a cause. For the week of 22 to 28 June, the agency reported on the order of 2,000 excess deaths, a figure broadly in line with the modelled estimate.
The two approaches are not measuring quite the same thing, and neither is simply "correct." Excess-mortality counts capture everyone who died above the baseline, whatever the reason, over a fixed window. The temperature-mortality method isolates the share the heat itself can explain, and it can span a wider set of days. That they land in the same rough neighbourhood (a couple of thousand deaths) is the useful signal. It is what gives a rapid estimate credibility: an independent, official measurement points the same way.
What the number is, and what it isn't
The 2,700 figure is a rapid estimate. The method behind it is peer-reviewed and was published in a leading journal; the application of it to this specific heatwave is not. It has not been through peer review as a result for the June 2026 event. It is a defensible, fast calculation by qualified scientists, not a settled finding.
It also carries the ordinary uncertainties of any statistical model: the exposure-response curve has error bars, populations adapt over time, and behaviour during a well-publicised heatwave can shift in ways a decades-long average will not perfectly capture. The honest reading is a range around 2,700, not a precise headcount.
What has genuinely changed is the speed. A decade ago, quantifying a heatwave's toll meant waiting months for mortality statistics to settle. Now, with a validated temperature-mortality relationship in hand, researchers can produce a credible estimate while the event is still fresh, fast enough to inform the public conversation rather than arrive as a historical footnote. That is the real development here: not that heat kills, which is old news, but that we can now put a number on it almost in real time, and check it against the official record within days. For a hazard that usually kills invisibly, that is a meaningful shift in how quickly its cost becomes visible.