A news item flashed through the morning digest from Copernicus: August 2026 tied the record for hottest month with July 2023 — the planet's average temperature 1.65°C above pre-industrial levels. Record chases record, the system no longer returns to old values. Striking, but familiar: we know how to read about daily records.
Then my eye caught on a phrase usually hidden in the footnotes: record minimum nighttime temperature. Paris, August 11–12, 2003, two nights in a row above 25.5°C — an absolute nighttime record that held as a symbol for barely any time at all. Twenty-three years ago Europe went through a canicule that killed, by the latest recalculations, more than 70,000 people, and the most interesting thing there — wasn't the heat. The most interesting thing — was that night killed them. And there's already a whole science about nighttime mortality, turns out, a fresh Nature paper from August 17, 2026, and a counter of lost sleep hours per planet inhabitant. This thread — from the Paris night of 2003 to "climate insomnia" in 2026 — hasn't really been assembled anywhere.
The August 2003 canicule was anomalous on all counts at once. A quasi-stationary anticyclone hung over Western Europe: clear skies, subsiding air, warm advection — classic heat wave construction. But on top of the circulation a second machine was working, about which in 2003 little was yet written: the ground. Spring 2003 came with a severe precipitation deficit, the soil dried out, and daytime heat stopped being spent on evaporation. Fischer, Seneviratne and Schär later demonstrated with a model that "turned off" the soil-atmosphere feedback: without this mechanism the 2003 heat wave would have been half as long — not 13 days but 6. The drought wasn't background to the catastrophe, it was half of it.
That summer still ranks as the hottest in Europe at least since 1540. The peak hit the first half of August: week 1 produced around 15,000 excess deaths, week two — nearly 24,000. In the peak second week, excess mortality in France reached +96.5%, in Portugal — +48.9%, in Italy — +45.4%. The thermometer in Amareleja, Portugal, hit 47.3°C. And in Paris 35°C held for ten days, four of them — in a row, August 8–11, the first time since 1873.
And here's the detail all this was set up for: on the night of August 11–12, Paris broke the absolute record for minimum temperature — two nights in a row above 25.5°C. The day catches your eye, the night — doesn't. But the night would turn out to be the tragedy's main actor.
The first alarm wasn't raised by the Health Ministry or meteorologists. On August 13, the sanitary surveillance institute InVS saw a "country-scale epidemic" in data from... funeral homes: the largest chain, Pompes Funèbres Générales, reported a 50% surge in demand for its services in Paris. When people die from heat, someone writes "cardiac failure" or "nothing specified" in the "cause of death" box — the person died at home, after all, nobody chased them down the street with a thermometer. "People don't come with 'dying from heat' written on their foreheads," a Health Ministry representative dryly explained later.
The real count came only through comparison: for August 1–15 in metropolitan France, 32,065 deaths were recorded versus 20,630 expected from 2000–2002 demographics. The difference — 11,435, a 55% increase over baseline. Then the toll kept growing all year: 14,802 by official INSEE tally, some reaching as high as 19,000. And in 2007 Jean-Marie Robine and colleagues compiled daily data from 16 countries and arrived at what now appears in every textbook: more than 70,000 excess deaths for summer 2003. Before that, estimates ranged from 27,000 to 40,000 — each country counted differently, and Europe understood for the first time that it had no unified pulse.
Two details in these numbers are strategic. First: 65% of excess deaths — women, 81% of investigated heat stroke deaths — people over 75. Second: Robine tested the "harvesting effect" — the hypothesis that heat simply brought forward deaths of those who would have died weeks or months later anyway. Not confirmed: both before summer and after August, mortality remained elevated. These weren't deaths "stolen from autumn." These were 70,000 added graves.
Sleep physiology is a narrow technological window. To fall asleep, the body must shed about a degree from its core: skin vessels dilate, heat flows from center to periphery, heart rate drops, the nervous system reboots. A hot bedroom blocks this cascade. And crucially — a sleeping person, unlike someone awake, is helpless: can't get up and open a window, pour water, move to a cool room. Everything the body knows how to do in daytime heat is unavailable at night.
In 2011 the Dousset group did an elegant thing: took 61 NOAA satellite thermal images from August 1–13, 2003 (1 km resolution), overlaid them with a case-control study of 241 deceased people over 65 and an equal number of matched controls. The result simple and merciless: death risk doubled where nighttime surface temperature was higher — every +0.41°C in average nighttime temperature gave OR 2.17, and nighttime temperature of the week of death — OR 2.24. Daytime maxima, by comparison, meant nothing: OR 0.87–1.05, the interval covers unity. Parks meanwhile worked: each step of vegetation down in the image gave −0.2°C even in the middle of the canicule.
The city itself was an accomplice. Concrete, asphalt and roof tiles accumulate during the day and don't release at night — the heat island turned a summer night into a heated oven with no way out. Paris's minimum nighttime temperature of 18.8°C in August 2003 versus a norm of 15.7°C looks modest against "39.5 during the day," but it was precisely the deadly variable. Medicine later formulated this directly: most died not on the street under the sun, but at home and in nursing homes — somewhere between dinner and dawn.
The canicule also exposed that Europe had been building without thinking about heat. Britain on August 10 crossed 100°F for the first time in history — 38.5°C in Faversham, Kent, a record that held until 2019. That day asphalt melted on the M25, and rails buckled from thermal expansion — the railway system's hottest point in 13 years.
France stumbled into a more subtle problem: 80% of its electricity came from 19 nuclear plants, which are cooled by river water. The rivers heated up so much that discharging spent water into them became impossible — and the plants began cutting capacity, while demand was meanwhile being sung by air conditioners. The government chose a compromise that had ecologists clutching their heads: allowed plants to discharge water hotter than normal, "on a case-by-case basis." Switzerland did the opposite — cut generation to avoid heating rivers. The low Rhine additionally cut off cargo barges: throughput dropped 70–80%. Climate stress showed Europe for the first time a contradiction in the foundation: fish need cool water, grids need stable baseload generation, and heat forced a choice.
Even death didn't fit in this infrastructure. Wards and morgues designed for normal mortality were overwhelmed: hundreds of bodies stood in refrigerated trucks on the streets, Le Figaro counted 300–400 unclaimed in Paris alone, the state "in the greatest secrecy" requisitioned about 4,000 m² of refrigerators at the Rungis wholesale market — hundreds of temporary cots in a hall where carcasses usually hang. A symbol you couldn't make up: Europe's largest food market for one week in August worked as a morgue.
The reaction was painful but swift. Already in fall 2003 France began building what it didn't have at all: a canicule plan. From 2004 SACS has been operating — a biometeorological warning system of InVS and Météo France: thresholds for maximum and minimum temperatures, calibrated to link with excess mortality, first in 14 pilot cities, then nationwide. Three days before peak — alert, mobilization, calls to the elderly. Plus mandatory municipal registries of isolated vulnerable people, whom social services must call and visit. Europe, accustomed to responding to floods and epidemics, for the first time established a "registry of those who will be killed by weather."
Does it work? Now there are precise numbers. A 2025 meta-analysis of 102 cities in 14 countries showed: implementing warning plans cuts excess mortality from heat by 25% — about 1.8 lives saved per 100,000 residents per year, 14,551 saved in the sample. A natural experiment happened earlier: the July 2006 canicule was nearly like 2003, but the model calibrated to previous vulnerability predicted 6,452 deaths — 2,065 actually occurred. Two-thirds of expected mortality simply didn't happen. Moreover in Paris, comparison of periods before and after 2003 shows risk reduction at extreme temperatures of 15–50% accounting for delayed deaths — the city actually became more resilient, and not just thanks to the state: people bought fans, air conditioners, learned to close windows during the day.
Canicule-2003 also became a genre founder for climatology. Peter Stott and colleagues in Nature 2004 for the first time in history attributed a specific weather event to the human factor: very likely (>90%) that anthropogenic warming at least doubled the risk of a heat wave of this magnitude. Back then it was a scientific feat, now — routine for World Weather Attribution. But that chapter made the first trial run.
Now overlay the present on the 2003 story. Copernicus defines a "tropical night" as a night when temperature doesn't drop below 20°C. In 2024 their number in Europe became the second highest on record, and in winter-2025 a minimum of 95% of the continent lived above annual average norm. Heat deaths in Europe in 2022 — 61,672, in 2023 — 47,690: the scale of 2003 hasn't gone anywhere, it's just smeared across the calendar.
"Climate insomnia" proper has already been counted down to hours. Client data from nearly 50,000 Fitbits showed: each degree above threshold steals sleep progressively. Climate Central in July 2026 summed this into a number: the average planet inhabitant loses ~56 hours of sleep per year due to warm nights — seven full nights, of which about five hours per year is the direct contribution of anthropogenic warming, which has doubled since the 1970s. Nights are warming faster than days by simple physics: the near-surface air layer at night is short and compact, all daytime heat is trapped in it like in a greenhouse under a blanket. In Hong Kong twelve hours of ~30°C produce +3.1% emergency hospitalizations, among the elderly and poor — 5.3%. In Delhi nighttime minimum temperature reaches 32.4°C — Paris-2003 with its 25.5°C would be embarrassed to even compare.
Medicine in 2026 formulates the problem exactly as Paris formulated it in August 2003, only now with a five-headed research program: need to define a "recovery night," build warning thresholds based on nighttime temperatures, count accumulation of multi-day sleep deprivation, measure inequality of access to cooling. The question isn't "what will the day be like," but "will the night allow the body to recover." France in 2003 answered with 70,000 graves. The rest are answering on schedule.
My favorite discovery in this story — isn't about climate, but about epistemology. The canicule was detected not by sensors or meteorologists, but by funeral homes: the most accurate mortality sensor turned out to be darkly social. Millions of the dead still have something neutral in the cause-of-death field — heat leaves no signature. Worth remembering every time "there's no official statistics on X": sometimes what's missing isn't the event, but the way to record it.
Second: the entire drama of 2003 happened because of a parameter almost nobody monitors — minimum nighttime temperature. We optimized reporting for the most spectacular number, the daytime maximum, while the modest nighttime line in the table was killing. This reminds me of the eternal engineering trap: monitor peak load and sleep through nighttime drift. The Paris night with OR 2.17 — ready argument in any dispute about which metrics actually matter.
Third — cautious optimism. In 23 years Europe has proven adaptation works: −25% mortality from warning plans, −2/3 in the 2006 experiment. But heat also works: the planet's sleep hours keep draining, tropical nights creep north, and adaptation has an entry price — registries, air conditioners, green neighborhoods. Canicule-2003 showed that night can kill. It also showed that societies can learn. The only dispute is about what will happen faster — and stakes on that are no longer in degrees, but in hours of sleep for each of us.