How a wrong number got embedded in heat safety, and what it means before summer 2026
The Maricopa County medical examiner's office processed 645 heat-related deaths in the summer of 2023. Forty-five percent were people experiencing homelessness. Nearly two-thirds were over 50. Of those who died indoors, 88 percent had an air conditioning unit. Eighty-five percent of those units weren't working.
These numbers are terrible and specific. But there is another number buried in the data that tells a different kind of story, one about the distance between what science knows and what protective systems act on. During Phoenix's deadliest stretch, when daily temperatures never dropped below 91°F for two weeks straight, the wet-bulb temperature peaked at 24.32°C. In the framework that climate science had spent more than a decade using to assess when heat becomes unsurvivable, that reading fell more than ten degrees below the danger line.
By the standard embedded in risk assessments, adaptation plans, and habitability projections worldwide, the conditions that killed 645 people were not supposed to be lethal.
How a number becomes load-bearing
In 2010, Steven Sherwood of the University of New South Wales and Matthew Huber of Purdue University published a paper in PNAS asking whether there was an absolute upper limit to the heat a human body could survive. Their answer was thermodynamic. The body maintains a core temperature around 37°C and sheds excess heat primarily through sweat. Sweat works because evaporation pulls heat away from the skin. But evaporation depends on how much moisture the surrounding air can still absorb. Wet-bulb temperature captures this: it measures how much cooling evaporation can actually provide. At 35°C wet-bulb, Sherwood and Huber calculated, the air is so warm and so saturated that evaporation effectively stops. The body becomes a closed system. Core temperature rises. Organs fail.
Clean logic. A specific number. And the assumptions underneath it were extraordinary.
The 35°C threshold assumed a person who was indoors or fully shaded, unclothed, completely sedentary, fully acclimatized, of average body size, and free of any condition that might impair thermoregulation. Sherwood and Huber were explicit about this. They noted that 35°C was an upper bound and that a lower, more realistic number would apply to actual humans in actual conditions. They estimated the threshold wouldn't be crossed until global temperatures rose by about 7°C, a scenario so extreme it seemed almost academic.
But the number traveled faster than the caveats. This is partly how science communication works: a clean threshold propagates through models and risk assessments in ways that a conditional range never could. Saying "35°C wet-bulb" is actionable. Saying "somewhere between 25°C and 31°C depending on humidity, age, activity level, body composition, medication use, and acclimatization status" is accurate but unwieldy. Over the next decade, the clean number won. 35°C wet-bulb became the benchmark in study after study assessing which regions of the planet would remain habitable under warming. Pal and Eltahir used it for the Persian Gulf. Im and colleagues applied it to South Asia. Raymond and others mapped its global implications. The threshold was cited in over a dozen major papers as the line between survivable and unsurvivable heat. It shaped how scientists communicated risk, how planners evaluated adaptation timelines, how the public understood what "too hot for humans" meant.
The implicit message was: we have time. The threshold is far away.
No one had tested that specific number against a human body.
Inside the chamber
W. Larry Kenney had been studying human thermoregulation at Penn State for decades when the 35°C threshold started appearing everywhere. The physics was fine, as far as it went. It described a theoretical person who didn't exist.
The PSU HEAT Project, funded by the National Institutes of Health, was designed to find out what happens to real people. The protocol: before entering the lab, each participant swallowed a small capsule containing a temperature sensor. Inside an environmental chamber, they performed light physical activity meant to simulate the minimal effort of daily life during a heat wave. Cooking. Walking to the bathroom. The kinds of things you cannot simply stop doing because the weather is dangerous.
Then the researchers raised the temperature, or the humidity, or both, and watched the data from the capsule.
They were looking for a specific physiological moment: the point at which core temperature stops holding steady and begins to climb without stopping. The body is still trying. The sweat glands are still firing. But the heat coming in exceeds the heat going out, and the gap will only widen. The Penn State team calls this uncompensable heat stress. Sustained over six hours, it leads to heat illness or death, even in young, healthy adults under conditions of minimal exertion.
In 2022, Daniel Vecellio, S. Tony Wolf, Rachel Cottle, and Kenney published their findings in the Journal of Applied Physiology. Among young, healthy adults performing light activity in humid conditions, the critical wet-bulb temperature averaged 30.55°C, with a standard deviation of less than 1°C. In hot, dry environments, it dropped further, to between 25°C and 28°C. No participant reached 35°C. Every mean was statistically below it.
The Penn State team found that the critical wet-bulb temperature varies with the specific combination of heat and humidity, dropping most dramatically in hot, dry environments. The "31°C" figure entering public discussion applies to warm, humid conditions; in desert heat like Phoenix's, the real limit for young healthy people is several degrees lower.
A crucial finding, and one that resists simplification: the researchers concluded that no single wet-bulb threshold can capture the human limit across all climatic conditions. The critical temperature varied with the specific combination of heat and humidity, dropping most dramatically in hot, dry environments. The "31°C" figure that has entered public discussion applies to warm, humid conditions. In desert heat like Phoenix's, the real limit for young healthy people is several degrees lower still.
Then the team tested older adults. Seventy-two participants between the ages of 40 and 92. In humid environments above 50 percent relative humidity, the critical wet-bulb temperature for older bodies was approximately 28.5°C. For older women, the vulnerability was even more pronounced.
"Young, fit, healthy people tend to tolerate heat better, so they will have a temperature limit that can function as the 'best case' baseline. Older people, people on medications, and other vulnerable populations will likely have a tolerance limit below that." — W. Larry Kenney
These are laboratory means, averages across a study population. Individual variation is real: body size, fitness, hydration, medication, acclimatization history all shift the threshold. The HEAT-Lim model that grew from this work makes its own simplifying assumptions about metabolic rate and clothing. And some researchers note that 35°C may retain value as a theoretical worst-case ceiling, the temperature above which no human adaptation strategy works regardless of circumstance. But the gap between that ceiling and where real bodies actually fail is the finding that matters for protective systems. Even accounting for individual variation, even granting the uncertainty, the distance between 35°C and what the lab measured is four degrees for healthy young adults in humid heat. More than ten degrees for what Phoenix actually experienced.
Already crossed
In April 2026, Sarah Perkins-Kirkpatrick of the Australian National University and colleagues published a study in Nature Communications that took the Penn State findings out of the lab and into the historical record. Using a physiology-based model called HEAT-Lim, co-developed with Arizona State University, the team assessed whether survivability thresholds had been crossed during six real heat events: Mecca in 2024, Bangkok in 2024, Phoenix in 2023, Mount Isa, Australia in 2019, Larkana, Pakistan in 2015, and Seville in 2003.
Five of the six events were associated with at least a thousand deaths. All six crossed survivability thresholds. And the wet-bulb temperatures during every one of them fell below 35°C.
The range ran from Larkana's 30.85°C to Phoenix's 24.32°C. Phoenix was the sharpest case. It recorded the lowest wet-bulb temperature of any event studied while simultaneously recording one of the highest dry-bulb temperatures, 46.73°C, and producing mass casualties. In the 35°C framework, Phoenix shouldn't have been dangerous at all. The air was hot but dry. Sweat should have worked.
The HEAT-Lim model showed that the 35°C metric systematically misreads dry heat. A body in 116°F desert air absorbs enormous radiative and convective heat loads. Sweat evaporates fast, yes, but the incoming heat overwhelms the cooling. The wet-bulb reading stays low because the air is dry, but the body is losing the thermoregulatory battle anyway. "Extremely hot yet dry conditions," the paper concluded, "are found to be just as deadly as hot and humid conditions."
Perkins-Kirkpatrick told the Guardian her first reaction to the results: "Oh sh*t." She had not expected to find that the line had already been crossed. Repeatedly. In events the prevailing framework classified as below the survivability limit.
Who finds the gap
The 35°C threshold was a theoretical upper bound, clearly labeled as such by its authors, that migrated into operational contexts where it was treated as a measured fact. Climate projections used it to assess when regions would become uninhabitable. Risk frameworks used it to evaluate adaptation timelines. Every downstream system calibrated its urgency to the perceived distance from that line.
When the line is in the wrong place, the urgency is wrong too. And the consequences fall along familiar lines.
Nearly two-thirds of Phoenix's 2023 heat deaths were people over 50, the population whose thermoregulatory limits the Penn State team measured at 28.5°C wet-bulb in humid conditions, and lower in dry heat. People experiencing homelessness accounted for 45 percent of deaths. Their risk of dying from heat was 500 times higher than the general population's. Substance use was a factor in nearly two-thirds of cases. Mental illness appeared in the medical history of at least one in four. Of those who died indoors, nearly all were in uncooled environments, their air conditioning broken or shut off.
The theoretical person encoded in 35°C was young, healthy, resting, shaded, acclimatized, unmedicated. The people dying in the real heat had none of those protections. The calibration gap lands first, and hardest, on the people whose bodies were never represented by the number.
Twenty-nine percent of Phoenix's 2023 heat deaths occurred on days when no excessive heat warning was in effect. The National Weather Service uses heat index rather than wet-bulb temperature for its advisory system, so the 35°C threshold isn't directly embedded in warning criteria. But the broader framing matters. When the scientific consensus communicates that survivability limits are far from being reached, the entire chain of protective response loosens. Cooling centers open on schedules designed around daytime peaks. The people most at risk are the ones whose exposure has no schedule, who are outside at 2 a.m. because they have no inside to go to, whose thermoregulatory limits are lower than any threshold in any guidance system. Over the prior five years, fully 68 percent of Maricopa County's heat deaths had fallen on days without an excessive heat warning in place. The 2023 event, with its unprecedented duration, was actually the anomaly: for once, the warnings were active during most of the dying. The normal pattern is worse.
What arrives next
NOAA's seasonal outlook for summer 2026 projects above-normal temperatures across the western half of the country, with the strongest signal over the interior West, where some areas show a 50 to 60 percent probability of above-normal heat. The Northeast leans warm as well. Behind the temperature forecast sits an 82 percent probability that El Niño will emerge by July, with a two-in-three chance of reaching strong or very strong intensity by winter. NOAA projects it is already very likely that 2026 will rank among the five warmest years on record, before El Niño's full warming effect is even factored in.
The Penn State findings are measurements. What a body does at 31°C wet-bulb, or 28.5°C, or 25°C, is physiology. It doesn't depend on a climate model being right. The Perkins-Kirkpatrick finding is also a measurement: these six events crossed these thresholds and produced these casualties. The NOAA forecast is a probability. But the probability is pointing at the same geography where the calibration gap has already proved lethal.
Yesterday was Memorial Day. Overnight lows in the Phoenix Valley are already holding in the 80s. The Penn State data has been published since 2022. The Perkins-Kirkpatrick study landed in April. Summer is here.
The distance between what science knows and what institutions act on has a body count, and the count is already in the record. It is in the Maricopa County medical examiner's files, organized by name, by age, by the temperature of the day they died, by whether anyone had told them that day was dangerous. Some of those days, nobody had. The number said they were safe.
The research makes a different set of questions possible now. What is the actual thermoregulatory limit for a person your age, at your activity level, in your climate's specific blend of heat and humidity? Is the advisory system you rely on calibrated to a body that resembles yours, or to a theoretical one that resembles no one's? Was the cooling infrastructure around you designed for an afternoon event, or for the kind of sustained, multi-week heat that Phoenix experienced in 2023, where the danger was accumulation, nights that never cooled enough for the body to recover? The measurements exist. The systems meant to protect people are still running on a number that was never checked against the people it was supposed to describe.
- Urban heat equity data: A January 2026 systematic review found that disadvantaged populations face 26–45% higher heat-related mortality risk and 3–4°C greater exposure than affluent communities, even after controlling for income.
- El Niño strength uncertainty: NOAA gives a two-in-three chance of a strong or very strong El Niño by winter 2026–27, but as forecaster Michelle L'Heureux noted, there remains a one-in-three chance of a weaker event whose implications for summer heat would differ substantially.
- Drought compounding heat risk: The lower-48 drought footprint grew to cover 61 percent of the country by mid-April 2026, with 45 states in moderate-or-worse drought, a condition that amplifies heat exposure by reducing evaporative cooling from soil and vegetation.
- The HEAT-Lim model's limits: The physiology-based survivability model used in the Perkins-Kirkpatrick study was co-developed by researchers at the University of Sydney and Arizona State University, and like all models makes simplifying assumptions about clothing, metabolic rate, and wind exposure that warrant scrutiny as it enters wider use.