Phoenix overnight heat · 134 neighbourhoods

Ask which blocks never cool down.

An agent that reads the temperature measurements for you, and tells you when they cannot answer your question.

Respite agent / starting up
18
tracts never dropped below 28 °C all night
58,176
people living in them
0.004
correlation with the index cities use to target help

Try asking

Why the night is what matters

Two tracts, the same night.

One drops below the line and recovers. The other never does.

Loading measured hourly temperatures.
The same numbers as a table
What the number actually measures

Hours of the night, midnight to dawn, that the outdoor air stayed above 28 °C (about 82 °F). Six hours means the whole window: it never once dropped below that line.

Why the night rather than the afternoon

Afternoon maps show the whole city as uniformly hot, because at four o'clock it is. The difference appears after dark, when some ground gives back the day's heat and some does not.

The body sheds accumulated heat when the air is cooler than skin, so a hot night removes the recovery window and the strain from the previous day carries into the next. Overnight minimum temperature predicts heat deaths better than the daytime peak, which is why multi-day heat events are more lethal than single hot days.

Indoors is worse than this

A building that absorbed heat all day keeps releasing it inside for hours after the outside air drops. These readings are outdoor air at head height, so the temperature people actually sleep in is generally higher.

The whole study area

Hours without relief, by neighbourhood.

Click any neighbourhood and the agent explains that one.

The finding

Heat and disadvantage do not line up here.

If the assumption cities target on held, these dots would climb from left to right. r = 0.004.

The same numbers as a table
Why 18 tracts, and not 32

A tract counts as having no relief only if its hours above 28 °C exactly equal the six-hour window. That is the strictest possible reading, and it is sensitive: several tracts dipped below the threshold for under a second, which is interpolation noise rather than a recovery window.

Counted by how much relief still counts as none: 18 tracts at exactly zero, 32 at under a minute, 34 at under fifteen minutes. The page quotes 18 because it is the claim that cannot be argued down, not because it is the robust one. The agent has the whole curve and will give it to you.

Why that flat cloud is the whole point

Heat programmes are usually aimed using a social vulnerability index, on the assumption that the most vulnerable places are also the hottest. In this study area the two are unrelated, so a programme aimed by the index alone will miss hot places and visit cool ones.

We checked the obvious explanations for a null result. It is not range restriction and not a ceiling effect: excluding the capped tracts still gives r = -0.070, and the mid-range gives -0.272.

The consequence

The places a vulnerability-led programme walks past.

How many tracts count as missed depends on where the targeting line is drawn, so here is the whole curve rather than one number:

What this does not show

The claims we tested and could not support.

A tool that only reports what it found is harder to trust than one that says where it stops. The agent reads this list before it answers.

We cannot tell you what an intervention would buy

We sampled surface composition at 50 tracts to test whether pavement and canopy explain overnight heat. Once position is accounted for, they do not: built surface gives t = +0.24, vegetation t = -1.46. So this tool never estimates how much cooling any intervention would buy, and the agent refuses to when asked.

We found position, not mechanism

Latitude and longitude alone explain about 70% of the variation. Something about where a tract sits drives this, but we did not test what. Elevation, drainage and distance from open desert are all untested candidates.

The metric has a ceiling, so the worst places are tied

The measure caps at six hours. Tracts at the cap were above 28 °C for the whole window and we cannot tell which was hottest, so they are not ranked against each other.

One night, one city

15 August 2026, 134 tracts in central Phoenix. Nothing here should be read as applying to another city, season or date.