↗ The Skin and the Air

A satellite reads the skin of a city and calls it heat.

v0.1 · OPA Air & Climate · call-sign Ground Truth · two temperatures, one mistake

The honest premise

Everyone's heat map is measuring the pavement. Almost nobody's measuring the air.

A satellite passes over a city and reads land surface temperature (Ts) — the skin of the ground: the roof, the road, the parking lot. It's global, it's sharp, it's free. So it got borrowed. For years now, Ts is what "urban heat" maps are built on.

But the skin is not the air. What a person standing on that street actually feels is Ta — near-surface air temperature — and the two come apart. On a summer afternoon the pavement runs hot enough to fry an egg while the air above it is a different, calmer number. Move the heat sideways with a little wind and the air smears out across the whole city; the pavement doesn't.

The map below is real Ta — reconstructed air temperature for U.S. cities. Flip on the satellite skin, click a city, and read the panel. Then ask the only question that matters: does the picture match what the panel actually says?

Averaged across these cities in summer, the satellite skin reports 2.37 °C of urban heat-island warming. The actual air? 0.15 °C. The satellite overstates the daytime effect by more than fifteen times — and that's the number a lot of heat policy has been quietly built on.

Whose class this is

Cmdr. Apollo Thorpe

Officer Development · on loan to the Air & Climate · call-sign Space Bolts

He teaches one thing, over and over: read the board before you trust it. He came up watching cadets ship an alarm that fired on the wrong variable and got lucky on one case — and he's been chasing that same mistake across every node since. This is the same bug in a new coat. A bright satellite hotspot looks like danger. It confirms what you walked in expecting: cities are hot, here's the heat, done. So it earns a trust it didn't earn.

"How hot is the ground" is not "how hot is the air" is not "who's actually in danger." Three different questions. Most heat maps ship the first and read it as the third.

"The satellite's screaming about a rooftop. Meanwhile grandma two neighborhoods over is the one cooking, and your bright red map's got nothing on her. Read the panel. When the color doesn't match the number — that's the tell."

The board — U.S. cities, air vs. skin

Zoomed out: national picture. Zoom past a state or two to break it into cities. Click any city to read its panel.

Read the panel — selected city

— pick a city

daytime · 9-yr mean

The gap · Ts − Ta

—°C

Skin vs air

—

Over-spread · Δσ

—K

Pick a wiring below, then click a city. Watch the skin run hot above the air by day, flip below it at night — and watch the satellite exaggerate how uneven the heat looks.

How the map is wired

Daytime gap — Ts − Ta at 1:30 PM. Every U.S. city runs positive: the skin is hotter than the air, most of all in the dry west. Nighttime gap — at 1:30 AM every city flips negative: the ground sheds heat faster than the air, so the skin is now cooler. Watch the colors invert. Over-spread — how much more uneven the skin looks than the air (σ_Ts − σ_Ta). This is the satellite exaggerating hotspots. Show city labels

What the colours mean

Big gap— the satellite overstates the heat here by a lot (dry western cities: Phoenix, Dallas)

Some gap— skin runs warmer than air, moderately

Small gap— skin and air are close (humid eastern cities)

Flipped— air is warmer than skin (nighttime, and a few pixels)

Circle size = size of the gap. Circle fill answers one question: how badly does the satellite overstate the heat right here? The dry-west / wet-east split falls right out of it.

Add a page to the notebook

Swap in your own numbers. Recomputed the gaps yourself, or added city names? Drop your us_cities.json here and the map redraws — stays on your screen only. Nothing uploaded, nothing shared, gone when you refresh.

Loaded 384 cities — real Ts−Ta gaps from the paper's Figure 3 source data.

What's real, what's mine

● Real. Every circle is the paper's own published number. The Ts−Ta gaps (daytime and nighttime) and the σ over-spread come straight from the Figure 3 source data of Zhang et al. 2026 — 384 CONUS cities, 9-year means, exact coordinates. Nothing here is modeled or estimated by this lab. The dry-west / wet-east split, the universal daytime overstatement, and the total nighttime sign flip are all in the numbers, not added by me.

◐ Partial. The source data ships coordinates but no city names — the figure didn't need them. 14 cities are named by matched coordinate (NYC, LA, Phoenix, Chicago, Dallas, Houston, Atlanta, Detroit, Boston, Philadelphia, DC, Miami, Minneapolis, Seattle); the rest are labeled by their lat/lon until a geocoder fills them in. Every point is real and correctly placed — only some labels are pending.

◐ Mine. Commander Apollo Thorpe is an authored OPA character; the "read the panel" framing is a story wrapped around a real finding. The map is a teaching render, not a climate model.

Source data: Zhang, Y., Zhao, L., Chakraborty, T.C., Mazumdar, P., Zhang, K. & Gentine, P. (2026). Transfer learning reveals large discrepancies between air and land surface temperatures in cities. Nature Communications. doi:10.1038/s41467-026-73716-7 · Fig. 3 & Fig. 2a source data files · full U-HAT dataset (CC BY 4.0): zenodo.20057651 · basemap © OpenStreetMap © CARTO / Esri.