UDA-city — Where is heat most dangerous to people?

In two sentences. In this hot-humid city, the hottest air is in the leafy, near-empty periphery — but the danger to people is concentrated in three dense informal settlements where many vulnerable residents have little defence against the heat. A warmer future makes the same places far more dangerous, and low-cost cool-roof + greening measures meaningfully help — but only if we judge them by real heat avoided, not by a relative score.

The headline finding. Jade Gardens — a low-rise green periphery — runs the hottest air of all ten, yet sits at the bottom of the risk ranking. The highest risk is Kampong Lama, a dense informal settlement that is only the third-hottest. Heat is a hazard; risk is what that hazard does to exposed, vulnerable people — and here the two point in opposite directions.

The city at a glance

Schematic map of the ten UDA-city neighbourhoods grouped by type and shaded by heat-risk index; hotspots are brightest (highest risk), refuges and most cores are darkest (lowest risk). — Each tile is a neighbourhood, shaded by its heat-**risk** index (bright = high risk). The four dense *hotspot* settlements carry almost all of the people-risk; the hottest place of all, Jade Gardens (☀), sits at zero risk because almost nobody lives there.

1. Hottest ≠ highest-risk

Hazard and risk are different questions. One is about the air; the other is about the people in it.

Scatter of dangerous-heat hours versus risk index for the ten neighbourhoods, showing the hottest neighbourhood at the bottom of risk and the highest-risk neighbourhood only mid-range on heat. — Horizontal = how hot (hours above 35 °C). Vertical = how dangerous to people (risk index). The hottest place (Jade Gardens, far right) is at the floor of risk; the highest-risk place (Kampong Lama, top) is only mid-pack on heat.

Why the divergence? It comes down to urban form:

Why Jade Gardens is hottest: it is smooth and low-rise. Aerodynamically smooth surfaces mix the air weakly, so daytime heat builds up near the ground — but almost nobody lives there (80 people per hectare, the lowest), with more air-conditioning and less deprivation. High hazard, near-zero exposure → low risk.
Why Kampong Lama is highest-risk: a dense informal settlement — still genuinely hot, but packed with 300 people per hectare, only 8% air-conditioning access, 62% outdoor workers and high deprivation. High on all three pillars of risk at once.

2. The risk table

Risk = Hazard × Exposure × Vulnerability, each scored 0–1 across the ten neighbourhoods and combined as a (deliberately cautious) geometric mean — a near-zero in any one pillar pulls risk down.

Horizontal bar chart of hazard, exposure and vulnerability pillar scores per neighbourhood. — The three pillars. Kampong Lama (top) scores high on all three; Jade Gardens has high hazard but its exposure bar is zero, which collapses its risk.

Rank	Neighbourhood	Type	Danger hrs (T2>35°C)	Hazard	Exposure	Vuln.	Risk index
1	Kampong Lama	hotspot	42	0.67	1.00	0.95	1.00
2	Dhobi Lines	hotspot	26	0.40	1.00	0.92	0.83
3	Fuzhou Lanes	hotspot	22	0.33	1.00	0.97	0.80
4	Mlima Moto	hotspot	5	0.05	1.00	1.00	0.43
5	Lusitano Square	core	5	0.05	0.77	0.09	0.18
6	Victoria Exchange	core	5	0.05	0.77	0.06	0.15
7=	Jade Gardens	refuge	62	1.00	0.00	0.32	0.00
7=	Taman Melati	refuge	47	0.75	0.00	0.36	0.00
7=	Serendib Rise	refuge	26	0.40	0.00	0.27	0.00
7=	Zheng He Towers	core	2	0.00	0.77	0.00	0.00

Take-away for action: the four hotspot settlements carry essentially all of the people-risk. The hottest refuge areas are a heat-management issue, not a life-safety priority — provided they stay sparsely populated.

3. What humidity and a +2.5 °C future change

Humidity makes everywhere more dangerous — but doesn't change who's worst

A dry-bulb 35 °C threshold understates danger in a city at ~81% humidity. Re-scoring the hazard with a humid-heat measure (wet-bulb temperature) shows every neighbourhood endures 90–265 hours of dangerous humid heat — far more than the dry-bulb count suggests.

Grouped bars comparing dry-bulb dangerous hours with the much larger wet-bulb dangerous hours for every neighbourhood. — Counting humid heat (wet-bulb > 28 °C, orange) reveals far more dangerous hours than dry-bulb alone (blue) — in *every* neighbourhood. Humidity raises the danger everywhere together, so the *ranking* is unchanged; the absolute danger is much higher.

One lesson we learned the hard way: switching to a humid-heat metric without also raising the threshold makes it "dangerous almost always" and destroys the signal — the metric and its threshold must move together.

+2.5 °C warming multiplies the hazard ~8-fold

Present versus future dangerous-heat hours by neighbourhood, future bars far taller. — Dangerous-heat hours, present vs a +2.5 °C hotter future. City-wide the total rises from 242 to 1,861 hours (×7.7). The ordering barely moves — warming hits everywhere.

Under the warmer future the people-risk ranking is stable (only a minor 2↔3 swap), because uniform warming raises hazard everywhere while exposure and vulnerability are unchanged. The message is not "the map reshuffles" — it's "the same places get far more dangerous, and the highest-risk settlements stay highest-risk."

4. A targeted intervention: cool roofs + street greening

We tested one concrete action on the three highest-risk settlements: brighten the roofs and pavements (reflect more sun) and convert some paved/bare ground to grass and trees (add evaporative cooling).

Dangerous-heat hours before and after the intervention for the three treated neighbourhoods, with reductions of 60 to 73 percent. — Cool roofs + greening cut dangerous-heat hours by 60–73% in the three treated settlements.

It works on the physics. Reflective surfaces cut absorbed sunlight, greening adds evaporative cooling, and together the sensible heating of the air falls by about a quarter — which is what cools the street. Dangerous-heat hours drop 42→17 (Kampong Lama, −60%), 26→9 (Dhobi Lines, −65%) and 22→6 (Fuzhou Lanes, −73%).

A scoring caution worth shouting about. The relative risk index barely moves after this intervention — Kampong Lama stays pinned at 1.00 — even though its real danger fell 60%. That is an artefact of scoring everything relative to the worst neighbourhood: improving the worst case simply re-baselines the scale. Judge interventions by the absolute hazard (hours of dangerous heat avoided), not by a relative index designed to always keep someone at 1.0.

5. Where this bridge holds — and where it breaks

The honest part. This pipeline is a decision aid, not an oracle.

✓ Where it holds

The headline is robust. Hottest ≠ highest-risk survives every humid-heat metric and the +2.5 °C future — a structural feature of the city, not an artefact of one threshold.
Relative ranking of the hotspots is stable and physically explainable (form → mixing → heat; density + low coping → risk).
The physics of the intervention is sound and traceable to the surface energy balance (less absorbed sun, more evaporation, less sensible heat).
Direction of travel — who to prioritise, and that warming worsens the same places — is trustworthy.

✗ Where it breaks

It is an environmental hazard, not a health outcome. Hours above a temperature are a proxy for danger, not a prediction of illness or death.
The socio-economic layer is synthetic. Read ranks, never absolute numbers; these are plausible magnitudes, not survey data for a real place.
Relative (min–max) scaling doesn't transfer and hides progress — it inflates the apparent risk of untouched areas when the worst case improves, and pins sparsely-populated places at zero risk no matter how hot they get.
District averages hide individuals — the most exposed people inside a "low-risk" area are invisible here.
The future is a pseudo-warming stress test, not a downscaled projection; and the greening benefit assumes the vegetation is kept watered.

6. So what should a city do?

Target the three informal settlements first. Kampong Lama, Dhobi Lines and Fuzhou Lanes are where dangerous heat meets dense, vulnerable, under-protected populations — they hold essentially all of the people-risk.
Deploy cool roofs + street greening there. A low-cost, proven lever: 60–73% fewer dangerous-heat hours in our runs, working through reflected sunlight and evaporative cooling.
Plan for the future now. +2.5 °C multiplies dangerous heat ~8-fold in the same places — adaptation is not optional.
Measure success in heat avoided, not in a relative score, and never read a "0" risk for a sparsely-populated area as "safe."

7. Methods & reproducibility

Model: SUEWS via supy 2026.6.5 — NARP net radiation + classic OHM storage heat (single-layer, laptop-runnable); 10 grids, present hot-humid forcing.
Hazard: dangerous-heat hours = hours with hourly-mean 2 m air temperature (T2) above 35 °C, after a 14-day spin-up. Humid variant: wet-bulb (Stull 2011) above 28 °C from the model's own T2+RH2.
Risk: UNDRR-style hazard × exposure × vulnerability, each min–max scaled to 0–1 and combined as a geometric mean (per risk_bridge.py).
Intervention: roof albedo 0.20→0.60, pavement 0.20→0.40; grass +0.10 and trees +0.07 of land cover (from paved/bare), on gridiv 4/5/9 only. Config validated before running.
Source & data: config uda-city.yml, bridge risk_bridge.py / caveats risk_bridge.md, scenario uda-city-intervention.yml — all in this repository. UDA-city is synthetic (realistic structure, not a real place); the socio-economic layer is synthetic — read ranks, not absolute values.

Modelled with SUEWS (Surface Urban Energy & Water Balance Scheme) via supy 2026.6.5. Figures are colour-blind-safe (Okabe–Ito / cividis). Risk decomposition follows the UNDRR hazard × exposure × vulnerability framing; see risk_bridge.md for full assumptions.

Please cite SUEWS: Järvi, Grimmond & Christen (2011), J. Hydrology 411, 219–237; and Ward, Kotthaus, Järvi & Grimmond (2016), Urban Climate 18, 1–32.