How it works
Landhi Town is divided into 265 areas of about 0.1 km² each. For every one, the model asks three questions and multiplies the answers together.
The three questions
| Heat | How hot does the ground get on a typical hot-season afternoon? Measured by satellite across five April–June seasons. |
|---|---|
| People | How many people live there? |
| Vulnerability | What makes heat more dangerous here: little greenery, distance from a clinic, distance from a relief centre. |
Why multiply rather than add?
Because all three have to be present for anyone to be harmed. A scorching but empty industrial yard scores near zero, which is the right answer — there is nobody there to help. Adding the three would let the heat and the lack of greenery cover for the fact that nobody lives there.
How sure is it?
The whole model was re-run 1,000 times with the assumptions varied. Of the 53 areas in the top fifth, 28 stayed there in at least 80% of runs and 12 did not survive a coin flip. Roughly half the list is solid and roughly half is not. The Confidence layer on the map shows which is which.
Why it says “provisional”
Two things are unfinished, and both are stated wherever they matter:
- The weights — how much each factor counts — are placeholders. They have not yet been set by considered judgement.
- Distance to a relief centre has no data at all. No centre has been verified on the ground yet, and OpenStreetMap does not list any inside Landhi — which is not evidence that none exist, because OpenStreetMap maps only about a tenth of Landhi’s buildings.
What it cannot tell you
- The people count is modelled, and comes to roughly half the 2023 census figure for Landhi. Use it to compare areas, never as a headcount.
- Satellites measure the temperature of the ground, not the air, and say nothing about humidity — which matters enormously in Karachi.
- Power cuts are not included, although long outages are a large part of why heat harms people here. The published schedules name electrical feeders, not places, so there is no honest way to put them on a map.
- Results describe areas, not the people in them, and are never a verdict on a neighbourhood.
The maths, for anyone who wants it
Each factor is scaled to a 0–1 range by clipping at the 5th and 95th percentile and stretching what remains, so one unusual area cannot squash everything else. The three are then combined as a geometric mean:
Priority = (Heat × People × Vulnerability)1/3
Heat and Vulnerability are given a small floor of 0.01 first, because a zero there is an artefact of the clipping rather than a fact. People is deliberately not floored: a zero there means nobody lives in the area, which is real, so those areas score zero priority by construction.
The full derivation, every limitation, and the independent checks are in the model report.