Find Population on a Map

How the estimate is built

The world is covered by a grid of cells roughly 100 metres across, each carrying a population count. When you close a shape on the map, the counts of the cells falling inside it are added up. That sum is the number you see.

The grid is WorldPop’s global population dataset, produced at the University of Southampton, for the year 2020. Its method is called top-down dasymetric modelling, and it works in two stages: start from the census counts each country reports for its smallest published administrative units, then redistribute each unit’s people within its own boundary according to where people arelikely to be — inferred from satellite-detected buildings, road networks, land cover and night-time lights.

That second stage is what buys the 100 metre detail, and it is also the thing to keep in mind. The national and district totals come from real censuses. The distribution withina district is a model’s best estimate, not a measurement. The figure is usually far more realistic than spreading people evenly across a district would be — but the fine-grained precision is partly inferred rather than observed.

Reading the number honestly

1. It describes 2020, not today. For a fast-growing city, or anywhere that has seen significant migration since, the figure will be low.
2. The detail is modelled, not observed. A district total is anchored to a census. Where those people sit within the district is the model’s inference from buildings and roads. It is usually close; it is not a count of who lives on which street.
3. Empty ground is rarely exactly zero. This dataset predicts across all land rather than restricting itself to detected buildings, so remote terrain picks up small fractional values. A 400 km² patch of open Sahara returns about 13 people — which is noise, not a settlement.
4. It counts residents, not people present. Census figures record where people live. A stadium, an office district at 11am, a beach in August and a ski resort in January all contain far more people than their resident count.
5. Census quality varies by country. The grid inherits whatever the underlying national census got right or wrong, including the age of the count and how finely administrative units are published.

How big is a cell, exactly?

The grid is defined in arc-seconds rather than metres: each cell is 3 arc-seconds on a side, one 1,200th of a degree. North to south that is a constant 93 m everywhere. East to west it shrinks as you move away from the equator, because the meridians converge.

So a cell is about 93 × 93 m at the equator — roughly 0.86 hectares — but 93 × 64 m in Ljubljana and 93 × 46 m in Oslo. At high latitude the cells are narrow rather than square, so “100 metre resolution” is really 100 metres in one direction and rather less in the other.

What that looks like in practice

Drawn around the island of Manhattan, this tool returns about 1.57 million against a true resident population near 1.6 million — close enough to be genuinely useful. Drawn around a single city block in Ljubljana it returns around 130, which is the right order of magnitude for a block of flats but should not be read as a headcount of the building.

That is the right way to use it: trust the magnitude, not the digits. If half your shape is a golf course the model will mostly know, because it can see there are no buildings there — but it is inferring, and it can be wrong about any individual hectare while still being close across the whole shape.

Why this tool changed its source

Until recently this page ran on NASA SEDAC’s Gridded Population of the World, which took the opposite approach: it spread each district’s census count evenly across the cells covering it, refusing to guess where within a district people lived. That restraint made GPW coarse — about 1 km cells — but traceable, since every figure led directly back to a census return rather than to a model.

The contract for the CIESIN-managed SEDAC was terminated in April 2025, and the service that answered these queries went offline with it. The underlying datasets were rehoused as downloadable files, but the ability to ask “how many people are inside this shape” on demand was not replaced.

WorldPop is the successor here, and on most measures it is an upgrade: ten times finer, five years more recent, and far more useful at small scales, where GPW’s flat district density was close to meaningless. The trade is the one described above — detail that is modelled rather than measured. If you need a figure whose provenance runs straight back to a census with no model in between, the archived GPW files are still the better source, and you will need to do the zonal statistics yourself.

A third option worth knowing: the European Commission’s GHS-POP disaggregates census counts onto satellite-detected built-up surface and publishes back to 1975, which makes it the strongest choice for studying how a place changed over time.