Hvor mennesker bor: befolkningstæthed 2025

How to read it

Each coloured cell is a place where people live, coloured by how many live there per square kilometre in 2025. Pale yellow is under 216 people per km², the darkest red 22,276 and more. Land left uncoloured is where the model places nobody.

Zoomed out, a pixel is an average. The data has one value per cell of about 1 km. A pixel of the world view covers many of those, and shows their average, empty cells included. So a city whose centre holds 30,000 people per km² can look orange from afar and turn dark red when you zoom in. The classes are fixed; what changes is how much each pixel averages.

Where the numbers come from

WorldPop, a model, not a count. The data is WorldPop’s estimate of the people in every 30 arc-second cell (about 1 km at the equator) in 2025, global mosaic, release R2025A version 1 (doi:10.5258/SOTON/WP00845). WorldPop spreads population totals over the cells with a model; in its words, “Random Forest-based dasymetric redistribution”. The result is “constrained”: people are placed only in cells WorldPop maps as settled. WorldPop calls R2025A an alpha release that may change over the coming year.

In total the file holds 8,196,931,816 people in 51,224,485 settled cells. That total is WorldPop’s, not a census figure.

What we did to the data

People per cell became people per km². WorldPop gives people per cell, and a 30 arc-second cell shrinks towards the poles. We divided each cell’s people by the cell’s area on a sphere of the mean Earth radius (6,371.0088 km), so that cells can be compared anywhere.

Unsettled cells became 0. WorldPop marks every cell it maps as unsettled, the sea included, as having no data. We wrote those as 0 people, which is what the model says of them, and draw 0 as nothing.

Nothing else was changed: the grid, the extent and every value are WorldPop’s. The highest density is 142,936 people per km², in the Nile delta at Damietta, Egypt (31.81° E, 31.42° N).

The seven classes

Head/tails breaks. Densities are very unequal: most settled cells are sparse, and a few are extremely dense. Equal steps would paint almost the whole world in the first colour. We split the settled cells at their mean, then split the denser part (the “head”) at its mean again, and so on while the head was under 40% of what was split (Jiang 2013). That gives the breaks 216, 1,299, 4,058, 8,952, 15,182 and 22,276 people per km². The rule would go on splitting; we stopped at seven classes so each colour stays distinguishable.

The classes hold very different numbers of cells, which is the point of the method: 43,637,092 cells are under 216 people per km², 5,918,561 are in the second class, 1,205,973 in the third, 309,847 in the fourth, 100,267 in the fifth, 34,706 in the sixth and 18,039 in the densest.

Our choices: the classes (above), the colours (six steps of ColorBrewer’s nine-step YlOrRd, leaving out its two palest so the lowest class stands out from the pale land, and a darker red of our own in place of its last), the projection and the frame.

Where it misleads

It is a model. Where censuses are old or coarse, WorldPop’s cells show where the model expects people, not where anyone counted them.

The sparsest cells are not drawn. GeoCanvas stores each value in 65,536 steps across the whole range, 2.2 people per km² apart here. A cell (or an average of cells, zoomed out) under about 1 person per km² falls on the step of 0 and is drawn like an empty cell.

Why not GHS-POP. We first built this map from the European Commission’s GHS-POP (release R2023A, epoch 2025) and dropped it. Around Sarvestan in Fars, Iran, a town OpenStreetMap gives 17,000 people, GHS-POP places 1,044,389 people in a 0.1° box, against 23,072 in WorldPop. Its 2025 values are also a projection made for its 2023 release.

Why this projection

Lambert azimuthal equal-area, centred on 80° E, 25° N: every area keeps its true proportion, so a coloured region’s size on the map is its size on the ground. Centred on Asia, where most people live; the Americas lie on the far edge, squeezed. Open the map to turn it.

What it is made of

Population: Bondarenko M. et al. (2025), The spatial distribution of population in 2015–2030 at a resolution of 30 arc seconds, R2025A v1, WorldPop, University of Southampton, doi:10.5258/SOTON/WP00845, CC BY 4.0, downloaded 28 September 2026. Countries and ocean: Natural Earth, public domain.