Data & methods

Overview

The Urban World measures how many people live in the world's urban centres, how densely they live, how much land they build on, and how all of that changes between 1975 and 2100. Every city figure comes from one release: the GHSL WUP Projections Data Package 2025 (GHS-WUP R2025A), published by the European Commission's Joint Research Centre (JRC).

Years up to 2020 are estimates built from satellite imagery and census data. 2025 is the current estimate. From 2030 onward every figure is a projection made by a model, and the site labels it as such (see Observed, current and projected).

The site shows all 21,777 urban centres in the release, each for the years it exists. Nothing is filtered out: figures that look suspect carry a confidence grade and are listed on a review page rather than hidden.

This page follows the data in the order it is processed: the two datasets you can choose on the map, the sources behind them, how cities are defined, what is projected, the grids, the city statistics, built-up surface and confidence, the review flags, and the radial density profiles.

Datasets

The map can show either of two datasets. Both are built from the same GHS-WUP R2025A population and built-up grids, cover the same 26 epochs from 1975 to 2100 in five-year steps, and include the same 21,777 urban centres. Neither filters any city out. They differ in the grid they count on and in the analysis built on top.

The two datasets can give slightly different totals for the same city, because a hexagon and a square do not cover the same ground along a city's edge. See Grids.

Urban World v2

Urban World v2 is the default dataset. It moves the GHS-WUP R2025A population and built-up grids onto H3 hexagons, which makes per-city heatmaps and radial density profiles possible. It covers 26 epochs, from 1975 to 2100; years after 2025 are projections.

What it includes

• H3 population and built-up grid: population and built-up surface summed into Uber's H3 hexagons at resolution 8 (~0.74 km² per cell), for every populated cell within 30 km of an urban centre. See Grids.
• City statistics: population, area, density, built-up surface and JRC's confidence grade for every city in every epoch it exists. See City statistics.
• Per-city heatmaps: population and built share per H3 cell within 30 km of each city, split into history (1975 to 2020) and projection (2025 to 2100) so a page loads only the years it shows.
• Radial density profiles: how population density and built-up share change with distance from the city centre, which you can re-centre on any cell. See Radial density profiles.

GHSL R2025A

GHSL R2025A shows the GHSL WUP Projections Data Package on its native 1 km grid, with city statistics computed directly from the published cells. Use it to see the JRC values without regridding, or to check a figure from Urban World v2. It covers 26 epochs, from 1975 to 2100; years after 2025 are projections.

What it includes

• 1 km grid: GHS-WUP-POP population and GHS-WUP-BUILT-S built-up surface on the original 1 km equal-area cells.
• City statistics: population, area, density, built-up surface and JRC's confidence grade, per epoch and per city, computed from the cells inside each city boundary.

Heatmaps and radial profiles are not available for this dataset. See Global Human Settlement Layer for the source products.

Sources

Every city figure on the site comes from the GHSL WUP Projections Data Package 2025. The world population total is the sum of its population grid, whose national totals follow the UN World Population Prospects 2024. The UN's own urban population, counted by national definitions, is shown as a reference point beside our urban-centre total up to 2030.

Global Human Settlement Layer (GHSL)

The Global Human Settlement Layer is produced by the European Commission's Joint Research Centre (JRC). It provides global, open, multi-temporal data on human presence on Earth. The site uses one GHSL release, the GHSL WUP Projections Data Package 2025 (GHS-WUP R2025A), the JRC's contribution to the UN World Urbanization Prospects 2025. We use three of its products:

• GHS-WUP-POP: population grids (residents per cell), 1975 to 2100 in five-year steps. We use the 1 km version, in the Mollweide equal-area projection, for both datasets.
• GHS-WUP-BUILT-S: built-up surface grids (km² of built-up surface per cell), on the same grids and epochs. See Built-up surface.
• GHS-WUP-MTUC, the Multi-Temporal Urban Centres: a boundary for each urban centre at each epoch it exists, with a statistics table giving its name, country, population-weighted centre and JRC's confidence grade for each year.

Temporal coverage

26 epochs: 1975, 1980, 1985 and every five years to 2100. The years up to 2020 are estimates from satellite imagery and census data; 2025 is the current estimate; 2030 to 2100 are projections. See Observed, current and projected.

How the grids are made

For 1975 to 2020 the population grid is the GHSL 2023 population grid, rescaled so that national totals follow the UN World Population Prospects 2024 instead of 2022, with corrected inputs in seven countries (DR Congo, Eritrea, Ethiopia, Myanmar, Nepal, Paraguay and Zimbabwe) where the earlier release had anomalous trends. The built-up grid for the same years is the GHSL 2023 built-up surface aggregated to 1 km, with only the total built-up surface and no residential split. The built-up surface is mapped from Landsat and Sentinel-2 imagery; census-based population totals are then distributed into cells using that built-up information. So the historical grids are the same satellite-derived built-up surface and the same census-based population allocation as the previous GHSL release, with national totals nudged to the newer UN series.

From 2025 onward every value is a projection produced by CRISP, the Cities and Rural Integrated Spatial Projections model (Jacobs-Crisioni et al. 2025). CRISP starts from the UN's national population projections and splits each country into "functional areas", large sub-national regions (about a thousand worldwide). Each functional area gets its own population path, extrapolated from its recent trend and converging on the national trend over a century, and an expected amount of new built-up surface derived from how much built land per person that area has been consuming.

Within each functional area CRISP then allocates the new built-up surface to 1 km cells, decade by decade, according to a suitability score built from terrain, distance to water and roads, flood, earthquake and landslide hazard, the existing settlement class and the built-up pattern from the previous step. Population change is allocated afterwards, using the modelled built-up surface as the guide for where people go. The resulting grid is reclassified into urban centres, towns and rural cells, which is how the urban-centre boundaries for 2025 to 2100 are drawn.

Everything after 2020 is a scenario, not an observation. The national totals are the UN's medium projection, the sub-national split is a trend extrapolation, and the cell-level pattern is a rule-based allocation. Two urban centres that look alike in 2020 can diverge in the model only because their functional areas had different recent trends.

How we use it

The population grids are the source of every city population and density figure, and the built-up grids of every built-up figure. GHS-WUP-MTUC decides what counts as a city, where its boundary is at each epoch, where its centre is, and how confident JRC is in its population. See Defining cities, Grids and Confidence.

JRC notes that these urban centres are a new, independent dataset: their IDs and figures cannot be matched one to one with the 2024 Urban Centre Database, and the 2025 and 2030 values of the two releases should not be compared.

How to cite

GHSL data are released under CC BY 4.0. JRC asks that each dataset be cited together with the papers describing its method.

Method, estimates 1975 to 2020 and the multi-temporal urban centres:

Method, projections 2025 to 2100:

Method, urban-centre population before 1975 (asked for by the GHS-WUP-MTUC citation):

The data package as a whole:

Data and documentation: human-settlement.emergency.copernicus.eu.

Data used in the guide

The guide's city population, density and area readings come from the site's native 1 km grid export. The city statistics extract preserves the values and source information used in those scenes.

The Guildford and Godalming example uses 2020 population on the 1 km Mollweide grid (GHS-WUP-POP), 2018 land-area estimates (GHS-LAND) and the published 2020 urban-centre footprint (GHS-WUP-MTUC). The source cells and extraction metadata include the dataset identifiers, extraction date and licence. These extracts remain available to download; the guide's source links open this explanation first.

UN World Population Prospects

The World Population Prospects (WPP) is produced by the Population Division of the United Nations Department of Economic and Social Affairs (UN DESA). It is the official source for national population estimates and projections, revised every two years.

What it measures

Total population under the de facto definition: all residents of a country regardless of legal status or citizenship. Values are midyear estimates, from 1950 to the present, with projections to 2100. The projections the site relies on are the medium variant.

How we use it

The "World Population" figure is the sum of the GHS-WUP population grid over the whole globe. JRC rescales that grid so each country's total follows WPP 2024: the estimates up to 2020 and the medium-variant projection from 2025 to 2100. The world figure therefore tracks the UN's, about 8.2 billion in 2025 and peaking near 10.3 billion in the 2080s, while being consistent with the city figures, which come from the same grid.

Because city and world totals share one grid, the share of the world living in urban centres is a like-for-like ratio in every year. After 2025 both are projections.

UN World Urbanization Prospects

The World Urbanization Prospects (WUP) is produced by the Population Division of the United Nations Department of Economic and Social Affairs (UN DESA). It provides estimates and projections of urban and rural populations for all countries. The GHSL data package the site uses was produced by the JRC as a contribution to the 2025 revision.

What it measures

The UN's urban population figure is based on national statistical definitions of "urban". Each country sets its own criteria. Some use population thresholds; others use administrative boundaries, density, or economic activity. The UN adds these national figures into a global total.

Why it differs from our total

The Urban World counts the people living inside urban centres, as defined by the GHSL grid (see Defining cities). That total is lower than the UN's urban population.

National definitions are broader. Many countries classify small towns, peri-urban areas and administrative units as urban even where the population grid shows sparse settlement. An urban centre is stricter: a contiguous cluster of 1 km cells each with at least 1,500 residents per km² of land and 50,000 residents in total, which leaves out towns and lower-density suburbs that national statistics count as urban. The unit of analysis differs too. The UN counts people living in nationally defined urban areas, while we count people living inside physically delineated city footprints.

The gap is not an error. The two sources give different answers to the question "what is urban?"

How we use it

The site's headline is the urban-centre population, which is consistent with every city figure and with the world total. The UN's urban population by national definitions is shown beside it as a reference point, from 1975 up to 2030: World Bank Open Data (indicator SP.URB.TOTL) up to 2020, and the UN projection for 2025 and 2030. It is not shown after 2030.

Defining cities

A city on this site is an urban centre from GHS-WUP-MTUC, the JRC's multi-temporal urban centres. JRC identifies urban centres from its own population grid with the Degree of Urbanisation method, so the definition is the same everywhere, independent of administrative borders or national definitions of "urban". It is also why our urban-centre total is lower than the UN's urban population (see UN World Urbanization Prospects).

The rule

For each epoch, an urban centre is a cluster of 1 km cells that:

• each have at least 1,500 residents per km² of land;
• touch along an edge (cells that meet only at a corner are not connected);
• together hold at least 50,000 residents.

The cluster's outline is then smoothed, and holes smaller than 15 km² inside it are filled, so a park or a rail yard in the middle of a city stays part of it. Clusters stop at national borders, except for Gibraltar, Monaco, the Vatican, Macao and Hong Kong. Only population decides: built-up surface plays no part in drawing the boundary.

21,777 urban centres over time

The release follows 21,777 urban centres across 1975 to 2100. Each has a boundary for every epoch it exists in, so its footprint can grow or shrink, and all of a city's statistics for an epoch use that epoch's boundary.

Of the 21,777, 12,140 exist in 2025, 1,422 no longer exist as separate centres by 2025, and 8,215 appear only after 2025, in the projection. You can browse them by country.

Lifespans

A city is born when a cluster first meets the rule, and it dies if it later falls below it. Its lifespan runs from its first to its last epoch with a boundary, without gaps. Outside that span a city has no figures: its page shows the years it exists in instead of numbers.

JRC links boundaries across epochs by their overlap. When clusters merge or split between two epochs, JRC assigns each piece to one city in its list, so a single city ID can cover what were several clusters in another year.

JRC also estimates urban-centre populations back to 1950, without boundaries. The site starts at 1975, the first year with boundaries.

Why some metros are several centres

Because the rule uses population density alone, a metropolitan area whose suburbs have fewer than 1,500 residents per km² breaks into several urban centres: dense cores separated by lower-density ground. Each is a separate city here, with its own page, figures and rank. This is what an urban centre is in this release, not an error to repair, and the site does not glue the pieces back into a metro.

Some examples, with 2025 populations:

• South Florida is three urban centres: Miami (about 2.9 million), Fort Lauderdale (1.4 million) and West Palm Beach (0.6 million).
• The Pearl River Delta holds Guangzhou (27.6 million) and Shenzhen (13.9 million) as two centres.
• Detroit (0.8 million) is one of four centres, with Sterling Heights (0.4 million), Lincoln Park (0.2 million) and Dearborn Heights (0.2 million), all within about 20 km of central Detroit.

Earlier GHSL releases used an optional step that also admitted cells at least half covered by buildings, which let built-up but thinly populated suburbs bridge the cores into one centre. JRC dropped that step in this release so that the settlement model runs on population alone and can be projected to 2100, where the built-up grid is itself model output. Dense, compact metros are barely affected; sprawling ones, common in North America, southern Africa and parts of China, are split.

City pages list the urban centres whose footprint lies within 30 km of the city's own (up to twelve, largest first), as a way into the rest of a metro: Miami's page lists Fort Lauderdale, Guangzhou's lists Shenzhen.

Names and countries

Names come from JRC. Where JRC gives a local name and an English one, such as "Tōkyō (Tokyo)", the site shows the local name first and searches both. Where JRC gives no name, the site uses the nearest place in GeoNames and marks the centre as unnamed. Countries follow the UN country codes JRC uses, so Hong Kong, Macao and Kosovo are listed separately.

Observed, current and projected

The site covers 26 epochs, 1975 to 2100 in five-year steps. They fall into three groups, and the site marks which one you are looking at.

"Observed" does not mean counted. Even the historical grids are estimates: census totals spread across cells with the help of satellite imagery, and years between satellite images are interpolated. But they are anchored in observations. The projected years are not.

What "projected" means

Everything from 2025 onward is produced by CRISP, the JRC's Cities and Rural Integrated Spatial Projections model. It starts from the UN's medium projection of each country's population, splits it across large sub-national regions by extending their recent trends, and allocates new built-up surface and then new residents to 1 km cells by a set of suitability rules. The urban-centre boundaries for those years are drawn from the modelled grid. The GHSL source section describes the model in more detail.

So a projected figure is a scenario: what the model's rules produce if national populations follow the UN's medium variant and regions keep their recent trends. It is not a forecast of what a particular city will do, and it cannot know about a new port, a war or a housing policy. Two cities that look alike in 2020 can diverge in the model only because the regions around them grew at different rates.

2025 is model output too, but it is five years from the last observed epoch and the least uncertain of the model years, so the site treats it as "now". Uncertainty grows with every step after it. JRC's own confidence grade is lower for more cities in later years for a related reason: the independent grids it checks against all describe the present.

How the site shows it

• The epoch strip marks the boundary at 2025; years after it carry a "projected" tag, and so do figures for those years.
• Charts draw projected segments dashed, over a shaded projection band.
• Growth is reported separately for the observed period (1975 to 2025) and the projected one (2025 to 2100), so a projected trend is never blended into a historical one.
• Urban centres born after 2025 exist only in projections. Their pages show the years they exist in, and all of their figures are projected.

Grids

Each dataset counts population and built-up surface on a different grid.

H3 hexagons (Urban World v2)

Urban World v2 moves the GHS-WUP population and built-up rasters (the native 1 km Mollweide editions, the grid JRC computes its urban-centre statistics on) onto Uber's H3 hexagonal grid at resolution 8. Each hexagon is projected onto the 1 km grid and receives the area-weighted share of every pixel it overlaps (exactextract), so the whole hexagon is always counted. A hexagon belongs to an urban centre when its centre lies inside the boundary, so city totals on H3 differ slightly from JRC's pixel-exact figures: about 3% lower overall in 2025, 1% for centres over 2 million, up to about 10% for centres under 100,000, whose dense edges are partly cut by hexagons centred outside. The GHSL R2025A dataset (1 km grid) matches JRC's statistics exactly. This is done for all 26 epochs and for every populated cell within 30 km of an urban centre in any year, which covers everything the site maps.

A hexagon belongs to a city in a given epoch when its centre lies inside that epoch's boundary; a hexagon claimed by two boundaries goes to the one that covers more of it. Cell area varies with latitude, from about 0.55 to 0.74 km², and city statistics use each cell's exact area. H3 cells have near-uniform area and compact shape, and every cell has six equidistant neighbours. That regularity is what makes the radial density profiles possible.

1 km grid (GHSL R2025A)

GHSL R2025A uses the native 1 km cells in the Mollweide equal-area projection, so every cell covers exactly 1 km². A cell belongs to a city when its centre lies inside the city's boundary for that epoch. The boundaries are drawn on this same grid, so the 1 km figures are the closest to JRC's own statistics.

City statistics

Every city has figures for every epoch in its lifespan, 1975 to 2100 at most. Figures after 2025 are projections (see Observed, current and projected).

Population and density

For each city and epoch, population is the sum of the grid cells inside that epoch's boundary. Area is the total area of those cells, and density is population divided by area. This is density over the whole urban centre, not over built-up land alone, and it is the site's headline density.

Built-up surface

Built-up surface is the sum of GHS-WUP-BUILT-S in the same cells. Beside every density the site shows the built share (built-up surface divided by area) and people per built-up km² (population divided by built-up surface). See Built-up surface.

Confidence

Each population figure carries JRC's confidence grade for that city and year, from very high to low. See Confidence.

Rankings

Rankings are computed separately for each epoch, over the urban centres that exist in that epoch: the 2100 ranking includes centres born in 2090 and leaves out centres gone by then. Cities are ranked by population worldwide, within their continent and within their country, and by density, built share and people per built-up km² worldwide. No city is left out; low-confidence figures are ranked like any other and carry their grade.

Growth

Growth between neighbouring epochs is the compound annual growth rate over those five years. Longer growth figures cover three windows: the city's own lifespan, 1975 to 2025 (observed) and 2025 to 2100 (projected), each only when the city exists at both ends. Built-up growth uses the same windows, and the ratio of the two is described under Built-up surface.

Built-up surface

Population says how many people live in a city. Built-up surface says how much of its ground is covered by buildings. The site treats the two as peers: every density figure has the built share and people per built-up km² beside it, the map can colour cells by built share, and radial profiles show both.

What it measures

GHS-WUP-BUILT-S gives, for every 1 km cell, the area covered by buildings in km²: the footprint of roofs, whatever their height or use, homes and factories alike. Roads, car parks and gardens are not built-up. For 1975 to 2020 it is the GHSL built-up surface mapped from Landsat and Sentinel-2 imagery (interpolated between image dates), aggregated to 1 km. From 2025 it is projected by the CRISP model, which estimates how much new built-up surface each region will need from its population and its recent use of built land per person, and places it cell by cell. The projected built-up surface is model output, like projected population.

The release publishes built-up surface on a 1 km grid only (and an equivalent 30 arc-second grid), not at the finer resolutions of earlier GHSL releases, so the site cannot map individual buildings. Urban World v2 carries the 1 km grid into H3 hexagons the same way as population.

Three numbers

• Built-up km²: the built-up surface inside the city's boundary.
• Built share: built-up km² divided by the boundary's area. In 2025 the median urban centre is about 20% built-up; most fall between about 5% and 30%.
• People per built-up km²: population divided by built-up surface. It says how intensively built land is used, and separates cities that have the same boundary density: Dhaka is about 7% built-up with about 80,000 people per built-up km², Tokyo about 30% built-up with about 22,000.

Why boundary density stays the headline

People per km² of boundary remains the site's main density figure. It is what readers usually mean by how dense a city is, and it rests only on population and the boundary. People per built-up km² depends on a second modelled layer, and it becomes unstable where there is little built-up surface: a boundary with almost no mapped buildings produces a huge, meaningless number. Where the built share is near zero, read the built density as a data problem, not a fact about the city. Cities whose built share falls below 1% are listed on the review page.

Land consumption and population growth

Comparing built-up growth with population growth shows whether a city is densifying or spreading out. The site reports the ratio used for UN Sustainable Development Goal indicator 11.3.1, the land consumption rate to population growth rate (LCRPGR):

LCRPGR = ln(built-up at end ÷ built-up at start) ÷ ln(population at end ÷ population at start)

A value above 1 means built-up land grew faster than population; below 1, population grew faster. It is computed for 1975 to 2025, 2000 to 2025 and 2025 to 2100.

The ratio divides by population growth, so it explodes when a city's population barely changes: a city growing 0.1% a year that adds a little built land gets a ratio in the dozens. The site therefore reports it only when population grows or shrinks by more than 0.5% a year on average over the window, and shows it as undefined otherwise.

Confidence

Every urban-centre population in the release comes with a grade from JRC, published as "plausibility", for that city and that year. The site calls it confidence and shows it next to population figures.

How JRC grades a figure

JRC estimates each city's population a second way, from seven grids made independently of GHSL: two population grids (LandScan and WorldPop, both for 2020) and five satellite land-cover maps of urban or settled surface from 2015 to 2021, each scaled to the city's residential share. It averages these estimates, using a regression model fitted separately for each income group and year, and compares the average with the population in the release:

What it tells you

A low grade means the independent sources disagree strongly with the release about how many people live there. JRC built the test to catch artefacts in the input data, such as erroneous census values or census areas placed in the wrong location, which move population to where it does not belong. A low grade is a reason to look closer before comparing that city with others; it does not prove the figure wrong.

The grade is about the population figure only. It does not grade the boundary, the built-up surface, or the projection method, and a high grade in 2080 does not make the 2080 projection an observation.

How it changes over time

In 2025, 16% of urban centres are graded moderate or low. By 2100 the share is 35%. The independent grids all describe the years around 2020. JRC fits the comparison separately for each year, but the further a year is from 2020, in either direction, the more a city's figure can drift from what those grids support, simply because the city has grown or shrunk since or before. In 1975, 38% of urban centres are graded moderate or low. Grades far from 2020 say as much about that distance as about the figure.

Where the site shows it

A small mark beside a population figure gives its grade and links to this page. Rankings can be sorted and filtered by confidence, but nothing is hidden by default. Cities graded low in any year are listed on the review page.

Map reference layers

Administrative boundaries

Administrative lines and local place names come from OpenStreetMap and Natural Earth through the self-hosted Protomaps map edition dated 1 January 2026. The edition date identifies this map snapshot; individual boundary effective dates are not supplied. These lines do not change when you select an earlier population year and must not be interpreted as historical administrative geography. Coverage and administrative levels vary by location.

The GHSL urban-centre footprint still defines city statistics. Administrative boundaries are reference information only. Sources: OpenStreetMap contributors and Natural Earth, packaged as Protomaps basemap layers. We serve this map archive from our own storage. The boundary layer includes country, region, county and locality boundaries where available; visibility depends on zoom.

Municipal outlines in the guide

The guide's Paris municipal outline comes from the French administrative API. Its Shanghai municipal outline comes from OpenStreetMap relation 913067, under the Open Database Licence. These outlines are separate reference datasets from the explorer's Protomaps archive and from GHSL urban-centre footprints.

Comparing years

Choose Compare years beside the epoch, then move either handle or choose its printed year. The URL preserves the interval. Done returns to the ending year.

In the Urban World v2 dataset, the change map subtracts the starting population from the ending population in the same H3 cell. The extract includes the city's surroundings. Missing observations are not treated as zero. The color scale is symmetric around zero and capped at the 95th percentile of absolute cell changes for the selected city and interval; stronger changes share the end color. Click a cell to inspect its full population history.

The dashed outline is the earlier urban footprint and the solid outline is the later footprint. City totals use each epoch's own footprint, so their change can differ from the sum of changes across the mapped cells. For the native GHSL grid dataset, comparison currently shows footprints and city totals; cell-level changes require a corresponding native-grid export.

Population figures from 2025 onward are model output, and from 2030 they are labelled as projections. A comparison that ends after 2025 compares an estimate with a projection, or two projections; see Observed, current and projected.

Data review

The site shows every urban centre in the release, for every year it exists. No city is removed from search, rankings, city pages, comparisons or the map, whatever its figures look like.

Some figures do look suspect: a population packed into a handful of cells, a boundary with almost no buildings, a city that doubles in five years. Instead of filtering those cities out, the site flags them. The review page lists every city that trips at least one of the rules below, largest first, with the years each rule applies to and a link to the city's page. It is where any future decision to curate the data will be made from; until then, the data is shown as JRC published it.

Rules

The first three rules look for density figures that rest on too few cells, where a small error in how population is placed moves the density a long way. A small, dense city is not necessarily wrong, and 20,000 people/km² is not a physical limit. The built-share rule catches boundaries where people per built-up km² becomes meaningless (see Built-up surface). The confidence rule is JRC's own check (see Confidence). A population jump can mark a boundary that merged with or split from a neighbour, or an artefact in the input data. The review page is computed from the H3 cells of Urban World v2; on the 1 km grid, cell counts and areas differ slightly, so a few cities near a threshold would be flagged differently.

A flag is a prompt to look closer, not a verdict. Values are compared unrounded; displayed values are rounded.

Radial density profiles

Radial density profiles follow the method developed by Alain Bertaud to characterize the spatial structure of cities. They measure how population density varies with distance from the city centre.

Method

1. Centre identification: the profile uses JRC's population-weighted centroid of the city for the selected year, published with the urban-centre boundaries (GHS-WUP-MTUC), snapped to its containing H3 cell. It is the point that balances the city's residents on the 1 km grid, and the same centroid shown on the population-centre drift path. Changing the year updates the centre, rings, and model fit. A year outside the city's lifespan has no centre or profile.
2. Ring construction: rings are circular bands around that cell, each one cell-spacing wide (about 0.9 km, the distance between neighbouring hexagon centres). Ring k holds every cell whose centre lies within k ± ½ spacings. Ring 0 is therefore the centre hexagon alone, ring 1 its six neighbours, and from there the rings grow as true circles. The chart plots each ring at its real mean distance from the centre.
3. Density aggregation: for each ring, population is summed over the cells that are part of the city at that epoch and divided by their exact area (persons/km²). Built-up surface is summed the same way, giving the ring's built share and its people per built-up km².
4. Profile output: the result is a distance-density curve that reveals whether a city is monocentric (steep gradient), polycentric (multiple peaks), or dispersed (flat profile), with a matching curve of how much of each ring is built up.

Choosing your own centre

Where a city "starts" is a judgement call, and for polycentric regions there is no single right answer. With the profile shown on the map, click any hexagon to rebuild the rings around it. The choice is kept in the page URL, so a re-centred profile can be shared, and stays fixed when you change years. Reset returns to the population centre for the currently selected year.

The population centre describes where residents are concentrated; it need not coincide with downtown or give the best Standard Urban Model fit. It can also fall between populated areas. Because the default centre moves over time, changes between profiles reflect both population redistribution and movement of the origin. Pick a fixed cell to compare years from the same location.

Dhaka shows how far off the default can be. Its 2025 boundary covers 6,812 km² and takes in the neighbouring towns around the capital, so the population-weighted centroid lands in a low-density gap between them and the profile peaks about 25 km out. For a boundary that joins several centres like this, re-centre on the core you want to study.

Reading the chart

• The bold line is a 3-ring average: population and area are pooled over each ring and its two neighbours. Pooling weights every ring by its area, so the single centre hexagon cannot swing the curve the way a 60-cell ring can.
• The faint dots are the individual rings. Their scatter around the line is mostly noise, since inner rings hold only a handful of cells.
• A dashed stretch marks rings that are less than 35% complete, typically the ragged urban edge or a coastline, where a few cells stand in for a whole ring.
• The density axis starts at zero, so a flat city reads as flat instead of being stretched into apparent volatility. Switch to log to see whether density falls off exponentially, which shows up as a straight line.

The built-up profile

Beside population density, the chart shows the built share of each ring: the fraction of its cells' area covered by buildings, from GHS-WUP-BUILT-S. It usually falls with distance too, and where it falls more slowly than population density the city's outer rings are built up but thinly occupied. You can also plot people per built-up km², which divides each ring's population by its built-up surface.

Density per built-up km² is much flatter than density per km² of land, because built share itself falls with distance and dividing it out removes most of the gradient. It is a measure of how intensively built land is used, not a second version of the density profile, and it becomes noisy in rings with very little built-up surface.

The Standard Urban Model

The dashed curve is the Standard Urban Model, the monocentric city of Alonso, Muth and Mills, which Bertaud uses as the baseline for reading real cities. It says density falls exponentially with distance from the centre:

D(r) = D₀ · e^(−β·r)

• D₀ is the modelled density at the centre
• β is the density gradient. We report it as the distance over which density halves (ln 2 / β): about 5 km for Paris, about 2 km for a typical small city
• R² is how much of the profile the curve explains

The fit is least squares on the density scale, with each ring weighted by its area. That is equivalent to fitting the individual cells, so the single centre hexagon and the thin, ragged outer rings count only in proportion to their area. It is recomputed in your browser whenever you change the year or move the centre. R² describes how well the model fits the selected profile; the default centre is determined by population, not by maximising R².

The curve is only drawn when the model actually describes the profile: at least 5 populated rings, R² of 0.2 or more, and density falling (β > 0). For profiles that do not meet these conditions, the sidebar reports the fit where enough rings are available and draws no curve. The model is always fitted to density per km² of land, never to the built-up profile. In projected years the fit describes the model's allocation of future residents, which tends to be smoother than observed cities, so projected profiles fit the model more often.

The drifting population centre

The map also traces JRC's population-weighted centre of the city at every epoch it exists, from 1975 to 2100 at most. The path shows which way growth has pulled the city's mass: toward new suburbs, along a coast, or back into the core.

Why H3?

Radial profiles use the H3 hexagonal grid (resolution 8, ~0.74 km² per cell) rather than the regular 1 km grid. H3 cells have near-uniform area and compact shape, and every cell has six equidistant neighbours, so a band exactly one spacing wide starts cleanly from a single centre cell. Fixed 1 km bands cut across the ~0.9 km spacing unevenly, so the innermost band takes in the centre cell together with its six neighbours. H3's own grid rings (all cells k steps away) avoid that, but they are hexagons. Their corners lie 15% farther from the centre than their sides, which smears distance.

Interpretation

• Steep exponential decay: classic monocentric city (e.g., Paris, Buenos Aires)
• Plateau then drop: large dense core (e.g., Mumbai, Dhaka)
• Multiple peaks: polycentric structure (e.g., Ruhr area, Randstad)
• Flat profile: dispersed, low-density sprawl (e.g., Atlanta, Houston)