The Geography of Human Pollution Exposure

The Global Human Settlement Layer estimates where every person on Earth lives in 2025 on a grid of about one kilometre. Aggregated to cells of about 7 km, it shows how extreme the concentration is. 75% of inhabited land is home to fewer than 50 people per square kilometre and holds only 5.5% of humanity. At the other end, 48% of people live at densities above 1,000 per square kilometre — towns and cities, but also the densely farmed plains of South Asia, eastern China and the Nile valley — on just 2.0% of inhabited land.

This matters for pollution in two ways. Density concentrates emissions: traffic, cooking fires, generators, workshops and power demand all scale with people per square kilometre, and in many of the densest places they are not controlled. And density concentrates receptors: whatever pollution exists, a dense area exposes more lungs to it. Average concentrations over land, or over a country's territory, therefore say little about what people breathe. The relevant average is the one weighted by population.

It also means that the world's air-quality problem is geographically compact. A few dense river plains, coastlines and urban regions — the Indo-Gangetic Plain, the North China Plain, the Pearl and Yangtze deltas, Java, the Nile, Lagos and the West African coast — hold a large share of exposed people. Improving air in those airsheds would change global exposure more than improving it almost anywhere else.

Across the 1,693 sample cells, fine-particle levels rise with the density at which the typical resident lives: from about 13 µg/m³ in cells where residents live at fewer than 150 per square kilometre to about 34 in cells above 1,500 per square kilometre, which hold 5.8 billion people, and 39 in the densest cells, home to 2.0 billion. The correlation between density and pollution across cells is 0.35 on a logarithmic scale — strong for a relationship that also reflects geography, climate and income.

The consequence is a co-location premium. Averaged over people, global exposure is 29.7 µg/m³; averaged over inhabited land, it is 19.5. The average person therefore breathes air 1.53 times dirtier than the average inhabited square kilometre. The premium is largest by far in East Asia (1.57), where dense urban and industrial regions sit inside much cleaner hinterlands. Elsewhere it is a more modest 1.16–1.22, smallest in sub-Saharan Africa (1.16) and Europe and Central Asia (1.16), where background pollution from dust, biomass burning or widespread secondary aerosol is spread more evenly over the land.

The season matters too. In January, when winter inversions trap smoke over northern India, Pakistan and China, the population-weighted level in this sample is 33.0 µg/m³; in July, when monsoon rains wash the air of South and East Asia, it is 19.4. Annual figures hide months when exposure is far higher. The map in section 6 shows January, July and calibrated annual values for every cell.

Air-quality scientists separate a city's pollution into a regional background — what the air already carries when it arrives — and an urban increment added by the city's own sources, a method associated with Lenschow and colleagues' work in Berlin. This dashboard applies it worldwide: for each of 149 large cities it compares PM2.5 at the city centre with the average of four points about 165 kilometres to the north, east, south and west, over four fortnights spread across 2024.

The background dominates. In the median city it is 79% as high as the centre, and it supplies more than half of centre-city PM2.5 in 87% of cities; weighted by city population, 61%. The median urban increment is only 3.2 µg/m³. Delhi illustrates the point: at about 71 µg/m³, its centre is among the most polluted sampled, but the countryside around it averages about 55 — 78% as much — because the whole Indo-Gangetic Plain is smoky. The largest local increments are in large, dense and often industrial cities: Beijing (58), Jakarta (46), Chengdu (39), Xi'an (32), Kuala Lumpur (29), Lahore (27) µg/m³.

Model resolution matters here. At about 40 km, CAMS blurs the city centre into its suburbs and can miss the sharpest local peaks, so true increments in city cores are larger than this sample shows. But the conclusion that regional background is the larger share is consistent with source-apportionment studies in many cities, and with global modelling: Zhang and colleagues estimated that about 12% of the 3.45 million PM2.5 deaths in 2007 were caused by pollution emitted in another world region entirely, and about 22% were linked to goods and services produced in one region and consumed in another.

If transport matters, a city's pollution should rise and fall with the background on its upwind side more than on its downwind side. Using hourly winds at 10 metres from NASA's POWER service, the analysis identifies, for every hour, which of the four ring points lies upwind and which downwind. In 96 of 149 cities, the city centre's hourly PM2.5 correlates more closely with the upwind point (median correlation 0.57) than with the downwind one (0.50) — the signature of polluted air arriving on the wind.

The other side of the ledger is export. In 94 of 149 cities the downwind ring point is, on average, more polluted than the upwind one, by a median of 0.7 µg/m³: cities add to the pollution of the countryside and towns downwind of them, often in another district, province or country. Every city is both a receptor and a source, and the people living downwind of large cities carry part of their burden.

Wind and size shape the local increment. Across cities, a tenfold larger population goes with a 50% larger ratio of centre to background (t = 6.6), and each additional metre per second of average wind speed with a 8% smaller one (t = 3.2), holding background constant. Stagnant, crowded cities trap their own emissions; windy ones dilute them — into their neighbours' air.

Weighted by population and calibrated to satellite-based annual means, PM2.5 exposure is 41.7 µg/m³ in lower-middle-income countries, home to 2.90 billion people in this sample; 30.4 in low-income countries; 26.0 in upper-middle-income countries and 12.9 in high-income countries. Exposure rises with industrialisation and urbanisation before falling as countries adopt cleaner energy and pollution controls — the inverted-U documented in this series' report on the unequal geography of air pollution. 32% of people in the sample live above 35 µg/m³, seven times the WHO guideline, and 98.5% above the guideline itself.

The source mix differs as much as the level. In the 22 low-income cities sampled, the regional background is a median 98% of centre-city PM2.5 and the local increment almost nil: their air is dominated by Saharan and Sahelian dust, savanna and crop fires, and household biomass burning spread across the countryside. In upper-middle-income cities the background share falls to 59% and the median increment rises to 8.4 µg/m³, reflecting traffic, industry and power plants concentrated in and around cities. High-income cities have low levels of both.

These differences turn an exposure map into a map of responsibility and leverage. People in poor countries breathe pollution that their own cities did little to create and that city policy alone cannot remove; people downwind of large cities and industrial regions breathe their neighbours' emissions. The World Bank's Rentschler and Leonova estimate that 7.3 billion people live with unsafe PM2.5, 80% of them in low- and middle-income countries, and that 716 million of the world's poorest people live with unsafe air.

The atlas shows the 1,693 sample points, each at the population-weighted centre of its 3-degree cell and sized by the number of people in it. The default layer shows calibrated annual PM2.5 on the WHO's scale; the dust layer shows CAMS desert dust of all particle sizes, which dominates across the Sahara, Sahel, Arabian Peninsula and Central Asian deserts and reaches 1.2 billion people at levels above 25 µg/m³; the density layer shows the density at which the typical resident of each cell lives.

Switch on the city rings to see the 149 cities analysed in sections 3 and 4. Hover over a cell or a city to see its values: population, resident density, January and July concentrations, the calibrated annual estimate, and dust; for cities, the centre, the background, the local increment and the wind.

Two geographies stand out. The Indo-Gangetic Plain and eastern China combine the highest densities with high levels from human sources. The dust belt from West Africa to Central Asia combines high levels with sparser populations, except where it meets the dense Nile valley and the Indus plain. Elsewhere — the Americas, Europe, southern Africa and Oceania — most people breathe air within a few times the guideline.

Where the regional background dominates — most of South Asia, much of Africa and the dust belt — city clean-air plans can only go so far. Delhi cannot clean its air while crop residue burns across Punjab and Haryana and millions of households cook with solid fuels across the plain. The effective unit is the airshed: the region within which emissions and receptors interact. That means coordinated plans across states and countries, clean household energy for rural as well as urban homes, alternatives to crop burning, controls on brick kilns and small industry, and, where dust dominates, land management and early warnings.

Where local increments are large — big, dense and industrial cities in East and Southeast Asia, Latin America and the Middle East — urban measures pay off directly: cleaner vehicles and fuels, public transport, controls on power plants and industry, and enforcement. China's experience shows how fast concentrated action can work: after its 2013 national action plan targeted coal, industry and regional coordination together, satellite-based population-weighted PM2.5 fell from 52 µg/m³ in 2014 to 31 in 2024, about 40%. Because windy cities export their pollution, these measures also benefit people downwind.

And because transport and trade carry pollution across borders, some of the responsibility is international. Regional agreements, like Europe's Convention on Long-range Transboundary Air Pollution, show that neighbouring countries can share commitments on the pollutants that drift between them; supply chains link consumers in rich countries to emissions where goods are made. The density result adds a final point: because exposure is concentrated, the people-weighted benefits of cleaning a few dense airsheds are very large.

Population comes from the Global Human Settlement Layer GHS-POP R2023A, 2025 epoch, 30 arc-second grid. It was read from the file's published overview at one-eighth resolution (about 7 km), whose averaged values reproduce the global total of 8.19 billion and a full-resolution tile to within 1%. Resident density is the population-weighted mean density of the ~7 km pixels in each 3-degree cell. Sample points are the population-weighted centres of the 1,693 3-degree cells with more than 10,000 people, covering 8.19 billion people.

PM2.5 and dust are hourly CAMS global atmospheric composition forecasts retrieved through the Open-Meteo air-quality API: for the grid, 8–21 January and 8–21 July 2024; for 149 cities (the largest city of every country with at least 5 million people, plus others in the most populous countries) and four ring points 1.5 degrees of latitude and the equivalent longitude away, four fortnights in January, April, July and October. Each country's grid sample is scaled to its population-weighted SatPM 2024 annual mean (Our World in Data); the median scale factor is 1.60. Hourly 10-metre winds for city centres come from NASA POWER (MERRA-2).

Regional share is the mean of the four ring points divided by the centre; upwind and downwind points are chosen hour by hour from the wind direction; the increment regression is OLS with robust errors. Caveats: two or four fortnights rather than full years (the four-fortnight city mean is 0.95 times the January–July mean); CAMS resolution blurs city cores; ring points over sea or other cities distort a few backgrounds; CAMS dust includes coarse particles. Related reports: the unequal geography of air pollution exposure, heat and air pollution and the pollution–climate feedback system.