The Unequal Geography of Human Air Pollution Exposure

In 2024 the most exposed countries formed a continuous belt. Bangladesh averaged 65 µg/m³, Pakistan 54, Nigeria 50, Cameroon 52, Niger 48 and India 44. The Gulf states were as high or higher — Qatar 76, Kuwait 57, the United Arab Emirates 56, Saudi Arabia 50 — but much of their exposure is desert dust rather than combustion. At the other end, only 17 small places, mostly Pacific islands with Iceland, Greenland, the Faroe Islands and Bermuda, met the WHO guideline of 5 µg/m³; together they hold well under 0.1% of humanity.

Between those extremes, the world divides roughly into thirds. About 17% of people live in countries averaging under 15 µg/m³, the WHO's third interim target. About 48% live in countries between 15 and 35 µg/m³, a range that includes China at 31 and Indonesia at 19. And 35% live in countries above 35 µg/m³, the least stringent interim target. That last share has fallen from 51% in 2010, mostly because China's average dropped below the line; the State of Global Air report, using a different exposure model, puts 36% of people above it in 2023.

The atlas below shows exposure in any year from 2000 to 2024, the change since 2010, the share of households without clean cooking, and the gap between each country's exposure and the level predicted from its income, industry, urban form, energy access and rainfall. That last layer separates countries whose air is cleaner than their circumstances would suggest from those whose air is dirtier — often a sign of dust, fires, geography or policy.

Grouped by income, exposure traces an inverted U. Low-income countries averaged 30.0 µg/m³ in 2024; lower-middle-income countries, the most exposed group, 42.5; upper-middle-income countries 26.0; and high-income countries 12.5. By income per head, exposure peaks among countries with $5,000–10,000 a year at purchasing-power parity, a band that includes India, Pakistan, Bangladesh and Nigeria and holds 2.5 billion people averaging 44.5 µg/m³. The shape echoes the environmental Kuznets curve, but the downturn is not automatic: it reflects policies, technologies and changes in economic structure that richer countries adopted, not income itself.

The concentration curve ranks the world's population from the poorest country to the richest and plots the cumulative share of exposure they bear. If exposure were equal, the curve would follow the diagonal. It bows above it: the poorest half of humanity bears 63% of all population-weighted exposure and 66% of exposure above the WHO guideline. The concentration index for exposure above the guideline is −0.19, where zero would mean no income gradient — a measure of environmental inequality between countries, before any inequality within them is counted.

The gap has widened. High-income countries cut their average from 15.3 µg/m³ in 2010 to 12.5 in 2024, and upper-middle-income countries from 32.0 to 26.0, led by China. Lower-middle-income countries peaked at 50.9 in 2016 and have since fallen to 42.5, still far above where they were in 2000, and low-income countries have barely moved from 30. In other words, the groups that were already cleanest improved the most, while the most exposed are only now starting to turn the corner.

A single cross-country model, estimated for 172 countries with 98% of the world's population, explains 62% of the variation in log exposure. To ask how much each driver contributes, the explained variation is split with a Shapley decomposition, which averages each group's added explanatory power over every order in which the groups could be entered. Income accounts for 0.18 of the R², industrialisation 0.13, rainfall 0.13, urban form 0.11 and energy access 0.07. The shares overlap heavily — income moves with the others — so they describe how much each tells us, not how much each causes.

Industrialisation carries the clearest independent signal. Holding the other drivers constant, a country at the 75th percentile of industry's share of GDP (32%) has exposure about 33% higher than one at the 25th (18%). Coal's share of electricity adds little once industry is controlled for, because the two move together. Urban form matters through crowding: moving from the 25th to the 75th percentile of population density raises predicted exposure by about 30%, and a larger share living in cities of over a million by about 13%. This fits evidence that dense, fast-growing cities concentrate traffic, construction and industrial emissions where people live.

Energy access matters beyond the home. Households burning wood, dung, charcoal and coal pollute the air outside as well as inside: modelling for India suggests that removing household sources would cut population-weighted ambient PM2.5 by roughly 30%, and a global source study attributes about a fifth of deaths from ambient PM2.5 to residential energy use. In the model, moving from the 25th to the 75th percentile of clean-cooking access lowers predicted exposure by about 18%. Natural dust runs the other way to policy: drier countries are dustier, and the 75th percentile of rainfall is associated with 22% lower exposure than the 25th. Income itself follows an inverted U once these are held constant.

Each driver tells part of the story and none tells all of it. Against industry's share of GDP, the industrial economies of South and East Asia sit high, while service economies sit low whatever their region. Against clean-cooking access, sub-Saharan Africa clusters at the low-access end, where 77% of people still cook with polluting fuels; South Asia sits in the middle, its exposure driven as much by industry, density and crop burning as by kitchens.

Population density separates the crowded river plains of Bangladesh and northern India, where pollution from many sources pools over hundreds of millions of people, from sparsely settled countries that can absorb more emissions per person. Rainfall separates the dust belt — the Gulf, North Africa, the Sahel — from humid countries where rain washes particles from the air. Several of the most exposed places are dusty rather than industrial, a reminder that national averages blend human and natural sources.

Selecting a country highlights it in every view and opens its profile in the atlas above. The colour can show region or income group, and bubble size shows population, so the countries that dominate global exposure — India, China, Pakistan, Bangladesh, Nigeria — stand out. Outliers relative to the model are often explained by things it cannot see: Bolivia's 2024 fire season, Saudi Arabia's dust or Nepal's position downwind of the Indo-Gangetic plain.

Population-weighted world exposure rose from 27.3 µg/m³ in 2000 to a peak of 36.0 in 2015 and fell to 30.0 by 2024. The fall was driven above all by China, whose average dropped 40% from its 2014 peak of 52.2 to 31.3 after the 2013 Air Pollution Prevention and Control Action Plan, which cut coal use in households and industry, closed or cleaned polluting plants and tightened vehicle standards. A study that removes the effect of weather still finds a 34% fall in PM2.5 across Chinese cities from 2013 to 2019, about 10 percentage points less than observed. Europe and North America continued long declines: since 2010 exposure has fallen by 46% in Poland, 41% in Germany, 39% in France and 19% in the United States.

Elsewhere the picture is mixed. South Asia's average fell from a 2016 peak of 59.6 to 46.0, with India down from 59.8 to 44.2, but Bangladesh remained near 65 and Pakistan near 54. Sub-Saharan Africa and the Middle East and North Africa have barely changed since 2010. Latin America is the one region where exposure rose, by 10%, with fire seasons pushing Bolivia to 41 and Brazil to 15 in 2024, both the highest in the series. Across countries, the median high-income country saw a 13% fall since 2010 and 75% of high-income countries improved; among lower-middle-income countries the median rose slightly and only 45% improved.

Year-to-year changes in satellite-derived exposure partly reflect weather — dry winters, dust storms and fire seasons — so single years should be read with care, and the trend lines are more informative than any one value. What the record shows clearly is that the descending side of the curve is reachable through policy: China moved down it within a decade, at an income level well below that at which Europe and North America made the same transition.

The experience of the countries that cleaned their air points to three groups of measures, matching the drivers. Industrial and power-sector controls — emission standards enforced with continuous monitoring, retrofits of scrubbers and filters, and the replacement of coal with cleaner generation — drove most of the declines in Europe, North America and China. Urban measures — public transport, cleaner vehicle fuels and standards, dust control on construction sites, and planning that keeps heavy industry and through-traffic away from dense neighbourhoods — address the crowding that concentrates exposure.

Clean household energy is the third, and the one most specific to the poorest countries. Liquefied petroleum gas, biogas and electric cooking cut household air pollution directly and reduce the outdoor pollution that household fires add. India's programme of free LPG connections and China's rural clean-heating programme show both the potential and the limits: connections rise quickly, but sustained use depends on fuel prices and supply. Because some exposure is regional — smoke from crop burning, haze crossing borders, dust — action also needs to be coordinated across administrative and national boundaries.

The calculator applies the model to a chosen country: change its industry share, clean-cooking access, urban share, million-city share or income and see how its predicted exposure moves, with its unexplained residual held fixed. It is a way to read the cross-country associations, not a forecast: the coefficients describe how exposure differs between countries, and a country's own path will depend on how it grows, what it regulates and what it burns.

Exposure is the national population-weighted annual mean PM2.5 from the SatPM product of the Atmospheric Composition Analysis Group at Washington University in St. Louis (2026 release), via Our World in Data, for 2000–2024. It combines satellite aerosol retrievals, a chemical transport model and ground monitors, and includes natural sources. World, income and regional averages are weighted by World Bank population for each year; income groups and regions use the current World Bank classification throughout. Bands follow the WHO 2021 guideline (5 µg/m³) and interim targets (10, 15, 25, 35); the 35–50 and above-50 split is added for detail.

Drivers are the latest 2015–2024 values from the World Development Indicators — GDP per capita (PPP, constant 2021 $), industry value added, urban population, population in agglomerations over one million (zero where none exists), population density, access to clean cooking fuels and technologies (total, urban and rural), electricity access and average precipitation — and coal's share of electricity from the Energy Institute and Ember via Our World in Data. The model regresses log PM2.5 in 2024 on log income and its square, industry share, coal share, urban share, million-city share, log density, clean-cooking access and log precipitation (OLS, n = 172, R² = 0.62); all but coal are significant at the 5% level with robust standard errors, urban share at about 5%.

The Shapley decomposition splits R² among five groups by averaging marginal contributions over all 120 orderings. The concentration curve and index rank 198 countries by GDP per capita. Results are cross-sectional associations: they cannot separate cause from correlation, and national averages hide large differences between cities and countryside, rich and poor neighbourhoods, and people near and far from major roads and plants. Rainfall is an imperfect proxy for dust. Related reports in this series: the pollution–climate feedback system and human exposure across the polycrisis century.