Satellites and AI Could Transform How Governments Track Air Pollution and Protect Public Health

Satellite observations and machine learning could help governments close major air-quality data gaps, tracking pollution across more than 1.5 million grid cells and around 8,000 urban areas. Better daily pollution data could guide health spending, environmental regulation and infrastructure investment while creating new opportunities for development partners and clean-technology businesses.

Air pollution is emerging as a major economic and development challenge, especially in countries where governments lack reliable monitoring networks. A World Bank study by Brian Blankespoor, Susmita Dasgupta and David Wheeler shows how satellite observations and machine learning could help governments track four major pollutants, fine particulate matter (PM2.5), ozone (O3), nitrogen dioxide (NO2) and carbon monoxide (CO), on a daily basis.

The stakes are high. The study cites estimates linking air-pollution exposure to around 8.1 million premature deaths annually and approximately 120 million disability-adjusted life years. In contrast, economic losses associated with PM2.5 exposure have been estimated at $8.1 trillion. Yet many developing countries have limited ground-based monitoring, making it difficult to identify pollution hotspots, measure population exposure or determine whether environmental policies are working.

Satellites Could Fill the Missing Gaps in Global Air-Quality Monitoring

The World Bank researchers developed pollution estimates for more than 1.5 million terrestrial grid cells at 0.1-degree resolution, equivalent to roughly 11.1 kilometres at the equator. The pilot covers April 10 to November 17, 2025, but the underlying satellite information means estimates could potentially be extended back to 2018.

The approach combines observations from the European Space Agency's TROPOMI satellite instrument with Copernicus ERA5-Land weather information, EDGAR emissions estimates, climate classifications, elevation and other geographic characteristics.

IQAir provides ground observations through its global platform, which includes more than 65,000 reporting sites. The researchers used daily information from 19,932 monitoring stations for part of the modelling exercise.

This matters particularly for developing regions where monitoring networks remain sparse. Satellite-supported estimates could provide information for secondary cities, rural areas and cross-border corridors that conventional monitoring stations cannot adequately cover.

Machine Learning Turns Environmental Data Into Actionable Evidence

Researchers developed separate random-forest machine-learning models for each pollutant using 500 decision trees. Monitoring stations were divided into 60% training, 20% calibration and 20% testing samples.

PM2.5 produced the strongest performance, recording an adjusted R-squared of 0.769 in the adjusted test sample. Comparable figures were about 0.464 for NO2, 0.439 for ozone and 0.341 for CO, showing that accuracy varies considerably by pollutant.

Satellite measurements were particularly important for PM2.5, NO2 and CO. Weather conditions were more influential for ozone, with wind speed and temperature each contributing around 10% of predictive importance.

Importantly, the models also captured daily changes. Across Functional Urban Areas with sufficient monitoring information, median correlations between predicted and monitored concentrations exceeded 0.69 for all four pollutants, while upper-quartile correlations were above 0.82.

Better Pollution Maps Could Reshape Health and Infrastructure Spending

The biggest development opportunity comes from combining pollution estimates with population, poverty and health information. The methodology can generate population-weighted exposure estimates for around 8,000 Functional Urban Areas and could be extended to municipalities, districts, provinces and countries.

For governments, this could improve decisions about where limited public money should be spent. Health ministries could identify communities facing high exposure, while finance ministries could compare pollution-control investments against potential reductions in healthcare and economic costs. Transport authorities could assess whether metro expansion, vehicle-emission standards or traffic policies actually improve air quality.

Daily information could also help evaluate fuel reforms, industrial interventions and emissions regulations. When combined with epidemiological evidence, pollution data could support estimates of premature deaths, hospital admissions and disability-adjusted life years at subnational levels.

There is also an equity dimension. Combining pollution estimates with poverty and census information could reveal whether disadvantaged communities face disproportionately high exposure and help governments target interventions more effectively.

New Opportunities for Development Partners and Private Investment

For international development institutions, satellite-based monitoring could strengthen project design, environmental assessments and impact evaluation. Development partners could establish pollution baselines before financing transport, energy and urban infrastructure projects and subsequently measure whether those investments produce environmental improvements.

The approach could be particularly useful in fragile and conflict-affected settings, where installing and maintaining conventional monitoring infrastructure may be difficult.

Private companies could also find opportunities in environmental analytics, monitoring technologies, pollution forecasting, clean transport, renewable energy and environmental-risk management. More detailed pollution information could help companies assess industrial locations, employee exposure and supply-chain risks.

However, greater transparency could also create challenges for pollution-intensive industries by increasing regulatory scrutiny and pressure for cleaner production. The study does not establish a mechanism for attributing observed pollution directly to individual companies, so satellite estimates should not automatically be interpreted as proof of firm-level responsibility.

The next step is to extend the dataset back to 2018, creating a multi-year record for evaluating long-term trends and policy interventions. Combining the system with weather forecasts and short-term satellite trends could eventually support pollution forecasting and early-warning systems.

For policymakers, development partners and businesses, the central message is clear: satellite technology and machine learning cannot replace reliable ground monitoring, but they can dramatically expand its reach. Used carefully, these tools could transform air pollution from a poorly measured environmental problem into actionable information for public health, investment and development planning.