Reviewed by Lexie CornerMar 19 2025
Researchers at the University of Utah found that the EPA's air quality monitoring network consistently fails to detect lead and sulfur dioxide, followed by ozone and carbon monoxide, in communities of color. The study highlights a disproportionate distribution of monitoring stations, with most located in predominantly white neighborhoods.
The Environmental Protection Agency regulates the monitor map by pollutant. The colors represent the area covered by each monitor. Image Credit: University of Utah
EPA regulatory monitors are the primary source of information for decisions on urban planning, pollution control, and public health programs. Unevenly distributed monitors may result in inaccurate pollution data, putting vulnerable populations at risk.
It’s the question behind the question. Researchers, policymakers, we all use air quality data, but whose air is it measuring? Even though this data is of really high quality, that doesn’t mean that it’s high quality for everyone.
Brenna Kelly, Doctoral Student and Study Lead Author, University of Utah
While previous studies assumed that the data represented all neighborhoods equally, they have shown that underprivileged populations face the highest rates of air pollution exposure. This study is the first to examine neighborhood-level differences in monitors across every US Census category. All non-white groups experienced inequalities, with Native Hawaiians and other Pacific Islanders facing the largest disparities, followed by American Indian and Alaska Native populations.
Artificial intelligence (AI) tools are often required in air quality research to manage large datasets. The study raises an additional ethical concern for big-data users: the potential bias within the datasets themselves, in addition to the known bias in AI systems.
If there was a disparity for just one type of monitor, it could conceivably be accidental or just poor design. The fact that it’s a consistent pattern across all pollutants suggests that the decision-making process needs to be looked at carefully—these monitors are not being distributed equitably.
Simon Brewer, Study Co-Author and Associate Professor, Geography, University of Utah
Simon Brewer is an executive committee member of the University’s One-U Responsible AI Initiative.
Air quality varies significantly from one street to another and is highly localized. The authors used the census-block level, one of the smallest units used by the US Census Bureau for residential patterns, to map and monitor locations and neighborhood demographics.
The researchers used the EPA Air Quality System Regulatory Monitoring Repository to determine monitors for the six major air pollutants harmful to human health: lead, ozone, nitrogen dioxide, sulfur dioxide, carbon monoxide, and particulate matter.
They estimated the racial and ethnic makeup of each census block using data from the 2022 American Community Survey Census. After controlling for population size, the researchers found systemic monitoring inequalities for each pollutant. All non-white demographics were associated with lower levels of monitoring for lead, ozone, nitrogen dioxide, and particulate matter compared to the white non-Hispanic population.
Kelly's dissertation research in population health sciences, which focused on the risks of air pollution exposure for pregnant women, sparked her interest in the EPA's air quality monitoring network. In epidemiology, research seeks to determine the causes of diseases in a community. Kelly noted that until now, it was assumed the data accurately reflected air quality issues across the world.
“It’s not just that we’re missing one pollutant type for one group; it’s that we understand less about everything for all these groups. That’s concerning. If I want to relate air pollution exposure to a disease, I need to measure it well. If I have a better understanding of air quality for one group of people, that’s going to produce biased results,” said Kelly.
The ethical use of AI and big data is a concern in many fields, including population health and air pollution. The University recently launched the One-U Responsible AI Initiative to bring together experts and develop best practices.
“This study is particularly relevant in an increasingly data-driven society. One of the goals of the Responsible AI Initiative is to study the fair application of artificial intelligence methods. Our results suggest that biases in the data may be as important to consider as any algorithmic bias,” said Brewer.
Other authors of the study include Michelle Debbink from the Departments of Obstetrics and Gynecology, Thomas Cova from the School of Environment, Society, and Sustainability, and Tracy Onega from Population Health Sciences at the University of Utah.
Journal Reference:
Kelly, C, B., et al. (2025) Racial and Ethnic Disparities in Regulatory Air Quality Monitor Locations in the US. JAMA Network Open. doi.org/10.1001/jamanetworkopen.2024.49005