Using Low-Cost Sensor Networks: Considerations to Help Reveal Neighborhood-Level Exposure Disparities

The growing availability of low-cost sensors can potentially democratize the process for reducing disparities in exposures to harmful air pollution. When used collaboratively with government agencies and researchers, sensors deployed by community organizations can build trust in environmental decision-making.¹ Low-cost continuous sensors can complement regulatory monitoring networks required by the Clean Air Act, which have high confidence but relatively low geographic coverage. Sensors are often portable or even mobile and can prove particularly useful if they measure some of the same air pollutants as regulatory monitors.

Sensors deployed on a neighborhood scale can reveal spatial and temporal variations in air quality, as Esie et al. (p. 1765) show. Increased temporal resolution can identify episodes of poor air quality that exacerbate existing inequities in exposure or if those episodes, when compared with mean air quality levels, create new inequities. Increased temporal resolution can show when exacerbations happen and, in combination with higher spatial resolution, can then reveal the cause. Identifying the sources of the emissions provides communities and decision-makers with the information needed for action in addressing inequities. However, communities and government agencies must work together to agree on how to interpret and evaluate sensor data, especially in cases when it may not agree with regulatory monitors, to prevent friction and loss of trust.¹

Sensors have a wide appeal and market availability, but the quality of data they generate must be considered. To help those using sensors as part of air monitoring, the US Environmental Protection Agency’s (EPA’s) Air Sensor Toolbox² provides the latest science on sensor performance and operation, and the EPA is providing $20 million in grants to enhance community and local efforts in monitoring air quality, including in or near underserved communities.³ In addition, the Inflation Reduction Act⁴ contains provisions to deploy air monitoring in communities, including deploying sensors in low-income and disadvantaged communities.

SELECTION OF DATES, TIMES, AND LOCATIONS IS CRITICAL

Even for low-cost sensors, air quality measurement campaigns can be resource intensive, and thus decisions often need to be made about where and when to take measurements. Esie et al. conducted measurements for July 2021 because July had historically shown higher fine particulate matter (diameter ≤|2.5 µm; PM_2.5) levels. Although overall PM_2.5 levels were below daily standards, there were relatively elevated PM_2.5 measurements, predictably on July 4 and unexpectedly on July 23 because of a wildfire smoke incursion event. The latter event showed minimal variation across neighborhoods with different sociodemographic profiles. Summer months often show a high contribution from regional sulfate from power generation (although this contribution has fallen over time), and, as such, more local contributions may be masked. Looking at other months may have revealed more significant disparities across neighborhoods, perhaps because of greater proportional contributions from local industries or from urban transportation or differences in heating emissions. Recent trends show that in many regions of the country, including Chicago, relative peaks in PM_2.5 now occur in the winter, and those peaks may be associated with more local emission sources.⁵ Other temporal events of concern would be short-term sources of emissions from industrial sources (such as shutdown/startup malfunctions or maintenance), particularly if those sources are proximate to communities with environmental justice concerns. Temporary increases in emissions such as these would be both isolated in time and space, in contrast to the two events in July.

For the purposes of understanding exposure disparities, focusing on spatial or temporal excursions from mean total PM_2.5 levels may be more useful than looking at total PM_2.5. Identifying a local “hot spot” that might contribute to disparities in exposure would entail subtracting citywide, regional, and national contributions until only the excess PM_2.5 associated with local contributors remained. An approach has recently been proposed to remove regional background and provide a decomposition of PM_2.5 air pollution into long-range, midrange, neighborhood, and near-source for all census tracts in the United States,⁶ and this approach may also remove autocorrelation in a more structural way rather than using spatial lags. Further removing longer-term temporal trends from these spatially decomposed PM_2.5 levels would highlight temporal excursions that may also lead to additional disparities. In both cases, the analyses not only would identify when and where disparities occur but also could help to diagnose the emission sources that cause the disparities.

Siting of a network of low-cost sensors can be focused on diagnosing where and when inequities in exposure occur and on identifying the cause(s) of the inequities. The siting should be done with community input. Esie et al. used sensors located at bus stations, which are convenient locations and could capture near-road PM_2.5 exposures. However, these locations might not be best for identifying PM_2.5 exposures from industrial or other sources.

UNDERSTANDING DISPARITIES REQUIRES EQUITABLE NETWORKS

When properly sited, and with a dense-enough sensor network, it becomes possible to predict PM_2.5 levels at other neighborhood locations. For example, a community may wish to identify places of neighborhood concern or places with sensitive populations. Inverse distance weighting (IDW) or cokriging approaches that incorporate additional information such as wind directions⁷ can provide spatially resolved predictions that are similar in quality to land use regressions or downscaled model predictions.⁸ However, it is not clear that the density or location of sensors at bus stops satisfies criteria for using IDW as employed by Esie et al., and thus statistical models that seek to identify the disparities in PM_2.5 concentrations across different races using IDW may suffer from exposure misclassification. To respond to the concern of Esie et al. about temporally invariant covariates, it may be possible to use new land use regression methods⁹ that allow both spatial and temporal decomposition.

Esie et al. importantly note that crowdsourced sensor networks tend to be located in White, high socioeconomic status neighborhoods. If higher-income neighborhoods have more access to air quality sensors and more ability to respond to the information they generate, disparities in air pollution health outcomes can be exacerbated.¹ This reveals a need for more consistent, government-sponsored networks, which could promote interoperability and equitable access. By allowing a cross-comparison of data gathered using disparate sensor networks, information could be compared and shared on a broader scale.

CONCLUSION

The Esie et al. study adds to evidence that disparities in exposure continue to exist in Chicago and that certain types of emission events can exacerbate those disparities. The types of emission events identified are difficult to regulate, and the study design is not able to identify harder-to-diagnose sources of air pollution excursions. A greater focus on the times and places that have substantially higher neighborhood air pollution levels would advance two goals: a greater ability to ascertain the sources of inequities and information that can empower communities working with government agencies to prevent those emission events and reduce exposures. Finally, low-cost sensors, with their affordability and ease of deployment, have the potential to collect data that can reveal air quality and exposure disparities, but the data will have the most impact in rectifying disparities when communities and government agencies agree, preferably in advance, on how to evaluate and interpret the data.