Considerations in the use of ozone and PM2.5 data for exposure assessment
- First Online:
- 462 Downloads
The US national ambient-air monitoring network, created to verify compliance with health-based standards, now doubles as an important source of exposure data for the epidemiological analyses on which these standards increasingly rest, particularly in the case of ozone and PM2.5. This paper was written for a workshop called to facilitate and inform the use of routine ozone and PM2.5 data by the Environmental Public Health Tracking Network. It examines the fit between priorities that shape regulatory monitoring and modeling and the data needs of public health tracking.
KeywordsOzone PM2.5 Ambient air quality standard Monitor Measurement Model
The 1963 Clean Air Act and its subsequent amendments identify “criteria pollutants” for which the US Environmental Protection Agency (EPA) is required to establish National Ambient Air Quality Standards (NAAQS) to protect human health with adequate margins of safety. Setting standards in terms of community air concentrations rather than source emission rates is a key feature of the statutory framework in which US air quality is managed (Seinfeld 1975, pp 27–32). This focus has led EPA and its local regulatory partners to develop a substantial national infrastructure of ambient air monitoring to assess communities’ compliance and atmospheric modeling to guide their emissions-reduction strategies. The national ambient air monitoring network, created to verify compliance with health-based regulatory standards, yields also a consistent and extensive record of air quality that is readily available for atmospheric research and other uses. It is an important source of exposure data for the epidemiological analyses that inform the NAAQS, bringing analytical closure to an ongoing cycle of air quality management (USEPA 2006b; USEPA 2008).
In 2002, the Centers for Disease Control and Prevention (CDC) initiated a program of Environmental Public Health Tracking (EPHT) seeking to promote more systematic linking of information on environmental hazards, exposures, and health effects (McGeehin et al. 2004). Ozone and PM2.5 are the air pollutants selected for initial study because air monitoring data are available, high levels of these contaminants are the main cause of poor air quality in much of the country, and they have been strongly linked with respiratory and cardiovascular health effects (CDC 2004). CDC joined EPA and the Health Effects Institute in sponsoring a January 2008 workshop in Baltimore titled Methodologies for Environmental Public Health Tracking of Air Pollution Effects. The workshop brought participants in the EPHT program together with US and international experts in biostatistics and air pollution epidemiology, as described elsewhere in this issue.
High concentrations of ozone and fine particle mass are indicators of unhealthful air, but these two parameters do not fully characterize the air’s composition and toxicity. An exclusive focus on improving the spatiotemporal resolution of ozone and PM2.5 data may not yield proportionate benefits for tracking public health effects.
Compliance determinations typically place more weight than do epidemiologic analyses on a measurement’s absolute accuracy. EPHT tools could usefully exploit some of the abundant additional air data that EPA collects or produces for other purposes.
Standard regulatory models of the atmosphere are optimized for predicting the impacts of known emissions from controllable major sources. They offer less reliable representations of the myriad emissions from smaller, less well-characterized sources that contribute to a population’s total exposure.
Parameterization of “air quality”
Community air is a complex and dynamic mix of gases and suspended particles in constant interaction; it would not be feasible for a routine national network to monitor all of its constituents. While carbon monoxide and particulate lead have NAAQS specific to their health effects, ozone and PM2.5 are monitored for NAAQS that address effects from mixtures of pollutants associated chemically or physically with their eponymous indicators (Arnold et al. 2007). Ozone is a unique species, but the ozone concentration serves as the indicator for a reactive mix of ozone and other photochemical oxidants. Fine particulate matter is a heterogeneous mixture, and its total mass concentration PM2.5 serves as an indicator that a recent review (NRC 2004, p 9) comments “greatly oversimplifies complex biological phenomena” addressed by its NAAQS.
A limitation of ozone and PM2.5 as a parameterization of air quality is that they do not represent the same oxidant and particulate mixes near a source as they do farther away. They do not, for example, capture the localized gradients in primary emissions from cars and trucks thought to account for chronic health risks found near roadways (e.g., Hoek et al. 2002). As noted above, highways are marked by ozone deficits even in oxidant-rich air. The combustion nuclei in vehicle exhaust, whatever their possibly disproportionate impact on health, generally represent a minor fraction of total PM2.5 mass (e.g., Roosbroeck et al. 2008). Air quality maps limited to ozone and PM2.5 will thus fail to fully characterize traffic-related air pollution regardless of their spatial resolution. Similar considerations apply to industrial or municipal point sources that may create localized “hot spots” for especially toxic trace species.
Availability of measurements
Ozone and PM2.5 are routinely monitored for compliance with their NAAQS by the Federal Reference Method (FRM) or a Federal Equivalent Method (FEM). (EPA specifies a reference method for measurement when it promulgates a standard; it designates other methods as equivalent if they demonstrate agreement with the reference method.) States, tribes, and local agencies establish and operate compliance networks following specific EPA guidelines for siting, instrumentation, and quality assurance. The resulting data are submitted to EPA’s Air Quality System (AQS) database by the end of the quarter following the quarter of their collection. AQS is an attractive source of uniform, timely, and quality-assured air monitoring data.
The different frequencies of the FRM/FEM data for ozone and PM2.5 reflect the networks’ design for compliance monitoring rather than time series analyses of acute health effects. The health-based ozone NAAQS has always targeted extreme values, specifying an 8-h (formerly 1 h) concentration not to be exceeded more than a handful of days each year. Verifying compliance with a standard of this form requires continuous monitoring, at least during seasons with the potential for high concentrations. In contrast, the new PM2.5 NAAQS introduced in 1997 included a limit on the annual mean as well as one on extreme 24-h concentrations. The annual standard was generally controlling, in the sense that it was hard to violate the 24-h standard without also exceeding the annual mean. Compliance monitoring for PM2.5, therefore, focused on the annual mean, which usually could be estimated from measurements on a representative sample of days. The EPA significantly tightened the 24-h standard in December 2006 and supported this change by moving to daily sampling at about 50 monitoring sites previously sampling 1 day in three (USEPA 2006a).
Non-FRM/FEM data for PM2.5 are available on the every-third-day schedule of most FRM/FEM monitors from two networks that monitor particle speciation (VIEWS 2007). The Interagency Monitoring of Protected Visual Environments (IMPROVE) operated about 160 sites in 2006 at predominantly rural or remote locations. The Chemical Speciation Network/Speciation Trends Network (CSN/STN) operated about 60 population-oriented sites every third day and about 125 more every sixth day. These networks weigh 24-h samples on Teflon filters as the FRM does, but use samplers with inlets and flow rates different from FRM specifications.
Daily PM2.5 measurements are made at many more locations by continuous monitors, in support of EPA’s AIRNow public-reporting program. About 580 sites throughout the US supply hourly data (Chan 2007, personal communication) that are reduced to broad ranges for real-time display on a national map (http://www.airnow.gov). Various measurement methods are used (Hanley 2006), including nephelometry, beta attenuation (BAM), and oscillating microbalance (TEOM). These real-time data are qualified as “not fully validated … [and] only approved for the expressed purpose of reporting and forecasting the Air Quality.”
EPA gave its first FEM designation to a continuous PM2.5 monitor shortly after the Baltimore workshop, and substantial growth in the availability of continuous FEM data can be expected in the near future. The FRM/FEM designation is important for compliance monitoring because particle measurements are sensitive to methodological details. Some ambient particles are in equilibrium with surrounding gases, an equilibrium that can shift after they are sampled onto an FRM/FEM filter through which air is drawn for 24 h. Because the PM2.5 NAAQS is set in terms of the FRM, a method that avoided such sampling artifacts (or exhibited different ones) would not be suited to monitoring NAAQS compliance. Additionally, nongravimetric methods require calibration factors that can vary with particle composition and ambient humidity. It is clearly undesirable to have site-specific calibrations influence compliance determinations that span diverse climates and regulatory jurisdictions.
Availability of models
The relationship of inhaled air to monitored air is embedded in any observed statistical association between community health and air quality. As fixed monitors collect only discrete samples from the continuous atmosphere to which people are actually exposed, this relationship must generally be modeled. City-scale studies have commonly modeled general populations’ individual exposures as Berksonian departures from a city-wide air quality that is estimated by averaging all local measurements in each monitoring period (e.g., Samet et al. 2000). Such simple models are consistent with the interpretation, sketched earlier, of ozone and PM2.5 as generic indicators for broader chemical mixes.
Some metropolitan-scale and regional-scale studies have used contouring, land-use, or atmospheric models to relate measured concentrations to individual exposures. EPA’s current operational-level understanding of emissions and transport factors is incorporated in its Community Multiscale Air Quality (CMAQ) grid model, which has been used since 2004 to produce real-time national forecasts of hourly ozone concentrations (NWS 2007). CMAQ’s PM2.5 routines are much younger than its ozone routines, which trace their ancestry back through generations of critical scrutiny (e.g., NRC 1991). A recent report (USEPA 2005) found its performance in the western US to be significantly poorer for all PM2.5 species than in the eastern US, which had been the early focus of evaluations. Users can expect CMAQ’s PM2.5 performance to evolve and improve as it undergoes more cycles of review and development.
CMAQ’s primary function has been to support the evaluation of alternative strategies for managing emissions to attain ambient standards. It is accordingly source-oriented, taking emissions and winds as known and predicting the ambient concentrations that result under various regulatory scenarios. One limitation of any source-oriented PM2.5 model is that elevated concentrations in the real world can result from sporadic and hard-to-characterize fugitive emissions. Unlike CO and SO2, which emerge predictably from tailpipes and stacks through which their fluxes can be measured and documented, episodes of dust and smoke typically reflect agricultural and construction activities, wildfires, and other erratic and diffuse sources. Even concentrations measured near such sources are not reliably convertible to the mass fluxes needed as model inputs. Evaluations of CMAQ repeatedly show the agreement of its predictions with measured 24-h concentrations to be better for sulfate than for any other pollutant (Mebust et al. 2003; USEPA 2005). It is no coincidence that our knowledge of emissions is also better for SO2, the primary precursor to sulfate, than it is for any other air pollutant. Because well-determined species can “steal” explanatory significance from poorly determined covariates in multivariable regression (White 1998), differential uncertainties in emissions can distort the modeled associations of individual species or source contributions with health effects.
CMAQ and the monitoring networks exhibit somewhat complementary strengths and weaknesses as sources of air data. Point measurements represent “true” concentrations, as defined by regulations and the epidemiological findings that motivate them. However, they have no necessary relationship to one another; in sparsely monitored areas, they leave uncertain the boundaries between clean and dirty air. In contrast, CMAQ yields a logically coherent grid of concentrations that reflects our understanding of emissions and atmospheric processes. These concentrations can be unrepresentative of reality, however, when they are derived from inaccurate descriptions of emissions and the atmosphere.
Model outputs would ideally be reconciled with available measurements by adjusting uncertain model inputs. Prior probability distributions would be assigned to the intensity and geographical distribution of emissions, to wind fields, and to empirical parameterizations of atmospheric transformations and then revised in light of observed concentrations. The iterations required for a fully Bayesian solution would be impractical with the massive CMAQ code, although initial steps in this direction have been taken with simpler models (Husar et al. 1986; Schichtel et al. 2006). A more tractable but still computationally demanding approach to assimilating observations with CMAQ is the hierarchical Bayesian (HB) fusion model described by McMillan et al. (2009, submitted for publication) and explored in early EPHT efforts (Boothe et al. 2005, 2006). Daily ozone and PM2.5 estimates from this statistical hybrid are currently available from 2001 to 2005 for 12- and 36-km grids over the contiguous United States (USEPA 2009).
McMillan et al. (2009, submitted for publication) used PM2.5 concentrations in the eastern US during 2001 for their initial development and evaluation of the HB fusion model. In 2001, there were no data quality objectives for continuous PM2.5 monitors (USEPA 2002) and no collation and mapping of continuous data by AIRNow. In this setting, the authors relied on FRM/FEM data for their observational inputs, reserving non-FEM measurements (from the IMPROVE and CSN/STN speciation networks) for cross-validation of the results. In 2004 and later years, the continuous PM2.5 data reviewed by Mintz and Schmidt (2007) provide much fuller observational coverage of the space–time grid than the predominantly 1-in-3-day FRM/FEM data offer, and any inequivalence between them can be accounted for within the Bayesian framework. Incorporating the continuous data seems a natural next step for tracking analyses to explore, particularly given EPA’s announced intention to facilitate their substitution for filter-based FRM measurements in the future (USEPA 2006a).
Spatial gradients in ozone and PM2.5 capture variations in the overall hazard from air pollution adequately at some scales and poorly at others, a consideration reflected in the guidelines by which air monitors are sited. Enhanced spatial resolution of these two indicators may, therefore, have limited value for health studies unless accompanied by the enhanced chemical resolution provided by measurement of additional species.
The common temporal modulation of all species concentrations by synoptic (i.e., regional-scale) weather patterns may, conversely, limit the value of enhanced chemical resolution unaccompanied by enhanced spatial resolution to map the gradients in the mix.
EPA generates various categories of ozone and PM2.5 data to support planning and public communication efforts as well as consequential regulatory actions. Data used for EPHT need not meet some of the more legalistic requirements placed on data used to determine compliance with regulatory standards.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.