A novel principal component analysis for spatially misaligned multivariate air pollution data

Characterizing Determinants of Near-Road Ambient Air Quality for an Urban Intersection and a Freeway Site

INTRODUCTION

Near-road ambient air pollution concentrations that are affected by vehicle emissions are typically characterized by substantial spatial variability with respect to distance from the roadway and temporal variability based on the time of day, day of week, and season. The goal of this work is to identify variables that explain either temporal or spatial variability based on case studies for a freeway site and an urban intersection site. The key hypothesis is that dispersion modeling of near-road pollutant concentrations could be improved by adding estimates or indices for site-specific explanatory variables, particularly related to traffic. Based on case studies for a freeway site and an urban intersection site, the specific aims of this project are to (1) develop and test regression models that explain variability in traffic-related air pollutant (TRAP) ambient concentration at two near-roadway locations; (2) develop and test refined proxies for land use, traffic, emissions and dispersion; and (3) prioritize inputs according to their ability to explain variability in ambient concentrations to help focus efforts for future data collection and model development.

The key pollutants that are the key focus of this work include nitrogen oxides (NOx), carbon monoxide (CO), black carbon (BC), fine particulate matter (PM2.5; PM ≤ 2.5 μm in aerodynamic diameter), ultrafine particles (UFPs; PM ≤ 0.1 μm in aerodynamic diameter), and ozone (O3). NOx, CO, and BC are tracers of vehicle emissions and dispersion. PM2.5 is influenced by vehicle table emissions and regional sources. UFPs are sensitive to primary vehicle emissions. Secondary particles can form near roadways and on regional scales, influencing both PM2.5 and UFP concentrations. O3 concentrations are influenced by interaction with NOx near the roadway. Nitrogen dioxide (NO2), CO, PM2.5, and O3 are regulated under the National Ambient Air Quality Standards (NAAQS) because of demonstrated health effects. BC and UFPs are of concern for their potential health effects. Therefore, these pollutants are the focus of this work.

METHODS

The methodological approach includes case studies for which variables are identified and assesses their ability to explain either temporal or spatial variability in pollutant ambient concentrations. The case studies include one freeway location and one urban intersection. The case studies address (1) temporal variability at a fixed monitor 10 meters from a freeway; (2) downwind concentrations perpendicular to the same location; (3) variability in 24-hour average pollutant concentrations at five sites near an urban intersection; and (4) spatiotemporal variability along a walking path near that same intersection.

The study boundary encompasses key factors in the continuum from vehicle emissions to near-road exposure concentrations. These factors include land use, transportation infrastructure and traffic control, vehicle mix, vehicle (traffic) flow, on-road emissions, meteorology, transport and evolution (transformation) of primary emissions, and production of secondary pollutants, and their resulting impact on measured concentrations in the near-road environment. We conducted field measurements of land use, traffic, vehicle emissions, and near-road ambient concentrations in the vicinity of two newly installed fixed-site monitors. One is a monitoring station jointly operated by the U.S. Environmental Protection Agency (U.S. EPA) and the North Carolina Department of Environmental Quality (NC DEQ) on I-40 between Airport Boulevard and I-540 in Wake County, North Carolina. The other is a fixed-site monitor for measuring PM2.5 at the North Carolina Central University (NCCU) campus on E. Lawson Street in Durham, North Carolina. We refer to these two locations as the freeway site and the urban site, respectively. We developed statistical models for the freeway and urban sites.

RESULTS

We quantified land use metrics at each site, such as distances to the nearest bus stop. For the freeway site, we quantified lane-by-lane total vehicle count, heavy vehicle (HV) count, and several vehicle-activity indices that account for distance from each lane to the roadside monitor. For the urban site, we quantified vehicle counts for all 12 turning movements through the intersection. At each site, we measured microscale vehicle tailpipe emissions using a portable emission measurement system.

At the freeway site, we measured the spatial gradient of NOx, BC, UFPs, and PM, quantified particle size distributions at selected distances from the roadway and assessed partitioning of particles as a function of evolving volatility. We also quantified fleet-average emission factors for several pollutants.

At the urban site, we measured daily average concentrations of nitric oxide (NO), NOx, O3, and PM2.5 at five sites surrounding the intersection of interest; we also measured high resolution (1-second to 10-second averages) concentrations of O3, PM2.5, and UFPs along a pedestrian transect. At both sites, the Research LINE-source (R-LINE) dispersion model was applied to predict concentration gradients based on the physical dispersion of pollution.

Statistical models were developed for each site for selected pollutants. With variables for local wind direction, heavy-vehicle index, temperature, and day type, the multiple coefficient of determination (R2) was 0.61 for hourly NOx concentrations at the freeway site. An interaction effect of the dispersion model and a real-time traffic index contributed only 24% of the response variance for NOx at the freeway site. Local wind direction, measured near the road, was typically more important than wind direction measured some distance away, and vehicle-activity metrics directly related to actual real-time traffic were important. At the urban site, variability in pollutant concentrations measured for a pedestrian walk-along route was explained primarily by real-time traffic metrics, meteorology, time of day, season, and real-world vehicle tailpipe emissions, depending on the pollutant. The regression models explained most of the variance in measured concentrations for BC, PM, UFPs, NO, and NOx at the freeway site and for UFPs and O3 at the urban site pedestrian transect.

CONCLUSIONS

Among the set of candidate explanatory variables, typically only a few were needed to explain most of the variability in observed ambient concentrations. At the freeway site, the concentration gradients perpendicular to the road were influenced by dilution, season, time of day, and whether the pollutant underwent chemical or physical transformations. The explanatory variables that were useful in explaining temporal variability in measured ambient concentrations, as well as spatial variability at the urban site, were typically localized real-time traffic-volume indices and local wind direction. However, the specific set of useful explanatory variables was site, context (e.g., next to road, quadrants around an intersection, pedestrian transects), and pollutant specific. Among the most novel of the indicators, variability in real-time measured tailpipe exhaust emissions was found to help explain variability in pedestrian transect UFP concentrations. UFP particle counts were very sensitive to real-time traffic indicators at both the freeway and urban sites. Localized site-specific data on traffic and meteorology contributed to explaining variability in ambient concentrations. HV traffic influenced near-road air quality at the freeway site more so than at the urban site. The statistical models typically explained most of the observed variability but were relatively simple. The results here are site-specific and not generalizable, but they are illustrative that near-road air quality can be highly sensitive to localized real-time indicators of traffic and meteorology.

Probabilistic predictive principal component analysis for spatially misaligned and high-dimensional air pollution data with missing observations

Accurate predictions of pollutant concentrations at new locations are often of interest in air pollution studies on fine particulate matters (PM2.5), in which data is usually not measured at all study locations. PM2.5 is also a mixture of many different chemical components. Principal component analysis (PCA) can be incorporated to obtain lower-dimensional representative scores of such multi-pollutant data. Spatial prediction can then be used to estimate these scores at new locations. Recently developed predictive PCA modifies the traditional PCA algorithm to obtain scores with spatial structures that can be well predicted at unmeasured locations. However, these approaches require complete data, whereas multi-pollutant data tends to have complex missing patterns in practice. We propose probabilistic versions of predictive PCA which allow for flexible model-based imputation that can account for spatial information and subsequently improve the overall predictive performance.

Adaptive predictive principal components for modeling multivariate air pollution

Air pollution monitoring locations are typically spatially misaligned with locations of participants in a cohort study, so to analyze pollution-health associations, exposures must be predicted at subject locations. For a pollution measure like PM_2.5 (fine particulate matter) comprised of multiple chemical components, the predictive principal component analysis (PCA) algorithm derives a low-dimensional representation of component profiles for use in health analyses. Geographic covariates and spatial splines help determine the principal component loadings of the pollution data to give improved prediction accuracy of the principal component scores. While predictive PCA can accommodate pollution data of arbitrary dimension, it is currently limited to a small number of pre-selected geographic covariates. We propose an adaptive predictive PCA algorithm, which automatically identifies a combination of covariates that is most informative in choosing the principal component directions in the pollutant space. We show that adaptive predictive PCA improves the accuracy of multi-pollutant exposure predictions at subject locations.

A Multivariate Dynamic Spatial Factor Model for Speciated Pollutants and Adverse Birth Outcomes

Evidence suggests that exposure to elevated concentrations of air pollution during pregnancy is associated with increased risks of birth defects and other adverse birth outcomes. While current regulations put limits on total PM2.5 concentrations, there are many speciated pollutants within this size class that likely have distinct effects on perinatal health. However, due to correlations between these speciated pollutants, it can be difficult to decipher their effects in a model for birth outcomes. To combat this difficulty, we develop a multivariate spatio-temporal Bayesian model for speciated particulate matter using dynamic spatial factors. These spatial factors can then be interpolated to the pregnant women's homes to be used to model birth defects. The birth defect model allows the impact of pollutants to vary across different weeks of the pregnancy in order to identify susceptible periods. The proposed methodology is illustrated using pollutant monitoring data from the Environmental Protection Agency and birth records from the National Birth Defect Prevention Study.