This dissertation investigates human exposure to vehicular air pollutant emissions in urban areas. Since resources for protecting human health from the adverse consequences of inadvertent environmental releases are constrained, it is often desirable to identify sources and settings in which emissions controls could lead to especially high human health benefits per unit effort. The three measurement and modeling studies that comprise this dissertation aim to contribute towards this goal by advancing a mechanistic understanding of the relationship between urban vehicle emissions and subsequent human exposures. Two key themes that permeate these investigations include the exposure consequences of vehicle emissions in low-income settings such as developing world cities, and the role of dynamic processes in influencing the emissions-to-exposure relationship for urban air pollutant sources. Chapter 1 introduces each of the dissertation chapters and provides context and background related to the broader themes motivating the investigation.

In Chapter 2, I report on exposures to particulate matter (PM) in the megacity of New Delhi, India. Previous work has identified New Delhi as a hotspot for ambient PM pollution. To investigate the degree to which in-vehicle exposures can be represented by ambient fixed-site measurements in New Delhi, I undertook a multi-month field campaign in 2010. In-vehicle measurements focused on concentrations of fine particulate matter (PM_2.5), black carbon (BC) and ultrafine particles (UFP, measured by particle number count, PN) inside the cabins of auto-rickshaws, a common type of unenclosed vehicle in South Asia. Supplemental measurements considered PM levels inside conventional (enclosed) automobiles. Contemporaneously with the in-vehicle measurements, I conducted routine ambient monitoring of PM_2.5, BC and PN at a rooftop fixed site.

In-vehicle particle concentrations measured during this field campaign were substantially elevated relative to the levels recorded at the ambient monitoring site. Geometric mean concentrations inside the auto-rickshaw, averaged over ~160 h of 1 Hz data, were 190 µg m^-3 PM_2.5, 42 µg m^-3 BC, and 280 × 10³ particles cm^-3. These concentrations rank among the highest levels ever reported for routine transportation microenvironments. Short-duration peak concentrations (averaging time: 10 s), attributable to exhaust plumes of nearby vehicles, were greater than 300 µg m^-3 for PM_2.5, 85 µg m^-3 for BC, and 650 × 10³ particles cm^-3 for PN. In-vehicle PM2.5 levels were 1.5× higher than the high ambient PM_2.5 concentrations (geometric mean: 120 µg m^-3) in Delhi. In-vehicle BC and PN levels were more substantially elevated above background levels (respectively 3.6× and 8.4×). The especially high degree of elevation for PN suggests that in-vehicle exposures might account for a large fraction of daily PN exposure for auto-rickshaw users. The in-vehicle amplification for PN is likely attributable to proximity to a major PN source (traffic emissions) as well as dynamic loss processes such as coagulation that may remove UFP from ambient air.

A small subset of measurements collected inside conventional cars with open windows resulted in similar mean concentrations to contemporaneous measurements inside auto-rickshaws. In contrast, concentrations were somewhat lower inside automobiles with air conditioning, likely owing to dynamic in-vehicle particle removal mechanisms. Overall, this investigation concludes that in-vehicle exposures in New Delhi substantially exceed the high ambient background concentrations recorded at fixed sites.

Chapter 3 presents a global analysis of the population exposure implications of urban vehicle emissions using the intake fraction (iF) metric. Intake fraction is a dimensionless parameter that represents the fraction of a source's emissions that are ultimately inhaled by all exposed individuals. In this chapter, I develop and apply a model to estimate iF for spatially distributed, ground-level emissions (e.g., from vehicles) in 3646 worldwide urban areas, each with year-2000 population > 100,000. This large dataset of cities accounts for ~ 2.0 billion people, roughly ~70% of the year-2000 urban population. The investigation develops the first-ever iF estimates for urban emissions in many regions outside of North America and Europe, including for numerous megacities for which iF data did not previously exist.

In particular, Chapter 3 considers the intraurban iF for each of the cities in this dataset, which accounts for the inhalation exposure to an urban area's emissions that occurs within that same city. Base-case model runs consider an archetypal primary, conserved, non-reactive pollutant. Sensitivity scenarios consider primary pollutants with first-order decay. These broad classifications are representative of many health-relevant pollutants emitted by vehicles and other urban sources. Moreover, they provide a point of comparison for understanding the exposure implications of non-conserved and secondary pollutants, which are explored in more detail in Chapter 4.

For conserved primary pollutants, population-weighted median, mean, and interquartile range iF values are 26, 39, and 14 - 52 ppm, respectively, where 1 ppm signifies 1 g inhaled per tonne emitted. The global mean urban iF determined here is roughly twice as large as previous estimates for cities in the United States and Europe, owing primarily to the inclusion of cities with higher iF located outside of these two regions. Intake fractions vary among cities owing to differences in population size, population density, and meteorology. Sorting by size, population-weighted mean iF values are 65, 35, and 15 ppm, respectively, for cities with populations larger than 3 million, 0.6 - 3 million, and 0.1 - 0.6 million. For the 20 worldwide megacities in the dataset, the population-weighted mean intraurban iF is 83 ppm. Overall, mean iF values are greatest in Asia and lowest in land-rich high-income regions, owing primarily to differing patterns in urban form between these two regions. Among the 10 countries with the largest urban populations, population-weighted mean intraurban iF varies by a factor of 3. Intake fraction results for individual cities are predicted well by a parsimonious regression model that incorporates metrics of urban land area, population density, and meteorology.

Chapter 4 extends the concept of intake fraction to incorporate semivolatile organic emissions. The analysis emphasizes the consequences of these emissions for population exposure to organic particulate matter, which is a major constituent in both vehicle exhaust and ambient urban air. Organic aerosols (OA) blur traditional notions of primary and secondary pollutants owing to dynamic exchange of material between the vapor and particle phases. Dilution of fresh organic PM emissions (primary organic aerosol, POA) with ambient air typically causes a profound shift of material from particle to vapor phase. Relatively more volatile vapor-phase material is then "aged" into lower-volatility products over its residence time in a regional airshed via oxidation reactions initiated by photochemically produced radicals (e.g., the hydroxyl radical OH). In turn, these lower volatility products oxidized from evaporated emissions then condense to form quasi-secondary particles that make up the so-called oxidized primary organic aerosol (OPOA).

In this analysis, I update the definition of intake fraction to accommodate the cumulative contributions of population exposure to primary and quasi-secondary organic particles (i.e., POA and OPOA) as well as vapor-phase material to the overall intake fraction for semivolatile organic emissions. As in Chapter 3, the primary emphasis of the analysis is on vehicles and other urban, ground level emissions sources. Because photochemical aging at the regional scale is the major mechanism for converting evaporated POA material into OPOA, I develop and employ a nested multi-compartment mechanistic model to consider exposures at the urban, periurban and regional scales with a 400-km domain. Base-case model simulations consider an archetypal medium-sized US city (population 1.5 M); alternative cases include a model of iF for a global megacity (population 12 M). Key transformation processes for semivolatile emissions (e.g., dilution, partitioning, aging) are represented using the Volatility Basis Set (VBS) framework. A major goal of the modeling exercise is to contrast the magnitude and spatial distribution of iF for semivolatile organic emissions with patterns in iF for nonreactive pollutants.

For urban emissions of non-reactive particles, ~75% of domain-wide population intake occurs in the same urban compartment as emissions. In contrast, for semivolatile emissions, spatial patterns and gas-particle partitioning of intake depend substantially on emissions volatility. Low volatility organic emissions in urban areas produce predominantly intraurban, particle-phase exposures (similar to inert pollutants). As volatility of material emitted in urban areas increases, three key trends emerge that reduce particle-phase iF: (1) the overall proportion of population exposure that takes place in the particle phase decreases and the proportion of exposure in the gas phase increases, (2) photochemically aged material (OPOA) accounts for a larger fraction of particle-phase population intake, and (3) regional-scale exposures account for the predominant fraction of organic aerosol exposure attributable to urban precursor emissions.

Since higher volatility compounds account for a large fraction of motor vehicle emissions, the overall iF for organic particles attributable to urban semivolatile organic emissions is lower than for inert pollutants. For example, for the default emissions volatility distribution and considering the base-case model, the particle- and vapor-phase intake fractions for semivolatile emissions are respectively 3.1 ppm and 12.5 ppm. Thus, exposure to organic particle-phase material accounts for only ~ 20% of the population intake of all semivolatile organic emissions. For comparison, the iF for primary, non-reactive, conserved pollutants is substantially higher (iF ~ 16.9 ppm) and is only modestly attenuated for inert particles that are subject to loss via dry deposition (iF ~ 12.9 ppm).

The dissertation concludes with a summary and synthesis chapter (Chapter 5), which explores policy implications and provides recommendations for future research.

Seasonally Optimized Calibrations Improve Low-cost Sensor Performance: Long-term Field Evaluation of PurpleAir Sensors in Urban and Rural India

Lower-cost air pollution sensors can fill critical air quality data gaps in India, which experiences very high fine particulate matter (PM2.5) air pollution but has sparse regulatory air monitoring. Challenges for low-cost PM2.5 sensors in India include high aerosol mass concentrations and pronounced regional and seasonal gradients in aerosol composition. Here, we report on a detailed, long-time performance evaluation of a popular sensor, the Purple Air PA-II, at multiple sites in India. We established 3 distinct sites in India across land-use categories and population density extremes (North India: Delhi [urban], Hamir- pur [rural]; South India: Bangalore [urban]), where we collocated the PA-II with reference beta-attenuation monitors. We evaluated the performance of uncalibrated sensor data, and then developed, optimized, and evaluated calibration models using a comprehensive feature selection process focused on reproducibility in the Indian context. We assessed the seasonal and spatial transferability of sensor calibration schemes, which is especially important in India because of the paucity of reference instrumentation. Without calibration, the PA-II was moderately correlated with the reference signal (R2: 0.55 - 0.74) but was inaccurate (NRMSE ≤ 40%). Relative to uncalibrated data, parsimonious annual calibration models improved PA performance at all sites (cross-validated NRMSE 20-30%, R2: 0.82-0.95), and greatly reduced seasonal and diurnal biases. Because aerosol properties and meteorology vary regionally, the form of these long-term models differed among our sites, suggesting that local calibrations are desirable when possible. Using a moving-window calibration, we found that using seasonally-specific information improves performance relative to a static annual calibration model, while a short-term calibration model generally does not transfer reliably to other seasons. Overall, we find that the PA-II can provide reliable PM2.5 data with better than ± 25% precision and accuracy when paired with a rigorous calibration scheme that accounts for seasonality and local aerosol composition.

Mapping Air Pollution in Polluted Data-Sparse Environments: Resolving Spatial-Temporal PM2.5 Trends with Lower-cost Sensors in North India

North India faces the world’s highest air pollution health burden across the urban-rural divide. Ground-based monitoring is essential for validating models and conducting holistic exposure assessments, but the high costs of reference instruments have limited their adoption to megacities. Here, we demonstrate an analytical approach to scaling a lower-cost sensor (LCS) network in North India using the commercially available PurpleAir PA-II sensor package. Our key constraint was to balance data completeness with representativeness, given the dynamic nature of LCS performance both out-of-the-box and for long-term deployment. We developed and applied our models across 13 diverse land-use collocations in the Delhi National Capital Region, resulting in robust out-of-sample seasonal performance (mean bias < 10%). Validating our framework in Lucknow, another megacity in the region, demonstrated a high correlation between aggregated community-hosted LCS nodes and reference networks (Pearson’s r ≥ 0.9) and well-constrained annual PM2.5 estimates (≤ 10%). While our data pipeline was developed in North India, our statistical framework offers paths forward for scaling similar regional networks in polluted, sparsely monitored environments.

Air Pollution Management Urgently Needed in Non-urban North India

High emissions from a diverse source mixture and unfavorable meteorology have resulted in extreme fine particulate matter (PM2.5) air pollution in the Indo-Gangetic Plain (IGP) of North India. Most studies and abatement strategies have focused on densely populated megacities, but the majority of the population in the region (60-70%) reside in non-urban settlements. Therefore, although non-urban air quality is understudied in the IGP, it is critical to understand the regional health burden of PM2.5 and develop informed policy. We implemented a lower-cost PM2.5 monitoring network of over 80 sites across more than 15 settlements using the popular PurpleAir sensor. Our network spanned megacities to regional background sites and represents the first ground-based observations of PM2.5 in most settlements. We observed sustained poor ambient air quality across the region (annual average PM2.5 concentration ≥ 60 µg m−3), with weak spatial gradients from urban core sites to regional background sites. Leveraging the high temporal resolution afforded by lower-cost sensors, we observed that although annual and seasonal trends were strongly correlated, diurnal patterns diverged across settlement population density strata. Megacities featured smooth diurnal profiles, with peaks later in the evening and larger intra-settlement variability- indicating the complex mixture of traffic, heavy industry, and other sources. Conversely, non-urban and small city sites featured earlier and higher magnitude diurnal peaks (1.5 -1.7 × daily minimum), with low intra-settlement variability, likely indicating a higher relative prevalence of biomass burning. Policymakers should ensure clean air programs do not solely focus on relatively well-studied megacities at the expense of the large non-urban population distributed across the IGP.

Background

Methods

Results

Conclusions