Geospatial methods are increasingly used to evaluate environmental exposures in epidemiologic studies. Geospatial data are often acquired from different sources and time periods and at different spatial resolutions. This dissertation presents methods for combining datasets that do not overlap in space and, or time and are essential for the accurate quantification of metrics used in Environmental Health Sciences. In three chapters, we demonstrated and compared advances in small-area geospatial methods used to reduce misalignment and misclassification of key variables (e.g., exposure).

First, we showed how small-area analyses of COVID-19 can benefit from spatial aggregation to account for areal misalignment. In an analysis of the association between air pollution levels and COVID-19 incidence and mortality, we used residential building footprints to combine misaligned COVID-19 outcomes recorded at the neighborhood level, population demographics recorded at the census-tract level, and NO2 interpolated surfaces recorded at 30-meter grids. We found NO2 to be positively associated with COVID-19 incidence and mortality for neighborhoods in Los Angeles.

Second, we used data acquired through the PASTA-LA study to attribute daily green space exposure for physically-active spaces (PASs) where activity, location, and green space data were misaligned. We used tracking data from accelerometers and smartphones to attribute green-space exposure using 21 geospatial methods. We found that exposures attributed to home-address buffers, a commonly used method, can result in exposure misclassification. We also found a large range in correlation depending on tracking device used and the way physical activity was categorized across the 21 methods, suggesting that the method selected is key to the findings.

Finally, third, also using PASTA-LA data, we studied the association between heat exposure and physical activity in which temperature, green space, and participant-level covariates were misaligned. We showed that heat exposure was associated with increased physical activity and that exposure to green space modified this association in only some models. In this chapter, and in all chapters of this dissertation, we demonstrated how choice in geospatial approach to spatiotemporal misalignment can yield different study results.

Automobile traffic has been a longstanding source of air pollution in human communities. The target of major regulations in the past few decades, the transportation sector has gone through significant changes, ranging from shifts in vehicle fleet composition to natural and artificial disruptions to traffic patterns. Both an essential form of transportation as well as a major source of air pollution, traffic, to this day, remains a human necessity and a public health challenge. As such, measuring and modeling the temporal and spatial distribution of traffic related air pollution (TRAP) is a critical necessity for exposure scientists, epidemiologists, and other public health professionals.

In this dissertation, we investigate methods to measure TRAP in response to recent trends and disruptions in traffic patterns and composition, with a particular focus on California State. To this end, we employ novel combinations of big data, including regulatory air quality data in addition to traffic, land-use, and internet-of-things network data. It is divide into five chapters: an introduction (chapter 1), three chapters of original research (chapters 2-4), and a discussion of the conclusions of the work (chapter 5).

Chapter 2 evaluates the near-road air quality impacts of traffic disruptions associated with the COVID-19 pandemic. Following global activity stoppages associated with stay-at-home measures, studies reported improved air quality in several cities and countries around the world. While widely observed, many studies could not properly attribute the decline of traffic in this chapter evaluates the relationship between traffic volume as reported by the California Department of Transportation and near-road NO and NO2 emissions at seven EPA monitoring sites in California state: four in Northern California and three in Southern California.

Chapter 3 models the spatial distribution of non-tailpipe emissions-related PM2.5 chemical species and oxidative potential in Southern California. Combining gold-standard filter samples, land-use data, and a novel internet-of-things low-cost sensor network dataset in a spatial regression (Co-Kriging with External Drift) model, we create a set of exposure surfaces for exposures that serve as tracers of TRAP. Results indicate that compared to typical modelling techniques, namely land-use regressions, the addition of low-cost sensor data improves model accuracy and precision.

Chapter 4 evaluates the associations between exposures modeled in Chapter 3 and the ischemic placental disease (IPD) in a prospectively-followed pregnancy cohort of 178 women. Air quality regulation have resulted in declines in tailpipe emissions in recent years. As stated earlier, TRAP is also generated from non-tailpipe sources, including brake and tire wear. Concerned that brake and tire wear particles contain metals and other organic compounds that could harm fetal health, this study uses a logistic regression model to estimate exposure outcome associations. Compared to conventional exposures, namely PM2.5 and black carbon, we find stronger associations between IPD and exposures more specific to brake and tire wear, such as barium, as well as oxidative potential markers.

At the time of filing, Chapter 2 has been published in Environmental Science and Technology Letters, Chapter 3 has been published in Environment International, and Chapter 4 is currently in preparation for submission to an academic journal.