Modelling and Assessing PM2.5 Exposure in Support of Ongoing Air Pollution Research in Mongolia

Abstract

This dissertation was conducted in support of an international research collaboration established to investigate how air pollution exposure may affect respiratory viral infections in two Mongolian cities (Darkhan City and Ulaanbaatar) where both air pollution levels and respiratory viral infection rates are among the highest in the world. Among the air pollutants present in these two cities, fine particulate matter with aerodynamic diameters less than or equal to 2.5 micrometers (PM2.5) was of key interest. A central research objective was to address missing PM2.5 measurements from historic, 3rd party datasets and within our primary datasets by imputing missing values using machine learning models. Another key objective was to evaluate residential PM2.5 concentrations measured during a recent field sampling initiative. These research objectives were developed through close collaboration with researchers within the United States and abroad in Mongolia. The key materials used to conduct this dissertation research included data from a variety of sources, computational software, and high-performance computing resources. The core methodologies used to achieve these research objectives included statistical analyses and data modelling using machine learning algorithms. This research increased the availability of PM2.5 datasets, identified the most important features driving the models which predicted these data, and yielded new insights pertaining to indoor and outdoor PM2.5 exposure in Darkhan City and Ulaanbaatar across the spatial temporal domain. The results and insights stemming from this research will be used to further ongoing research efforts and offer guidance to researchers investigating similar topics in other settings around the globe.