The role of the spatial representation in the relation between chronic exposure to NO2 and respiratory health outcomes is studied through a spatial approach encompassing three conceptual components: the geography of susceptibility, the geography of exposure and the geography of risk. A spatially explicit methodology that defined natural neighbourhoods for the city of Ottawa is presented; it became the geography of analysis in this research. A LUR model for Ottawa is developed to study the geography of exposure. Model sensitivity to the spatial representation of population showed that dasymetric population mapping did not provide significant improvements to the LUR model over population at the dissemination block level. However, both the former were significantly better than population represented at the dissemination area. Spatial representation in the geography of exposure was also evaluated by comparing four kriging and cokriging interpolation models to the LUR. Geostatistically derived NO2 concentration maps were weakly correlated with LUR model results. The relationship between mean NO2 concentrations and respiratory health outcomes was assessed within the natural neighbourhoods. We find a statistically significant association between NO2 concentrations and respiratory health outcomes as measured by global bivariate Moran’s I. However, for regression model building, NO2 had to be forced into the model, demonstrating that NO2 is not one of the main contributing variables to respiratory health outcomes in Ottawa. The results point toward the importance of the socioeconomic status on the health condition of individuals. Finally, the role of spatial representation was assessed using three different spatial structures, which also permitted to better understand the role of the modifiable areal unit problem (MAUP) in the study of the relationship between exposure to NO2 and health. The results confirm that NO2 concentration is not a major contributing factor to the respiratory health in Ottawa but clearly demonstrate the implications that the use of opportunistic administrative boundaries can have on results of exposure studies. The effects of the MAUP, the scale effect and the zoning effect, were observed indicating that a spatial structure that embodies the scale of major social processes behind the health condition of individuals should be used when possible.

The role of the spatial representation in the relation between chronic exposure to NO2 and respiratory health outcomes is studied through a spatial approach encompassing three conceptual components: the geography of susceptibility, the geography of exposure and the geography of risk. A spatially explicit methodology that defined natural neighbourhoods for the city of Ottawa is presented; it became the geography of analysis in this research. A LUR model for Ottawa is developed to study the geography of exposure. Model sensitivity to the spatial representation of population showed that dasymetric population mapping did not provide significant improvements to the LUR model over population at the dissemination block level. However, both the former were significantly better than population represented at the dissemination area. Spatial representation in the geography of exposure was also evaluated by comparing four kriging and cokriging interpolation models to the LUR. Geostatistically derived NO2 concentration maps were weakly correlated with LUR model results. The relationship between mean NO2 concentrations and respiratory health outcomes was assessed within the natural neighbourhoods. We find a statistically significant association between NO2 concentrations and respiratory health outcomes as measured by global bivariate Moran’s I. However, for regression model building, NO2 had to be forced into the model, demonstrating that NO2 is not one of the main contributing variables to respiratory health outcomes in Ottawa. The results point toward the importance of the socioeconomic status on the health condition of individuals. Finally, the role of spatial representation was assessed using three different spatial structures, which also permitted to better understand the role of the modifiable areal unit problem (MAUP) in the study of the relationship between exposure to NO2 and health. The results confirm that NO2 concentration is not a major contributing factor to the respiratory health in Ottawa but clearly demonstrate the implications that the use of opportunistic administrative boundaries can have on results of exposure studies. The effects of the MAUP, the scale effect and the zoning effect, were observed indicating that a spatial structure that embodies the scale of major social processes behind the health condition of individuals should be used when possible.

The role of the spatial representation in the relation between chronic exposure to NO2 and respiratory health outcomes is studied through a spatial approach encompassing three conceptual components: the geography of susceptibility, the geography of exposure and the geography of risk. A spatially explicit methodology that defined natural neighbourhoods for the city of Ottawa is presented; it became the geography of analysis in this research. A LUR model for Ottawa is developed to study the geography of exposure. Model sensitivity to the spatial representation of population showed that dasymetric population mapping did not provide significant improvements to the LUR model over population at the dissemination block level. However, both the former were significantly better than population represented at the dissemination area. Spatial representation in the geography of exposure was also evaluated by comparing four kriging and cokriging interpolation models to the LUR. Geostatistically derived NO2 concentration maps were weakly correlated with LUR model results. The relationship between mean NO2 concentrations and respiratory health outcomes was assessed within the natural neighbourhoods. We find a statistically significant association between NO2 concentrations and respiratory health outcomes as measured by global bivariate Moran’s I. However, for regression model building, NO2 had to be forced into the model, demonstrating that NO2 is not one of the main contributing variables to respiratory health outcomes in Ottawa. The results point toward the importance of the socioeconomic status on the health condition of individuals. Finally, the role of spatial representation was assessed using three different spatial structures, which also permitted to better understand the role of the modifiable areal unit problem (MAUP) in the study of the relationship between exposure to NO2 and health. The results confirm that NO2 concentration is not a major contributing factor to the respiratory health in Ottawa but clearly demonstrate the implications that the use of opportunistic administrative boundaries can have on results of exposure studies. The effects of the MAUP, the scale effect and the zoning effect, were observed indicating that a spatial structure that embodies the scale of major social processes behind the health condition of individuals should be used when possible.

The role of the spatial representation in the relation between chronic exposure to NO2 and respiratory health outcomes is studied through a spatial approach encompassing three conceptual components: the geography of susceptibility, the geography of exposure and the geography of risk. A spatially explicit methodology that defined natural neighbourhoods for the city of Ottawa is presented; it became the geography of analysis in this research. A LUR model for Ottawa is developed to study the geography of exposure. Model sensitivity to the spatial representation of population showed that dasymetric population mapping did not provide significant improvements to the LUR model over population at the dissemination block level. However, both the former were significantly better than population represented at the dissemination area. Spatial representation in the geography of exposure was also evaluated by comparing four kriging and cokriging interpolation models to the LUR. Geostatistically derived NO2 concentration maps were weakly correlated with LUR model results. The relationship between mean NO2 concentrations and respiratory health outcomes was assessed within the natural neighbourhoods. We find a statistically significant association between NO2 concentrations and respiratory health outcomes as measured by global bivariate Moran’s I. However, for regression model building, NO2 had to be forced into the model, demonstrating that NO2 is not one of the main contributing variables to respiratory health outcomes in Ottawa. The results point toward the importance of the socioeconomic status on the health condition of individuals. Finally, the role of spatial representation was assessed using three different spatial structures, which also permitted to better understand the role of the modifiable areal unit problem (MAUP) in the study of the relationship between exposure to NO2 and health. The results confirm that NO2 concentration is not a major contributing factor to the respiratory health in Ottawa but clearly demonstrate the implications that the use of opportunistic administrative boundaries can have on results of exposure studies. The effects of the MAUP, the scale effect and the zoning effect, were observed indicating that a spatial structure that embodies the scale of major social processes behind the health condition of individuals should be used when possible.

No description available.

Increasing levels of air pollution are prompting researchers to develop more reliable air pollution modeling approaches in order to protect the public and the environment from toxic contaminants and airborne pathogens. Although land use regression has long been used to assess exposure to air pollution, researchers are increasingly using machine learning algorithms to quantify the concentration of harmful pollutants—for this study black carbon (BC) and particle number (PN). Additionally, researchers are moving away from using fixed-site data in favor of using mobile monitoring data in a variety of locations to develop hourly empirical models of particulate air pollution. This study uses secondary data describing BC and PN pollutant levels, which are obtained from roads that bikers share in the more rural location of Blacksburg (VA). Machine learning (ML) algorithms are then built to develop accurate and reliable short-term empirical prediction models. Different pre-processing methods for the mobile monitoring data and various input variables are tested to assess how ML can be used effectively in this process. Three types of time-average models are developed (daytime, hourly average, and one second models). Various combinations of spatial and temporal input variables are used in the short-term models. The impact of adding more spatiotemporal variables (e.g., emissions) to machine learning models to improve model performance is assessed in the short-term models. Incorporating spatial and temporal autocorrelation is intended to develop more sophisticated validation approaches for identifying ML performance patterns—the goal of which is to predict concentration levels more accurately in comparison to using raw data without data reprocessing. The results show that the model developed using refined disaggregated data is able to detect the spatial distribution of the pollutant concentration at equivalent levels as the smoothed data models, although the latter display fewer errors. The performance of the short-term model including all variables is equivalent to the model omitting emissions. The ML results are compared to earlier stepwise regression model results, suggesting that ML has the ability to improve both long-term and short-term model accuracy. Our findings indicate that ML demonstrates higher predictive capacity in comparison to stepwise regression. The results from this study may be useful in enhancing the performance of ML through the incorporation of different data preprocessing tasks, as well as showing how different input variables contribute to the ML modeling process. The findings from this study could be used toward the development of environmental/eco-friendly routes that would decrease the risk for exposure to harmful vehicle-related emissions. / Doctor of Philosophy / Air pollution is a major environmental threat to human health, claiming the lives of millions of people each year, primarily as a result of fine particulate matter entering the respiratory system. As such, it is important to develop reliable and accurate air pollution modeling approaches in order to protect the public and the environment from toxic contaminants and pathogens in the air. Although an approach known as land use regression has long been used to assess exposure to air pollution, researchers are increasingly using machine learning (ML) algorithms to quantify the concentration of harmful pollutants—for this study black carbon and particle number, which is a generic assessment that captures a number of known airborne hazards. Additionally, researchers are moving away from using fixed-site data in favor of using mobile monitoring data in a variety of locations to develop hourly empirical models of particulate air pollution. In this study, machine learning algorithms are developed using secondary data collected from roads that bikers share, which are representative of pollution levels of particle number and black carbon in the more rural location of Blacksburg (VA), in order to develop accurate and reliable short-term empirical prediction models. Different pre-processing methods of the mobile monitoring data and various input variables are tested to assess how machine learning can be efficiently used in this process. Our findings indicate that machine learning demonstrates higher predictive capacity in comparison to stepwise regression. The results from this study are expected to be useful in enhancing the performance of machine learning through the incorporation of different data preprocessing tasks, as well as how different input variables contribute to the machine learning modeling process. The findings from this study could assist transportation planners and other stakeholders better assess pollution risks for bike riders and pedestrians. As such, this study's findings could be used toward the development of environmental/eco-friendly routes that would decrease the risk for exposure to harmful vehicle-related emissions.

Air pollution is associated with various adverse health impacts and is identified as one of the leading risk factors for global disease burden. Further, air pollution is one of the pathways through which climate change could negatively impact health. Field studies have shown that air pollution has high spatiotemporal variability and pollutant concentrations vary substantially within neighborhoods. Characterizing air pollution at a fine-grained level is essential for accurately estimating human exposure, assessing its impact to human health, and further aiding localized air pollution policy. Air quality models are developed to estimate air pollution at locations and time periods without monitors, and these estimates are commonly used for exposure and health effects studies. Traditional land use regression [LUR] models are one of the cost-effective empirical air quality models. LUR typically relies on fixed-site measurements, GIS-derived variables with limited spatial resolution, and captures linear relationships. In recent years, innovative open-source imagery datasets and their associated features (e.g., street view imagery, remote sensing imagery) have emerged and show potential to augment or replace traditional LUR predictors. Such imagery data sources embody abundant information of natural and built environment features. Advanced computer vision techniques enable feature extraction and quantification through these extensive imagery datasets. The overarching objective of this dissertation is to investigate the feasibility of leveraging open-source imagery datasets (i.e., Google Street View [GSV] imagery, Landsat imagery, etc.) and advanced machine learning algorithms to develop image-based empirical air quality models at both local and national scale. The first study of this work established a pipeline of feature extraction through street view imagery sematic segmentation. The resulting street view features were used to predict street-level particulate air pollution for a single city. The results showed that solely using GSV-derived features can achieve comparable model fits as using traditional GIS-derived variables. Feature engineering improved model stability and interpretability through reducing spurious variables from potential misclassifications from computer vision algorithms. The second study further developed GSV-based models at national scale across multiple years. Random forest models were developed to capture the nonlinear relationship between air pollution and its impacting factors. The results showed that with sufficient street view images, GSV imagery alone may explain the variation of long-term national NO2 concentrations. Adding satellite-derived aerosol estimates (i.e., OMI column density) can significantly boost model performance when GSV images are insufficient, but the addition narrows when more GSV images are available. Our systematic assessment of the impact of image availability on model performance suggested that a parsimonious image sampling strategy (i.e., one GSV image per 100m grid) may be sufficient and most cost-effective for model development and application. Our third study explored the feasibility of combining street view and remote sensing derived features for national NO2 and PM2.5 modeling and projection at high spatial resolution. We found that GSV-based models captured both the highest and lowest pollutant concentrations while remote sensing features tended to smooth the air pollution variations. The results suggested that GSV features may have the capability to better capture fine-scale air pollution variability. The resulting air pollution prediction product may serve a variety of applications, including providing new insights into environmental justice and epidemiological studies due to its high spatial resolution (i.e., street level). Collectively, the result of this dissertation suggests that GSV imagery, processed with computer vision techniques, is a promising data source to develop empirical air quality models with high spatial resolution and consistent predictor variables processing protocol. Image-based features assisted with advanced ML approaches have the potential to greatly improve air quality modeling estimates, and successfully show comparable and even superior model performance than other modeling studies. Moreover, the ever-growing public imagery data sources are particularly promising for remote or less developed areas where traditional curated geodatabases are sparse or nonexistent. / Doctor of Philosophy / Air pollution is detrimental to human health and well-being. Further, air pollutants concentrations can change rapidly within a short distance and temporal frame. Monitoring air pollution with high spatial-temporal resolution is important. Traditional air quality monitoring networks are expensive and sparsely distributed, leading to gaps in capturing the air pollution at small spatial scales. Air quality models are developed to estimate air pollution at locations and time periods without monitors. Empirical air quality models often use air measurements from stationary sites and GIS-derived features (e.g., traffic, population density, land use types, etc.) to develop regression models and use the regression formula to estimate air pollutant concentrations in unmonitored areas. However, GIS-derived features are often collected from curated GIS databases, which often have coarse resolution when available across large geography. Street view imagery and remote sensing imagery contains rich information of natural and built environments. Computer vision techniques can be applied to extract such information to replace or augment traditional GIS-derived features. Combined with advanced machine learning algorithms, features derived from open-access images are promising to develop air quality models with a consistent image collection and processing protocol. This dissertation examines the feasibility of using street view imagery (i.e., Google Street View [GSV] Imagery) and remote sensing imagery to develop air quality models at both local and national scales. Our results found that solely using GSV features to build local and national models can achieve good model performance, which is consistent or even better than other models using traditional GIS-derived variables. For areas without sufficient GSV images, adding satellite observations for air pollution can significantly enhance model performance. Remote sensing features tend to smooth air pollution variation while GSV features tend to better capture fine-scale intra-urban air pollution variation. In conclusion, leveraging open-source imagery datasets with advanced machine learning methods are promising for estimating air pollution at high spatial resolution with good model fits.

Outdoor air pollution is a major contributor to adverse health effects of citizens, in particular those living in urban environments. Air quality monitoring networks are set up to measure air quality in different environments in compliance with national and European legislation. Generally, only a few fixed monitoring sites are located within a city and thus cannot represent air pollutant concentrations in urban areas accurately enough to allow for a detailed human exposure assessment. Other approaches to derive detailed urban air pollutant concentration estimates exist, such as dispersion models and land-use regression (LUR) models. Low-cost portable air quality monitors are also emerging, which have the potential to add value to existing monitoring networks by providing measurements at greater spatial resolution and also to provide individual-level exposure assessment. The aim of this thesis is to demonstrate how measurements and modelling in combination allow detailed investigations of the variability of air pollutants in space and time in urban area, and in turn improve on the current exposure assessment methods. Three types of low-cost portable monitors measuring NO2, O3 (Aeroqual monitors) and PM2.5 (microPEM monitor) were evaluated against their respective reference instruments. The Aeroqual O3 monitor showed very good correlation (r2 > 0.9) with the respective reference instruments, but biases in the slope and intercept coefficients indicated that calibration of Aeroqual O3 monitor was needed. The Aeroqual NO2 monitor was subject to cross-sensitivity from O3, which, as demonstrated, can be effectively corrected by making O3 and NO2 measurements in tandem. Correlation between the microPEM monitor and its reference instrument was poor (r2 < 0.1) when PM2.5 concentrations were low (< 10 μg m-3), but significantly improved (r2 > 0.69) during periods with elevated PM2.5 concentrations. Relative humidity was not found to affect the raw results of PM2.5 measurements in a consistent manner. All three types of monitors cannot be used as equivalent or indicative methods instead of reference methods in studies that require quantification of absolute pollutant concentrations. However, the generally good correlations with reference instruments reassure their application in studies of relative trends of air pollution. Concentrations of PM2.5, ultrafine particles (UFP) and black carbon (BC) were quantified using portable monitors through a combination of mobile and static measurements in the city of Edinburgh, UK. The spatial variability of UFP and BC was large, of similar magnitude and about 3 times higher than the spatial variability of PM2.5. Elevated concentrations of UFP and BC were observed along streets with high traffic volumes whereas PM2.5 showed less variation between streets and a footpath without road traffic. Both BC and UFP significantly correlated with traffic counts, while no significant correlation between PM2.5 and traffic counts was observed. The relationships between UFP, NO2 and inorganic components of PM2.5 were further investigated through long-term measurements at roadside, urban background and rural sites. UFP moderately correlated with NOx (NO2 + NO) and showed varying relationships with NOx depending on the particle size distribution. Principal component analysis and air-mass back trajectory analysis revealed that PM2.5 concentrations were dominated by long-range transport of secondary inorganic aerosols, whereas UFP were mainly related to varying local emissions and meteorological conditions. These findings imply the need for different policies for managing human exposure to these different particle components: control of much BC and UFP appears to be manageable at local scale by restricting traffic emissions; however, abatement of PM2.5 requires a more strategic approach, in cooperation with other regions and countries on emissions control to curb long-range transport of PM2.5 precursors. A dispersion model (ADMS-Urban) was used to simulate high resolution NO2 and O3 concentrations in Edinburgh. The effects of different emission and meteorological input datasets on the resulting modelled NO2 concentrations were investigated. The modelled NO2 and O3 concentrations using the optimal model setup were validated against reference instrument and diffusion tube measurements. Temporal variability of NO2 was predicted well at locations that were not heavily influenced by local effects, such as road junctions and bus stops. Temporal variability of O3 was predicted better than for NO2. Long-term spatial variability of NO2 was found to correlate well with diffusion tube measurements, while modelled spatial variability of O3 in ADMS-Urban compared poorly with diffusion tube measurements. However, it was found that the O3 diffusion tube measurements may be subject to some unidentified biases affecting their accuracy. Land-use regression (LUR) models are widely used to estimate exposure to air pollution in urban areas. An appropriately sized and designed monitoring network is an important component for the development of a robust LUR model. Concentrations of NO2 were simulated by ADMS-Urban at ‘virtual’ monitoring sites in 54 different network designs of varying numbers and types of site, using a 25 km2 area including much of the Edinburgh city area. Separate LUR models were developed for each network. These LUR models were then used to estimate ambient NO2 concentrations at all residential addresses, which were evaluated against the ADMS-Urban modelled concentration at these addresses. The improvement in predictive capability of the LUR models was insignificant above ~30 monitoring sites, although more sites tended to yield more precise LUR models. Monitoring networks containing sites located within highly populated areas better estimated NO2 concentrations across all residential locations. LUR models constructed from networks containing more roadside sites better characterised the high end of residential NO2 concentrations but had increased errors when considering the whole range of concentrations. No particular composition of monitoring network resulted in good estimation simultaneously across all residential NO2 concentration and of the highest NO2 levels implying a lack of spatial contrast in LUR-modelled pollution surface compared with the dispersion model. Finally, the results from the measurement and modelling studies presented in thesis are synthesised in the context of current exposure assessment studies. Low-cost air-quality monitors currently do not possess and are unlikely in the near future to provide the robustness and accuracy to replace the existing routine monitoring network. Development of the low-cost air-quality should be aiming at upgrading them as the indicative method as defined in the data quality objective in the EU directive. The monitoring sites used to build LUR models should capture well the population distribution in the study area as opposed to capturing the greatest pollution contrast. The traditional methods of evaluating LUR models are also ineffective in characterising the models’ capability at estimating pollutant concentration at residential address. Given that the dispersion models are also subject to the availability and uncertainties in the input data, future air quality model development should endeavour to incorporate both dispersion and land-use regression models, where the uncertainty in the input data can be reduced by using LUR models built on actual measurements, and the limitation in the statistical modelling can be replaced by adopting the deterministic approach used in the dispersion model.