Estimating Air Pollution with Regulatory Monitors and Supplementary Data from Low-cost Sensors and Remotely Sensed Imagery

This thesis has two main objectives: assessing the accuracy of long-term air pollution concentration estimates produced from portable monitoring campaigns, using temporal adjustments; and second, developing spatial models of air pollution concentrations with remotely sensed imagery and machine learning within a land use regression framework. Objective one was achieved by simulating mobile monitoring campaigns and estimating long-term concentrations with multiple temporal adjustments. The results indicated that sample size greatly influenced the accuracy of long-term estimates produced with temporal adjustments and that adjustments which accounted for the log-normal distribution of air pollution observations produced more accurate estimates. Objective two was achieved through the development of annual and seasonal models for NO2 and PM2.5 at three study sites. The results showed that deep learning models were the most accurate however these models could not capture spatial patterns of air pollution from the spectral signature of the land use surrounding monitors.M.Sc