Operationalising Ambient Air Pollution Estimation

Abstract

Air pollution presents a significant health risk to individuals worldwide, with the estimation of ambient air pollution levels being an essential step in tackling the global health burden of air pollution due to the high cost of individual monitoring stations. This thesis outlines research done in fulfilment of a Ph.D. in Environmental Intelligence that looks at the use of machine learning to provide an alternative to traditional methods that fills a need for a scalable method of estimating ambient air pollution with varying spatial and temporal resolution to meet the demands of various stakeholders. Machine learning allows for estimations to be made at speed, allowing for deployment in an operational capacity to improve decision velocity and pivot interventions from reactive to proactive. The thesis presents how machine learning can estimate air pollution in the UK at the annual level, with a subsequent approach presented to estimate air pollution concentrations at the hourly and 1km2 temporal and spatial resolution in England. The merit of the approach is further explored with a global air pollution concentration model presented to address global inequality of air pollution intelligence, ensuring that no one is left behind when tackling the air pollution crisis. Finally, an open-source Python package and companion analysis of how the intelligence provided can help decision-making concerning decisions such as outdoor school activities timing, and clean air zones are presented, ensuring dissemination of the work to various stakeholders.