Space and time modelling of intra-urban air pollution

• Main Author: Tang, Ho Kin Robert
• Other Authors: Gulliver, John ; Blangiardo, Marta
• Published: Imperial College London 2014
• Subjects: 614
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.676783

Exposures to air pollution have adverse effects on health. Traditionally, epidemiological studies used monitoring data to investigate the relationship between air pollution and health. In recent decades, modelling tools have been developed to predict pollutant concentrations for population exposure assessments. Whilst gradual improvements have been made to these techniques, such as dispersion and land use regression (LUR), results have exhibited spatial inconsistencies at times. The processes involved are often time- and data- consuming, and outputs generally do not account for short-term variations in pollution. Improving model prediction capabilities can avoid exposure misclassifications, and provide better estimates for health risk assessment. The aim of this project is to increase the accuracy and efficiency of current exposure modelling techniques to capture spatial and temporal variability of urban air pollution. As part of this study, air pollution models were developed in a GIS framework for London for PM10, NOX and NO2, using dispersion, LUR, hybrid and Bayesian statistical methods. Predictors derived from traffic, land use, population datasets were incorporated in a geographical information system for modelling. For the first time, newly available city-wide datasets were used to extract enhanced geographical variables, including building height/ area, street canyon and detailed urban green space, which may have significant influence on pollution in local dispersion environment. Developed models were cross-validated and compared to concentrations obtained from routine monitoring network. LUR models were found to have higher prediction capabilities over other techniques, providing accurate explanations of spatial variability in urban air pollution. Significant improvements in model performance were seen with addition of buildings and street configuration variables, particularly for traffic-related pollutants. LUR require less computational demands than conventional dispersion methods; therefore can be easily applied over large urban areas. Introducing Bayesian statistical techniques has enabled spatio-temporal predictions which accounted uncertainties, allowing detection of pollution trends and episodes.