Modelling NO2 Exposure in Greater Glasgow Jonathan Gillespie [email protected] University of Strathclyde Aims • To develop a land use regression model for nitrogen dioxide in Greater Glasgow • To apply this model to estimate residential exposure in a cohort group What is Land Use Regression Modelling? Statistical approach based on nearby features and measured concentrations Very low cost Good spatial prediction Applicable to multiple pollutants/ scales Easy to apply and understand No emission/met. data required × Sampling prior to modelling × Network design crucial × May be temporally/spatially limited × No physical basis × Subjective Extensively used in epidemiological studies Building and applying a Land Use Regression Model 1. Design monitoring network . 5 Local Authority monitoring networks 2. Measure pollutant . 181 monitoring locations Monitoring Site Locations ¯ Monitoring Sites 0 1 2 4 Miles Building and applying a Land Use Regression Model 1. Design monitoring network . 5 Local Authority monitoring networks 2. Measure pollutant . 181 monitoring locations 3. Calculate proximal features in GIS Building and applying a Land Use Regression Model 1. Design monitoring network . 5 Local Authority monitoring networks 2. Measure pollutant . 181 monitoring locations 3. Calculate proximal features in GIS . Length of roads . Traffic (HGV/Total Traffic) . Influence of Buildings . Population . Altitude . Land use class Building and applying a Land Use Regression Model 1. Design monitoring network . 5 Local Authority monitoring networks 2. Measure pollutant . 181 monitoring locations 3. Calculate proximal features in GIS 4. Perform linear regression Evaluation of Model on Hold-Out Data Model R2 0.73 Validation R2 0.82 Y = 0.82x + 8.1 Building and applying a Land Use Regression Model 1. Design monitoring network . 5 Local Authority monitoring networks 2. Measure pollutant . 181 monitoring locations 3. Calculate proximal features in GIS 4. Perform linear regression 5. Create pollution surface in GIS Modelled NO2 Concentration in Greater Glasgow Legend Modelled NO2 (ug/m3) 9.7 - 20 ¯ 20.1 - 30 30.1 - 40 40.1 - 50 50.1 - 60 601 - 70 70.1 - 80 80.1 - 90 0 1 2 4 Miles Building and applying a Land Use Regression Model 1. Design monitoring network 2. Measure pollutant 3. Calculate proximate features in GIS 4. Perform linear regression 5. Create pollution surface . Participant locations from Glasgow blood 6. Geocode participant locations and pressure clinic estimate exposure . Health data available for follow up study Patient Distribution Legend Patient Location Modelled NO2 (ug/m3) 9.7 - 20 ¯ 20.1 - 30 30.1 - 40 40.1 - 50 50.1 - 60 601 - 70 70.1 - 80 80.1 - 90 0 1 2 4 Miles Exposure Estimation for Cohort Exposure Statistics Mean 26.6 Median 24.5 Minimum 9.7 Maximum 78.8 Percent > 40 µg/m3 7.5 Count 13679 Model Limitations • Networks designed and run by 5 local authorities • Sampling bias from preferential sampling – 25% background sites? • Accuracy of GPS coordinates? • Consistency of measurements • Only annual average for single year • Extrapolation is possible • Exposure assessment limited to outdoor exposure at the home • Personal exposure may differ Summary • Developed a simple LUR model for Greater Glasgow based on Local Authority data • Highest concentrations were found in City Centre and localised hotspots -3 • ~7 % of participants exposed to > 40 µg m NO2 • Participant health statistics are available for follow up work Acknowledgments • Dr Iain Beverland (University of Strathclyde) • Dr Scott Hamilton (Ricardo-AEA) • Professor Sandosh Padmanabhan (University of Glasgow) .