Understanding the spatial variation of air pollution and its impacts through Geographically Weighted Regression (GWR)

We explore the potential correlation between income and exposure to air pollution for the city of Madrid, Spain, and its neighboring municipalities. Statistical analyses were carried out using electoral district level data on gross household income, and NO₂ and PM_2.5 concentrations in air obtained from a mesoscale air quality model for the study area. Household income data were summarized at the grid level through a zonal statistics operation, carried out at four different cell resolutions for both 1 x 1 km and 2 x 2 km grids in order to account for modifiable aerial units and data conversion uncertainty. Our findings point to a clear negative correlation between level of household income and exposure to both pollutants, which was clearly present at all resolutions and both grid sizes, though with a high degree of variability depending on the resolution and grid size chosen. The strongest association between income and air pollution was found for minimum gross household income (MGHI) and NO_2, and MGHI and PM_2.5. The global regression models explained between 10% and 20% of the variance for MGHI and NO₂, and between 12% and 19% of the variance for MGHI and PM_2.5, depending on resolution and grid size chosen. Standard residual error varied between 0.55-0.58 for MGHI and NO₂ and between 0.28-0.30 for MGHI and PM_2.5. To address the high degree of variability in the models we explored the spatial heterogeneity of the correlation effect, finding stronger decrease in contamination exposure as minimum rent increases in the north of the city of Madrid and the municipalities of Tres Cantos and Colmenar Viejo. In the centre and south of the metropolitan area, the slope of the regression line is shallower. This may be partly due to the fact that contaminant concentrations in the centre of the study area are uniformly higher than elsewhere, offering less opportunity to reduce exposure within this area. In the east of the metropolitan area no relationship between the variables was detected at the scale of the analysis. Our results suggest income-based inequality in exposure to air pollution in Madrid. They highlight the usefulness of electoral district level income data and simulated concentrations from Eulerian photochemical air quality models (a Community Multiscale Air Quality model – CMAQ, in this case) for understanding environmental inequality. Further work would be needed to explore the patterns observed at the district and neighborhood level.