While looking into the modelling framework related to areal level data, the first
statistical challenge regards the spatial misalignment between health and exposure
data. Here the consequence that pollutant concentration has to be scaled up at the
area level, while being measured at finite number of point-referenced monitoring
stations which is a procedure known as change of support (Gelfand et al. 2010). A
simple solution would be to compute the average of the point measurements
collected for each area, possibly using distance or population-based weights (Elliott
et al. 2007; Madsen et al. 2008; Young et al. 2009). However, this approach may not
be feasible when the monitoring stations network is sparse, that is, in areas without
monitoring stations, or when there is a strong spatial heterogeneity of the considered
pollutants.
Therefore, estimation of the area-level concentrations through weighted averages
over grid level concentrations is possible by using deterministic atmospheric dispersion models (Bell 2006; Rushworth et al. 2014; Lee and Sarran 2015). Another
alternative approach would be model-based perspective, where a spatial statistical
model can be built which combines pollutant measurements from stations with the
output of numerical dispersion models in a data fusion approach (Fuentes et al. 2006;
Peng and Bell 2010; Sahu et al. 2010; Berrocal et al. 2010; Pannullo et al. 2015;
Moraga et al. 2017). Under this modelling framework, other than accounting for
spatial correlation and measurement error, additional covariates like meteorological
variables can also be included.
The authors of this paper conducted a study where the satellite aerosol concentrations are linked to lung cancer incidences using spatial regression models. The
lung cancer incident data was aggregated and sorted for each taluk and the age
standardized incidence rates were calculated. The remotely sensed grid data of the
PM 2.5 concentrations was acquired from the Socioeconomic Data and Applications
Centre (SEDAC) of NASA. The Global Annual PM 2.5 grid data sets from MODIS,
MISR and SeaWiFS Aerosol Optical Depth (AOD) represent a series of three-year
running mean grids (1998–2012) of fine particulate matter (solid particulate and
liquid droplets) derived from a combination of the above satellite retrievals.
Together, the grids provide a continuous surface of PM 2.5 concentrations per
cubic metre, that is, particulate matter 2.5 micrometres or smaller, for health and
environmental research. A global decadal (2001–2010) mean PM 2.5 concentration
grid was also produced. The raster grids have a cell resolution of 6 arc-minutes (0.1
degree or approximately 10 km at the equator) and cover the global land surface. The
decadal PM 2.5 concentration grids were used for the present study. Using zonal
statistics, the mean PM 2.5 value for each taluk was computed and utilized for further
analysis. In addition, the age standardized tobacco and bidi smoking prevalence rates
were incorporated to the present assessment and the data was acquired from the
study conducted by Fu et al. (2014) (Tables 18.1, 18.2, 18.3 and Fig. 18.1).
The statistically significant positive association between standardized lung cancer
incidence rates and aerosol particle pollution can be taken as indicator of a potential
air pollution effect. Additionally, the MODIS aerosol data lends itself to populationbased exposure assessment and the ecological approach. Seamless remote sensing
images represent aggregate exposure better than averaging ground measurements
18 Geospatial Modelling of Air Pollution and its Impact on Health of Urban. . .
383
statistical challenge regards the spatial misalignment between health and exposure
data. Here the consequence that pollutant concentration has to be scaled up at the
area level, while being measured at finite number of point-referenced monitoring
stations which is a procedure known as change of support (Gelfand et al. 2010). A
simple solution would be to compute the average of the point measurements
collected for each area, possibly using distance or population-based weights (Elliott
et al. 2007; Madsen et al. 2008; Young et al. 2009). However, this approach may not
be feasible when the monitoring stations network is sparse, that is, in areas without
monitoring stations, or when there is a strong spatial heterogeneity of the considered
pollutants.
Therefore, estimation of the area-level concentrations through weighted averages
over grid level concentrations is possible by using deterministic atmospheric dispersion models (Bell 2006; Rushworth et al. 2014; Lee and Sarran 2015). Another
alternative approach would be model-based perspective, where a spatial statistical
model can be built which combines pollutant measurements from stations with the
output of numerical dispersion models in a data fusion approach (Fuentes et al. 2006;
Peng and Bell 2010; Sahu et al. 2010; Berrocal et al. 2010; Pannullo et al. 2015;
Moraga et al. 2017). Under this modelling framework, other than accounting for
spatial correlation and measurement error, additional covariates like meteorological
variables can also be included.
The authors of this paper conducted a study where the satellite aerosol concentrations are linked to lung cancer incidences using spatial regression models. The
lung cancer incident data was aggregated and sorted for each taluk and the age
standardized incidence rates were calculated. The remotely sensed grid data of the
PM 2.5 concentrations was acquired from the Socioeconomic Data and Applications
Centre (SEDAC) of NASA. The Global Annual PM 2.5 grid data sets from MODIS,
MISR and SeaWiFS Aerosol Optical Depth (AOD) represent a series of three-year
running mean grids (1998–2012) of fine particulate matter (solid particulate and
liquid droplets) derived from a combination of the above satellite retrievals.
Together, the grids provide a continuous surface of PM 2.5 concentrations per
cubic metre, that is, particulate matter 2.5 micrometres or smaller, for health and
environmental research. A global decadal (2001–2010) mean PM 2.5 concentration
grid was also produced. The raster grids have a cell resolution of 6 arc-minutes (0.1
degree or approximately 10 km at the equator) and cover the global land surface. The
decadal PM 2.5 concentration grids were used for the present study. Using zonal
statistics, the mean PM 2.5 value for each taluk was computed and utilized for further
analysis. In addition, the age standardized tobacco and bidi smoking prevalence rates
were incorporated to the present assessment and the data was acquired from the
study conducted by Fu et al. (2014) (Tables 18.1, 18.2, 18.3 and Fig. 18.1).
The statistically significant positive association between standardized lung cancer
incidence rates and aerosol particle pollution can be taken as indicator of a potential
air pollution effect. Additionally, the MODIS aerosol data lends itself to populationbased exposure assessment and the ecological approach. Seamless remote sensing
images represent aggregate exposure better than averaging ground measurements
18 Geospatial Modelling of Air Pollution and its Impact on Health of Urban. . .
383