05.05.2015 Views

Download REHIS Journal 21/3 (Autumn 2009) - The Royal ...

Download REHIS Journal 21/3 (Autumn 2009) - The Royal ...

Download REHIS Journal 21/3 (Autumn 2009) - The Royal ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Environmental Health Scotland<br />

Figure 5. Predicted NO 2<br />

concentrations in Glasgow, 1998-99 (μg/m 3 ).<br />

Other useful predictor variables included postcodes<br />

within 300m and junctions within 250m. Figure 5<br />

shows the surface of predicted pollution concentrations<br />

for NO 2<br />

in Glasgow, 1998-99.<br />

<strong>The</strong> variables that predicted the highest amount of<br />

variation in pollution concentrations for the BS model<br />

were the road length within 250m of the monitoring<br />

site, the distance to the edge of the urban area and the<br />

altitude (table 2). Other useful predictor variables<br />

were the road length within 50m, junctions within<br />

250m and households within 300m. Figure 6 shows<br />

the surface of predicted pollution concentrations for<br />

BS in Central Scotland, 1970-79.<br />

<strong>The</strong> best model predicted 38% of the variation in<br />

BS concentrations, which means that the BS model<br />

was not very reliable. <strong>The</strong> main problems with this<br />

model were that contemporary road and urban edge<br />

data were used to predict historical BS concentrations<br />

and that the black smoke measurements from the<br />

1970s were incomplete with many extended periods<br />

of missing data. Using contemporary surrogate<br />

emission and advection data was not ideal, but<br />

historical road and urban area data were not available.<br />

Statisticians working on an epidemiological study<br />

on long-term black smoke exposure and mortality<br />

(Beverland et al, 2007) predicted the missing<br />

pollution data using sophisticated imputation methods<br />

(Yap, 2005), which improved the reliability of<br />

the pollution concentrations. However, bivariate<br />

correlations between the BS measurements and each<br />

predictor variable were weak, which in turn led to a<br />

model with poor predictive power.<br />

Ideal conditions for using land use regression modelling<br />

would be when good quality data for pollution<br />

concentrations and predictor variables are measured<br />

during the same time period.<br />

Table 2. Multiple regression model for BS in Central Scotland, 1970-79.<br />

8

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!