Supplementary material
Yüklə 262,86 Kb.
|
- Bu səhifədəki naviqasiya:
- 1.4 Monitoring sites Table S3
- State/Territory Population (2011) Area (km 2 ) n
- TOTAL 21,504,688 7,687,591 68 0.9
- Variable Centre Standardise
- 2. RESULTS 2.1 Measured NO 2 concentrations
- Figures S1 and S2
|
NOTES: a average daily minimum and maximum temperature were also included in the model b major roads were defined as national/state highways, arterial roads (which are major connector roads for national and state highways) and sub-arterial roads (which are connectors between highways and/or arterial roads, or serve as an alternative for arterial roads) (PSMA, 2013) c minor roads were defined as collector roads (which are connectors between sub-arterial roads, and distribute traffic to local roads) and local roads (which provide property access) (PSMA, 2013) d four land use categories were examined – residential, commercial, industrial, and open space (which was the sum of water, parks and agricultural land [Rose et al., 2011]) e total (fugitive + non-fugitive) estimated NOX emissions from the 1,857 industrial and commercial sites around Australia f the average density (sites/km2) of industrial and commercial sites emitting NOX in each buffer was also included in the model g there were few airports in the majority of buffers so they were defined as either 0 (not present) or 1 (present) (Hystad et al., 2011) h a mesh block is the smallest spatial unit used in the Australian census and their size varies - on average they contain 62 people i average of variable within buffer j sum of variable within buffer * 22 circular buffers were created with radii of 100 m, 200 m, 300 m, 400 m, 500 m, 600 m, 700 m, 800 m, 1000 m, 1200 m, 1500 m, 1800 m, 2000 m, 2500 m, 3000 m, 3500 m, 4000 m, 5000 m, 6000 m, 7000 m, 8000 m, and 10,000 m (Novotny et al., 2011). ** Positional accuracy is ±2 m in urban areas, ±10 m in rural and remote areas. Attribute accuracy is 99.09% for key attributes (name and unique identifier) (PSMA, 2013). 1.4 Monitoring sites Table S3. Location and number of monitoring sites in the 8 states and territories of Australia during the study period (2006-2011).
1.5 Model building Variable processing To improve model convergence and make the parameter estimates more interpretable, we centred and standardised some of the independent variables using the values shown below (Table S4). So, for example, the parameter estimates for elevation will be for a 100 metre increase, and the intercept of the model will be for an elevation of 20 metres. Table S4. Values used to centre and standardise variables.
* NPI NOX refers to the total emissions (kg/yr) from point source sites within each buffer. The median and IQR were used for each buffer (from 100 m to 10,000 m). Model validation We used cross-validation rather than leave-one-out validation because some important predictors were only present at single sites, and hence the model failed to converge when these sites were removed. We examined the importance of these sites using the df-beta statistics. We did not examine the spatial autocorrelation of the residuals because we used a generalised estimating equation (GEE) model (see discussion section of paper), which means that the residuals include any remaining differences between monitors. This could have been avoided using a mixed model with a random station intercept, which would have given similar predictions and parameter estimates. However, we found that the GEE models had a better convergence and were easier to use in the cross-validation. Another advantage is that the GEEs give R2 statistics that are based only on the predictors, whereas mixed models use the predictors and the random intercepts and hence would have given overly optimistic R2 statistics for our models. 2. RESULTS 2.1 Measured NO2 concentrations Figures S1 and S2 present annual and monthly average NO2 concentrations measured at the monitoring sites, respectively. 
Figures S1 and S2. Time series of annual (top) monthly (bottom) NO2 averages by monitoring site. 2.2 Results of variable selection process Table S5 shows the results of the initial lasso variable selection process for the annual column model for the years 2006 to 2011 (1=Selected, 0=Not selected). Variables are ordered by selection frequency. Variables selected at least once are shown (26 variables), as only these were used in the second stage of variable selection (see methods section of main paper). Tables S6 shows the results of the lasso process for the annual surface model. Tables are not shown for the monthly models due to the large number of variables. Table S5. Results of lasso variable selection for the annual column model.
Table S6. Results of lasso variable selection for the annual surface model.
Yüklə 262,86 Kb. Dostları ilə paylaş:
|