Development of metrics for individual exposure assessment to traffic related air pollution
Woolcock Institute of Medical Research, June 2008
- Development of metrics for individual exposure assessment to traffic related air pollution (PDF - 3,735 KB) | (RTF - 19,399 KB)
This project has explored the use of a variety of methods for assigning exposure to traffic related air pollution. The work has been conducted simultaneously with the collection of health outcome data for an epidemiological study investigating the respiratory and irritant effects of changes in traffic related air pollution. Epidemiological studies investigating the effects of environmental exposures often suffer from poor or inadequate exposure assessment. This CARP project has provided an ideal opportunity to test a number of methods for assigning values for exposure to traffic related air pollution to households in four discrete geographic locations near a roadway tunnel in northern Sydney. We used nitrogen dioxide as a marker of exposure to traffic related air pollution. The methods we developed and tested for estimating the spatial distribution of NO2 included: land use regression modelling based on data from dispersed passive samplers; interpolation based on data from passive samplers; dispersion modelling; proximity measures (eg distance to main road); fixed site monitor data; and, personal monitoring.
We found that of the models tested, the land use regression models were the most reliable and agreed most accurately with the actual measured NO2 levels. The three LUR models were able to explain a high degree of the variability in NO2 ranging from 70-85%, and were also found to reliably predict NO2, with agreement with the measured passive NO2 samplers ranging from 87-95%. The most complex LUR model we developed incorporated a temporal component using data obtained from four fixed site monitors in the area, and used mixed model analytical methods. This mixed model LUR demonstrated the greatest ability to predict concentrations of NO2 for the study area. The predictor variables for the LUR models were consistent with those reported internationally and included traffic density, population density, commercial land use, and for the mixed model, data from the fixed site monitors.
Of the other methods, the dispersion models (LWM, TAPM and TAPM-interpolated) were found to perform reasonably well when compared to the fixed site monitors, but were found to have poorer agreement with the passive NO2 sampler readings.
We found that the interpolation of data from passive samplers using kriging was unreliable for these data, probably due to the lack of spatial autocorrelation in the data given the small-scale urban environment. The proximity technique, while readily available and easily implemented did not perform as well as the LUR model. This was expected because the LUR incorporates additional explanatory factors and a more sophisticated measure of traffic exposure than that used in the proximity model.
Personal sampling of NO2 was found to be useful for describing the range of NO2 exposure levels for subjects and was found to correlate reasonably well with household NO2 for adults but less so for children. On average personal NO2 was about 2ppb lower than for household ambient NO2. After adjusting for outdoor household NO2, personal levels were related to indoor sources of NO2 (cooking with gas stoves and ovens) supporting the validity of these measures.