Comparison of regression models with land-use and emissions data to predict the spatial distribution of traffic-related air pollution in Rome

Spatial modeling of traffic-related air pollution typically involves either regression modeling of land-use and traffic data or dispersion modeling of emissions data, but little is known to what extent land-use regression models might be improved by incorporating emissions data. The aim of this study was to develop a land-use regression model to predict nitrogen dioxide (NO2) concentrations and compare its performance with a model including emissions data. The association between each land-use variable and NO2 concentrations at 68 locations in Rome in 1995 and 1996 was assessed by univariate linear regression and a multiple linear regression model that was constructed based on the importance of each variable. Traffic emissions (particulate matter, carbon monoxide, nitrogen oxides, and benzene) were estimated for 164 areas of the city based on vehicle type, traffic counts and driving patterns. Mean NO2 concentration across the 68 sites was 46.8 microg/m3 (SD 9.8 microg/m3; inter-quartile range 11.5 microg/m3; min 24 microg/m3; max 73 microg/m3). The most important predicting variables were the circular traffic zones (main ring road, green strip, inner ring road, traffic-limited zone), distance from busy streets, size of the census block, the inverse population density, and altitude. A multiple regression model including these variables resulted in an R2 of 0.686. The best-fitting model adding an emission term of benzene resulted in an R2 of 0.690, but was not significantly different from the model without emissions (P=0.147). In conclusion, these results suggest that a land-use regression model explains the traffic-related air pollution levels with reasonable accuracy and that emissions data do not significantly improve the model.