Comparison of the Performances of Land Use Regression Modeling and Dispersion Modeling for Estimating Intra-Urban Air Pollution Concentrations:

ISEE-0311 Background and Objective: Exposure assessment of intra-urban air pollution concentrations for epidemiological studies remains difficult. Current approaches include dispersion modeling and land use regression (LUR) modeling. There is however not much known about the comparability of the performances of both methods. We compared the performances of a LUR model and a dispersion model in a Dutch study area. Methods: For the Rijnmond area, i.e. Rotterdam and surroundings, NO2 concentrations for 2001 were estimated for 231,191 grid cells. The LUR model was developed using monitoring data from the national monitoring network and potential Geographic Information System (GIS) predictor variables. First, concentrations measured at regional background sites were interpolated and residual concentrations at urban background and traffic sites were calculated. LUR models were then developed to explain these residual concentrations. The URBIS model was used to estimate NO2 concentrations using dispersion modeling. Results: The LUR model for the residual concentration at urban and traffic sites included as predictor variables the area of residential land in a 300 m buffer, the traffic intensity in a 200 m buffer, the area of industry in a 1,000 m buffer, and the population density in a 100 m buffer, with R2 value of 88% and RMSE value of 3.7 μg/m3. Because of the large number of grid cells, the data were sorted and averaged over groups of 1,000 records. The correlation between estimates of both methods was 0.98. The average predicted concentration was however higher for the LUR estimates, especially for the lower concentrations. Differences in model assumptions and model development partly explained the differences in model predictions. Conclusion: LUR and dispersion modeling have their own advantages and limitations, but both were able to estimate intra-urban NO2 concentrations. Predictions were highly correlated, and could be used in epidemiological studies.