A comparison of different approaches to estimate small-scale spatial variation in outdoor NO₂ concentrations

Background In epidemiological studies, small-scale spatial variation in air quality is estimated using land-use regression (LUR) and dispersion models. An important issue of exposure modeling is the predictive performance of the model at unmeasured locations. Objective In this study, we aimed to evaluate the performance of two LUR models (large area and city specific) and a dispersion model in estimating small-scale variations in nitrogen dioxide (NO2) concentrations. Methods Two LUR models were developed based on independent NO2 monitoring campaigns performed in Amsterdam and in a larger area including Amsterdam, the Netherlands, in 2006 and 2007, respectively. The measurement data of the other campaign were used to evaluate each model. Predictions from both LUR models and the calculation of air pollution from road traffic (CAR) dispersion model were compared against NO2 measurements obtained from Amsterdam. Results and conclusion The large-area and the city-specific LUR models provided good predictions of NO2 concentrations [percentage of explained variation (R2) = 87% and 72%, respectively]. The models explained less variability of the concentrations in the other sampling campaign, probably related to differences in site selection, and illustrated the need to select sampling sites representative of the locations to which the model will be applied. More complete traffic information contributed more to a better model fit than did detailed land-use data. Dispersion-model estimates for NO2 concentrations were within the range of both LUR estimates.