Equivalence of using nested buffers and concentric adjacent rings as predictors in land use regression models

Land use regression (LUR) models have been widely used to estimate traffic air pollution within cities taking into account spatial distributions. These prediction models are often built based on multiple linear models that regress concentrations measured at locations spread over the cities on land use characteristics of those locations. Land use variables, as for example total road length, are sometimes calculated in buffers of different radii that are then included simultaneously in the models to account for a distance gradient in the influence of sources. Some authors favour including disjoint concentric ring areas instead. Both approaches have their advantages, and adjacent concentric rings tend to make model interpretation more intuitive. The present paper shows that both approaches lead to identical models and the results achieved from one can be used to calculate the results of the other when absolute continuous exposure measures such as total length of roads are used. It is also shown that this holds as well for relative exposures such as area-weighted total length of streets, inhabitants per area, and % urban area, where the measure of interest is scaled by a constant factor.