A Semi-parametric Regression Model to Estimate Variability of NO2

Abstract

The purpose of this analysis was to derive a land-use regression (LUR) model using a semi-parametric method (based on penalized splines) to estimate the geographical characteristics that influence ambient concentrations of nitrogen dioxide (NO2) in Montreal, Quebec, Canada. Linear and semi-parametric multivariate regression analyses were conducted to assess the dependency between logarithms of concentrations of NO2 and land-use variables. In the published multiple linear regression analyses for this study, distance from the nearest highway, length of highways and major roads within 100 m, traffic count on the nearest highway, and population density showed significant associations with NO2. This work used a semi-parametric model with a nonparametric part incorporating the variables "area of open space within 100 m" and "length of minor roads within 500 m". These variables were non-significant in the linear regression model and showed nonlinear associations with the level of NO2. The semi-parametric model improves the fit of the model for land-use regression when comparing observed and predicted results.