Application of nonparametric regression methods to study the relationship between NO 2 concentrations and local wind direction and speed at background sites

Abstract

Background concentrations of nitrogen dioxide (NO 2) are not constant but vary temporally and spatially. The current paper presents a powerful tool for the quantification of the effects of wind direction and wind speed on background NO 2 concentrations, particularly in cases where monitoring data are limited. In contrast to previous studies which applied similar methods to sites directly affected by local pollution sources, the current study focuses on background sites with the aim of improving methods for predicting background concentrations adopted in air quality modelling studies. The relationship between measured NO 2 concentration in air at three such sites in Ireland and locally measured wind direction has been quantified using nonparametric regression methods. The major aim was to analyse a method for quantifying the effects of local wind direction on background levels of NO 2 in Ireland. The method was expanded to include wind speed as an added predictor variable. A Gaussian kernel function is used in the analysis and circular statistics employed for the wind direction variable. Wind direction and wind speed were both found to have a statistically significant effect on background levels of NO 2 at all three sites. Frequently environmental impact assessments are based on short term baseline monitoring producing a limited dataset. The presented non-parametric regression methods, in contrast to the frequently used methods such as binning of the data, allow concentrations for missing data pairs to be estimated and distinction between spurious and true peaks in concentrations to be made. The methods were found to provide a realistic estimation of long term concentration variation with wind direction and speed, even for cases where the data set is limited. Accurate identification of the actual variation at each location and causative factors could be made, thus supporting the improved definition of background concentrations for use in air quality modelling studies.