Author(s)

H. Al-Madfai, D. G. Snelson & A. J. Geens

Abstract

Modelling and forecasting of time series of concentrations of air pollutants is essential in monitoring air quality and assessing whether set targets will be achieved. While many established time series modelling approaches transform the data to stationarity a priori, the explicit modelling and presentation of the non-stationary components of the series in this application is essential to allow for further understanding of variability and hence more informed policies. The Hierarchical Profiling Approach (HPA) was used to model the multi-year daily air quality data gathered at St. Leonards in Edinburgh, UK spanning from 1st of March 2004 to 15th July 2007. The HPA is an avant-garde approach that explicitly models the non stationary component of time series data at different levels depending on the span of the component, so that within-year disturbances are at Level 1 and year-long variability such as seasonality is at Level 2, and so on. HPA decomposes the variability into deterministic, stochastic and noise and uses continuous models to describe the non-stationary components using the deterministic part of the model. The stationary stochastic component is then modelled using established approaches. The dataset modelled was the total daily concentrations of carbon monoxide. After modelling within-year events at Level 1, a harmonic regression with trend model was used to describe the weekly aggregates of the data at Level 2. This model was then sampled back in the daily domain and no evidence of a larger cyclical component was found. Wind-speed and a dummy intervention-variable indicating the implementation of the smoking ban were considered in a transfer function model. The concluding model included the present and one lagged observations of wind speed. The smoking ban variable was not significant. Keywords: monitoring air quality, Hierarchical Profiling Approach, smoking.

Keywords

monitoring air quality, Hierarchical Profiling Approach, smoking.