Paper detail

Multiple change-point Poisson model for threshold exceedances of air pollution concentrations

A Bayesian multiple change-point model is proposed to analyse violations of air quality standards by pollutants such as nitrogen oxides (NO2 and NO) and carbon monoxide (CO). The model is built on the assumption that the occurrence of threshold exceedances may be described by a non-homogeneous Poisson process with a step rate function. Unlike earlier approaches, our model is not restricted by a predetermined number of change-points, nor does it involve any covariates. Possible short-range correlations in the exceedance data (e.g., due to chemical and meteorological factors) are removed via declusterisation. The unknown rate function is estimated using a reversible jump MCMC sampling algorithm adapted from Green (1995), which allows for transitions between parameter subspaces of varying dimension. This technique is applied to the 17-year (1993-2009) daily NO2, NO and CO concentration data in the City of Leeds, UK. The results are validated by running the MCMC estimator on simulated data replicated via a posterior estimate of the rate function. The findings are interpreted and discussed in relation to some known traffic control actions. The proposed methodology may be useful in the air quality management context by providing quantitative objective means to measure the efficacy of pollution control programmes.