Prediction of seasonal extremes of one-hour average urban CO concentrations

Abstract

The annual and seasonal extremes of pollutant concentrations in urban areas tend to represent samples from a long-term non-stationary series so that purely non-causal, statistical methods for their prediction are largely inapplicable. The paper described a method to determine the seasonal extremes of 1-h average CO concentrations from vehicle patterns and emissions, basic meteorological measurements and historical records of ambient concentrations. The method links the output of a deterministic Gaussian plume line source model (which provides average winter trends on an annual basis) with knowledge of a suitable parametric form of the probability density function (pdf) of the daily peak 1-h CO concentrations. The deterministic model requires only average emission and meteorological data as input, although the approach outlined can be extended to include more complex deterministic models with more detailed dynamic input information. Knowledge of the pdf of ambient concentrations is gained from past data by applying goodness-of-fit tests based upon maximum likelihood estimation and its accuracy is assessed by examining prediction performance for the extremes of interest. Problems of non-stationarity and autocorrelation are minimized by restricting attention to the winter season and to the evening peak concentration. The method is used to predict maxima of 1-h CO concentrations for winter seasons in Canberra, Australia, although it applies to other extremes at other time averages, such as 8-h averages, and to other pollutants where they are dispersed predominantly from mobile sources.