A time series analysis of aerosol composition measurements

Time sequence measurements of the elemental composition of aerosols, on an hourly to fewhourly basis, may be analyzed statistically for chemical associations which are characteristic of sources before modification of these associations during transport through the atmosphere. If correlations between elemental abundances are computed in the measured data set, high serial correlations in the series can introduce a bias. Therefore, Box-Jenkins techniques are used to obtain the residual series as an estimate of the sequence of independent "shocks" for each element. It is argued that, due to the mixing effects of atmospheric factors, these shocks are the appropriate random variables for investigating the elemental associations within a data set of time sequence aerosol composition measurements. Examples from measurements at different locations in Florida of the concentrations of nine elements, every 2 h for about two weeks, are used to illustrate the method. In particular it is shown that principal component analysis of the residuals gives rise to different interpretations than analysis of the observed series.