• Autoregressive time series;
  • Box-Cox power transformation;
  • Transform Both Sides (TBS) regression;
  • air pollution;
  • ozone concentration


The statistical analysis of the concentration of air pollutants in the atmosphere is generally applied to a time series of average concentrations. Statistical difficulties arise from the non stationarity of the series, the nonnormality of the measured concentration, the lack of independence and the important influence of external factors. A new model is proposed for the time series consisting of daily averages of ground level ozone. The annual trend is modelled by a function of the length of day, the nonnormality is taken into account by a Box-Cox transformation and the errors are supposed to be autoregressive of order 1. This method is applied to the analysis of NABEL (Nationales Beobachtungsnetz für Luftfremdstoffe) data (two years data from two locations), Switzerland.