Statistical Analyses of Ozone Temporal Trends: An Application of Multivariate Geostatistics

Space-time modeling of atmospheric pollutants has been actively attempted by many workers, including Kyriakidis (1999) who successfully integrated the deterministic trend m(t) and the probabilistic residual R(t) components of a random variable RV Z(t) through a stochastic simulation approach. However, many of the spatiotemporal studies carried a major assumption that the temporal aspect is fully understood and thus focused primarily on spatial modeling. The main objective of this work is concerned with evaluating the accuracy and suitability of the techniques used for modeling the temporal phenomena. Various statistical methodologies, i.e., linear regression, kriging and stochastic simulation, were performed in the case of predicting tropospheric ozone concentrations in Calgary, Alberta for 1998-2000. It should be emphasized that the more accurate the temporal modeling is performed at various environmental monitoring stations, the higher the probability of success in estimating ozone values at unknown locations. Therefore, exhaustive studies of ozone phenomena must be carried out at as many cities in Canada as possible.