Use of kriging models to predict 12-hour mean ozone concentrations in Metropolitan Toronto-A pilot study

Ozone concentrations exhibit spatial variability within metropolitan areas, resulting in significantly different personal exposures among individuals. This paper uses the statistical technique, kriging, to explore the 12-h daytime (8am–8pm) ozone spatial variation and to predict mean outdoor ozone levels at home sites within the Toronto metropolitan area. Outdoor ozone measurements taken in the Toronto metropolitan area between June and August 1992 are used in kriging models to predict outdoor ozone concentrations. The performance of the model is evaluated by comparing actual home outdoor measurements with predicted values. Results indicate that kriging predictions are more accurate than using only the closest stationary ambient site measurements for determining home outdoor ozone concentrations within the metropolitan area. The average variogram obtained from pooling data throughout the entire sampling period shows a clear spatial trend in the outdoor ozone variation. Kriging predictions using the parameters from the average variogram perform as well as those using variograms from individual days. An approach for minimizing sample bias can be used to increase the accuracy of the predictions; cross-validation suggests that it is a reasonable procedure.