Correcting air pollution time series for meteorological variability. With an application to regional PM10 concentrations

In this report we have demonstrated how a series of binary decision rules, known as Classification And Regression Trees (CART), can be used to calculate pollution concentrations that are standardized to levels expected to occur under a fixed (reference) set of meteorological conditions. Such meteo-corrected concentration measures can then be used to identify "underlying" air quality trends resulting from changes in emissions that may otherwise be difficult to distinguish due to the interfering effects of unusual weather patterns.

The examples here concern air pollution data (daily concentrations of SO₂ and PM₁₀). However, the methodology could very well be applied to water and soil applications. Classification trees, where the response variable is categorical, have important applications in the field of public health. Furthermore, Regression Trees, which have a continuous response variable, are very well suited for situations where physically oriented models explain (part of) the variability in the response variable. Here, CART analysis and physically oriented models are not exclusive but complementary tools.