An Automated Classification Scheme Designed to Better Elucidate the Dependence of Ozone on Meteorology

Abstract

This paper utilizes a two-stage (average linkage then convergent k means) clustering approach as part of an automated meteorological classification scheme designed to better elucidate the dependence of ozone on meteorology. When applied to 10 years (1981?90) of meteorological data for Birmingham, Alabama, the classification scheme identified seven statistically distinct meteorological regimes, the majority of which exhibited significantly different daily 1-h maximum ozone concentration distributions. Results from this two-stage clustering approach were then used to develop seven ?refined? stepwise regression models designed to 1) identify the optimum set of independent meteorological parameters influencing the O3 concentrations within each meteorological cluster, and 2) weigh each independent parameter according to its unique influence within that cluster. Large differences were noted in the number, order, and selection of independent variables found to significantly contribute (α = 0.10) to the variability of O3. When this unique dependence was taken into consideration through the development and subsequent amalgamation of the seven individual regression models, a better parameterization of O3's dependence on meteorology was achieved. This ?composite? model exhibited a significantly larger R2 (0.59) and a smaller rmse (12.80 ppb) when compared to results achieved from an ?overall? model (R2 = 0.53, rmse = 13.85) in which the meteorological data were not clustered.