A Comparison of Two Approaches to Classification of Air Pollution Data
A. Narayanan and Thomas W. Sager
Weather and emissions are the primary determinants of air pollution. The classification of days into categories in terms of their meteorological potential for producing harmful air pollution and help regulators in controlling it. The Texas Air Control Board (TACB), the state agency whose responsibilities include monitoring and controlling air pollutants in the State of Texas, recently classified a selected set of days into nine well-defined categories (WPC) for this purpose. Guided by a written protocol and the exercise of professional judgment, meteorologists assigned each day to a WPC category on the basis of their examination of the weather chart for the day and independently of the air pollution level. This is a laborious, time-consuming task. The categories are then used in understanding the movement of air and relating the weather pattern classification to formation of elevated levels of ground-level ozone, a significant air pollutant.
The aim of this paper is two-fold: (1) to imitate the labor-intensive judgemental WPC as nearly as possible by a purely automatic statistical classification based on discriminant analysis, and (2) to determine the extent to which either WPC or the statistical classification (called STATCLASS) successfully discriminates high- from low-ozone-potential days. It is found that STATCLASS was able to assign about 60–70% of the days to the same category assigned by WPC – an agreement rate comparable to that of the TACB in cross-validation checks made on WPC. We also found that both schemes were reasonably successful in discriminating high-ozone from low-ozone days but that STATCLASS was more successful than was WPC.
Discriminant analysis; multiple regression; stepwise discrimination; stepwise regression; selection of variables; weather pattern classification; air pollution data.