A stepwise cluster analysis method for predicting air quality in an urban environment

Abstract

A stepwise cluster analysis method has been advanced and applied to air quality prediction. The method has improved monovariate A.I.D. (Automatic Interaction Detection) Algorithm, and can effectively deal with continuous and discrete variables, as well as nonlinear relations between the variables. In the application to air quality prediction, all source variables can carry information about air quality variations, and clustering results are given by cluster trees, so that a set of forecasting systems, which is flexible to reflect changes in source value distributions, can be formed.

In a case study, the method was applied to air quality prediction in the urban district of Xiamen, China. Data concerning three pollutant concentrations and four source types from 31 grid squares during 1984–1988 were used in the calculation. The results of cluster analysis were applied to the prediction of air quality in 1989. Through graphical and statistical tests, it was indicated that 82.8% of monitored concentrations were within the predicted radius, and, compared with the predicted mean concentrations, 76.3% of the predicted data had relative errors lower than ±20%, and 61.3% had errors lower than ±15%; thus showing the good performance of the method.