A cluster multivariate statistical method for environmental quality management

This study advances a Linear-regression-based stepwise cluster analysis (LSCA) and a quadratic-regression-based stepwise cluster analysis (QSCA) model. The models have the advantages of (1) being independent of complex atmospheric, meteorological and topographical information, (2) dealing with continuous and discrete variables, as well as nonlinear relationships among the variables, (3) facilitating finer analysis of within-cluster variations for the stepwise cluster analysis (SCA) outputs, leading to an improved forecasting accuracy, and (4) providing a reasonable result–interpretation since any variation of the explanatory variable will lead to the corresponding change of the response level. The models are then applied to the city of Xiamen in China for forecasting in air quality management. They are also compared with several alternative statistical models: SCA, decision trees (DT) and quadratic regression (QR). The results show that, among the five prediction models, the QSCA has the best forecasting performance followed by the LSCA. Since the diverse and potentially nontechnical user communities׳ interests are embraced by the approaches, they could be robust in terms of the varying levels of knowledge backgrounds, application capabilities, and data availability.