A Bayesian hierarchical model for urban air quality prediction under uncertainty

Abstract

Urban air quality is subject to the increasing pressure of urbanization, and, consequently, the potential impact of air quality changes must be addressed. A Bayesian hierarchical model was developed in this paper for urban air quality predication. Literature data on three pollutants and four external driving factors in Xiamen City, China, were studied. The air quality model structure and prior distributions of model parameters were determined by multivariate statistical methods, including correlation analysis, classification and regression trees (CART), hierarchical cluster analysis (CA), and discriminant analysis (DA). A multiple linear regression (MLR) equation was proposed to measure the relationship between pollutant concentrations and driving variables; and Bayesian hierarchical model was introduced for parameters estimation and uncertainty analysis. Model fit between the observed data and the modeled values was demonstrated, with mean and median values and two credible levels (2.5% and 97.5%). The average relative errors between the observed data and the mean values of SO 2, NO x , and dust fall were 6.81%, 6.79%, and 3.52%, respectively.