Prediction of surface ozone episodes using clusters based generalized linear mixed effects models in Houston–Galveston–Brazoria area, Texas

Abstract

A two–stage strategy is proposed to predict regional peak ozone episodes in the Houston–Galveston–Brazoria (HGB) area of Texas, USA. With the forecasted meteorological information, ozone episodes can be predicted one day in advance. Three generalized linear mixed effects models (GLMMs) are built with air quality and meteorological data monitored at CAMS35, CAMS403 and CAMS1015; wind field data from 8 monitoring sites in HGB area are used to generate clusters which represent distinct weather patterns. Air quality and meteorological data during ozone seasons (Apr. 1st – Oct. 31st) from 2003 to 2005 are used to build site–specific prediction models. Data of ozone season from 2006 to 2007 are used to test these models. Compared to linear regression models (LM), generalized linear models (GLMs), multilayer perceptron (MLP) and support vector machine (SVM), GLMM which considers differences in ozone formation and diffusion in distinct weather patterns has the smallest fitting and prediction error on ozone exceedances and can detect the most number of exceedance days correctly.