An efficient and robust air quality prediction system for describing and forecasting the daily maximum of the 8-hr averaged ground-level ozone concentration in Macau is developed in the present study. When the ozone concentration is modeled statistically, it is common to use one selected model for all seasons, which may not deliberate the effect of the seasonal variations. In this study, seasonal Kalman filter based models, namely the Mnon-episode model, the Mepisode model and the Mtransition model were proposed by using the Bayesian information criterion. Then, the Bayesian model averaging approach was implemented to enhance the efficiency and robustness of the developed models by incorporating the air quality models for different seasons into one merged model Msystem. The method was found to be efficient and the Msystem was proved to be the most plausible one after comparing its performance with other models constructed in this study. The model Msystem is the first of its kind that assembles air quality models of different seasons together to provide adaptive estimation of ozone concentration for all seasons.