Air pollution is a major environmental threat to human health. Therefore, multiple systems have been developed for early prediction of air pollution levels in large cities. However, deterministic models produce uncertainties due to the complexity of the physical and chemical processes of individual systems and transport. In turn, statistical and machine learning techniques require a large amount of historical data to predict the behavior of a variable. In this paper, we propose a data fusion model to spatially and temporally predict air quality and assess its situation and risk for public health. Our model is based on the Joint Directors of Laboratories (JDL) model and focused on Information Quality (IQ), which allows us to fine tune hyper-parameters in different processes and trace information from raw data to knowledge. Expert systems use the information assessment to select and process data, information, and knowledge. The functionality of our model is tested using an environmental database of the Air Quality Monitoring Network of Área Metropolitana del Valle de Aburrá (AMVA in Spanish) in Colombia. Different levels of noise are added to the data to analyze the effects of information quality on the systems' performance throughout the process. Finally, our system is compared with two conventional machine learning-based models: Deep Learning and Support Vector Regression (SVR). The results show that our proposed model exhibits better performance, in terms of air quality forecasting, than conventional models. Furthermore, its capability as a mechanism to support decision making is clearly demonstrated.