Abstract
The aim of this study is to investigate the potential, methods, and benefits of applying supervised machine-learning techniques to enhance deterministic air quality forecasts. These forecasts are produced at the Institute of Environmental Protection–National Research Institute using a numerical grid-based model that solves a system of conservation equations describing atmospheric dynamics as well as pollutant transport and transformation. Four alternative machine-learning models were tested, yielding similar results. The outcomes indicate the possibility of achieving a near-perfect forecast at the locations of measurement stations. It also turns out that if the pollutant concentration values predicted by the deterministic model are not used as features in the machine-learning model, the quality of the final forecast drops drastically.