Regression and stochastic models for air pollution—I. Review, comments and suggestions

Abstract

Regression and stochastic models for air pollution are considered and their advantages and disadvantages are analysed. General theoretical aspects of these models are reviewed, focusing on the points that are of particular interest to air pollution modelling. It is argued that, from both the conceptual and the practical points of view, stochastic models are preferred to regression models. The misuse of the statistical methods in regression models is rather common in the air pollution literature and some examples are discussed. Emphasis is also given to the distinction between deterministic and stochastic trends, as there is some confusion in the literature. It is argued that linear or higher-order deterministic trends cannot be incorporated in univariate stochastic models in a parsimonious way. Some suggestions are made on how elements of the two methodologies (regression and stochastic models) can be combined. Finally, some aspects of the theory of co-integration are reviewed and the possibility raised of applying the methodology to air pollution modelling and to atmospheric sciences.