Statistical surface ozone models: an improved methodology to account for non-linear behaviour

Abstract

Using UK data as a case study, this paper demonstrates that statistical models of hourly surface ozone concentrations require interactions and non-linear relationships between predictor variables in order to accurately capture the ozone behaviour. Comparisons between linear regression, regression tree and multilayer perceptron neural network models of hourly surface ozone concentrations quantify these effects. Although multilayer perceptron models are shown to more accurately capture the underlying relationship between both the meteorological and temporal predictor variables and hourly ozone concentrations, the regression tree models are seen to be more readily physically interpretable.