Probabilistic modelling of air pollution from road traffic

Abstract

A newly developed deterministic numerical model for air pollution from road traffic is combined with stochastic models in order to predict hourly average concentrations of nitrogen oxides (NOx) with estimated uncertainty. Four stochastic models are considered: Three non-hierarchical models, treating the air pollution model as a black box, and a fourth, hierarchical model, where some of the input variables of this model are also treated as uncertain. The probabilistic models are evaluated by comparing sample or ensemble based probability distributions of concentrations with hourly observed values of NOx at Nordbysletta, Norway, during a 3.5 months campaign period in 2002, where we focus on verification issues such as calibration and sharpness of the predictive distributions.