Re-framing the Gaussian dispersion model as a nonlinear regression scheme for retrospective air quality assessment at a high spatial and temporal resolution

Abstract Regression models (e.g. Land-Use Regression) are currently the most popular way to estimate retrospective exposures to air pollution. However, these models lack important features of atmospheric dispersion. We developed a new non-linear air quality regression model which is based on the physical grounds of the well-established and commonly applied Gaussian dispersion model. This was achieved through parametrization of the basic Gaussian model (including its standard deviations) and optimizing the parameters to provide a least-squares fit with ambient measurements at each individual time-point. The new model (GaussODM) outperformed both a simpler regression model and a benchmark interpolation model in predicting spatial ambient nitrogen oxides (NOx) concentrations. The GaussODM enables a deeper understanding of the relationship between air pollution and adverse health effects. This is partly because it is better adapted at incorporating meteorological data and the effects of elevated emissions compared with previously available air pollution regression models.