Modélisation inverse des sources de pollution atmosphérique accidentelle: Progrès récents

This review discusses recent progress of the inverse modelling of sources of atmospheric pollution, especially from accidental origin, from regional to continental scale. Typically, it amounts to estimate and characterise a pollution source from an industrial site, with a chemistry and transport model, and with measurements of concentration of this pollutant distributed in space and time. The mathematical formalism that allows to fuse these sources of information is given by the Bayesian framework of data assimilation. We show how to define mathematically such a problem, then how to solve it, in particular thanks to nonparametrical methods. Because the chemistry of species related to such a dispersion event can often be considered linear in the short term, more advanced mathematical methods can be implemented with better results. Beyond the estimation of these sources, the computation of the uncertainties attached to this estimation is discussed. Recent examples from the literature are then presented: the ETEX case, the Chernobyl case, and the Algeciras case. Lastly, we describe the impact of the spatio-temporal resolution of the parameter space on the retrieval. In the atmospheric dispersion context, the estimation of the source can depend significantly on this choice.