A METHODOLOGY TO CHARACTERISE THE SOURCES OF UNCERTAINTIES IN ATMOSPHERIC TRANSPORT MODELLING

The atmospheric dispersion modelling of pollutants is based on models, but also on data and users, who lead to uncertainties, i.e. to differences between the results of the models and the physical reality to describe. The question of the uncertainty of dispersion models is a subject of increasing interest for primarily two reasons: • In spite of the significant number of research works on atmospheric dispersion in the last 30 years, results of simulations preserve an important level of uncertainty. Since the reduction of this uncertainty will be more and more difficult, it thus seems today necessary to characterize it and, if possible, to quantify it. • The development of computer performances allows today the use of models in real time for decision-making aid, e.g. for the control of an industrial facility or for the management of an accident. In this context, where important decisions must be made quickly, it becomes extremely important to provide to the results of models an information on the uncertainty associated with these results. For better characterising uncertainties associated with atmospheric dispersion models, we propose a methodology of analysis which led us to make an inventory of all the sources of uncertainties of a model, considering all kind of models (Gaussian, Lagrangian, Eulerian). This methodology is based on a decomposition of the modelling process in four steps: • The collection of input data (sources, meteorology, topography, land cover…), with uncertainties in the measurement of these data, in their treatment and in their representativeness. • The modelling of the meteorological field, with uncertainties associated with intrinsic nature of atmospheric phenomena and with the quality of the models/approaches used to describe these phenomena. • The modelling of dispersion processes and of the physicochemical transformations. • The statistical treatment of results: it is shown that the choice of the output parameters (average annual or hourly maximum, maximum ground value or position of this maximum, etc.) modify, in an important way, the uncertainty of the results and how to evaluate it. The methodology proposed in this work should make it possible to progress towards a more systematic quantification of uncertainties associated with modelling of atmospheric dispersion