Simulation numérique et assimilation de données variationnelle pour la dispersion atmosphérique de polluants

摘要

This work has led to the development of a three-dimensional chemistry-transport model Polair3D which simulates photochemistry. Model-to-data comparison of ozone and nitrogen oxides measurements over Lille in 1998 has proven its reliability at regional scale. 4 D-var data assimilation has been implemented. It relies on the adjoint model of Polair3D obtained through automatic differentiation. An application of inverse modelling of emissions over Lille with real measurements has been performed. It has proven that the inversion of temporal parameters of nitrogen oxides emissions leads to a significant improvement of forecasts. The so-called second-order sensitivity allows to study the sensitivity of the inversion with respect to the data assimilation system itself by computing its conditioning. This is illustrated by two test cases: short-range dispersion of radionuclides and gas-phase atmospheric chemistry characterized by a wide range of timescales.