Reconstruction of the Fukushima accident atmospheric source term using inverse modelling & Inverse modelling of regional carbon monoxide emissions using ground observations
Wednesday, Januray 25th, CERFACS Conference Room - 14h00
Abstract.
In the context of atmospheric dispersion of pollutant and atmospheric chemistry, the main uncertainty is often the source term, or the emission fluxes of the pollutants. Inverse modelling, that makes optimal use of observed concentrations and of a chemistry and transport model, allows to estimate this source term. I shall address two distinct problems on this subject.
Inverse modelling techniques are first applied to the Fukushima-Daiichi accident, and estimates for iodine 131 and caesium 137 source term are obtained. One major difficulty is that the number of direct measurements of radionuclide activity concentration does not exceed a few hundreds. As a consequence, one needs to fully exploit any available information, such as the positivity of the source term.
Another difficulty is the estimation of the prior errors magnitude which significantly influences the retrieval. To address these two issues, we generalise the Desroziers parameter estimation technique used in meteorology to take into account the source term positivity. Our retrieval is consistent with the Japanese's estimation.
A second example addresses the inverse modelling of atmospheric pollutant emissions using in situ concentration measurements that are impacted by representativity errors. We focus on carbon monoxide at the scale of France, using observations from the BDQA database. To do so we couple a 4D-Var for inverse modelling to a simple subgrid statistical model. Not only does the formalism lead to emission estimations close to the ones that can be obtained using space-borne instruments, but we can also quantitatively characterise the representativity of each station. Using this formalism, one can forecast carbon monoxide concentrations on the ground with a station-averaged Pearson correlation coefficient of about 70% and a normalised bias of about 1%.
Marc Bocquet (1,2), Mohammad Reza Koohkan (1,2), Victor Winiarek (1,2), Olivier Saunier (3), Anne Mathieu (3),
(1) Université Paris-Est, CEREA joint laboratory Ecole des Ponts ParisTech and EdF R&D, France.
(2) INRIA, Paris Rocquencourt research center, France.
(3) IRSN, Fontenay-aux-Roses, France.
Inverse modelling techniques are first applied to the Fukushima-Daiichi accident, and estimates for iodine 131 and caesium 137 source term are obtained. One major difficulty is that the number of direct measurements of radionuclide activity concentration does not exceed a few hundreds. As a consequence, one needs to fully exploit any available information, such as the positivity of the source term.
Another difficulty is the estimation of the prior errors magnitude which significantly influences the retrieval. To address these two issues, we generalise the Desroziers parameter estimation technique used in meteorology to take into account the source term positivity. Our retrieval is consistent with the Japanese's estimation.
A second example addresses the inverse modelling of atmospheric pollutant emissions using in situ concentration measurements that are impacted by representativity errors. We focus on carbon monoxide at the scale of France, using observations from the BDQA database. To do so we couple a 4D-Var for inverse modelling to a simple subgrid statistical model. Not only does the formalism lead to emission estimations close to the ones that can be obtained using space-borne instruments, but we can also quantitatively characterise the representativity of each station. Using this formalism, one can forecast carbon monoxide concentrations on the ground with a station-averaged Pearson correlation coefficient of about 70% and a normalised bias of about 1%.
Marc Bocquet (1,2), Mohammad Reza Koohkan (1,2), Victor Winiarek (1,2), Olivier Saunier (3), Anne Mathieu (3),
(1) Université Paris-Est, CEREA joint laboratory Ecole des Ponts ParisTech and EdF R&D, France.
(2) INRIA, Paris Rocquencourt research center, France.
(3) IRSN, Fontenay-aux-Roses, France.
