Integrating measurement representativeness and release temporal variability to improve the Fukushima-Daiichi 137Cs source reconstruction

    The Fukushima-Daiichi accident involved massive and complex releases of radionuclides in the atmosphere. The releases assessment is a key issue and can be achieved by advanced inverse modelling techniques combined with a relevant dataset of measurements. A Bayesian inversion is particularly suitable to deal with this case. Indeed, it allows for rigorous statistical modelling and enables easy incorporation of informations of different natures into the reconstruction of the source and the associated uncertainties.
    We propose several methods to better quantify the Fukushima-Daiichi 137Cs source and the associated uncertainties. Firstly, we implement the Reversible-Jump MCMC algorithm, a sampling technique able to reconstruct the distributions of the 137Cs source magnitude together with its temporal discretisation. Secondly, we develop methods to (i) mix both air concentration and deposition measurements, and to (ii) take into account the spatial and temporal information from the air concentration measurements in the error covariance matrix determination.
    Using these methods, we obtain distributions of hourly 137Cs release rates from 11 to 24 March and assess the performance of our techniques by carrying out a model-to-data comparison. Furthermore, we demonstrate that this comparison is very sensitive to the statistical modelling of the inverse problem.