Keywords

Source term estimation; Variational Bayes inference; Gamma dose rate measurement; CVX optimization toolbox; Covariance matrix model

Location

Session B1: Data Assimilation Techniques for Uncertainty Reduction

Start Date

13-7-2016 2:50 PM

End Date

13-7-2016 3:10 PM

Abstract

Detection of release of an atmospheric pollutant is a problem of interest in environmental sciences. We are concerned with estimation of unknown source term of the release. Formally, the problem is formulated as a linear model where measurements are explained using source-receptor- sensitivity matrix obtained from atmospheric transport model and source term vector which has to be estimated. Specifically, we estimate the release of radioactivity from measurements of gamma dose rate (GDR). The problem of isolation of activity of individual nuclides is poorly conditioned. We propose a probabilistic model with the prior information on intervals of nuclide ratios. The model parameters are estimated using the Variational Bayes method. The proposed algorithm is tested on simulated scenario with 16 nuclides and compared with state-of-the-art optimization approaches.

Share

COinS
 
Jul 13th, 2:50 PM Jul 13th, 3:10 PM

Bayesian Estimation of Source Term of Atmospheric Radiation Release with Interval Prior

Session B1: Data Assimilation Techniques for Uncertainty Reduction

Detection of release of an atmospheric pollutant is a problem of interest in environmental sciences. We are concerned with estimation of unknown source term of the release. Formally, the problem is formulated as a linear model where measurements are explained using source-receptor- sensitivity matrix obtained from atmospheric transport model and source term vector which has to be estimated. Specifically, we estimate the release of radioactivity from measurements of gamma dose rate (GDR). The problem of isolation of activity of individual nuclides is poorly conditioned. We propose a probabilistic model with the prior information on intervals of nuclide ratios. The model parameters are estimated using the Variational Bayes method. The proposed algorithm is tested on simulated scenario with 16 nuclides and compared with state-of-the-art optimization approaches.