Bayesian Inference and Markov Chain Monte Carlo Sampling to Reconstruct a Contaminant Source on a Continental Scale

• DOI: 10.1175/2008jamc1766.1
• Publication Date: 2008-02-28T19:53:17Z
• Authors: Luca Delle Monache; Julie K. Lundquist; Branko Kosović; Gardar Johannesson; Kathleen M. Dyer; Roger D. Aines; Fotini K. Chow; Rich D. Belles; William G. Hanley; Shawn C. Larsen; Gwen A. Loosmore; John J. Nitao; Gayle A. Sugiyama; Philip J. Vogt
• Keywords: 01 natural sciences; 0105 earth and related environmental sciences

Abstract A methodology combining Bayesian inference with Markov chain Monte Carlo (MCMC) sampling is applied to a real accidental radioactive release that occurred on continental scale at the end of May 1998 near Algeciras, Spain. The source parameters (i.e., location and strength) are reconstructed from limited set measurements release. Annealing adaptive procedures implemented ensure robust effective parameter-space exploration. simulation setup similar an emergency response scenario, simplifying assumptions geometry time known. stochastic algorithm provides likely locations within 100 km true source, after exploring domain covering area approximately 1800 × 3600 km. strength distribution values same order magnitude as upper range reported by Spanish Nuclear Security Agency. By running MCMC large parallel cluster inversion results could be obtained in few hours required for continental-scale releases. With additional testing refinement (e.g., tests also include among unknown parameters), well continuous rapid growth computational power, approach can potentially used real-world future.