Meeting Abstracts
Stochastic Plume Simulations with Multiple Sources of Uncertainty
Emanuel Coelho, Gregg Jacobs, Germana Peggion, Clark Rowley, Patrick Hogan
University of New Orleans (resident at the Naval Research Laboratory)
(Abstract received 05/09/2012 for session C)
ABSTRACT
Stochastic plumes are here defined through the Risk Assessment Codes (RAC) that associate numbers 1 to 6 to each geographical location on a grid, based on the likelihood (0.25, 0.5, 0.75) of having tracer particles within prescribed threshold ranges (very near, near, far, very far) or by the likelihood of reaching prescribed thresholds of tracer concentrations (very high, high, medium, low). Such that 1 corresponds to likelihood above 0.75 of tracers inside the “very near” threshold or high concentrations, and 6 corresponds to likelihoods below 0.25 of tracers within the “very far” range or low concentrations. The likelihoods are taken from probability distribution functions derived from ensemble runs of ocean models and perturbed Lagrangian tracer simulations or Eulerian diffusion-advection runs.
This work discusses the application of this framework for two test cases. The Fukushima Dai-ichi power plant accident off Japan, where the unknown amounts of radiation being released motivated a Lagrangian methodology, and the Deep Water Horizon Oil Spill accident using an Eulerian simulation. Both simulations are based on velocity fields from ensembles of the Navy Coastal Ocean Model (NCOM), configured over the two regions (NE Japan and Northern Gulf of Mexico), using perturbed initialization and atmospheric forcing. For the Fukushima case tracers were released on each ensemble member every hour. For the Deep Water Horizon oil spill case each ensemble member was coupled with a diffusion-advection solution to estimate possible scenarios of oil concentrations as defined in the Bonn Agreement Oil Appearance Code, using perturbed estimates of the released amounts as the source terms at the surface. In the Deep Water Horizon case the stochastic plume estimates showed a significant improvement relative to a deterministic estimate, with accurate stochastic dispersion maps 1 month after the spill.