Master of Science (MS)
Meteorology and Physical Oceanography (Marine)
Date of Defense
First Committee Member
Second Committee Member
Tamay M. Özgökmen
Third Committee Member
Any prediction based on numerical model contains uncertainty. The Deepwater Horizon oil spill forecast is a challenge due to the difficulty in measuring key parameters and the complexity in predicting the oil fate. Polynomial Chaos Expansion (PCE), Gaussian Process Regression (GPR) and Monte-Carlo (MC) sampling are applied to quantify the forecast uncertainties in an integral oil plume model simulating the Deepwater Horizon oil-gas blowout. The main goal is to propagate the inputs and model parametrizations uncertainties, to understand the sensitivity of different uncertain parameters using efficient uncertainty quantification methods, and to validate their results against the MC reference solution. The PCE and GPR approaches require only 100 simulations to reach an acceptable level of accuracy; while the MC approach typically needs thousands or even tens of thousands of simulations. Therefore PCE and GPR approaches are ideal tools for uncertainty analysis in real-time operational oil plume predictions given their efficiency. The uncertainty analysis focused on five uncertain input parameters and on several model output uncertainties for the oil/gas plume: the trap height, the peel height, and different gas mass fluxes. The probability density functions of these outputs and their sensitivity indices are estimated. Sensitivity analysis of trap height and peel height indicates that entrainment parameters and 95th percentile of droplet size (d95) are the most sensitive parameters. Sensitivity analysis of gas mass fluxes as a function of depth shows that the gas to oil ratio dominates the sensitivity near the wellhead and the d95 dominates the sensitivity above the trap height level. Finally. we repeat the UQ experiment by including the flow rate as an additional uncertain parameter. The huge uncertainty range for the flow rate leads to substantial uncertainties in the ensuing quantities of interest whereas constraining the range to reasonable values shifts the dominance to droplet size distribution uncertainties and turbulence parametrization uncertainties.
Uncertainty Quantification; Polynomial Chaos; Gaussian Process; Integral Plume model; Sensitivity Analysis; Deepwater Horizon
Wang, Shitao, "Forward Propagation of Uncertainty and Sensitivity Analysis in an Integral Oil-Gas Plume Model" (2015). Open Access Theses. 570.