Propagation of uncertainty and sensitivity analysis in an integral oil‐gas plume model

Abstract

Polynomial Chaos expansions are used to analyze uncertainties in an integral oil‐gas plume model simulating the Deepwater Horizon oil spill. The study focuses on six uncertain input parameters—two entrainment parameters, the gas to oil ratio, two parameters associated with the droplet‐size distribution, and the flow rate—that impact the model's estimates of the plume's trap and peel heights, and of its various gas fluxes. The ranges of the uncertain inputs were determined by experimental data. Ensemble calculations were performed to construct polynomial chaos‐based surrogates that describe the variations in the outputs due to variations in the uncertain inputs. The surrogates were then used to estimate reliably the statistics of the model outputs, and to perform an analysis of variance. Two experiments were performed to study the impacts of high and low flow rate uncertainties. The analysis shows that in the former case the flow rate is the largest contributor to output uncertainties, whereas in the latter case, with the uncertainty range constrained by aposteriori analyses, the flow rate's contribution becomes negligible. The trap and peel heights uncertainties are then mainly due to uncertainties in the 95% percentile of the droplet size and in the entrainment parameters.

Numerical Study on the Influence of Model Uncertainties on the Transport of Underwater Spilled Oil

“…As indicated by Gonçalves et al [ 5 ], the uncertainties in the oil spill location, the amount of the spilled oil, the flow rate of the oil, the oil droplet size distribution, the ocean current, the subgrid-scale parametrization, the wind drag at the surface, the oil degradation, and other parameters would influence the simulated and predicted fate of the spilled oil. It is necessary to investigate the influence of model uncertainties [ 6 , 14 , 16 , 20 , 21 ]. The start time and spill site are the basic temporal and spatial information for simulating the underwater oil spill, and the wind and current are the necessary input for the underwater oil spill model.…”

“…An analysis of model uncertainties is required for reliable numerical simulation and prediction [ 16 ]. Elliott and Jones [ 17 ] indicated that the wind was a much more important requirement for simulating an oil slick than the current, buoyancy, and weathering processes.…”

“…Li et al [ 19 ] studied the impact of the vertical diffusion scheme on oil dispersion and transportation, and found that the scheme by solving the Langevin equation was the best one. Wang et al [ 16 ] used the polynomial chaos expansions to analyze the uncertainties in six input parameters in an oil-gas plume model, and found the sensitive and insensitive parameters. Gonçalves et al [ 5 ] analyzed the effect of the uncertainties in ocean current and oil droplet size data on the simulated oil concentration in the oil-fate model using a nonintrusive polynomial chaos method.…”

Quantifying uncertainties in fault slip distribution during the Tōhoku tsunami using polynomial chaos

“…In those applications, traditional spectral projection methods [12,15,16] to construct the PC model were successfully implemented. In recent studies, however, the spectral projection technique failed to construct faithfully a PC expansion that represents the forward model [17,55]. This was due to the non-linearity of the forward model and to the internal noise that was present, leading to PC expansion convergence issues.…”

Polynomial Chaos–Based Bayesian Inference of K-Profile Parameterization in a General Circulation Model of the Tropical Pacific

“…The BPDN method offers an additional advantage that a smaller number of model runs compared to the spectral projection method is required to determine the PC coefficients as shown in Section 4c. BPDN was recently employed to build a proxy model for an integral oil-gas plume model (Wang et al 2015) and for an ocean model with initial and wind forcing uncertainties (Li et al 2015). In the former, the model output was noisy due to the iterative solver used in the double-plume calculation (Socolofsky et al 2008) and BPDN proved to be successful in filtering the noise and thus building a representative PC model.…”