Quantifying Uncertainty in an Oil Fate Model Using a Polynomial Chaos Surrogate

An uncertainty quantification framework is presented for oil fate simulations based on a non-intrusive polynomial chaos (PC) method. The method consists in constructing a computational cheap surrogate for the model, where an output is defined as a polynomial expansion of the uncertain input parameters. The surrogate is built through an ensemble of model realizations, and once constructed, it should be able to approximate the model's output for any given value of the input parameters inside a pre-established range. The capabilities of the method are illustrated by simulating the far field dispersal of oil in a Deep Water Horizon blowout scenario. Two sources of uncertainty are considered. The advecting ocean currents, which are provided by the outputs of a general ocean circulation model, and the oil droplet size distribution, which affects the buoyancy of the spilled oil. The forward propagation of uncertainty is carried out in a hindcast simulation of the first 30 days of the oil spill. We quantify the uncertainty in the simulated oil distribution and estimate the contribution of each uncertain input parameter considered. Also, probabilistic hazard maps of oil impact are presented. The skill of the PC surrogate in representing the model's output is checked by comparing it to a set of actual realizations of the model.