Title

Use of ensemble based methods for oil spill risk assessment investigations

Document Type

Conference Proceeding

Date of Original Version

1-1-2014

Abstract

Oil spill risk assessment studies typically involve two major steps: (1) estimating the risk of an oil release and its associated characteristics (location, duration, amount, and type of oil) and (2) the transport, fate, and effects of the oil that is released. The risk of release is normally based on an analysis of historical data for the area of interest and the potential sources of oil (vessels, ruptured pipelines, and blowouts). The transport, fate, and effects risk is estimated by performing stochastic simulations with a spill model, where the physical data (winds, currents, waves, and ice dynamics/coverage) used to force the spill model are selected from model hindcasts of these environmental variables. The physical model hindcasts are derived from one source that uses one model type/formulation, solution methodology, turbulence parameterization, grid system, and data assimilation method. Hindcasts are typically performed for 8 to 15 yrs and ideally are from models that have been validated with and assimilate observations (winds, currents, sea surface temperature, sea surface heights, drifter data, density profiles. HE radar measurements, and ice cover). Predictions from stochastic simulations are provided in the form of surface and shoreline oiling probability maps at selected times after release. These maps are then compared to environmental and cultural resources of interest. If the simulations are three dimensional, then seabed oiling and oil concentrations in the water column are provided. The inherent assumption in this strategy is that the use of hindcasts from a carefully selected, validated, data assimilating model adequately represents the risk associated with uncertainties in environmental forcing. It also assumes that estimates of the risk improve as the length of the environmental model hindcasts increase, as more environmental variability is included as input to the spill model, and that the models accurately reflect environmental forcing. Evaluation of the predictive performance of many models for ocean currents of shelf waters in many parts of the world, from both inter-comparison studies and model validation efforts, show that in general predictive performance is generally poor to marginal and that no model seems to systematically outperform other models when evaluated against a variety- of metrics. In the present study, spill risk assessments have been performed using hindcasts from several different validated, state-of-the- Art. data assimilating environmental models and show substantial differences in the surface and shorehne oiling probability maps for the Gulf of Mexico. These differences can be wide spread and are not simply restricted to low probability areas. It is clear from these comparisons that the use of hindcasts from any one model is unlikely to represent the uncertainty associated with environmental forcing. To address this problem and optimize the use of available hindcasts, an ensemble methodology has been developed and implemented. In the ensemble approach, the environmental forcing is selected randomly from one of the available hindcasts for the study area of interest, rather than from just one hindcast. These hindcasts can be from different implementations (grid system, turbulence closure scheme, initialization, data assimilation method, etc.) of one environmental model or from the application of different models. A protocol has also been developed and implemented to allow assigning various weights to different hindcast model products based on an assessment of their performance in validation studies. The use of the ensemble approach provides a substantially improved representation of the spill transport risk that more accurately reflects the uncertainty in environmental forcing.

Publication Title, e.g., Journal

Proceedings of the 37th AMOP Technical Seminar on Environmental Contamination and Response

This document is currently not available here.

Share

COinS