Quantitative risk analysis in numerical oil spill modeling system in offshore bays and estuaries
Although oil exploration companies and ship operators try to prevent oil spills, there appears to be no way to completely ensure a spill-free coastal environment. To help with rapid spill response, operational agencies have developed oil spill forecasting systems that provide the oil spill path predicted by sophisticated models. These models have focused on mechanistic prediction - using forecast winds and tides with a hydrodynamic model to predict the water surface currents, and an oil spill transport model to track representative particles of oil. There are several layers of uncertainty and error that are unavoidable in such systems, including such obvious elements as wind/tide forecast error and more subtle model elements such as the numerical discretization, type of governing equations, or choice of model grid. Prior to the present work, there has been no attempt to quantify the uncertainty associated with all the modeling levels of an operational oil spill forecast system.
Herein, concepts from the discipline of risk analysis are used to examine the performance of the oil spill modeling system developed under the support of the Texas General Land Office (TGLO) and the Texas Water Development Board (TWDB). In the present context "risk'' is interpreted as the risk of an oil spill forecast track being wrong; that is, we are not examining the risk of a spill occurring, but instead the risks that operational managers must evaluate when deciding on a spill mitigation strategy based on the predicted track for an oil spill.
This study develops a modeling risk analysis approach - Fuzzy Analytic Hierarchy Process with Sensitivity Analysis (FAHP-SA), which integrates several ideas from risk analysis and applies them in a unique context. The method is demonstrated with a case study in Galveston Bay (Texas, USA), which indicates that the overall choice of the hydrodynamic model might be the critical role and introduce the most uncertainty in prediction of an oil spill track. To quantitatively analyze the forecast uncertainty effects, a new Forecast Uncertainty Probability Map (FUPM) method is developed as a way to visualize results of forecast uncertainty based on Monte Carlo simulations. New metrics have been created to quantify uncertainty under different beaching scenarios. Within the Hydrodynamic and oil spill Python (HyosPy) modeling system, we have developed a HyosPy Forecast Reliability Assessment (HFRA) method to evaluate prediction reliability for a particular forecast period. These new methods are applied in a case study in Corpus Christi Bay (Texas, USA). The results indicate that the oil spill forecast of the study area has good predictability up to 18 hours in the future, but the reliability of the forecast degrades very rapidly thereafter.
The technologies developed in this study are a prototype for new methods of quantitative risk, uncertainty, and reliability assessment in a numerical oil spill modeling system. These methods can be readily implemented within existing operational oil spill systems to provide emergency managers improved real-time understanding of the quality of model predictions. Furthermore, these methods can be applied to a priori assess the uncertainty and error of a modeling system, so that future improvements can be targeted at the issues that are the key drivers of uncertainty.