A predictive Bayesian data-derived multi-modal Gaussian model of sunken oil mass

Hydrodynamic modeling of sunken oil is hindered by insufficient knowledge of bottom currents. In this paper, the development of a predictive Bayesian model, SOSim, for inferring the location of sunken oil in time, based on sparse, qualitative or quantitative near-real time field data collected immediately following a spill, is described. Mapped output represents unconditional multi-modal Gaussian relative probabilities of finding oil at points across a relatively flat bay bottom, in time. The method of images is extended to address curvilinear reflecting shorelines. The model is demonstrated to locate the entire DBL-152 spill, given field data covering part of the area affected, and to project oil movement near curvilinear shoreline boundaries given simulated field data at two points in time. Limitations include accountability for discontinuous boundary conditions. Further development is recommended, including development of capability for accepting bathymetric data, for modeling continuous oil releases, and for 3-D modeling of suspended oil.