Sep 2023
Forecasting the fate and transport of marine oil spills is an important application for operational oceanography. It plays a pivotal role in designing emergency response plans and in decision making during oil spill incidents. This thesis addresses the challenges of modeling oil spill transport in marine environments under meteorological and oceanographic (metocean) data uncertainty. A novel approach is proposed for efficiently propagating data uncertainties and forecasting oil spill trajectories in stochastic metocean data fields. Realistic data generated by an ensemble ocean data assimilation system in the Red Sea are utilized to demonstrate the effectiveness of the approach in improving the oil spill transport prediction. Moreover, the uncertainty propagation method is validated using the Sabiti tanker oil spill incident in the Red Sea, showing superior performance compared to deterministic methods. The thesis also presents an algorithm for oil spill source identification from remote sensing images, an essential step in environmental forensics. A Bayesian approach is proposed for estimating the source location, time, duration and volume of an oil spill from images of oil contours. A Markov chain Monte Carlo technique is employed to sample the posterior distribution of the oil spill source parameters. To accelerate the source identification algorithm, an iterative data-driven approach is presented for constructing a localized surrogate of the oil spill model. By utilizing polynomial chaos expansion, the approach builds an inexpensive surrogate enabling faster sampling of the posterior distribution. In conclusion, the methodologies developed in this thesis offer valuable insights for decision makers to improve oil spill response and management strategies.
