Bayesian Source Identification of Marine Pollution from Moving Vessels in the Mediterranean

Identifying the source of marine pollution is critical for environmental protection, yet it remains computationally challenging when sources are moving and ocean currents are uncertain. We propose a Bayesian inference framework to identify single and multiple release events from vessels moving along predefined paths in the Mediterranean Sea. Our approach utilizes a Markov Chain Monte Carlo (MCMC) algorithm with an adaptive scheme to robustly infer release locations, injection times, and relative source contributions.

The likelihood function is constructed using logistic regression to quantify the discrepancy between binary satellite-like observations and a probabilistic spill distribution generated by a stochastic Lagrangian Particle Tracking (LPT) model driven by realistic ocean currents. We demonstrate the efficiency of this method through synthetic scenarios involving both separate and overlapping pollution patches. The results highlight the framework's ability to successfully reconstruct release parameters even in complex, stochastic flow fields, showing strong agreement when compared against global optimization baselines. This work offers a rigorous tool for environmental forensics in maritime contexts.