Probabilistic Inversion of Magnetic UXO Data: Implementing Prior UXO Data from the North Sea

Characterization and discrimination of UXO in magnetic offshore seabed surveys usually have limited success. We investigate how to utilize available prior knowledge on UXO type and quantity in a probabilistic framework, in order to improve on discrimination capabilities. It has previously been demonstrated how Bayesian inference can be utilized in a Markov chain Monte Carlo (MCMC) framework, where the solution can be sampled in a stochastic process. In this project, we extend the previous work containing independent 1-D prior distributions to a more complex case, introducing real quantitative data of actual UXO findings in the North Sea. Here, we develop the methodology to take into account knowledge about known size and shape of different UXO as well as the expected quantities of each type. This enables us to not only sample the posterior distribution of the model parameters, but also assign a probability of each UXO type with respect to the data at hand.