It is generally accepted that models of the deep ocean must assimilate observations in order to make realistic forecasts in regions dominated by mesoscale variability (i.e., “ocean weather”). The present study is an attempt to quantify the information on ocean weather that is contained in Lagrangian trajectories, and the best way to extract it. Following a review of ocean data assimilation in a Bayesian framework, including the Ensemble Kalman Filter and the Particle Filter, a new class of idealized models of self advecting vortices is introduced. Through a large number of carefully designed Monte Carlo experiments it is shown when, where and why the Ensemble Kalman Filter will fail. The study concludes with a discussion of a hybrid scheme that takes advantage of the lower computational cost of the Ensemble Kalman Filter and the ability of the Particle Filter to handle highly non-Gaussian probability density functions. / MSc Thesis