Ocean acoustic tomography involves the observation of several versions of a transmitted signal at the receiver side of an underwater acoustic communication system, that are due to multipath propagation in the ocean. The estimation and the tracking of multipath propagation parameters is considered in this contribution. We propose a bayesian approach to this problem in two steps. First, a deconvolution method is performed record by record and yields an initial estimation of the arrival times. Then, the slow evolution of arrival times over several observations enables applying image processing techniques for peak tracking. In addition, the proposed method accounts for possible distortions of the transmitted signal during the propagation, which significantly influences the arrival time estimation performance. The proposed method is tested on real data.