Trans-dimensional reflection coefficient inversion of seabed sediments in two spatial dimensions

We introduce an adaptive spatially two-dimensional (2-D) inference method for seabed sediment structure and geoacoustic parameters from wide-angle reflection coefficient spectra which does not require piece-wise one-dimensional (1-D) inversions or fixed assumptions about the 2-D parameter space. This is an advance in reflection coefficient inversion, more accurately capturing the information content of the data while retaining a parsimonious representation of the seabed. The approach employs (a) Bayesian inference with the reversible jump Markov chain Monte Carlo algorithm to allow the number of model parameters to change (i.e., trans-dimensional) and (b) 2-D Voronoi tessellations to enable a spatially irregular model grid with a variable number of cells. Synthetic tests indicate that the method estimates the 2-D geoacoustic model and its uncertainties in a more objective and straightforward way than approaches with fixed dimensions or multi-step 1-D inversions, while the computational cost remains similar to previous approaches. The resulting structure and uncertainties are more directly interpretable than those of fixed dimensional modeling methods.