A tunneling approach for clustered priors in full-waveform inversion

Abstract

While seismic data is an extremely useful source of information, many seismic inversion problems can be difficult or impossible without the use of supplementary prior information. Knowledge of plausible rock-types in a given setting, available from well logs and/or geological understanding, can, in particular, be an effective supplement to seismic data. However, when this prior information describes clustering in rock physics or elastic-parameter properties, it can be difficult to effectively include in full waveform inversion due to the limitations of the local optimization strategies used. In particular, regularization terms based on clustering-type information can create local minima, which are prone to hindering convergence with conventional FWI optimization approaches. Here, we propose an optimization strategy for full waveform inversion, in which global regularization information is partially accounted for, and in which a model can, under the right circumstances, “tunnel” between basins, to honour prior information. This tunneling is implemented in conjunction with conventional, local optimization strategies; after each local model update, a second update follows, in which each grid cell tunnels to a new cluster if the move is warranted by the update history of the model (its “momentum”) and the model regularization penalty (its “potential”). With a synthetic example, we illustrate that this approach offers the potential to improve upon conventional strategies when prior information represents rock-type clustering.