Sparsity promoting formulations and algorithms for FWI

Full Waveform Inversion (FWI) is a computational procedure to extract medium parameters from seismic data. FWI is typically formulated as a nonlinear least squares optimization problem, and various regularization techniques are used to guide the optimization because the problem is ill-posed. We propose a novel sparse regularization which exploits the ability of curvelets to efficiently represent geophysical images. We then formulate a corresponding sparsity promoting constrained optimization problem, which we solve using an open source algorithm. The techniques are applicable to any inverse problem where sparsity modeling is appropriate. We demonstrate the efficacy of the formulation on a toy example (stylized cross-well experiment) and on a realistic Seismic example (partial Marmoussi model). We also discuss the tradeoff between model fit and sparsity promotion, with a view to extend existing techniques for linear inverse problems to the case where the forward model is nonlinear.