Constrained optimization framework for joint inversion of geophysical data sets
Many experimental techniques in geophysics advance the understanding of Earth processes by estimating and interpreting Earth structure (e.g. velocity and/or density structure). Different types of geophysical data can be collected and analysed separately, sometimes resulting in inconsistent models of the Earth depending on the data used. We present a constrained optimization approach for a joint inversion least-squares (LSQ) algorithm to characterize 1-D Earth's structure. We use two geophysical data sets sensitive to shear velocities: receiver function and surface wave dispersion velocity observations. We study the use of bound constraints on the regularized inverse problem, which are more physical than the regularization parameters required by conventional unconstrained formulations. Specifically, we develop a constrained optimization formulation that is solved with a primal-dual interior-point (PDIP) method, and validate our results with a traditional, unconstrained formulation that is solved with a truncated singular value decomposition (TSVD) for a set of numerical experiments with synthetic crustal velocity models. We conclude that the PDIP results are as accurate as those from the regularized TSVD approach, are less affected by noise, and honour the geophysical constraints.