Scaling of the objective function gradient for full wavefield inversion

Abstract

Computation of the gradient in full wavefield inversion using the adjoint state method requires back‐propagation of the data residual. We show that if proper scaling of the gradient is desired, the data residual should be properly preprocessed before back‐propagation. We describe an approach that can be followed in order to obtain meaningful gradient amplitudes and illustrate its application for the cases of pressure and multi‐component (particle velocity) recordings in isotropic elastic media.