Using band-tridiagonal preconditioners for 1D seismic waveform inversion

Abstract

Geophysical inversion is quite a challenge. Many different methods and strategies have been developed in the past decades. Global methods like simulated annealing and genetic algorithms do not depend on an initial model, although they may be time consuming using a huge number of fitness evaluations. Local methods such Gauss-Newton require a ’good’ start model and may be not convergent to the global minimum of the objective function. In this paper we evaluate the use a band-tridiagonal preconditioner in the multiple reweighted least square inversion method (MRLS) to the seismic waveform inversion. The MRLS inversion method is a type-gradient method that generates many candidate models for each sensitivity matrix, exploring the model space in a very effective way. A numeric example shows that the inverted model has better fitness and may be obtained with less computational CPU time, when compared to the results given by the Gauss-Newton approach.