Seismic velocity structure of oceanic crust by inversion using genetic algorithms

Abstract

SUMMARY We determine the velocity structure along two expanding-spread seismic profiles, shot near the Blake Spur fracture zone in the western North Atlantic. We use the genetic algorithm as an optimization method in our inversion scheme. The genetic algorithm requires a forward modelling tool, for which we use kinematic ray tracing when traveltimes are required, and Chapman's (WKBJ seismogram) method when waveforms are needed. We optimize the seismic problem by first making a traveltime fit with velocity functions consisting of linear velocity gradients: these calculations are fast. Subsequently, we fit waveforms using B-splines for the velocity function. The splines give more consistent synthetic seismograms than linear velocity gradients because caustics caused by the model discretization do not introduce amplitude distortions, so we remain within the region of validity of asymptotic theory. We introduce a stopping criterion for genetic algorithms similar to the one used in crude Monte Carlo methods. Finally, we illustrate the whole procedure by applying the method to P- and S-wave refraction data, and compare the results of automatic inversion for the velocity-depth structure with results from trial-and-error forward modelling.