Least squares inversion of field seismic data for an elastic 1-D Earth

Elastic 1-D inversion of seismic reflection data is successfully applied to common midpoint field data of varying offset ranges. The inversion, actually iterative forward modeling, operates on plane wave seismic data and yields P-wave velocity, (Vₚ), density, and S-wave velocity, (Vₛ) for each of many (typically hundreds) equal time thickness layers. The problem is viewed in terms of a multi-dimensional optimization problem where the error function is defined as the square of model and data misfit. The error is minimized by the Newton method which requires slope (sensitivity of each datum to each model parameter) and curvature (full Hessian) of the error function at each of several iterations. The algorithm is tested on realistic synthetic data, and when applied to conventional common midpoint data is very helpful in determining the nature of a bottom simulating reflector in the Carolina Trough off shore South Carolina. The algorithm is also applied to deep water data acquired in the Nankai Trough offshore Japan as part of a tectonic study of the accretionary prism. Here the waveform inversion results from 8 very wide aperture expanding spread profiles (ESPs), and 610 conventionally acquired common midpoint gathers are used to generate sediment property cross sections across the accretionary wedge deformation front, highlighting the vertical and lateral sediment property changes.