Seismic slope estimation: analyses in 2D/3D

Abstract

Slope estimation is a critical step for many post-processing seismic image techniques. Accurate slope images allow for automatic interpretation techniques to effectively and efficiently follow seismic horizons, and identify structurally discontinuous features with little to no information from the interpreter. However, accurately estimating slope, while simultaneously mitigating slope discontinuities caused by noise, is difficult. The structure tensor method estimates slope from local structure within ellipsoids whose half-widths are specified by the user. This method performs well for seismic images with highly variable structure and computes slope fastest among three slope estimation methods analyzed in this thesis. Although, no slope derivative constraints exist, which can produce slope discontinuities that are caused by noise. The plane-wave destructor method solves a non-linear optimization problem using the Gauss-Newton method to estimate slope. This method has an optional input parameter for initial slope, which can contain valuable information. Yet, the smoothing regularization is performed on each slope perturbation and not the slope, thereby allowing slope discontinuities from the initial slope image to persist through iterations. The smooth dynamic warping method, proposed in this thesis, estimates slope by finding a globally optimal shift solution. This method is the first slope estimation method to constrain slope derivatives, preventing slope discontinuities caused by noise. However, some parameter choices may significantly increase computational time or memory requirements. Through qualitative and quantitative analyses of 2D and 3D real and synthetic seismic images, I identify the advantages and disadvantages between three slope estimation methods.