Introduction

Flattening algorithms Lomask et al. (2005); Lomask and Claerbout (2002); Lomask (2003) are able to flatten seismic data cubes with faults by summing the dips into time shifts around the faults and ignoring the dips across the faults. In order to know which dips to ignore and which dips to honor during inversion, we require a fault indicator (data residuals weight that throw away bad dips). This indicator could be either manually picked or automatically generated by an automatic fault detector.

While the smoothness of the summed time shifts are often justifiable in non-faulted areas, the time shifts can change abruptly across the faults. In these situations, we desire an inversion technique that yields smooth time shifts in non-faulted areas while preserving sharp time shifts across the faults. In addition, no pre-defined fault indicator should be supplied.

In this paper, we present an automatic edge-preserving method for flattening faulted data without requiring an input fault model. The method uses iterative re-weighted least-squares (IRLS). A Geman-McClure weight function (fault indicator) of data residuals is computed at each non-linear iteration to allow outliers in the data residuals.

The only requirements are that part of the fault tip-lines are encased in the data and that the faults are oriented vertically. The resulting weight generated by this IRLS method is a fault indicator cube that best flattens the data. This is an important difference from many traditional automatic fault detectors that are defined by local discontinuities.