Automated Salt Recognition in 2-D Seismic and Mapping Basin-Wide Salt in the Gulf of Mexico

Abstract

Seismic technology’s advance over the last three decades, particularly the transition from 2D to 3D, has seemingly rendered enormous volumes of vintage 2D data obsolete. However, innovative methods applied to these legacy assets provide paths for extraction of valuable new insights at very low marginal costs. Here, automated pattern recognition techniques were utilized in the analysis of a set of more than 7,000 2D post-stack, migrated SEG-Y lines, covering approximately 400,000 km² (154,441 mi²) across the main petroliferous area of the Gulf of Mexico, producing a basin-wide map of top of salt.

Two analytic approaches were taken to the unsupervised machine identification of salt in old (pre-1992) seismic time sections. The first, based on image texture analysis, assigned a salt-likelihood score to each pixel in a seismic section through application of a grey-level co-occurrence matrix and statistics on the resultant rasters. Several texture statistics were found efficient for differentiating salt and non-salt regions. The second, based on vector analysis of reflectors extracted from seismic images, discriminated salt/non-salt based on the densities of low-dip and high-dip reflectors.

These measures were evaluated, without interpreter intervention, and combined to estimate salt/non-salt boundaries in sections with salt and to identify sections with no salt. Levels of confidence in the boundaries were statistically estimated to convey certainty in the existence and location of the salt boundaries.

Final estimates of the top of salt were contoured in time. Based on a data set of more than 3,000 velocity surveys, the time contours were transformed to depth. A 3D surface was estimated on the depth map so the salt surface could be visualized and further studied in a basin-wide 3D geographic information system in which other geologic, geophysical, production and facility data were available.

Processing each line took several minutes of computer time. However, the analysis can be easily parallelized, reducing the computational impediment to batch machine analysis of thousands of lines.

AAPG Datapages/Search and Discovery Article #90323 ©2018 AAPG Annual Convention and Exhibition, Salt Lake City, Utah, May 20-23, 2018