Machine-learning-assisted seismic prediction of massive (up to 500 m) sands and their geologic meaning: Lower wilcox intraslope basins, south-central Texas, USA

Abstract

We studied deep (up to 6000 m) Lower Wilcox strata (Guadalupe A) immediately above the Midway shale using a 1000-km2 3D seismic volume and sparse six deep-well, wireline-log data in south-central Texas. Although some of the wells encounter sandstones collectively as thick as 500 m, traditional seismic attributes and seismic inversion are unable to reveal the sign and extent of the massive sand distribution in the intraslope minibasins. A new workflow that incorporates machine learning into seismic lithology and geomorphology was tested for quantitative mapping of sand-content sandstone thickness. The procedure significantly updates current workflows for better geologic and reservoir prediction with higher resolution and accuracy. In addition to good fit to impedance in training wells, results tied gamma-ray and sand-content curves at a blind-test well, ensuring high confidence in stratigraphic and depositional interpretation. The study recognized multiple thick, sandy intraslope minibasins in a 30-km-wide upper slope. Clinoforms on the shelf, erosional channels at the shelf edge, sandy channels, and fans in intraslope minibasins were found to support the interpretation of a unique sediment-disposal system triggered by hyperpycnal flows. Existence of massive sand accumulations in the upper slope zone in the Gulf Coast suggest the existence of another major mechanism of large sand transport during the early Wilcox in a high sea-level stand and possibly a future deep gas play in the area.