Classifying basin-scale stratigraphic geometries from subsurface formation tops using K-nearest neighbor classification

Abstract

Stratal geometries are the generalized shape of stratigraphic packages at scales ranging from meters to kilometers in thickness. Large-scale stratal geometries (100-1,000 m) in sedimentary basins document the complex interplay of sediment supply and deposition, subsidence, and topography. Despite the historical use of stratigraphic geometries to predict hydrocarbon reservoirs at the field scale, there has been limited work to couple stratigraphic predictions and machine learning algorithms. Herein, we present a machine-learning model used to classify basin-scale stratigraphic geometries from subsurface formation tops. The classifier was trained on a conceptual geometric model and validated on real-world subsurface data. The classification model correctly predicts the distribution of stratigraphic geometries of the Upper Cretaceous and Paleocene strata of the Eastern Greater Green River Basin in south-central Wyoming. Results from this model include documenting new areas of Paleocene age uplift and erosion in Wyoming’s Eastern Greater Green River Basin that might have gone unnoticed using traditional structure and isochore maps. The classification model also documents areas of uplift and truncation surrounding areas of onlap, which has the potential to affect the sealing mechanism, net-sand content, and in-place natural gas volumes in these fluvial reservoirs. Furthermore, I discuss how this classifier is useful to guide geologic interpretation of spatial changes in stratigraphic geometries.

AAPG Datapages/Search and Discovery Article #90372 © 2020 AAPG Pacific Section Convention, 2020 Vision: Producing the Future, Mandalay Beach, Oxnard, CA, April 4-8, 2020 (Cancelled)