Geologically Realistic Fluvial Point Bar Geocellular Models: Conditioning Algorithms With Outcrop Statistics

Abstract

Geostatistical reservoir modeling necessitates assumptions that often over-simplify heterogeneity, causing realizations to belie complex geologic reality. Outcrop studies provide robust suites of subseismic-scale (bed- to geobody-scale) statistics that characterize internal geobody architecture (e.g., bed thicknesses and lengths, grain size distributions). However, direct integration of such data into subsurface modeling workflows that capture geological complexity, remains challenging. This study: 1) presents bed-scale statistics from Late Cretaceous Horseshoe Canyon Formation fluvial point bar deposits that outcrop in southeastern Alberta; 2) investigates multiple modeling methodologies to test how different algorithms can integrate such statistics into reservoir modeling workflows; and (3) generates models that retain the geological essence of the outcrop deposits.

Stratigraphic sections (n = 40) record grain size, sedimentary structures, and bedding characteristics, providing the basis for reservoir model facies classification. Differential GPS surveys delineate individual lateral accretion packages (LAPs), capturing the stratigraphic framework for modeling outcrop architecture. Vertical and horizontal facies proportions and transition probabilities are calculated from measured sections, and constrain the probability that a facies is present at specific stratigraphic positions within each LAP. Bed-scale correlations are crucial for robust characterization of outcrop heterogeneity with such statistics. Probability cubes derived from these outcrop statistics guide simulations. Nested Truncated Gaussian Simulation (nTGS) is compared with Plurigaussian Simulation (PGS). nTGS realizations produce geologically-realistic outcomes with a small loss of fidelity to input facies proportions. PGS realizations produce less-geologically-realistic realizations, but better-preserve global facies proportions. Neither reproduce the LAP internal architecture observed in outcrop. Training images derived from outcrop statistics for multiple-point simulation achieve what nTGS and PGS could not, generating models with a better visual match to the outcrop that honor input statistics. This study presents a pathway to directly constrain models to measured outcrop statistics while also reproducing visual outcrop heterogeneity. As such, flow simulations on such models will test the flow-impacts of outcrop-derived architecture rather than purely stochastic heterogeneity.

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