Hugoton Geomodel: a Hybrid Stochastic-Deterministic Approach
Bohling, Geoffrey C.1,
Martin K. Dubois1, Alan P. Byrnes1 (1) Kansas Geological Survey, University of Kansas, Lawrence, KS
Development of the geomodel
for the 170-m thick, 16,000 km2 Permian-age Hugoton giant gas field in
southwest Kansas involved various
modeling methodologies on the continuum from stochastic to deterministic.
Significant differences in lithofacies-specific
permeability-porosity and capillary pressure properties and inability to
determine saturations from wireline logs because of
deep filtrate invasion prompted development of a lithofacies-based
matrix properties model. The 108-million cell geomodel
was developed by: 1) defining lithofacies in wells
with neural network models trained on core lithofacies-to-log
correlations and estimating porosity from wireline
logs at 1600 node wells; 2) modeling between wells using sequential indicator
simulation (SIS) for lithofacies and sequential
Gaussian simulation (SGS) for porosity; and 3) calculating permeability,
capillary pressure, and relative permeability for each lithofacies-porosity
combination using empirical transforms with water saturation calculated using
the lithofacies/porosity-specific capillary pressure
and a location-specific height above free-water level. Although the SIS and SGS
processes are both stochastic, the large horizontal ranges of the estimated variograms lead to multiple realizations that are nearly
identical and approaches deterministic prediction. The neural network lithofacies predictions at node wells are also
probabilistic. Aspects of the geomodel were validated
using reservoir flow simulations at the well and multi-well, multi-section
scales. The Hugoton geomodel workflow illustrates the
continuum between stochastic and deterministic modeling and the dependence of
the methodology used for each property on the available data, the scale of
prediction, and the order (predictability) of the system relative to the
property being modeled.