(688c) Reservoir Model Optimization Using the Seismic Response to CO2 Injection and Stochastic Inversion

Reservoir Model Optimization Using the Seismic Response to CO₂
Injection and Stochastic Inversion

Abelardo L. Ramirez¹, Kathy Dyer¹,
Donald J. White², Yue Hao¹,
James Johnson³

1-Lawrence Livermore National Laboratory

2-Geological Survey of Canada

3-Schlumberger-Doll Research Center

Abstract

4D reflection seismic data were used to map CO₂ migration during
Phase 1 of the Weyburn-Midale Project, while an extensive fluid sampling program
documented the geochemical evolution triggered by CO₂-brine-oil-mineral
interactions. Our goal is to use these data to optimize the reservoir model and
thereby improve site characterization and dependent predictions of long-term CO₂
storage in the Weyburn-Midale reservoir.

We have developed a stochastic inversion tool that identifies porosity/permeability
models that optimize agreement between the observed and predicted seismic
response. The tool explicitly integrates reservoir flow modeling, facies-based
geostatistical methods, and a novel stochastic inversion technique to identify
optimal porosity/permeability models. Reservoir model optimization is
accomplished through stepwise refinement of its permeability magnitude,
anisotropy, and heterogeneity.

Our algorithm starts
by proposing one reservoir model realization. Then, using a
multi-phase/multi-component flow simulator, we predict reservoir conditions
associated with CO₂/H₂O injection, HC/H₂O
withdrawal, and CO₂ migration. Using these conditions, we predict P-wave
velocities as a function of fluid-phase saturations, pressure, and porosity
using Gassmann's equation.  The velocity stack is used together with
calculated bulk densities to generate an intra-reservoir reflectivity series.
These series are convolved with an estimate of the source wavelet to predict
time-series waveforms that can be directly compared to observed seismic
response. The likelihood for each model is computed and used to decide whether
the proposed reservoir model produces acceptable agreement between predicted
and observed 4D seismic reflection data. Then, a new reservoir model is
proposed and the process is repeated until the process converges. We will
present the results from a synthetic data study to illustrate the method's
performance.

This work has been
performed under the auspices of the U.S. Department of Energy by Lawrence
Livermore National Laboratory under Contract DE-AC52-07NA27344.