HR: 0800h
AN: U31A-0001    [Abstracts]
TI: Integration of Multiresolution Data using Geostatistics, Stratigraphy, and Seismic Inversions to Build Realistic Flow Simulation Models
AU: Kalla, S
EM: skalla2@lsu.edu
AF: Louisiana State University, 3516 Taylor Hall, Baton Rouge, LA 70803, United States
AU: * White, C D
EM: cdwhite@lsu.edu
AF: Louisiana State University, 3516 Taylor Hall, Baton Rouge, LA 70803, United States
AU: Gunning, J
EM: james.gunning@csiro.au
AF: CSIRO, Bayview Avenue, Clayton, Vic 3150, Australia
AU: Glinsky, M E
EM: Michael.E.Glinsky@bhpbilliton.com
AF: BHP Billiton, 1360 Post Oak Blvd Suite 150, Houston, TX 77056, United States
AB: Flow models commonly require high-resolution (~ 1 m vertical) grids of properties. Seismic data are areally dense, but their vertical resolution (~ 10 m) may be too coarse for flow models. We propose two methods to downscale seismic data to flow models. Both methods combine stochastic seismic inversion (for interval sum constraints), geologic modeling (for stratigraphic frameworks), rock physics (to interrelate acoustic and flow properties), and geostatistics (for spatial correlation); both methods also match well data. Stochastic ensembles of geomodels capture variability in a Bayesian framework. A Markov Chain Monte Carlo downscaling algorithm provides downscaled models for net thickness, gross thickness, and porosity. The fine- scale models have nonuniform stratigraphy with rich layer truncation or "pinchout" behavior. The cascading method uses realizations from a stochastic seismic inversion as exact constraints. Many seismic inversion realizations are used to sample seismic uncertainty; each is downscaled: thus, the workflow cascades. The constraints propagate rock physics and covariances into the flow model. This problem is challenging because seismic constraints confine sampling to a constraint hypersurface. The inexact constraint method uses means and variances estimated from an ensemble of seismic inversions. The downscaled models weight well and seismic data by their covariances. Each geomodel appropriately but only approximately matches all data, avoiding overtuning to inexact data. Because of pinchouts, the likelihood is nonlinear (but piecewise linear), complicating sampling and motivating use of auxiliary variables for this difficult configurational problem. Both downscaling methods are demonstrated for an oil field in offshore West Australia. Cornerpoint flow models are constructed, and screening and simulation studies are discussed. Three-dimensional flow simulations illustrate stratigraphic variability and flow behavior: prior models for geologic continuity are statistically significant and stochastic fluctuations, although smaller, are not negligible.
DE: 0520 Data analysis: algorithms and implementation
DE: 0545 Modeling (4255)
DE: 0902 Computational methods: seismic
DE: 1873 Uncertainty assessment (3275)
SC: Union [U]
MN: 2007 Fall Meeting