Stochastic fault network modeling, honoring both statistical parameters and hard data.

@inproceedings{CherpeauGM2010,
 abstract = { Fault network modeling is a crucial step in 3D reservoir modeling. Indeed, faults control the spatial layout of reservoir formations and behave as barriers or conduits for fluid flows, hence play a major role in reservoir oil-in-place (OIP) estimation and compartmentalization.
During the resource exploration phase, or when a few data are available, lots of uncertainties may exist concerning fault geometry and connections. Typically, a 3D structure is inferred from a few 2D seismic lines and wells. To reduce the risk of using only one deterministic model, we propose to sample this uncertainty space by simulating stochastic fault networks, honoring both statistical parameters (i.e. fault families and their relative ages, orientation, size distributions) as well as hard data. Stochastic data point clustering enables to better sample the uncertainty space by considering the uncertainty of the association of data points to possibly different fault surfaces.
A fault object is represented in an implicit manner, i.e. by an isovalue of a scalar field, within a domain of interest of the total modeling volume. A binary tree, similar to a Binary Space Partition tree (BSP tree), is used to describe the spatial layout of the faults in the volume of interest, each fault representing a node of the tree, its two children nodes being the spatial regions on both sides of the fault. Fault connections, i.e. the topology of the fault array, are inferred from the binary tree, and specific care is taken to represent synsedimentary and laterally terminating faults.
A case study, based on the interpretation of 2D seismic lines, illustrates the method and enables to compare the results to the deterministic model built from a full 3D seismic survey. },
 author = { Cherpeau, Nicolas AND Caumon, Guillaume AND Levy, Bruno },
 booktitle = { Proc. 30th Gocad Meeting, Nancy },
 title = { Stochastic fault network modeling, honoring both statistical parameters and hard data. },
 year = { 2010 }
}