A hierarchical sampler for multi-scenario interpretation of structural observations constrained by geological rules.
in: 2021 RING Meeting, ASGA
Abstract
Recently, Godefroy et al. (2021) proposed a graph-based method to assess the uncertainty arising when associating incomplete fault observations on sparse datasets. Their sampler sequentially draws fault surfaces from a global graph representing possibilities for associating observations using a notion of fault importance represented by maximal cliques of the graph. This generates a set of possible association scenarios. As a result of this sampler, independent realizations are obtained, and it is not obvious to determine whether two output realizations are the same, or whether they are close one from another. To address this challenge, we propose a new hierarchical sampler which sequentially insert the simulated faults on a tree data structure. This strategy allows for an adaptive exploration of the search space by generating first order fault scenarios insert in a tree data structure, then child scenarios corresponding to various realizations of smaller faults. This method also allows for updating conditional probabilities while generating at each tree node a fault knowing that another was previously drawn. The proposed approach is implemented in C++ in the FaultMod2 plugin of SKUA-GOCAD. We demonstrate the approach on a synthetic Santos Basin case study, which shows that distant realizations in the tree are visually different. This methodology opens interesting perspectives for solving inverse problems in the presence of sparse structural observations.
Download / Links
BibTeX Reference
@inproceedings{TATYMOUKATI_RM2021,
abstract = { Recently, Godefroy et al. (2021) proposed a graph-based method to assess the uncertainty arising when associating incomplete fault observations on sparse datasets. Their sampler sequentially draws fault surfaces from a global graph representing possibilities for associating observations using a notion of fault importance represented by maximal cliques of the graph. This generates a set of possible association scenarios. As a result of this sampler, independent realizations are obtained, and it is not obvious to determine whether two output realizations are the same, or whether they are close one from another. To address this challenge, we propose a new hierarchical sampler which sequentially insert the simulated faults on a tree data structure. This strategy allows for an adaptive exploration of the search space by generating first order fault scenarios insert in a tree data structure, then child scenarios corresponding to various realizations of smaller faults. This method also allows for updating conditional probabilities while generating at each tree node a fault knowing that another was previously drawn. The proposed approach is implemented in C++ in the FaultMod2 plugin of SKUA-GOCAD. We demonstrate the approach on a synthetic Santos Basin case study, which shows that distant realizations in the tree are visually different. This methodology opens interesting perspectives for solving inverse problems in the presence of sparse structural observations. },
author = { Taty-Moukati, Fabrice AND Caumon, Guillaume AND Antoine, Christophe AND Stoica, Radu-Stefan },
booktitle = { 2021 RING Meeting },
publisher = { ASGA },
title = { A hierarchical sampler for multi-scenario interpretation of structural observations constrained by geological rules. },
year = { 2021 }
}