Numerical homogenization for seismic wave propagation in 3D geological media

Abstract:

Despite the important increase of the computational power in the last decades, simulating the seismic wave propagation through realistic geological models is still a challenge. By realistic models we here mean 3D media in which a broad variety (in terms of amplitude and extent) of heterogeneities lies, including discontinuities with complex geometry such as faulted and folded horizons, intrusive geological contacts and fault systems. To perform accurate numerical simulations, these discontinuities require complicated meshes which usually contain extremely small elements, yielding large, sometimes prohibitive, computation costs. Fortunately, the recent development of the non-periodic homogenization technique now enables to overcome this problem by computing smooth equivalent models for which a coarse mesh is sufficient to get an accurate wavefield.

In this work, we present an efficient implementation of the technique which now allows for the homogenization of large 3D geological models. This implementation relies on a tetrahedral finite-element solution of the elasto-static equation behind the homogenization problem. Because this equation is time-independent, solving it is numerically cheaper than solving the wave equation, but it nevertheless requires some care because of the large size of the stiffness matrix arising from the fine mesh of realistic geological structures. A domain decomposition is therefore adopted. In our strategy, the obtained sub-domains overlap but they are independent so the solution within each of them can be computed either in series or in parallel. In addition, well-balanced loads, efficient search algorithms and multithreading are implemented to speed up the computation. The resulting code enables the homogenization of 3D elastic media in a time that is neglectable with respect to the simulation time of the wave propagation within. This is illustrated through a sub-surface model of the Furfooz karstic region, Belgium.

Back to: General Contributions in Seismology II

<< Previous Abstract | Next Abstract >>