Elsevier

Journal of Applied Geophysics

Volume 138, March 2017, Pages 185-197
Journal of Applied Geophysics

Comparison between staggered grid finite–volume and edge–based finite–element modelling of geophysical electromagnetic data on unstructured grids

https://doi.org/10.1016/j.jappgeo.2017.01.016Get rights and content

Highlights

  • Two finite–element and three finite–volume schemes for the modelling of geophysical electromagnetic data are compared.

  • The FE schemes use tetrahedral grids and the FV schemes use staggered tetrahedral–Voronoï grids.

  • The FE schemes are slightly more accurate but also more expensive than the FV schemes.

Abstract

This study compares two finite–element (FE) and three finite–volume (FV) schemes which use unstructured tetrahedral grids for the modelling of electromagnetic (EM) data. All these schemes belong to a group of differential methods where the electric field is defined along the edges of the elements. The FE and FV schemes are based on both the EM–field and the potential formulations of Maxwell's equations. The EM–field FE scheme uses edge–based (vector) basis functions while the potential FE scheme uses vector and scalar basis functions. All the FV schemes use staggered tetrahedral–Voronoï grids. Three examples are used for comparisons in terms of accuracy and in terms of the computation resources required by generic iterative and direct solvers for solving the problems. Two of these examples represent survey scenarios with electric and magnetic sources and the results are compared with those from the literature while the third example is a comparison against analytical solutions for an electric dipole source. Exactly the same mesh is used for all examples to allow for direct comparison of the various schemes. The results show that while the FE and FV schemes are comparable in terms of accuracy and computation resources, the FE schemes are slightly more accurate but also more expensive than the FV schemes.

Introduction

Common numerical methods use structured rectilinear grids for the modelling of geophysical electromagnetic (EM) data (e.g., Avdeev et al., 2002, Farquharson and Miensopust, 2011, Farquharson and Oldenburg, 2002, Fomenko and Mogi, 2002, Hursan and Zhdanov, 2002, Madden and Mackie, 1989, Mitsuhata and Uchida, 2004, Newman and Alumbaugh, 1995, Streich, 2009, Sugeng, 1998, Weiss and Constable, 2006, Weiss and Newman, 2002). While these methods are efficient in terms of accuracy and are simple to work with, they lack the flexibility required for representing arbitrary, complicated interfaces and for local refinement of the mesh. More flexibility is offered by the finite–volume (FV) and finite–element (FE) methods which can use unstructured grids. In unstructured grids, the facets of the elements conform to the irregular interfaces and, hence, avoid further refinements at these interfaces. These grids also allow local refinement of the mesh at observation points or at locations of high solution curvature (e.g., at the source in total–field approaches).

The FV method directly discretizes the integral form of the governing equations while the FE method approximates the solution of the weak form of a partial differential equation by minimizing an error function. Compared to the FE techniques, the FV methods are generally simpler in idea and more physically meaningful (see, e.g., Hirsch, 2007, Hermeline, 2009, Hermeline et al., 2008). The more rigorous mathematical background of the FE method, on the other hand, provides better control over the scheme. Moreover, unstructured FE schemes are usually more convenient for programming because, unlike FV discretizations, FE discretizations do not deal with complex geometrical structures in the mesh. For these reasons, the FE technique has been favored in geophysics by researchers who have worked with unstructured grids (e.g., Badea et al., 2001, Puzyrev et al., 2013, Schwarzbach et al., 2011, Um et al., 2012a, Um et al., 2012b).

Different classes of FE and FV methods have been used for solving EM problems on unstructured grids. Methods which solve for the electric and/or magnetic fields are called EM–field methods, and those which solve for the electric and magnetic potentials are commonly called potential (or A–ϕ) methods. (The main reason for using the potential method is the better conditioning of the problems in this method which suits iterative solvers.) FE methods use nodal–based and/or edge–based (vector) basis functions. Nodal basis functions were used by, e.g., Badea et al., 2001, Key and Ovall, 2011, Li and Key, 2007 and Puzyrev et al. (2013). Nédéléc (1980) introduced vector basis functions which were used by, e.g., Börner et al., 2008, Schwarzbach et al., 2011, Um et al., 2010 and Um et al. (2013). These basis functions have the advantage of allowing the discontinuity of the normal component of electric field at conductivity contrasts, and also of being divergence free (inside source/charge free elements).

There are also different classes of FV methods for unstructured grids. In the cell–centred FV methods, all the fields are co–located at the centres of the control–volumes (e.g., Remaki, 2000, Shankar et al., 1990). In the staggered methods, the unknowns are located at separate locations: Madsen and Ziolkowski (1990) employed a staggered grid in which the full 3–D vectors of electric and magnetic fields are defined on the edges of the primary and dual cells, respectively (Weiss, 2010, and Jahandari and Farquharson, 2014, Jahandari and Farquharson, 2015, belong to this category). Yee and Chen (Yee and Chen, 1994, Yee and Chen, 1997) proposed a slightly different approach by locating the electric and magnetic fields at the vertices of primary and dual cells. In terms of the location of the unknown fields, cell–centred FV schemes can be seen as counterparts of nodal–based FE schemes (e.g., a cell–centred FV scheme with Voronoï cells as control–volumes corresponds to a nodal–based FE scheme on a tetrahedral mesh), and staggered FV schemes as equivalents of edge–based FE schemes.

During recent years, we developed several FV and FE schemes for the modelling of EM data on tetrahedral grids (Ansari and Farquharson, 2014, Jahandari and Farquharson, 2014, Jahandari and Farquharson, 2015). The fact that all these schemes belong to classes of FE and FV techniques where the unknown vector fields are defined at the edges of the elements motivated the comparison of these schemes. The EM–field FE scheme uses vector basis functions while the potential FE scheme uses both vector and scalar basis functions (Ansari, 2014). The EM–field and potential FV schemes all use staggered tetrahedral–Voronoï grids (Jahandari and Farquharson, 2014, Jahandari and Farquharson, 2015). In the EM–field FV and FE schemes, the unknown fields are defined along the edges of the tetrahedra while in the potential FV and FE schemes, the vector and scalar potentials are defined along the edges and at the nodes of the tetrahedra, respectively. There is only one FE potential scheme while there are two potential FV schemes (gauged and ungauged). Therefore, there are five frequency–domain total–field schemes (two FE and three FV). In the following sections, the theoretical background of the methods is briefly described and then three examples are used for comparing the schemes in terms of accuracy and in terms of computational resources required by generic direct and iterative solvers for solving the problems.

Section snippets

Theory

In the direct EM–field method the problem is formulated in terms of electric and magnetic fields (E and H) while in the potential method, these fields are replaced by their representations in terms of magnetic vector potential and electric scalar potential (A and ϕ): E=iωAϕ×A=μH,where ω is angular frequency, i is the imaginary unit, μ is magnetic permeability and a time–dependence of eiωt is assumed (see, e.g., Griffiths, 1999, Ward and Hohmann, 1988).

In low–frequency geophysical

Examples

This section includes three examples: the first two examples represent benchmark models in which the responses of halfspaces with anomalous blocks due to magnetic and electric sources were calculated. In these examples, the solutions from different schemes are compared with each other and with results from the literature. In the third example, the source is an electric dipole on a homogeneous halfspace and the numerical results are compared with analytical solutions. The MUMPS sparse direct

Discussion and conclusions

In this study, we compared two FE and three FV schemes which are used for the modelling of EM data on unstructured grids. The reason for this comparison was the similarity between these methods in terms of spatial discretization. In the two E–field FV and FE schemes, the electric field was defined at the edges of the tetrahedra while in the three potential FV and FE schemes the scalar and vector potentials were defined at the vertices and at the edges of the tetrahedra, respectively. The FV

Acknowledgements

The authors are grateful for the financial support provided by the Natural Sciences and Engineering Research Council of Canada (NSERC), the Hibernia Geophysics Support Fund, and the School of Graduate Studies at Memorial University.

References (44)

  • C.G. Farquharson et al.

    Comparison of integral equation and physical scale modeling of the electromagnetic responses of models with large conductivity contrasts

    Geophysics

    (2006)
  • C.G. Farquharson et al.

    An integral-equation solution to the geophysical electromagnetic forward-modelling problem

  • E.Y. Fomenko et al.

    A New Computation Method for a Staggered Grid of 3-D EM Field Conservative Modeling

    Earth, Planets and Space

    (2002)
  • D.J. Griffiths

    Introduction to Electrodynamics, Third Edition

    (1999)
  • C. Hirsch

    Numerical Computation of Internal and External Flows: The Fundamentals of Computational Fluid Dynamics (2nd ed.)

    (2007)
  • G. Hursan et al.

    Contraction integral equation method in three-dimensional electromagnetic modeling

    Radio Sci.

    (2002)
  • H. Jahandari et al.

    A finite-volume solution to the geophysical electromagnetic forward problem using unstructured grids

    Geophysics

    (2014)
  • H. Jahandari et al.

    Finite-volume modelling of geophysical electromagnetic data on unstructured grids using potentials

    Geophys. J. Int.

    (2015)
  • K. Key et al.

    A parallel goal-oriented adaptive finite element method for 2.5-D electromagnetic modelling

    Geophys. J. Int.

    (2011)
  • Y. Li et al.

    2D marine controlled-source electromagnetic modeling: part 1 - an adaptive finite element algorithm

    Geophysics

    (2007)
  • T.R. Madden et al.

    Three-dimensional magnetotelluric modeling and inversion

    Proc. IEEE

    (1989)
  • N.K. Madsen et al.

    A three-dimensional modified finite volume technique for Maxwell's equations

    Electromagnetics

    (1990)
  • Cited by (35)

    • A hybrid spectral element-infinite element approach for 3D controlled-source electromagnetic modeling

      2022, Journal of Applied Geophysics
      Citation Excerpt :

      A fast and high-precision 3D forward modeling algorithm is the important engine for 3D inversion. Typically, the mainstream 3D geoelectromagnetic numerical simulation methods include integral equation method (Xiong, 1992; Kuvshinov et al., 2002; Zhdanov et al., 2006), finite volume method (Jahandari and Farquharson, 2015; Jahandari et al., 2017; Lu and Farquharson, 2020), finite difference method (Haber et al., 2000; Weiss and Newman, 2003), and finite element method (Tang et al., 2010; Li et al., 2012; Ren et al., 2013, 2014; Cai et al., 2014; Grayver, 2015; Yin et al., 2016; Liu et al., 2018). Spectral element method is a high-precision numerical simulation method developed from the combination of finite element method and spectral method.

    • 3D magnetotelluric modeling using high-order tetrahedral Nédélec elements on massively parallel computing platforms

      2022, Computers and Geosciences
      Citation Excerpt :

      Given its reduced implementation effort, the FV method has recently been employed for 3D MT modeling, either for 2D problems on arbitrary topographies (e.g., Du et al., 2016) or 3D Earth models with general anisotropy (e.g., Guo et al., 2020). But, although it supports unstructured grids, the accuracy of FV solutions is, in general, inferior to FE computations when comparing meshes with similar characteristics (Bondeson et al., 2012; Jahandari et al., 2017). The rest of the paper is organized as follows.

    • A focused and constrained 2D inversion of potential field geophysical data through Delaunay triangulation, a case study for iron-bearing targeting at the Shavaz deposit in Iran

      2020, Physics of the Earth and Planetary Interiors
      Citation Excerpt :

      In this context to tackle these issues, a complex geological model can be represented using triangular meshes, which is difficult to make using the common discretization approach. Nowadays, researches are being undertaken in discretizing physical model domain through the unstructured grids, as these can better approximate the geological bodies using less cells and reconstruct any arbitrary topographic relief (Jahandari and Farquharson, 2015; Ren et al., 2017; Jahandari et al., 2017). A novel strategy is the non-structured triangulation mesh such as the Delaunay triangulation as an active area of research in the geophysical community (Liu et al., 2015a; Jahandari and Farquharson, 2015; Singh, 2020).

    • Modular finite volume approach for 3D magnetotelluric modeling of the Earth medium with general anisotropy

      2020, Physics of the Earth and Planetary Interiors
      Citation Excerpt :

      In addition, tensor conductivity representing anisotropy may also substantially increase the number of non-zero elements when assembling the FEM coefficient matrix. Thus, theoretically, FVM can combine the advantages of FEM in mesh flexibility and superior accuracy compared to FDM, and the advantages of FDM in straightforward implementation and economic computation load compared to FEM (Jahandari and Farquharson, 2014; Jahandari et al., 2017; Rochlitz et al., 2019). In light of above-mentioned justifications, along with the facts that physical meaning intuitively originates from the integral representations of differential operators, and straightforward implementation could be partly achieved by using regular meshes, we present a new staggered-grid finite volume approximation of MT responses with general anisotropy for the 3D Earth, which solves electric fields tangential to cell edges and discretizes integral form of frequency-domain Maxwell's equations.

    View all citing articles on Scopus
    View full text