Workflow for the integration of a realistic 3D geomodel in process simulations
using different cell types and advanced scientific visualization: Variations on
a synthetic salt diapir

Ines Görz, Martin Herbst, Jana H. Börner, Björn Zehner

PII: S0040-1951(17)30020-3
DOI: doi:10.1016/j.tecto.2017.01.011
Reference: TECTO 127383

To appear in: Tectonophysics

Received date: 13 July 2016

Revised date: 8 January 2017
Accepted date: 13 January 2017

Please cite this article as: Görz, Ines, Herbst, Martin, Börner, Jana H., Zehner, Björn,
Workflow for the integration of a realistic 3D geomodel in process simulations using
different cell types and advanced scientific visualization: Variations on a synthetic salt
diapir, Tectonophysics (2017), doi:10.1016/j.tecto.2017.01.011

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Workflow for the integration of a realistic 3D geomodel in process simulations using

different cell types and advanced scientific visualization: Variations on a synthetic salt
diapir

Ines Görz1, Martin Herbst1, Jana H. Börner1, Björn Zehner2

1)
TU Bergakademie Freiberg, Faculty of Geosciences, Geoengineering and Mining, Gustav-Zeuner-

PT
Straße 12, 09599 Freiberg, Germany
2)
Federal Institute for Geosciences and Natural Resources (BGR), Wilhelmstraße 25-30,
13593 Berlin, Germany

RI
Abstract

SC
The purpose of this study is to use one complex geological 3D model for numerical simulations of
various physical processes in process-specific simulation software. To do this, the 3D model has to be
discretized according to different cell types, depending on the requirements of the simulation
method. We used a salt structure with a diapir and its deformed host rock to produce two 3D models

NU
describing the boundary surfaces of the structure: one very simplified model consisting of cuboid
surfaces and a realistic model consisting of irregular boundary surfaces. We provide a workflow for
how to generate hexahedral, tetrahedral and spherical volume representations of these two
MA
geometries. We utilized the volume representations to simulate temperature, displacement and
transient electromagnetic fields. We can show that the simulation results closely reflect the input
geometry and that it is worth the effort to produce geometric models that are as realistic as possible.
Additionally, we provide a workflow for simultaneous visualization and analysis of the simulation
results. Scientific visualization is an important tool for deriving knowledge from complex
D

investigations.
TE

Keywords: geological 3D modeling, structured grid, tetrahedral mesh, distinct elements,

electromagnetic monitoring, displacement and thermal fields, salt diapir
P

1. Introduction
CE

The geological subsurface is of utmost economic importance for the production of mineral
resources, drinking water, geothermal energy or even for traffic. Economic exploitation of the
subsurface reservoirs requires a careful risk assessment, e.g. to prevent perforations of barriers
between formations and mixing of aquifers (e.g., brines with fresh water) or the formation of
AC

sinkholes, landslides and structural damage around mines. The swelling of rocks like clay or anhydrite
may also cause land movements and damage to buildings.
The structure and physical properties of the subsurface must therefore be investigated in detail
before any commercial use in order to avoid hazards. The effect of human actions has to be
validated, taking into account superimposed or triggered geological processes. Exploitation requires
that the subsurface domain is monitored in order to detect tectonic movements, water ingress or
pollution and structural damage. These investigations usually include simulating the physics of
processes which control the structure and properties of the Earth's crust as preliminary studies or
monitoring tools.
To that end, numerical simulations are a powerful tool for understanding how geological
processes operate and which effects they cause. In addition to the physical laws applied, the
geometry and petrophysical properties of the geological bodies heavilly influence physical fields in
the subsurface as well as the simulation results. Therefore, the simulation of geological and physical
processes can only yield meaningful and feasible results, if all available information about the
structure and lithology of the modeling domain is included in a realistic model. Process modelers
often face practical problems when they want to set up a realistic detailed subsurface geometry,
since many software packages can only generate bodies with known geometry and topology and an
arbitrary amount of available data.

1
 ACCEPTED MANUSCRIPT

However, the geometry and topology of geological objects are complex and not known a priori
and have to be modeled from sparse data with heterogeneous spatial distribution. Specialized
geomodeling software and a large amount of work is necessary to generate feasible subsurface
models while respecting geological concepts and all geological data related to one geological object.
3D geomodeling was introduced by geological surveys and other administrative institutions as an
experimental complement to digital maps, but 3D models are increasingly replacing maps and cross-

PT
sections and are being established as standard communicational means in geology. Many European
states have started to document their subsurface in 3D models (van der Meulen et al., 2013; Alberta
Geological Survey, 2016, British Geological Survey, 2016; Geological Survey of Denmark and

RI
Greenland, 2016). Some of these 3D models have been used as geometry for simulations of fluid flow
in oil or groundwater reservoirs (Oladyshkin et al., 2007; Wycisk et al., 2009; Park et al., 2014), the
migration of CO2 at a sequestration site (Kempka et al., 2013), the temperature and heat flow (Kohl

SC
et al., 2003), seismic wavefields (Zehner et al., 2016), and electromagnetic fields (Börner et al., 2015).
These studies demonstrate the great potential that 3D models have for use in process simulations.
Modern geoscientific investigations have to deal with complex questions, e.g. how the quality of

NU
potable water resources is influenced by structural geology, raw material production and geo-
engineering activities. The investigation methods are very specialized and so are the process
simulations, which are based on different physical laws. The physical state variables are usually
described by partial differential equations, which are solved numerically by different solution
MA
methods like the Finite Difference Method (FDM), the Finite Element Method (FEM) or the Distinct
Element Method (DEM). These methods require of the modeling domain to be discretized in
different ways. If complex models with an irregular geometry have to be used for numerical
simulations, the creation of the model discretization is complicated and still under investigation
D

(Pellerin et al., 2014; Wieczorek et al., 2015; Watson et al., 2015; Zehner et al., 2015). Discretization
is a crucial element in pre-processing, since no numerical simulation is possible without it. In previous
TE

years, the coupling of simulation software packages has been performed for methods that use the
same cell type (Kolditz et al., 2012;
Kempka et al. 2013, Kempka and Kühn 2013, Börner et al., 2015). However, specific simulation
P

software that works with special physical laws and numerical methods often requires different
discretization types (Mallet, 2002; 2014; Royer et al., 2015). If geomodeling specialists want to use the
CE

power of the state-of-the-art geomodeling software and the geological knowledge provided in the 3D
subsurface models as a basis for complex investigations and predictions, they have to prepare, save
and transfer the 3D models in such a way that various types of model representations can be
AC

generated easily and flexibly, as required by the simulation software. If complex geoscientific
questions have to be addressed, one 3D model has to be discretized by various cell types and the
results of several calculations or simulations have to be combined, analyzed and interpreted in
common.
The objective of our study is to find workflows for the efficient integration of one realistic 3D
model into several software packages specially tailored to the simulation of one physical process.
This requires preparing geological 3D models for the needs of various numerical methods each of
which has been optimized for a special case and model representation. To test the workflows, one
structural model was used in two implementations: a simplified model of cuboid bodies and a
realistic one described by irregular boundary faces. For both we applied three methods which are
different in terms of their physics, numerical method and volumetric discretization:
• the simulation of a temperature field, using the finite difference method (FDM) on a regular
hexahedral grid;
• the simulation of stress state and displacement field using the distinct element method
(DEM) on an assembly of spherical particles;
• the simulation of a transient electromagnetic experiment with the finite element method
(FEM) on a tetrahedral mesh.
Comparing the simulation results for the simplified model with those for the realistic one, we
demonstrate the advantages of working with realistic complex subsurface models.
2
 ACCEPTED MANUSCRIPT

Finally, we show how to visualize all the modeling results in order to make combined queries,
analyses and interpretations of the simulation results generated by various methods and to detect
correlations or similarities, which is an important task of modern geoscientific investigation.

PT
RI
SC
NU
MA
D
P TE
CE
AC

3
 ACCEPTED MANUSCRIPT

2. Geological context and geometry description

The structure we used as a geometrical basis for this study represents a synthetic salt diapir
combining typical characteristics of salt bodies in the Central German Basin. Sedimentation started in
the Upper Permian with the deposition of salinar rocks, which were buried by marine and
terrigenous clastic successions. Due to the sedimentary overload, the salt evolved into diapirs which
pierced through the overlying sequences and dragged them upward (Stille, 1932, Ziegler, 1982). In

PT
this way, a complicated geometry evolved with a salt body consisting of overturned flanks and with
stratigraphic horizons comprising holes or ending inside of the investigation area. Some sedimentary
units thinned sharply due to their deformation. These structures are interesting from the perspective
of 3D modeling. However, the resulting models are simple enough that the simulation results can be

RI
evaluated intuitively.
The geological data comprise top markers of all sedimentary horizons and the borders of the

SC
diapir (Fig. 1a, b). From this data, two kinds of geometric descriptions were generated, since the
complexity of every geological model is controlled by geoscientists, their knowledge and the
software they work with. In order to compare the simulation results dependent upon the input

NU
geometry, we choose two levels of complexity to represent the geological structure:
• The “block model” describes the geological objects in terms of simple cuboids (Fig. 1c). The
diapir has a rectangular shape with vertical side walls; the stratigraphic units are horizontal and have
a constant thickness. This model is a rough approximation of the geological data and could be
MA
generated in a standard preprocessor that is often part of simulation software.
• The second model is a “realistic model” which describes the geological boundaries by
irregular, partly non-convex and overturned shapes (Fig. 1d). The dip and thickness of the
sedimentary units are variable; the diapir has overturned irregular side walls. The generation of such
D

a model requires a geomodeling software that can handle sparse irregular data and geological
concepts.
TE

3. Discretization of the models

3.1. Representation of geo-objects in 3D models
P

Structural 3D models describe the geometry and topology of geological objects while respecting
all data and knowledge referring to these objects. The geometry of the modeling domain can be
CE

described by different mathematical approaches and data structures.

One common way is to describe geological discontinuities by discrete polygonal meshes, e.g.
irregular triangulated networks as used in Skua-Gocad or GSI3D (Kessler et al., 2004, Paradigm,
AC

2011). If the spatial extent of a geological object is only described by its boundaries, this is called a
boundary representation (Weiler, 1988; Royer et al., 2015). A coherent boundary representation is
achieved when the volume of the modeling domain is completely confined and partitioned by
surfaces with identical triangle nodes at the contact lines of intersecting surfaces, such that no holes
and overlaps exist (Fig. 1e). In this case it can be unequivocally determined if the point lies inside a
geological object or not.
If the structural 3D model is to be used for process simulations, the full modeling volume has to
be tessellated. Volume discretization is achieved when the full volume of the 3D model is filled with
cells. Depending on the simulation method applied to solve the process equation numerically,
different cell types are used.
Structured grids are characterized by a regular topology and often consist of hexahedrons (Fig.
1f). They are stored in a raster format and are represented by the origin, the cell number and the
step width in each dimension. This grid type approximates geological discontinuities by staircase-like
boundaries, which means that a high-resolution grid is necessary in order to represent complex
geometries realistically. Structured hexahedral grids are often used by finite difference (FD) codes.
In order to achieve a representation of the subsurface that is closer to reality, the structured grids
can be deformed to follow the horizon geometry and the cells can become disconnected in order to
represent faults. However, these more advanced features must then be supported by the finite
difference codes (Bennis et al., 1996).

4
 ACCEPTED MANUSCRIPT

Unstructured discretizations consist of an irregular pattern of points with neither pre-defined

topology nor fixed cell geometry. They are usually described by a vector data format giving the
corner nodes of each cell and a topological model specifying the neighborhood relations of the cells.
The simplest type of unstructured discretization is a tetrahedral mesh which can adapt to
complicated geometries (Fig. 1g). Tetrahedral meshes are often used in finite element (FE) codes.
Distinct elements (DE) are another type of unstructured discretization, where the volume is

PT
discretized by unconnected cells. One way to generate this type of discretization is to fill the
modeling domain with spherical particles with a given center point and radius (Fig. 1h). A contact
tree serves as a topological model to describe the neighborhood of the elements (Cundall and Strack,
1979). This kind of discretization can be used to simulate the motion of discrete bodies by means of a

RI
distinct element method (DEM, Cundall and Strack, 1979).
The volume discretization usually has to be derived from surface-based representations as

SC
produced in many geomodeling software packages (e.g. Paradigm’s Skua-Gocad, Midland Valley’s 3D
Move, GSI3D from Insight GmbH and Schlumberger’s Petrel). For this task, special meshing software
is needed (Pellerin et al. 2011, Zehner 2011, Watson et al. 2015, Wieczoreck 2015). The conversion

NU
process from the surface-based to the volume-based representation is complicated and is still under
investigation. All existing methods have advantages, disadvantages and limitations regarding the
input and output models. MA
3.2. Method-specific tessellation of the models
The temperature, displacement and electromagnetic field simulations we want to perform on the
salt diapir models, require discretizations with hexahedral, spherical and tetrahedral cells
respectively (Fig. 2). Each has to be derived in two versions from the block and the realistic model.
D

The structural models were generated in Paradigm’s geomodeling software Skua-Gocad. They
consist of boundary surfaces that are, depending on the modeling workflow used, either non-
TE

conformable surfaces with overlaps and gaps at the surface contacts or conformable surfaces which
often have triangles with a poor quality (large ratio of side lengths) that may cause problems in some
discretization workflows. The 3D models are characterized by geometrical and topological
P

peculiarities which require a robust meshing tool.

First, the surface models were prepared by generating a coherent boundary representation with
CE

mainly equilateral triangles. This step is necessary to make the tetrahedral tessellation, since the
triangulation of the boundary surfaces is identical with the triangulation of the tetrahedrons. Only if
the triangle nodes at the surface contacts are identical is it possible to generate the tetrahedrons
AC

completely belonging to one geological unit, such that the geological boundaries are represented by
connected facets of the tetrahedral mesh. The quality of the boundary representation is also
important for this kind of discretization. Additionally, the coherent boundary representation was
used for the discretization by spherical elements, since the simulation software requires this. The
discretization with a hexahedral grid can also be performed without a coherent boundary
representation, but is more easily accomplished with it. Therefore, the coherent boundary
representation is a good basis to start from. To generate the conformable boundary representations,
various tools are available. We used the workflows described in Pellerin et al. (2011), Zehner (2011)
or Zehner et al. (2015) which interact with the external Graphite (ALICE, 2008) and Gmesh (Geuzaine
and Remacle, 2009) meshing software. Both workflows provided good results. While Tweedle is
faster, CompGeom allows for local mesh refinement and the inclusion of special points into the
surface mesh.
The conformable boundary representations were passed to various tools in order to obtain
volume discretizations with three cell types. The rectangular hexahedral grids were generated with
Skua-Gocad standard tools, but can also be produced with the workflows and software modules
described in Zehner et al. (2016), Watson et al. (2015), Wieczoreck et al. (2015) or Marschallinger
(2015).
The distinct element models can be discretized in the ITASCA PFC3D 5.0 software (Itasca, 2015)
from the structural models, if each unit is represented by a closed mesh. This means that contact

5
 ACCEPTED MANUSCRIPT

surfaces between two geological units have to be imported twice. The model domain was discretized
by spheres with specified radius distribution and density named as balls. We used different ball
radius distributions according to the geological layer thicknesses. The balls were generated with
overlaps to ensure the fast generation of the ball assembly. The resulting models were cycled to
generate ball assemblies with minimum overlaps which represent mechanical equilibrium and can be
used as initial model states.

PT
To make/carry out the electromagnetic simulation, a local mesh refinement is required at the
location of the electromagnetic source (e.g. Zehner et al. 2015). In order to generate the tetrahedral
mesh, we used the software Ansys (AnsysWorkbench, 2014). This software loads the boundary
representation and transforms it into NURBS, which are continuous mathematical functions.

RI
Therefore, the tetrahedral tessellation is not restricted to the position and shape of the triangles and
can be easily adapted locally.

SC
As an alternative, the software TetGen (Si, 2011) was tested to generate the tetrahedral mesh in
correspondence with the triangulation of the boundary representation. Using this method, the mesh
refinement is already part of the boundary representation which is exported to TetGen.

NU
As a result of performing the meshing workflow, three different discretizations of each of the two
structural models were obtained, where the material parameters were assigned averaged constant
values per geological unit (Table 1). Note the major contrast between the material parameters of the
salt and the clastic sediments.
MA
4. Simulation examples
4.1. Simulation of the temperature field using the finite difference method (FDM)
D

First, the 3D model with the hexahedral discretization was used to simulate a temperature field as
might be required to explore geothermal resources. The temperature field and its fluctuations are
TE

described by the heat conservation equation:

డ் డ௞డ் డ௞డ் డ௞డ்

∙∙ = + + +  , (1)
P

డ௧ డ௫ మ డ௬ మ డ௭ మ
CE

where
is the density,  the specific heat capacity, T the temperature, t the time, k the specific
heat conductivity, x, y, z the spatial dimensions and H the heat production.

The equation was solved with the finite difference method on a grid with constant step width in
AC

each coordinate direction after Gerya (2010) using the software Matlab and an implicit solution
scheme.
The temperature field was simulated for the static case, where for the left side of equation (1) the
temporal change of temperature is set to zero. A constant temperature and heat flux boundary
condition were set at the Earth’s surface and the model base respectively. A zero horizontal heat flux
condition was set at the lateral model boundaries.
Before interpreting the simulation results, the numerical solution has to be checked for accuracy
and reliability. One way of doing this is to compare with the analytical solution (Fig. 3a). This method
is not directly applicable to our study, since no analytical solution exists for the complex models we
generated here. We therefore applied homogeneous material to the grid, checked the solution for
varying heat productions on a grid with 8,000 cells and compared the numerical to the analytical
solution. In the case that the heat production is zero, the temperature-depth function is linear and
the numerical and analytical solutions are identical at all grid points. If heat production is applied, the
temperature-depth function is a parabola which is more curved, the larger the heat production is.
Since the finite difference solution is a piecewise linear approximation, the deviation of the
numerical solution from the analytical solutions is bigger, the greater the heat production is. For the
heat production applied in our model, a deviation of 0.5 °C is achieved at the final depth.
Additionally, we tested the numerical solution of grids with 1,000, 8,000 and 162,000 cells against
one another (Fig. 3b). A maximum temperature deviation of 3 °C was achieved between the grids,
6
 ACCEPTED MANUSCRIPT

such that we have to assume that the accuracy of the simulation results is valid within an uncertainty
range of ±3 °C.
One sensitive parameter that can cause large deviations in the simulation results is the specific
thermal conductivity. This parameter depends on the mineralogical composition, porosity, bulk
density and temperature and can vary by the factor two or three for any given rock type (Clauser and
Huenges, 1995). Since we work on a synthetic model, we checked the sensitivity of the temperature

PT
field to the temperature dependence of the thermal conductivity in order to validate the influence of
this parameter. Based on the analysis of experimental data, Sass et al. (1992) propose an empirical
formula to describe the relationship of the specific thermal conductivity and the temperature. We
used this formula and calculated the difference in the results with constant and temperature-

RI
dependent conductivities. Down to a depth of approximately 4000 m, the difference is smaller than
3 °C, which is within the uncertainty range of the method. At a greater depth, the difference

SC
increases up to 13 °C, which is a deviation of a maximum of 10 %. If detailed data about the
structure, mineralogical composition and porosity are available, the temperature dependency of the
thermal conductivity should therefore be taken into account as well. If no data are available, one

NU
should keep in mind that the modelling results may deviate from reality by more than 10 % due to
thermal conductivity. Consequently, the accuracy and reliability of the simulation result drop with
depth.
The modeling results show a temperature field which is increasing to a maximum temperature of
MA
110 °C at the model base. The highest temperatures occur at the southern model bottom of the
realistic models, where the salt unit is thinnest.
All models show a positive temperature anomaly above the salt diapir and a negative anomaly
below the salt diapir (Fig. 3c). Thus, the temperature difference inside the diapir is only 85 % in
D

comparison to the difference in the sedimentary host rock. This modeling result is feasible, since the
heat conductivity of the salt is bigger than that of the country rock, so that the heat is better
TE

transported from the Earth’s interior to the surface. At the head of the diapir, the heat conductivity
decreases, so that the heat cannot be transported to the groundsurface adequately and a positive
temperature anomaly forms.
P

By comparing the realistic models with the block models (Fig. 4), differences in the shape, spatial
extent and temperature differences in the anomalies can be seen. The block models have a broad
CE

anomaly with an angular shape at the top of the diapir. The peak temperatures are slightly higher
(5 °C) than in the realistic models (Fig. 3c). The thermal anomaly at the base of the salt is also broader
than in the realistic models, but the temperatures are smaller. The realistic models have an anomaly
AC

at the top of the diapir which is only half as big as the anomaly in the block models and is
characterized by an irregular shape.
One major difference between the block and realistic models can be seen in the shape of iso-
temperature surfaces which were extracted from the simulation results. Figs. 4e, f show the 10 °C-
iso-surface of the 30 x 60 x 90 model. The color codes the difference of the z-value and the median of
the z-value of the iso-surface. The 10 °C-iso-surface of the block model reflects the geometry of the
geological units. The global minimum of the z-values of the iso-surface can be found in the “sand”
unit and the global maximum above the salt diapir. The shape of the iso-surface is smooth and
rectangular.
The 10 °C-iso-surface of the realistic model has a local minimum of its z-values in the “sand” unit
and a local maximum where the “clay” unit is cropping out, which has a bigger heat production and
smaller heat conductivity than the “sand” unit. The local maximum clearly reflects the shape of the
“clay” unit. The global minimum can be found above the salt diapir and has an irregular shape. This
indicates that the geometry of the model heavily influences the simulation result.

4.2. Displacement field simulation with a distinct element method (DEM)

A simulation of the displacement field was performed with a distinct element method (DEM),
which is well suited to describing the behavior and the decoupled motion of material blocks. In this
study, we used the ITASCA PFC3D 5.0 software, which solves the Newtonian equation of motion for

7
 ACCEPTED MANUSCRIPT

each element and minimizes the resultant force for the whole ball assembly (Itasca 2015). The forces
are calculated from the overlap at the ball contacts in this study for a linear contact model. The
resultant contact force is the sum of the normal and shear forces at all contacts and equals the
external force:

௜ = ∑௜ ௖ே + ∑௜ ௖ௌ = ∑௜ ௖ே ∙ ௖ே ∙ ୧ − ∑௜ ௖ௌ ∙ ∆௖ௌ ∙

୧ = ௜ ∙ ( ప − ) , (2)

PT
where ௜ is the resultant particle force, ௖∗ the contact force, *=N the normal component, *=S the
shear component, n the normal unit vector, t the tangential unit vector, K is the contact stiffness, U

RI
the contact overlap,  ప the ball acceleration, g the gravity constant, m the ball mass.
A slip model is defined for every contact and allows slip to occur by limiting the shear force. The
slip criterion is controlled by the friction coefficient at each contact. In PFC3D, the interaction of

SC
particles is treated as a dynamic process and a state of equilibrium develops when all forces in the
model balance.
In our model, all geological units consist of ball assemblies with linear contact behaviour,

NU
because the model is supposed to represent a large volume and long term deformation. The
geodynamic behaviour of all geological units can therefore be assumed to be a very slow viscose or
granular flow with very large pieces (Ranalli, 1995).
Large-scale brittle rocks contain discontinuities where deformations of the whole unit can be
MA
accommodated. Here, the contacts between intact rock pieces can be modelled as linear contacts
with great friction and stiffness.
Salt has a visco-elastic material behaviour (Haupt, 1991) when typical geological strain rates are
considered. During tectonic processes, deformation velocities are mainly small and therefore the
D

strain rates are small too. If such a deformation regime continues for a geological time period, the
relaxation rate is small and the viscous material properties dominate. Consequently salt can be
TE

described as a Newtonian fluid, where a linear relationship exists between stress and strain (Hudec
and Jackson, 2007). The linear contact model is used as a simplification and first approximation of the
salt flow. This approach does not assign the correct viscous behaviour, but reproduces the flow of
P

the salt by the motion of a granular material. Small values for friction as well as stiffness were
assigned to the salt unit in order to model the strength contrast between salt and its host rock.
CE

The simulation was performed without applying external forces, neglecting the regional stress
field, so that only system-internal stresses and ball displacements are calculated (Fig. 5). Confined
traverse strain boundary conditions were applied at all boundaries except the groundsurface, where
AC

free motion in all directions is allowed (Fig. 5b). In addition to the forces, the stresses, ball
displacements and velocities were computed.
The accuracy of the simulated vertical stress can only be checked approximately against the
analytical solution for the lithostatic stress along vertical lines. Depending on the thickness and
density of the geological units, the lithostatic stress increases unit-wise in a linear fashion. Calculating
the stress state at given locations (Fig. 5a) allows the deviations of the numerical solution from the
analytical to be compared as shown in Fig. 5c. Deviations occur, since the ball assembly is not a
continuous medium and stress maxima or minima may emerge at individual ball contacts. The
precision of stress calculations is limited, since small stress calculation spheres only contain few
contacts and large stress calculation spheres may be larger than the thickness of one geological unit.
However, the trend of the stress development is correct. The differences between analytical and
numerical solutions are caused by the pressure of the confining particles since horizontal stress
components appear and cannot be neglected. Boundary effects are also more important here. The
stresses in the realistic model in part deviate more from the analytical solution of simple overburden
pressure than in the block model because of the irregular geometry.
In our model, the complex visco-elastic state equations were neglected. We therefore cannot
interpret the deformation style and velocities but can obtain information about the force balance in
the domain and evaluate whether the upwelling equilibrium is reached or not. Additionally, the

8
 ACCEPTED MANUSCRIPT

model results can be used to compare the block and the realistic model, which is one of the main
purposes of this study.
The investigation of the movement (displacement directions and magnitudes) is suitable for this
type of DEM analysis, because it can clarify whether the upwelling of the salt diapir is finished. Fig. 6
shows the vertical component of the final ball displacement under gravity. In both the block and the
realistic model, the upper parts of the salt diapir are almost mechanically equilibrated. The roof of

PT
the diapir is exposed to a small downward-directed ball displacement, such that the salt diapir does
not ascend anymore. In both models the base of the salt diapir is exposed to upward-directed ball
displacement, which is compensated by a downward-directed displacement of the host rock above
the diapir.

RI
The total displacement plot of the block model (Fig. 6) shows the transition from ascending to
settling displacement where the arrows meet. In the realistic model (Fig. 6), this transition in the

SC
total displacement cannot be seen so clearly, because the material is flowing mainly southward. This
effect can be explained by density difference which forms because the two dense host rock units
(limestone and claystone) are much thinner to the south of the diapir than to the north. This causes a

NU
stress difference. The downward motion at the margins of both models has to be interpreted as a
boundary effect.
One significant difference between the block and realistic models can be seen in the general
displacement pattern. The block model shows a symmetrical vertical displacement pattern with a
MA
buoyancy maximum in the salt diapir and maximum settlements near the groundsurface.
The realistic model yields an asymmetrical displacement field related to the joint upward motion
of the limestone and the salt. In both models, the simulation results reflect the input geometry. That
demonstrates the importance of the geometry and its control on the modeling result.
D

4.3. Simulation of a transient electromagnetic field with the finite element method (FEM)
TE

The structural 3D model can also be used for planning geophysical field surveys. For example, the
simulation of a transient electromagnetic experiment can be used to perform virtual measurements
with the aim of testing different source-receiver configurations with respect to coverage, resolution
P

and detectability.
The transient electromagnetic experiment works with a cable loop which is located at the Earth’s
CE

surface. At a given time t0 the direct current flowing through the cable loop is switched off, and as a
consequence the electromagnetic field decays. The resulting eddy current system propagates
downwards within the conductive subsurface in accordance with Faraday’s Law. The resulting
AC

transient electromagnetic fields can be observed at the surface or in boreholes. In order to describe
the propagation of the electromagnetic field, the curl-curl equation of the electric field describing the
inductive response in the time domain is used:

∇ ×  ିଵ ∇ ×  + ௧  = −௧ ௘ , (3)

where e is the time-dependent electric field, t is the time, µ the magnetic permeability, ௘ is the
electric source current density and  is the electric conductivity of the rock. The measured voltage
induced in the receiver coil can be calculated from the electrical field by

డ࢈
 = డ௧ = −∇ ×  . (4)

The weak formulation of equation 3 is solved with the finite element method on a tetrahedral
mesh (Afanasjew et al. 2013). This method requires the volume discretization of the modeling
domain by a tetrahedral mesh which can be unstructured in order to adapt to complex structural
situations.
An ASCII file containing the tetrahedral mesh and the material parameters was transferred to the
MATLAB based TEM software (Afanasjew et al., 2013). On all boundaries except the Earth’s surface,
the tangential component of the electric field was set to zero. At the Earth’s surface, an exact
9
 ACCEPTED MANUSCRIPT

boundary condition was implemented which avoids the discretization of the air half space (Goldman
et al., 1986).
In order to estimate the accuracy of the tetrahedral mesh (Spitzer and Wurmstich, 1999), we
checked the numerical solution for a homogeneous model against the analytical solution for a
homogeneous half-space (Nabighian, 1979). The electrical resistivities were set to a constant value of
80 Ωm. In Fig. 7, the analytical solution in terms of electric field and magnetic induction is plotted

PT
within a virtual vertical borehole at 500 m depth versus time. The transient electromagnetic fields
show the correct shape, sign and order of magnitude as the analytical solution. Consequently, we are
confident that all finite element solutions are reliable and accurate enough to yield meaningful

RI
results for the following discussion relating to the complete 3D models.
Finite element simulations of the transient electromagnetic fields were performed for both the
diapir and the block model. Fig. 9 shows the computational results at 5e-4 s and 1.3e-2 s after switch-

SC
off of the VMD both in terms of the x-component of the electric field and the z-component of the
time derivative of the magnetic induction. Additionally the differences between the different models
are shown.

NU
First of all, the transient electromagnetic fields show the expected behaviour. The electric field
reflects the predominantly horizontal, ring-shaped eddy-current system, which is surrounded by a
toroidal magnetic field. The fields react to the conductivity features of the subsurface as expected. It
is refracted at the corners of top horizons in the block model, while it is propagating smoothly in the
MA
realistic model. At greater depths, the electromagnetic fields are propagating in the host rocks and
avoid the salt dome due to its large resistivity. The electromagnetic field is distorted and deformed,
which may be explained by the low damping and an increased diffusion velocity in the resistive salt
rocks (Fig. 8).
D

This sensitivity to the conductivity distribution in the subsurface is the cause of the differences
between the numerical solutions for the two models, which become evident when computing the
TE

absolute difference between the field values (Fig. 9). At the early point in time (left column of Fig. 9)
the eddy-current system is located at relatively shallow depths and consequently reacts to the
geological structures there. The difference in the x-component of the electric field and the z-
P

component of the time derivative of magnetic induction at this stage is caused by the top layer,
which smoothly gains in thickness in the case of the diapir model, but changes its thickness with a
CE

distinct step in case of the block model. At the sensitive early time span, the fields are therefore not
comparable.
At the later point in time (right column in Fig. 9), the eddy-current system has propagated further
AC

downwards. Consequently it is now sensitive to structures in other parts of the model. The main
difference between the computed fields is now caused by the different representations of the salt
diapir. Where the current-system has just reached the salt structure for the diapir model, the upper
left corner of the blocky salt dome influences the fields at an earlier stage. As a result, significant
differences occur at the top of the diapir. At the same time, the different representations of the top
of the claystone layer cause a smaller deviation between the solutions in the left part of the model as
well.
The described behaviour shows that the transient electromagnetic fields have a time-dependent
sensitivity to different parts of the computational domain. This means that special care has to be
taken when describing the geological setting for numerical simulations close to the source, because
this region is particularly sensitive at early times. Furthermore, the region around the potential
receivers and the transition zone of the eddy-current system has to be covered correctly as well.
Upon reversion, when the target of interest is shallow and far away from, e.g., a deep diapir
structure, simplifying the dome structure might be acceptable. However, when the target of interest
is in the vicinity of the said structure, it is crucial to cover its geometry correctly.

5. Visualization of the simulation results

5.1. Visualization tasks

10
 ACCEPTED MANUSCRIPT

The simulation results consist of very large arrays of numbers that have to be analyzed,
interpreted and validated. This is possible through the 3D visualization of the complex data. In this
study the free software tool Paraview (www.paraview.org) was used to visualize our data. Paraview
is based on the Visualization Toolkit (VTK, Schröder et al, 1996), a C++-based library for scientific
visualization. VTK implements a pipeline-based approach: data are read by elements of the pipeline
which act as a source and are than processed by subsequent filters which each implement a certain

PT
processing step, such as iso-surface extraction or streamline calculation. To explore our simulation
results systematically, we used Paraview for a number of visualization tasks, such as:

1. preprocessing of the data;

RI
2. data visualization and feature extraction;
3. comparative visualization - the influence of the model geometry;

SC
4. visualization of one physical problem dependent upon another one.

Fig. 10 shows a visualization of the results of the six simulation runs (three different simulation

NU
types, each run on the block and on the realistic model). Clearly this display is cluttered and contains
too much information for a printed image in the given size of a journal. However, in a 3D visualization
software, such as Paraview, it is possible to explore the data interactively, e.g. rotate the model or
zoom in on certain features of interest. Further the users’ comprehension can be improved by using
MA
stereoscopic visualization or by letting them explore the data in a virtual environment, as described
in Zehner (2010). In the remaining part of this section we explain the execution of the above-
mentioned visualization tasks.
D

4.2. Preprocessing
VTK provides its own data format for loading and saving data and can import data provided in the
TE

suitable external formats. We used the Skua plugin GocadExporters (Zehner, 2011; Zehner et al.,
2015) to export the data for the geometrical models as well as the results of the thermal simulation
in the VTK format. The simulation software for the electromagnetic simulation directly outputs the
P

data in VTK format. However, the software PFC3D which we used for the mechanical simulation with
the distinct element method outputs the results in form of a comma separated file, describing the
CE

displacement vector components for each ball in a row. Since many VTK functions require data to be
organized on a topologically consistent grid or mesh, this output had to be processed in Paraview.
We first read the data from the CSV file into Paraview as a table and converted it to a point set using
AC

the TableToPoints filter. In order to turn this into a topologically consistent mesh, we used the
Delaunay3D filter, which converted our point set into a tetrahedral mesh with the attributes on the
vertices. At the end of the pipeline we used the Programmable filter which allowed a python script to
be applied to the attributes. This python script allows the displacement vector to be assembled from
the different components. As this computation was already work intensive, we saved the result as a
new data set which is then further used for visualization.

4.3. Data visualization and feature extraction

At the beginning it is necessary simply to visualize the data in order to gain an insight into the
simulation results and their intrinsic features. One important feature that has to be displayed is the
course of vector fields in 3D space. We used two ways to envision vector fields: fieldlines and arrows.
The electric and magnetic fields were visualized by fieldlines. Fig. 10 shows electrical fieldlines for
the block model in orange and the realistic model in brown as well as the corresponding magnetic
fieldlines for the realistic mode in yellow (additionally shown in Fig. 11a). All the electric fieldlines
were calculated using Paraview’s StreamTracer filter. The torus-shaped streamlines of the magnetic
field were calculated by using the vertices of an electric fieldline as seed points with the
StreamTracerWithCustomSource filter. For technical reasons, the electromagnetic simulation was
performed in another coordinate system with the source in the origin. When using Paraview, it is
important to note that the seed-points/seed-lines for the StreamTracer are defined in the local

11
 ACCEPTED MANUSCRIPT

coordinate system and are pictured there, even when the data were transformed beforehand to the
global coordinate system using the Transform filter. For this reason it is more intuitive to do the
necessary computations in the original coordinate system and then to transform the result.
For the mechanical data, the Glyph filter could be used to visualize the displacement field by
vector arrows indicating the direction and magnitude of the displacement in one point. In Fig. 10, the
displacement field of the block model is illustrated by light green arrows, the displacement field of

PT
the realistic model by dark green arrows.
The visibility of the scalar temperature fields was enhanced by rendering contour lines of the
temperature along a N-S section with the Contour filter (Figs. 10, 11a right).

RI
4.4. Comparative visualization – influence of the simplification of model geometry
It is easiest to compare data generated on the same grid, such as the temperature fields of the

SC
block and realistic models. In this case the Calculator filter could be used to generate a new attribute
that contains the difference in the temperature. The blue and red clouds in the Figs. 10, 11 indicate
volumes where the temperature difference between the two models is larger than 4.5°C or smaller

NU
than -4.5°C. They were generated by applying the Threshold filter for the two values, subsequently
using the ShrinkCells filter for generating volume filling geometry, and then rendering the result
semi-transparent. Additionally, the temperature difference was displayed on a vertical section along
the long axis of the data set with Paraview’s Slice filter with the temperature difference as a color
MA
map (Fig. 10, 11c).
If the data sets were generated on different grids, they then have to be interpolated on one grid
for comparison. This is only possible for scalar properties. The electromagnetic simulations for the
realistic and the block model were performed on two tetrahedral meshes with different topology. In
D

order to show how this kind of data could be compared numerically, we interpolated the z-
component of the time derivative of the magnetic induction of the realistic model onto the vertices
TE

of the block model using the ResampleWithDataset filter. We then generated a new field, which
contains the difference between the two models, with the Calculator filter. Using the Contour filter,
we generated iso-surfaces (pink and green iso-surfaces in Figs.10a, 11b). For the interpolation of
P

vectors, specialized methods for interpolation such as Quaternions (Lengyel, 2004) are necessary, but
are not available in Paraview so far.
CE

In order to show where the displacement vector fields differ, we used the ResampleWithDataset
filter to interpolate the displacement vector of the block model onto the grid with the simulation
results for the realistic model. We then used the Calculator filter to calculate the angle between the
AC

displacement vectors of the two simulation runs. The white clouds in Figs. 10, 11b describe the iso-
surfaces within which the angle between the displacement-vectors is larger than 70° and were
generated using the Contour filter. Further we used the Threshold filter to clip all vertices/cells where
the angle is lower than 70°.

4.5. Visualize or query one physical problem in dependency on another one

Since we simulated three independent physical fields, one major advantage of passing the
simulation results to the visualization software is that the data can be combined for an exploratory
data analysis. Paraview provides the functionality to render one field locally dependent upon the fact
that another field exhibits a certain feature or is of a certain value. Many variables can be added and
integrated arbitrarily and several types of rendering can be generated. The visualization of one
physical problem dependent upon another one is possible, if the data are interpolated on a common
grid. This could be done for a scalar field by first applying the Threshold filter, which suppresses all
cells outside a given range. Then the visualized data could be interpolated onto the remaining cells
and rendered using a subsequent filter.
An example is shown in Fig. 11c where we suppressed all cells where the thermal difference
between the realistic model and the block model is smaller than 4°C and visualized the displacement
vector using the Glyph filter. This kind of visualization could, in practical applications, be used for
projects, where the production of geothermal energy is investigated with regard to geomechanical

12
 ACCEPTED MANUSCRIPT

problems. A correlation of fields or of important features can thus be achieved, which is valuable for
the knowledge discovery process.

5. Discussion
We have chosen a salt diapir to present our method, since the geological structures involved are
often complicated to model, due to their irregular shape, complex geometry and topology. The

PT
synthetic salt diapir serves as a standard salt structure including typical challenges a meshing tool has
to deal with. However, salt layers are also particularly relevant for geoscience as well as for the
economy and can be found worldwide in continental depositional settings (Jackson et al., 1995;
Hudec and Jackson, 2007).

RI
Due to their special physical properties, low density, high water solubility and viscous rheology, salt
layers initiate and control tectonic and morphogenetic processes. The vertical displacement of salt by

SC
upwelling from a source layer buried below a sedimentary cover is called halokinesis. Salt structures
like pillows, rollers, diapirs, domes and walls may form only due to gravitational instability caused by
the density contrast of the salt and its overburden sediments (Nettleton, 1934; Parker and McDowell,

NU
1955). This explains the occurrence of salt structures in regions where no indications for regional
tectonic processes are recorded, e.g. in cratonic basins. In compressional and extensional tectonic
settings, salt layers may efficiently be activated as viscous detachments for folding or thrusting
(Buxtorf, 1916; Goguel, 1952; De Sitter, 1956; Laubscher, 1977, Colman-Sadd, 1978, Davis and
MA
Engelder, 1985,1987) or normal faulting (Vendeville and Jackson, 1995, Brun and Mauduit, 2009, Tari
et al, 2016). Salt flows into the fault zones or into the cores of detachment faults and forms roller,
wall and diapir structures (Sherkati et al., 2005). If the influence of salt layers on the evolution of
structures has to be studied, the deformation processes can be simulated by analogue or numerical
D

experiments. Many numerical simulations were performed on simplified modeling set-ups and often
in 2D. These models provide a good insight into the dynamics of salt tectonic processes and the
TE

parameters controlling it (Talbot and Jackson, 1988, Poliakow et al., 1993; Schultz-Ela et al., 1993;
Weijermars et al., 1993; Bahroudi and Koyi, 2003; Massimi et al., 2007; Koyi et al., 2008;
Nilfouroushan et al., 2013, Ghazian and Buiter, 2014, Li et al., 2012).
P

A more detailed description is required, if salt bodies are subject to economic interventions. Salt
diapirs host mineral deposits of halite and sylvite. Sole cavities can be re-used as gas caverns. Mineral
CE

oil and gas traps are often associated with salt domes (Tittman, 1991; Petersen and Lerche, 1996;
Mello et al., 1995; Stover et al., 2001; Dorjnamjaa et al., 2006; Wan et al., 2007). The small
permeability of salt can promote trap and seal formation and the temperature anomalies induced by
AC

its high thermal conductivity can accelerate the maturation of supra salt rocks (Selling and Wallick,
1966; O’Brien and Lerche, 1984; Mello et al., 1995). The trap and seal structures are also of interest
for CO2 sequestration. Due to the unusual thermal properties of salt, diapirs are also interesting as a
geothermal resource. In northern Germany, for example, the temperatures above a salt diapir are
10-15 °C higher than in the same depths where no salt structures are present (Frome et al. 2010).
Additionally, the great specific thermal conductivity results in a great heat extraction capacity of
geothermal heat pumps. The salt structures have to be investigated in detail prior to any intervention
in the subsurface. While the economic intervention is in progress, the salt diapir has to be monitored
in order to detect water ingress, fluid or tectonic movements. These investigations usually require
simulations of the stress, displacement and temperature fields and of fluid and heat flow. The
simulations have to be run on geometries that describe the real structure as accurately as possible in
order to obtain detailed predictions about the profitability of a resource and risk assessment of
exploitation.
To generate a realistic 3D model of a salt diapir, detailed geological data of the surface and
subsurface are needed. The data set should comprise a geological map and at least two pairs of
crossing seismic sections, or even better a 3D seismic survey, or a set of drilling data describing the
flanks of the salt body and the host rock sediments. The challenges presented by modelling salt
diapirs are their irregular shape, complex internal structures and fault patterns, which cause complex
ray travel paths in seismic surveys and complicate the seismic interpretation (Bernitsas et al., 1997;

13
 ACCEPTED MANUSCRIPT

Hoetz et al., 2011; Jones and Davison, 2014). The interpretation can be improved by implementing
detection techniques and by processing algorithms removing source and receiver ghost signals (Asjad
and Deriche, 2015), by adding gravity data to better constrain the geometry of the salt diapir (Dubey
et al., 2014; Alvers et al., 2014) or by performing magnetotelluric surveys (Aveddeva et al., 2012).
Harding and Huuse (2015) provide a 3D model of a salt diapir from the Netherlands North Sea sector
based on a 3D seismic survey. They use thickness variations in the host rock sediments of the 3D

PT
model for reconstructing the syn-sedimentary growth of the diapir. Strozyk et al. (2012) provide a
seismic-based 3D model of the internal structure of a salt diapir from the Cleaver Bank High in the
North Sea. They describe brittle intra-salt layers which may control the deformation of the whole salt
body. Wu et al. (2016) use the 3D model of a salt diapir in the Australian shelf for kinematic

RI
interpretations of the regional fault pattern and its local deviation around the salt body. These 3D
models can provide the geometry for numerical simulations. In order to transfer the 3D models to

SC
different simulation methods and tools, it may be necessary to mesh them with various cell types.
Our workflow is a contribution to this important step of pre-processing and can be applied for tasks
of the primary investigation of a site as well as for the monitoring during an operational phase.

NU
Our temperature simulation could be part of a primary reconnaissance for geothermal potential
above a salt diapir. It could try to estimate the temperature of porous sedimentary units above a salt
body from where geothermal energy can be extracted. Heat flow and temperature simulations may
be used to evaluate the potential kind of heat use (heating of residential districts, greenhouse
MA
farming, production of electricity) and the depth of the heat production well. Adding a thermal
convection term to the heat equation enables the dimensioning of the production rate and mass. If
the heat extraction by a production well has to be modelled, adaptive meshing of the 3D model is
necessary, such that the borehole can be simulated accurately. If several scenarios with different well
D

locations need to be analyzed, the different discretizations of the 3D model have to be produced,
such that our workflow may be used for the flexible meshing of the models.
TE

The mechanical model presented here offers a very simple analysis of whether the diapir is
mechanically equilibrated or still ascending. If the diapir is not equilibrated, it still can grow and has
to be assessed as a tectonically active setting. In our example, the diapir is already equilibrated. The
P

distinct element method is well suited to describing the piercement of the salt diapir through its host
rock. The simulation method can also be used to investigate how stress orientations around the salt
CE

diapir are deflected due to the contrast between the mechanical properties of salt and sedimentary
host rock. This has implications on the drilling risk, since the most stable drilling direction may not be
vertical anymore (King, 2012, Girardi, 2012). A detailed investigation of stress and displacement in a
AC

salt diapir with a distinct element analysis is laborious, since the material parameters have to be
calibrated for the particle contacts of the specific ball assembly.
In order to monitor the water ingress in the salt diapir or the contamination of a fresh water aquifer
with brine, electromagnetic monitoring methods are suitable, since a great resistivity contrast is
produced by both salt and water or fresh water and brine. Different electromagnetic and geoelectric
methods are available, like the controlled-source, the transient electromagnetic and the direct-
current resistivity method. Each electromagnetic source creates an individual current system in the
subsurface which, together with a suitable receiver location, yields a particular sensitivity distribution
in space, time and frequency. In complicated conductivity environments, however, it is impossible to
predict these sensitivity patterns. Which methods and configurations appear most promising
depends on the local background resistivities, the borehole infrastructure, and the casing material.
Simulations of electromagnetic experiments enable the resolution of a number of configurations to
be assessed, leading to a most suitable survey configuration and provide information, in which time
range and spatial configuration an impact of the water is traceable. If various methods are simulated,
this may also require different types of discretizations of the same 3D model.

All simulated physical fields are strongly controlled / heavily influenced? by the salt diapir and its
geometry. The electric field develops circularly around the source with the torus-shaped magnetic
field rectangular to it. When propagating to the salt diapir, the electrical field is distorted and

14
 ACCEPTED MANUSCRIPT

deformed in both the block and the realistic model, due to the differing electromagnetic properties
of the salt. It is almost angular in the block model, but smooth and oval in the realistic model. A
positive temperature anomaly forms in the roof of the salt dome and its shape and size corresponds
to the shape and size of the diapir: It is rectangular in the block model and oval in the realistic model.
In the mechanical model, the roof of the diapir is equilibrated and corresponds to a plane of zero
vertical displacement. An upward directed displacement appears inside of the salt body. This

PT
buoyancy is compensated by the overload of the host rock.
The sedimentary layers also control the simulation results. The electric and the temperature fields
react mainly to the top clay and sand layers. Due to their different thermal properties, a temperature
anomaly in the shape of the clay layer forms. The electric field follows the clay layer due to its small

RI
resistivity. The sharp geometric changes in the block model cause major differences to occur in the
electric and thermal fields of the block and realistic models. In the mechanical model, the upper units

SC
are not prominent. Instead, this model reacts sensitively to the dense claystone and limestone layers.
These units are very thick in the northern part of the salt diapir in the realistic model. Thus, they
compensate the buoyancy of the less dense salt rocks. To the south of the diapir, these two layers

NU
are very thin, so that they cannot compensate the buoyancy of the salt and are dragged upward with
the diapir. In the block model the two units have a relatively small and constant thickness, such that
a symmetrical displacement pattern forms and the buoyancy of the salt is only compensated in the
top of the claystone layer. All simulation results clearly reflect the input geometry of the two models.
MA
6. Conclusions
We have shown how several simulations that differ regarding the type of the numerical
calculation scheme can be run on one 3D subsurface model. The input model could be generated
D

using a standard package for geoscientific 3D modeling, namely Skua-Gocad. However, we also used
a simplified model, the block model, which could easily be generated using e.g. Boolean operations in
TE

a CAD system.
Depending on the simulation method, different types of volume discretization of the 3D structural
model are required. The different grids for the numerical simulations could be generated without
P

substantially simplifying the model. The 3D structural model should be stored in a data format that
allows for a flexible volume discretization. Boundary representations of the model are particularly
CE

appropriate, because they describe only the boundaries of the geo-objects and allow for any
discretization with various cell types and resolutions. Conformable boundary representations without
holes and overlaps are especially advantageous, since only in this representation is it unequivocal
AC

whether a point of the volume is inside a polyhedron or not. Many meshing algorithms can only work
with conformable boundary representations, while other algorithms can handle holes and overlaps.
The triangulation of the structural model should be as equilateral as possible, since some meshing
algorithms use the surface triangulation for the construction of the cells, e.g. tetrahedrons. Triangles
with equilateral shape are especially stable during the simulation and enable a particularly accurate
solution to be achieved.
A special workflow for volume discretization has to be found for every simulation method,
depending on which mesh is required by the code used. In this study, we presented a workflow for
the discretization of Skua models with regular hexahedral cells, distinct elements and unstructured
tetrahedral meshes (Fig. 2).
The examples presented show that the results of numerical simulations closely reflect the input
geometry, e.g. in the case of the thermal simulation, we could work with exactly the same
discretization to represent the block and the realistic model; and the simulated temperature iso-
surfaces are different and reproduce the geometry of structural models. That is why it is necessary to
use or to generate the best possible 3D structural model for numerical simulations of physical fields
in the subsurface. Simplifications produce deviations and artifacts in the simulation results. It is
worth taking the time to construct the geometry of the modeling domain in a special geomodeling
software which includes all available geological data and geological concepts and to import this 3D
model into the desired simulation software.

15
 ACCEPTED MANUSCRIPT

The integrative visualization of the multiple physical simulations requires a special visualization
software. In this study, we presented a Paraview-based workflow. In order to accomplish this
visualization, the simulation results have to be exported in a file type that can be imported into
Paraview and interpolated onto one uniform grid. Then the various physical fields can be visualized
systematically and simultaneously. Differences between simulation results can be investigated and it
is possible to query minima, maxima and critical volumes. This integrative visualization enables

PT
geoscientists to answer complex geoscientific questions based on multiple simulations of physical
processes.
All examples considered in this study concern the direct problem, e.g. the calculation of physical
fields (stress, electromagnetic fields, temperature) for a given geometry and geology. The next step

RI
in practical application is to solve the inverse problem, e.g. reconstructing the distribution of material
parameters (density, electrical resistivity, thermal conductivity) directly from measured physical

SC
fields. Joint inversion techniques allow for the simultaneous inversion of contrasting datasets, which
requires the combination of physical methods operating independently of one another. Each method
requires an individually adapted mesh, notably around sources and receivers. In order to develop a

NU
user-friendly software for joint inversion problems, the workflow presented here has to be extended
to a uniform and flexible meshing concept.

Acknowledgements
MA
Ines Görz and Martin Herbst were funded by the Ministry of Science and Education of the Federal
State of Saxony. Jana Börner thanks the German Research Foundation for funding. Björn Zehner was
funded by the German Federation. We thank Martin Afanasjew for the transient electromagnetic
simulation code and Ralph-Uwe Börner for his implementation of the TEM analytical solution. We
D

thank two anonymous reviewers and Philippe Agard, Editor of Tectonophysics, for their constructive
comments and suggestions which helped to improve the manuscript. Further we thank Alison E.
TE

Martin for proofreading the manuscript and so making it more readable.

P

References
Afanasjew, M., Börner, R.-U., Eiermann, M., Ernst, O., Spitzer, K., 2013. Efficient Three-Dimensional
CE

Time Domain TEM Simulation Using Finite Elements, a Nonlocal Boundary Condition, Multigrid, and
Rational Krylov Subspace Methods. Expanded Abstracts, 5th International Symposium on Three-
Dimensional Electromagnetics, May 7 - 9, 2013, Sapporo, Japan, 4p.
AC

Alberta Geological Survey, 2016. 3D geological framework. http://ags.aer.ca/3d-geological-

framework.htm, last visited 11-9-2016.

ALICE, 2008. Graphite software. http://alice.loria.fr/WIKI/index.php/Graphite/Graphite, last visited

11-9-2016.

Alvers, M. R., Götze, H. J., Barrio-Alvers, L., Schmidt, S., Lahmeyer, B., Plonka, C., 2014. A novel
warped-space concept for interactive 3D-geometry-inversion to improve seismic imaging. First
Break, 32(4), 1-42.

Ansys Workbench 14.5, 2014. Ansys Inc., URL: http://www.ansys.com.

Archie, G. E., 1942. The electrical resistivity log as an aid in determining some reservoir
characteristics. Transactions of the American Institute of Mining and Metallurgical Engineers 146,
54-62.

Asjad, A., Deriche M., 2015. A new approach for salt dome detection using a 3D multidirectional edge
detector. Applied Geophysics 12.3 (2015), 334-342.

16
 ACCEPTED MANUSCRIPT

Avdeeva, A., Avdeev, D., Jegen, M., 2012. Detecting a salt dome overhang with magnetotellurics: 3D
inversion methodology and synthetic model studies. Geophysics 77(4), E251-E263.

Bahroudi, A., Koyi H. A., 2003. Effect of spatial distribution of Hormuz salt on the deformation style in
the Zagros fold and thrust belt: an analogue modelling approach. Journal of the Geological Society

PT
160, 719-733.

Bauer, M., Freeden, W., Jacobi, H., Neu, T., 2014. Handbuch tiefe Geothermie. Springer Berlin
Heidelberg, 842 pp.

RI
Bennis, C., Sassi, W., Faure, J. L., Chehade, F., 1996. One more step in Gocad stratigraphic grid

SC
generation: Taking into account faults and pinchouts. Society of Petroleum Engineers 35326, 307-
316.

NU
Bernitsas, N., Sun, J., Sicking C., 1997. Prism waves.- An explanation for curved seismic horizons
below the edge of salt bodies. 59th Annual International Conference and Exhibition, EAGE,
Extended Abstracts, E38. MA
Börner, J. H., Wang, F., Weißflog, J., Bär, M., Görz, I., Spitzer, K., 2015. Multi-method virtual
electromagnetic experiments for developing suitable monitoring designs: A fictitious CO2
sequestration scenario in Northern Germany. Geophysical Prospecting, 1430-1449.
D

British Geological Survey, 2016. 3D Geology. http://www.bgs.ac.uk/services/3Dgeology/home.html,

last visited 11-9-2016
TE

Brun J.-P., Mauduit T.P.O., 2009. Salt rollers: Structure and kinematics from analogue modelling.
Marine and Petroleum Geology 26 (2), 249-258.
P

Buxtorf, A., 1916. Prognosen und Befunde beim Hauensteinbasis und Grenchenberg-Tunnel und die
CE

Geologie des Juragebirges. Verhandlungen der Naturforschenden Gesellschaft in Basel 27, 184-205.

Byerlee, J. D., 1978. Friction of Rocks. Pure and Applied Geophysics 116 (4-5), 615-626.
AC

Carmichael, R. S., 1982. CRC handbook of physical properties of rocks. CRC Press. Inc, Boca Raton,
Florida USA.

Clauser, C., Huenges, E., 1995. Thermal conductivity of rocks and minerals. Rock physics & phase
relations. A handbook of physical constants, Wiley Online Library, 105-126.

Colman-Sadd, S.P., 1978. Fold development in Zagros Simply Folded Belt, southwest Iran. AAPG Bull.
62 (6), 984-1003.

Cundall, P. A., Strack, O. D. L., 1979. A discrete numerical model for granular assemblies.
Geotechnique 29 (1), 47-65.

Davis, D.M., Engelder, T., 1985. The role of salt in fold-and-thrust belts. Tectonophysics 119, 67-88.

Davis, D.M., Engelder, T., 1987. Thin-skinned deformation over salt. In: Lerche, I., O'Brien, J.J. (Eds.),
Dynamical Geology of Salt and Related Structures. Academic Press, Orlando, Florida, pp. 301-338.

De Sitter, L., 1956. Structural Geology. McGraw-Hill, London. 375pp.

17
 ACCEPTED MANUSCRIPT

Dorjnamjaa, D; Kondratov, L S; Voinkov, D M; Selenge, D, 2006. Mongolian Mesozoic intracontinental

basin and new technology of perspective evaluation by adsorbed form of gas in salt domes.
Geochimica et Cosmochimica Acta 70.18S, A144.

Dubey, C. P., Götze, H. J., Schmidt, S., Tiwari, V. M., 2014. A 3D model of the Wathlingen salt dome in

PT
the Northwest German Basin from joint modeling of gravity, gravity gradient, and curvature.
Interpretation 2(4), SJ103-SJ115.

Franke-Börner, A., 2013. Three-dimensional finite element simulation of magnetotelluric fields on

RI
unstructured grids – on the efficient formulation of the boundary value problem. PhD thesis,
Technical University Bergakademie Freiberg.

SC
Fromme, K., Michalzik, D., Wirth, W., 2010. Das geothermische Potenzial von Salzstrukturen in
Norddeutschland. Zeitschrift der Deutschen Gesellschaft für Geowissenschaften 161(3), 323-333.

NU
Geological Survey of Denmark and Greenland, 2016. 3D geological modelling.
http://www.geus.dk/UK/water-soil/mapping/Pages/3d_geol_model.aspx, last visited 11-9-2016.
MA
Gerya, T., 2010 Introduction to numerical geodynamic modeling. Cambridge University Press, 358 pp.

Geuzaine, C., Remacle, J.-F., 2009. Gmsh: a Three-Dimensional Finite Element Mesh Generator with
Built-in Pre- and Post-Processing Facilities. International Journal for Numerical Methods in
D

Engineering 79, 1309-1331.

TE

Ghazian, R. K., Buiter, S. J., 2014. Numerical modelling of the role of salt in continental collision: An
application to the southeast Zagros fold-and-thrust belt. Tectonophysics 632, 96-110.
P

Girardi, O., 2012. In-situ stresses and Palaeostresses around salt diapirs: a structural analysis from
the Gulf of Mexico and Amadeus Basin, Central Australia. University of Adelaide, PhD thesis, 54 pp.
CE

Goguel, J., 1952. Traite´ de tectonique. Masson et Cie, Paris. 383pp.

AC

Goldman, Y., Hubans, C., Nicoletis, S., Spitz, S., 1986. A finite-element solution for the transient
electromagnetic response of an arbitrary two dimensional resistivity distribution. Geophysics 51,
1450-1461.

Harding, R., Huuse, M., 2015. Salt on the move: Multi stage evolution of salt diapirs in the
Netherlands North Sea. Marine and Petroleum Geology 61, 39-55.

Haupt, M., 1991. A constitutive law for rock salt based on creep and relaxation tests. Rock Mechanics
and Rock Engineering, 24(4), 179-206.

Hoetz, G., Steenbrinkl, J., Bekkers, N., Vogelaar, A., Luthi, S., 2011. Salt-induced stress anomalies: An
explanation for variations in seismic velocity and reservoir quality. Petroleum Geoscience 17, 385-
396.

Hudec, M. R., Jackson, M. P. (2007). Terra infirma: Understanding salt tectonics. Earth-Science
Reviews 82(1), 1-28.

ITASCA, 2015. PFC3D – Particle Flow Code in 3 Dimensions. Manual, Itasca Consulting Group Inc.,
Mineapolis.

18
 ACCEPTED MANUSCRIPT

Jackson, M.P.A., 1995. Retrospective salt tectonics. In: Jackson, M.P.A., Roberts, D.G., Snelson, S.
(Eds.), Salt Tectonics: a Global Perspective. AAPG Memoir, vol. 65, pp. 1-28.

Jones, I. F., Davison, I., 2014. Seismic imaging in and around salt bodies. Interpretation 2(4), SL1-SL20.

PT
Kempka, T., Class, H., Görke, U.-J., Norden, B., Kolditz, O., Kühn, M., Walther, L., Wang, W., Zehner B.,
2013. A dynamic flow simulation code intercomparison based on the revised static model of the
Kezin pilot site. Energy Procedia 40, 418-427.

RI
Kempka, T., Kühn, M., 2013. Numerical simulations of CO2 arrival times and reservoir pressure
coincide with observations from the Ketzin pilot site. Germany Environ Earth Sci 70, 3675-3685.

SC
Kessler, H., Mathers, S. J., Sobisch, H.-G., 2004. GSI3D – The software and methodology to build near-
surface 3-D geological models. British Geological Survey (IR/04/029), 96pp.

NU
http://nora.nerc.ac.uk/4019/1/IR04029.pdf.

King, R., Backé, G., Tingay, M., Hillis, R., Mildren, S., 2012. Stress deflections around salt diapirs in the
Gulf of Mexico. Geological Society, London, Special Publications 367(1), 141-153.
MA
Kohl, T., Andenmatten, N., Rybach, L., 2003. Geothermal resource mapping – example from northern
Switzerland. Geothermics 32, 721-732.
D

Kolditz, O., Bauer, S., Bilke, L., Böttcher, N., Delfs, J. O., Fischer, T., Park, C. H., 2012. OpenGeoSys: an
open-source initiative for numerical simulation of thermo-hydro-mechanical/chemical (THM/C)
TE

processes in porous media. Environmental Earth Sciences 67(2), 589-599.

Koyi, H. A., Ghasemi, A., Hessami, K., Dietl, C., 2008. The mechanical relationship between strike-slip
P

faults and salt diapirs in the Zagros fold–thrust belt. Journal of the Geological Society 165(6), 1031-
1044.
CE

Kukkonen, I.T.,Golovanova, I.V., Khachay, Yu.V., Druzhinin, V.S., Kosarev, A.M., Schapov, V.A., 1997.
Low geothermal heat flow of the Urals fold belt implication of low heat production, fluid circulation
AC

or palaeoclimate? Tectonophysics, 276, 63-85.

Laubscher, H.P., 1977. Fold development in the Jura. Tectonophysics 37, 337–362.

Lengyel, E., 2004. Mathematics for 3D Game Programming & Computer Graphics. Second Edition,
Charles River Media Inc., Hingham, Massachusetts, 551p.

Li, S., Abe, S., Reuning, L., Becker, S., Urai, J. L., Kukla, P. A., 2012. Numerical modelling of the
displacement and deformation of embedded rock bodies during salt tectonics: A case study from
the South Oman Salt Basin. Geological Society, London, Special Publications 363(1), 503-520.

Lotz, B., 2004. Neubewertung des rezenten Wärmestroms im Norddeutschen Becken. PhD thesis,
Free University, Berlin, Germany.

Mallet, J. L., 2002. Geomodeling. Oxford University Press, Inc.

Mallet, J. L., 2014. Elements of mathematical sedimentary geology: The GeoChron model. EAGE
publications BV.

19
 ACCEPTED MANUSCRIPT

Marschallinger, R., Jandrisevits, C., Zobl, F., 2015. A visual LISP program for voxelizing AutoCAD solid
models.Computers and Geosciences 74,110-120.

Massimi, P., Quarteroni, A., Saleri, F., Scrofani, G., 2007. Modeling of salt tectonics. Computer
methods in applied mechanics and engineering 197(1), 281-293.

PT
Mello, U.T., Karner, G.D., Anderson, R.N., 1995. The role of salt rock in modifying the maturation
history of sedimentary basins, Marine Petrol. Geol. 12 (7), 697-716.

Nabighian, M.N., 1979. Quasi-static Transient Response of a Conducting Half-Space – an Approximate

RI
Representation. Geophysics 44 (10), 1700-1705.

SC
Nettleton, L.L., 1934. Fluid mechanics of salt domes. AAPG Bulletin 18, 1175-1204.

Nilfouroushan, F., Pysklywec, R., Cruden, A., Koyi, H., 2013. Thermal-mechanical modeling of

NU
salt-based mountain belts with pre-existing basement faults: Application to the Zagros fold and
thrust belt, southwest Iran. Tectonics 32(5), 1212-1226.

O’Brien, J.J., Lerche, I., 1984. The influence of salt domes on paleotemperature distributions.
MA
Geophysics 49, 2032-2043.

Oladyshkin, S., Royer, J.-J., Panfilov, M., 2007. Modeling Compositional Flows in oil reservoirs using
gocad streamline simulator and HT-splitting technique. 27th Gocad meeting, Nancy, 18 pp.
D

PARADIGM Gocad-Skua® software 2011. URL: http://www.pdgm.com/products/skua-gocad, last

TE

visited 11-9-2016.

Park, C.-H., Shinn, Y.J., Park, Y.-C., Huh, D.-G., Lee, S.K., 2014. PET2OGS: Algorithms to link the static
P

model of Petrel with the dynamic model of OpenGeoSys. Computers and Geosciences 62, 95-102.
CE

Parker, T.J., McDowell, A. N., 1955. Model studies of salt-dome tectonics. Bulletin of the American
Association of Petroleum Geologists 39.12, 2384-2470.
AC

Pellerin, J., Levy, B., Caumon, G., 2011. Topological control for isotropic remeshing of nonmanifold
surfaces with varying resolution: application to 3D structural models. Conference of the
International Association of Mathematical Geosciences (IAMG 2011), Salzburg, Austria, 5th-9th
September 2011, doi:10.5242/iamg.2011.0158.

Pellerin, J., Lévy, B., Caumon, G., Botella, A., 2014. Automatic surface remeshing of 3D structural
models at specified resolution: A method based on Voronoi diagrams. Computers & Geosciences
62, 103-116.

Petersen, K., & Lerche, I., 1996. Temperature dependence of thermal anomalies near evolving salt
structures: importance for reducing exploration risk. Geological Society, London, Special
Publications 100(1), 275-290.

Poliakov, A. N. B., Podladchikov, Y., Talbot, C., 1993. Initiation of salt diapirs with frictional
overburdens: numerical experiments. Tectonophysics 228(3-4), 199-210.

Ranalli, G., 1995. Rheology of the Earth. Springer Science & Business Media.

20
 ACCEPTED MANUSCRIPT

Royer, J.-J. , Mejia, P., Caumon, G., Collon, P., 2015. 3 and 4D Geomodeling Applied to Mineral
Resources Exploration - An Introduction. In P. Weihed (Ed.), 3D, 4D and Predictive Modelling of
Major Mineral Belts in Europe, Mineral Resource Reviews, Ch. 4, 73-89.

Sass, J. H., Lachenbruch, A. H., Moses Jr., T. H., 1992. Heat flow from a scientific research well at
Cajon Pass, California, J. Geophys. Res. 97(B4), 5017-5030.

PT
Schön, J., 2015. Physical Properties of Rocks – Fundamentals and Principles of Petrophysics.
Handbook of Geophysical Exploration 18, 2nd ed, Elsevier.

RI
Schröder, W., Martin, K., Lorensen, B., 1996. The Visualization Toolkit, An Object- Oriented Approach
to 3D Graphics. Prentice-Hall PTR, Upper Saddle River, NJ, USA 826pp.

SC
Schultz-Ela, D. D., Jackson, M. P., Vendeville, B. C., 1993. Mechanics of active salt diapirism.
Tectonophysics 228(3), 275-312.

NU
Sellig, F., Wallick, G. C., (1995): Temperature distribution in salt domes and surrounding sediments.
Geophysics 31 (2), 346-361. MA
Sherkati, S., Molinaro, M., de Lamotte, D. F., Letouzey, J., 2005. Detachment folding in the Central
and Eastern Zagros fold-belt (Iran): salt mobility, multiple detachments and late basement control.
Journal of structural Geology 27(9), 1680-1696.
D

Si, H., 2011. TetGen, A Quality Tetrahedral Mesh Generator and a 3D Delaunay Triangulator.
http://wias-berlin.de/software/tetgen/, last visited 11-9-2016.
TE

Spitzer, K., Wurmstich, B., 1999. Speed and acuracy in 3d resistivity modeling. in Three-dimensional
electromagnetics: Society of Exploration Geophysicists, SEG Book Series Geophysical Developments
P

7,161-176.
CE

Stille, H., 1932. Zur Paläogeographie des nordöstlichen Niedersachsens. Weidmann.

Stover, S. C., Ge, S., Weimer, P., McBride, B. C., 2001. The effects of salt evolution, structural
AC

development, and fault propagation on Late Mesozoic-Cenozoic oil migration: A two-dimensional

fluid-flow study along a megaregional profile in the northern Gulf of Mexico Basin. AAPG bulletin
85(11), 1945-1966.

Strozyk, F., Van Gent, H., Urai, J. L., Kukla, P. A., 2012. 3D seismic study of complex intra-salt
deformation: An example from the Upper Permian Zechstein 3 stringer, western Dutch offshore.
Geological Society, London, Special Publications 363(1), 489-501.

Talbot, C. J., Jackson, M. P. A., 1988. Salt tectonics. Science, Washington, DC, USA.

Tari, G., Dellmour, R., Rodgers, E., Asmar, C., Hagedorn, P., Salman, A., 2016. Styles of salt tectonics in
the Sab’atayn basin, onshore Yemen. Arabian Journal of Geosciences 9(10), 570.

Tittman, J., 1991. Finding oil next to salt domes, Oilfield Review 3.1, 4-7.

Van der Meulen, M. J., Doornenbal, J. C., Gunnink, J. L., Stafleu, J., Schokker, J., Vernes, R. W., Bakker,
M. A. J., 2013. 3D geology in a 2D country: perspectives for geological surveying in the Netherlands.
Netherlands Journal of Geosciences 92(04), 217-241.

21
 ACCEPTED MANUSCRIPT

Vendeville, B.C., Ge, H., Jackson, M.P.A., 1995. Scale models of salt tectonics during basement-
involved extension. Petroleum Geoscience 1, 179-183.

Wan, G.; Tang, Liangjie; J., Wenzheng; Y., Wenjing; Y Y., 2007. Control of salt-related tectonics on oil
and gas accumulation in the western Kuqa Depression. Dizhixue Bao = Acta Geologica Sinica 81.2,
187-196.

PT
Watson, C., Richardson, J., Wood, B., Jackson, C., Hughes, A., 2015. Improving geological and process
model integration through TIN to 3D grid conversion. Computers and Geosciences 82, 45-54.

RI
Weijermars, R., Jackson, M. T., Vendeville, B., 1993. Rheological and tectonic modeling of salt
provinces. Tectonophysics 217(1-2), 143-174.

SC
Weiler, K., 1988. The radial edge structure: A topological representation for non-manifold geometric
boundary modeling. In Wozny M., McLaughlin H., Encarna (Eds.), Geometric modeling for CAD

NU
application. Elsevier Amsterdam, 3-36.

Wieczoreck, B., Semmler, G., Gietzel, J., Schaeben, H., 2015. Converting corrupted 3d geomodels into
voxel models, IAMG2015 Conference, Freiberg, 67-75.
MA
Wu, L., Trudgill, B. D., Kluth, C. F., 2016. Salt diapir reactivation and normal faulting in an oblique
extensional system, Vulcan Sub-basin, NW Australia. Journal of the Geological Society, jgs2016-008.
D

Wycisk, P., Hubert, T., Gossel, W., Neumann, C., 2009. High-resolution 3D spatial modeling of
complex geological structures for an environmental risk assessment of abundant mining and
TE

industrial megasites. Computers and Geosciences 35, 165-182.

Zehner, B., 2010. Mixing Virtual Reality and 2D Visualization - Using Virtual Environments as Visual 3D
P

Information Systems for Discussion of Data from Geo- and Environmental Sciences. Conference
Proceedings of the International Conference on Computer Graphics Theory and Applications 2010
CE

(GRAPP 2010), Angers, France, May 2010, 364-369.

Zehner, B., 2011. Constructing Geometric Models of the Subsurface for Finite Element Simulation.
AC

Conference of the International Association of Mathematical Geosciences (IAMG 2011), Salzburg,

Austria, 5th-9th September 2011, doi:10.5242/iamg.2011.0069.

Zehner, B., Börner, J., Görz, I., Spitzer, K., 2015. Workflows for generating tetrahedral meshes for
finite element simulations on complex geological structures. Computers and Geosciences 79, 105-
117. http://dx.doi.org/10.1016/j.cageo.2015.02.009.

Zehner, B., Hellwig, O., Linke, M., Görz, I. , Buske, S., 2016. Rasterizing geological models for parallel
finite difference simulation using seismic simulation as an example. Computer and Geosciences 86,
83-91.

Ziegler, P. A., 1982. Geological atlas of western and central Europe. Shell International Petroleum
Maatschappij u. a., Den Haag.

22
 ACCEPTED MANUSCRIPT

Fig. 1. The diapir data set and its representation in 3D models: a) the data points, b) the
stratigraphic column, c) structure represented in a simplified block model and d) in a realistic 3D
model that can represent complex geometry, the realistic model in various types of representation: e)
in a boundary representation, f) discretized by a structured hexahedral grid, g) by an unstructured
tetrahedral mesh and h) by distinct elements.

PT
Fig. 2. Workflows for preparation of a 3D model produced with the Paradigm software Skua-
Gocad for volume discretization with various cell types. The generation of a conformable boundary
representation is a step that allows for the efficient discretization of the model volume, since many

RI
meshing tools require it, while most of the meshing tools which do not require it, can handle it as well.

Fig. 3. Accuracy test of the numerical solution: a) comparison with the analytical solution. The

SC
diagram shows the end temperature at 6,464 m depth dependent upon the heat production. The
greater the heat production, the larger the difference is between the numerical and analytical
solution. For the parameter range used in the diapir models, a deviation of 0.5 °C is achieved. b)

NU
Comparison of the temperature-depth curve for the realistic model and the block model calculated on
grids with different solution. The temperature variation is about 3 °C. c) Section through the
temperature fields of all simulated models above and below the diapir.
MA
Fig. 4. Temperature field simulations: a) discretization of the block model and b) of the realistic
model. Temperature filed along a N-S and an E-W section for c) the block model d) the realistic model.
A positive temperature anomaly can be seen above the salt diapir. e) and f) show the 10°C iso-
temperature surface for the block and the realistic model respectively. The shape of both surfaces
D

clearly reproduces the input geometry. The last set of images shows the differences between the
block and realistic model. g) displays the differences in the id of the geological units (compare with
TE

table 1) and h) displays the differences in the simulated temperatures.

Fig. 5. Distinct element discretization of the salt diapir model: a) The color code shows differences
P

in the unit assignment for the block and the realistic model, the lines indicate the locations where an
analytical solution for the lithostatic stress was calculated, the spheres represent calculation spheres,
CE

where the vertical stress component in the numerical models was obtained; b) shows the boundary
conditions and densities used for the numerical simulation. c) comparison of the analytical solution
and the numerical solution for the lithostatic stress along the lines marked in Fig. 5a.
AC

Fig. 6. Simulation results for the distinct element models: vertical displacement and total
displacement of the particles after the gravity was applied.

Fig. 7. Comparison between the analytical solution (black lines) and the numerical solution
(triangles) for a 80 Ωm half-space in terms of the x-component of the electric field (left) and the z-
component of the time derivative of the magnetic induction (right) at (0 m, 1,000 m, 500 m). Dashed
lines and empty markers denote negative field values, solid lines and filled markers denote positive
values.

Fig. 8. Deformation of the electric field in the vicinity of the salt bodies. The representation of the
diapir in both the block and the realistic model is displayed for comparison.

Fig. 9. Numerical finite element simulation results for two time steps (columns) and in terms of the
x-component of the electric field (column 1 and 2) and the z-component of the time derivative of the
magnetic induction (column 3 and 4). In each case, the results along the plane x = 0 are shown for
both the realistic and the block 3D model (the dominant geometric features are plotted as black
lines). The absolute difference between the results for the two models is shown in a third subplot for
each case.

23
 ACCEPTED MANUSCRIPT

Fig. 10. Scientific visualization of the simulation results of the six model runs. The electric and
magnetic fields are represented by fieldlines, the displacement fields are represented by arrows, the
temperature field by iso-lines along a N-S section. Differences between the block and realistic model
are visualized clouds.

PT
Fig. 11. Visualization of the simulation results. a) Details of the visualized physical fields, b)
differences between the block and the realistic model, c) visualization of one physical problem
dependent upon another one.

RI
SC
NU
MA
D
P TE
CE
AC

24
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

25
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

26
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

27
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

28
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

29
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

30
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

31
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

32
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

33
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

34
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

35
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
D
TE
P
CE
AC

36
 ACCEPTED MANUSCRIPT

Table 1. Material parameters used for the numerical simulations. The parameters for each geological unit are constant.
References: (1) Bauer et al. (2014), (2) Kukkonen et al. (1997), (3) Lotz (2004), (4) Augner (2007), (5) Byerlee (1978), (6)
Carmichael (1982), (7) Schön (2015).
Temperature Stress and Electromagnetic
field displacement field fields
Unit Id Heat Radiogenic heat Friction Density Resistivity
conductivity production coefficient

PT
W/(m*K) 10-6 W/m3 unitless kg/m3 Ωµ
sand 1 2.1(1) 0.7(1) 0.5(4) 2000(6) 80(7)
clay 2 1.6(1) 1.8(1) 0.3(4) 2400(6) 50(7)

RI
clay+sand 3 1.8(1) 1.3(1) 0.4(4) 2400(6) 50(7)
marl 4 2.7(1) 1.1(2) 0.6(5) 2250(6) 15(7)
claystone 5 2.0(1) 1.8(1) 0.6(5) 2650(6) 10(7)

SC
limestone 6 2.6(1) 0.4(2) 0.6(5) 2700(6) 20(7)
sandstone 7 3.2(1) 0.7(1) 0.6(5) 2450(6) 25(7)
saltstone 8 5.4(1) 0.5(3) 0.6(5) 2100(6) 1000(7)

NU
MA
D
P TE
CE
AC

37