GEOPHYSICAL RESEARCH LETTERS, VOL. 29, NO. 9, 1320, 10.

1029/2002GL014901, 2002

Core flooding laboratory experiment validates numerical simulation of

induced permeability change in reservoir sandstone
Jorn Bartels1
Geothermie Neubrandenburg Ltd., USA

Michael Kuhn and Wilfried Schneider

Technical University of Hamburg-Harburg, Department of Water Management and Water Supply, USA

Christoph Clauser and Hansgeorg Pape

Aachen University of Technology (RWTH), Applied Geophysics, USA

Volker Meyn and Ivo Lajcsak

German Petroleum Institute, USA
Received 12 February 2002; revised 21 March 2002; accepted 28 March 2002; published 10 May 2002.

[1] Numerical simulation of reactive transport was validated in a

core flooding experiment simulating conditions in a managed
geothermal reservoir. Permeability was measured along a sandstone
core prepared with anhydrite and subjected to a temperature
gradient. Anhydrite was dissolved and precipitated in the cold
upstream and hot downstream regions of the core, respectively. The
numerical code SHEMAT was used to simulate coupled transport
and chemical reactions at the temperature front (http://www.rwthaachen.de/geop/shemat / ). It comprises an extended version of the
geochemical speciation code PHRQPITZ for calculating chemical
reactions in brines of low-high ionic strength and temperatures of 0
150
C. Permeability is updated to porosity via a novel, calibrated
power-law based on a fractal pore-space model resulting in a large
exponent of 11.3. Simulation results agree well with measured
permeability. This both validates the model and demonstrates that
the fractal relationship is crucial for a successful simulation of this
type of reactive transport.
INDEX TERMS: 3210 Mathematical
Geophysics: Modeling; 5114 Physical Properties of Rock:
Permeability and porosity; 8135 Evolution of the Earth:
Hydrothermal systems (8424); 1854 Hydrology: Precipitation
(3354); 1045 Geochemistry: Low-temperature geochemistry

1. Introduction
[2] Hydrothermal heat mining installations are major capital
investments. They usually involve two or more production and
injection wells. In order to pay off, operation times of 30 years and
more are required during which sufficiently large flow rates must
be maintained at the injection and production wells. Therefore,
permeability is a key parameter for a successful heat mining,
particularly in the vicinity of the wells.
[3] Reservoir fluids are often highly saline brines, such as, for
instance, in the north German sedimentary basin where thermal
waters are almost saturated with sodium chloride, earth alkaline
sulfates and carbonates. When cooled brines are re-injected into a
reservoir, the chemical equilibrium between the formation water
and the reservoir rocks is disturbed. This could cause temperature
induced dissolution or precipitation of minerals, if not prevented by
1
Formerly at Aachen University of Technology (RWTH), Applied
Geophysics.

Copyright 2002 by the American Geophysical Union.

0094-8276/02/2002GL014901

kinetic barriers. In the worst case this may cause a more or less
irreversible sealing of the reservoir.
[4] In order to predict the amount of precipitation or dissolution
over the entire period of operation the complex interaction of the
involved processesflow, heat transfer, multi-species transport,
and chemical reactionsneeds to be understood. This complex
regime is characterized by steep fronts in temperature and concentration of dissolved salts as well as high flow rates and large
temperature differences which are required for an economic heat
mining. These processes can be predicted only by numerical
simulation. For this purpose a new simulation code combined with
a customized graphical user interface was developed (SHEMAT/
Processing SHEMAT; [Bartels et al., 2000]).
[5] To achieve a certain degree of confidence with respect to the
predicted reservoir behavior numerical results need to be validated
by comparison with observations. On the one hand the evolution of
the mineral and permeability distribution in the reservoir in space
and time cannot be observed directly. On the other hand 30 years of
operation experience of geothermal installations in sandstone
reservoirs are not yet available. Therefore, a one-dimensional
laboratory model was set up to study processes at a thermal front
such as encountered in heat mining. In this core flooding experiment temperature, pressure, flow rate, and salinity in a sandstone
core of approximately 500 mm length were made comparable with
the conditions at the temperature front of a deep hot-water
reservoir. Model validation was obtained by comparison of simulation results with permeability observed after the core flooding
experiment.

2. Core Flooding Experiment

[6] A series of flooding experiments was performed on cores of
Bentheim sandstone which is very clean and consists virtually only
of quartz-cemented quartz grains. Diameter and length of the cores
were 29 mm and approximately 500 mm, respectively. Prior to
flooding small amounts of iron were flushed out using hydrochloric acid. Then a total of 9 12 g of anhydrite was deposited in
the pore space along the entire length of the core. A special device
was used to deposit anhydrite in the core at high salinity and
elevated temperature of 90
C to avoid the precipitation of metastable solids [Power et al., 1966]. Most probably, this anhydrite
precipitate is not homogeneously distributed. Therefore, the variation of permeability along the core was measured using a custommade ring gas-permeameter. The device is moved in increments
along the core, and gas flows into the core via the circumference
and out against ambient pressure. The pressure differences are used

34 - 1

34 - 2

BARTELS ET AL.: REACTIVE FLOW: EXPERIMENT AND SIMULATION

to determine numerically the variation of permeability along the

core, assuming negligible radial variation. With this device the
permeability distribution was measured after both the preparation
of the cores and the flooding experiment. Therefore, all information about porosity and mineral redistribution had to be inferred
from permeability as the only rock property measured continuously
along the core. It had not been possible to measure the permeability
along the untreated core in the described experiment.
[7] For monitoring the permeability variation within the three
core segments during the transport experiments, four pressure tips
were installed through the core sleeve. The end segment was
heated and held at a constant temperature of 100
C at the outflow.
The central segment was thermally insolated. Fluid flow was from
the cold to the warm end and the out-flowing brine was re-injected
at a lower temperature level of 80
C, so that a steady-state
temperature gradient of about 1 K cm
1 built up over a distance
of 20 cm in the central segment. For a period of 20 days flooding
of the core was maintained at a specific discharge rate of 1.1 meter
per day. This corresponds to a distance of 10 100 m from an
injection well in an operated heat-mining field. The overburden
pressure was 10 MPa. Salinity varied from 0 100 g L
1 NaCl
within this series of transport experiments. The concentration of
dissolved calcium at the outlet of the cores was determined. After
flooding permeability was measured continuously along the core
with the ring gas permeameter. Where significant changes in
permeability were observed samples were taken from the core
and permeability, porosity and anhydrite content were determined.
Also scanning electron microscopy was used to display changes in
the internal pore structure.
[8] One of the flood experiments was selected for a comparison
with a numerical reactive transport simulation. The parameters of
this anhydrite redistribution experiment are listed in Table 1.

Table 1. Observation Data Used for Simulating the Core Flooding

Experiment
Average porosity after preparation with anhydrite
Anhydrite content in the core after preparation
NaCl concentration of the fluid
Total duration of flooding
Fractal exponent exp3
Sealing pressure
Inflow temperature
Outflow temperature

terms on the right-hand side of Equation (1) correspond to the

porosity ranges f  0.01, 0.01 < f  0.1, and 0.1 < f  0.5, from
left to right. The coefficients A, B, and C, in turn, need to be
calibrated for each rock type. Different processes are responsible for
changes in porosity and permeability in different ranges of porosity
and may be explained by fractal or other models such as percolation
theory. In SHEMAT, this is approximated via Equation (2) by
assigning different exponents for up to three different ranges of
porosity. In Equation (2), (k 0 , f0) denote the initial values which
represent the unaltered reservoir properties corresponding to the
coefficients A, B and C in Equation (1):
k k0

 expi
f
; i 1; 2 or 3
f0

k Afexp1 Bfexp 2 Cfexp 3

where exp1 = 1 and the exponents exp2 and exp3 depend on the
fractal dimension of the internal surface of the pore space. The three

In our case of a high-porosity sediment, the first two terms on the

right-hand side of Equation (1), corresponding to i = 1, 2 in
Equation (2) can be neglected. Equation (2) needs to be applied
twice in order to set up a numerical model for the flooding experiment: First, to calculate the initial distribution of anhydrite in the
prepared core from the measured initial permeability via porosity:

3. Numerical Simulation
[9] The mutually coupled partial differential equations for flow,
heat transfer, multi-component transport, and chemical reactions
were solved numerically using an extension of the finite difference
code SHEMAT [Clauser and Villinger, 1990; Bartels et al., 2000].
The Ilin-flux-blending scheme [Clauser and Kiesner, 1987] was
chosen from the three available numerical procedures implemented
in SHEMAT for solving the equations of heat and species transport.
[10] All processes involved, i.e. flow, heat transfer, multi-component transport, and geochemical reactions, are coupled (1) via the
dependence of the material and thermodynamic properties on
temperature, pressure, and species concentrations, and (2) by solving simultaneously the equations of flow, transport, and chemical
reactions. A chemical reaction model was developed to calculate the
amount of precipitation and dissolution of each mineral species at
high temperature and salinity. It is a modification of the geochemical
simulation code PHRQPITZ [Plummer et al., 1988] and permits to
calculate geochemical reactions in brines using the Pitzer virial
coefficient approach for the correction of activity coefficients. The
data base of Pitzer interaction parameters was established for the
system Na-K-Mg-Ca-H-Cl-SO4-OH-(HCO3-CO3-CO2)-H2O and is
valid for temperatures from 0 150
C with the exception of the
carbonic acid system (set in parenthesis) which is restricted to
0 90
C [Kuhn et al., 1999; Kuhn et al., 2002].
[11] In general, the variation of permeability with porosity
within a limited porosity range can be expressed by an empirical
power law. For average sandstones whose diagenesis was dominated by mechanical compaction we applied a three-term power
series [Pape et al., 1999]:

24.6 %
10.8 g
100 g L
1
474 h
11.3
10 Mpa
80
C
100
C

c x

fprep x
funprep x
;
vmol

where c is the change of mineral concentration per volume unit in

mol m
3, and vmol the molar volume of the precipitated mineral. The
indices of f refer to the porosity of the unprepared and the prepared
core before the experiment (Table 1).
[12] Since k unprep (x) and therefore funprep (x) were unavailable
it had to be substituted by its mean value hfunprepi which was
obtained from:


funprep fprep

c total
Vcore rmineral ;

where c total is the total amount of mineral used for preparation,

rmineral is the density of the precipitated mineral, and Vcore the total
core volume. Calculation of the initial mineral distribution by this
method is based on the measured permeability distribution, average
porosity, and the fractal exponent derived from statistical analysis
of data from core samples.
[13] Then, Equation (2) needs to be applied a second time in
order to calculate the transient permeability change in the core
during the experiment from porosity, i.e. from the time dependent
mineral distribution:
f x ftt
ft Vspec x vmol ct x;

where Vspec is the volume change per unit volume, t the time and
t the time step.
[14] The steady-state temperature distribution along the flooded
core was obtained by an independent two-dimensional simulation
of coupled flow and heat transport only. Boundary conditions were
the measured inflow and outflow temperatures. The heating of the

BARTELS ET AL.: REACTIVE FLOW: EXPERIMENT AND SIMULATION

34 - 3

[18] Additional simulations, not discussed here, involving lower

values for exp3 and different kinetic reaction rates, were performed
for comparison and sensitivity studies. They yielded significantly
lower maximum anhydrite concentrations in the precipitation zone.
[19] In our preferred model, the position and shape of the
dissolution front propagating downstream still deviate somewhat
from the observations: at the end of the simulation the reduction of
porosity and permeability occurs 5 8 cm too far downstream and
is too steep (Figure 3, top panel). However, the simulated anhydrite
content agrees with the measured one within the uncertainty of the
laboratory analysis (Figure 3, bottom panel).

5. Discussion
Figure 1. Variation of permeability with position x along the core
after preparation with anhydrite measured with a ring permeameter
before (circles) and after flooding (triangles). The simulated
steady-state temperature field is shown as a dashed line.

core-mantle over a length of 165 mm in the last segment and the

thermal insulation over a length of 121 mm in the central segment
were also taken into account. The flow rate in the model was held
constant and equal to the measured one. The resulting temperature
distribution was held constant in the numerical simulation of the
complete flood experiment discussed in the following section.

[20] The core flooding experiment shows that increase and

reduction of permeability are associated with dissolution and
precipitation of anhydrite, respectively. A dissolution front accompanied by a permeability increase up to the level of the unprepared
core is shifting downstream. This suggests strongly that anhydrite
is completely removed from this region.
[21] Dissolution in this low temperature region is due to the fact
that the amount of soluble anhydrite at the inflow at 80
C is larger
than at the outflow at 100
C where the concentration of dissolved
anhydrite obtained from fluid analyses (0.0245 mol kg
1) is in
equilibrium [Kuhn et al., 2002]. Therefore, chemical equilibrium can
be assumed, as was done in the successful simulation described here.
This means that the chemical reaction rates are sufficiently rapid, and
any deviations from equilibrium at a fixed core position due to

4. Results
[15] Figure 1 displays the observed permeability distribution
before and after the flooding experiment and the simulated steadystate temperature profile along the core. Comparison of the initial
and final curve in Figure 1 shows a permeability increase in the
upstream, low-temperature core section and a permeability reduction at the temperature front, i.e. in the region of the steepest
positive temperature gradient. Figure 2 shows the precipitated finegrained crystals with a size of 0.5 15 mm in this region of
maximum permeability reduction after flooding. Due to the special
preparation of the core described above the only process capable of
producing these permeability changes is transport of anhydrite
dissolved in the low-temperature region upstream, followed by
precipitation at an elevated temperature downstream.
[16] Corresponding values of porosity and permeability from
the regions of dissolution and precipitation were used for a semilogarithmic fit of Equation (2). Linear regression on a logarithmic
scale with 10 data points yields an exponent exp3 = 11.3 for the
porosity range f > 0.1 with R2 = 0.82, where R is the correlation
coefficient. This value was used to (1) calculate the initial mineral
distribution (Figure 3 below) from the measured permeability, and
(2) calculate permeability changes in the numerical simulation
from Equation (2). The resulting total amount of anhydrite in the
core is 10.9 g compared to 10.8 g from the chemical analysis. In
contrast, simulations with exponents in the range of 3 5, which
are typical for sandstones with a smooth internal surface, yielded
poor fits to the data.
[17] The simulation results are shown in Figure 3. The transient
changes in the distribution of anhydrite are illustrated by three
snapshots before, during and after flooding (Figure 3, bottom
panel). Clearly, anhydrite is continuously redistributed downstream
across the temperature front as observed in the core flooding
experiment. Generally, the final numerical simulation result fits
the observed mineral content, porosity and permeability. In detail,
both the porosity and permeability reduction in the precipitation
zone and the maximum measured value of anhydrite concentration
are reproduced well (Figure 3, top panel). The simulated outflow
concentration of dissolved anhydrite (24.7 mmol kg
1 H2O) agrees
with the measured one (24.5 mmol kg
1 H2O) within 1%.

Figure 2. REM micrographs of the pore space in the region of

maximum permeability reduction after flooding. Large figure:
Overview of the distribution of precipitated anhydrite in pore
space. Small inset: Typical shape and size distribution of the
precipitated anhydrite crystals; length scale in both parts of figure
is 10 mm.

34 - 4

BARTELS ET AL.: REACTIVE FLOW: EXPERIMENT AND SIMULATION

model are calculated from simulated changes of the mineral
concentration (Equation (2) and Equation (5)).
[25] Also a number of idealizing assumptions were made to
express anhydrite structure by a single fractal exponent. Therefore,
it is an important feature of this simulation that the good fit of the
data by the model was achieved without further adapting exp3 in
Equation (2). It may be related to the fractal dimension D of the
pore wall structure [Pape et al., 1999; Pape et al., 2000]. Its value
of exp3 = 11.3 in our experiment, for instance, corresponds to a
fractal dimension of D = 2.45, which reflects rough pore walls like
the one shown in Figure 2. The results indicate therefore that the
frequently made assumption of smooth and spherical grains (corresponding to D = 2 and exp3 = 3) is not justified in the case of
precipitation of anhydrite in a comparatively rapid flow regime,
typical for managed geothermal reservoirs.
[26] In summary, laboratory and numerical experiments prove
that at least for a simplified system under well-controlled laboratory conditions, it has been possible to successfully model
dissolution-precipitation effects and associated porosity and permeability changes. In particular, the numerical code SHEMAT is
well suited to simulate transient processes and to predict changes in
properties during heat mining in reservoirs.
[27] Acknowledgments. The research reported in this paper was
supported by the German Federal Ministries for Education, Science,
Research, and Technology (BMBF) under grant 032 69 95A-D and for
Economics and Technology (BMWi) under grant 0327095.

References
Figure 3. Comparison of simulation results with data. Top panel:
Simulated final porosity (bold dashed line), permeability (bold full
line), and data from measurements on core specimens (horizontal
bars) and with the ring permeameter (fine line); Bottom panel:
Simulated mineral content before core flooding (long dash), after
118 hours of flooding (short dash), and after core flooding (20 days)
(solid line) compared to data measured on core specimens after
flooding (horizontal bars with error bars). The length of the
horizontal bars in both panels corresponds to the size of the
analyzed core specimen.
transport are equilibrated more or less instantaneously by dissolution
or precipitation. As a consequence, the largest permeability reduction is found where the gradient of temperature is largest.
[22] A second justification for neglecting the influence of
reaction kinetics in simulations of anhydrite sealing or flushing
during the production of hot water from deep reservoirs, is provided
by the good agreement between measured and simulated rock
properties: Both permeability increase and reduction occur, in the
experiment as well as in the simulation. The net effect of this
simultaneous improvement and deterioration of reservoir properties
is discussed in more detail in companion papers [Kuhn et al., 2002;
Kuhn et al., Long term behavior of reservoir properties of a
geothermally used aquifer - Interaction between flow, heat transfer,
transport, and chemical reactions, submitted to Geothermics, 2001].
[23] Finally, this simulation demonstrates that our improved
chemical reaction model implemented in SHEMAT and verified by
solubility data [Kuhn et al., 2002] performs correctly also for
complex flow situations typical for managed hot water reservoirs,
such as reactive flow of high salinity brines at varying temperature
and elevated pressure.
[24] The k-f-relation (Equation (2)) is applied twice at crucial
points of our model validation procedure. It is used to relate the
distribution of permeability (the only continuously logged property
in the experiment) to the prognostic quantity of the simulation
model, the distribution of the mineral concentration: (1) for
initializing the model (long-dashed curve in Figure 3), and (2)
for a comparison with data, as the variations of permeability in the

Bartels J., M. Kuhn, H. Pape, and C. Clauser, A New Aquifer Simulation

Tool for Coupled Flow, Heat Transfer, Multi-Species Transport and Chemical Water-Rock Interactions, in Proceedings of World Geothermal
Congress 2000, Kyushu Tohuku, Japan, May 28 June 10, edited by
E. Iglesias, D. Blackwell, T. Hunt, J. Lund, and S. Tamanyu, 3997
4002, 2000.
Clauser, C., and S. Kiesner, A Conservative, Unconditionally Stable, Second-Order Three Point Differencing Scheme for the Diffusion Convection Equation, Geophys. J. R. Astr. Soc., 91, 557 568, 1987.
Clauser, C., and H. Villinger, Analysis of Conductive and Convective Heat
Transfer in a Sedimentary Basin, Demonstrated for the Rheingraben,
Geophys. J. Int., 100(3), 393 414, 1990.
Kuhn, M., W. Schneider, J. Bartels, H. Pape, and C. Clauser, Numerical
simulation of chemically induced permeability changes during heat
mining in hot aquifers, in Proceedings of 21st New Zealand Geothermal
Workshop 1999, Auckland, November 10 12, 1999, edited by S. F.
Simmons, O. E. Morgan, and M. G. Dunstall, 223 228, Auckland University, Auckland, New Zealand, 1999.
Kuhn, M., J. Bartels, H. Pape, W. Schneider, and C. Clauser, Modeling
chemical brine-rock interaction in geothermal reservoirs, in Water Science
And Technology Library, Vol. 40: Water Rock Interaction, edited by
I. Stober and K. Bucher, 147 165, Kluwer Publishers, Dordrecht, 2002.
Pape, H., C. Clauser, and J. Iffland, Permeability prediction based on fractal
pore-space geometry, Geophysics, 64(6), 1447 1460, 1999.
Pape, H., C. Clauser, and J. Iffland, Variation of permeability with porosity
in sandstone diagenesis interpreted with a fractal pore space model, Pure
Appl. Geophysics, 157, 603 619, 2000.
Plummer, L. N., D. L. Parkhurst, G. W. Fleming, and S. A. Dunkle, A
computer program incorporating Pitzers equations for calculation of
geochemical reactions in brines, U.S. Geological Survey Water-Resources Investigations Report 88-4153, 310 pp., 1988.
Power, W. H., B. M. Fabuss, and C. N. Satterfield, Transient Solute Concentrations and Phase Changes of Calcium Sulfate in Aqueous Sodium
Chloride, Chem. Eng. Data, 11(2), 149, 1966.

J. Bartels, Geothermie Neubrandenburg Ltd., Postfach 110120, D-17041

Neubrandenburg, Germany.
C. Clauser and H. Pape, Aachen University of Technology (RWTH),
Applied Geophysics, Lochnerstr. 4-20, D-52056 Aachen, Germany.
M. Kuhn and W. Schneider, Technical University of Hamburg-Harburg,
Department of Water Management and Water Supply, Schwarzenbergstr.
95, D-21073 Hamburg, Germany.
I. Lajcsak and V. Meyn, German Petroleum Institute (If E), WalterNernst-Str. 7, D-38678 Clausthal-Zellerfeld, Germany.