Simmons 2001

Journal of Contaminant Hydrology 52 Ž2001.
245–275
www.elsevier.comrlocaterjconhyd
Variable-density groundwater flow and solute

transport in heterogeneous porous media:
approaches, resolutions and future challenges
Craig T. Simmons a,) , Thomas R. Fenstemaker b,
John M. Sharp Jr. b
a
School of Chemistry, Physics and Earth Sciences, Flinders UniÕersity of South Australia,
GPO Box 2100 Adelaide, SA 5001, Australia
b
Department of Geological Sciences, The UniÕersity of Texas at Austin, Austin, TX 78712-1101, USA
Abstract
In certain hydrogeological situations, fluid density variations occur because of changes in the
solute or colloidal concentration, temperature, and pressure of the groundwater. These include
seawater intrusion, high-level radioactive waste disposal, groundwater contamination, and geother-
mal energy production. When the density of the invading fluid is greater than that of the ambient
one, density-driven free convection can lead to transport of heat and solutes over larger spatial
scales and significantly shorter time scales than compared with diffusion alone. Beginning with
the work of Lord Rayleigh in 1916, thermal and solute instabilities in homogeneous media have
been studied in detail for almost a century. Recently, these theoretical and experimental studies
have been applied in the study of groundwater phenomena, where the assumptions of homogeneity
and isotropy rarely, if ever, apply. The critical role that heterogeneity plays in the onset as well as
the growth andror decay of convective motion is discussed by way of a review of pertinent
literature and numerical simulations performed using a variable-density flow and solute transport
numerical code. Different styles of heterogeneity are considered and range from continuously
AtrendingB heterogeneity Žsinusoidal and stochastic permeability distributions. to discretely frac-
tured geologic media. Results indicate that both the onset of instabilities and their subsequent
growth and decay are intimately related to the structure and variance of the permeability field.
While disordered heterogeneity tends to dissipate convection through dispersive mixing, an
ordered heterogeneity Že.g., sets of vertical fractures. allows instabilities to propagate at modest
combinations of fracture aperture and separation distances. Despite a clearer understanding of the
processes that control the onset and propagation of instabilities, resultant plume patterns and their
)
Corresponding author.
E-mail address: craig.simmons@es.flinders.edu.au ŽC.T. Simmons..
0169-7722r01r$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 9 - 7 7 2 2 Ž 0 1 . 0 0 1 6 0 - 7
246 C.T. Simmons et al.r Journal of Contaminant Hydrology 52 (2001) 245–275
migration rates and pathways do not appear amenable to prediction at present. The classical
Rayleigh number used to predict the occurrence of instabilities fails, in most cases, when
heterogeneous conditions prevail. The incorporation of key characteristics of the heterogeneous
permeability field into relevant stability criteria and numerical models remains a challenge for
future research. q 2001 Elsevier Science B.V. All rights reserved.
Keywords: Variable density; Groundwater flow; Solute transport; Heterogeneity; Rayleigh number; Instability
1. Introduction
Over the years, there has been considerable interest in the migration of solutes in
groundwater. Many of the current pressing problems in groundwater and of environmen-
tal concern involve transport issues including groundwater contamination, seawater
intrusion in coastal aquifers, radioactive waste disposal, geothermal energy development,
groundwater–surface water interaction, and subsurface storage of materials, fluids, and
energy.
Field and laboratory studies have shown that fluid density gradients caused by
variations in concentration andror temperature can play an important role in the
transport of solutes in such groundwater systems. When the density of the invading
plume is significantly greater than that of the ambient groundwater, density-driven flow
can manifest itself in the form of lobe-shaped instabilities or fingers. The density
stratification, where a dense fluid overlies a less dense one, is the main cause of the
instability Žfree convection.. The instability mechanisms have been studied for almost a
century now and are called, variously, Benard, Rayleigh, solute, or thermal instability.
This field of fluid mechanics probably began in 1900 when Benard observed hexagonal
patterns in an unstably stratified horizontal fluid layer heated from below. Rayleigh
Ž1916. then analysed an idealised model of such instability and determined the charac-
teristics of the fluid layer that caused the instability to occur. The dimensionless number
most commonly used in fluid mechanics to predict and describe the occurrence of
instability, the Rayleigh Number, was named in his honor.
The free convection process enhances hydrodynamic mixing of the dense solute with
the less dense ambient groundwater. It is significant for three main reasons: Ž1. the total
quantity of solute transport involved in the convective mixing process is typically far
greater than that of diffusional transport, Ž2. the time scales associated with the mixing
process are significantly reduced, and Ž3. the dimensions of the mixing zone are
typically larger and, therefore, enable solutes to spread over much greater distances.
There are many examples described in the literature that demonstrate the importance
of both stable and unstable variable-density groundwater flow and solute transport.
These include intrusion of seawater in coastal aquifers ŽHuyakorn et al., 1987; Pinder
and Cooper, 1970; Voss and Souza, 1987.; infiltration of leachates from waste disposal
sites ŽFrind, 1982; Oostrom et al., 1992a,b; Schincariol and Schwartz, 1990.; transport
of salts due to agricultural practices ŽMulqueen and Kirkham, 1972.; groundwater flow
through salt formations in connection with high-level radioactive waste disposal ŽHas-
sanizadeh and Leijnse, 1988; Herbert et al., 1988.; and accumulation and subsequent
C.T. Simmons et al.r Journal of Contaminant Hydrology 52 (2001) 245–275 247
downward migration of salt near the land surface of a saline lake or playa ŽSimmons and
Narayan, 1997; Wooding et al., 1997a,b.. Most of these studies, however, assume
homogeneous and isotropic hydrogeologic settings.
In this study, we investigate the critical role that heterogeneity plays in the onset as
well as the growth andror decay of free convective motion. We begin by presenting a
brief review of the concepts and approaches that have been used in the study of
heterogeneity and variable-density flow. The small body of research work that has been
carried out to examine the relationship between hydrogeologic heterogeneity and the
nature of density-induced instability is then described.
Fig. 1. Different styles of geologic media: Ža. homogeneous porous media, Žb. heterogeneity represented by a
stochastic random spatial function, Žc. continuous periodic sinusoidal heterogeneity, Žd. fractured geologic
media represented by regular vertical fracturing and Že. fractured geologic media with nonuniform fracture
aperture, separation distances and orientation.
In order to further examine variable-density flow and transport processes in heteroge-

neous geologic media, we perform three numerical studies using the Saturated–Un-
saturated TRAnsport ŽSUTRA. variable-density groundwater flow and solute transport
code ŽVoss, 1984.. The style of heterogeneity varies from seemingly random heteroge-
neous porous media through to a more regular periodic style of continuous heteroge-
neous porous media and, finally, a discretely fractured system ŽFig. 1.. More specifi-
cally, the numerical studies performed are: Ž1. variable-density flow and transport
beneath a contaminant source with a sinusoidal hydraulic conductivity distribution for
the continuous heterogeneous hydraulic conductivity field, Ž2. a variant on Ž1. that
employs continuously AtrendingB heterogeneity in hydraulic conductivity represented by
a random spatial function ŽRSF. with known stochastic parameters, and Ž3. the develop-
ment of convective instabilities in fractured rock of low matrix permeability relevant to
the formation, maturation, and migration of hydrocarbons in the Gulf of Mexico
sedimentary basin of North America. The similarities and differences between the
developing convection patterns in these three heterogeneous systems are discussed to
elucidate the key features of heterogeneous permeability fields that control the onset,
growth, and decay of the instability process.
Finally, we examine the applicability of classical Rayleigh stability criteria for
predicting the onset conditions for density-driven convection in heterogeneous porous
media. The numerical examples show that the Rayleigh number is not a useful indicator
of instability in heterogeneous porous systems. We revisit the assumptions made in the
derivation of the Rayleigh number in order to understand why its application may be
limited in many groundwater studies.
2. Heterogeneity and variable-density flow: a review of concepts and approaches
Below, we review briefly heterogeneity in geologic systems, variable-density flow

phenomena, and how they interrelate.
2.1. Heterogeneity
Geologic systems are not truly homogeneous or uniform. Therefore, single-number

representations of porous media are not true reflections of reality. This concept, although
widely accepted, is commonly simplified or ignored because of our limited ability to
characterise and observe detailed hydraulic conductivity distributions in actual forma-
tions over large spatial scales. As a result, we cannot know the exact nature of the
hydraulic conductivity variations. Recent experimental evidence and theoretical results
suggest that the transport of passive solutes in natural porous formations is controlled by
the spatial variations in heterogeneous conductivity fields Žsee Dagan, 1989 for an
exhaustive review..
In recent years, significant research has been devoted to the development of system-
atic and predictive modeling techniques that explicitly account for natural heterogeneity.
The underlying reason for the move toward a stochastic approach was the realisation
that hydrogeological environments are exceedingly heterogeneous and, even in areas
where considerable data are available, we can never fully observe the exact spatial
variations. In the stochastic approach, a statistical model is adopted whereby input
parameters, such as hydraulic conductivity, are considered to be random spatial func-
tions with an associated probability density function for the entire flow domain.
Numerous studies have used this approach in groundwater flow and solute transport
modeling Že.g., Bakr et al., 1978; Dagan, 1984, 1986, 1989, 1990, 1994; Freeze, 1975;
Gelhar, 1986, 1987, 1993; Gelhar and Axness, 1983; Graham and McLaughlin, 1991;
Indelman and Abramovich, 1994; Neuman et al., 1987; Smith and Freeze, 1979; Smith
and Schwartz, 1981, 1980; Tompson and Gelhar, 1990.. Stochastic approaches can
provide a quantitative description of bulk hydraulic conductivity behaviour over large
spatial scales and account for the inherent spatial variability. Fundamental experimental
validation from accurate field analyses and data collection support the validity of
stochastic theory, at least for mildly heterogeneous aquifers, such as the Borden aquifer
ŽGraham and McLaughlin, 1991; Sudicky, 1986. and Cape Cod ŽLe Blanc et al., 1991..
A variety of modeling approaches has been developed to simulate groundwater flow
and solute transport in fractured rock aquifers. These include equivalent porous media,
double porosity and discrete fracture methodologies ŽSharp, 1993.. There is a vast body
of work that has examined flow and solute transport in discretely fractured geologic
media ŽBerkowitz et al., 1988; Grisak and Picken, 1980, 1981; Harrison et al., 1992;
Robinson et al., 1998; Smith and Schwartz, 1984; Sudicky and Frind, 1982; Therrien
and Sudicky, 1996..
2.2. Variable-density flow phenomena
Groundwater contamination often occurs when surface-based pollution sources re-

lease considerable amounts of heavy leachates into the soil. The higher density of these
contaminants often results in modified transport behaviour once the plume penetrates
into the groundwater region. Freeze and Cherry Ž1979, p. 435. note that the total
dissolved solid concentrations for leachate from sanitary landfills typically range from
5000 to 40,000 mg ly1 Ži.e., approximately 0.35–2.8% density difference between the
leachate and the groundwater.. This density difference is sufficiently high to cause the
leachate to sink as it penetrates into and flows with the groundwater. When density
differences are significant, solute transport is the result not only of forced Žhydraulically
driven. convection Žalso commonly called advection. and dispersionrdiffusion, but also
of free conÕection ŽGhebart et al., 1988.. Such mixed conÕectiÕe flow regimes occur
when both hydraulically driven transport and buoyancy-driven transport operate together
to control solute concentrations. Some mixed convective systems can be characterised
by instability development that superimposes perturbations in concentration distributions
on the mixed flow. The density stratification Žwhere a dense fluid overlies a less dense
one. is the main cause of the instability and the process of instability causes the fluids to
mix to achieve a stable density gradient ŽSchincariol and Schwartz, 1990..
Historically, researchers in the field of unstable convective flows have categorised
these flows into broad categories. For example, instabilities or convective flows within a
bounded region, or layer, have been called Benard, Rayleigh, or thermal instabilities
ŽGhebart et al., 1988.. Variations on this theoretical problem have been studied for
nearly a century Že.g., Combarnous and Bories, 1975; Elder, 1967; Horton and Rogers,
1945; Lapwood, 1948; Pratts, 1966; Rayleigh, 1916. and the area of convection in
porous media continues to be an active area of research. In their comprehensive text on
this subject, Nield and Bejan Ž1999. report that research on convection in porous media
continues to be published at a rate of over 100 papers per year.
Saffman and Taylor Ž1958. examined the stability of a fluid–fluid interface along
which fingers may develop to form a corrugated interface. They examined the stability
of the interface between immiscible fluids moving vertically through a porous medium.
These concepts of interfacial instabilities were then generalised to miscible systems by
Wooding Ž1962., who examined the effects of longitudinal dispersion across the
interface of two miscible fluids. Wooding, 1959, 1962, 1963, 1969. and Bachmat and
Elrick Ž1970. provide some of the earliest work on convective instability in porous
media. Biggar and Nielsen Ž1964., Krupp and Elrick Ž1969., and Rose and Passioura
Ž1971. performed column displacement studies with variable-density salt solutions.
These studies confirmed that the occurrence of instabilities was related to density
differences and the average pore-water velocities. Another notable achievement in the
field of free convection was the Elder Ž1967. Ashort heater problemB that examined
transient thermal convection in a Hele-Shaw cell. In the Elder experiment, part of the
base of the Hele-Shaw cell was heated from below and the development of convection
cells was observed as they grew with time. Interestingly, the Elder Ž1967. problem is
now commonly used as a test of the accuracy and reliability of variable-density
groundwater flow and solute transport simulators Že.g., Kolditz et al., 1998; Oldenburg
and Pruess, 1995; Simmons et al., 1999a; Voss and Souza, 1987..
Most mass transport studies in groundwater have focused on transport that occurs
primarily by forced convection Žadvection.. Paschke and Hoopes Ž1984. investigated
buoyant plume behaviour of dense aqueous phase liquids in an analogous manner to that
used for plumes and jets discharging into an ambient cross-flow. They studied the effect
of variable density on the trajectory of the plumes, but an investigation of plume
stability was not included. To date, only List Ž1965. has theoretically analysed interfa-
cial instabilities for miscible fluids with density stratification in a horizontal flow field.
In that study, dense solutions were injected into a confined homogeneous aquifer model
to study the stability of fluids of different densities during horizontal flow through an
isotropic medium. List Ž1965. found that when a heavier fluid overlies a lighter fluid,
the system is always unstable. However, under certain conditions, the growth rate is
bounded so that quasi-stable plumes develop. In general, by virtue of the analytical
approach and its need for simplifying assumptions, List’s analysis is limited in its
general applicability.
Based on the behaviour of some leachate plumes in the Netherlands, van der Molen
and van Ommen Ž1988. found that unstable fluid density stratification in geological
systems is probably more common than is usually assumed. Investigating the plume
behaviour under landfills in New York, Kimmel and Braids Ž1980. found that highest
salt concentrations occurred at the bottom of the aquifer. They made the assumption that
the high-density leachate flowed out of the landfill as pockets of high-density fluid after
recharge, and moved diagonally downward through the aquifer. Le Blanc Ž1984. studied
the movement of a sewage plume at Cape Cod and partly attributed its vertical
movement to density differences. However, not every mixed convective system will
result in gravitational convective instabilities and their development will depend upon
the relative magnitudes of the hydraulic and buoyancy-driven forces ŽRaffensperger and
Vlassopoulos, 1999.. Convective instabilities have recently been studied in relation to
groundwater contamination problems, where a more dense plume is enclosed by and is
moving along in a body of less dense fluid Že.g., Fan and Kahawita, 1994; Koch and
Zhang, 1992; Liu and Dane, 1996; Oostrom et al., 1992a,b; Schincariol and Schwartz,
1990; Schincariol et al., 1994, 1997; Zhang and Schwartz, 1995.. Oostrom et al.
Ž1992a,b. studied the behaviour of dense surface injected leachate plumes in homoge-
neous porous media. Their results show that for a given porous medium, the magnitude
of horizontal flow velocity, the contaminant leachate rate, and the density difference
between the contaminant solution and the ambient groundwater influenced the stability
of the plumes. Simmons and Narayan Ž1997. examined the onset conditions for
gravitational plumes to develop beneath a salt disposal basin. Wooding et al. Ž1997b.
performed laboratory experiments of an idealised salt lake using a Hele-Shaw cell
system. The boundary layer beneath the salt lake was seen to be unstable and resulted in
downward convection of salt fingers. Simmons et al. Ž1999a. performed numerical
simulations of the Wooding et al. Ž1997b. experiment and found that the numerically
generated plume patterns and their growth rates were in good spatial and temporal
agreement with those seen in the laboratory experiment. The study revealed that
numerical errors in the solute transport code could trigger unrealistic instabilities to form
and also alter their subsequent shape and growth rates. This was also a major conclusion
of Schincariol et al. Ž1994.. Simmons et al. Ž1999a. emphasised the need for proper
spatial and temporal discretisation, particularly in the simulation of unstable convective
flow phenomena and encouraged further intercode comparisons and benchmarks for
variable-density flow simulators.
2.3. Variable-density flow in heterogeneous systems
In variable-density flow systems, heterogeneity in hydraulic properties can perturb

flow over many length scales, ranging from slight differences in pore geometry to larger
heterogeneities at the formational or regional scale, to generate instabilities in density
stratified systems. Few studies have examined the way in which heterogeneities in
hydrogeologic properties can promote the onset, growth andror decay of density-driven
instabilities. McKibbin and O’Sullivan Ž1980. and McKibbin and Tyvand Ž1983, 1982.
investigated the effect that multiple hydraulic conductivity layers had on thermal
convection. They showed that, as the number of layers increased, the system approached
the behaviour of an equivalent anisotropic, homogeneous single layer. However, each
layer was assumed homogeneous in these studies.
Schincariol and Schwartz Ž1990. carried out an experimental investigation of vari-
able-density groundwater flow in homogeneous, layered, and lenticular media. They
found that in a layered medium, reductions in hydraulic conductivity of half an order of
magnitude or less influence the flow of the dense plume. Dense water tended to
accumulate along bedding interfaces. They noted that while density differences as low as
0.0008 grcm3 were sufficient to produce gravitational instabilities in a lenticular
medium, the combination of convective dispersion and nonuniform flow due to hetero-
geneities caused relatively large dispersion that tended to dissipate instabilities.
Schincariol et al. Ž1994. numerically reproduced the experimental results of Schincar-
iol and Schwartz Ž1990. and examined how the style of interfacial perturbations
controlled the pattern of instability development. They showed that whether or not
perturbations became unstable depended not only on the Rayleigh number of the system,
but also upon the perturbation wavelength, which must exceed some critical wavelength
determined by flow and transport parameters. Schincariol et al. Ž1997. carried out
numerical simulations incorporating the effects of heterogeneity in the hydraulic conduc-
tivity field and found that the statistical characteristics of the permeability field Žmean,
variance, correlation length scales. played a critical role in the onset and subsequent
growth or decay of instabilities, and must be incorporated in the stability criteria.
Schincariol et al. Ž1997. concluded from their simulations that, essentially, the more
anisotropic and heterogeneous a permeability field is, the greater the dampening of
instability growth. This should be interpreted as true only once a perturbation of a long
enough wavelength has been propagated. They also state that while heterogeneities on
all scales probably dampen instability growth once an unstable perturbation has been
propagated, heterogeneities in the porous medium provide the most likely source of
initial perturbations to the plume. In the study of Schincariol et al. Ž1997., random
permeability fields were used for the first time in connection with gravitational
instabilities. Schincariol Ž1998. studied natural perturbation initiation by local-scale
heterogeneities and concluded that based on the controlling influence local field
characteristics exhibit on instability growth and decay, the applicability of stability
criteria to natural or field-scale porous media is likely inappropriate. Schincariol Ž1998.
states that homogeneous field criteria will not be applicable. There are still no criteria
for describing the onset of instabilities as well as their subsequent growth or decay in
heterogeneous porous media in the scientific literature.
In a numerical study that examined density-dependent solute transport in discretely
fractured geologic media, Shikaze et al. Ž1998. found that fractures with aperture values
as small as 25 mm can significantly enhance contaminant migration in the saturated zone
relative to the case where the fractures are absent. This is also demonstrated analytically
by Simmons et al. Ž1999b.. Shikaze et al. Ž1998. concluded that, in typical field settings,
there is a high degree of nonuniformity in fracture, apertures and spacings that the
migration rates, directions, and patterns of dense plumes might not be amenable to
prediction. Based upon the conclusions presented by Schincariol et al. Ž1997. and
Shikaze et al. Ž1998., it appears reasonable to suggest that heterogeneity can either
enhance plume migration rates or dissipate them. The present study suggests that what is
critical is the careful delineation of the style and geometrical structure of the heteroge-
neous permeability field.
3. Dimensionless parameters as indicators of stability
Consider an unstably stratified system where a fluid of higher density overlies a fluid
of lower density. If the density difference is sufficiently high, instabilities form in which
lobes of dense fluid move downwards, counterbalanced by less dense fluid moving
upward. In simple free convective systems where mechanical dispersion is assumed
independent of convective flow velocity, the onset of instability is determined by the
value of a nondimensional number called the Rayleigh number Ž Ra.. This dimensionless
number is the ratio between buoyancy-driven forces and resisting forces caused by
diffusion and dispersion.
Ra is defined by:
Uc H gk b Ž Cmax y Cmin . H Buoyancy and Gravitation

Ra s s s , Ž 1.
D0 un 0 D 0 Diffusion and Dispersion
where Uc is the convective velocity; H is the thickness of the porous layer; D 0 is the
molecular diffusivity; g is the acceleration due to gravity; k is the intrinsic permeabil-
ity; b s ry1
0
ŽE rrEC . is the linear expansion coefficient of fluid density with changing
fluid concentration; Cmax and Cmin are the maximum and minimum values of concentra-
tion, respectively; u is the aquifer porosity, n 0 s m 0rr 0 is the kinematic viscosity of the
fluid. The choice of characteristic length scale, H, in the Rayleigh number is often
difficult and ambiguous. This is discussed in further detail in Section 5.4.
For sufficiently high Rayleigh numbers greater than some critical Rac , gravity-based
instability will occur in the form of waves in the boundary layer that develop into
fingers or plumes. This critical Rayleigh number defines the transition between disper-
siverdiffusive solute transport Žat lower Rayleigh numbers. and convective transport by
density-driven fingers Žat higher Rayleigh numbers.. Instabilities observed in the labora-
tory typically take the form of fingers, first of a very small wavelength, which with time
become dominated by several large fingers ŽGreen and Foster, 1975; Wooding, 1969..
4. Mathematical modeling
Below, we discuss the numerical model used for simulations and permeability
distributions for the continuous periodic models, stochastic permeability models, and
fractured low-permeability shales.
4.1. The SUTRA model
The Saturated–Unsaturated TRAnsport ŽSUTRA. model developed by Voss Ž1984.

simulates two-dimensional Ž2-D. density-dependent groundwater flow and solute trans-
port. SUTRA employs a hybrid finite-element mesh discretisation and integrated finite
difference time-stepping method to approximate the governing equations for the coupled
processes being simulated. SUTRA solves the governing equations of ‘conservation of
mass of fluid’ and ‘conservation of mass of salt’ and computes flow velocities using
Darcy’s Law. The governing equations are briefly discussed below and are discussed
more fully in Voss Ž1984. and Voss and Souza Ž1987..
4.1.1. Mass conserÕation

The fluid mass balance equation is:
E Ž ur .
s y= P Ž ur V . q Q p , Ž 2.
Et
where u Ž x, y, t . is a dimensionless porosity; r Ž x, y, t . is the fluid density; V Ž x, y, t .
is average fluid velocity; Q p Ž x, y, t . is a fluid mass source; x and y are spatial
coordinates; t is time; and = is the divergence operator.
The coupling between density and concentration is assumed to be a linear function of
the form:
Er
r s r0 q Ž C y C0 . , Ž 3.
EC
where r 0 is the fluid density at a base concentration, C0 ; and E rrEC is a constant
coefficient of density variability.
With variable-fluid density, the fluid flow equation is expressed in terms of the
pressure variable since the potential head function does not exist. The pressure gradient
form of Darcy’s Law is:
k
Vsy
ž /
um
P Ž =p y r g . , Ž 4.
where pŽ x, y, t . is the fluid pressure; g is the gravity vector; m is fluid dynamic

viscosity; and k Ž x, y . is the intrinsic permeability tensor.
4.1.2. Salt conserÕation

For a single species stored in solution, the solute mass balance equation may be
expressed as:
E Ž ´r C .
s y= P Ž ´r VC . q = P ´r Ž D 0 I q D . P =C q Q p C ) , Ž 5.
Et
where D 0 is the apparent molecular diffusivity in a porous medium of solutes in
solution, I is the dimensionless identity tensor, C ) is the concentration of fluid sources
expressed as a mass fraction. Bear Ž1979. has formulated the components of the
mechanical dispersion tensor, D, to account for both transverse and longitudinal
dispersivities, respectively.
It can be seen that Eqs. Ž2., Ž4. and Ž5. are coupled since density is used to compute
the velocity distribution in the flow field. The density and velocity terms are used in the
fluid mass balance and solute mass balance equations and the solution for the concentra-
tion field is then in turn used to compute the density distribution.
4.2. Continuous periodic and stochastic hydraulic conductiÕity distributions
4.2.1. Conceptual model

Previous studies by Schincariol and Schwartz Ž1990. and Schincariol et al. Ž1994,
1997. considered the sideways injection of a dense contaminant plume into a horizontal
ambient velocity. In this study, we use a simplified conceptual model that considers a
dense contaminant source that lies on the surface of the model. A horizontal ambient
velocity is not considered so that only free convective Žand not mixed convective.
systems can result. This simplifies the resultant processes, analysis and interpretation
and allows us to better understand the way in which heterogeneity controls free
convection.
Consider the conceptual model shown in Fig. 2. A dense fluid contaminant source
with concentration, C U , overlies groundwater of initial concentration, C L , and is located
on the surface of the model with position described mathematically by x L - x - x R ,
where x is the horizontal Cartesian coordinate relative to the left-hand boundary. This
configuration could be representative of a number of possible contaminant sources,
including leachate from a surficial landfill, or brine movement from a salt lake, or a
saline disposal basin.
A cross-sectional slice with depth, H s 10 m, and horizontal length scale, L s 250
m, was used. The boundary conditions for flow and transport are also shown in Fig. 2.
All boundaries are considered to be no flow, except the dense fluid source. Solute may
neither advect nor disperse across these no flow boundaries. Initially, pressure through-
out is hydrostatic with an initial concentration in the aquifer given by C L .
4.2.2. Spatial and temporal discretisation

For the cases studied, the spatial domain was discretised to form a completely
uniform mesh containing 2761 nodes and 2500 square elements, by choosing 250 finite
Fig. 2. Simulation geometry, boundary and initial conditions for the simulation of a dense contaminant plume
source. Here, permeability is considered to be a function k Ž x, y . or is represented by a random spatial
function ŽRSF..
elements horizontally Ž D x s 1 m. and 10 finite elements vertically Ž D y s 1 m.. All

transient simulations were run with 100 time steps to cover a simulation time of about
15 years. Time-step sizes varied between 1-month steps at earlier simulation times and
1-year steps at later simulation times. Several other spatial and temporal discretisations
were tested to ensure that convergence in the solution had been achieved. To assess this,
solute mass flux across the upper constant concentration boundary as well as the
resultant concentration field plots were examined at a number of different spatial and
temporal discretisations. Voss Ž1984. notes that numerical oscillations in a finite element
solution can be virtually eliminated if the element size is chosen such that D L F 4a L ,
where D L is the local distance between sides of an element measured in the direction
parallel to local flow and a L is the longitudinal dispersivity. This is an empirical
measure used to estimate adequate mesh sizes based upon the chosen values of
dispersivity. The spatial discretisation employed in the study easily satisfied this criteria.
4.2.3. Hydraulic conductiÕity distributions

To test the effect of heterogeneity structure and, in particular, the concept of a critical
perturbing horizontal wavelength, the hydraulic conductivity value at any given point x
and for all y was assigned using:
2p x
K x , y Ž x , y . s K AV 10 c Žsin l
.
. Ž 6.
Any heterogeneous hydraulic conductivity distribution can be defined in terms of its
average conductivity, K AV , its amplitude, c Žrepresenting the order of magnitude
variation around the average value., and the wavelength of the periodic structure, l.
Also note that unless otherwise specified, the formations are fully isotropic as described
by the notation K x, y . Although Eq. Ž6. represents a continuous hydraulic conductivity
distribution, in the numerical model, this continuity is limited by choice of numerical
discretisation. Thus, numerically, the distribution is represented by a series of step
functions whose width are related to the horizontal discretisation, D x.
We now turn to stochastic hydraulic conductivity distributions. Gelhar Ž1993. pro-
vides a review of the theory underlying stochastic groundwater hydrology. A number of
studies Že.g., Freeze, 1975; Law, 1944; McMillan, 1966. suggest that the probability
density function for hydraulic conductivity is lognormal. If the hydraulic conductivity K
is log-normally distributed, we can define a new parameter Y s log K that is normally
distributed; i.e., Y: N w m y , sy x where m y is its mean and sy its standard deviation.
Therefore, a normal distribution, Y, with sample size given by the number of elements in
the mesh can be formed with a given mean, m y , and a given standard deviation, sy , and
then converted to the log-normal hydraulic conductivity distribution using the exponen-
tial transform, K s e 2.3Y.
For hydraulic conductivity distributions represented by a random spatial function, we
must acknowledge that these fields are not random and that structure and correlations do
exist between hydraulic conductivities throughout the flow domain. The spatial depen-
dence between neighbouring values of the random variable can be defined in terms of a
stochastic process model that defines the spatial autocorrelation, or alternatively, its
variogram throughout the system. The variogram, g Ž h., is half the average squared
difference between the paired data values separated by a distance, h, and is used to
describe the spatial continuity of a variable, which, in the present case, is hydraulic
conductivity, K. The total number of paired points separated by a distance, h, is N Ž h..
The variogram model can be written mathematically as:
1 2
g Ž h. s Ý Ž KiyKj. . Ž 7.
2 N Ž h. Ž i , j .Nh i jsh
In this study, the exponential model, a commonly used variogram is used. Its
standardised equation in one direction or dimension is:
h
g Ž h . s 1 y exp y ž / t
. Ž 8.
This variogram model reaches its sill or plateau asymptotically, with the practical
range 3t being defined as that distance at which the variogram value is 95% of its
plateau or sill. For h ) 3t , pairs of data are uncorrelated.
In a two-dimensional vertical slice, it is necessary to specify an anisotropy factor.
Geologic considerations suggest that in describing the spatial dependence between
neighbouring conductivity values, statistically anisotropic variograms are probably the
rule rather than the exception. For example, conductivity values within any one
stratigraphic layer are likely to be more highly correlated with each other than with
conductivity values outside the layer. This typically requires the horizontal range, 3t H ,
within any system to be larger than the vertical range, 3t V , and the specification of an
anisotropy factor, anis, which is given by anis s t H rt V .
4.2.4. Model parameters

The list of model parameters is given in Table 1. These parameters were held
constant throughout all simulations unless otherwise specified. Parameters, such as k AV ,
C U , l, and c , were allowed to vary in specific simulations for the sinusoidal hydraulic
conductivity distributions. For the stochastic hydraulic conductivity distributions, m y ,
sy , t H , and anis s t H rt V were varied between simulations.
4.3. Fractured low-permeability shales: Gulf of Mexico case study
4.3.1. Introduction
The Gulf of Mexico sedimentary basin is a major region of natural resources
production, including hydrocarbons, uranium, lignite, and numerous industrial minerals
ŽSalvador, 1991.. It has been penetrated by many thousands of exploratory drill holes.
Nevertheless, significant questions remain about fundamental processes of mass and
energy transport, such as how can the fluid flow and solute transport required to account
for high levels of sediment diagenesis occur in sedimentary basins dominated by
low-permeability shaly sediments Že.g., Land, 1997, 1991; Raffensperger and Vlas-
sopoulos, 1999; Sharp et al., 1988.. Diffusive transport of diagenetic reactants and
products, which cannot account for the mass fluxes required for diagenesis and
density-driven free convection, has been suggested by several authors Že.g., Heydari,
Table 1
Model parameters used in sinusoidal and stochastic permeability studies
Freshwater density Ž r . s1000 kg my3
Groundwater concentrations 0 mg ly1
Lake concentration Ž C U . sC U mg ly1
Fluid dynamic viscosity Ž m . s10y3 kg my1 sy1
Coefficient of fluid density change ŽE r rEC . s 700 kg my3
Water compressibility s 4.5=10y1 0 Pay1
Soil compressibility s1=10y8 Pay1
Porosity Ž ´ . s 0.35
Intrinsic permeability Ž k . s k Ž x, y . m2
Longitudinal dispersivity Ž a L . s1 m
Transverse dispersivity Ž a T . s1 m
Molecular diffusivity Ž D 0 . s10y9 m2 sy1
Acceleration due to gravity Ž g . s9.81 m sy2
Aquifer depth Ž H . s10 m
Aquifer length Ž L. s 250 m
Basin length Ž L B . s 200 m
Coordinate left basin edge Ž x L . s 25 m
Coordinate right basin edge Ž x R . s 225 m
2000; Land, 1991; Wood and Hewett, 1984. to account for observed levels of sediment
diagenesis. Land Ž1991., Morton and Land Ž1987., and Sharp et al. Ž1988. note salinity
inversions in the Gulf of Mexico Basin in which more saline waters Žincluding brines of
) 100,000 ppm total dissolved solids. in formations overlie less saline formation waters
Žabout 20,000 ppm. as shown in Fig. 3. The high-salinity fluids are perched above or
just below the top of overpressure within the transition zone. There are few salt domes
in this area and data indicate that some of these brines are derived from the deep
Mesozoic rocks of the Basin. We contend that AburpsB of deep hot saline fluid are
expelled vertically upwards along fault zones episodically, most notably along the
Wilcox and Frio Fault zones ŽFig. 3.. It appears that the brines are not commonly found
in the zone of extreme overpressures Žalthough deep salinity data are sparse., except at
depths not yet generally penetrated by drilling. Where these inversions occur, a strong
buoyancy gradient exists because these shallower pore waters are typically both cooler
and more saline than the deeper pore waters. Because shales dominate the sedimentary
sequence of the Gulf of Mexico Basin, defining their role on limiting free convection is
critical.
Understanding these processes is important for explaining: Ž1. diagenetic effects that
control reservoir porosity and permeability; and Ž2. fluid flow and solute transport in the
context of petroleum maturation and migration. In this part of the study, we analyze the
potential for salinity-driven free convection on a large, basinal scale through low-per-
meability strata.
4.3.2. Conceptual model

The conceptual model is shown in Fig. 4 and uses a layer cake conceptualisation
ŽGalloway and Sharp, 1998. of isotropic sandstone and shale layers consisting of upper
Fig. 3. Cross-section through the Rio Grande Embayment showing salinities, the top of overpressure, and
Wilcox and Frio growth fault zones Žafter McKenna and Sharp, 1997..
and lower sandstone units with an intervening shale layer. Table 2 gives the relevant
model parameters. Because we focus on transport across low-permeability units, all
model scenarios presented herein use homogeneous sand layers with a permeability of
10y1 2 m2 . The initial solute concentration is 200,000 mg ly1 in the upper sand at and
20,000 mg ly1 elsewhere. The upper sandstone layer has constant head boundaries on
the left- and right-hand sides. The top and the bottom as well as the left- and right-hand
sides of the shale and the lower sandstone layers are no-flow boundaries. Initial fluid
pressures are set at hydrostatic for the initial fluid conditions specified with an arbitrary
value of zero assigned to the upper right-hand corner of the model. The constant
pressure nodes along the left- and right-hand boundaries in the upper sandstone unit are
adjusted to maintain a uniform horizontal hydraulic gradient of 10y4 throughout the
upper layer. Using this pressurisation regime, the initial flow in the upper sand is
subhorizontal flowing from left to right.
A series of paired scenarios were conducted to investigate the effect of permeability
variations in the shale layer. The average shale permeability across the entire layer in the
paired scenarios is the same; however, each scenario has a different shale permeability
distribution. The permeability of the shale layers are either heterogeneous, containing
Fig. 4. Conceptual model for fractured low-permeability shale used in the Gulf of Mexico case study: Layer
cake sand, shale, sand stratigraphy with two conduits. Simulation geometry, boundary and initial conditions
are shown.
300-m-wide Žthe width of one element. zones of higher permeability AconduitsB, or

homogeneous. We hypothesise that localised zones of micro fracturing or facies changes
may create such AconduitsB in shales. The permeabilities of these conduits are set one
order of magnitude greater than the rest of the shale layer and occur at 900–1200 and
7800–8100 m from the left-hand boundary in the shale layer.
Table 2
Model parameters used in fractured low permeability shale study
Freshwater density Ž r . s1000 kg my3
Groundwater concentrations 20,000 mg ly1
Concentration of invading fluids 200,000 mg ly1
Fluid dynamic viscosity Ž m . s 2.83=10y4 kg my1 sy1
Coefficient of fluid density change ŽE r rEC . s 700 kg my3
Water compressibility s 5=10y9 Pay1
Soil compressibility s1=10y8 Pay1
Porosity sand and shale Ž ´ . s 0.15
Intrinsic permeability sand Ž k sand . s10y1 2 m2
Intrinsic permeability shale Ž k shale . s10y15 –10y22 m2
Longitudinal dispersivity Ž a L . s60 m
Transverse dispersivity Ž a T . s15 m
Molecular diffusivity Ž D 0 . s10y9 m2 sy1
Acceleration due to gravity Ž g . s9.81 m sy2
Aquifer depth Ž H . s150 m
Aquifer length Ž L. s9300 m
Thickness sand layerss 50 m
Thickness shale layer s 50 m
4.3.3. Spatial and temporal discretisation

The dimensions of the model grid are 9300 m horizontally and 150 m vertically. For
the cases studied, the spatial domain was discretised to form a completely uniform mesh
containing 992 nodes and 930 elements, by choosing 31 finite elements horizontally
Ž D x s 300 m. and 30 finite elements vertically Ž D y s 5 m.. All transient simulations
covered a total simulated time of 14,055 years and were run with an initial time step of 1
day that increased to a maximum of 10 4 days using a time-step multiplier of 1.1.
5. Results and discussion
5.1. Continuous periodic hydraulic conductiÕity distributions
The first set of simulations examined the effect of varying wavelength on the stability
of the dense plume problem. A set of numerical simulations were run with k AV s 10y1 3
m2 , c s 1 and with C U s 5000, 10,000, 20,000, 30,000, 40,000, 50,000, and 100,000
mg ly1 and with permeability distribution wavelength allowed to vary as l s 2, 5, 50,
and 100 m.
Instability is manifested by lobe-shaped protrusions or AfingersB which sink vertically
into the underlying aquifer under gravitational influence. By defining a stability index
where simulations, say at some simulation time, T s 12 years, were classed as either
stable Žstability index, SI s 0. or unstable ŽSI s 1., the dependence of stability on both
concentration and wavelength is seen. The two criteria used for instability are: Ži. the
amplitude of the initial perturbation grows with time as evidenced by lobe-shaped
protrusions or fingers in the concentration fields; and Žii. zones of convective motion
Župward and downward fluid movement. as evident in the velocity vector plot. For small
concentration differences, progressively larger perturbation wavelengths are required to
propagate instabilities ŽFig. 5.. Results show that as the concentration difference
increases, the spectrum of unstable wavelengths increases and the minimum critical
wavelength required to perturb the system is reduced. The curve in Fig. 5 appears to be
asymptotic to both axes although more data points would be needed to truly prove
asymptotic conditions. The data suggest that when density differences are small, only a
small spectrum of long wavelengths will be unstable, and when density differences are
large, most disturbing wavelengths will be unstable. These findings are consistent with
those of Schincariol et al. Ž1994., who studied the movement of a dense plume enclosed
by and moving along in a body of less dense fluid.
In flow through natural porous media, interfacial disturbances are continuously
generated due to the heterogeneities of the medium ŽMoissis and Wheeler, 1990.. It
must be recognised that perturbations to flow will, therefore, occur over many spatial
scales. Slight differences in pore geometry to larger heterogeneities on the scale of the
dense plume source can be sufficient to generate instabilities. There is some maximum
wavelength obtainable and this should be somewhat similar to the dimensions of the
reservoir ŽMarle, 1981.. In the present case, the upper limit to this wavelength will be
imposed by the effective dimensions of the constant concentration source and, hence, the
length of the plume interfaces. The results of the present study agree with the following
Fig. 5. Stability diagram showing wavelength, l, vs. concentration difference DC. Here, the stability index
ŽSI. is A1B when a simulation was classed as unstable and A0B when it was classed as stable. The results show
the dependence of stabilityrinstability on the concentration difference DC and the perturbation wavelength, l.
The dashed curve Ž----. indicates the approximate locus of minimum critical wavelength required to produce
instabilities.
aspects of List’s Ž1965. study of dense plume stability in a homogeneous porous

medium: Ži. that horizontal two-dimensional flows in saturated porous media with a
denser fluid overlying a less dense fluid are always potentially unstable, although
instability may not occur because the perturbing waves are below the necessary critical
wavelength; Žii. that longer wavelengths are more unstable; and Žiii. that an increase in
density contrast increases the spectrum of unstable perturbation wavelengths.
In miscible fluid systems, the concept of a critical wavelength is understood by
examining the dispersive effects within the system. With miscible fluids, dispersion
tends to enhance the stability of a displacement and this is comparable to the effects of
capillarity for immiscible fluids. Shorter perturbation wavelengths make Žhorizontal.
diffusionrdispersion between neighbouring fingers more effective at smoothing out
fingers before they can grow Žsee Fig. 6a.. In fact, the existence of a mixing zone where
the viscosity and density of the fluid vary progressively from those of the displacing
fluid to those of the displaced fluid, diminishes the principle cause of the instability,
namely, differences in fluid properties ŽMarle, 1981..
Fig. 6. A conceptual model of the instability process: Ža. homogeneous media and Žb. heterogeneous media. In
both cases, vertically downward buoyancy forces that drive instabilities compete with diffusiverdispersive that
dissipate them. In the homogeneous case, instabilities of shorter wavelength tend to stabilise at earlier times
because lateral diffusion and dispersion are more effective at AsmearingB fingers. In the heterogeneous case,
long and vertically connected high-permeability conduits enhance instabilities and laterally extensive low-per-
meability conduits tend to dissipate them.
To test the effect of average permeability of a porous layer on the stability of a dense
plume, a series of simulations were run with the average permeability now increased by
one order of magnitude. Here, k AV s 10y1 2 m2 , C U s 5000 mg ly1 , and c s 1.
Whereas each of these systems was previously stable with an average permeability
k AV s 10y1 3 m2 , they were now all unstable even at the low concentration difference of
C U s 5000 mg ly1 . An increase in average permeability changed the spectrum of
unstable wavelengths and reduced the critical wavelength in a similar way that increas-
ing the concentration did. Similarly, a set of simulations performed with k AV s 10y1 4
m2 , C U s 100,000 mg ly1 , and c s 1 showed that only the larger wavelengths l s 50
and 100 m were capable of producing instabilities.
It is interesting to examine what effect changing the amplitude of the permeability
perturbation has on the stability of the dense plume. It is also a valuable exercise to
verify the usefulness of traditional stability criteria in predicting the onset of instability
in systems, which are subject to large amplitude perturbations, such as those which are
more than likely to be encountered in field situations. Previous definitions and stability
criteria based upon a Rayleigh number have had the underlying fundamental assumption
that the system under consideration was almost perfectly homogeneous. In addition,
without it necessarily being explicitly stated, the Rayleigh number is valid only in
systems where very small amplitude perturbations are assumed to exist. This appears to
be a reasonable assumption for almost perfectly homogeneous systems where c s 0.
Consider a set of simulations performed with variability over four orders of magni-
tude, namely, c s 2 and with k AV s 10y1 3 m2 , C U s 5000 mg ly1 . Results for all
wavelengths tested in the range l s 2 m to l s 100 m now showed the occurrence of
instabilities. Previous simulations conducted with c s 1 and with other parameters held
constant were all stable below C U s 10,000 mg ly1 . Other simulations conducted also
confirmed that permeability amplitude alone could alter system stability, even with
average permeability, k AV , and basin salinity, C U , kept constant. An increase in the
amplitude of the permeability function will reduce the permeability of troughs between
fingers and, therefore, reduces horizontal dispersion. Therefore, increasing the amplitude
of a permeability distribution is destabilising for two reasons, namely, Ži. it creates zones
of higher permeability where instabilities may be more likely to occur and are easier to
propagate; and Žii. it creates zones of lower permeability which hinder cross or lateral
dispersion between fingers ŽFig. 6.. These results suggest that any stability criterion that
uses average values for hydraulic conductivity and does not account for spatial variabil-
ity in a heterogeneous distribution will not be a useful descriptor of the onset of
instability in large amplitude spatially variable heterogeneous systems.
5.2. Stochastic hydraulic conductiÕity distributions
Internally correlated normal distributions with mean 0, standard deviation 1, horizon-

tal range, a, and anisotropy, anis, were created using GSLIB software ŽDeutsch and
Journel, 1992. modified for specific use with the numerical code. The mean and
standard deviation could then be transformed accordingly to produce normal distribu-
tions for Y, namely, Y: N w m y , sy x. Any simulation could then be specified in terms of
m y , sy , t H , and anis s t H rt V .
A set of 10 random hydraulic conductivity fields was generated for use in the model.
Each of these fields were defined by Y: N w m y s y6.0, sy s 0.5x, where the variable K
to be generated had units of wmrsx. The structural constants used were t H s 300 m and
anis s 200. The resulting field had an average hydraulic conductivity K AV ; 0.1 mrday
or alternatively average permeability k AV ; 10y1 3 m2 . The permeability distributions,
plotted as ylogŽ k . are given for four stochastic fields in Fig. 7. These figures are
vertically exaggerated with VE s 10 and when drawn to scale, it is easy to see that they
Fig. 7. Permeability distributions, plotted as ylogŽ k . for four stochastic fields defined by Y: N w m y sy6.0,
sy s 0.5x. The structural constants used were t H s 300 and aniss 200. The resulting field had an average
hydraulic conductivity, K AV ; 0.1 mrday or alternatively average permeability, kAV ;10y1 3 m2 . These
figures are vertically exaggerated with VE s10 and when drawn to scale, it is easy to see that these fields are
lenticular or layered. Lighter regions correspond to high-permeability regions and darker regions to lower
permeability regions.
are lenticular or layered. The development of instabilities in this system was studied for
the case where C U s 50,000 mg ly1 . Fig. 8 gives the computed concentration profiles
for these permeability distributions at a simulation time of 12 years.
For statistically equivalent systems, entirely different behaviour is observed depend-
ing upon the spatial location of the low- and high-permeability zones. A spectrum of
Fig. 8. Cross-sectional distribution of computed concentration for the permeability distributions of Fig. 7. Each
conductivity field is statistically equivalent Y: N w m y sy6.0, sy s 0.5x and identical structure constants,
t H s 300 and aniss 200. The resulting field had an average hydraulic conductivity, K AV ; 0.1 mrday or
alternatively average permeability, kAV ;10y1 3 m2 . Here, basin salinity C U s 50,000 mg ly1 and for
statistically equivalent systems a range of behavior is observed from entirely erratic and unstable behavior in
Ža. to almost completely stable behavior in Žd.. These figures are vertically exaggerated with VE s10 and
simulation time, T s12 years. Lighter regions correspond to low concentration regions and darker regions to
high concentration regions.
cases was noted, from entirely erratic and unstable behaviour to almost completely
stable behaviour for the same average permeability. Plume stability is determined locally
as a function of the changing permeability field in the vicinity of the plume boundary
layer. Lower permeability regions tend to dissipate instabilities and higher permeability
regions tend to promote their growth. Clearly, a Rayleigh number based upon the
distribution mean cannot predict which cases would be stable or those that would be
unstable. These findings are completely consistent with those of Schincariol Ž1998., who
suggested that homogeneous field criteria will not be applicable.
To test the effect of varying sample standard deviation on the onset of instabilities, a
set of 10 random hydraulic conductivity fields were then generated with fields defined
by Y: N w m y s y6.0, sy s 1.0x. Increasing s increases the spread of the hydraulic
conductivity values. Larger scale mixing tends to enhance dispersion within the system
and the resultant number of unstable cases appears to be lower as the standard deviation
of the permeability distribution increases. Increasing the standard deviation of the
permeability field produces local zones of lower permeability that stabilise the flow by
providing a barrier to the necessary vertical upflows and downflows associated with
unstable convective instabilities ŽFig. 6b.. This finding is in contrast to that of the
sinusoidal hydraulic conductivity distributions where increased amplitude or variance in
the permeability distribution promoted instability. This fundamental difference is related
to the structure or style of the heterogeneous field. The sinusoidal distributions are
highly ordered with longer continuous conduits and vertically propagating fingers do not
encounter low-permeability barriers. Furthermore, the larger variance in the sinusoidal
case reduces lateral dispersion. By contrast, the stochastic distributions tend to be more
disordered. Even if a finger begins to propagate in a sandy conduit, it is likely to
encounter a lower permeability lense as it propagates downwards. At the same time,
dispersive mixing is enhanced due to the irregular heterogeneous structure and continu-
ally dissipates the buoyancy force that drives the instability.
The effect of permeability structure or correlation length scale was examined in a set
of numerical experiments with Y: N w m y s y6.0, sy s 0.5x. The structural constants
were varied. In particular, the horizontal continuity values tested were t H s 1000, 300,
100 and 10 m. The anisotropy factor was varied in each case to keep the vertical
continuity constant at t V s 1.5 m. Consistent with Schincariol et al. Ž1994., it was found
that increased horizontal correlation length promotes stability. This is because higher
correlation length scales produce more laterally extensive low-permeability zones that
impede or dampen the vertical development of instabilities. A sensitivity analysis to
vertical continuity was performed by generating a distribution Y: N w m y s y6.0,
sy s 0.5x. The structural constants used were t H s 300 m, while the anisotropy factor,
anis, was allowed to vary such that t V s 100, 10 and 5 m were tested. Results indicated
that instabilities tended to be dampened in cases with hydraulic conductivity fields with
low vertical-correlation length scales.
5.3. Fractured low-permeability shales
We now consider the results of simulations in which a dense brine is introduced into
a sandy layer that sits above a low-permeability shale. Fig. 9 compares a 50-m-thick
Fig. 9. A comparison of isocons for 50-m shales with two zones or AconduitsB of higher permeability. Ža.
Homogeneous shale, k s1.585=10y1 8 m2 . Žb. Heterogeneous shale with two conduits with permeability one
order of magnitude higher than the shale permeability with equivalent average permeability to that of case Ža.
Ž k m s10y1 8 m2 ; k c s10y17 m2 ..
homogeneous shale to a heterogeneous shale with two AconduitsB as described previ-

ously. The introduction of AconduitsB only changes the arithmetic mean permeability of
the shale layer minimally. In Fig. 9a, the permeability of the homogeneous shale layer is
1.585 = 10y1 8 m2 . For the heterogeneous case ŽFig. 9b., the shale matrix permeability
Ž k m . is 10y1 8 m2 and the permeability of the AconduitsB is 10y17 m2 Ž k c .. Results show
that simulations with homogeneous shale permeabilities require more time Ž6936 years
in Fig. 9a. for the initial breakthrough of denser solution into the lower sand layer than
simulations with AconduitsB Ž3651 years in Fig. 9b.. In a number of simulations Žnot
shown here., it was found that cases where heterogeneity was explicitly accounted for
required less time for the initial breakthrough of denser solution into the lower sand than
in the equivalent homogeneous case. Importantly, compared with the other heteroge-
neous cases considered in this study, it appears that a set of vertical parallel fractures or
conduits in a low-permeability matrix unit is a very conducive to the formation and
growth of instabilities. Other numerical simulations suggest that the larger the perme-
ability contrast between matrix and conduits, the more easily instabilities are propagated.
Furthermore, low-permeability Žhomogeneous. shale layers that did not previously
permit convection were seen to do so when conduits were introduced. This has
important implications for the understanding of processes controlling solute migration in
low-permeability shales as well as for the way in which we might more accurately
model them.
5.4. A comment on the Rayleigh number
Numerical simulations have shown that it is not possible to select a single value of
each flow parameter, define an equivalent porous medium, and calculate an equivalent
Rayleigh number. The averaging process smears the variability that triggers instability
and controls its growth andror decay. A Rayleigh number based upon average hydraulic
properties cannot be a useful predictor of instability if it ignores key properties of the
heterogeneous permeability distribution. While it may be possible to use such criteria at
the laboratory scale, where the homogeneity of the porous medium can be closely
controlled, traditional Rayleigh number approaches should be questioned in their
application at the field scale in heterogeneous porous media.
In addition to problems with averaging spatially variable properties into single valued
parameters, several other difficulties currently exist in using the Rayleigh number in
many hydrogeologic applications. The choice of an appropriate nondimensionalising
length scale used in the Rayleigh number formulation is not trivial. For transient cases
and for determining onset conditions, the aquifer depth H is not an appropriate length
scale. The dimensional analysis that leads to dimensionless Ra makes two assumptions
that are not relevant or appropriate in the cases under consideration. First, a steady-state
flow field is assumed. Second, in the nondimensionalising of parameters, porous layer
thickness, H, was used as the appropriate length scale because the density profile across
the entire layer was assumed to vary linearly and uniformly with density Žor concentra-
tion. gradient, DCrH. In most, if not all, transport problems, plumes tend to be
localised and do not extend across entire aquifers and the concentration distributions
within them are typically nonlinear. The boundary layer where instabilities are generated
and begin to propagate is not affected by aquifer depth at early times. Instability is
determined on a local basis at the boundary layer by perturbations to flow in the vicinity
of the boundary layer interface. When we consider heterogeneous porous media, aquifer
depth, H, is even less relevant as a length scale and the thickness of more permeable
zones or conduits is likely to become more important.
Oostrom et al. Ž1992b. also pointed out the potential problems with choice of length
scale for characterising instabilities in transient groundwater flow. In that study, a dense
plume was injected along part of the top boundary of a saturated sand tank in which
there is a horizontal ambient flow imposed by an external head gradient. They found that
the depth of the sand tank was not an appropriate length scale for examining the
likelihood of convection in a transient flow field and defined a new length scale, Hdp ,
which was effectively a measure of the plume thickness at any point in time. Unlike the
depth of the sand tank as the length scale, the use of this new length scale gave
consistent results for the range of Ra over which plumes are unstable and the range over
which they are not. Oostrom et al. Ž1992b. also noted significant problems in determin-
ing dispersion associated with the flow and incorporating this into the stability criteria.
Another problem with the use of a Rayleigh number arises when considering transient
density-driven flow. By virtue of the steady flow assumptions made in Rayleigh’s
original paper, the Rayleigh number can only provide information about onset conditions
and nothing about subsequent temporal development. The instability process itself
occurs in order to produce a stable density stratification and it is, therefore, expected that
the behaviour is inherently time-dependent and that fingers would not grow indefinitely.
Furthermore, a missing term that is likely to be important in most transient hydrogeo-
logic applications is aquifer storage. Interestingly, all numerical modeling studies and
conceptual models of instability to date, including our own, have assumed some value of
storativity in their transient analyses, but do not comment on what effect varying this
parameter has on the onset, growth or decay of instabilities. Transient groundwater flow
processes are affected by changes in aquifer storativity as water is released from or
taken into storage by the porous matrix. However, neither storativity nor its two key
independent variables Žmatrix compressibility and fluid compressibility. appear in any
Rayleigh stability formulation to date even when used in the description of transient
instability phenomenon. The problem lies in the steady-state flow assumptions made in
deriving these formulations. How can a stability criterion be a complete and universal
descriptor of a process if it does not contain all parameters that affect it? The effect of
storativity is likely to be important, but how this is incorporated into relevant stability
criteria is not presently clear. Further work is required to determine what effect it has on
instability processes.
6. Conclusions
This study demonstrates that the onset and subsequent growth or decay of convective
instabilities is intimately related to the style or structure of the heterogeneous porous
media under consideration. It is clear that heterogeneity is important for two main Žyet
opposing. reasons: Ži. it serves as the triggering mechanism for the onset of instabilities;
and Žii. it is the most important factor controlling whether instabilities once they are
generated will grow or decay. The structure of the heterogeneous field appears to be the
single most important factor in controlling the fate of instabilities. Long and vertically
continuous high-permeability regions tend to enhance growth conditions Že.g., vertical
fracturing or conduits, sinusoidal conductivity distribution. where intermediate low-per-
meability regions provide resistance to horizontal dispersive mixing. However, in
lenticular structures Že.g., stochastic distributions., increased variance and increased
horizontal correlation in the permeability field create laterally extensive low-permeabil-
ity barriers to the vertical upflow and downflow necessary to maintain convective flow.
Instabilities tend to be dampened compared with the homogeneous cases. Only simple
vertical conduits or high-permeability zones have been studied here. More realistic
nonuniform distributions for fracture apertures, separation distances and orientations
must be the subject of further study and will help to elucidate onset and propagation
conditions ŽFig. 1.. While some of the controlling factors appear to be evident, it is clear
that no broad generalisation on the impact of heterogeneity on variable-density flow can
be made because certain styles of heterogeneity provide growth conditions, while others
do not. These differences mean that it is necessary to carefully delineate between the
styles of heterogeneity in order to understand their impact on unstable free convection.
Several questions arise: Are the results obtained using pump tests for aquifer hydraulic
properties meaningful in the simulation of density-driven flow processes if the variabil-
ity in properties is lost? Would sedimentological evidence about deposition and structure
be more useful here than a handful of hydraulic conductivity estimates made from pump
tests? Furthermore, we might ask whether field-scale free convection is even less
common than is presently believed because of unaccounted heterogeneity?
The onset conditions, growth rates, shapes, and resultant plume directions do not, at
present, appear amenable to prediction. The relevant stability criteria for describing
these phenomena are complex and presently unclear. The classical Rayleigh number
fails, in most cases, to predict the occurrence of instabilities under heterogeneous
conditions and should be used with caution in transient contaminant plume studies.
Fundamental problems arise due to the transient nature of the transport processes,
nonlinear density gradients between the plume and the surrounding groundwater, an
inability to use spatially averaged or single number representations for the permeability
distribution, and failure to account for aquifer storage properties.
Thus, if the Rayleigh number does not work, what then are the relevant stability
criteria? How do we incorporate key characteristics of the heterogeneous permeability
field into such criteria and our models? If the resultant processes are so sensitive to a
detailed knowledge of heterogeneity, can we predict the fate of a dense contaminant
plume? What level of explicit permeability characterisation is required to model
transport processes accurately? Will we ever be able to uncover the requisite details and
if we cannot, what level of simplification is permissible before density-dependent
processes are no longer accurately depicted? How do different variable-density flow and
solute transport codes perform and compare in simulating these processes? How are
instability processes altered in a three-dimensional flow field compared to previous
experimental and numerical work that has focussed largely on two-dimensional flow
fields? These are just some of the unanswered questions and future challenges that face
us in understanding variable-density groundwater flow and solute transport in heteroge-
neous porous media. Without adequate answers to these questions, we contend that it is
difficult to predict the fate of a dense contaminant plume in the field. However, given
the significance of variable-density flow in contaminant transport processes, the study of
these phenomena is clearly warranted.
Acknowledgements
We acknowledge the United States Department of Energy, Geosciences Research

Program, for the partial support of this research under grant number DE-FG03-
97ER14772. Author CTS wishes to acknowledge the partial support provided by an
Australian Research Council Žsmall. grant. The authors wish to thank Robert Schincar-
iol, Rene Therrien, Neville Robinson, Paul Pavelic and Awadhesh Prasad for their useful
comments and feedback on this manuscript.
References
Bachmat, Y., Elrick, D.E., 1970. Hydrodynamic instability of miscible fluids in a vertical porous column.
Water Resour. Res. 6 Ž1., 156–171.
Bakr, A.A., Gelhar, L.W., Gutjahr, A.L., MacMillan, J.R., 1978. Stochastic analysis of spatial variability in
subsurface flows: 1. Comparison of one- and three-dimensional flows. Water Resour. Res. 14 Ž2., 263.
Bear, J., 1979. Hydraulics of Groundwater. McGraw-Hill, New York, 567 pp.
Berkowitz, B., Bear, J., Braester, C., 1988. Continuum models for contaminant transport in fractured porous
formations. Water Resour. Res. 24 Ž8., 1225–1236.
Biggar, J.W., Nielsen, D.R., 1964. Chloride-36 diffusion during stable and unstable flow through glass beads.
Soil Sci. Soc. Am. Proc. 28, 591–594.
Combarnous, M.A., Bories, S.A., 1975. Hydrothermal convection in saturated porous media. Adv. Hydrosci.
10, 231–307.
Dagan, G., 1984. Solute transport in heterogeneous porous formations. J. Fluid Mech. 145, 151–177.
Dagan, G., 1986. Statistical theory of groundwater flow and transport: pore to laboratory, laboratory to
formation, and formation to regional scale. Water Resour. Res. 22 Ž9., 120S–135S.
Dagan, G., 1989. Flow and Transport in Porous Formations. Springer-Verlag, New York.
Dagan, G., 1990. Transport in heterogeneous porous formations: spatial moments, ergodicity and effective
dispersion. Water Resour. Res. 26, 1281–1290.
Dagan, G., 1994. The significance of heterogeneity of evolving scales to transport in porous formations. Water
Resour. Res. 30 Ž12., 3327–3336.
Deutsch, C.V., Journel, A.G., 1992. GSLIB Geostatistical Software Library and Users Guide. Oxford Univ.
Press, New York, 340 pp.
Elder, J.W., 1967. Transient convection in a porous medium. J. Fluid Mech. 27 Ž3., 609–623.
Fan, Y., Kahawita, R., 1994. A numerical study of variable density flow and mixing in porous media. Water
Resour. Res. 30 Ž10., 2707–2716.
Freeze, R.A., 1975. A stochastic-conceptual analysis of one-dimensional groundwater flow in nonuniform
homogeneous media. Water Resour. Res. 11 Ž5., 725–741.
Freeze, R.A., Cherry, J.A., 1979. Groundwater. Prentice-Hall, Englewood Cliffs, NJ.
Frind, E.O., 1982. Simulation of long term density-dependent transport in groundwater. Adv. Water Resour. 5,
73–97.
Galloway, W.E., Sharp Jr., J.M., 1998. Characterizing aquifer heterogeneity within terrigenous clastic
depositional systems. In: Fraser, G.S., Davis, M.W. ŽEds.., Sedimentology and Stratigraphy in Aquifer
Heterogeneity Concepts in Hydrogeology and Environmental Geology. Society for Sedimentary Geology,
Tulsa, OK, pp. 85–90.
Gelhar, L.W., 1986. Stochastic subsurface hydrology from theory to applications. Water Resour. Res. 22 Ž9.,
135S–145S.
Gelhar, L.W., 1987. Stochastic analysis of transport in saturated and unsaturated porous media. In: Bear, J.,
Corapcioglu, M. ŽEds.., Advances in Transport Phenomena in Porous Media. Martinus Nijhof, Dordrecht,
The Netherlands, pp. 657–700.
Gelhar, L.W., 1993. Stochastic Subsurface Hydrology. Prentice-Hall, New Jersey, 390 pp.
Gelhar, L.W., Axness, C.L., 1983. Three-dimensional stochastic analysis of macro-dispersion in aquifers.
Ghebart, B., Jaluria, Y., Mahajan, R.L., Sammakia, B., 1988. Buoyancy-Induced Flows and Transport,
HemisphererHarper and Row, New York.
Graham, W.D., McLaughlin, D.B., 1991. A stochastic model of solute transport in groundwater: application to
the Borden, Ontario, tracer test. Water Resour. Res. 27 Ž6., 1345–1359.
Green, L.L., Foster, T.D., 1975. Secondary convection in a Hele-Shaw cell. J. Fluid Mech. 71 Ž4., 675–687.
Grisak, G.E., Pickens, J.F., 1980. Solute transport through fractured media: I. The effect of matrix diffusion.
Grisak, G.E., Pickens, J.F., 1981. An analytical solution for solute transport through fractured media with
matrix diffusion. J. Hydrol. 52, 47–57.
Harrison, B., Sudicky, E.A., Cherry, J.A., 1992. Numerical analysis of solute migration through fractured
clayey deposits into underlying aquifers. Water Resour. Res. 28 Ž2., 515–526.
Hassanizadeh, S.M., Leijnse, T., 1988. On the modeling of brine transport in porous media. Water Resour.
Res. 24 Ž2., 321–330.
Herbert, A.W., Jackson, C.P., Lever, D.A., 1988. Coupled groundwater flow and solute transport with fluid
density strongly dependent upon concentration. Water Resour. Res. 24, 1781–1795.
Heydari, E., 2000. Porosity loss, fluid flow, and mass transfer in limestone reservoirs: application to the Upper
Jurassic Smackover Formation, Mississippi. Am. Assoc. Pet. Geol. Bull. 84, 100–118.
Horton, C.W., Rogers Jr., F.T., 1945. Convection currents in a porous medium. J. Appl. Phys. 16, 367–370.
Huyakorn, P.S., Andersen, P.F., Mercer, J.W., White, H.O., 1987. Saltwater intrusion in aquifers: development
and testing of a three-dimensional finite element model. Water Resour. Res. 23 Ž2., 293–312.
Indelman, P., Abramovich, B., 1994. Nonlocal properties of nonuniform averaged flows in heterogeneous
media. Water Resour. Res. 30 Ž12., 3385–3393.
Kimmel, G.E., Braids, O.C., 1980. Leachate plumes in groundwater from Babylon and Islip landfills, Long
Island, New York. U.S. Geol. Surv. Prof., 1085.
Koch, M., Zhang, G., 1992. Numerical solution of the effects of variable density in a contaminant plume.
Groundwater 30 Ž5., 731–742.
Kolditz, O., Ratke, R., Diersch, H.-J., Zielke, G., 1998. Coupled groundwater flow and transport: 1.
Verification of variable density flow and transport models. Adv. Water Resour. 21 Ž1., 27–46.
Krupp, H.K., Elrick, D.E., 1969. Density effects in miscible displacement experiments. Soil Sci. 107,
372–380.
Land, L.S., 1991. Evidence for vertical movement of fluids, Gulf Coast sedimentary basin. Geophys. Res. Lett.
18, 919–922.
Land, L.S., 1997. Mass transfer during burial diagenesis in the Gulf of Mexico Basin: an overview. In:
Montaneez, I.P. ŽEd.., Basin-Wide Diagenetic Patterns; Integrated Petrologic, Geochemical, and Hydro-
logic Considerations. SEPM ŽSociety for Sedimentary Geology., Lake Ozark, MO, pp. 29–39, Special
Publication.
Lapwood, E.R., 1948. Convection of a fluid in a porous medium. Proc. Cambridge Philos. Soc. 44, 508–521.
Law, J., 1944. A statistical approach to the interstitial heterogeneity of sand reservoirs. Trans. AIME 155,
202–222.
Le Blanc, D.R., 1984. Sewerage plume in a sand and gravel aquifer, Cape Cod, Massachusetts. U.S. Geol.
Surv. Water Supply Pap., 2218.
Le Blanc, D.R., Garabedian, S.P., Hess, K.M., Gelhar, L.W., Quadri, R.D., Stollenwerk, K.G., Wood, W.W.,
1991. Large-scale natural gradient test in sand and gravel, Cape Cod, Massachusetts: 1. Experimental
design and observed tracer movement. Water Resour. Res. 27 Ž5., 895–910.
List, E.J., 1965. The stability and mixing of a density-stratified horizontal flow in a saturated porous medium,
Rep. KH-R-11, Calif. Inst. of Technol., Pasadena.
Liu, H.H., Dane, J.H., 1996. A criterion for gravitational instability in miscible dense plumes. J. Contam.
Hydrol. 23, 233–243.
Marle, C.M., 1981. Multiphase Flow in Porous Media. Gulf, Houston, TX, 257 pp.
McKenna, T.E., Sharp Jr., J.M., 1997. Subsurface temperatures, fluid pressures, and salinities in the Rio
Grande Embayment, Gulf of Mexico Basin, USA. Proceedings, 30th International Geological Congress,
Beijing, vol. 8, pp. 263–274.
McKibbin, R., O’Sullivan, M.J., 1980. Onset of convection in a layered porous medium heated from below. J.
Fluid Mech. 96, 375–393.
McKibbin, R., Tyvand, P.A., 1982. Anisotropic modeling of thermal convection in a multilayered porous
media. J. Fluid Mech. 118, 315–339.
McKibbin, R., Tyvand, P.A., 1983. Thermal convection is a porous medium composed of alternating thick and
thin layers. Int. J. Heat Mass Transfer 26, 761–780.
McMillan, W.D., 1966. Theoretical analysis of groundwater basin operations. Water Resour. Cent. Contrib.
vol. 114 Univ. California, Berkeley, 167 pp.
Morton, R.A., Land, L.S., 1987. Regional variations in formation water chemistry, Frio Formation ŽOligocene.,
Texas Gulf Coast. Am. Assoc. Pet. Geol. Bull. 71, 191–206.
Moissis, D.E., Wheeler, M.F., 1990. Effect of the structure of the porous medium on unstable miscible
displacement. In: Cushman, J.H. ŽEd.., Dynamics of Fluids in Hierarchical Porous Media. Academic Press,
San Diego, CA, pp. 243–271.
Mulqueen, J., Kirkham, D., 1972. Leaching of a surface layer of sodium chloride into tile drains in a sand-tank
model. Soil Sci. Soc. Am. J. 36, 3–9.
Neuman, S.P., Winter, C.L., Newman, C.M., 1987. Stochastic theory of field-scale Fickian dispersion in
anisotropic porous media. Water Resour. Res. 23 Ž3., 453–466.
Nield, D.A., Bejan, A., 1999. Convection in Porous Media. Springer-Verlag, New York, 546 pp.
Oldenburg, C.M., Pruess, K., 1995. Dispersive transport dynamics in a strongly coupled groundwater–brine
flow system. Water Resour. Res. 31 Ž2., 289–302.
Oostrom, M., Dane, J.H., Guven,¨ O., Hayworth, J.S., 1992a. Experimental investigation of dense solute
plumes in an unconfined aquifer model. Water Resour. Res. 28 Ž9., 2315–2326.
Oostrom, M., Hayworth, J.S., Dane, J.H., Guven,¨ O., 1992b. Behavior of dense aqueous leachate plumes in
homogeneous porous media. Water Resour. Res. 28 Ž8., 2123–2134.
Paschke, N.W., Hoopes, J.A., 1984. Buoyant contaminant plumes in groundwater. Water Resour. Res. 20 Ž9.,
1183–1192.
Pinder, G.F., Cooper, H.H., 1970. A numerical technique for calculating the transient position of the saltwater
front. Water Resour. Res. 6 Ž3., 875–882.
Pratts, M., 1966. The effects of horizontal fluid flow on thermally induced convection currents in porous
mediums. J. Geophys. Res. 71, 4835–4838.
Raffensperger, J.P., Vlassopoulos, D., 1999. The potential for free and mixed convection in sedimentary
basins. Hydrogeol. J. 7, 505–520.
Rayleigh, L., 1916. On convection currents in a horizontal layer of fluid when the higher temperature is on the
under side. Philos. Mag., Ser. 6 32, 529–546 ŽJ.W. Strutt..
Robinson, N.I., Sharp Jr., J.M., Kreisel, I., 1998. Contaminant transport in sets of parallel finite fractures with
fracture skins. J. Contam. Hydrol. 31, 83–109.
Rose, D.A., Passioura, J.B., 1971. Gravity segregation during miscible displacement. Soil Sci. 111, 258–265.
Salvador, A. ŽEd.., The Gulf of Mexico Basin. The Geology of North America, vol. J, Geol. Soc. America,
Boulder, CO.
Saffman, P.G., Taylor, G.I., 1958. The penetration of a fluid into a porous medium or Hele-Shaw cell
containing a more viscous liquid. Proc. R. Soc. London, Ser. A 245, 312–329.
Schincariol, R.A., 1998. Dispersive mixing dynamics of dense miscible plumes: natural perturbation initiation
by local-scale heterogeneities. J. Contam. Hydrol. 34, 247–271.
Schincariol, R.A., Schwartz, F.W., 1990. An experimental investigation of variable density flow and mixing in
homogeneous and heterogeneous media. Water Resour. Res. 26 Ž10., 2317–2329.
Schincariol, R.A., Schwartz, F.W., Mendoza, C.A., 1994. On the generation of instabilities in variable density
flow. Water Resour. Res. 30 Ž4., 913–927.
Schincariol, R.A., Schwartz, F.W., Mendoza, C.A., 1997. Instabilities in variable density flows: stability
analyses for homogeneous and heterogeneous media. Water Resour. Res. 33 Ž1., 31–41.
Sharp Jr., J.M., 1993. Fractured aquifersrreservoirs: approaches, problems, and opportunities. In: Banks, D.,
Banks, S. ŽEds.., Hydrogeology of Hard Rocks, Memoires of the 24th Congress, vol. 24, International
Assoc. Hydrogeologists, Oslo, Norway, pp. 23–38, part 1.
Sharp, J.M., Galloway, W.E., Land, L.S., McBride, E.F., Blanchard, P.E., Dutton, S.P., Farr, M.R., Gold,
P.B., Jackson, T.J., Lundegard, P.D., MacPherson, G.L., Milliken, K.L., 1988. Diagenetic processes in
Northwestern Gulf of Mexico sediments. In: Chilingarian, G.V., Wolf, K.H. ŽEds.., Developments in
Sedimentology. Diagenesis II, vol. 43, Elsevier, Amsterdam, pp. 43–113.
Shikaze, S.G., Sudicky, E.A., Schwartz, F.W., 1998. Density-dependent solute transport in discretely-fractured
geologic media: is prediction possible? J. Contam. Hydrol. 34, 273–291.
Simmons, C.T., Narayan, K.A., 1997. Mixed convection processes below a saline disposal basin. J. Hydrol.
194, 263–285.
Simmons, C.T., Narayan, K.A., Wooding, R.A., 1999a. On a test case for density-dependent groundwater flow
and solute transport models: the salt lake problem. Water Resour. Res. 35 Ž12., 3607–3620.
Simmons, C.T., Sharp Jr., J.M., Robinson, N.I., 1999b. Density-driven free convection in zones of inverted
salinity through fractured low-permeability units in the Gulf of Mexico Basin, Texas, USA. Water 99 Jt.
Congr. vol. 2, Inst. of Engineers, Brisbane, Australia, pp. 739–744.
Smith, L., Freeze, R.A., 1979. Stochastic analysis of steady state groundwater flow in a bounded domain: 2.
Two-dimensional simulations. Water Resour. Res. 15 Ž6., 1543–1559.
Smith, L., Schwartz, F.W., 1980. Mass transport: 1. A stochastic analysis of macroscopic dispersion. Water
Resour. Res. 16 Ž2., 303–313.
Smith, L., Schwartz, F.W., 1981. Mass transport: 2. Analysis of uncertainty in prediction. Water Resour. Res.
17 Ž2., 351–369.
Smith, L., Schwartz, F.W., 1984. An analysis of the influence of fracture geometry on mass transport in
fractured media. Water Resour. Res. 20 Ž9., 1241–1252.
Sudicky, E.A., 1986. A natural gradient experiment on solute transport in a sand aquifer: spatial variability of
hydraulic conductivity and its role in the dispersion process. Water Resour. Res. 22 Ž13., 2069–2082.
Sudicky, E.A., Frind, E.O., 1982. Contaminant transport in fractured porous media: analytical solutions for a
series of parallel fractures. Water Resour. Res. 18 Ž6., 1634–1642.
Therrien, R., Sudicky, E.A., 1996. Three-dimensional analysis of variably-saturated flow and solute transport
in discretely-fractured porous media. J. Contam. Hydrol. 23, 1–44.
Tompson, A.F.B., Gelhar, L.W., 1990. Numerical simulation of solute transport in three-dimensional randomly
heterogeneous porous media. Water Resour. Res. 26 Ž10., 2541–2562.
van der Molen, W.H., van Ommen, H.C., 1988. Transport of solutes in soils and aquifers. J. Hydrol. 100,
433–451.
Voss, C.I., 1984. SUTRA: a finite-element simulation model for saturated-unsaturated fluid density-dependent
groundwater flow with energy transport or chemically reactive single-species solute transport. U. S. Geol.
Surv., Water-Resour. Invest. Rep. 84-4369, 409 pp.
Voss, C.I., Souza, W.R., 1987. Variable density flow and solute transport simulation of regional aquifers
containing a narrow freshwater–saltwater transition zone. Water Resour. Res. 23 Ž10., 1851–1866.
Wood, J.R., Hewett, T.A., 1984. Reservoir diagenesis and convective fluid flow. In: McDonald, D.A., Surdam,
R. ŽEds.., Clastic Diagenesis. Am. Assoc. Pet. Geol. Mem., vol. 37, pp. 99–111.
Wooding, R.A., 1959. The stability of a viscous liquid in a vertical tube containing porous material. Proc. R.
Soc. London, Ser. A 252, 120–134.
Wooding, R.A., 1962. The stability of an interface between miscible fluids in a porous medium. Z. Angew.
Math. Phys. 13, 255–265.
Wooding, R.A., 1963. Convection in a saturated porous medium at large Rayleigh number or Peclet number. J.
Fluid Mech. 15, 527–544.
Wooding, R.A., 1969. Growth of fingers at an unstable diffusing interface in a porous medium or Hele-Shaw
cell. J. Fluid Mech. 39, 477–495.
Wooding, R.A., Tyler, S.W., White, I., 1997a. Convection in groundwater below an evaporating salt lake: 1.
Onset of instability. Water Resour. Res. 33, 1199–1217.
Wooding, R.A., Tyler, Scott, W., White, I., Anderson, P.A., 1997b. Convection in groundwater below an
evaporating salt lake: 2. Evolution of fingers or plumes. Water Resour. Res. 33, 1219–1228.
Zhang, H., Schwartz, F.W., 1995. Multispecies contaminant plumes in variable density flow systems. Water
Resour. Res. 31 Ž4., 837–847.

Simmons 2001

Uploaded by

Copyright:

Available Formats

Simmons 2001

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Simmons 2001

Uploaded by

Copyright:

Available Formats

Journal of Contaminant Hydrology 52 Ž2001.

Variable-density groundwater flow and solute

In order to further examine variable-density flow and transport processes in heteroge-

2. Heterogeneity and variable-density flow: a review of concepts and approaches

Below, we review briefly heterogeneity in geologic systems, variable-density flow

Geologic systems are not truly homogeneous or uniform. Therefore, single-number

2.2. Variable-density flow phenomena

Groundwater contamination often occurs when surface-based pollution sources re-

2.3. Variable-density flow in heterogeneous systems

In variable-density flow systems, heterogeneity in hydraulic properties can perturb

3. Dimensionless parameters as indicators of stability

Uc H gk b Ž Cmax y Cmin . H Buoyancy and Gravitation

4.1. The SUTRA model

The Saturated–Unsaturated TRAnsport ŽSUTRA. model developed by Voss Ž1984.

4.1.1. Mass conserÕation

where pŽ x, y, t . is the fluid pressure; g is the gravity vector; m is fluid dynamic

4.1.2. Salt conserÕation

4.2. Continuous periodic and stochastic hydraulic conductiÕity distributions

4.2.1. Conceptual model

4.2.2. Spatial and temporal discretisation

elements horizontally Ž D x s 1 m. and 10 finite elements vertically Ž D y s 1 m.. All

4.2.3. Hydraulic conductiÕity distributions

4.2.4. Model parameters

4.3. Fractured low-permeability shales: Gulf of Mexico case study

4.3.2. Conceptual model

300-m-wide Žthe width of one element. zones of higher permeability AconduitsB, or

4.3.3. Spatial and temporal discretisation

5. Results and discussion

5.1. Continuous periodic hydraulic conductiÕity distributions

aspects of List’s Ž1965. study of dense plume stability in a homogeneous porous

5.2. Stochastic hydraulic conductiÕity distributions

Internally correlated normal distributions with mean 0, standard deviation 1, horizon-

5.3. Fractured low-permeability shales

homogeneous shale to a heterogeneous shale with two AconduitsB as described previ-

We acknowledge the United States Department of Energy, Geosciences Research

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.