3 Fluids in Porous Media

The behavior of uids in porous media is quite dierent from that in large uid compartments like for instance in a lake. It is essentially determined by the boundaries which pervade the entire volume and dissipate energy and momentum very eciently. In contrast, large uid compartments are dominated by the internal dynamics of the uid, most importantly by the inertia term [v ]v which leads to turbulence.

3.1 Natural Porous Media

Most natural solid materials are porous, at least to some extent. This is in particular the case for soils which evolve through a complicated process from rocks and organic debris. It involves the physical and chemical weathering of the original rocks, i.e., the disintegration of the mineral conglomerate whereby some minerals are dissolved while the more stable ones are released, eventually as individual grains. Some rocks, for instance limestone, contain no stable minerals and over time they are dissolved completely. The dissolved components yield nutrients for bacteria, fungi, and eventually for plants. These in turn contribute to the soils evolution on all scales, from the formation of organo-mineral complexes to root channels which provide a direct link between the atmosphere and deeper soil layers. Eventually, the lower life forms are followed by animals, from the miniscule enchytrae and nematodes through earthworms, typical by far the largest mass fraction of soil biota, all the way to moles and larger mammals. They all burrow their living space into the ground, thereby mixing the soil and creating structures in a wide range of scales. Physical processes of course continue to contribute to soil formation also after the initial weathering. These are primarily freezethaw cycles in cold regions, shrinking-swelling cycles in moderately humid regions, and erosion-sedimentation processes almost everywhere. A still most readable and comprehensive description of soil formation is given by Jenny [1941]. The important aspect here is that the soil environment is dynamic, not static, and that structures may be expected over a wide range of scales. 35

36

3 Fluids in Porous Media

Figure 3.1. Thin section of about 1 mm length from a loamy-clay soil. Clearly distinguishable are the system of macropores with diameters of some 0.1 mm, small soil aggregates (lighter shades of brown) with sizes of about 0.3 mm, and the system of mesoand micropores. (Image courtesy of H.-J. Vogel)

While this might be reminiscent of turbulence, although at time scales of years to centuries for a typical soil layer, we notice a qualitative dierence: Turbulence originates from a single generator, the inertia term [v ]v, and thus leads to the same structures at all scales and the feasibility of comparatively simple scaling laws [Frisch, 1995]. In contrast, soil structures result from vastly dierent generators and have vastly dierent forms. We thus cannot expect simple transitions between scales. The multiscale nature of the pore space, which is typical for most soils, is easily appreciated by looking at even small samples as illustrated in the following examples.
Example: Micropore System of a Loamy-Clay Soil At the smallest scale of interest here, we consider a thin section of about 1 mm length from a loamyclay soil (Figure 3.1). The largest pores have a diameter of about 0.1 mm, corresponding to a capillary height of about 0.3 m as may be calculated from (2.40). These pores are thus air-lled whenever the water table is more than 0.3 m below them. In addition to the macropores, there appears a continuum of smaller scale pores all the way down to the resolution limit of a few micrometers. This network of so called meso- and micropores penetrates the aggregates as well as the darker material between them. Their wide range of sizes assures that some water is retained in the soil even at large distances from a water table and under rather dry conditions. Example: Macropore System of a Loamy-Clay Soil Looking at a soil sample at a larger scale, only the macropores are perceptible. Figure 3.2 shows three cross-sections, out of a stack of 120, that were obtained from a larger sample extracted from the so called B-horizon the one below the densely rooted topmost layer with a high organic content of a loamy-clay soil. After excavation, the sample was impregnated for stabilization and was grinded down in steps of 0.1 mm. The faces were photographed and digitized with a resolution of 0.12 mm [Cousin et al., 1996]. Visible structures range from the prominent cavity in the leftmost image it is some 30 mm long, 5 mm wide, and, as inspection of images not shown here reveals, at least 2 mm high through narrow elongated voids, typically remnants of root channels and excrements of earthworms, and almost circular voids which result from the vertical worm channels, all the way to very small voids that originate from thin plant roots and enchytrae. As illustrated in Figure 3.1, much smaller pores exist in the soil matrix but they are not resolved here.

3.1 Natural Porous Media

37

Figure 3.2. Horizontal cross-sections through a sample taken at 0.4 m depth from a loamy-clay soil near Beauce, France [Cousin et al., 1996]. The side length of the square sections is 48 mm with a resolution of 0.12 mm. The smallest visible pores thus are comparable to the largest pores in Figure 3.1. The vertical distance between the sections shown here is 6 mm. White represents the pore space, black the soil matrix which itself is again porous at a smaller scale. (Data courtesy of I. Cousin)

A three-dimensional reconstruction of the macropore system of a sub-volume illustrates that this network of large pores is connected rather well (Figure 3.3). In a well-drained soil, macropores are only lled with water during short times after heavy rainfall events. During such events, inltrating water bypasses the ner textured soil matrix, is thus lost for the plants but recharges groundwater, and it may carry contaminants past the reactive surface layer. Most of the time, however, the macropore system acts as fast conduit for the exchange of gases between soil and atmosphere.

We notice in passing that the soil environment also provides a wide spectrum of physicochemical micro-sites as is apparent from the various shades of red and brown visible in Figure 3.1. They indicate the presence of organic material as well as of iron- and manganese-oxides. Obviously, this adds signicantly to the complexity of the soil environment.

Figure 3.3. Three-dimensional reconstruction of the macropore system for a selection from the dataset shown in Figure 3.2. Resolution is 0.12 mm horizontally and 0.10 mm vertically. (Image courtesy of H.-J. Vogel)

38

3 Fluids in Porous Media

Representing a typical soil in its full complicated beauty is next to impossible and we will have to restrict our consideration to a greatly simplied model. To this end, we will assume in the following that the porous medium is rigid and that its surface properties are uniform, typically completely water-wet which would lead to a thin lm of adsorbed water that covers the entire surface.

3.2 Transition to Continuum Scale

The detailed description of the pore geometry, with relevant lengths in the sub-micrometer range, and of pore-scale phenomena is not feasible for any domain of practical environmental interest. These have extents on the order of 1 m, often far more. Even for just 1 m3 , we would have to cope with some 1018 degrees of freedom! Obviously, we must come up with an averaged description along a similar line as the one that leads from gas dynamics to thermodynamics. Hence, we want to make the transition from the discrete pore space to the continuum and to macroscopic eld variables which describe the observed phenomena.

3.2.1 Representative Elementary Volume

In a pore-scale representation, each point belongs to one the phases, either to the solid matrix or to some uid. For a formal representation, we use the indicator function which for phase i is dened as i (x) = 1 , x phase i, 0 , x phase i. (3.1)

As an example of an indicator function consider Figure 3.2. If we choose white to represent 1, then this is the indicator function for the pore space. A natural way for reducing the pore-scale description to a more manageable form is the averaging of the indicator functions with some convenient weight function (x) to obtain i (x) := i (x) :=

i (x x)(x ) dx ,

(3.2)

where is the support of and i is the volume fraction of phase i, sometimes also called the macroscopic phase density. Similarly, we obtain for the average of some microscopic property of phase i the macroscopic value m (x) := i (x) :=

(x x)i (x x)(x ) dx .

(3.3)

3.2 Transition to Continuum Scale microscopic Vw dA macroscopic

39

V V Figure 3.4. Transition from pore-scale (microscopic) to continuum (macroscopic) representation. Consider a macroscopic volume V with boundary V (dotted line). Microscopically, the detailed distribution of all the phases is available, e.g., of the water phase Vw V with external boundary Vw V (red line). Macroscopically, the phases and possibly other quantities are replaced by the superposition of continuous elds (uniformly colored regions). These elds may vary in space, but on a much larger scale than that of the averaging volume.

For such averages to be unbiased, we require (x) 0 and (x) dx = 1. Popular choices for are a constant within a spherical support of radius and 0 outside or the n-dimensional Gaussian [2 2 ]n/2 exp(|x|2 /[2 2 ]) with zero mean and variance 2 which averages over a typical distance . The latter has the advantage that the averages are smooth, i.e., their spatial derivatives of arbitrary order are continuous. We notice that averaging transforms the discrete microscopic description, where each point belongs to exactly one phase, to a continuous representation where at each point the densities of the macroscopic quantities are given (Figure 3.4). Hence the notion of a transition to the continuum scale. In general, denition (3.2) depends on the choice of which makes all subsequent considerations rather cumbersome. However, if the indicator function i is suciently uniform, enlarging the averaging volume will eventually lead to a value for i that depends neither on the size nor on the form of . Such an averaging volume is called a representative elementary volume. It is the means for an objective transition from a pore-scale to a continuum representation and is thus of fundamental importance.
Example: REV for Porosity of Macropore System We consider the volume fraction of the pore space, the so called porosity, of the soil sample illustrated in Figure 3.2 and Figure 3.3 and use a cube of side length as averaging volume with a weight function that is constant within the cube. From a coverage of the entire sample volume with such cubes, we obtain the distribution of local porosities as a function of (Figure 3.5). For very small values of , this distribution is very broad with extremes given by cubes that are completely within the

40 1 averaged porosity

3 Fluids in Porous Media

Figure 3.5. Estimated porosity of soil sample from Figure 3.2 as a function of averaging cubes length. The cyan curve represents a particular location. The other curves represent the ensemble of all cubes: average (magenta), minimum and maximum, and the two quartiles. Half of all values are within the gray band. The linear extent of a reasonable REV would be some 17 mm.

0 0 5 10 15 length of averaging cube [mm]

matrix and in the pore space, respectively. With increasing values of , the representation of both, matrix and pore space, also increase and the dierence between cubes decreases and we may postulate an REV with a size of about 17 mm. Fluctuations at larger scales would then be attributed to macroscopic variability.

Early studies of porous media considered averaging functions that were constant within some bounded support and 0 outside, and introduced the notion of a representative elementary volume (REV) for a volume of linear extent [Bear, 1972]. In the following, we will use REV as shorthand for the requirement that some microscopic quantity becomes approximately constant when averaged with a weight function of characteristic extent , even if this function is smooth and possibly even has unbounded support. We realize that the transition to continuous elds replaces the original pore scale representation by a new one. While we hope that the solutions of both representations approach each other at a suciently large scale, ideally that of the REV, one may expect large deviations between the two at smaller scales. Hence, despite the fact that we will eventually formulate dierential equations for the dynamics of transport processes in porous media, evaluating the ensuing solutions at scales smaller than the REV will generally not reect the physical processes correctly. This becomes particularly virulent near boundaries. Much the same discussion applies to all transitions between scales that involve some kind of averaging and also to measuring processes where the measuring volume takes the role of the REV.

3.2.2 Texture
At the continuum scale, the complicated geometry of the porous medium is reduced to a few statistical quantities like the volume and surface density of the pore space or correlation functions and lengths. Here, we only consider

3.2 Transition to Continuum Scale

41

the lowest order description, the volume density of the pore space which is usually referred to as the porosity . Formally, = is obtained from the indicator function of the pore space through (3.2). It is often convenient to also introduce the bulk density b which is obtained by averaging the mass density m of the solid matrix. With (3.3) this may be written as b = m (x)[1 (x)] . (3.4) Assuming m to be constant, which is a good approximation for mineral porous media for which m 2.65 103 kg m3 , we obtain b = m [1 ] . (3.5)

Since both b and m can be measured with reasonable eort for many porous media, this equality is often used to calculate the porosity .

3.2.3 State Variables

To describe the macroscopic state of an isothermal uid in a porous medium, we anticipate that at least the amount of uid and its potential energy must be given. Amount of Fluid The volume fraction i of phase i the volume of phase i per volume of porous medium was already dened in (3.2). Sometimes, the saturation i (3.6) i := is used instead of the volume fraction. These quantities are most useful for dealing with incompressible media, which will be the case in the following. In analogy to the bulk density (3.5), the macroscopic mass density of phase i could be introduced. We will not do this, however, and will always write it explicitly as i i , where i is the mass density of the pure uid. Potential Energy Potentials are more convenient than forces for describing uid ow through porous media. We dene the potential energy, more precisely its density, by the energy required to move a unit volume of uid from some reference state to a particular state in the porous medium. In general, we dene the state of an element of uid i by its height z, its pressure pi and temperature Ti , and by the concentrations Cij of chemical species j. The energy of this state is referred to that in a reference state which we choose as z = z0 , p = p0 , T = T0 , and Cij = 0.

42

3 Fluids in Porous Media

In the following, we consider the simple case of isothermal ow of a pure uid. The only contributions to the potential energy i per unit volume of uid i then come from gravity and from the pressure dierence. Hence,
z

i (x) = pi (x) p0
z0 p

i (z )g dz ,
g

(3.7)

where z is the downward pointing component of x, p is called the pressure potential and g the gravity potential. For an incompressible medium, this reduces to i (x) = [pi (x) p0 ] [z z0 ]i g . (3.8) The pressure term pi , while being measurable quite easily, may consist of several components and actually be of a rather complicated nature. To appreciate this, we start out from the most simple case and proceed to more complicated ones. First consider the two uids water and air in a water-wet, rigid porous medium and choose the ambient air pressure as p0 . Since the density of air is much smaller than that of water, the hydrostatic pressure in the air is negligible. For bound water, pw < p0 , the only contribution to the pressure potential comes from the interfacial forces and is given by (2.39). This is the typical situation in a well-aerated soil at a sucient height above the water table. Things get a bit more complicated when the air is under pressure pa > p0 . This occurs for instance when an irrigation front in an initially unsaturated soil compresses the air ahead or in a gas reservoir. Then, pa p0 gets added to the interfacial pressure jump. Similarly, when the density of some overlying uid i is not negligible and it is not bound, the hydrostatic pressure i gh, where h is the vertical extent of uid i, has to be added. For free water in the above porous medium, pw p0 . Now, water is not kept in place by interfacial forces anymore and the hydrostatic pressure pw = w gz develops, where z is the depth below the corresponding free water table. This is the typical situation in groundwater. A note in passing: bound and free states are often, in a somewhat sloppy manner, associated with an unsaturated and saturated medium, respectively. The dierence between the two notions becomes apparent when we consider the capillary in Figure 2.10: for heights less than h, the capillary is saturated with uid 2 but this is held by capillary forces nevertheless. In a porous medium, this zone of full saturation, but still bound uid, is referred to as the capillary fringe. The situation becomes considerable more complicated with saturated compressible porous medium. Here, the pressure in the uid phase compensates a fraction of the weight of the overlying mass. Then, the hydrostatic pressure does not only depend on i but also on the density of the matrix and on the relative compressibilities of uid and matrix. This component is sometimes referred to as overload pressure or overload potential.

3.2 Transition to Continuum Scale

43

Finally, we consider the case of a medium that dries out ever more and again assume a water-wet rigid porous medium. As water is removed, the largest pores empty rst because water is bound less tightly there as may be seen from (2.39). Assume circular and cylindrical pores for simplicity and rearrange (2.39) for the water-air system as pw = pa + 2wa /r, where r < 0 is the radius of the interface. With pa = 105 Pa and wa = 0.0725 N m2 , we nd that pw would become negative once the interfacial radius falls below 1.45 m. It is then useful to replace the notion of a pressure by that of an energy density, actually by the chemical potential of the uid phase. We will thus write (3.8) as i (x) = im [z z0 ]i g (3.9) and refer to im as the matric potential. In our notation, this potential is negative when the uid is bound and positive when it is free. For the case of water, w is usually called the water potential. Then we have m = p w p a . (3.10)

A thermodynamic denition of the matric potential is given in Section 7.1.2. Head It is sometimes convenient to express the potential energy per unit weight i g instead of unit volume. Then, (3.9) becomes hi (x) = him [z z0 ] , (3.11)

where hi = i /[i g] is the hydraulic head and him the matric head. The interpretation of these quantities is that the energy density is expressed in terms of the height of an equivalent uid column. Notice that a negative value of the matric head indicates bound water with the meaning that the corresponding potential could hold a hanging uid column of the given height.

3.2.4 Mass Balance

We rst envisage a macroscopically uniform and rigid porous medium whose pore space is completely lled with a single uid, say water of mass density w . We consider a macroscopic volume element V with boundary V . This element is chosen to be an REV and its shape shall be arbitrary but invariable (Eulerian perspective). Let Vw V be the water-lled part and let Vw V be its external boundary (Figure 3.4). Conservation of the mass of water means that the rate of change of the mass within a volume equals the mass ow across the volumes boundary. Microscopically, the volume to be considered is Vw with its boundary Vw . Macroscopically however, there exists no explicit representation of water phase and solid matrix anymore and we need to operate with V and V ,

44

3 Fluids in Porous Media

respectively. Formulating the microscopic conservation of the mass of water yields t

Vw

dV = w

Vw

v dA = w

Vw

[ v ] dV , w

(3.12)

where the superscript indicates microscale quantities, the minus sign comes from the area element dA pointing outwards, and Gauss theorem has been applied for the second equality. Next, we use that the shape of V is xed, hence does not change as we move it through the porous medium. As a consequence, divergence and integral may be interchanged in the last term and we obtain v dV = 0 , (3.13) dV + t w w
Vw Vw Vw w w Vw v w
w

where Vw is the volume of Vw and . . . w indicates averaging over Vw . Dividing this by V , and noting that Vw / V equals the volume fraction w of the water phase, yields t [w w
w]

[w v w

w]

=0.

(3.14)

This is the conservation of mass formulated at the macroscopic scale and we may identify the terms in brackets as the macroscopic mass density and mass ux, respectively. The second term is dicult since it depends on the microscopic correlation between density and velocity. This is not severe for the case of pure and weakly compressible uid, however. For this case, the correlation is small because depends only weakly on pressure w whereas v depends on its gradient. Hence, v w w v w is a w w good approximation. We choose to nally formulate the mass balance in terms of the real density of water and of the macroscopic volume ux w := w
w

and

jw := w v

(3.15)

respectively, where superscripts are dropped in favor of a more compact notation. We thus obtain t [w w ] + [w jw ] = 0 . (3.16)

Up to now, we only considered exchange of water across some boundary V . We will in the following also look at situations, where water is extracted directly out of V , for instance by pumping wells or plant roots. Let w be the volume extraction rate, that is the volume of water that is extracted per unit time and unit volume. Since this formulation is already at the macroscopic scale, it is easily accommodated by modifying (3.16) into t [w w ] + [w jw ] = w w . (3.17)

3.2 Transition to Continuum Scale

45

Finally, we comment that (3.16)(3.17) may be written down directly as conservation equations at the macroscopic scale. However, while formally correct of course: it is just a mathematical statement this is not so useful because the quantities w and jw could then not be related to microscopic quantities and pure material properties as is the case with (3.15).

3.2.5 Empirical Flux Law

Macroscopic conservation of linear momentum could be formulated in analogy to the conservation of mass by averaging the microscopic momentum equation [Gray and Hassanizadeh, 1998]. We will follow a dierent path, however, and employ the previously introduced similarity analysis which will lead to a macroscopic ux law. Again, we will only consider slow stationary ow of a Newtonian uid that is driven by a pressure gradient alone. Some comments on releasing these assumptions are added at the end of this section. Stationary Newtonian Flow Flow of an incompressible Newtonian uid in cylindrical voids is simple because isobars surfaces of constant pressure are planes orthogonal to the cylinders, pressure gradients hence are well-dened. The situation becomes much more complicated in a general porous medium, see for instance Figure 3.3, because (i) isobars are generally neither plane nor parallel to each other and (ii) streamtubes in the strongly bifurcating network of ow channels are hard to dene. Using the previously introduced similarity analysis, we may nevertheless gain insight into the relation between the velocity v and pressure gradient p. In the following, we assume that (i) external forcing changes so slowly that the ow may be considered as stationary, (ii) ow is so slow that inertia may be neglected, and (iii) the extent of the ow domain is so small that gravity can be neglected, relative to friction. Remember that these conditions can be quantied through the dimensionless numbers dened in (2.22). For instance, (iii) may be formulated as Fr Re, or, using (2.22), as 2 g/[u] 1. These assumptions lead to the simplied Stokes equation 2 v = p which was already given in (2.30). We notice that this equation is linear, i.e., if {v, p} is a solution then {v, p} is also one. We further recall that v and p are parallel (Figure 2.4). Hence, we may write v x ; = x ; p x ; , (3.18)

where ( x ; ) is some scalar function, typically with a very complicated form. Notice that in general, we would write v( x ; Re) and correspondingly for the other two terms. However, because of the linearity just mentioned, v and p change proportionally with u and we may neglect this dependency

46

3 Fluids in Porous Media

jw

bn [v n]b b n bn [jw n]b b n

Figure 3.6. Flow through macroscopic planar REV with orthonormal vector b n in microscopic (left) and macroscopic (right) representation, respectively. The red arrows indicate the projection of velocity or ow vectors (blue) into the direction b of n. w is the region occupied by the water (light blue).

altogether. Since both sides of (3.18) must scale identically we nd with (2.29) that x; x , (3.19) = f 2 / with f some dimensionless function of space alone. This establishes a constant of proportionality between v and p that scales with the square of a characteristic length, usually a typical diameter in the pore network, and inversely with the uids viscosity. We consider a plane REV with orthonormal vector n and denote the intersection of with the pore space by w (Figure 3.6). The macroscopic volume ow through w is then given by jw n =
w

v n dA = w v n

(3.20)

where w is the area of w and . . . w means the average over this area. Dividing by , noticing that w / = Vw / V = w if is an REV and if the pore space is suciently irregular and that n is arbitrary, we again obtain jw = w v w . With (3.18), v (x) = (x) p where (x) = 1 2 f (x) with f (x) a dimensionless scalar function and is some characteristic length of the pore space. Inserting this in (3.15) yields
2

jw = w

f p

(3.21)

Notice that f is not constant but varies in space at least as rapidly as p . As a consequence, the average depends strongly on correlations between f

3.2 Transition to Continuum Scale

47

p jw

Figure 3.7. Macroscopic ux jw and pressure gradient p (large arrows) in anisotropic media need not be parallel, even though their microscopic counterparts v and p are (small arrows).

and p , hence on the details of the pore space. However, due to the linearity of Stokes equation (2.30), f does not depend on the mean velocity, hence not on the macroscale pressure gradient. Thus, | pm | is proportional to | p w | although the two vectors are in general not parallel (Figure 3.7). From this, we nally obtain the macroscopic ux law 1 jw = k p , (3.22)

where k is a second rank tensor and where we have dropped the superscripts from the macroscopic quantities jw and p. This law is attributed to Darcy [1856] who, while studying water ow through sand lters in order to optimize the water supplies of the city of Dijon, suggested a linear relation between the water ow and the height dierence between the water levels at the inlet and outlet ends of the columns. The tensor k is considered as a material property of the pore space, called permeability. It can be shown to be symmetric and its elements have dimension L2 . Sometimes, particularly in the engineering literature, the elements of k are given in units of darcy with 1 d = 0.987 1012 m2 . For an isotropic medium, jw and p are parallel and k may be replaced by a scalar constant. We notice that the volume ux jw may be interpreted as a velocity, it is sometimes referred to as Darcy velocity, and corresponds to the velocity that would be required in the absence of the porous matrix, to sustain the given volume ux. It is related to average microscopic velocity v w by (3.15). This average is usually called the pore water velocity and, in formulations at the macroscopic scale, denoted by v. Hence, we have the relation jw = w v . (3.23)

The two constants that occur in Darcys law are often merged into one, the conductivity 1 K := k , (3.24) which may be interpreted as a material property of some geologic environments, for instance of an aquifer or an oil reservoir. Obviously, such a denition is only useful when the uid stays always the same.

48

3 Fluids in Porous Media

Multiple Driving Forces We deduced Darcys law under the premiss of stationary and slow ow of a Newtonian uid that is only driven by a pressure gradient. Extension to gravity as an additional driving force is straightforward whereas releasing the other assumptions leads to a much more complicated analysis and may even invalidate the very concept of relating macroscale velocity and macroscale driving force. When there are driving forces other then the negative pressure gradient, we can include them in our analysis by simply replacing p with the sum F of all driving forces per unit volume. This is correct for an isotropic uid and as long as the ow may be assumed to be stationary (Figure 2.4). The case encountered most often is a uid driven by gravity and pressure gradient for which Darcys law may be written as 1 jw = k[ p g] . (3.25)

We notice in passing that static equilibrium, that is jw = 0, implies p = g. Choosing z to point upwards with z = 0 at the free surface, this leads for incompressible media to p(z) = p0 gz or m (z) = gz . (3.26)

Non-Newtonian Fluids For non-Newtonian uids, a number of empirical viscosity models have been proposed to replace (2.11). A popular one is the power-law yx = |dy vx |n1 dy vx with coecients and n. One can show that this also leads to a power-law for the macroscopic velocity, |jw |n1 jw = 1 k p, e (3.27)

where the coecients n and e depend on the uid [Bird et al., 1960]. Weak Inertia For ow that cannot be considered as slow, hence Re Recrit , the eect of inertia is not yet dominating but also cannot be neglected. Probably the rst modication of Darcys law for this regime has been proposed by Forchheimer [1901]. For one-dimensional ow in a uniform and isotropic porous media, he suggested the empirical relation
n jw + jw = Kx p ,

(3.28)

where , n, and K are coecients that depend on the shape of the pore space. Using a homogenization approach, Mei and Auriault [1991] found n = 3 and concluded that Darcys law is thus also applicable for moderately high ow velocities. Non-Stationary Flow While weak inertia and non-Newtonian behavior can be handled at least in an empirical way, the situation is more severe for nonstationary ow. If the external time scale that is given by the temporal change

3.3 Material Properties

49

of the pressure at the boundary V is comparable or even smaller than the internal time scale which is given by the dynamics inside V , we cannot expect a functional relation between jw and p anymore and the very formulation of a ux law becomes impossible. This is generally not a problem for suciently small volumes and slowly changing external forcing. However, it becomes a dominating issue when the reference volume increases, as is the case when moving from the laboratory through the eld to the regional scale, and if the forcing is fast as for instance when starting to pump a well.

3.3 Material Properties

Combining the conservation of mass, for instance (3.16), and Darcys empirical ux law (3.25) readily yields a generic formulation for the macroscopic dynamics of uids in porous media. Using the matric potential m instead of pressure p and dropping subscripts this may be written as t [] + [j] = 0 j = K[ m g] . (3.29)

Obviously, this formulation must be supplemented by descriptions of the material properties. These must include at least the relation between and m , the hydraulic conductivity K, and possibly the density in case the uid is compressible. At this point, an important remark is in order: While (3.29) describes temporal changes of uid content and ow, we will in general invoke the local equilibrium hypothesis which in essence states that microscale processes are fast on the time scale of the macroscopic processes and hence in equilibrium. Consequently, material properties which reect the action of the microscopic processes at the macroscale do not depend on the speed of macroscopic evolution.

3.3.1 Capacity
The pressure jump across an interface the matric potential is related to the mean curvature of the interface (2.39). This curvature in turn is bound by the radius of the respective pore since the interfacial radius cannot be smaller than that of the pore. Hence, we expect for a porous medium a relation between the saturated volume fraction and the matric potential m and we dene the hydraulic or soil water capacity as C(m ) := d . dm (3.30)

50

3 Fluids in Porous Media z m g

rmax

Figure 3.8. Hydraulic capacity of porous medium illustrated for a bundle of capillaries.

z capillary fringe 0

m /[g]

Traditionally, the integral of this function, the uid fraction

0

(m ) = 0
m

C(m ) dm

(3.31)

as a function of the matric potential is more popular in the soil physical and in the engineering literature and graphical representations almost invariably show (m ), not C(m ). Depending on context and discipline (m ) is referred to as soil water characteristic, pressure-saturation relation, or desaturation-imbibition curve. Bundle of Capillaries Consider a bundle of capillaries that sticks in a free wetting uid with uid height given by (2.40) as illustrated in Figure 3.8. We rst notice the capillary fringe that is determined by the largest radius in the bundle. Above the capillary fringe, the uid fraction decreases monotonically with height. Consider a thin horizontal slab with height increment z, which corresponds to the increment m = gz of the matric potential according to (3.26). Denoting the corresponding increment of the uid fraction by , the hydraulic capacity C is then the ratio /m , in the limit m 0. We thus already recognize the important fact that (m ) can be read directly from the uid fraction above a free uid table for a stationary situation. This remains true also for natural porous media. We notice in passing, that the distribution of radii of a bundle of capillaries can be calculated easily from measurements of C(m ). This cannot be extended to natural porous media, however. Porous Medium While C(m ) and (m ) are unique functions for a bundle of cylindrical capillaries, the situations becomes much more complicated for a general porous medium where the pore radius is almost never constant. To understand the impact of this, we consider a single pore that is wider in the middle (Figure 3.9). Starting from a pore that is initially saturated with uid 1, pressure at one end shall decrease gradually. Once it falls below the entry pressure for uid 2, the interface moves gradually into the pore up to the next pore throat, where the pore radius is minimal. This corresponds to the uid distribution I in Figure 3.9. Reducing the pressure any further

3.3 Material Properties

51

2 I

III II Figure 3.9. Hysteresis of hydraulic capacity in natural porous media during invasion of uid 2 into pore initially lled with uid 1, and vice versa.

Figure 3.10. Pressure-saturation relation reported by Topp and Miller [1966] for water and air in thin packing of glass beads with mean diameter of 180 m. Water content was measured by -absorption and pressure with transducers. Top: Wetting loops, with the dashed line indicating the primary drying. Bottom: Drying loops. The arrows indicate the direction of movement. Notice that pressure p is given as equivalent height of a water column and saturation S equals the ratio between water content and porosity. Notice that, compared to the representation in Figure 3.8, the curves are rotated by 90 . This gure was extracted and combined from gures 2 and 3 of the original publication.

empties the entire cavity and leads to distribution II because the pore radius in the cavity is too large to sustain the interface whose radius is determined by the pressure jump across the interface. Reverting at this point and gradually increasing the pressure will not lead back to I. Instead, uid 1 will invade the cavity until the pressure is increased such that the jump at the interface corresponds to the largest radius. Increasing it any further has uid 1 ll the entire cavity and actually also the adjacent throat because the pore radii are smaller than the interfacial radius. Such discontinuous changes of the uid content are referred to as Haines jumps [Haines, 1930]. Understanding a single pore, we expect that for a porous medium both the hydraulic capacity C(m ) and also (m ) are hysteretic. This was already found in the rst experiments reported by Haines [1930] and Richards [1931] and it has been studied in great detail ever after. Topp and Miller [1966] did some of the early precise measurements of air-water systems in ow cells lled with various mixtures of glass beads. Exemplary results are shown in Figure 3.10.

52

3 Fluids in Porous Media

We rst consider the wetting or imbibition cycles that start with an initial drying of the water-saturated medium by reducing the matric head from 0 to about 65 cm. This is the so-called primary drainage branch. The imbibition branch that starts from this end point does not reach a complete saturation anymore: the saturation only reaches about 0.85, even for p = 0. This results from residual air, i.e., from air which is completely enclosed by water and does not form a connected phase anymore. Draining again produces the secondary drainage branch which is now only followed up to a certain matric head before imbibition is initiated again. Various of these so-called scanning loops are shown. Next, we consider the drying or desaturation cycles that start from different locations on the main imbibition branch and approach the secondary drainage branch. Apparently, the wetting and drying loops are dierent and resemble in shape the respective main branches that also limit them. What is the reason for the dierent shapes of the hysteresis loops? The initial preparation of the system was such that water lled the entire pore space. Through drainage and re-wetting a very complicated conguration of the water phase evolves. As long as both, the water- and the air-phase are connected, the curvature of their interface is the same everywhere and determined by the matric head. As is illustrated in Figure 3.9, this condition can be satised by many dierent congurations each of which is associated with a dierent water content. Moving the system through a succession of states along equilibrium paths, we thus cannot expect it to return to the same conguration by merely re-establishing the initial matric head. Also from Figure 3.9 we expect that the drainage branches reect the pore volume associated with bottle-necks while, vice versa, imbibition branches reect the cavities. We mention in closing that the experiments could also be started from a completely air-saturated (dry) state. This would lead to the primary imbibition branch and to corresponding scanning branches. Such experiments are dicult, however, because the associated equilibration times are much longer. As will be shown below, this is due to the greatly reduced hydraulic conductivity at low water saturations.

3.3.2 Conductivity
The hydraulic conductivity relates two vector quantities and is thus in general a second order tensor. While its directional nature has been addressed in a number of theoretical and numerical studies in the past, there is a paucity of experimental data on this aspect. Almost exclusively K is presumed to be isotropic and only the corresponding scalar K is measured. As we found previously with (3.24), the conductivity of a medium saturated with a single uid is a conglomerate of the mediums permeability and of the uids dynamic viscosity. The question now is: How does K

3.3 Material Properties

53

Figure 3.11. Pressure-conductivity relation corresponding to Figure 3.10. Left: Wetting Cycles. Right: Drying Cycles. The inset shows the primary drainage and main imbibition curves. This gure was extracted and combined from gures 2 and 3 of Topp and Miller [1966].

Figure 3.12. Saturation-conductivity relation corresponding to Figures 3.10 3.11. Left: Wetting Cycles. Right: Drying Cycles. Notice that the cycles practically overlay but are shifted relative to each other in order to separate them. This gure was extracted and combined from gures 2 and 3 of Topp and Miller [1966].

change in the presence of a second uid and is it a unique function of matric potential m or of uid fraction ? We rst convince ourselves that K(m ) is not unique by looking at Figure 3.9 and envisaging ow perpendicular to the drawing plane. Experimental evidence shown in Figure 3.11 nicely supports this intuition. Whether K() is a unique function is dicult to decide intuitively. However, experimental evidence reveals that this is indeed the case to a fair degree of accuracy as is corroborated by Figure 3.12). Evidently, the conductivity varies over several orders of magnitude even for rather moderate variations of . There are three main causes for this: With decreasing saturation (i) the cross-sectional area decreases, (ii) the uid is restricted to an every smaller class of pores to small ones if the uid is wetting and to large ones if it is nonwetting, (iii) the microscopic potential gradient decreases because the path within the uid of interest between any two points becomes larger.

54

3 Fluids in Porous Media

Comment on Anisotropic Media Most experiments for measuring the hydraulic conductivity impose a constant gradient along the axis of some cylindrical sample and measure the resulting uid ow. Unless the porous medium is indeed isotropic or at least one of its main axes is aligned with the cylinder axis, such a measurement does not provide a correct value for the conductivity component along the axis. The reason for this is that in the general case the anisotropy of K will lead to ow components orthogonal to the sample walls. Since this ow is impeded, a corresponding gradient of the hydraulic potential will be generated which in turn modies the ow along the axis, again through the anisotropy of K. Denoting the axis of the sample by z and the two orthogonal directions by x and y, a quick calculation shows that the measurement yields Kzz = Kzz
2 2 Kxz Kyy 2Kxy Kxz Kyz + Kxx Kyz , 2 Kxx Kyy Kxy

(3.32)

where Kij is the true ij-component of K. Clearly, the measured value Kzz only equals the true one if all o-diagonal elements vanish.

3.3.3 Flux Law

The empirical ux law for a single uid was introduced in Section 3.2.5 as Darcys law (3.22). The situation becomes much more dicult when more than one uid is present. Here, Buckinghams conjecture is typically invoked which states that Darcys law remains valid and the only modication required is to write K as a function of . Hence, for uid i, ji = Ki (i ) i , (3.33)

which is called the Buckingham-Darcy law in soil physics and hydrology and the generalized Kozeny-Carman law in petroleum industry. This law is remarkable for two reasons: (i) It states that the volume fraction of uid i is the sole factor that modies the conductivity which is corroborated by experiments. (ii) It states that the ux in uid i only depends on the hydraulic gradient in the same uid. From Onsagers theorem, we might instead expect a ux law of the form ji =
l

Kil (i , l ) l ,

(3.34)

where summation is over all phases l. This issue has not yet been explored, neither theoretically nor experimentally.

3.3 Material Properties

55

3.3.4 Compressibility
Porous media are generally compressible, many of them quite strongly as for instance most soils. In the following, we only consider weakly compressible media and think of water-saturated sandstone as a generic example. The compressibility of these media water and porous sandstone are so small that greatly simplifying assumptions are warranted. In particular, we assume that the conguration of grains does not change with pressure p which implies that the load on the grain contacts is large compared to the pressure change. Then, and are simple functions of p and the storage term of (3.29) may be written as t [(p)(p)] = d d + t p = water + [1 ]matrix t p p p dp dp
storage coecient S

= St p ,

(3.35)

where the compressibility p has been dened in (2.10) and S is the volumetric storage coecient. Since the compressibilities are constant to a very good approximation, S is also constant. Next, we look into the ux term [j] = K[ pg] , where m in (3.29) was replaced by p since we only consider water-saturated media. We assume for simplicity that K is isotropic and write it as K(p). Then K[ p g] = K and we further transform [ p g] + [K] [ p g] K (3.36)

[K]/[K] into p = water + p 1 dK K dp p. (3.37)

[K] 1 d 1 dK = + K dp K dp

For the relative change of K = 1 k with pressure we nd 1 d1 1 dk 1 dK = 1 + . K dp dp k dp (3.38)

Since is practically independent of p for liquids, we only need to calculate the relative change of the permeability k with pressure. To this end, we invoke a simple deformation model for the sandstone: Assume that the grains deform isotropically except at their contacts where only the area shall shrink (Figure 3.13). Then, the water phase remains similar with changing pressure. Since k is proportional to the square of the mean pore diameter , we obtain 1 dk 2d 2 = = matrix , (3.39) k dp dp 3 p

56

3 Fluids in Porous Media

Figure 3.13. In a simple model of weakly compressible media, the grains shrink isotropically with increasing pressure (dashed lines) except at the contact regions between grains where only the contact area decreases. Notice that shrinking is grossly exaggerated in this sketch and also porosity is rendered too large.

water grain

where the sign for the rst equality indicates that the permeability increases with decreasing grain diameter, and the last equality follows from V 3 1 together with p = V dV . Inserting everything into (3.37) yields dp 2 [K] = water + matrix p K 3 p p. (3.40)

With this, a quick analysis of scales shows that for weakly compressible media as we consider them here, the second term in parenthesis of (3.36) is many orders of magnitude smaller than the rst one. To a very good approximation, we may thus write for the ux term [j] = K[ p g] = K [ p g] , (3.41)

where and K are now constant.

Example: Sandstone Aquifer Consider a typical formation with porosity = 0.1 and compressibility matrix 1010 Pa1 . The compressibility of sand grains p is somewhat lower, about 3 1011 Pa1 , with the higher value of the matrix reecting the reduced contact area between the grains. Storage Coecient With 5 1010 Pa1 for the compressibility of water, the storage coecient becomes S 1.4 1010 Pa1 . Reducing the pressure in this formation by 105 Pa thus releases some 14 grams of water per cubic meter of aquifer, or some 140 ppm of the stored mass. Conductivity With (3.39) and assuming as constant, the relative change of K is about 109 Pa1 . To set this in proportion, we employ the hydrostatic pressure gradient dz p = g and nd a relative increase of K with depth due to the compressibility of water and matrix of some 105 m1 , which is quite negligible, indeed.

Notice that the treatment of the inuence of compressibility on the storage term t [] was quite dierent from that on the ux term: While both are exceedingly small for the example of the sandstone S 1.4 1010 Pa1 , dp K/K 109 Pa1 , and dp / 5 1010 Pa1 the storage coecient was retained in the formulation while the ux was approximated by its incompressible part. The reason for this is that with S = 0 the character of the dynamics changes qualitatively in that ow is always stationary. We will elaborate on this in Section 4.2.

3.3 Material Properties

57

Exercises
3.1 Units of Conductivity Darcys law may be written in terms of the hydraulic potential, jw = K w , or in terms of the hydraulic head, jw = K hw . For simplicity, the same symbol is often used for the hydraulic conductivity K. What are its units in the two formulations? 3.2 Conductivity Consider a uniform and isotropic porous medium with permeability k = 1010 m2 . What is its hydraulic conductivity at 10 C and at 20 C? Assuming the medium to be completely dry, what is its conductivity for air? If the same medium were scaled up by a factor of 10, i.e., pores are stretched by a factor of 10, how does its conductivity change?

Consider a vertical, perfectly waterwet capillary with radius R. Assume a constant lm ow in analogy to Exercise 2.2 and calculate the relation between lm thickness d and water ux jw := q/[R2 ], where q is the ow. Film ow in a capillary may be used as the most simple model for water ow in an unsaturated medium. Calculate and plot the relation between saturation and conductivity K. Discuss. Envisage a porous medium with a fractal distribution of its mass. Sketch and discuss the average density as a function of the averaging volume.
3.4 REV and Fractal Media

3.3 Conductivity Function of Capillary

Consider a vertical, 1 m long column uniformly lled with quartz sand. The columns porosity is 0.3 and the hydraulic conductivity Kw = 107 m s1 . It is water-saturated and the water table is held constant at the columns upper end. At the lower end, water ows out freely. The column is thermally isolated and initally at T = 10 C as is the inowing water. Assume that the water equilibrates thermally with the column as it ows through. Estimate the rate of temperature increase of the outowing water.

3.5 Heating by Water Flow