Documente Academic
Documente Profesional
Documente Cultură
1, JANUARY 1977 49
criteria for multispectral remote sensing," in Int. Joint Conf. 161 D. M.Gates, H. J. Keegan, J. C. Schleter, and V. R. Weidner,
Pattern Recognition, 5 pp., 1973; also Laboratory for Applications "Spectral properties of plants," Appl. Opt., vol. 4, pp. 11-20,
of Remote Sensing, Purdue Univ., W. Lafayette, IN, LARS Infor- 1965.
mation Note 042673. 1171 A. I. Obukhov and D. S. Orlov, Pochvovedeniye, 174, no. 2, 1964.
1151 K. S. Fu and P. J. Min, "On feature selection in multiclass pattern [181 R. Kumar, "Radiation from plants-Reflection and emission: A
recognition," School of Electrical Engineering, Purdue Univ., W. review," Purdue Univ., W. Lafayette, IN, AA & ES 72-2-2, Feb.
Lafayette, IN, Tech. Rep. TR-EE 68-17, July 1968. 1972. 8 pp.
Abstract-A model of soil composed of a multiphase mixture of solid form of the curve is similar even near microwave frequencies
particles, water, and air voids is proposed from which the complex where water dipole relaxation begins to dominate and where
permittivity, or dielectric constant and conductivity, is calculated. It is
based on the Hanai/Bruggelman/Wagner theory of mixtures and con- this theory is not intended to apply [8]. In other words, a
siders the ionic conducting water as partly dispersed and partly the "normal curve" exists which approximately describes the be-
dispersing medium, an important distinction with this theory. The havior of most soils [4], [9]. One aim of this paper is to show
permittivity as a function of frequency and water content is predicted. that the approximate normal curve is predictable from an
The increase in dielectric constant with water volume fraction does not extension of existing dielectric theories of mixtures. This
differ greatly with soil type in the high-frequency limit and is approxi-
mated by a normal curve specified by the theory. A normal curve and theory, termed the semidisperse soil model, also predicts the
the marked increase in dielectric constant at lower frequencies has been variation in permittivity with frequency, specifically relating
observed experimentally. It is concluded that this semidisperse theory it to the ionic conductivity of the water and the manner and
of the dielectric permittivity is successful in describing the behavior of degree of dispersion in water in the soil.
soils containing moisture in the high-frequency range (1 MHz-1 GHz).
Approximations to the more detailed theory and a series-parallel RC HANAI/BRUGGELMAN/WAGNER (HBW) THEORY
equivalent circuit are given.
A theory of two-phase heterogeneous dielectrics with com-
plex dielectric constants was developed by Hanai [10]. It is an
INTRODUCTION extension of Bruggelman's theory of the conductivity of mix-
IT HAS LONG BEEN recognized that the dielectric permit- tures, and is based on extrapolation of Wagner's more exact
tivity of soil increased with its water content [1]. Indeed theory of suspended dilute spheres to the concentrated state.
such an effect is predicted by the theory of heterogeneous di- More recently Pearce, Hulse, and Walker [11] applied the
electrics which was developed before this century. An exten- equations derived by Hanai to the calculations of the dielectric
sive review of this and subsequent work is given by Hasted [21 permittivity of soil considered as a three-phase system (parti-
and by van Beek [3]. However it is not widely recognized, cles, air, and water). They were able to obtain rather good
although observed experimentally by Thomas [4] some time agreement with experiment at high frequencies but the increase
ago, that the increase in permittivity with volume fraction of in permittivity with decreasing frequency (dispersion) was un-
water at high frequencies is quite similar for most soils in spite realistically small.
of their diverse composition. The measurements we have In Fig. 1 the steps in the development of the theory are
made [5 ] on a number of soils at radio frequencies is consistent summarized. Wagner's theory [2] is derived for spheres widely
with this observation. Arulanandan et al. [6] and Hipp et al. dispersed or suspended in a medium where both phases are
[7] observe generally higher permittivity for clays, but the partially conducting dielectrics. For each phase a complex
permittivity e* can be defimed as
Manuscript received July 29, 1976. This work was supported in part
by the National Transportation Research Board under a contract to e _+ jc"Keo -
(1)
develop a soil moisture sensor. Co
The author is with the Departments of Electrical Engineering and
Biophysical Sciences, State University of New York at Buffalo, Buffalo, where K is the real part of the dielectric constant, co is the
NY 14226. permittivity of free space (8.854 X 10-12 F/m), a is the con-
50 IEEE TRANSACTIONS ON GEOSCIENCE ELECTRONICS, JANUARY 1977
I.
plex permittivity of the mixture. Since 0', e*, and ed are
presumably known, e* can be obtained from (2), a cubic
X\ 'k ; equation with complex coefficients.
II Equation (2) was solved by a computer routine (ZANLYT)
\
capable of finding the roots to a third-degree polynomial with
(a)I complex coefficients. A procedure in which the imaginary
and real parts of (2) were solved iteratively also proved
efficient. Hanai gives the real part of (2) as
[(C Cd)2 +(C"- Cd)2 1 [(CXI)2 (Cl)2 1
+
2
[(Cl ElC)2 + (C' E )2 )2 +(EC )2 J
(3a)
and the imaginary part as
m
(b)
e e X(3- X2)
(3b)
" 1- 3X2
1+ ,
Em E
where
e Ed em Ed
X- Ed em Ed
(c) (3c)
Fig. 1. Theories of mixtures. (a) Dilute dielectric spheres (Wagner).
(b) Two-phase concentrated dielectric mixtures (Bruggelman/Hanai). ( Ed -
em E Ed Ed
(c) Three-phase mixture of solid particles, water, and air (Pearce et al.).
Trial values of c' and e" were inserted into (3a, b, c) and then
and C" were varied by computer iteration until both
ductivity (mho/m), w is the angular frequency (rad/s), and sides of the equations were equal.' Identification of the cor-
j = .-1 The dielectric permittivity of the mixture can be rect solution from the three roots was not always obvious
calculated from the permittivities and volume fraction of the since under some conditions more than one root is physically
components. The derivation involves the calculation of the allowable (e' > 1, e" < 0) but ambiguities were removed by
electric field in both phases through a matching of the bound- tracing the evolution of the root from the unambiguous high-
ary conditions at the sphere surface. frequency limit.
In applying the theory to soil it is assumed that soil particles, In applying (2) to any mixture, it is necessary to know which
although of irregular shape, are randomly distributed and thus phase is disperse because the results of this inexact theory can
have a spherical average. While objections can be raised be quite different if e* and Ed are interchanged (and O re-
against this assumption, a more serious problem is the require- placed by Om = 1 0). A lack of symmetry between phases
-
ment that the particle concentration be dilute, which is far for this approximate relation is not unexpected since it was
from true. Bruggelman [2], [3] approached this problem by derived assuming 0 << 1, but the distinction is more funda-
calculating the dielectric constant produced by a series of mental since dispersions of oil in water and water in oil show
infinitesimal additions to the disperse phase (spheres) with a a marked difference experimentally. Surprisingly in many
corresponding reduction in the dispersing medium. While the cases (2) has been found experimentally [10] to be approxi-
validity of this approach is not clear theoretically, in particular mately valid for ranging up to 0.7 or even 0.95, well beyond
that the boundary conditions still hold, the formulas derived the range which can be theoretically justified. Therefore a
have been verified experimentally in many mixtures. Bruggel- phase present even in high concentrations, such as particles in
man's derivation was intended to cover the case of real permit- soil, may be considered disperse, but the assignment of the
tivities, or to the analogous case of pure conductivities, but not disperse and suspending medium is not arbitrary and ultimately
complex permittivities. However, Hanai [10] found that the must be justified experimentally.
result for complex permittivities (e*) was the same as that for It is instructive to examine the character of the solutions of
real permittivities (e) or conductivities (a), specifically for a (2) for mixtures of a conducting high-dielectric phase (water)
two-phase mixture and a low-dielectric phase (soil particles, for example) which
may be conducting or nonconducting. Frequency plots of the
Ce -d (>) (2) real and imaginary parts of the permittivity expressed in terms
cm Ed
of dielectric constant and conductivity are given in Fig. 2 for
where ¢ is the volume fraction of the disperse phase, Ed and
e* are the complex permittivities of the disperse phase and Computer programs may be obtained from the author. ZANLYT is
medium in which dispersed, respectively, and e* is the com- an IMSL routine.
WOBSCHALL: DIELECTRIC PERMITTIVITY OF SOIL 51
Log (Frequency) - Hz
Log (Frequency) - Hz
Fig. 2. Permittivity of solid-water dispersions as a function of frequency
as calculated by the Hanai theory. The solid is the low dielectric
phase. Comparisons of the conducting, nonconducting, disperse, and
nondisperse solid phases are shown.
the cases of a) the high-dielectric-constant phase (H) dispersed conductor-dielectric interface. For appreciable charge to be
and b) the low-dielectric-constant phase (L) dispersed. It accumulated the low-dielectric material (here considered a
should be noted that marked variation of the dielectric con- solid particle) must act as a conductor at the measurement
stant with frequency2 occurs only if L is conducting. Very frequency. This will occur when the relaxation time (e/a) is
high dielectric constants at low frequencies are predicted if small compared to the period (27r/c). The higher dielectric
L is disperse as well as conducting. Some frequency depen- material (water) acts as a pure dielectric (or at least has a
dence occurs for L nonconducting, but it is quite small. The substantially longer relaxation time). In effect the conducting
frequency dependence of the conductivity of the mixture particles separated by thin insulating regions act as a series of
follows a similar behavior. Further, if L is nonconducting with capacitors. Since the electric field is concentrated across the
H disperse, the net conductivity is small and indeed approaches thin high-dielectric material the charge accumulation can be
zero in the low-frequency limit. Clearly the theory predicts very large. At high frequencies the particle conductivity be-
quite different results for H rather than L disperse. comes unimportant and the material acts as a mixture of pure
The physical reason for the increase in dielectric constant at dielectrics.
lower frequencies is the accumulation of charge at the
SEMIDISPERSE SOIL MODEL
2Care must be taken to avoid confusion of the variation of the di-
Soil consists of irregular particles containing crevices and
electric constant with frequency, usually referred to as dispersion, pores (air-filled voids) which are gradually filled by water as
with the disperse phase. the moisture content increases. No single disperse phase is
52 IEEE TRANSACTIONS ON GEOSCIENCE ELECTRONICS, JANUARY 1977
two-phase volume fraction of the disperse phase is ductivity is due to water in small crevices is physically more
reasonable than attributing the conductivity to the solid
O=1-¢'I. (12) particle. For these reasons we prefer the four-stage calculation
even though it is lengthier. Actually a five-stage calculation
Referring to (2), here Ed = coKp (nonconducting particles), may be more realistic for very inhomogeneous soils such as
e* = e*, and the mixture (conducting particle) permittivity clays which have not been thoroughly mixed.
C* is calculated where It might be supposed, in view of the rather large number of
parameters and variables introduced, that the solutions ob-
* = eoK _ i ( . (13) tained by this procedure would be complicated and conse-
quently would have little predictive value. This is not the case
If a distinction between the conductivity of microcrevice for the range of parameters appropriate to most soils, and, in
water a' and pore water a, is made, then a' replaces ow for fact, rather a fairly narrow range of solutions exist which, at
the first stage only. high frequencies, approach the normal curve experimentally
In the second stage, water is considered dispersed in pores observed. To demonstrate this, the dependence of the solu-
within the conducting particle phase to make up the moist tions (K, and a,) on the various parameters will be examined.
particle phase q, the two-phase volume fraction is The variables and parameters may be grouped as follows:
Oq= e
+ 4/e
(14) Variables:
volume fraction of water,
Here ed = c , e* = eZ, and e* is calculated by (2), or (3). f frequency in Hz (= cl/21r),
In the third stage, the moist particle phase is considered dis- Fixed Parameters:
persed in the remaining water to form the air-free phase c. The Kw dielectric constant of water (- 79),
volume fraction of the disperse phase is Kp dielectric constant of soil particles (n3.5),
Measurable Parameters:
_4/m4/a
/= I- I- (15) ow conductivity of pore water (containing dissolved salts)
in mmho,
Here Ed =4,e = *,and e* is calculated. ay conductivity of microcrevice water (or' = ac unless
E*
noted),
*
e* =eoKs_j
(- ) (17) The fixed parameters are nearly constant for all soils. Ac-
tually Kw varies slightly with temperature and frequency [7]
where, in accordance with the previous definition, Ks is the (see (4)) but this variation is insignificant. Neglect of the
(real) dielectric constant of the soil and rs is the soil conduc- variation of Kp with soil type is justified by the argument that
tivity at the measurement frequency. Kp is nearly the same for all soils (Kp 3.5). It is not clear
-
An alternate form of the semidisperse theory was studied in that this argument is valid for soils in which a substantial part
which the first step, the calculation of the conducting particle of the water is chemiadsorbed (some clays).
permittivity c*, was eliminated by directly assuming a value Measurable parameters can be determined for any soil by
for e* = Eo0K - jap/w, where ap is the particle conductivity. laboratory tests. For most soils below saturation the void
In this case up takes the place of : as an adjustable parameter. volume fraction is between 0.3 and 0.6; a typical value of 0.5
There is no reduction in the number of parameters even has been chosen for many of the calculations given here.
though the number of steps in the calculation is reduced from Water conductivities typically range from 0.01 to 50 mmho/
four to three. The difference between the three- and four-step cm, depending on the concentration and kind of dissolved salt.
process is primarily whether the conducting particle permittiv- Further ow might vary with moisture content, specifically de-
ity is independent of or dependent on frequency. However creasing at high values of 4w as the salt is diluted by the added
because the low-dielectric-constant nonconducting phase is water which presumably contains less salt.
disperse in the first step of the four-step process, the variation Both ao and are quantities associated with the hypothetical
of the dielectric constant and conductivity with frequency is moist particle phases. They are treated here as theoretical
small (see Fig. 2) and the distinction is not very significant. parameters which can be adjusted to f1tthe model to the ob-
We feel also that the mechanism by which the particle con- served permittivity data. Independent experimental measure-
54 IEEE TRANSACTIONS ON GEOSCIENCE ELECTRONICS, JANUARY 1977
o Iw=0.47 w- 1.0
0e9 =O.5O 3:0.05
o0.7
_0ao.3
VI,
-o~~~~~~~~~~~~~~~\o-3
a:~~~~~~~~~~~~~~~=.
z°t
zo
-
r
u a's
M a-oos3 am-0.o*s0.7
ro
C?~~~~~4 O azo.9
00
.oo 6'so a.oo 6'.GO 7.00 7 .60 8'.oo 8.5 B'.oo -lb
L0G(FREQUENCY) -- HZ
Fig. 4. Permittivity of soil as a function of frequency as calculated by the semidisperse theory. Results for several values
of the disperse index (oe) are shown. Note the approach to the high- and low-frequency limits of dielectric constant
(K., and Ko) and conductivity (a., and o0).
100(-
/Low
Frequency
7 " / Limit (KJ Kw
7~~~~~'
1.0 o.oI-L e7-0 02 Lneor , /
1' 50 / Proportion-r- /
Fig. 7. High-frequency limit of the permittivity as a function of water content. Results for several values of a are shown.
much higher than those obtained theoretically by Pearce et al. Fig. 8. Soil equivalent circuit.
[11]. It should be pointed out that very large values of dielec-
tric constant at low frequencies, much above that of water, are
theoretically possible with an appropriate selection of parame- tion of water content, with a and : as parameters, is
ters. The theoretical dielectric-constant limits are sketched in
Fig. 6, which is intended to show an overview of the dielectric- K.-~Ao +A l w +A24iw (19a)
constant variation with water volume fraction. The low- where
frequency limit, in particular, depends critically on the water Ao = (Kp - 1)(1 - Qv) + 0.8 + 12.40
dispersion and conductivity. In the high-frequency limit, how- (19b)
ever, the composite dielectric constant cannot exceed, and is A1 = 29.4- 21.1a- 90a43- 80f (19c)
generally less than, the average of the dielectric constant of the A2 =29+l150
individual components, weighted according to their volume
. (1 9d)
fractions. The conductivity limit is
Of particular practical interest is the high-frequency limit of
u*. (wu(Bo +B1jw +B244) (20a)
K, because it depends primarily on the volume fraction of
water and its dispersion, and not on its conductivity (see where
Fig. 7). Soil conductivity as is proportional to water conduc-
tivity a,, and nonlinearly dependent on volume fraction 4. Bo = 0.12(, (20b)
It is convenient to plot the high-frequency limit of the ratio B1 =0.28+1.7cta-0.31a- (20c)
aslaw as a function of 4,, as shown in Fig. 7 because this ratio B2 =0.62 + 2f3- 0.25 p3- 1.2a. (20d)
is nearly independent of or,
These approximations are accurate to at least ±5 percent in
POLYNOMIAL APPROXIMATION AND AN the range: 0.2 > 4, > 0.6, 0.3 > a > 0.9, 0.01 > 3 > 0.1, and
EQUIVALENT CIRCUIT 1.5 >Kp >5.0.
Because the calculations required by the theory are tedious An approximation for the frequency dependence in terms of
even when done by computer, a simpler mathematical expres- D and f, is
sion is desirable. An approximate expression of the results
(Fig. 7) was obtained by a polynomial curve fitting procedure.
The high-frequency limit of the dielectric constant as a func- +1 +(flfXk)2] (21)
56 IEEE TRANSACTIONS ON GEOSCIENCE ELECTRONICS, JANUARY 1977
TABLE I
PARAMETERS OF THEORETICAL CURVES IN FIGS. 9 AND 10
ID Suil Type Ref f aw
C Comments
~~~ 1
!_
v
Units
it Mhz Mmho
and grettably a complete set of such data was not found published.
Partial data are available for many soils, however, and a few
selected examples of such data [4], [6], [7] will be compared
UsUO [1+1 +(fxa/f)2l (22) with the theory in order to demonstrate that it can be fitted to
a range of soil types.
where Dk, D, fxk, and fxa are given by Fig. 5. It should be emphasized that the permittivity is a function
Frequency dependence can be modeled in terms of the of water content expressed as a volume fraction, that is,
equivalent circuit of Fig. 8. It is represented, in terms of soil volume of water per total soil volume (including solids, air,
impedance defined for a specific test chamber with a given and water). When water content is expressed as mass percent
electrode configuration, as which is more conventional, the correlation is much poorer
1 = jcoc*a = jweoaKs + aa (23) experimentally. With this theory the poor correlation is
expected since the dielectric permittivity is only slightly de-
where a is the ratio of capacitance to dielectric constant which pendent on porosity (4'). Soil density is an irrelevant variable
depends on the geometry of the test chamber (a = coA/d for a and if put into the expression for moisture content (by weight)
parallel plate capacitor; A is the electrode area and d the and element of randomness is introduced.
sample thickness, assumed small). Expressing Zs explicitly in The volume fraction of water l,1 is related to the moisture
terms of frequency as implied by Fig. 8 and inserting a* into content Wc, expressed as weight of water per weight of soil in
(1 7), an equation similar in form to (21) and (22) is obtained. percent, by the following equation:
A defect of this model is that the same break frequency
(f D=D uO/27rDKK.) is predicted for both the capacitance w = 0-
WC (24)
and conductance components while in fact fxK and fx a (equa- W1000'y
tions (21) and (22)) are unequal except for higher values of,B where 'y is the density of the soil. In the cm-gm-s system of
(see Fig. 5). It was found by curve fitting that log (fx) units, Qw is numerically equal to Mu, the mass of water per
log (f,) = 7.24 log (u,). unit volume.
An alternative equivalent circuit for soil with four capacitors Data obtained on the high-frequency dielectric constant as a
and four resistors has been described by Arulanandan and co- function of moisture content (by volume) are compared with
workers [6]. It is based on an intuitive, and in our view over- the theory in Fig. 9. (See Table I for the parameters of the
simplified, model introduced by Sachs and Spiegler [12] in curves in Figs. 9 and 10.) An adequate fit in the high-frequency
which the particle and water capacitance and conductance are limit is obtained for a = 0.9 (,B- 0.07), which corresponds to
individually lumped together to form a resultant series-parallel the normal curve discussed above. A fairly narrow range of
RC network. Because of the larger number of adjustable limiting curves is expected by the theory because the dielectric
parameters, matching of the frequency dependence can be constant in the high-frequency limit is independent of particle
done more accurately than with the simpler model of Fig. 8, shape and size, is nearly independent of soil porosity and pore
but at the expense of a greater number of arbitrarily adjustable water conductivity, and varies only moderately with dispersion
parameters. A more serious disadvantage is that the depen- (as characterized by a and ,B). Although the theory is intended
pendence on water content is inaccurately predicted. to apply to all types of soil, it is more likely to be valid for
sand than for clay since sand particles more closely approxi-
COMPARISON WITH EXPERIMENT mate the solid low-dielectric-constant spherical particles as-
A critical test of the theory would require data on the sumed in the derivation. In spite of our theoretical reserva-
dielectric constant and conductivity as a function of water tions, it appears experimentally that clays also approach nearly
volume fraction and frequency on a soil sample of known the same normal curve as sand although the limit is reached at
porosity, water conductivity, and particle permittivity. Re- much higher frequencies (0.3-3 GHz). Data taken in the
WOBSCHALL: DIELECTRIC PERMITTIVITY OF SOIL 57
orS
A Normal Curve #
La
at 30 MHZ ,
Fig. 9. Dielectric constant of the soil high-frequency limit as a function of water content. A comparison of experiment
with theory (Table I) is shown.
c.-I
* Montmorillonite/Sand (Ref 6)
* Cloy Loorm, San Antonio (Ref 7)
X Sandy Loom, New Hampshire
I-
a)
zo
cl:
z
o)
(JO.
.
to
u
Cl-.
Lo-
t
LL
_; D
0 O.
EoW
x > S_Va
v
10-1 00-MHz range may differ appreciably from the high- above 0kw 0.35. Such a spread is expected theoretically
frequency limit. In this range especially, it is better to think because Ks depends on the conductivity below the high-
of the normal curve as a band centered about some mean frequency limit, and the conductivity varies greatly with soil
value. At 30 MHz, for example, the values of K. observed for type. At higher frequencies the percent difference between
various soils [4] lie 5 percent to over 50 percent above the Ks and its high-frequency limit is roughly proportional to the
high-frequency limit with the deviation increasing rapidly soil conductivity (of the order of 3-15 percent per mmho/cm
58 IEEE TRANSACTIONS ON GEOSCIENCE ELECTRONICS, JANUARY 1977